Skip to contentSkip to conversation

CXL 2.0 Support

RFC PATCH Superseded Ben Widawsky 6 participants 65 messages Feb 2, 2021 – Dec 2, 2021 Arm/Amlogic Meson SoC support

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v3 cover CXL 2.0 Support

Major changes since v2 [1]:
* Removed all register endian/alignment/size checking. Using core functionality instead. This untested on big endian hosts, but Should Work(tm).
* Fix component capability header generation (off by 1).
* Fixed HDM programming (multiple issues).
* Fixed timestamp command implementations.
* Added commands: GET_FIRMWARE_UPDATE_INFO, GET_PARTITION_INFO, GET_LSA, SET_LSA

Things have remained fairly stable since since v2. The biggest change here is definitely the HDM programming which has received limited (but not 0) testing in the Linux driver.

Jonathan Cameron has gotten this patch series working on ARM [2], and added some much sought after functionality [3].

---

I've started #cxl on OFTC IRC for discussion. Please feel free to use that channel for questions or suggestions in addition to #qemu.

---

Introduce emulation of Compute Express Link 2.0
(https://www.computeexpresslink.org/). Specifically, add support for Type 3 memory expanders with persistent memory.

The emulation has been critical to get the Linux enabling started [4], it would be an ideal place to land regression tests for different topology handling, and there may be applications for this emulation as a way for a guest to manipulate its address space relative to different performance memories.

Three of the five CXL component types are emulated with some level of
functionality: host bridge, root port, and memory device. All components and devices implement basic MMIO. Devices/memory devices implement the mailbo interface. Basic ACPI support is also included. Upstream ports and downstream ports aren't implemented (the two components needed to make up a switch).

CXL 2.0 is built on top of PCIe (see spec for details). As a result, much of the implementation utilizes existing PCI paradigms. To implement the host bridge, I've chosen to use PXB (PCI Expander Bridge). It seemed to be the most natural fit even though it doesn't directly map to how hardware will work. For persistent capacity of the memory device, I utilized the memory subsystem (hw/mem).

We have 3 reasons why this work is valuable:
1. Linux driver feature development benefits from emulation both due to a lack of initial hardware availability, but also, as is seen with NVDIMM/PMEM emulation, there is value in being able to share topologies with system-software developers even after hardware is available.

2. The Linux kernel's unit test suite for NVDIMM/PMEM ended up injecting fake resources via custom modules (nfit_test). In retrospect a QEMU emulation of nfit_test capabilities would have made the test environment more portable, and allowed for easier community contributions of example configurations.

3. This is still being fleshed out, but in short it provides a standardized mechanism for the guest to provide feedback to the host about size and placement needs of the memory. After the host gives the guest a physical window mapping to the CXL device, the emulated HDM decoders allow the guest a way to tell the host how much it wants and where. There are likely simpler ways to do this, but they'd require inventing a new interface and you'd need to have diverging driver code in the guest programming of the HDM decoder vs. the host. Since we've already done this work, why not use it?

There is quite a long list of work to do for full spec compliance, but I don't believe that any of it precludes merging. Off the top of my head:
- Main host bridge support (WIP)
- Interleaving
- Better Tests
- Hot plug support
- Emulating volatile capacity
- CDAT emulation [3]

The flow of the patches in general is to define all the data structures and registers associated with the various components in a top down manner. Host bridge, component, ports, devices. Then, the actual implementation is done in the same order.

The summary is:
1-5: Infrastructure for component and device emulation 6-9: Basic mailbox command implementations
10-19: Implement CXL host bridges as PXB devices 20: Implement a root port
21-22: Implement a memory device
23-26: ACPI bits
27-29: Add some more advanced mailbox command implementations 30: Start working on enabling the main host bridge 31: Basic test case

---

[1]: https://lore.kernel.org/qemu-devel/[email protected]/
[2]: https://lore.kernel.org/qemu-devel/[email protected]/
[3]: https://lore.kernel.org/qemu-devel/[email protected]/
[4]: https://lore.kernel.org/linux-cxl/[email protected]/

---

Ben Widawsky (31):
hw/pci/cxl: Add a CXL component type (interface) hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5) hw/cxl/device: Introduce a CXL device (8.2.8) hw/cxl/device: Implement the CAP array (8.2.8.1-2) hw/cxl/device: Implement basic mailbox (8.2.8.4) hw/cxl/device: Add memory device utilities hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1) hw/cxl/device: Timestamp implementation (8.2.9.3) hw/cxl/device: Add log commands (8.2.9.4) + CEL hw/pxb: Use a type for realizing expanders hw/pci/cxl: Create a CXL bus type
hw/pxb: Allow creation of a CXL PXB (host bridge)
qtest: allow DSDT acpi table changes
acpi/pci: Consolidate host bridge setup
tests/acpi: remove stale allowed tables
hw/pci: Plumb _UID through host bridges hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142) acpi/pxb/cxl: Reserve host bridge MMIO
hw/pxb/cxl: Add "windows" for host bridges
hw/cxl/rp: Add a root port
hw/cxl/device: Add a memory device (8.2.8.5) hw/cxl/device: Implement MMIO HDM decoding (8.2.5.12) acpi/cxl: Add _OSC implementation (9.14.2) tests/acpi: allow CEDT table addition
acpi/cxl: Create the CEDT (9.14.1)
tests/acpi: Add new CEDT files
hw/cxl/device: Add some trivial commands hw/cxl/device: Plumb real LSA sizing
hw/cxl/device: Implement get/set LSA
qtest/cxl: Add very basic sanity tests
WIP: i386/cxl: Initialize a host bridge

 MAINTAINERS                         |   6 +
 hw/Kconfig                          |   1 +
 hw/acpi/Kconfig                     |   5 +
 hw/acpi/cxl.c                       | 173 ++++++++++
 hw/acpi/meson.build                 |   1 +
 hw/arm/virt.c                       |   1 +
 hw/core/machine.c                   |  26 ++
 hw/core/numa.c                      |   3 +
 hw/cxl/Kconfig                      |   3 +
 hw/cxl/cxl-component-utils.c        | 208 ++++++++++++
 hw/cxl/cxl-device-utils.c           | 264 +++++++++++++++
 hw/cxl/cxl-mailbox-utils.c          | 498 ++++++++++++++++++++++++++++
 hw/cxl/meson.build                  |   5 +
 hw/i386/acpi-build.c                |  87 ++++-
 hw/i386/microvm.c                   |   1 +
 hw/i386/pc.c                        |   2 +
 hw/mem/Kconfig                      |   5 +
 hw/mem/cxl_type3.c                  | 405 ++++++++++++++++++++++
 hw/mem/meson.build                  |   1 +
 hw/meson.build                      |   1 +
 hw/pci-bridge/Kconfig               |   5 +
 hw/pci-bridge/cxl_root_port.c       | 231 +++++++++++++
 hw/pci-bridge/meson.build           |   1 +
 hw/pci-bridge/pci_expander_bridge.c | 209 +++++++++++-
 hw/pci-bridge/pcie_root_port.c      |   6 +-
 hw/pci/pci.c                        |  32 +-
 hw/pci/pcie.c                       |  30 ++
 hw/ppc/spapr.c                      |   2 +
 include/hw/acpi/cxl.h               |  27 ++
 include/hw/boards.h                 |   2 +
 include/hw/cxl/cxl.h                |  29 ++
 include/hw/cxl/cxl_component.h      | 187 +++++++++++
 include/hw/cxl/cxl_device.h         | 255 ++++++++++++++
 include/hw/cxl/cxl_pci.h            | 160 +++++++++
 include/hw/pci/pci.h                |  15 +
 include/hw/pci/pci_bridge.h         |  25 ++
 include/hw/pci/pci_bus.h            |   8 +
 include/hw/pci/pci_ids.h            |   1 +
 monitor/hmp-cmds.c                  |  15 +
 qapi/machine.json                   |   1 +
 tests/data/acpi/pc/CEDT             | Bin 0 -> 36 bytes
 tests/data/acpi/pc/DSDT             | Bin 5065 -> 5065 bytes
 tests/data/acpi/pc/DSDT.acpihmat    | Bin 6390 -> 6390 bytes
 tests/data/acpi/pc/DSDT.bridge      | Bin 6924 -> 6924 bytes
 tests/data/acpi/pc/DSDT.cphp        | Bin 5529 -> 5529 bytes
 tests/data/acpi/pc/DSDT.dimmpxm     | Bin 6719 -> 6719 bytes
 tests/data/acpi/pc/DSDT.hpbridge    | Bin 5026 -> 5026 bytes
 tests/data/acpi/pc/DSDT.hpbrroot    | Bin 3084 -> 3084 bytes
 tests/data/acpi/pc/DSDT.ipmikcs     | Bin 5137 -> 5137 bytes
 tests/data/acpi/pc/DSDT.memhp       | Bin 6424 -> 6424 bytes
 tests/data/acpi/pc/DSDT.numamem     | Bin 5071 -> 5071 bytes
 tests/data/acpi/pc/DSDT.roothp      | Bin 5261 -> 5261 bytes
 tests/data/acpi/q35/CEDT            | Bin 0 -> 36 bytes
 tests/data/acpi/q35/DSDT            | Bin 7801 -> 7801 bytes
 tests/data/acpi/q35/DSDT.acpihmat   | Bin 9126 -> 9126 bytes
 tests/data/acpi/q35/DSDT.bridge     | Bin 7819 -> 7819 bytes
 tests/data/acpi/q35/DSDT.cphp       | Bin 8265 -> 8265 bytes
 tests/data/acpi/q35/DSDT.dimmpxm    | Bin 9455 -> 9455 bytes
 tests/data/acpi/q35/DSDT.ipmibt     | Bin 7876 -> 7876 bytes
 tests/data/acpi/q35/DSDT.memhp      | Bin 9160 -> 9160 bytes
 tests/data/acpi/q35/DSDT.mmio64     | Bin 8932 -> 8932 bytes
 tests/data/acpi/q35/DSDT.numamem    | Bin 7807 -> 7807 bytes
 tests/qtest/cxl-test.c              |  93 ++++++
 tests/qtest/meson.build             |   4 +
 64 files changed, 3004 insertions(+), 30 deletions(-)
 create mode 100644 hw/acpi/cxl.c
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/cxl-device-utils.c
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 hw/mem/cxl_type3.c
 create mode 100644 hw/pci-bridge/cxl_root_port.c
 create mode 100644 include/hw/acpi/cxl.h
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_device.h
 create mode 100644 include/hw/cxl/cxl_pci.h
 create mode 100644 tests/data/acpi/pc/CEDT
 create mode 100644 tests/data/acpi/q35/CEDT
 create mode 100644 tests/qtest/cxl-test.c
Signature · 1 line
2.30.0
#1 lore raw

RFC PATCH v3 1/31 hw/pci/cxl: Add a CXL component type (interface)

A CXL component is a hardware entity that implements CXL component registers from the CXL 2.0 spec (8.2.3). Currently these represent 3 general types.
1. Host Bridge
2. Ports (root, upstream, downstream)
3. Devices (memory, other)

A CXL component can be conceptually thought of as a PCIe device with extra functionality when enumerated and enabled. For this reason, CXL does here, and will continue to add on to existing PCI code paths.

Host bridges will typically need to be handled specially and so they can implement this newly introduced interface or not. All other components should implement this interface. Implementing this interface allows the core pci code to treat these devices as special where appropriate.

---
 hw/pci/pci.c         | 10 ++++++++++
 include/hw/pci/pci.h |  8 ++++++++
 2 files changed, 18 insertions(+)
hw/pci/pci.c +10 -0
index 512e9042ff..a45ca326ed 100644
@@ -194,6 +194,11 @@ static const TypeInfo pci_bus_info = {
     .class_init = pci_bus_class_init,
 };
 
+static const TypeInfo cxl_interface_info = {
+    .name          = INTERFACE_CXL_DEVICE,
+    .parent        = TYPE_INTERFACE,
+};
+
 static const TypeInfo pcie_interface_info = {
     .name          = INTERFACE_PCIE_DEVICE,
     .parent        = TYPE_INTERFACE,
@@ -2091,6 +2096,10 @@ static void pci_qdev_realize(DeviceState *qdev, Error **errp)
         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
     }
 
+    if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
+        pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
+    }
+
     pci_dev = do_pci_register_device(pci_dev,
                                      object_get_typename(OBJECT(qdev)),
                                      pci_dev->devfn, errp);
@@ -2817,6 +2826,7 @@ static void pci_register_types(void)
     type_register_static(&pci_bus_info);
     type_register_static(&pcie_bus_info);
     type_register_static(&conventional_pci_interface_info);
+    type_register_static(&cxl_interface_info);
     type_register_static(&pcie_interface_info);
     type_register_static(&pci_device_type_info);
 }
include/hw/pci/pci.h +8 -0
index 66db08462f..528cef341c 100644
@@ -195,6 +195,8 @@ enum {
     QEMU_PCIE_LNKSTA_DLLLA = (1 << QEMU_PCIE_LNKSTA_DLLLA_BITNR),
 #define QEMU_PCIE_EXTCAP_INIT_BITNR 9
     QEMU_PCIE_EXTCAP_INIT = (1 << QEMU_PCIE_EXTCAP_INIT_BITNR),
+#define QEMU_PCIE_CXL_BITNR 10
+    QEMU_PCIE_CAP_CXL = (1 << QEMU_PCIE_CXL_BITNR),
 };
 
 #define TYPE_PCI_DEVICE "pci-device"
@@ -202,6 +204,12 @@ typedef struct PCIDeviceClass PCIDeviceClass;
 DECLARE_OBJ_CHECKERS(PCIDevice, PCIDeviceClass,
                      PCI_DEVICE, TYPE_PCI_DEVICE)
 
+/*
+ * Implemented by devices that can be plugged on CXL buses. In the spec, this is
+ * actually a "CXL Component, but we name it device to match the PCI naming.
+ */
+#define INTERFACE_CXL_DEVICE "cxl-device"
+
 /* Implemented by devices that can be plugged on PCI Express buses */
 #define INTERFACE_PCIE_DEVICE "pci-express-device"
 
Signature · 1 line
2.30.0
#2 lore raw

RFC PATCH v3 2/31 hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5)

In reply to Ben Widawsky

A CXL 2.0 component is any entity in the CXL topology. All components have a analogous function in PCIe. Except for the CXL host bridge, all have a PCIe config space that is accessible via the common PCIe mechanisms. CXL components are enumerated via DVSEC fields in the extended PCIe header space. CXL components will minimally implement some subset of CXL.mem and CXL.cache registers defined in 8.2.5 of the CXL 2.0 specification. Two headers and a utility library are introduced to support the minimum functionality needed to enumerate components.

The cxl_pci header manages bits associated with PCI, specifically the DVSEC and related fields. The cxl_component.h variant has data structures and APIs that are useful for drivers implementing any of the CXL 2.0 components. The library takes care of making use of the DVSEC bits and the CXL.[mem|cache] registers. Per spec, the registers are little endian.

None of the mechanisms required to enumerate a CXL capable hostbridge are introduced at this point.

Note that the CXL.mem and CXL.cache registers used are always 4B wide. It's possible in the future that this constraint will not hold.

---
 MAINTAINERS                    |   6 +
 hw/Kconfig                     |   1 +
 hw/cxl/Kconfig                 |   3 +
 hw/cxl/cxl-component-utils.c   | 208 +++++++++++++++++++++++++++++++++
 hw/cxl/meson.build             |   3 +
 hw/meson.build                 |   1 +
 include/hw/cxl/cxl.h           |  17 +++
 include/hw/cxl/cxl_component.h | 187 +++++++++++++++++++++++++++++
 include/hw/cxl/cxl_pci.h       | 138 ++++++++++++++++++++++
 9 files changed, 564 insertions(+)
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_pci.h
MAINTAINERS +6 -0
index bcd88668bc..981dc92e25 100644
@@ -2234,6 +2234,12 @@ F: qapi/block*.json
 F: qapi/transaction.json
 T: git https://repo.or.cz/qemu/armbru.git block-next
 
+Compute Express Link
+M: Ben Widawsky <[email protected]>
+S: Supported
+F: hw/cxl/
+F: include/hw/cxl/
+
 Dirty Bitmaps
 M: Eric Blake <[email protected]>
 M: Vladimir Sementsov-Ogievskiy <[email protected]>
hw/Kconfig +1 -0
index 5ad3c6b5a4..c03650c5ed 100644
@@ -6,6 +6,7 @@ source audio/Kconfig
 source block/Kconfig
 source char/Kconfig
 source core/Kconfig
+source cxl/Kconfig
 source display/Kconfig
 source dma/Kconfig
 source gpio/Kconfig
hw/cxl/Kconfig new file +3 -0
new file mode 100644
index 0000000000..8e67519b16
@@ -0,0 +1,3 @@
+config CXL
+    bool
+    default y if PCI_EXPRESS
hw/cxl/cxl-component-utils.c new file +208 -0

208 lines not on this page · raw message

hw/cxl/meson.build new file +3 -0
new file mode 100644
index 0000000000..00c3876a0f
@@ -0,0 +1,3 @@
+softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
+  'cxl-component-utils.c',
+))
hw/meson.build +1 -0
index 010de7219c..3e440c341a 100644
@@ -6,6 +6,7 @@ subdir('block')
 subdir('char')
 subdir('core')
 subdir('cpu')
+subdir('cxl')
 subdir('display')
 subdir('dma')
 subdir('gpio')
include/hw/cxl/cxl.h new file +17 -0
new file mode 100644
index 0000000000..55f6cc30a5
@@ -0,0 +1,17 @@
+/*
+ * QEMU CXL Support
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_H
+#define CXL_H
+
+#include "cxl_pci.h"
+#include "cxl_component.h"
+
+#endif
+
include/hw/cxl/cxl_component.h new file +187 -0

187 lines not on this page · raw message

include/hw/cxl/cxl_pci.h new file +138 -0

138 lines not on this page · raw message

Signature · 1 line
2.30.0
#3 lore raw

Re: [RFC PATCH v3 02/31] hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5)

Quoted 16 lines · On Mon, 1 Feb 2021 16:59:19 -0800 Ben Widawsky <[email protected]> wrote:

A CXL 2.0 component is any entity in the CXL topology. All components have a analogous function in PCIe. Except for the CXL host bridge, all have a PCIe config space that is accessible via the common PCIe mechanisms. CXL components are enumerated via DVSEC fields in the extended PCIe header space. CXL components will minimally implement some subset of CXL.mem and CXL.cache registers defined in 8.2.5 of the CXL 2.0 specification. Two headers and a utility library are introduced to support the minimum functionality needed to enumerate components.

The cxl_pci header manages bits associated with PCI, specifically the DVSEC and related fields. The cxl_component.h variant has data structures and APIs that are useful for drivers implementing any of the CXL 2.0 components. The library takes care of making use of the DVSEC bits and the CXL.[mem|cache] registers. Per spec, the registers are little endian.

Quoted 7 lines

None of the mechanisms required to enumerate a CXL capable hostbridge are introduced at this point.

Note that the CXL.mem and CXL.cache registers used are always 4B wide. It's possible in the future that this constraint will not hold.

A few minor discrepancies from the spec, + naming suggestions.

Otherwise LGTM.

Quoted 149 lines
---
 MAINTAINERS                    |   6 +
 hw/Kconfig                     |   1 +
 hw/cxl/Kconfig                 |   3 +
 hw/cxl/cxl-component-utils.c   | 208 +++++++++++++++++++++++++++++++++
 hw/cxl/meson.build             |   3 +
 hw/meson.build                 |   1 +
 include/hw/cxl/cxl.h           |  17 +++
 include/hw/cxl/cxl_component.h | 187 +++++++++++++++++++++++++++++
 include/hw/cxl/cxl_pci.h       | 138 ++++++++++++++++++++++
 9 files changed, 564 insertions(+)
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_pci.h
MAINTAINERS +6 -0
index bcd88668bc..981dc92e25 100644
@@ -2234,6 +2234,12 @@ F: qapi/block*.json
 F: qapi/transaction.json
 T: git https://repo.or.cz/qemu/armbru.git block-next
 
+Compute Express Link
+M: Ben Widawsky <[email protected]>
+S: Supported
+F: hw/cxl/
+F: include/hw/cxl/
+
 Dirty Bitmaps
 M: Eric Blake <[email protected]>
 M: Vladimir Sementsov-Ogievskiy <[email protected]>
hw/Kconfig +1 -0
index 5ad3c6b5a4..c03650c5ed 100644
@@ -6,6 +6,7 @@ source audio/Kconfig
 source block/Kconfig
 source char/Kconfig
 source core/Kconfig
+source cxl/Kconfig
 source display/Kconfig
 source dma/Kconfig
 source gpio/Kconfig
hw/cxl/Kconfig new file +3 -0
new file mode 100644
index 0000000000..8e67519b16
@@ -0,0 +1,3 @@
+config CXL
+    bool
+    default y if PCI_EXPRESS
hw/cxl/cxl-component-utils.c new file +87 -0
new file mode 100644
index 0000000000..8d56ad5c7d
@@ -0,0 +1,208 @@
+/*
+ * CXL Utility library for components
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/pci/pci.h"
+#include "hw/cxl/cxl.h"
+
+static uint64_t cxl_cache_mem_read_reg(void *opaque, hwaddr offset,
+                                       unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->read) {
+        return cregs->special_ops->read(cxl_cstate, offset, size);
+    } else {
+        return cregs->cache_mem_registers[offset / 4];
+    }
+}
+
+static void cxl_cache_mem_write_reg(void *opaque, hwaddr offset, uint64_t value,
+                                    unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->write) {
+        cregs->special_ops->write(cxl_cstate, offset, value, size);
+    } else {
+        cregs->cache_mem_registers[offset / 4] = value;
+    }
+}
+
+/*
+ * 8.2.3
+ *   The access restrictions specified in Section 8.2.2 also apply to CXL 2.0
+ *   Component Registers.
+ *
+ * 8.2.2
+ *   • A 32 bit register shall be accessed as a 4 Bytes quantity. Partial
+ *   reads are not permitted.
+ *   • A 64 bit register shall be accessed as a 8 Bytes quantity. Partial
+ *   reads are not permitted.
+ *
+ * As of the spec defined today, only 4 byte registers exist.
+ */
+static const MemoryRegionOps cache_mem_ops = {
+    .read = cxl_cache_mem_read_reg,
+    .write = cxl_cache_mem_write_reg,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+
+void cxl_component_register_block_init(Object *obj,
+                                       CXLComponentState *cxl_cstate,
+                                       const char *type)
+{
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    memory_region_init(&cregs->component_registers, obj, type,
+                       CXL2_COMPONENT_BLOCK_SIZE);
+
+    /* io registers controls link which we don't care about in QEMU */
+    memory_region_init_io(&cregs->io, obj, NULL, cregs, ".io",
+                          CXL2_COMPONENT_IO_REGION_SIZE);
+    memory_region_init_io(&cregs->cache_mem, obj, &cache_mem_ops, cregs,
+                          ".cache_mem", CXL2_COMPONENT_CM_REGION_SIZE);
+

hw/cxl/cxl-component-utils.c line 97

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+    memory_region_add_subregion(&cregs->component_registers, 0, &cregs->io);
+    memory_region_add_subregion(&cregs->component_registers,
+                                CXL2_COMPONENT_IO_REGION_SIZE,
+                                &cregs->cache_mem);
+}
+
+static void ras_init_common(uint32_t *reg_state)
+{
+    reg_state[R_CXL_RAS_UNC_ERR_STATUS] = 0;
+    reg_state[R_CXL_RAS_UNC_ERR_MASK] = 0x1efff;  

This should be everything up to bit 11 then bits 14-16 I believe. 0x1cfff

hw/cxl/cxl-component-utils.c line 98

Quoted 1 line
hw/cxl/cxl-component-utils.c new file +1 -0
+    reg_state[R_CXL_RAS_UNC_ERR_SEVERITY] = 0x1efff;

0x1cfff as well

Quoted 25 lines
hw/cxl/cxl-component-utils.c new file +25 -0
+    reg_state[R_CXL_RAS_COR_ERR_STATUS] = 0;
+    reg_state[R_CXL_RAS_COR_ERR_MASK] = 0x3f;
+    reg_state[R_CXL_RAS_ERR_CAP_CTRL] = 0; /* CXL switches and devices must set */
+}
+
+static void hdm_init_common(uint32_t *reg_state)
+{
+    ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY, DECODER_COUNT, 0);
+    ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_GLOBAL_CONTROL, HDM_DECODER_ENABLE, 0);
+}
+
+void cxl_component_register_init_common(uint32_t *reg_state, enum reg_type type)
+{
+    int caps = 0;
+    switch (type) {
+    case CXL2_DOWNSTREAM_PORT:
+    case CXL2_DEVICE:
+        /* CAP, RAS, Link */
+        caps = 2;
+        break;
+    case CXL2_UPSTREAM_PORT:
+    case CXL2_TYPE3_DEVICE:
+    case CXL2_LOGICAL_DEVICE:
+        /* + HDM */
+        caps = 3;

hw/cxl/cxl-component-utils.c line 133

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+        break;
+    case CXL2_ROOT_PORT:
+        /* + Extended Security, + Snoop */
+        caps = 5;
+        break;
+    default:
+        abort();
+    }
+
+    memset(reg_state, 0, 0x1000);

Better to pass the size in so it's apparent where that came from?

Quoted 18 lines
hw/cxl/cxl-component-utils.c new file +18 -0
+
+    /* CXL Capability Header Register */
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ID, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ARRAY_SIZE, caps);
+
+
+#define init_cap_reg(reg, id, version)                                        \
+    _Static_assert(CXL_##reg##_REGISTERS_OFFSET != 0, "Invalid cap offset\n");\
+    do {                                                                      \
+        int which = R_CXL_##reg##_CAPABILITY_HEADER;                          \
+        reg_state[which] = FIELD_DP32(reg_state[which],                       \
+                                      CXL_##reg##_CAPABILITY_HEADER, ID, id); \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER,       \
+                       VERSION, version);                                     \
+        reg_state[which] =                                                    \

hw/cxl/cxl-component-utils.c line 161

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER, PTR,  \
+                       CXL_##reg##_REGISTERS_OFFSET);                         \
+    } while (0)
+
+    init_cap_reg(RAS, 2, 1);
+    ras_init_common(reg_state);
+
+    init_cap_reg(LINK, 4, 2);
+
+    if (caps < 3) {

It strikes me that this approach of basing it purely on number of caps is not particularly flexible or maintainable but I guess it will do for until we need something more sophisticated. (i.e. when they aren't a series of expanding inclusive sets of entries)

Quoted 170 lines
hw/cxl/cxl-component-utils.c new file +47 -0
+        return;
+    }
+
+    init_cap_reg(HDM, 5, 1);
+    hdm_init_common(reg_state);
+
+    if (caps < 5) {
+        return;
+    }
+
+    init_cap_reg(EXTSEC, 6, 1);
+    init_cap_reg(SNOOP, 8, 1);
+
+#undef init_cap_reg
+}
+
+/*
+ * Helper to creates a DVSEC header for a CXL entity. The caller is responsible
+ * for tracking the valid offset.
+ *
+ * This function will build the DVSEC header on behalf of the caller and then
+ * copy in the remaining data for the vendor specific bits.
+ */
+void cxl_component_create_dvsec(CXLComponentState *cxl, uint16_t length,
+                                uint16_t type, uint8_t rev, uint8_t *body)
+{
+    PCIDevice *pdev = cxl->pdev;
+    uint16_t offset = cxl->dvsec_offset;
+
+    assert(offset >= PCI_CFG_SPACE_SIZE &&
+           ((offset + length) < PCI_CFG_SPACE_EXP_SIZE));
+    assert((length & 0xf000) == 0);
+    assert((rev & ~0xf) == 0);
+
+    /* Create the DVSEC in the MCFG space */
+    pcie_add_capability(pdev, PCI_EXT_CAP_ID_DVSEC, 1, offset, length);
+    pci_set_long(pdev->config + offset + PCIE_DVSEC_HEADER1_OFFSET,
+                 (length << 20) | (rev << 16) | CXL_VENDOR_ID);
+    pci_set_word(pdev->config + offset + PCIE_DVSEC_ID_OFFSET, type);
+    memcpy(pdev->config + offset + sizeof(struct dvsec_header),
+           body + sizeof(struct dvsec_header),
+           length - sizeof(struct dvsec_header));
+
+    /* Update state for future DVSEC additions */
+    range_init_nofail(&cxl->dvsecs[type], cxl->dvsec_offset, length);
+    cxl->dvsec_offset += length;
+}
hw/cxl/meson.build new file +3 -0
new file mode 100644
index 0000000000..00c3876a0f
@@ -0,0 +1,3 @@
+softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
+  'cxl-component-utils.c',
+))
hw/meson.build +1 -0
index 010de7219c..3e440c341a 100644
@@ -6,6 +6,7 @@ subdir('block')
 subdir('char')
 subdir('core')
 subdir('cpu')
+subdir('cxl')
 subdir('display')
 subdir('dma')
 subdir('gpio')
include/hw/cxl/cxl.h new file +17 -0
new file mode 100644
index 0000000000..55f6cc30a5
@@ -0,0 +1,17 @@
+/*
+ * QEMU CXL Support
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_H
+#define CXL_H
+
+#include "cxl_pci.h"
+#include "cxl_component.h"
+
+#endif
+
include/hw/cxl/cxl_component.h new file +73 -0
new file mode 100644
index 0000000000..762feb54da
@@ -0,0 +1,187 @@
+/*
+ * QEMU CXL Component
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_COMPONENT_H
+#define CXL_COMPONENT_H
+
+/* CXL 2.0 - 8.2.4 */
+#define CXL2_COMPONENT_IO_REGION_SIZE 0x1000
+#define CXL2_COMPONENT_CM_REGION_SIZE 0x1000
+#define CXL2_COMPONENT_BLOCK_SIZE 0x10000
+
+#include "qemu/range.h"
+#include "qemu/typedefs.h"
+#include "hw/register.h"
+
+enum reg_type {
+    CXL2_DEVICE,
+    CXL2_TYPE3_DEVICE,
+    CXL2_LOGICAL_DEVICE,
+    CXL2_ROOT_PORT,
+    CXL2_UPSTREAM_PORT,
+    CXL2_DOWNSTREAM_PORT
+};
+
+/*
+ * Capability registers are defined at the top of the CXL.cache/mem region and
+ * are packed. For our purposes we will always define the caps in the same
+ * order.
+ * CXL 2.0 - 8.2.5 Table 142 for details.
+ */
+
+/* CXL 2.0 - 8.2.5.1 */
+REG32(CXL_CAPABILITY_HEADER, 0)
+    FIELD(CXL_CAPABILITY_HEADER, ID, 0, 16)
+    FIELD(CXL_CAPABILITY_HEADER, VERSION, 16, 4)
+    FIELD(CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 20, 4)
+    FIELD(CXL_CAPABILITY_HEADER, ARRAY_SIZE, 24, 8)
+
+#define CXLx_CAPABILITY_HEADER(type, offset)                  \
+    REG32(CXL_##type##_CAPABILITY_HEADER, offset)             \
+        FIELD(CXL_##type##_CAPABILITY_HEADER, ID, 0, 16)      \
+        FIELD(CXL_##type##_CAPABILITY_HEADER, VERSION, 16, 4) \
+        FIELD(CXL_##type##_CAPABILITY_HEADER, PTR, 20, 12)
+CXLx_CAPABILITY_HEADER(RAS, 0x4)
+CXLx_CAPABILITY_HEADER(LINK, 0x8)
+CXLx_CAPABILITY_HEADER(HDM, 0xc)
+CXLx_CAPABILITY_HEADER(EXTSEC, 0x10)
+CXLx_CAPABILITY_HEADER(SNOOP, 0x14)
+
+/*
+ * Capability structures contain the actual registers that the CXL component
+ * implements. Some of these are specific to certain types of components, but
+ * this implementation leaves enough space regardless.
+ */
+/* 8.2.5.9 - CXL RAS Capability Structure */
+#define CXL_RAS_REGISTERS_OFFSET 0x80 /* Give ample space for caps before this */
+#define CXL_RAS_REGISTERS_SIZE   0x58
+REG32(CXL_RAS_UNC_ERR_STATUS, CXL_RAS_REGISTERS_OFFSET)
+REG32(CXL_RAS_UNC_ERR_MASK, CXL_RAS_REGISTERS_OFFSET + 0x4)
+REG32(CXL_RAS_UNC_ERR_SEVERITY, CXL_RAS_REGISTERS_OFFSET + 0x8)
+REG32(CXL_RAS_COR_ERR_STATUS, CXL_RAS_REGISTERS_OFFSET + 0xc)
+REG32(CXL_RAS_COR_ERR_MASK, CXL_RAS_REGISTERS_OFFSET + 0x10)
+REG32(CXL_RAS_ERR_CAP_CTRL, CXL_RAS_REGISTERS_OFFSET + 0x14)
+/* Offset 0x18 - 0x58 reserved for RAS logs */
+
+/* 8.2.5.10 - CXL Security Capability Structure */
+#define CXL_SEC_REGISTERS_OFFSET (CXL_RAS_REGISTERS_OFFSET + CXL_RAS_REGISTERS_SIZE)

include/hw/cxl/cxl_component.h line 83

Quoted 10 lines
include/hw/cxl/cxl_component.h new file +10 -0
+#define CXL_SEC_REGISTERS_SIZE   0 /* We don't implement 1.1 downstream ports */
+
+/* 8.2.5.11 - CXL Link Capability Structure */
+#define CXL_LINK_REGISTERS_OFFSET (CXL_SEC_REGISTERS_OFFSET + CXL_SEC_REGISTERS_SIZE)
+#define CXL_LINK_REGISTERS_SIZE   0x38
+
+/* 8.2.5.12 - CXL HDM Decoder Capability Structure */
+#define HDM_DECODE_MAX 10 /* 8.2.5.12.1 */
+#define CXL_HDM_REGISTERS_OFFSET \
+    (CXL_LINK_REGISTERS_OFFSET + CXL_LINK_REGISTERS_SIZE) /* 8.2.5.12 */

The positioning of this isn't really defined by 8.2.5.12 that I can see. So I'd drop that trailing comment.

include/hw/cxl/cxl_component.h line 84

Quoted 1 line
include/hw/cxl/cxl_component.h new file +1 -0
+#define CXL_HDM_REGISTERS_SIZE (0x20 + HDM_DECODE_MAX * 10)

This doesn't look quite right. 0x10 + HDM_DECODE_MAX * 0x20 I think. Offset to decoder 0 + number of HDM decoders * size of each decode description.

include/hw/cxl/cxl_component.h line 87

Quoted 3 lines
include/hw/cxl/cxl_component.h new file +3 -0
+#define HDM_DECODER_INIT(n)                                                    \
+  REG32(CXL_HDM_DECODER##n##_BASE_LO,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x10)                          \

this might be easier to read if you define something like CXL_HDM_REGS_DECODER0_OFFSET CXL_HDM_REGISTERS_OFFSET + 0x10 then use that for the base.

Quoted 7 lines
include/hw/cxl/cxl_component.h new file +7 -0
+            FIELD(CXL_HDM_DECODER##n##_BASE_LO, L, 28, 4)                      \
+  REG32(CXL_HDM_DECODER##n##_BASE_HI,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x14)                          \
+  REG32(CXL_HDM_DECODER##n##_SIZE_LO,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x18)                          \
+  REG32(CXL_HDM_DECODER##n##_SIZE_HI,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x1C)                          \

include/hw/cxl/cxl_component.h line 104

Quoted 10 lines
include/hw/cxl/cxl_component.h new file +10 -0
+  REG32(CXL_HDM_DECODER##n##_CTRL,                                             \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x20)                          \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, IG, 0, 4)                         \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, IW, 4, 4)                         \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, LOCK_ON_COMMIT, 8, 1)             \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, COMMIT, 9, 1)                     \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, COMMITTED, 10, 1)                 \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, ERROR, 11, 1)                     \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, TYPE, 12, 1)                      \
+  REG32(CXL_HDM_DECODER##n##_TARGET_LIST_LO, 0x24)                             \

Offset should I think be per 'n'.

include/hw/cxl/cxl_component.h line 105

Quoted 1 line
include/hw/cxl/cxl_component.h new file +1 -0
+  REG32(CXL_HDM_DECODER##n##_TARGET_LIST_HI, 0x28)

Hmm. There is a hole here in the spec. Probably needs a reserved 4 bytes at CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x2c given next entry is at 0x30

Should be added to 8.2.5.12 table

include/hw/cxl/cxl_component.h line 113

Quoted 8 lines
include/hw/cxl/cxl_component.h new file +8 -0
+
+REG32(CXL_HDM_DECODER_CAPABILITY, CXL_HDM_REGISTERS_OFFSET)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, DECODER_COUNT, 0, 4)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, TARGET_COUNT, 4, 4)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, INTERLEAVE_256B, 8, 1)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, INTELEAVE_4K, 9, 1)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, POISON_ON_ERR_CAP, 10, 1)
+REG32(CXL_HDM_DECODER_GLOBAL_CONTROL, CXL_HDM_REGISTERS_OFFSET + 4) 

I'd be consistent on using hex for all offsets (even when it clearly makes no difference like here!)

include/hw/cxl/cxl_component.h line 126

Quoted 13 lines
include/hw/cxl/cxl_component.h new file +13 -0
+    FIELD(CXL_HDM_DECODER_GLOBAL_CONTROL, POISON_ON_ERR_EN, 0, 1)
+    FIELD(CXL_HDM_DECODER_GLOBAL_CONTROL, HDM_DECODER_ENABLE, 1, 1)
+
+HDM_DECODER_INIT(0);
+
+/* 8.2.5.13 - CXL Extended Security Capability Structure (Root complex only) */
+#define EXTSEC_ENTRY_MAX        256
+#define CXL_EXTSEC_REGISTERS_OFFSET (CXL_HDM_REGISTERS_OFFSET + CXL_HDM_REGISTERS_SIZE)
+#define CXL_EXTSEC_REGISTERS_SIZE   (8 * EXTSEC_ENTRY_MAX + 4)
+
+/* 8.2.5.14 - CXL IDE Capability Structure */
+#define CXL_IDE_REGISTERS_OFFSET (CXL_EXTSEC_REGISTERS_OFFSET + CXL_EXTSEC_REGISTERS_SIZE)
+#define CXL_IDE_REGISTERS_SIZE   0

0x20 (given we seem to be sizing other things we aren't using yet)

Quoted 111 lines
include/hw/cxl/cxl_component.h new file +61 -0
+
+/* 8.2.5.15 - CXL Snoop Filter Capability Structure */
+#define CXL_SNOOP_REGISTERS_OFFSET (CXL_IDE_REGISTERS_OFFSET + CXL_IDE_REGISTERS_SIZE)
+#define CXL_SNOOP_REGISTERS_SIZE   0x8
+
+_Static_assert((CXL_SNOOP_REGISTERS_OFFSET + CXL_SNOOP_REGISTERS_SIZE) < 0x1000,
+               "No space for registers");
+
+typedef struct component_registers {
+    /*
+     * Main memory region to be registered with QEMU core.
+     */
+    MemoryRegion component_registers;
+
+    /*
+     * 8.2.4 Table 141:
+     *   0x0000 - 0x0fff CXL.io registers
+     *   0x1000 - 0x1fff CXL.cache and CXL.mem
+     *   0x2000 - 0xdfff Implementation specific
+     *   0xe000 - 0xe3ff CXL ARB/MUX registers
+     *   0xe400 - 0xffff RSVD
+     */
+    uint32_t io_registers[CXL2_COMPONENT_IO_REGION_SIZE >> 2];
+    MemoryRegion io;
+
+    uint32_t cache_mem_registers[CXL2_COMPONENT_CM_REGION_SIZE >> 2];
+    MemoryRegion cache_mem;
+
+    MemoryRegion impl_specific;
+    MemoryRegion arb_mux;
+    MemoryRegion rsvd;
+
+    /* special_ops is used for any component that needs any specific handling */
+    MemoryRegionOps *special_ops;
+} ComponentRegisters;
+
+/*
+ * A CXL component represents all entities in a CXL hierarchy. This includes,
+ * host bridges, root ports, upstream/downstream switch ports, and devices
+ */
+typedef struct cxl_component {
+    ComponentRegisters crb;
+    union {
+        struct {
+            Range dvsecs[CXL20_MAX_DVSEC];
+            uint16_t dvsec_offset;
+            struct PCIDevice *pdev;
+        };
+    };
+} CXLComponentState;
+
+void cxl_component_register_block_init(Object *obj,
+                                       CXLComponentState *cxl_cstate,
+                                       const char *type);
+void cxl_component_register_init_common(uint32_t *reg_state,
+                                        enum reg_type type);
+
+void cxl_component_create_dvsec(CXLComponentState *cxl_cstate, uint16_t length,
+                                uint16_t type, uint8_t rev, uint8_t *body);
+
+#endif
include/hw/cxl/cxl_pci.h new file +44 -0
new file mode 100644
index 0000000000..a53c2e5ae7
@@ -0,0 +1,138 @@
+/*
+ * QEMU CXL PCI interfaces
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_PCI_H
+#define CXL_PCI_H
+
+#include "hw/pci/pci.h"
+#include "hw/pci/pcie.h"
+
+#define CXL_VENDOR_ID 0x1e98
+
+#define PCIE_DVSEC_HEADER1_OFFSET 0x4 /* Offset from start of extend cap */
+#define PCIE_DVSEC_ID_OFFSET 0x8
+
+#define PCIE_CXL_DEVICE_DVSEC_LENGTH 0x38
+#define PCIE_CXL1_DEVICE_DVSEC_REVID 0
+#define PCIE_CXL2_DEVICE_DVSEC_REVID 1
+
+#define EXTENSIONS_PORT_DVSEC_LENGTH 0x28
+#define EXTENSIONS_PORT_DVSEC_REVID 0
+
+#define GPF_PORT_DVSEC_LENGTH 0x10
+#define GPF_PORT_DVSEC_REVID  0
+
+#define PCIE_FLEXBUS_PORT_DVSEC_LENGTH_2_0 0x14
+#define PCIE_FLEXBUS_PORT_DVSEC_REVID_2_0  1
+
+#define REG_LOC_DVSEC_LENGTH 0x24
+#define REG_LOC_DVSEC_REVID  0
+
+enum {
+    PCIE_CXL_DEVICE_DVSEC      = 0,
+    NON_CXL_FUNCTION_MAP_DVSEC = 2,
+    EXTENSIONS_PORT_DVSEC      = 3,
+    GPF_PORT_DVSEC             = 4,
+    GPF_DEVICE_DVSEC           = 5,
+    PCIE_FLEXBUS_PORT_DVSEC    = 7,
+    REG_LOC_DVSEC              = 8,

include/hw/cxl/cxl_pci.h line 54

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+    MLD_DVSEC                  = 9,
+    CXL20_MAX_DVSEC
+};
+
+struct dvsec_header {
+    uint32_t cap_hdr;
+    uint32_t dv_hdr1;
+    uint16_t dv_hdr2;
+} __attribute__((__packed__));
+_Static_assert(sizeof(struct dvsec_header) == 10,

Be consistent on decimal or hex for these checks. I don't really care which but odd to mix and match.

Quoted 4 lines
include/hw/cxl/cxl_pci.h new file +4 -0
+               "dvsec header size incorrect");
+
+/*
+ * CXL 2.0 devices must implement certain DVSEC IDs, and can [optionally]

include/hw/cxl/cxl_pci.h line 68

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+ * implement others.
+ *
+ * CXL 2.0 Device: 0, [2], 5, 8
+ * CXL 2.0 RP: 3, 4, 7, 8
+ * CXL 2.0 Upstream Port: [2], 7, 8
+ * CXL 2.0 Downstream Port: 3, 4, 7, 8
+ */
+
+/* CXL 2.0 - 8.1.5 (ID 0003) */
+struct extensions_dvsec_port {

Probably want consistent naming (as can be done anyway) so cxl_dvsec_port_extensions maybe?

Quoted 7 lines
include/hw/cxl/cxl_pci.h new file +7 -0
+    struct dvsec_header hdr;
+    uint16_t status;
+    uint16_t control;
+    uint8_t alt_bus_base;
+    uint8_t alt_bus_limit;
+    uint16_t alt_memory_base;
+    uint16_t alt_memory_limit;

include/hw/cxl/cxl_pci.h line 85

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+    uint16_t alt_prefetch_base;
+    uint16_t alt_prefetch_limit;
+    uint32_t alt_prefetch_base_high;
+    uint32_t alt_prefetch_base_low;
+    uint32_t rcrb_base;
+    uint32_t rcrb_base_high;
+};
+_Static_assert(sizeof(struct extensions_dvsec_port) == 0x28,
+               "extensions dvsec port size incorrect");
+#define PORT_CONTROL_OVERRIDE_OFFSET 0xc

What's this one? Looks to just be the PORT_CONTROL_OFFSET though admittedly the spec does refer to this as OVERRIDE_OFFSET in one place.

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+#define PORT_CONTROL_UNMASK_SBR      1
+#define PORT_CONTROL_ALT_MEMID_EN    4
+
+/* CXL 2.0 - 8.1.6 GPF DVSEC (ID 0004) */
+struct dvsec_port_gpf {
+    struct dvsec_header hdr;
+    uint16_t rsvd;
+    uint16_t phase1_ctrl;
+    uint16_t phase2_ctrl;
+};

include/hw/cxl/cxl_pci.h line 105

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+_Static_assert(sizeof(struct dvsec_port_gpf) == 0x10,
+               "dvsec port GPF size incorrect");
+
+/* CXL 2.0 - 8.1.8/8.2.1.3 Flexbus DVSEC (ID 0007) */
+struct dvsec_port_flexbus {
+    struct dvsec_header hdr;
+    uint16_t cap;
+    uint16_t ctrl;
+    uint16_t status;
+    uint32_t rcvd_mod_ts_data;

Whilst it is wordy, I'd keep the full naming of that field. rcvd_mod_ts_data_phase_1;

include/hw/cxl/cxl_pci.h line 111

Quoted 6 lines
include/hw/cxl/cxl_pci.h new file +6 -0
+};
+_Static_assert(sizeof(struct dvsec_port_flexbus) == 0x14,
+               "dvsec port flexbus size incorrect");
+
+/* CXL 2.0 - 8.1.9 Register Locator DVSEC (ID 0008) */
+struct dvsec_register_locator {

I'd prefix these structures with cxl_ for all the normal namespacing collision reasons.
Same for #defines

Quoted 27 lines
include/hw/cxl/cxl_pci.h new file +27 -0
+    struct dvsec_header hdr;
+    uint16_t rsvd;
+    uint32_t reg0_base_lo;
+    uint32_t reg0_base_hi;
+    uint32_t reg1_base_lo;
+    uint32_t reg1_base_hi;
+    uint32_t reg2_base_lo;
+    uint32_t reg2_base_hi;
+};
+_Static_assert(sizeof(struct dvsec_register_locator) == 0x24,
+               "dvsec register locator size incorrect");
+
+/* BAR Equivalence Indicator */
+#define BEI_BAR_10H 0
+#define BEI_BAR_14H 1
+#define BEI_BAR_18H 2
+#define BEI_BAR_1cH 3
+#define BEI_BAR_20H 4
+#define BEI_BAR_24H 5
+
+/* Register Block Identifier */
+#define RBI_EMPTY          0
+#define RBI_COMPONENT_REG  (1 << 8)
+#define RBI_BAR_VIRT_ACL   (2 << 8)
+#define RBI_CXL_DEVICE_REG (3 << 8)
+
+#endif
#4 lore raw

Re: [RFC PATCH v3 02/31] hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5)

Quoted 193 lines · Jonathan Cameron
Quoted 16 lines · On Mon, 1 Feb 2021 16:59:19 -0800 Ben Widawsky <[email protected]> wrote:

A CXL 2.0 component is any entity in the CXL topology. All components have a analogous function in PCIe. Except for the CXL host bridge, all have a PCIe config space that is accessible via the common PCIe mechanisms. CXL components are enumerated via DVSEC fields in the extended PCIe header space. CXL components will minimally implement some subset of CXL.mem and CXL.cache registers defined in 8.2.5 of the CXL 2.0 specification. Two headers and a utility library are introduced to support the minimum functionality needed to enumerate components.

The cxl_pci header manages bits associated with PCI, specifically the DVSEC and related fields. The cxl_component.h variant has data structures and APIs that are useful for drivers implementing any of the CXL 2.0 components. The library takes care of making use of the DVSEC bits and the CXL.[mem|cache] registers. Per spec, the registers are little endian.

Quoted 7 lines

None of the mechanisms required to enumerate a CXL capable hostbridge are introduced at this point.

Note that the CXL.mem and CXL.cache registers used are always 4B wide. It's possible in the future that this constraint will not hold.

A few minor discrepancies from the spec, + naming suggestions.

Otherwise LGTM.

Quoted 149 lines
---
 MAINTAINERS                    |   6 +
 hw/Kconfig                     |   1 +
 hw/cxl/Kconfig                 |   3 +
 hw/cxl/cxl-component-utils.c   | 208 +++++++++++++++++++++++++++++++++
 hw/cxl/meson.build             |   3 +
 hw/meson.build                 |   1 +
 include/hw/cxl/cxl.h           |  17 +++
 include/hw/cxl/cxl_component.h | 187 +++++++++++++++++++++++++++++
 include/hw/cxl/cxl_pci.h       | 138 ++++++++++++++++++++++
 9 files changed, 564 insertions(+)
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_pci.h
MAINTAINERS +6 -0
index bcd88668bc..981dc92e25 100644
@@ -2234,6 +2234,12 @@ F: qapi/block*.json
 F: qapi/transaction.json
 T: git https://repo.or.cz/qemu/armbru.git block-next
 
+Compute Express Link
+M: Ben Widawsky <[email protected]>
+S: Supported
+F: hw/cxl/
+F: include/hw/cxl/
+
 Dirty Bitmaps
 M: Eric Blake <[email protected]>
 M: Vladimir Sementsov-Ogievskiy <[email protected]>
hw/Kconfig +1 -0
index 5ad3c6b5a4..c03650c5ed 100644
@@ -6,6 +6,7 @@ source audio/Kconfig
 source block/Kconfig
 source char/Kconfig
 source core/Kconfig
+source cxl/Kconfig
 source display/Kconfig
 source dma/Kconfig
 source gpio/Kconfig
hw/cxl/Kconfig new file +3 -0
new file mode 100644
index 0000000000..8e67519b16
@@ -0,0 +1,3 @@
+config CXL
+    bool
+    default y if PCI_EXPRESS
hw/cxl/cxl-component-utils.c new file +87 -0
new file mode 100644
index 0000000000..8d56ad5c7d
@@ -0,0 +1,208 @@
+/*
+ * CXL Utility library for components
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/pci/pci.h"
+#include "hw/cxl/cxl.h"
+
+static uint64_t cxl_cache_mem_read_reg(void *opaque, hwaddr offset,
+                                       unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->read) {
+        return cregs->special_ops->read(cxl_cstate, offset, size);
+    } else {
+        return cregs->cache_mem_registers[offset / 4];
+    }
+}
+
+static void cxl_cache_mem_write_reg(void *opaque, hwaddr offset, uint64_t value,
+                                    unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->write) {
+        cregs->special_ops->write(cxl_cstate, offset, value, size);
+    } else {
+        cregs->cache_mem_registers[offset / 4] = value;
+    }
+}
+
+/*
+ * 8.2.3
+ *   The access restrictions specified in Section 8.2.2 also apply to CXL 2.0
+ *   Component Registers.
+ *
+ * 8.2.2
+ *   • A 32 bit register shall be accessed as a 4 Bytes quantity. Partial
+ *   reads are not permitted.
+ *   • A 64 bit register shall be accessed as a 8 Bytes quantity. Partial
+ *   reads are not permitted.
+ *
+ * As of the spec defined today, only 4 byte registers exist.
+ */
+static const MemoryRegionOps cache_mem_ops = {
+    .read = cxl_cache_mem_read_reg,
+    .write = cxl_cache_mem_write_reg,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+
+void cxl_component_register_block_init(Object *obj,
+                                       CXLComponentState *cxl_cstate,
+                                       const char *type)
+{
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    memory_region_init(&cregs->component_registers, obj, type,
+                       CXL2_COMPONENT_BLOCK_SIZE);
+
+    /* io registers controls link which we don't care about in QEMU */
+    memory_region_init_io(&cregs->io, obj, NULL, cregs, ".io",
+                          CXL2_COMPONENT_IO_REGION_SIZE);
+    memory_region_init_io(&cregs->cache_mem, obj, &cache_mem_ops, cregs,
+                          ".cache_mem", CXL2_COMPONENT_CM_REGION_SIZE);
+

hw/cxl/cxl-component-utils.c line 97

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+    memory_region_add_subregion(&cregs->component_registers, 0, &cregs->io);
+    memory_region_add_subregion(&cregs->component_registers,
+                                CXL2_COMPONENT_IO_REGION_SIZE,
+                                &cregs->cache_mem);
+}
+
+static void ras_init_common(uint32_t *reg_state)
+{
+    reg_state[R_CXL_RAS_UNC_ERR_STATUS] = 0;
+    reg_state[R_CXL_RAS_UNC_ERR_MASK] = 0x1efff;  

This should be everything up to bit 11 then bits 14-16 I believe. 0x1cfff

Quoted 3 lines

hw/cxl/cxl-component-utils.c line 98

Quoted 1 line
hw/cxl/cxl-component-utils.c new file +1 -0
+    reg_state[R_CXL_RAS_UNC_ERR_SEVERITY] = 0x1efff;

0x1cfff as well

Thanks. This changed since the spec version I originally wrote to...

Quoted 29 lines
Quoted 29 lines
hw/cxl/cxl-component-utils.c new file +29 -0
+    reg_state[R_CXL_RAS_COR_ERR_STATUS] = 0;
+    reg_state[R_CXL_RAS_COR_ERR_MASK] = 0x3f;
+    reg_state[R_CXL_RAS_ERR_CAP_CTRL] = 0; /* CXL switches and devices must set */
+}
+
+static void hdm_init_common(uint32_t *reg_state)
+{
+    ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_CAPABILITY, DECODER_COUNT, 0);
+    ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_GLOBAL_CONTROL, HDM_DECODER_ENABLE, 0);
+}
+
+void cxl_component_register_init_common(uint32_t *reg_state, enum reg_type type)
+{
+    int caps = 0;
+    switch (type) {
+    case CXL2_DOWNSTREAM_PORT:
+    case CXL2_DEVICE:
+        /* CAP, RAS, Link */
+        caps = 2;
+        break;
+    case CXL2_UPSTREAM_PORT:
+    case CXL2_TYPE3_DEVICE:
+    case CXL2_LOGICAL_DEVICE:
+        /* + HDM */
+        caps = 3;
+        break;
+    case CXL2_ROOT_PORT:
+        /* + Extended Security, + Snoop */
+        caps = 5;
Quoted 8 lines

hw/cxl/cxl-component-utils.c line 133

Quoted 6 lines
hw/cxl/cxl-component-utils.c new file +6 -0
+        break;
+    default:
+        abort();
+    }
+
+    memset(reg_state, 0, 0x1000);

Better to pass the size in so it's apparent where that came from?

I think #define for now is a better fit. As you allude to below, I suspect this gets messy with future versions of the spec.

Quoted 26 lines
Quoted 26 lines
hw/cxl/cxl-component-utils.c new file +25 -0
+
+    /* CXL Capability Header Register */
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ID, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ARRAY_SIZE, caps);
+
+
+#define init_cap_reg(reg, id, version)                                        \
+    _Static_assert(CXL_##reg##_REGISTERS_OFFSET != 0, "Invalid cap offset\n");\
+    do {                                                                      \
+        int which = R_CXL_##reg##_CAPABILITY_HEADER;                          \
+        reg_state[which] = FIELD_DP32(reg_state[which],                       \
+                                      CXL_##reg##_CAPABILITY_HEADER, ID, id); \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER,       \
+                       VERSION, version);                                     \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER, PTR,  \
+                       CXL_##reg##_REGISTERS_OFFSET);                         \
+    } while (0)
+
+    init_cap_reg(RAS, 2, 1);
+    ras_init_common(reg_state);
+
Quoted 9 lines

hw/cxl/cxl-component-utils.c line 161

Quoted 3 lines
hw/cxl/cxl-component-utils.c new file +3 -0
+    init_cap_reg(LINK, 4, 2);
+
+    if (caps < 3) {

It strikes me that this approach of basing it purely on number of caps is not particularly flexible or maintainable but I guess it will do for until we need something more sophisticated. (i.e. when they aren't a series of expanding inclusive sets of entries)

SUPER inflexible. I am not yet convinced we need flexible here and I liked simple. It's totally fine to change this when it's needed.

Quoted 184 lines
Quoted 170 lines
hw/cxl/cxl-component-utils.c new file +47 -0
+        return;
+    }
+
+    init_cap_reg(HDM, 5, 1);
+    hdm_init_common(reg_state);
+
+    if (caps < 5) {
+        return;
+    }
+
+    init_cap_reg(EXTSEC, 6, 1);
+    init_cap_reg(SNOOP, 8, 1);
+
+#undef init_cap_reg
+}
+
+/*
+ * Helper to creates a DVSEC header for a CXL entity. The caller is responsible
+ * for tracking the valid offset.
+ *
+ * This function will build the DVSEC header on behalf of the caller and then
+ * copy in the remaining data for the vendor specific bits.
+ */
+void cxl_component_create_dvsec(CXLComponentState *cxl, uint16_t length,
+                                uint16_t type, uint8_t rev, uint8_t *body)
+{
+    PCIDevice *pdev = cxl->pdev;
+    uint16_t offset = cxl->dvsec_offset;
+
+    assert(offset >= PCI_CFG_SPACE_SIZE &&
+           ((offset + length) < PCI_CFG_SPACE_EXP_SIZE));
+    assert((length & 0xf000) == 0);
+    assert((rev & ~0xf) == 0);
+
+    /* Create the DVSEC in the MCFG space */
+    pcie_add_capability(pdev, PCI_EXT_CAP_ID_DVSEC, 1, offset, length);
+    pci_set_long(pdev->config + offset + PCIE_DVSEC_HEADER1_OFFSET,
+                 (length << 20) | (rev << 16) | CXL_VENDOR_ID);
+    pci_set_word(pdev->config + offset + PCIE_DVSEC_ID_OFFSET, type);
+    memcpy(pdev->config + offset + sizeof(struct dvsec_header),
+           body + sizeof(struct dvsec_header),
+           length - sizeof(struct dvsec_header));
+
+    /* Update state for future DVSEC additions */
+    range_init_nofail(&cxl->dvsecs[type], cxl->dvsec_offset, length);
+    cxl->dvsec_offset += length;
+}
hw/cxl/meson.build new file +3 -0
new file mode 100644
index 0000000000..00c3876a0f
@@ -0,0 +1,3 @@
+softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
+  'cxl-component-utils.c',
+))
hw/meson.build +1 -0
index 010de7219c..3e440c341a 100644
@@ -6,6 +6,7 @@ subdir('block')
 subdir('char')
 subdir('core')
 subdir('cpu')
+subdir('cxl')
 subdir('display')
 subdir('dma')
 subdir('gpio')
include/hw/cxl/cxl.h new file +17 -0
new file mode 100644
index 0000000000..55f6cc30a5
@@ -0,0 +1,17 @@
+/*
+ * QEMU CXL Support
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_H
+#define CXL_H
+
+#include "cxl_pci.h"
+#include "cxl_component.h"
+
+#endif
+
include/hw/cxl/cxl_component.h new file +73 -0
new file mode 100644
index 0000000000..762feb54da
@@ -0,0 +1,187 @@
+/*
+ * QEMU CXL Component
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_COMPONENT_H
+#define CXL_COMPONENT_H
+
+/* CXL 2.0 - 8.2.4 */
+#define CXL2_COMPONENT_IO_REGION_SIZE 0x1000
+#define CXL2_COMPONENT_CM_REGION_SIZE 0x1000
+#define CXL2_COMPONENT_BLOCK_SIZE 0x10000
+
+#include "qemu/range.h"
+#include "qemu/typedefs.h"
+#include "hw/register.h"
+
+enum reg_type {
+    CXL2_DEVICE,
+    CXL2_TYPE3_DEVICE,
+    CXL2_LOGICAL_DEVICE,
+    CXL2_ROOT_PORT,
+    CXL2_UPSTREAM_PORT,
+    CXL2_DOWNSTREAM_PORT
+};
+
+/*
+ * Capability registers are defined at the top of the CXL.cache/mem region and
+ * are packed. For our purposes we will always define the caps in the same
+ * order.
+ * CXL 2.0 - 8.2.5 Table 142 for details.
+ */
+
+/* CXL 2.0 - 8.2.5.1 */
+REG32(CXL_CAPABILITY_HEADER, 0)
+    FIELD(CXL_CAPABILITY_HEADER, ID, 0, 16)
+    FIELD(CXL_CAPABILITY_HEADER, VERSION, 16, 4)
+    FIELD(CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 20, 4)
+    FIELD(CXL_CAPABILITY_HEADER, ARRAY_SIZE, 24, 8)
+
+#define CXLx_CAPABILITY_HEADER(type, offset)                  \
+    REG32(CXL_##type##_CAPABILITY_HEADER, offset)             \
+        FIELD(CXL_##type##_CAPABILITY_HEADER, ID, 0, 16)      \
+        FIELD(CXL_##type##_CAPABILITY_HEADER, VERSION, 16, 4) \
+        FIELD(CXL_##type##_CAPABILITY_HEADER, PTR, 20, 12)
+CXLx_CAPABILITY_HEADER(RAS, 0x4)
+CXLx_CAPABILITY_HEADER(LINK, 0x8)
+CXLx_CAPABILITY_HEADER(HDM, 0xc)
+CXLx_CAPABILITY_HEADER(EXTSEC, 0x10)
+CXLx_CAPABILITY_HEADER(SNOOP, 0x14)
+
+/*
+ * Capability structures contain the actual registers that the CXL component
+ * implements. Some of these are specific to certain types of components, but
+ * this implementation leaves enough space regardless.
+ */
+/* 8.2.5.9 - CXL RAS Capability Structure */
+#define CXL_RAS_REGISTERS_OFFSET 0x80 /* Give ample space for caps before this */
+#define CXL_RAS_REGISTERS_SIZE   0x58
+REG32(CXL_RAS_UNC_ERR_STATUS, CXL_RAS_REGISTERS_OFFSET)
+REG32(CXL_RAS_UNC_ERR_MASK, CXL_RAS_REGISTERS_OFFSET + 0x4)
+REG32(CXL_RAS_UNC_ERR_SEVERITY, CXL_RAS_REGISTERS_OFFSET + 0x8)
+REG32(CXL_RAS_COR_ERR_STATUS, CXL_RAS_REGISTERS_OFFSET + 0xc)
+REG32(CXL_RAS_COR_ERR_MASK, CXL_RAS_REGISTERS_OFFSET + 0x10)
+REG32(CXL_RAS_ERR_CAP_CTRL, CXL_RAS_REGISTERS_OFFSET + 0x14)
+/* Offset 0x18 - 0x58 reserved for RAS logs */
+
+/* 8.2.5.10 - CXL Security Capability Structure */
+#define CXL_SEC_REGISTERS_OFFSET (CXL_RAS_REGISTERS_OFFSET + CXL_RAS_REGISTERS_SIZE)

include/hw/cxl/cxl_component.h line 83

Quoted 10 lines
include/hw/cxl/cxl_component.h new file +10 -0
+#define CXL_SEC_REGISTERS_SIZE   0 /* We don't implement 1.1 downstream ports */
+
+/* 8.2.5.11 - CXL Link Capability Structure */
+#define CXL_LINK_REGISTERS_OFFSET (CXL_SEC_REGISTERS_OFFSET + CXL_SEC_REGISTERS_SIZE)
+#define CXL_LINK_REGISTERS_SIZE   0x38
+
+/* 8.2.5.12 - CXL HDM Decoder Capability Structure */
+#define HDM_DECODE_MAX 10 /* 8.2.5.12.1 */
+#define CXL_HDM_REGISTERS_OFFSET \
+    (CXL_LINK_REGISTERS_OFFSET + CXL_LINK_REGISTERS_SIZE) /* 8.2.5.12 */

The positioning of this isn't really defined by 8.2.5.12 that I can see. So I'd drop that trailing comment.

Quoted 4 lines

include/hw/cxl/cxl_component.h line 84

Quoted 1 line
include/hw/cxl/cxl_component.h new file +1 -0
+#define CXL_HDM_REGISTERS_SIZE (0x20 + HDM_DECODE_MAX * 10)

This doesn't look quite right. 0x10 + HDM_DECODE_MAX * 0x20 I think. Offset to decoder 0 + number of HDM decoders * size of each decode description.

Not sure how I botched that. It's literally in the spec as:
20h *n+ 10h

Thanks.

Quoted 35 lines

include/hw/cxl/cxl_component.h line 87

Quoted 5 lines
include/hw/cxl/cxl_component.h new file +3 -0
+#define HDM_DECODER_INIT(n)                                                    \
+  REG32(CXL_HDM_DECODER##n##_BASE_LO,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x10)                          \

this might be easier to read if you define something like CXL_HDM_REGS_DECODER0_OFFSET CXL_HDM_REGISTERS_OFFSET + 0x10 then use that for the base.

Quoted 7 lines
include/hw/cxl/cxl_component.h new file +7 -0
+            FIELD(CXL_HDM_DECODER##n##_BASE_LO, L, 28, 4)                      \
+  REG32(CXL_HDM_DECODER##n##_BASE_HI,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x14)                          \
+  REG32(CXL_HDM_DECODER##n##_SIZE_LO,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x18)                          \
+  REG32(CXL_HDM_DECODER##n##_SIZE_HI,                                          \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x1C)                          \

include/hw/cxl/cxl_component.h line 104

Quoted 10 lines
include/hw/cxl/cxl_component.h new file +10 -0
+  REG32(CXL_HDM_DECODER##n##_CTRL,                                             \
+        CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x20)                          \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, IG, 0, 4)                         \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, IW, 4, 4)                         \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, LOCK_ON_COMMIT, 8, 1)             \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, COMMIT, 9, 1)                     \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, COMMITTED, 10, 1)                 \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, ERROR, 11, 1)                     \
+            FIELD(CXL_HDM_DECODER##n##_CTRL, TYPE, 12, 1)                      \
+  REG32(CXL_HDM_DECODER##n##_TARGET_LIST_LO, 0x24)                             \

Offset should I think be per 'n'.

include/hw/cxl/cxl_component.h line 105

Quoted 1 line
include/hw/cxl/cxl_component.h new file +1 -0
+  REG32(CXL_HDM_DECODER##n##_TARGET_LIST_HI, 0x28)

Hmm. There is a hole here in the spec. Probably needs a reserved 4 bytes at CXL_HDM_REGISTERS_OFFSET + (0x20 * n) + 0x2c given next entry is at 0x30

Should be added to 8.2.5.12 table

Quoted 12 lines

include/hw/cxl/cxl_component.h line 113

Quoted 8 lines
include/hw/cxl/cxl_component.h new file +8 -0
+
+REG32(CXL_HDM_DECODER_CAPABILITY, CXL_HDM_REGISTERS_OFFSET)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, DECODER_COUNT, 0, 4)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, TARGET_COUNT, 4, 4)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, INTERLEAVE_256B, 8, 1)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, INTELEAVE_4K, 9, 1)
+    FIELD(CXL_HDM_DECODER_CAPABILITY, POISON_ON_ERR_CAP, 10, 1)
+REG32(CXL_HDM_DECODER_GLOBAL_CONTROL, CXL_HDM_REGISTERS_OFFSET + 4) 

I'd be consistent on using hex for all offsets (even when it clearly makes no difference like here!)

I'm going to punt on this for now to avoid thrash. But I'm fine to do this or take a patch from someone else who does it.

Quoted 141 lines

include/hw/cxl/cxl_component.h line 126

Quoted 13 lines
include/hw/cxl/cxl_component.h new file +13 -0
+    FIELD(CXL_HDM_DECODER_GLOBAL_CONTROL, POISON_ON_ERR_EN, 0, 1)
+    FIELD(CXL_HDM_DECODER_GLOBAL_CONTROL, HDM_DECODER_ENABLE, 1, 1)
+
+HDM_DECODER_INIT(0);
+
+/* 8.2.5.13 - CXL Extended Security Capability Structure (Root complex only) */
+#define EXTSEC_ENTRY_MAX        256
+#define CXL_EXTSEC_REGISTERS_OFFSET (CXL_HDM_REGISTERS_OFFSET + CXL_HDM_REGISTERS_SIZE)
+#define CXL_EXTSEC_REGISTERS_SIZE   (8 * EXTSEC_ENTRY_MAX + 4)
+
+/* 8.2.5.14 - CXL IDE Capability Structure */
+#define CXL_IDE_REGISTERS_OFFSET (CXL_EXTSEC_REGISTERS_OFFSET + CXL_EXTSEC_REGISTERS_SIZE)
+#define CXL_IDE_REGISTERS_SIZE   0

0x20 (given we seem to be sizing other things we aren't using yet)

Quoted 111 lines
include/hw/cxl/cxl_component.h new file +61 -0
+
+/* 8.2.5.15 - CXL Snoop Filter Capability Structure */
+#define CXL_SNOOP_REGISTERS_OFFSET (CXL_IDE_REGISTERS_OFFSET + CXL_IDE_REGISTERS_SIZE)
+#define CXL_SNOOP_REGISTERS_SIZE   0x8
+
+_Static_assert((CXL_SNOOP_REGISTERS_OFFSET + CXL_SNOOP_REGISTERS_SIZE) < 0x1000,
+               "No space for registers");
+
+typedef struct component_registers {
+    /*
+     * Main memory region to be registered with QEMU core.
+     */
+    MemoryRegion component_registers;
+
+    /*
+     * 8.2.4 Table 141:
+     *   0x0000 - 0x0fff CXL.io registers
+     *   0x1000 - 0x1fff CXL.cache and CXL.mem
+     *   0x2000 - 0xdfff Implementation specific
+     *   0xe000 - 0xe3ff CXL ARB/MUX registers
+     *   0xe400 - 0xffff RSVD
+     */
+    uint32_t io_registers[CXL2_COMPONENT_IO_REGION_SIZE >> 2];
+    MemoryRegion io;
+
+    uint32_t cache_mem_registers[CXL2_COMPONENT_CM_REGION_SIZE >> 2];
+    MemoryRegion cache_mem;
+
+    MemoryRegion impl_specific;
+    MemoryRegion arb_mux;
+    MemoryRegion rsvd;
+
+    /* special_ops is used for any component that needs any specific handling */
+    MemoryRegionOps *special_ops;
+} ComponentRegisters;
+
+/*
+ * A CXL component represents all entities in a CXL hierarchy. This includes,
+ * host bridges, root ports, upstream/downstream switch ports, and devices
+ */
+typedef struct cxl_component {
+    ComponentRegisters crb;
+    union {
+        struct {
+            Range dvsecs[CXL20_MAX_DVSEC];
+            uint16_t dvsec_offset;
+            struct PCIDevice *pdev;
+        };
+    };
+} CXLComponentState;
+
+void cxl_component_register_block_init(Object *obj,
+                                       CXLComponentState *cxl_cstate,
+                                       const char *type);
+void cxl_component_register_init_common(uint32_t *reg_state,
+                                        enum reg_type type);
+
+void cxl_component_create_dvsec(CXLComponentState *cxl_cstate, uint16_t length,
+                                uint16_t type, uint8_t rev, uint8_t *body);
+
+#endif
include/hw/cxl/cxl_pci.h new file +44 -0
new file mode 100644
index 0000000000..a53c2e5ae7
@@ -0,0 +1,138 @@
+/*
+ * QEMU CXL PCI interfaces
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_PCI_H
+#define CXL_PCI_H
+
+#include "hw/pci/pci.h"
+#include "hw/pci/pcie.h"
+
+#define CXL_VENDOR_ID 0x1e98
+
+#define PCIE_DVSEC_HEADER1_OFFSET 0x4 /* Offset from start of extend cap */
+#define PCIE_DVSEC_ID_OFFSET 0x8
+
+#define PCIE_CXL_DEVICE_DVSEC_LENGTH 0x38
+#define PCIE_CXL1_DEVICE_DVSEC_REVID 0
+#define PCIE_CXL2_DEVICE_DVSEC_REVID 1
+
+#define EXTENSIONS_PORT_DVSEC_LENGTH 0x28
+#define EXTENSIONS_PORT_DVSEC_REVID 0
+
+#define GPF_PORT_DVSEC_LENGTH 0x10
+#define GPF_PORT_DVSEC_REVID  0
+
+#define PCIE_FLEXBUS_PORT_DVSEC_LENGTH_2_0 0x14
+#define PCIE_FLEXBUS_PORT_DVSEC_REVID_2_0  1
+
+#define REG_LOC_DVSEC_LENGTH 0x24
+#define REG_LOC_DVSEC_REVID  0
+
+enum {
+    PCIE_CXL_DEVICE_DVSEC      = 0,
+    NON_CXL_FUNCTION_MAP_DVSEC = 2,
+    EXTENSIONS_PORT_DVSEC      = 3,
+    GPF_PORT_DVSEC             = 4,
+    GPF_DEVICE_DVSEC           = 5,
+    PCIE_FLEXBUS_PORT_DVSEC    = 7,
+    REG_LOC_DVSEC              = 8,

include/hw/cxl/cxl_pci.h line 54

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+    MLD_DVSEC                  = 9,
+    CXL20_MAX_DVSEC
+};
+
+struct dvsec_header {
+    uint32_t cap_hdr;
+    uint32_t dv_hdr1;
+    uint16_t dv_hdr2;
+} __attribute__((__packed__));
+_Static_assert(sizeof(struct dvsec_header) == 10,

Be consistent on decimal or hex for these checks. I don't really care which but odd to mix and match.

...

Quoted 30 lines
Quoted 4 lines
include/hw/cxl/cxl_pci.h new file +4 -0
+               "dvsec header size incorrect");
+
+/*
+ * CXL 2.0 devices must implement certain DVSEC IDs, and can [optionally]

include/hw/cxl/cxl_pci.h line 68

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+ * implement others.
+ *
+ * CXL 2.0 Device: 0, [2], 5, 8
+ * CXL 2.0 RP: 3, 4, 7, 8
+ * CXL 2.0 Upstream Port: [2], 7, 8
+ * CXL 2.0 Downstream Port: 3, 4, 7, 8
+ */
+
+/* CXL 2.0 - 8.1.5 (ID 0003) */
+struct extensions_dvsec_port {

Probably want consistent naming (as can be done anyway) so cxl_dvsec_port_extensions maybe?

Quoted 13 lines
include/hw/cxl/cxl_pci.h new file +13 -0
+    struct dvsec_header hdr;
+    uint16_t status;
+    uint16_t control;
+    uint8_t alt_bus_base;
+    uint8_t alt_bus_limit;
+    uint16_t alt_memory_base;
+    uint16_t alt_memory_limit;
+    uint16_t alt_prefetch_base;
+    uint16_t alt_prefetch_limit;
+    uint32_t alt_prefetch_base_high;
+    uint32_t alt_prefetch_base_low;
+    uint32_t rcrb_base;
+    uint32_t rcrb_base_high;
Quoted 9 lines

include/hw/cxl/cxl_pci.h line 85

Quoted 4 lines
include/hw/cxl/cxl_pci.h new file +4 -0
+};
+_Static_assert(sizeof(struct extensions_dvsec_port) == 0x28,
+               "extensions dvsec port size incorrect");
+#define PORT_CONTROL_OVERRIDE_OFFSET 0xc

What's this one? Looks to just be the PORT_CONTROL_OFFSET though admittedly the spec does refer to this as OVERRIDE_OFFSET in one place.

This has changed since earlier versions of the spec. I think 9.12.3 needs update...

Quoted 24 lines
Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+#define PORT_CONTROL_UNMASK_SBR      1
+#define PORT_CONTROL_ALT_MEMID_EN    4
+
+/* CXL 2.0 - 8.1.6 GPF DVSEC (ID 0004) */
+struct dvsec_port_gpf {
+    struct dvsec_header hdr;
+    uint16_t rsvd;
+    uint16_t phase1_ctrl;
+    uint16_t phase2_ctrl;
+};

include/hw/cxl/cxl_pci.h line 105

Quoted 10 lines
include/hw/cxl/cxl_pci.h new file +10 -0
+_Static_assert(sizeof(struct dvsec_port_gpf) == 0x10,
+               "dvsec port GPF size incorrect");
+
+/* CXL 2.0 - 8.1.8/8.2.1.3 Flexbus DVSEC (ID 0007) */
+struct dvsec_port_flexbus {
+    struct dvsec_header hdr;
+    uint16_t cap;
+    uint16_t ctrl;
+    uint16_t status;
+    uint32_t rcvd_mod_ts_data;

Whilst it is wordy, I'd keep the full naming of that field. rcvd_mod_ts_data_phase_1;

Quoted 11 lines

include/hw/cxl/cxl_pci.h line 111

Quoted 6 lines
include/hw/cxl/cxl_pci.h new file +6 -0
+};
+_Static_assert(sizeof(struct dvsec_port_flexbus) == 0x14,
+               "dvsec port flexbus size incorrect");
+
+/* CXL 2.0 - 8.1.9 Register Locator DVSEC (ID 0008) */
+struct dvsec_register_locator {

I'd prefix these structures with cxl_ for all the normal namespacing collision reasons.
Same for #defines

Agreed

Quoted 30 lines
Quoted 28 lines
include/hw/cxl/cxl_pci.h new file +27 -0
+    struct dvsec_header hdr;
+    uint16_t rsvd;
+    uint32_t reg0_base_lo;
+    uint32_t reg0_base_hi;
+    uint32_t reg1_base_lo;
+    uint32_t reg1_base_hi;
+    uint32_t reg2_base_lo;
+    uint32_t reg2_base_hi;
+};
+_Static_assert(sizeof(struct dvsec_register_locator) == 0x24,
+               "dvsec register locator size incorrect");
+
+/* BAR Equivalence Indicator */
+#define BEI_BAR_10H 0
+#define BEI_BAR_14H 1
+#define BEI_BAR_18H 2
+#define BEI_BAR_1cH 3
+#define BEI_BAR_20H 4
+#define BEI_BAR_24H 5
+
+/* Register Block Identifier */
+#define RBI_EMPTY          0
+#define RBI_COMPONENT_REG  (1 << 8)
+#define RBI_BAR_VIRT_ACL   (2 << 8)
+#define RBI_CXL_DEVICE_REG (3 << 8)
+
+#endif
#5 lore raw

Re: [RFC PATCH v3 02/31] hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5)

In reply to Ben Widawsky
Quoted 16 lines · On Mon, 1 Feb 2021 16:59:19 -0800 Ben Widawsky <[email protected]> wrote:

A CXL 2.0 component is any entity in the CXL topology. All components have a analogous function in PCIe. Except for the CXL host bridge, all have a PCIe config space that is accessible via the common PCIe mechanisms. CXL components are enumerated via DVSEC fields in the extended PCIe header space. CXL components will minimally implement some subset of CXL.mem and CXL.cache registers defined in 8.2.5 of the CXL 2.0 specification. Two headers and a utility library are introduced to support the minimum functionality needed to enumerate components.

The cxl_pci header manages bits associated with PCI, specifically the DVSEC and related fields. The cxl_component.h variant has data structures and APIs that are useful for drivers implementing any of the CXL 2.0 components. The library takes care of making use of the DVSEC bits and the CXL.[mem|cache] registers. Per spec, the registers are little endian.

Quoted 7 lines

None of the mechanisms required to enumerate a CXL capable hostbridge are introduced at this point.

Note that the CXL.mem and CXL.cache registers used are always 4B wide. It's possible in the future that this constraint will not hold.

A few additions to previous comments.

Quoted 72 lines
---
 MAINTAINERS                    |   6 +
 hw/Kconfig                     |   1 +
 hw/cxl/Kconfig                 |   3 +
 hw/cxl/cxl-component-utils.c   | 208 +++++++++++++++++++++++++++++++++
 hw/cxl/meson.build             |   3 +
 hw/meson.build                 |   1 +
 include/hw/cxl/cxl.h           |  17 +++
 include/hw/cxl/cxl_component.h | 187 +++++++++++++++++++++++++++++
 include/hw/cxl/cxl_pci.h       | 138 ++++++++++++++++++++++
 9 files changed, 564 insertions(+)
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_pci.h
hw/cxl/cxl-component-utils.c new file +46 -0
new file mode 100644
index 0000000000..8d56ad5c7d
@@ -0,0 +1,208 @@
+/*
+ * CXL Utility library for components
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/pci/pci.h"
+#include "hw/cxl/cxl.h"
+
+static uint64_t cxl_cache_mem_read_reg(void *opaque, hwaddr offset,
+                                       unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->read) {
+        return cregs->special_ops->read(cxl_cstate, offset, size);
+    } else {
+        return cregs->cache_mem_registers[offset / 4];
+    }
+}
+
+static void cxl_cache_mem_write_reg(void *opaque, hwaddr offset, uint64_t value,
+                                    unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->write) {
+        cregs->special_ops->write(cxl_cstate, offset, value, size);
+    } else {
+        cregs->cache_mem_registers[offset / 4] = value;
+    }
+}
+
+/*
+ * 8.2.3

hw/cxl/cxl-component-utils.c line 56

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+ *   The access restrictions specified in Section 8.2.2 also apply to CXL 2.0
+ *   Component Registers.
+ *
+ * 8.2.2
+ *   • A 32 bit register shall be accessed as a 4 Bytes quantity. Partial
+ *   reads are not permitted.
+ *   • A 64 bit register shall be accessed as a 8 Bytes quantity. Partial
+ *   reads are not permitted.
+ *
+ * As of the spec defined today, only 4 byte registers exist.

The exciting exception to this is the RAS header log which is defined as 512 bits. Will seek clarification but I think the spec should probably say that is a set of 32 bit registers.

A bunch of the other elements that we probably want to block in plausible values for also seem to use 64 bit registers.

Quoted 16 lines
hw/cxl/cxl-component-utils.c new file +16 -0
+ */
+static const MemoryRegionOps cache_mem_ops = {
+    .read = cxl_cache_mem_read_reg,
+    .write = cxl_cache_mem_write_reg,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+

..

Quoted 36 lines
hw/cxl/cxl-component-utils.c new file +36 -0
+
+void cxl_component_register_init_common(uint32_t *reg_state, enum reg_type type)
+{
+    int caps = 0;
+    switch (type) {
+    case CXL2_DOWNSTREAM_PORT:
+    case CXL2_DEVICE:
+        /* CAP, RAS, Link */
+        caps = 2;
+        break;
+    case CXL2_UPSTREAM_PORT:
+    case CXL2_TYPE3_DEVICE:
+    case CXL2_LOGICAL_DEVICE:
+        /* + HDM */
+        caps = 3;
+        break;
+    case CXL2_ROOT_PORT:
+        /* + Extended Security, + Snoop */
+        caps = 5;
+        break;
+    default:
+        abort();
+    }
+
+    memset(reg_state, 0, 0x1000);
+
+    /* CXL Capability Header Register */
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ID, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ARRAY_SIZE, caps);
+
+
+#define init_cap_reg(reg, id, version)                                        \
+    _Static_assert(CXL_##reg##_REGISTERS_OFFSET != 0, "Invalid cap offset\n");\
+    do {                                                                      \

hw/cxl/cxl-component-utils.c line 154

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+        int which = R_CXL_##reg##_CAPABILITY_HEADER;                          \
+        reg_state[which] = FIELD_DP32(reg_state[which],                       \
+                                      CXL_##reg##_CAPABILITY_HEADER, ID, id); \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER,       \
+                       VERSION, version);                                     \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER, PTR,  \
+                       CXL_##reg##_REGISTERS_OFFSET);                         \
+    } while (0)

Seems like this would be cleaner using ARRAY_FIELD_DP32 as you did for the header.

    #define init_cap_reg(reg, id, version)                                        \
        _Static_assert(CXL_##reg##_REGISTERS_OFFSET != 0, "Invalid cap offset\n");\
        do {                                                                    \
            ARRAY_FIELD_DP32(reg_state, CXL_##reg##_CAPABILITY_HEADER, ID, id); \
            ARRAY_FIELD_DP32(reg_state, CXL_##reg##_CAPABILITY_HEADER,          \
                             VERSION, version);                                 \
            ARRAY_FIELD_DP32(reg_state, CXL_##reg##_CAPABILITY_HEADER,          \
                             PTR, CXL_##reg##_REGISTRS_OFFSET);                 \
	} while (0)

I think gives the same result.

hw/cxl/cxl-component-utils.c line 159

Quoted 5 lines
hw/cxl/cxl-component-utils.c new file +5 -0
+
+    init_cap_reg(RAS, 2, 1);
+    ras_init_common(reg_state);
+
+    init_cap_reg(LINK, 4, 2);

Feels like we'll want to block some values for the rest of these to at least ensure whatever is read isn't crazy.

Quoted 49 lines
hw/cxl/cxl-component-utils.c new file +49 -0
+
+    if (caps < 3) {
+        return;
+    }
+
+    init_cap_reg(HDM, 5, 1);
+    hdm_init_common(reg_state);
+
+    if (caps < 5) {
+        return;
+    }
+
+    init_cap_reg(EXTSEC, 6, 1);
+    init_cap_reg(SNOOP, 8, 1);
+
+#undef init_cap_reg
+}
+
+/*
+ * Helper to creates a DVSEC header for a CXL entity. The caller is responsible
+ * for tracking the valid offset.
+ *
+ * This function will build the DVSEC header on behalf of the caller and then
+ * copy in the remaining data for the vendor specific bits.
+ */
+void cxl_component_create_dvsec(CXLComponentState *cxl, uint16_t length,
+                                uint16_t type, uint8_t rev, uint8_t *body)
+{
+    PCIDevice *pdev = cxl->pdev;
+    uint16_t offset = cxl->dvsec_offset;
+
+    assert(offset >= PCI_CFG_SPACE_SIZE &&
+           ((offset + length) < PCI_CFG_SPACE_EXP_SIZE));
+    assert((length & 0xf000) == 0);
+    assert((rev & ~0xf) == 0);
+
+    /* Create the DVSEC in the MCFG space */
+    pcie_add_capability(pdev, PCI_EXT_CAP_ID_DVSEC, 1, offset, length);
+    pci_set_long(pdev->config + offset + PCIE_DVSEC_HEADER1_OFFSET,
+                 (length << 20) | (rev << 16) | CXL_VENDOR_ID);
+    pci_set_word(pdev->config + offset + PCIE_DVSEC_ID_OFFSET, type);
+    memcpy(pdev->config + offset + sizeof(struct dvsec_header),
+           body + sizeof(struct dvsec_header),
+           length - sizeof(struct dvsec_header));
+
+    /* Update state for future DVSEC additions */
+    range_init_nofail(&cxl->dvsecs[type], cxl->dvsec_offset, length);
+    cxl->dvsec_offset += length;
+}

...

#6 lore raw

Re: [RFC PATCH v3 02/31] hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5)

Quoted 19 lines · Jonathan Cameron
Quoted 16 lines · On Mon, 1 Feb 2021 16:59:19 -0800 Ben Widawsky <[email protected]> wrote:

A CXL 2.0 component is any entity in the CXL topology. All components have a analogous function in PCIe. Except for the CXL host bridge, all have a PCIe config space that is accessible via the common PCIe mechanisms. CXL components are enumerated via DVSEC fields in the extended PCIe header space. CXL components will minimally implement some subset of CXL.mem and CXL.cache registers defined in 8.2.5 of the CXL 2.0 specification. Two headers and a utility library are introduced to support the minimum functionality needed to enumerate components.

The cxl_pci header manages bits associated with PCI, specifically the DVSEC and related fields. The cxl_component.h variant has data structures and APIs that are useful for drivers implementing any of the CXL 2.0 components. The library takes care of making use of the DVSEC bits and the CXL.[mem|cache] registers. Per spec, the registers are little endian.

Quoted 8 lines
Quoted 7 lines

None of the mechanisms required to enumerate a CXL capable hostbridge are introduced at this point.

Note that the CXL.mem and CXL.cache registers used are always 4B wide. It's possible in the future that this constraint will not hold.

A few additions to previous comments.

Thanks for continuing to look.

Quoted 88 lines
Quoted 73 lines
---
 MAINTAINERS                    |   6 +
 hw/Kconfig                     |   1 +
 hw/cxl/Kconfig                 |   3 +
 hw/cxl/cxl-component-utils.c   | 208 +++++++++++++++++++++++++++++++++
 hw/cxl/meson.build             |   3 +
 hw/meson.build                 |   1 +
 include/hw/cxl/cxl.h           |  17 +++
 include/hw/cxl/cxl_component.h | 187 +++++++++++++++++++++++++++++
 include/hw/cxl/cxl_pci.h       | 138 ++++++++++++++++++++++
 9 files changed, 564 insertions(+)
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_pci.h
hw/cxl/cxl-component-utils.c new file +46 -0
new file mode 100644
index 0000000000..8d56ad5c7d
@@ -0,0 +1,208 @@
+/*
+ * CXL Utility library for components
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/pci/pci.h"
+#include "hw/cxl/cxl.h"
+
+static uint64_t cxl_cache_mem_read_reg(void *opaque, hwaddr offset,
+                                       unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->read) {
+        return cregs->special_ops->read(cxl_cstate, offset, size);
+    } else {
+        return cregs->cache_mem_registers[offset / 4];
+    }
+}
+
+static void cxl_cache_mem_write_reg(void *opaque, hwaddr offset, uint64_t value,
+                                    unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+
+    assert(size == 4);
+
+    if (cregs->special_ops && cregs->special_ops->write) {
+        cregs->special_ops->write(cxl_cstate, offset, value, size);
+    } else {
+        cregs->cache_mem_registers[offset / 4] = value;
+    }
+}
+
+/*
+ * 8.2.3

hw/cxl/cxl-component-utils.c line 56

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+ *   The access restrictions specified in Section 8.2.2 also apply to CXL 2.0
+ *   Component Registers.
+ *
+ * 8.2.2
+ *   • A 32 bit register shall be accessed as a 4 Bytes quantity. Partial
+ *   reads are not permitted.
+ *   • A 64 bit register shall be accessed as a 8 Bytes quantity. Partial
+ *   reads are not permitted.
+ *
+ * As of the spec defined today, only 4 byte registers exist.

The exciting exception to this is the RAS header log which is defined as 512 bits. Will seek clarification but I think the spec should probably say that is a set of 32 bit registers.

Quoted 3 lines

A bunch of the other elements that we probably want to block in plausible values for also seem to use 64 bit registers.

IIRC, it was only the link caps, but I can look again. (I don't ever intend to emulate link caps). The RAS log was a mistake...

FWIW, I bunch of feedback about a few of register mixups in this vein and I think there's been errata published but it was before I was sitting in on the consortium calls, so I'm not sure.

Quoted 76 lines
Quoted 16 lines
hw/cxl/cxl-component-utils.c new file +16 -0
+ */
+static const MemoryRegionOps cache_mem_ops = {
+    .read = cxl_cache_mem_read_reg,
+    .write = cxl_cache_mem_write_reg,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+

..

Quoted 36 lines
hw/cxl/cxl-component-utils.c new file +36 -0
+
+void cxl_component_register_init_common(uint32_t *reg_state, enum reg_type type)
+{
+    int caps = 0;
+    switch (type) {
+    case CXL2_DOWNSTREAM_PORT:
+    case CXL2_DEVICE:
+        /* CAP, RAS, Link */
+        caps = 2;
+        break;
+    case CXL2_UPSTREAM_PORT:
+    case CXL2_TYPE3_DEVICE:
+    case CXL2_LOGICAL_DEVICE:
+        /* + HDM */
+        caps = 3;
+        break;
+    case CXL2_ROOT_PORT:
+        /* + Extended Security, + Snoop */
+        caps = 5;
+        break;
+    default:
+        abort();
+    }
+
+    memset(reg_state, 0, 0x1000);
+
+    /* CXL Capability Header Register */
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ID, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, CACHE_MEM_VERSION, 1);
+    ARRAY_FIELD_DP32(reg_state, CXL_CAPABILITY_HEADER, ARRAY_SIZE, caps);
+
+
+#define init_cap_reg(reg, id, version)                                        \
+    _Static_assert(CXL_##reg##_REGISTERS_OFFSET != 0, "Invalid cap offset\n");\
+    do {                                                                      \

hw/cxl/cxl-component-utils.c line 154

Quoted 10 lines
hw/cxl/cxl-component-utils.c new file +10 -0
+        int which = R_CXL_##reg##_CAPABILITY_HEADER;                          \
+        reg_state[which] = FIELD_DP32(reg_state[which],                       \
+                                      CXL_##reg##_CAPABILITY_HEADER, ID, id); \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER,       \
+                       VERSION, version);                                     \
+        reg_state[which] =                                                    \
+            FIELD_DP32(reg_state[which], CXL_##reg##_CAPABILITY_HEADER, PTR,  \
+                       CXL_##reg##_REGISTERS_OFFSET);                         \
+    } while (0)

Seems like this would be cleaner using ARRAY_FIELD_DP32 as you did for the header.

    #define init_cap_reg(reg, id, version)                                        \
        _Static_assert(CXL_##reg##_REGISTERS_OFFSET != 0, "Invalid cap offset\n");\
        do {                                                                    \
            ARRAY_FIELD_DP32(reg_state, CXL_##reg##_CAPABILITY_HEADER, ID, id); \
            ARRAY_FIELD_DP32(reg_state, CXL_##reg##_CAPABILITY_HEADER,          \
                             VERSION, version);                                 \
            ARRAY_FIELD_DP32(reg_state, CXL_##reg##_CAPABILITY_HEADER,          \
                             PTR, CXL_##reg##_REGISTRS_OFFSET);                 \
	} while (0)
Quoted 2 lines

I think gives the same result.

I think it looks better too. I don't remember why I didn't do this.

Could I entice you to send a tested patch to change it? I'll gladly put it on top. I'm trying to not mess with the original patches at this point and do everything on top, until someone yells to squash it in.

Quoted 9 lines

hw/cxl/cxl-component-utils.c line 159

Quoted 5 lines
hw/cxl/cxl-component-utils.c new file +5 -0
+
+    init_cap_reg(RAS, 2, 1);
+    ras_init_common(reg_state);
+
+    init_cap_reg(LINK, 4, 2);

Feels like we'll want to block some values for the rest of these to at least ensure whatever is read isn't crazy.

Yep. I've pretty much left everything as a TODO in the component register block. I only did RAS as an example on how one would add things, but then I ended up adding HDM as a better example.

Would be good as part of the "project plan" to identify what registers are interesting to implement.

Quoted 52 lines
Quoted 49 lines
hw/cxl/cxl-component-utils.c new file +49 -0
+
+    if (caps < 3) {
+        return;
+    }
+
+    init_cap_reg(HDM, 5, 1);
+    hdm_init_common(reg_state);
+
+    if (caps < 5) {
+        return;
+    }
+
+    init_cap_reg(EXTSEC, 6, 1);
+    init_cap_reg(SNOOP, 8, 1);
+
+#undef init_cap_reg
+}
+
+/*
+ * Helper to creates a DVSEC header for a CXL entity. The caller is responsible
+ * for tracking the valid offset.
+ *
+ * This function will build the DVSEC header on behalf of the caller and then
+ * copy in the remaining data for the vendor specific bits.
+ */
+void cxl_component_create_dvsec(CXLComponentState *cxl, uint16_t length,
+                                uint16_t type, uint8_t rev, uint8_t *body)
+{
+    PCIDevice *pdev = cxl->pdev;
+    uint16_t offset = cxl->dvsec_offset;
+
+    assert(offset >= PCI_CFG_SPACE_SIZE &&
+           ((offset + length) < PCI_CFG_SPACE_EXP_SIZE));
+    assert((length & 0xf000) == 0);
+    assert((rev & ~0xf) == 0);
+
+    /* Create the DVSEC in the MCFG space */
+    pcie_add_capability(pdev, PCI_EXT_CAP_ID_DVSEC, 1, offset, length);
+    pci_set_long(pdev->config + offset + PCIE_DVSEC_HEADER1_OFFSET,
+                 (length << 20) | (rev << 16) | CXL_VENDOR_ID);
+    pci_set_word(pdev->config + offset + PCIE_DVSEC_ID_OFFSET, type);
+    memcpy(pdev->config + offset + sizeof(struct dvsec_header),
+           body + sizeof(struct dvsec_header),
+           length - sizeof(struct dvsec_header));
+
+    /* Update state for future DVSEC additions */
+    range_init_nofail(&cxl->dvsecs[type], cxl->dvsec_offset, length);
+    cxl->dvsec_offset += length;
+}

...

#7 lore raw

RFC PATCH v3 3/31 hw/cxl/device: Introduce a CXL device (8.2.8)

In reply to Ben Widawsky

A CXL device is a type of CXL component. Conceptually, a CXL device would be a leaf node in a CXL topology. From an emulation perspective, CXL devices are the most complex and so the actual implementation is reserved for discrete commits.

This new device type is specifically catered towards the eventual implementation of a Type3 CXL.mem device, 8.2.8.5 in the CXL 2.0 specification.

---
 include/hw/cxl/cxl.h        |   1 +
 include/hw/cxl/cxl_device.h | 155 ++++++++++++++++++++++++++++++++++++
 2 files changed, 156 insertions(+)
 create mode 100644 include/hw/cxl/cxl_device.h
include/hw/cxl/cxl.h +1 -0
index 55f6cc30a5..23f52c4cf9 100644
@@ -12,6 +12,7 @@
 
 #include "cxl_pci.h"
 #include "cxl_component.h"
+#include "cxl_device.h"
 
 #endif
 
include/hw/cxl/cxl_device.h new file +155 -0

155 lines not on this page · raw message

Signature · 1 line
2.30.0
Reviewed-by on v3 3/31
#8 lore raw

Re: [RFC PATCH v3 03/31] hw/cxl/device: Introduce a CXL device (8.2.8)

Quoted 10 lines · On Mon, 1 Feb 2021 16:59:20 -0800 Ben Widawsky <[email protected]> wrote:

A CXL device is a type of CXL component. Conceptually, a CXL device would be a leaf node in a CXL topology. From an emulation perspective, CXL devices are the most complex and so the actual implementation is reserved for discrete commits.

This new device type is specifically catered towards the eventual implementation of a Type3 CXL.mem device, 8.2.8.5 in the CXL 2.0 specification.

Really minor comments inline.

In the interests of avoiding giving myself a headache again

Quoted 81 lines
---
 include/hw/cxl/cxl.h        |   1 +
 include/hw/cxl/cxl_device.h | 155 ++++++++++++++++++++++++++++++++++++
 2 files changed, 156 insertions(+)
 create mode 100644 include/hw/cxl/cxl_device.h
include/hw/cxl/cxl.h +1 -0
index 55f6cc30a5..23f52c4cf9 100644
@@ -12,6 +12,7 @@
 
 #include "cxl_pci.h"
 #include "cxl_component.h"
+#include "cxl_device.h"
 
 #endif
 
include/hw/cxl/cxl_device.h new file +57 -0
new file mode 100644
index 0000000000..a85f250503
@@ -0,0 +1,155 @@
+/*
+ * QEMU CXL Devices
+ *
+ * Copyright (c) 2020 Intel
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#ifndef CXL_DEVICE_H
+#define CXL_DEVICE_H
+
+#include "hw/register.h"
+
+/*
+ * The following is how a CXL device's MMIO space is laid out. The only
+ * requirement from the spec is that the capabilities array and the capability
+ * headers start at offset 0 and are contiguously packed. The headers themselves
+ * provide offsets to the register fields. For this emulation, registers will
+ * start at offset 0x80 (m == 0x80). No secondary mailbox is implemented which
+ * means that n = m + sizeof(mailbox registers) + sizeof(device registers).
+ *
+ * This is roughly described in 8.2.8 Figure 138 of the CXL 2.0 spec.
+ *
+ * n + PAYLOAD_SIZE_MAX  +---------------------------------+
+ *                       |                                 |
+ *                  ^    |                                 |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |         Command Payload         |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  n    +---------------------------------+
+ *                  ^    |                                 |
+ *                  |    |    Device Capability Registers  |
+ *                  |    |    x, mailbox, y                |
+ *                  |    |                                 |
+ *                  m    +---------------------------------+
+ *                  ^    |     Device Capability Header y  |
+ *                  |    +---------------------------------+
+ *                  |    | Device Capability Header Mailbox|
+ *                  |    +------------- --------------------
+ *                  |    |     Device Capability Header x  |
+ *                  |    +---------------------------------+
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |      Device Cap Array[0..n]     |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  |    |                                 |
+ *                  0    +---------------------------------+
+ */
+

include/hw/cxl/cxl_device.h line 67

Quoted 10 lines
include/hw/cxl/cxl_device.h new file +10 -0
+#define CXL_DEVICE_CAP_HDR1_OFFSET 0x10 /* Figure 138 */
+#define CXL_DEVICE_CAP_REG_SIZE 0x10 /* 8.2.8.2 */
+#define CXL_DEVICE_CAPS_MAX 4 /* 8.2.8.2.1 + 8.2.8.5 */
+
+#define CXL_DEVICE_REGISTERS_OFFSET 0x80 /* Read comment above */
+#define CXL_DEVICE_REGISTERS_LENGTH 0x8 /* 8.2.8.3.1 */
+
+#define CXL_MAILBOX_REGISTERS_OFFSET \
+    (CXL_DEVICE_REGISTERS_OFFSET + CXL_DEVICE_REGISTERS_LENGTH)
+#define CXL_MAILBOX_REGISTERS_SIZE 0x20

Perhaps a ref to 8.2.8.4 or Figure 139 here somewhere?
Thanks for all the refs by the way. They make checking this a lot quicker!

Quoted 22 lines
include/hw/cxl/cxl_device.h new file +22 -0
+#define CXL_MAILBOX_PAYLOAD_SHIFT 11
+#define CXL_MAILBOX_MAX_PAYLOAD_SIZE (1 << CXL_MAILBOX_PAYLOAD_SHIFT)
+#define CXL_MAILBOX_REGISTERS_LENGTH \
+    (CXL_MAILBOX_REGISTERS_SIZE + CXL_MAILBOX_MAX_PAYLOAD_SIZE)
+
+typedef struct cxl_device_state {
+    MemoryRegion device_registers;
+
+    /* mmio for device capabilities array - 8.2.8.2 */
+    MemoryRegion caps;
+
+    /* mmio for the device status registers 8.2.8.3 */
+    MemoryRegion device;
+
+    /* mmio for the mailbox registers 8.2.8.4 */
+    MemoryRegion mailbox;
+
+    /* memory region for persistent memory, HDM */
+    MemoryRegion *pmem;
+
+    /* memory region for volatile  memory, HDM */
+    MemoryRegion *vmem;

include/hw/cxl/cxl_device.h line 99

Quoted 10 lines
include/hw/cxl/cxl_device.h new file +10 -0
+} CXLDeviceState;
+
+/* Initialize the register block for a device */
+void cxl_device_register_block_init(Object *obj, CXLDeviceState *dev);
+
+/* Set up default values for the register block */
+void cxl_device_register_init_common(CXLDeviceState *dev);
+
+/* CXL 2.0 - 8.2.8.1 */
+REG32(CXL_DEV_CAP_ARRAY, 0) /* 48b!?!?! */

Also missing a reserved 64 bits to fill in below the device capability headers which are offset by 0x10

Quoted 29 lines
include/hw/cxl/cxl_device.h new file +29 -0
+    FIELD(CXL_DEV_CAP_ARRAY, CAP_ID, 0, 16)
+    FIELD(CXL_DEV_CAP_ARRAY, CAP_VERSION, 16, 8)
+REG32(CXL_DEV_CAP_ARRAY2, 4) /* We're going to pretend it's 64b */
+    FIELD(CXL_DEV_CAP_ARRAY2, CAP_COUNT, 0, 16)
+
+/*
+ * Helper macro to initialize capability headers for CXL devices.
+ *
+ * In the 8.2.8.2, this is listed as a 128b register, but in 8.2.8, it says:
+ * > No registers defined in Section 8.2.8 are larger than 64-bits wide so that
+ * > is the maximum access size allowed for these registers. If this rule is not
+ * > followed, the behavior is undefined
+ *
+ * Here we've chosen to make it 4 dwords. The spec allows any pow2 multiple
+ * access to be used for a register (2 qwords, 8 words, 128 bytes).
+ */
+#define CXL_DEVICE_CAPABILITY_HEADER_REGISTER(n, offset)                            \
+    REG32(CXL_DEV_##n##_CAP_HDR0, offset)                 \
+        FIELD(CXL_DEV_##n##_CAP_HDR0, CAP_ID, 0, 16)      \
+        FIELD(CXL_DEV_##n##_CAP_HDR0, CAP_VERSION, 16, 8) \
+    REG32(CXL_DEV_##n##_CAP_HDR1, offset + 4)             \
+        FIELD(CXL_DEV_##n##_CAP_HDR1, CAP_OFFSET, 0, 32)  \
+    REG32(CXL_DEV_##n##_CAP_HDR2, offset + 8)             \
+        FIELD(CXL_DEV_##n##_CAP_HDR2, CAP_LENGTH, 0, 32)
+
+CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
+CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
+                                               CXL_DEVICE_CAP_REG_SIZE)
+

include/hw/cxl/cxl_device.h line 138

Quoted 10 lines
include/hw/cxl/cxl_device.h new file +10 -0
+REG32(CXL_DEV_MAILBOX_CAP, 0)
+    FIELD(CXL_DEV_MAILBOX_CAP, PAYLOAD_SIZE, 0, 5)
+    FIELD(CXL_DEV_MAILBOX_CAP, INT_CAP, 5, 1)
+    FIELD(CXL_DEV_MAILBOX_CAP, BG_INT_CAP, 6, 1)
+    FIELD(CXL_DEV_MAILBOX_CAP, MSI_N, 7, 4)
+
+REG32(CXL_DEV_MAILBOX_CTRL, 4)
+    FIELD(CXL_DEV_MAILBOX_CTRL, DOORBELL, 0, 1)
+    FIELD(CXL_DEV_MAILBOX_CTRL, INT_EN, 1, 1)
+    FIELD(CXL_DEV_MAILBOX_CTRL, BG_INT_EN, 2, 1)

Is it not worth defining the
CXL_DEV_MAILBOX_CMD register for completeness? off set 0x8

Quoted 5 lines
include/hw/cxl/cxl_device.h new file +5 -0
+
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_MAILBOX_STS, 0x10)
+    FIELD(CXL_DEV_MAILBOX_STS, BG_OP, 0, 1)
+    FIELD(CXL_DEV_MAILBOX_STS, ERRNO, 32, 16)

include/hw/cxl/cxl_device.h line 153

Quoted 10 lines
include/hw/cxl/cxl_device.h new file +10 -0
+    FIELD(CXL_DEV_MAILBOX_STS, VENDOR_ERRNO, 48, 16)
+
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_BG_CMD_STS, 0x18)
+    FIELD(CXL_DEV_BG_CMD_STS, BG, 0, 16)
+    FIELD(CXL_DEV_BG_CMD_STS, DONE, 16, 7)
+    FIELD(CXL_DEV_BG_CMD_STS, ERRNO, 32, 16)
+    FIELD(CXL_DEV_BG_CMD_STS, VENDOR_ERRNO, 48, 16)
+
+REG32(CXL_DEV_CMD_PAYLOAD, 0x20)

Probably want a comment for this one that it might be huge.

Quoted 2 lines
include/hw/cxl/cxl_device.h new file +2 -0
+
+#endif
#9 lore raw

RFC PATCH v3 4/31 hw/cxl/device: Implement the CAP array (8.2.8.1-2)

In reply to Ben Widawsky

This implements all device MMIO up to the first capability. That includes the CXL Device Capabilities Array Register, as well as all of the CXL Device Capability Header Registers. The latter are filled in as they are implemented in the following patches.

Endianness and alignment are managed by softmmu memory core.

---
 hw/cxl/cxl-device-utils.c   | 105 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl_device.h |  27 +++++++++-
 3 files changed, 132 insertions(+), 1 deletion(-)
 create mode 100644 hw/cxl/cxl-device-utils.c
hw/cxl/cxl-device-utils.c new file +105 -0
new file mode 100644
index 0000000000..bb15ad9a0f
@@ -0,0 +1,105 @@
+/*
+ * CXL Utility library for devices
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "qemu/log.h"
+#include "hw/cxl/cxl.h"
+
+/*
+ * Device registers have no restrictions per the spec, and so fall back to the
+ * default memory mapped register rules in 8.2:
+ *   Software shall use CXL.io Memory Read and Write to access memory mapped
+ *   register defined in this section. Unless otherwise specified, software
+ *   shall restrict the accesses width based on the following:
+ *   • A 32 bit register shall   be accessed as a 1 Byte, 2 Bytes or 4 Bytes
+ *     quantity.
+ *   • A 64 bit register shall be accessed as a 1 Byte, 2 Bytes, 4 Bytes or 8
+ *     Bytes
+ *   • The address shall be a multiple of the access width, e.g. when
+ *     accessing a register as a 4 Byte quantity, the address shall be
+ *     multiple of 4.
+ *   • The accesses shall map to contiguous bytes.If these rules are not
+ *     followed, the behavior is undefined
+ */
+
+static uint64_t caps_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    return cxl_dstate->caps_reg_state32[offset / 4];
+}
+
+static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    return 0;
+}
+
+static const MemoryRegionOps dev_ops = {
+    .read = dev_reg_read,
+    .write = NULL, /* status register is read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+    },
+};
+
+static const MemoryRegionOps caps_ops = {
+    .read = caps_reg_read,
+    .write = NULL, /* caps registers are read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+
+void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
+{
+    /* This will be a BAR, so needs to be rounded up to pow2 for PCI spec */
+    memory_region_init(&cxl_dstate->device_registers, obj, "device-registers",
+                       pow2ceil(CXL_MMIO_SIZE));
+
+    memory_region_init_io(&cxl_dstate->caps, obj, &caps_ops, cxl_dstate,
+                          "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);
+    memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
+                          "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+
+    memory_region_add_subregion(&cxl_dstate->device_registers, 0,
+                                &cxl_dstate->caps);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_DEVICE_REGISTERS_OFFSET,
+                                &cxl_dstate->device);
+}
+
+static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
+
+void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
+{
+    uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
+    const int cap_count = 1;
+
+    /* CXL Device Capabilities Array Register */
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_VERSION, 1);
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY2, CAP_COUNT, cap_count);
+
+    cxl_device_cap_init(cxl_dstate, DEVICE, 1);
+    device_reg_init_common(cxl_dstate);
+}
hw/cxl/meson.build +1 -0
index 00c3876a0f..47154d6850 100644
@@ -1,3 +1,4 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
+  'cxl-device-utils.c',
 ))
include/hw/cxl/cxl_device.h +26 -1
index a85f250503..f3bcf19410 100644
@@ -58,6 +58,8 @@
 #define CXL_DEVICE_CAP_HDR1_OFFSET 0x10 /* Figure 138 */
 #define CXL_DEVICE_CAP_REG_SIZE 0x10 /* 8.2.8.2 */
 #define CXL_DEVICE_CAPS_MAX 4 /* 8.2.8.2.1 + 8.2.8.5 */
+#define CXL_CAPS_SIZE \
+    (CXL_DEVICE_CAP_REG_SIZE * CXL_DEVICE_CAPS_MAX + 1) /* +1 for header */
 
 #define CXL_DEVICE_REGISTERS_OFFSET 0x80 /* Read comment above */
 #define CXL_DEVICE_REGISTERS_LENGTH 0x8 /* 8.2.8.3.1 */
@@ -70,11 +72,18 @@
 #define CXL_MAILBOX_REGISTERS_LENGTH \
     (CXL_MAILBOX_REGISTERS_SIZE + CXL_MAILBOX_MAX_PAYLOAD_SIZE)
 
+#define CXL_MMIO_SIZE                                       \
+    CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_REGISTERS_LENGTH + \
+        CXL_MAILBOX_REGISTERS_LENGTH
+
 typedef struct cxl_device_state {
     MemoryRegion device_registers;
 
     /* mmio for device capabilities array - 8.2.8.2 */
-    MemoryRegion caps;
+    struct {
+        MemoryRegion caps;
+        uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
+    };
 
     /* mmio for the device status registers 8.2.8.3 */
     MemoryRegion device;
@@ -126,6 +135,22 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+#define cxl_device_cap_init(dstate, reg, cap_id)                                   \
+    do {                                                                           \
+        uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \
+        int which = R_CXL_DEV_##reg##_CAP_HDR0;                                    \
+        cap_hdrs[which] =                                                          \
+            FIELD_DP32(cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_ID, cap_id); \
+        cap_hdrs[which] = FIELD_DP32(                                              \
+            cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_VERSION, 1);            \
+        cap_hdrs[which + 1] =                                                      \
+            FIELD_DP32(cap_hdrs[which + 1], CXL_DEV_##reg##_CAP_HDR1,              \
+                       CAP_OFFSET, CXL_##reg##_REGISTERS_OFFSET);                  \
+        cap_hdrs[which + 2] =                                                      \
+            FIELD_DP32(cap_hdrs[which + 2], CXL_DEV_##reg##_CAP_HDR2,              \
+                       CAP_LENGTH, CXL_##reg##_REGISTERS_LENGTH);                  \
+    } while (0)
+
 REG32(CXL_DEV_MAILBOX_CAP, 0)
     FIELD(CXL_DEV_MAILBOX_CAP, PAYLOAD_SIZE, 0, 5)
     FIELD(CXL_DEV_MAILBOX_CAP, INT_CAP, 5, 1)
Signature · 1 line
2.30.0
#10 lore raw

Re: [RFC PATCH v3 04/31] hw/cxl/device: Implement the CAP array (8.2.8.1-2)

Quoted 8 lines · On Mon, 1 Feb 2021 16:59:21 -0800 Ben Widawsky <[email protected]> wrote:

This implements all device MMIO up to the first capability. That includes the CXL Device Capabilities Array Register, as well as all of the CXL Device Capability Header Registers. The latter are filled in as they are implemented in the following patches.

Endianness and alignment are managed by softmmu memory core.

A few trivials

Quoted 23 lines
---
 hw/cxl/cxl-device-utils.c   | 105 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl_device.h |  27 +++++++++-
 3 files changed, 132 insertions(+), 1 deletion(-)
 create mode 100644 hw/cxl/cxl-device-utils.c
hw/cxl/cxl-device-utils.c new file +10 -0
new file mode 100644
index 0000000000..bb15ad9a0f
@@ -0,0 +1,105 @@
+/*
+ * CXL Utility library for devices
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"

hw/cxl/cxl-device-utils.c line 20

Quoted 10 lines
hw/cxl/cxl-device-utils.c new file +10 -0
+#include "qemu/log.h"
+#include "hw/cxl/cxl.h"
+
+/*
+ * Device registers have no restrictions per the spec, and so fall back to the
+ * default memory mapped register rules in 8.2:
+ *   Software shall use CXL.io Memory Read and Write to access memory mapped
+ *   register defined in this section. Unless otherwise specified, software
+ *   shall restrict the accesses width based on the following:
+ *   • A 32 bit register shall   be accessed as a 1 Byte, 2 Bytes or 4 Bytes

odd spacing

Quoted 50 lines
hw/cxl/cxl-device-utils.c new file +50 -0
+ *     quantity.
+ *   • A 64 bit register shall be accessed as a 1 Byte, 2 Bytes, 4 Bytes or 8
+ *     Bytes
+ *   • The address shall be a multiple of the access width, e.g. when
+ *     accessing a register as a 4 Byte quantity, the address shall be
+ *     multiple of 4.
+ *   • The accesses shall map to contiguous bytes.If these rules are not
+ *     followed, the behavior is undefined
+ */
+
+static uint64_t caps_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    return cxl_dstate->caps_reg_state32[offset / 4];
+}
+
+static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    return 0;
+}
+
+static const MemoryRegionOps dev_ops = {
+    .read = dev_reg_read,
+    .write = NULL, /* status register is read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+    },
+};
+
+static const MemoryRegionOps caps_ops = {
+    .read = caps_reg_read,
+    .write = NULL, /* caps registers are read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },

hw/cxl/cxl-device-utils.c line 80

Quoted 10 lines
hw/cxl/cxl-device-utils.c new file +10 -0
+};
+
+void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
+{
+    /* This will be a BAR, so needs to be rounded up to pow2 for PCI spec */
+    memory_region_init(&cxl_dstate->device_registers, obj, "device-registers",
+                       pow2ceil(CXL_MMIO_SIZE));
+
+    memory_region_init_io(&cxl_dstate->caps, obj, &caps_ops, cxl_dstate,
+                          "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);

Specifying a size in terms of the offset of another region isn't exactly intuitive so perhaps a comment on why or better yet actually use a size parameter covering what is there rather than simply the region below the CXL_DEVICE_REGISTERS_OFFSET.

Quoted 54 lines
hw/cxl/cxl-device-utils.c new file +25 -0
+    memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
+                          "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+
+    memory_region_add_subregion(&cxl_dstate->device_registers, 0,
+                                &cxl_dstate->caps);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_DEVICE_REGISTERS_OFFSET,
+                                &cxl_dstate->device);
+}
+
+static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
+
+void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
+{
+    uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
+    const int cap_count = 1;
+
+    /* CXL Device Capabilities Array Register */
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_VERSION, 1);
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY2, CAP_COUNT, cap_count);
+
+    cxl_device_cap_init(cxl_dstate, DEVICE, 1);
+    device_reg_init_common(cxl_dstate);
+}
hw/cxl/meson.build +1 -0
index 00c3876a0f..47154d6850 100644
@@ -1,3 +1,4 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
+  'cxl-device-utils.c',
 ))
include/hw/cxl/cxl_device.h +5 -0
index a85f250503..f3bcf19410 100644
@@ -58,6 +58,8 @@
 #define CXL_DEVICE_CAP_HDR1_OFFSET 0x10 /* Figure 138 */
 #define CXL_DEVICE_CAP_REG_SIZE 0x10 /* 8.2.8.2 */
 #define CXL_DEVICE_CAPS_MAX 4 /* 8.2.8.2.1 + 8.2.8.5 */
+#define CXL_CAPS_SIZE \
+    (CXL_DEVICE_CAP_REG_SIZE * CXL_DEVICE_CAPS_MAX + 1) /* +1 for header */
 
 #define CXL_DEVICE_REGISTERS_OFFSET 0x80 /* Read comment above */
 #define CXL_DEVICE_REGISTERS_LENGTH 0x8 /* 8.2.8.3.1 */
@@ -70,11 +72,18 @@
 #define CXL_MAILBOX_REGISTERS_LENGTH \
     (CXL_MAILBOX_REGISTERS_SIZE + CXL_MAILBOX_MAX_PAYLOAD_SIZE)
 
+#define CXL_MMIO_SIZE                                       \
+    CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_REGISTERS_LENGTH + \
+        CXL_MAILBOX_REGISTERS_LENGTH

include/hw/cxl/cxl_device.h line 86

Quoted 10 lines
include/hw/cxl/cxl_device.h +5 -1
+
 typedef struct cxl_device_state {
     MemoryRegion device_registers;
 
     /* mmio for device capabilities array - 8.2.8.2 */
-    MemoryRegion caps;
+    struct {
+        MemoryRegion caps;
+        uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
+    };

With this unnamed,w hat is the benefit of having these two in a struct? The naming makes it clear they are related anyway.

Quoted 26 lines
include/hw/cxl/cxl_device.h +16 -0
     /* mmio for the device status registers 8.2.8.3 */
     MemoryRegion device;
@@ -126,6 +135,22 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+#define cxl_device_cap_init(dstate, reg, cap_id)                                   \
+    do {                                                                           \
+        uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \
+        int which = R_CXL_DEV_##reg##_CAP_HDR0;                                    \
+        cap_hdrs[which] =                                                          \
+            FIELD_DP32(cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_ID, cap_id); \
+        cap_hdrs[which] = FIELD_DP32(                                              \
+            cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_VERSION, 1);            \
+        cap_hdrs[which + 1] =                                                      \
+            FIELD_DP32(cap_hdrs[which + 1], CXL_DEV_##reg##_CAP_HDR1,              \
+                       CAP_OFFSET, CXL_##reg##_REGISTERS_OFFSET);                  \
+        cap_hdrs[which + 2] =                                                      \
+            FIELD_DP32(cap_hdrs[which + 2], CXL_DEV_##reg##_CAP_HDR2,              \
+                       CAP_LENGTH, CXL_##reg##_REGISTERS_LENGTH);                  \
+    } while (0)
+
 REG32(CXL_DEV_MAILBOX_CAP, 0)
     FIELD(CXL_DEV_MAILBOX_CAP, PAYLOAD_SIZE, 0, 5)
     FIELD(CXL_DEV_MAILBOX_CAP, INT_CAP, 5, 1)
#11 lore raw

Re: [RFC PATCH v3 04/31] hw/cxl/device: Implement the CAP array (8.2.8.1-2)

Quoted 105 lines · Jonathan Cameron
Quoted 8 lines · On Mon, 1 Feb 2021 16:59:21 -0800 Ben Widawsky <[email protected]> wrote:

This implements all device MMIO up to the first capability. That includes the CXL Device Capabilities Array Register, as well as all of the CXL Device Capability Header Registers. The latter are filled in as they are implemented in the following patches.

Endianness and alignment are managed by softmmu memory core.

A few trivials

Quoted 23 lines
---
 hw/cxl/cxl-device-utils.c   | 105 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl_device.h |  27 +++++++++-
 3 files changed, 132 insertions(+), 1 deletion(-)
 create mode 100644 hw/cxl/cxl-device-utils.c
hw/cxl/cxl-device-utils.c new file +10 -0
new file mode 100644
index 0000000000..bb15ad9a0f
@@ -0,0 +1,105 @@
+/*
+ * CXL Utility library for devices
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"

hw/cxl/cxl-device-utils.c line 20

Quoted 10 lines
hw/cxl/cxl-device-utils.c new file +10 -0
+#include "qemu/log.h"
+#include "hw/cxl/cxl.h"
+
+/*
+ * Device registers have no restrictions per the spec, and so fall back to the
+ * default memory mapped register rules in 8.2:
+ *   Software shall use CXL.io Memory Read and Write to access memory mapped
+ *   register defined in this section. Unless otherwise specified, software
+ *   shall restrict the accesses width based on the following:
+ *   • A 32 bit register shall   be accessed as a 1 Byte, 2 Bytes or 4 Bytes

odd spacing

Quoted 57 lines
hw/cxl/cxl-device-utils.c new file +57 -0
+ *     quantity.
+ *   • A 64 bit register shall be accessed as a 1 Byte, 2 Bytes, 4 Bytes or 8
+ *     Bytes
+ *   • The address shall be a multiple of the access width, e.g. when
+ *     accessing a register as a 4 Byte quantity, the address shall be
+ *     multiple of 4.
+ *   • The accesses shall map to contiguous bytes.If these rules are not
+ *     followed, the behavior is undefined
+ */
+
+static uint64_t caps_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    return cxl_dstate->caps_reg_state32[offset / 4];
+}
+
+static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    return 0;
+}
+
+static const MemoryRegionOps dev_ops = {
+    .read = dev_reg_read,
+    .write = NULL, /* status register is read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+    },
+};
+
+static const MemoryRegionOps caps_ops = {
+    .read = caps_reg_read,
+    .write = NULL, /* caps registers are read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 4,
+    },
+};
+
+void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
+{
+    /* This will be a BAR, so needs to be rounded up to pow2 for PCI spec */
+    memory_region_init(&cxl_dstate->device_registers, obj, "device-registers",
+                       pow2ceil(CXL_MMIO_SIZE));
Quoted 9 lines

hw/cxl/cxl-device-utils.c line 80

Quoted 3 lines
hw/cxl/cxl-device-utils.c new file +3 -0
+
+    memory_region_init_io(&cxl_dstate->caps, obj, &caps_ops, cxl_dstate,
+                          "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);

Specifying a size in terms of the offset of another region isn't exactly intuitive so perhaps a comment on why or better yet actually use a size parameter covering what is there rather than simply the region below the CXL_DEVICE_REGISTERS_OFFSET.

I didn't have a simple way to accommodate this before, but now I do have CXL_CAPS_SIZE which is what we want here.

You also made me realize I have a PEMDAS bug in CXL_CAPS_SIZE, so thanks.

Quoted 97 lines
Quoted 55 lines
hw/cxl/cxl-device-utils.c new file +25 -0
+    memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
+                          "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+
+    memory_region_add_subregion(&cxl_dstate->device_registers, 0,
+                                &cxl_dstate->caps);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_DEVICE_REGISTERS_OFFSET,
+                                &cxl_dstate->device);
+}
+
+static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
+
+void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
+{
+    uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
+    const int cap_count = 1;
+
+    /* CXL Device Capabilities Array Register */
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_VERSION, 1);
+    ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY2, CAP_COUNT, cap_count);
+
+    cxl_device_cap_init(cxl_dstate, DEVICE, 1);
+    device_reg_init_common(cxl_dstate);
+}
hw/cxl/meson.build +1 -0
index 00c3876a0f..47154d6850 100644
@@ -1,3 +1,4 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
+  'cxl-device-utils.c',
 ))
include/hw/cxl/cxl_device.h +5 -0
index a85f250503..f3bcf19410 100644
@@ -58,6 +58,8 @@
 #define CXL_DEVICE_CAP_HDR1_OFFSET 0x10 /* Figure 138 */
 #define CXL_DEVICE_CAP_REG_SIZE 0x10 /* 8.2.8.2 */
 #define CXL_DEVICE_CAPS_MAX 4 /* 8.2.8.2.1 + 8.2.8.5 */
+#define CXL_CAPS_SIZE \
+    (CXL_DEVICE_CAP_REG_SIZE * CXL_DEVICE_CAPS_MAX + 1) /* +1 for header */
 
 #define CXL_DEVICE_REGISTERS_OFFSET 0x80 /* Read comment above */
 #define CXL_DEVICE_REGISTERS_LENGTH 0x8 /* 8.2.8.3.1 */
@@ -70,11 +72,18 @@
 #define CXL_MAILBOX_REGISTERS_LENGTH \
     (CXL_MAILBOX_REGISTERS_SIZE + CXL_MAILBOX_MAX_PAYLOAD_SIZE)
 
+#define CXL_MMIO_SIZE                                       \
+    CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_REGISTERS_LENGTH + \
+        CXL_MAILBOX_REGISTERS_LENGTH

include/hw/cxl/cxl_device.h line 86

Quoted 10 lines
include/hw/cxl/cxl_device.h +5 -1
+
 typedef struct cxl_device_state {
     MemoryRegion device_registers;
 
     /* mmio for device capabilities array - 8.2.8.2 */
-    MemoryRegion caps;
+    struct {
+        MemoryRegion caps;
+        uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
+    };

With this unnamed,w hat is the benefit of having these two in a struct? The naming makes it clear they are related anyway.

Quoted 26 lines
include/hw/cxl/cxl_device.h +16 -0
     /* mmio for the device status registers 8.2.8.3 */
     MemoryRegion device;
@@ -126,6 +135,22 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+#define cxl_device_cap_init(dstate, reg, cap_id)                                   \
+    do {                                                                           \
+        uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \
+        int which = R_CXL_DEV_##reg##_CAP_HDR0;                                    \
+        cap_hdrs[which] =                                                          \
+            FIELD_DP32(cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_ID, cap_id); \
+        cap_hdrs[which] = FIELD_DP32(                                              \
+            cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_VERSION, 1);            \
+        cap_hdrs[which + 1] =                                                      \
+            FIELD_DP32(cap_hdrs[which + 1], CXL_DEV_##reg##_CAP_HDR1,              \
+                       CAP_OFFSET, CXL_##reg##_REGISTERS_OFFSET);                  \
+        cap_hdrs[which + 2] =                                                      \
+            FIELD_DP32(cap_hdrs[which + 2], CXL_DEV_##reg##_CAP_HDR2,              \
+                       CAP_LENGTH, CXL_##reg##_REGISTERS_LENGTH);                  \
+    } while (0)
+
 REG32(CXL_DEV_MAILBOX_CAP, 0)
     FIELD(CXL_DEV_MAILBOX_CAP, PAYLOAD_SIZE, 0, 5)
     FIELD(CXL_DEV_MAILBOX_CAP, INT_CAP, 5, 1)
#12 lore raw

RFC PATCH v3 5/31 hw/cxl/device: Implement basic mailbox (8.2.8.4)

In reply to Ben Widawsky

This is the beginning of implementing mailbox support for CXL 2.0 devices. The implementation recognizes when the doorbell is rung, handles the command/payload, clears the doorbell while returning error codes and data.

Generally the mailbox mechanism is designed to permit communication between the host OS and the firmware running on the device. For our purposes, we emulate both the firmware, implemented primarily in cxl-mailbox-utils.c, and the hardware.

No commands are implemented yet.

---
 hw/cxl/cxl-device-utils.c   | 125 ++++++++++++++++++++++-
 hw/cxl/cxl-mailbox-utils.c  | 197 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl.h        |   3 +
 include/hw/cxl/cxl_device.h |  28 ++++-
 5 files changed, 349 insertions(+), 5 deletions(-)
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
hw/cxl/cxl-device-utils.c +124 -1
index bb15ad9a0f..6602606f3d 100644
@@ -40,6 +40,111 @@ static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
     return 0;
 }
 
+static uint64_t mailbox_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    switch (size) {
+    case 8:
+        return cxl_dstate->mbox_reg_state64[offset / 8];
+    case 4:
+        return cxl_dstate->mbox_reg_state32[offset / 4];
+    default:
+        g_assert_not_reached();
+    }
+}
+
+static void mailbox_mem_writel(uint32_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CTRL:
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_CAP:
+        /* RO register */
+        break;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 32-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        break;
+    }
+
+    reg_state[offset / 4] = value;
+}
+
+static void mailbox_mem_writeq(uint64_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CMD:
+        break;
+    case A_CXL_DEV_BG_CMD_STS:
+        /* BG not supported */
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_STS:
+        /* Read only register, will get updated by the state machine */
+        return;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 64-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        return;
+    }
+
+
+    reg_state[offset / 8] = value;
+}
+
+static void mailbox_reg_write(void *opaque, hwaddr offset, uint64_t value,
+                              unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    if (offset >= A_CXL_DEV_CMD_PAYLOAD) {
+        memcpy(cxl_dstate->mbox_reg_state + offset, &value, size);
+        return;
+    }
+
+    /*
+     * Lock is needed to prevent concurrent writes as well as to prevent writes
+     * coming in while the firmware is processing. Without background commands
+     * or the second mailbox implemented, this serves no purpose since the
+     * memory access is synchronized at a higher level (per memory region).
+     */
+    RCU_READ_LOCK_GUARD();
+
+    switch (size) {
+    case 4:
+        mailbox_mem_writel(cxl_dstate->mbox_reg_state32, offset, value);
+        break;
+    case 8:
+        mailbox_mem_writeq(cxl_dstate->mbox_reg_state64, offset, value);
+        break;
+    default:
+        g_assert_not_reached();
+    }
+
+    if (ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                         DOORBELL))
+        cxl_process_mailbox(cxl_dstate);
+}
+
+static const MemoryRegionOps mailbox_ops = {
+    .read = mailbox_reg_read,
+    .write = mailbox_reg_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 8,
+    },
+};
+
 static const MemoryRegionOps dev_ops = {
     .read = dev_reg_read,
     .write = NULL, /* status register is read only */
@@ -80,20 +185,33 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
                           "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);
     memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
                           "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+    memory_region_init_io(&cxl_dstate->mailbox, obj, &mailbox_ops, cxl_dstate,
+                          "mailbox", CXL_MAILBOX_REGISTERS_LENGTH);
 
     memory_region_add_subregion(&cxl_dstate->device_registers, 0,
                                 &cxl_dstate->caps);
     memory_region_add_subregion(&cxl_dstate->device_registers,
                                 CXL_DEVICE_REGISTERS_OFFSET,
                                 &cxl_dstate->device);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_MAILBOX_REGISTERS_OFFSET,
+                                &cxl_dstate->mailbox);
 }
 
 static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
 
+static void mailbox_reg_init_common(CXLDeviceState *cxl_dstate)
+{
+    /* 2048 payload size, with no interrupt or background support */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CAP,
+                     PAYLOAD_SIZE, CXL_MAILBOX_PAYLOAD_SHIFT);
+    cxl_dstate->payload_size = CXL_MAILBOX_MAX_PAYLOAD_SIZE;
+}
+
 void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 {
     uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
-    const int cap_count = 1;
+    const int cap_count = 2;
 
     /* CXL Device Capabilities Array Register */
     ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
@@ -102,4 +220,9 @@ void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 
     cxl_device_cap_init(cxl_dstate, DEVICE, 1);
     device_reg_init_common(cxl_dstate);
+
+    cxl_device_cap_init(cxl_dstate, MAILBOX, 2);
+    mailbox_reg_init_common(cxl_dstate);
+
+    assert(cxl_initialize_mailbox(cxl_dstate) == 0);
 }
hw/cxl/cxl-mailbox-utils.c new file +197 -0

197 lines not on this page · raw message

hw/cxl/meson.build +1 -0
index 47154d6850..0eca715d10 100644
@@ -1,4 +1,5 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
   'cxl-device-utils.c',
+  'cxl-mailbox-utils.c',
 ))
include/hw/cxl/cxl.h +3 -0
index 23f52c4cf9..362cda40de 100644
@@ -14,5 +14,8 @@
 #include "cxl_component.h"
 #include "cxl_device.h"
 
+#define COMPONENT_REG_BAR_IDX 0
+#define DEVICE_REG_BAR_IDX 2
+
 #endif
 
include/hw/cxl/cxl_device.h +24 -4
index f3bcf19410..af91bec10c 100644
@@ -80,16 +80,27 @@ typedef struct cxl_device_state {
     MemoryRegion device_registers;
 
     /* mmio for device capabilities array - 8.2.8.2 */
+    MemoryRegion device;
     struct {
         MemoryRegion caps;
         uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
     };
 
-    /* mmio for the device status registers 8.2.8.3 */
-    MemoryRegion device;
-
     /* mmio for the mailbox registers 8.2.8.4 */
-    MemoryRegion mailbox;
+    struct {
+        MemoryRegion mailbox;
+        uint16_t payload_size;
+        union {
+            uint8_t mbox_reg_state[CXL_MAILBOX_REGISTERS_LENGTH];
+            uint32_t mbox_reg_state32[CXL_MAILBOX_REGISTERS_LENGTH / 4];
+            uint64_t mbox_reg_state64[CXL_MAILBOX_REGISTERS_LENGTH / 8];
+        };
+        struct {
+            uint16_t opcode;
+            uint16_t effect;
+        } cel_log[1 << 16];
+        size_t cel_size;
+    };
 
     /* memory region for persistent memory, HDM */
     MemoryRegion *pmem;
@@ -135,6 +146,9 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate);
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate);
+
 #define cxl_device_cap_init(dstate, reg, cap_id)                                   \
     do {                                                                           \
         uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \
@@ -162,6 +176,12 @@ REG32(CXL_DEV_MAILBOX_CTRL, 4)
     FIELD(CXL_DEV_MAILBOX_CTRL, INT_EN, 1, 1)
     FIELD(CXL_DEV_MAILBOX_CTRL, BG_INT_EN, 2, 1)
 
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_MAILBOX_CMD, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND, 0, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND_SET, 8, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, LENGTH, 16, 20)
+
 /* XXX: actually a 64b register */
 REG32(CXL_DEV_MAILBOX_STS, 0x10)
     FIELD(CXL_DEV_MAILBOX_STS, BG_OP, 0, 1)
Signature · 1 line
2.30.0
#13 lore raw

Re: [RFC PATCH v3 05/31] hw/cxl/device: Implement basic mailbox (8.2.8.4)

Quoted 29 lines · On Mon, 1 Feb 2021 16:59:22 -0800 Ben Widawsky <[email protected]> wrote:

This is the beginning of implementing mailbox support for CXL 2.0 devices. The implementation recognizes when the doorbell is rung, handles the command/payload, clears the doorbell while returning error codes and data.

Generally the mailbox mechanism is designed to permit communication between the host OS and the firmware running on the device. For our purposes, we emulate both the firmware, implemented primarily in cxl-mailbox-utils.c, and the hardware.

No commands are implemented yet.

---
 hw/cxl/cxl-device-utils.c   | 125 ++++++++++++++++++++++-
 hw/cxl/cxl-mailbox-utils.c  | 197 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl.h        |   3 +
 include/hw/cxl/cxl_device.h |  28 ++++-
 5 files changed, 349 insertions(+), 5 deletions(-)
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
hw/cxl/cxl-device-utils.c +0 -0
index bb15ad9a0f..6602606f3d 100644
@@ -40,6 +40,111 @@ static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
     return 0;
 }

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 51

Quoted 10 lines
hw/cxl/cxl-device-utils.c +9 -0
 
+static uint64_t mailbox_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    switch (size) {
+    case 8:
+        return cxl_dstate->mbox_reg_state64[offset / 8];
+    case 4:
+        return cxl_dstate->mbox_reg_state32[offset / 4];

Numeric order seems more natural and I can't see a reason not to do it.
+ you do them in that order below.

Quoted 216 lines
hw/cxl/cxl-device-utils.c +115 -1
+    default:
+        g_assert_not_reached();
+    }
+}
+
+static void mailbox_mem_writel(uint32_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CTRL:
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_CAP:
+        /* RO register */
+        break;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 32-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        break;
+    }
+
+    reg_state[offset / 4] = value;
+}
+
+static void mailbox_mem_writeq(uint64_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CMD:
+        break;
+    case A_CXL_DEV_BG_CMD_STS:
+        /* BG not supported */
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_STS:
+        /* Read only register, will get updated by the state machine */
+        return;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 64-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        return;
+    }
+
+
+    reg_state[offset / 8] = value;
+}
+
+static void mailbox_reg_write(void *opaque, hwaddr offset, uint64_t value,
+                              unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    if (offset >= A_CXL_DEV_CMD_PAYLOAD) {
+        memcpy(cxl_dstate->mbox_reg_state + offset, &value, size);
+        return;
+    }
+
+    /*
+     * Lock is needed to prevent concurrent writes as well as to prevent writes
+     * coming in while the firmware is processing. Without background commands
+     * or the second mailbox implemented, this serves no purpose since the
+     * memory access is synchronized at a higher level (per memory region).
+     */
+    RCU_READ_LOCK_GUARD();
+
+    switch (size) {
+    case 4:
+        mailbox_mem_writel(cxl_dstate->mbox_reg_state32, offset, value);
+        break;
+    case 8:
+        mailbox_mem_writeq(cxl_dstate->mbox_reg_state64, offset, value);
+        break;
+    default:
+        g_assert_not_reached();
+    }
+
+    if (ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                         DOORBELL))
+        cxl_process_mailbox(cxl_dstate);
+}
+
+static const MemoryRegionOps mailbox_ops = {
+    .read = mailbox_reg_read,
+    .write = mailbox_reg_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 8,
+    },
+};
+
 static const MemoryRegionOps dev_ops = {
     .read = dev_reg_read,
     .write = NULL, /* status register is read only */
@@ -80,20 +185,33 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
                           "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);
     memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
                           "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+    memory_region_init_io(&cxl_dstate->mailbox, obj, &mailbox_ops, cxl_dstate,
+                          "mailbox", CXL_MAILBOX_REGISTERS_LENGTH);
 
     memory_region_add_subregion(&cxl_dstate->device_registers, 0,
                                 &cxl_dstate->caps);
     memory_region_add_subregion(&cxl_dstate->device_registers,
                                 CXL_DEVICE_REGISTERS_OFFSET,
                                 &cxl_dstate->device);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_MAILBOX_REGISTERS_OFFSET,
+                                &cxl_dstate->mailbox);
 }
 
 static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
 
+static void mailbox_reg_init_common(CXLDeviceState *cxl_dstate)
+{
+    /* 2048 payload size, with no interrupt or background support */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CAP,
+                     PAYLOAD_SIZE, CXL_MAILBOX_PAYLOAD_SHIFT);
+    cxl_dstate->payload_size = CXL_MAILBOX_MAX_PAYLOAD_SIZE;
+}
+
 void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 {
     uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
-    const int cap_count = 1;
+    const int cap_count = 2;
 
     /* CXL Device Capabilities Array Register */
     ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
@@ -102,4 +220,9 @@ void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 
     cxl_device_cap_init(cxl_dstate, DEVICE, 1);
     device_reg_init_common(cxl_dstate);
+
+    cxl_device_cap_init(cxl_dstate, MAILBOX, 2);
+    mailbox_reg_init_common(cxl_dstate);
+
+    assert(cxl_initialize_mailbox(cxl_dstate) == 0);
 }
hw/cxl/cxl-mailbox-utils.c new file +66 -0
new file mode 100644
index 0000000000..466055b01a
@@ -0,0 +1,197 @@
+/*
+ * CXL Utility library for mailbox interface
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "hw/cxl/cxl.h"
+#include "hw/pci/pci.h"
+#include "qemu/log.h"
+#include "qemu/uuid.h"
+
+/*
+ * How to add a new command, example. The command set FOO, with cmd BAR.
+ *  1. Add the command set and cmd to the enum.
+ *     FOO    = 0x7f,
+ *          #define BAR 0
+ *  2. Forward declare the handler.
+ *     declare_mailbox_handler(FOO_BAR);
+ *  3. Add the command to the cxl_cmd_set[][]
+ *     CXL_CMD(FOO, BAR, 0, 0),
+ *  4. Implement your handler
+ *     define_mailbox_handler(FOO_BAR) { ... return CXL_MBOX_SUCCESS; }
+ *
+ *
+ *  Writing the handler:
+ *    The handler will provide the &struct cxl_cmd, the &CXLDeviceState, and the
+ *    in/out length of the payload. The handler is responsible for consuming the
+ *    payload from cmd->payload and operating upon it as necessary. It must then
+ *    fill the output data into cmd->payload (overwriting what was there),
+ *    setting the length, and returning a valid return code.
+ *
+ *  XXX: The handler need not worry about endianess. The payload is read out of
+ *  a register interface that already deals with it.
+ */
+
+/* 8.2.8.4.5.1 Command Return Codes */
+typedef enum {
+    CXL_MBOX_SUCCESS = 0x0,
+    CXL_MBOX_BG_STARTED = 0x1,
+    CXL_MBOX_INVALID_INPUT = 0x2,
+    CXL_MBOX_UNSUPPORTED = 0x3,
+    CXL_MBOX_INTERNAL_ERROR = 0x4,
+    CXL_MBOX_RETRY_REQUIRED = 0x5,
+    CXL_MBOX_BUSY = 0x6,
+    CXL_MBOX_MEDIA_DISABLED = 0x7,
+    CXL_MBOX_FW_XFER_IN_PROGRESS = 0x8,
+    CXL_MBOX_FW_XFER_OUT_OF_ORDER = 0x9,
+    CXL_MBOX_FW_AUTH_FAILED = 0xa,
+    CXL_MBOX_FW_INVALID_SLOT = 0xb,
+    CXL_MBOX_FW_ROLLEDBACK = 0xc,
+    CXL_MBOX_FW_REST_REQD = 0xd,
+    CXL_MBOX_INVALID_HANDLE = 0xe,
+    CXL_MBOX_INVALID_PA = 0xf,
+    CXL_MBOX_INJECT_POISON_LIMIT = 0x10,
+    CXL_MBOX_PERMANENT_MEDIA_FAILURE = 0x11,
+    CXL_MBOX_ABORTED = 0x12,
+    CXL_MBOX_INVALID_SECURITY_STATE = 0x13,
+    CXL_MBOX_INCORRECT_PASSPHRASE = 0x14,
+    CXL_MBOX_UNSUPPORTED_MAILBOX = 0x15,
+    CXL_MBOX_INVALID_PAYLOAD_LENGTH = 0x16,
+    CXL_MBOX_MAX = 0x17
+} ret_code;

hw/cxl/cxl-mailbox-utils.c line 76

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+
+struct cxl_cmd;
+typedef ret_code (*opcode_handler)(struct cxl_cmd *cmd,
+                                   CXLDeviceState *cxl_dstate, uint16_t *len);
+struct cxl_cmd {
+    const char *name;
+    opcode_handler handler;
+    ssize_t in;
+    uint16_t effect; /* Reported in CEL */
+    uint8_t *payload;

This payload pointer feels somewhat out of place. Perhaps it should be a parameter passed to the opcode_handler()? The address of the payload doesn't feel like part of the command as such so you are justing using it as somewhere to stash the address when passing to the handler.

Quoted 59 lines
hw/cxl/cxl-mailbox-utils.c new file +59 -0
+};
+
+#define define_mailbox_handler(name)                \
+    static ret_code cmd_##name(struct cxl_cmd *cmd, \
+                               CXLDeviceState *cxl_dstate, uint16_t *len)
+#define declare_mailbox_handler(name) define_mailbox_handler(name)
+
+#define define_mailbox_handler_zeroed(name, size)                         \
+    uint16_t __zero##name = size;                                         \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = __zero##name;                                              \
+        memset(cmd->payload, 0, *len);                                    \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_const(name, data)                          \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = sizeof(data);                                              \
+        memcpy(cmd->payload, data, *len);                                 \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_nop(name)                                  \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+
+#define CXL_CMD(s, c, in, cel_effect) \
+    [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
+
+static struct cxl_cmd cxl_cmd_set[256][256] = {};
+
+#undef CXL_CMD
+
+QemuUUID cel_uuid;
+
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
+{
+    uint16_t ret = CXL_MBOX_SUCCESS;
+    struct cxl_cmd *cxl_cmd;
+    uint64_t status_reg;
+    opcode_handler h;
+
+    /*
+     * current state of mailbox interface
+     *  mbox_cap_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CAP];
+     *  mbox_ctrl_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CTRL];
+     *  status_reg = *(uint64_t *)&cxl_dstate->reg_state[A_CXL_DEV_MAILBOX_STS];
+     */
+    uint64_t command_reg =
+        *(uint64_t *)&cxl_dstate->mbox_reg_state[A_CXL_DEV_MAILBOX_CMD];
+
+    /* Check if we have to do anything */
+    if (!ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                          DOORBELL)) {

hw/cxl/cxl-mailbox-utils.c line 145

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+        qemu_log_mask(LOG_UNIMP, "Corrupt internal state for firmware\n");
+        return;
+    }
+
+    uint8_t set = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET);
+    uint8_t cmd = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND);
+    uint16_t len = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH);
+    cxl_cmd = &cxl_cmd_set[set][cmd];
+    h = cxl_cmd->handler;
+    if (!h) {

This path seems to not convey information it perhaps should. Maybe some feedback that a command was requested that isn't registered?

Quoted 29 lines
hw/cxl/cxl-mailbox-utils.c new file +29 -0
+        goto handled;
+    }
+
+    if (len != cxl_cmd->in) {
+        ret = CXL_MBOX_INVALID_PAYLOAD_LENGTH;
+    }
+
+    cxl_cmd->payload = cxl_dstate->mbox_reg_state + A_CXL_DEV_CMD_PAYLOAD;
+    ret = (*h)(cxl_cmd, cxl_dstate, &len);
+    assert(len <= cxl_dstate->payload_size);
+
+handled:
+    /*
+     * Set the return code
+     * XXX: it's a 64b register, but we're not setting the vendor, so we can get
+     * away with this
+     */
+    status_reg = FIELD_DP64(0, CXL_DEV_MAILBOX_STS, ERRNO, ret);
+
+    /*
+     * Set the return length
+     */
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH, len);
+
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_CMD / 8] = command_reg;
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_STS / 8] = status_reg;
+

hw/cxl/cxl-mailbox-utils.c line 184

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+    /* Tell the host we're done */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                     DOORBELL, 0);
+}
+
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate)
+{
+    const char *cel_uuidstr = "0da9c0b5-bf41-4b78-8f79-96b1623b3f17";
+
+    for (int i = 0; i < 256; i++) {

Instead of indexing with i and j, perhaps this would be more consistent using the naming you have above cmd and set?

Quoted 28 lines
hw/cxl/cxl-mailbox-utils.c new file +13 -0
+        for (int j = 0; j < 256; j++) {
+            if (cxl_cmd_set[i][j].handler) {
+                struct cxl_cmd *c = &cxl_cmd_set[i][j];
+
+                cxl_dstate->cel_log[cxl_dstate->cel_size].opcode = (i << 8) | j;
+                cxl_dstate->cel_log[cxl_dstate->cel_size].effect = c->effect;
+                cxl_dstate->cel_size++;
+            }
+        }
+    }
+
+    return qemu_uuid_parse(cel_uuidstr, &cel_uuid);
+}
hw/cxl/meson.build +1 -0
index 47154d6850..0eca715d10 100644
@@ -1,4 +1,5 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
   'cxl-device-utils.c',
+  'cxl-mailbox-utils.c',
 ))
include/hw/cxl/cxl.h +0 -0
index 23f52c4cf9..362cda40de 100644

include/hw/cxl/cxl.h line 18

Quoted 5 lines
include/hw/cxl/cxl.h +2 -0
@@ -14,5 +14,8 @@
 #include "cxl_component.h"
 #include "cxl_device.h"
 
+#define COMPONENT_REG_BAR_IDX 0
+#define DEVICE_REG_BAR_IDX 2

I'd argue for prefixing all defines

CXL_COMPONENT_REG_BAR_IDX etc

Will make it clear they are generic CXL related things.

Quoted 10 lines
include/hw/cxl/cxl.h +1 -0
+
 #endif
 
include/hw/cxl/cxl_device.h +0 -0
index f3bcf19410..af91bec10c 100644
@@ -80,16 +80,27 @@ typedef struct cxl_device_state {
     MemoryRegion device_registers;
 

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 90

Quoted 10 lines
include/hw/cxl/cxl_device.h +1 -3
     /* mmio for device capabilities array - 8.2.8.2 */
+    MemoryRegion device;
     struct {
         MemoryRegion caps;
         uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
     };
 
-    /* mmio for the device status registers 8.2.8.3 */
-    MemoryRegion device;
-

Move this block back to where it was originally introduced rather than introduce it then move it later?

Quoted 6 lines
include/hw/cxl/cxl_device.h +4 -1
     /* mmio for the mailbox registers 8.2.8.4 */
-    MemoryRegion mailbox;
+    struct {
+        MemoryRegion mailbox;
+        uint16_t payload_size;
+        union {

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 103

Quoted 10 lines
include/hw/cxl/cxl_device.h +10 -0
+            uint8_t mbox_reg_state[CXL_MAILBOX_REGISTERS_LENGTH];
+            uint32_t mbox_reg_state32[CXL_MAILBOX_REGISTERS_LENGTH / 4];
+            uint64_t mbox_reg_state64[CXL_MAILBOX_REGISTERS_LENGTH / 8];
+        };
+        struct {
+            uint16_t opcode;
+            uint16_t effect;
+        } cel_log[1 << 16];
+        size_t cel_size;
+    };

If the structure is unnamed, chances of a naming clash seem rather high if you don't prefix all the elements with mbx_ or something like that.

Quoted 14 lines
include/hw/cxl/cxl_device.h +3 -0
     /* memory region for persistent memory, HDM */
     MemoryRegion *pmem;
@@ -135,6 +146,9 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate);
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate);
+
 #define cxl_device_cap_init(dstate, reg, cap_id)                                   \
     do {                                                                           \
         uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \

include/hw/cxl/cxl_device.h REG32(CXL_DEV_MAILBOX_CTRL, 4) · line 184

Quoted 9 lines
include/hw/cxl/cxl_device.h +6 -0
@@ -162,6 +176,12 @@ REG32(CXL_DEV_MAILBOX_CTRL, 4)
     FIELD(CXL_DEV_MAILBOX_CTRL, INT_EN, 1, 1)
     FIELD(CXL_DEV_MAILBOX_CTRL, BG_INT_EN, 2, 1)
 
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_MAILBOX_CMD, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND, 0, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND_SET, 8, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, LENGTH, 16, 20)
+

Ah. this is where this definition went. Perhaps pull it back into patch 3? That patch defines plenty of other things that aren't used until later patches I think, so one more won't hurt and will save me asking why you skipped it:)

Quoted 3 lines
include/hw/cxl/cxl_device.h +0 -0
 /* XXX: actually a 64b register */
 REG32(CXL_DEV_MAILBOX_STS, 0x10)
     FIELD(CXL_DEV_MAILBOX_STS, BG_OP, 0, 1)
#14 lore raw

Re: [RFC PATCH v3 05/31] hw/cxl/device: Implement basic mailbox (8.2.8.4)

Quoted 33 lines · On Tue, 2 Feb 2021 14:58:30 +0000 Jonathan Cameron…
Quoted 30 lines · On Mon, 1 Feb 2021 16:59:22 -0800 Ben Widawsky <[email protected]> wrote:

This is the beginning of implementing mailbox support for CXL 2.0 devices. The implementation recognizes when the doorbell is rung, handles the command/payload, clears the doorbell while returning error codes and data.

Generally the mailbox mechanism is designed to permit communication between the host OS and the firmware running on the device. For our purposes, we emulate both the firmware, implemented primarily in cxl-mailbox-utils.c, and the hardware.

No commands are implemented yet.

---
 hw/cxl/cxl-device-utils.c   | 125 ++++++++++++++++++++++-
 hw/cxl/cxl-mailbox-utils.c  | 197 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl.h        |   3 +
 include/hw/cxl/cxl_device.h |  28 ++++-
 5 files changed, 349 insertions(+), 5 deletions(-)
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
hw/cxl/cxl-device-utils.c +0 -0
index bb15ad9a0f..6602606f3d 100644
@@ -40,6 +40,111 @@ static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
     return 0;
 }
 
Quoted 5 lines

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 47

Quoted 5 lines
hw/cxl/cxl-device-utils.c +5 -0
+static uint64_t mailbox_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    switch (size) {

As per the thread on the linux driver and the infinite loop I saw there as a result of doing 1 byte reads.

With the current setup of min_access_size = 4 and this function QEMU will helpfully issue a series of unaligned 4 byte reads to this function. It will then mask those down to 1 byte each and combine them. Given the integer division that results in the bottom byte of offset / 4 being returned up to 4 times.

To handle 2 and 1 byte reads we need explicit support in here and the MemoryRegionOps need to reflect that as well.

All the similar cases where such reads are allowed need to do the same.

Quoted 476 lines

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 51

Quoted 4 lines
hw/cxl/cxl-device-utils.c +4 -0
+    case 8:
+        return cxl_dstate->mbox_reg_state64[offset / 8];
+    case 4:
+        return cxl_dstate->mbox_reg_state32[offset / 4];  

Numeric order seems more natural and I can't see a reason not to do it.
+ you do them in that order below.

Quoted 216 lines
hw/cxl/cxl-device-utils.c +115 -1
+    default:
+        g_assert_not_reached();
+    }
+}
+
+static void mailbox_mem_writel(uint32_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CTRL:
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_CAP:
+        /* RO register */
+        break;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 32-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        break;
+    }
+
+    reg_state[offset / 4] = value;
+}
+
+static void mailbox_mem_writeq(uint64_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CMD:
+        break;
+    case A_CXL_DEV_BG_CMD_STS:
+        /* BG not supported */
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_STS:
+        /* Read only register, will get updated by the state machine */
+        return;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 64-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        return;
+    }
+
+
+    reg_state[offset / 8] = value;
+}
+
+static void mailbox_reg_write(void *opaque, hwaddr offset, uint64_t value,
+                              unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    if (offset >= A_CXL_DEV_CMD_PAYLOAD) {
+        memcpy(cxl_dstate->mbox_reg_state + offset, &value, size);
+        return;
+    }
+
+    /*
+     * Lock is needed to prevent concurrent writes as well as to prevent writes
+     * coming in while the firmware is processing. Without background commands
+     * or the second mailbox implemented, this serves no purpose since the
+     * memory access is synchronized at a higher level (per memory region).
+     */
+    RCU_READ_LOCK_GUARD();
+
+    switch (size) {
+    case 4:
+        mailbox_mem_writel(cxl_dstate->mbox_reg_state32, offset, value);
+        break;
+    case 8:
+        mailbox_mem_writeq(cxl_dstate->mbox_reg_state64, offset, value);
+        break;
+    default:
+        g_assert_not_reached();
+    }
+
+    if (ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                         DOORBELL))
+        cxl_process_mailbox(cxl_dstate);
+}
+
+static const MemoryRegionOps mailbox_ops = {
+    .read = mailbox_reg_read,
+    .write = mailbox_reg_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 8,
+    },
+};
+
 static const MemoryRegionOps dev_ops = {
     .read = dev_reg_read,
     .write = NULL, /* status register is read only */
@@ -80,20 +185,33 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
                           "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);
     memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
                           "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+    memory_region_init_io(&cxl_dstate->mailbox, obj, &mailbox_ops, cxl_dstate,
+                          "mailbox", CXL_MAILBOX_REGISTERS_LENGTH);
 
     memory_region_add_subregion(&cxl_dstate->device_registers, 0,
                                 &cxl_dstate->caps);
     memory_region_add_subregion(&cxl_dstate->device_registers,
                                 CXL_DEVICE_REGISTERS_OFFSET,
                                 &cxl_dstate->device);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_MAILBOX_REGISTERS_OFFSET,
+                                &cxl_dstate->mailbox);
 }
 
 static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
 
+static void mailbox_reg_init_common(CXLDeviceState *cxl_dstate)
+{
+    /* 2048 payload size, with no interrupt or background support */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CAP,
+                     PAYLOAD_SIZE, CXL_MAILBOX_PAYLOAD_SHIFT);
+    cxl_dstate->payload_size = CXL_MAILBOX_MAX_PAYLOAD_SIZE;
+}
+
 void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 {
     uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
-    const int cap_count = 1;
+    const int cap_count = 2;
 
     /* CXL Device Capabilities Array Register */
     ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
@@ -102,4 +220,9 @@ void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 
     cxl_device_cap_init(cxl_dstate, DEVICE, 1);
     device_reg_init_common(cxl_dstate);
+
+    cxl_device_cap_init(cxl_dstate, MAILBOX, 2);
+    mailbox_reg_init_common(cxl_dstate);
+
+    assert(cxl_initialize_mailbox(cxl_dstate) == 0);
 }
hw/cxl/cxl-mailbox-utils.c new file +66 -0
new file mode 100644
index 0000000000..466055b01a
@@ -0,0 +1,197 @@
+/*
+ * CXL Utility library for mailbox interface
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "hw/cxl/cxl.h"
+#include "hw/pci/pci.h"
+#include "qemu/log.h"
+#include "qemu/uuid.h"
+
+/*
+ * How to add a new command, example. The command set FOO, with cmd BAR.
+ *  1. Add the command set and cmd to the enum.
+ *     FOO    = 0x7f,
+ *          #define BAR 0
+ *  2. Forward declare the handler.
+ *     declare_mailbox_handler(FOO_BAR);
+ *  3. Add the command to the cxl_cmd_set[][]
+ *     CXL_CMD(FOO, BAR, 0, 0),
+ *  4. Implement your handler
+ *     define_mailbox_handler(FOO_BAR) { ... return CXL_MBOX_SUCCESS; }
+ *
+ *
+ *  Writing the handler:
+ *    The handler will provide the &struct cxl_cmd, the &CXLDeviceState, and the
+ *    in/out length of the payload. The handler is responsible for consuming the
+ *    payload from cmd->payload and operating upon it as necessary. It must then
+ *    fill the output data into cmd->payload (overwriting what was there),
+ *    setting the length, and returning a valid return code.
+ *
+ *  XXX: The handler need not worry about endianess. The payload is read out of
+ *  a register interface that already deals with it.
+ */
+
+/* 8.2.8.4.5.1 Command Return Codes */
+typedef enum {
+    CXL_MBOX_SUCCESS = 0x0,
+    CXL_MBOX_BG_STARTED = 0x1,
+    CXL_MBOX_INVALID_INPUT = 0x2,
+    CXL_MBOX_UNSUPPORTED = 0x3,
+    CXL_MBOX_INTERNAL_ERROR = 0x4,
+    CXL_MBOX_RETRY_REQUIRED = 0x5,
+    CXL_MBOX_BUSY = 0x6,
+    CXL_MBOX_MEDIA_DISABLED = 0x7,
+    CXL_MBOX_FW_XFER_IN_PROGRESS = 0x8,
+    CXL_MBOX_FW_XFER_OUT_OF_ORDER = 0x9,
+    CXL_MBOX_FW_AUTH_FAILED = 0xa,
+    CXL_MBOX_FW_INVALID_SLOT = 0xb,
+    CXL_MBOX_FW_ROLLEDBACK = 0xc,
+    CXL_MBOX_FW_REST_REQD = 0xd,
+    CXL_MBOX_INVALID_HANDLE = 0xe,
+    CXL_MBOX_INVALID_PA = 0xf,
+    CXL_MBOX_INJECT_POISON_LIMIT = 0x10,
+    CXL_MBOX_PERMANENT_MEDIA_FAILURE = 0x11,
+    CXL_MBOX_ABORTED = 0x12,
+    CXL_MBOX_INVALID_SECURITY_STATE = 0x13,
+    CXL_MBOX_INCORRECT_PASSPHRASE = 0x14,
+    CXL_MBOX_UNSUPPORTED_MAILBOX = 0x15,
+    CXL_MBOX_INVALID_PAYLOAD_LENGTH = 0x16,
+    CXL_MBOX_MAX = 0x17
+} ret_code;

hw/cxl/cxl-mailbox-utils.c line 76

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+
+struct cxl_cmd;
+typedef ret_code (*opcode_handler)(struct cxl_cmd *cmd,
+                                   CXLDeviceState *cxl_dstate, uint16_t *len);
+struct cxl_cmd {
+    const char *name;
+    opcode_handler handler;
+    ssize_t in;
+    uint16_t effect; /* Reported in CEL */
+    uint8_t *payload;  

This payload pointer feels somewhat out of place. Perhaps it should be a parameter passed to the opcode_handler()? The address of the payload doesn't feel like part of the command as such so you are justing using it as somewhere to stash the address when passing to the handler.

Quoted 59 lines
hw/cxl/cxl-mailbox-utils.c new file +59 -0
+};
+
+#define define_mailbox_handler(name)                \
+    static ret_code cmd_##name(struct cxl_cmd *cmd, \
+                               CXLDeviceState *cxl_dstate, uint16_t *len)
+#define declare_mailbox_handler(name) define_mailbox_handler(name)
+
+#define define_mailbox_handler_zeroed(name, size)                         \
+    uint16_t __zero##name = size;                                         \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = __zero##name;                                              \
+        memset(cmd->payload, 0, *len);                                    \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_const(name, data)                          \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = sizeof(data);                                              \
+        memcpy(cmd->payload, data, *len);                                 \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_nop(name)                                  \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+
+#define CXL_CMD(s, c, in, cel_effect) \
+    [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
+
+static struct cxl_cmd cxl_cmd_set[256][256] = {};
+
+#undef CXL_CMD
+
+QemuUUID cel_uuid;
+
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
+{
+    uint16_t ret = CXL_MBOX_SUCCESS;
+    struct cxl_cmd *cxl_cmd;
+    uint64_t status_reg;
+    opcode_handler h;
+
+    /*
+     * current state of mailbox interface
+     *  mbox_cap_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CAP];
+     *  mbox_ctrl_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CTRL];
+     *  status_reg = *(uint64_t *)&cxl_dstate->reg_state[A_CXL_DEV_MAILBOX_STS];
+     */
+    uint64_t command_reg =
+        *(uint64_t *)&cxl_dstate->mbox_reg_state[A_CXL_DEV_MAILBOX_CMD];
+
+    /* Check if we have to do anything */
+    if (!ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                          DOORBELL)) {

hw/cxl/cxl-mailbox-utils.c line 145

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+        qemu_log_mask(LOG_UNIMP, "Corrupt internal state for firmware\n");
+        return;
+    }
+
+    uint8_t set = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET);
+    uint8_t cmd = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND);
+    uint16_t len = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH);
+    cxl_cmd = &cxl_cmd_set[set][cmd];
+    h = cxl_cmd->handler;
+    if (!h) {  

This path seems to not convey information it perhaps should. Maybe some feedback that a command was requested that isn't registered?

Quoted 29 lines
hw/cxl/cxl-mailbox-utils.c new file +29 -0
+        goto handled;
+    }
+
+    if (len != cxl_cmd->in) {
+        ret = CXL_MBOX_INVALID_PAYLOAD_LENGTH;
+    }
+
+    cxl_cmd->payload = cxl_dstate->mbox_reg_state + A_CXL_DEV_CMD_PAYLOAD;
+    ret = (*h)(cxl_cmd, cxl_dstate, &len);
+    assert(len <= cxl_dstate->payload_size);
+
+handled:
+    /*
+     * Set the return code
+     * XXX: it's a 64b register, but we're not setting the vendor, so we can get
+     * away with this
+     */
+    status_reg = FIELD_DP64(0, CXL_DEV_MAILBOX_STS, ERRNO, ret);
+
+    /*
+     * Set the return length
+     */
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH, len);
+
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_CMD / 8] = command_reg;
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_STS / 8] = status_reg;
+

hw/cxl/cxl-mailbox-utils.c line 184

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+    /* Tell the host we're done */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                     DOORBELL, 0);
+}
+
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate)
+{
+    const char *cel_uuidstr = "0da9c0b5-bf41-4b78-8f79-96b1623b3f17";
+
+    for (int i = 0; i < 256; i++) {  

Instead of indexing with i and j, perhaps this would be more consistent using the naming you have above cmd and set?

Quoted 28 lines
hw/cxl/cxl-mailbox-utils.c new file +13 -0
+        for (int j = 0; j < 256; j++) {
+            if (cxl_cmd_set[i][j].handler) {
+                struct cxl_cmd *c = &cxl_cmd_set[i][j];
+
+                cxl_dstate->cel_log[cxl_dstate->cel_size].opcode = (i << 8) | j;
+                cxl_dstate->cel_log[cxl_dstate->cel_size].effect = c->effect;
+                cxl_dstate->cel_size++;
+            }
+        }
+    }
+
+    return qemu_uuid_parse(cel_uuidstr, &cel_uuid);
+}
hw/cxl/meson.build +1 -0
index 47154d6850..0eca715d10 100644
@@ -1,4 +1,5 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
   'cxl-device-utils.c',
+  'cxl-mailbox-utils.c',
 ))
include/hw/cxl/cxl.h +0 -0
index 23f52c4cf9..362cda40de 100644

include/hw/cxl/cxl.h line 18

Quoted 5 lines
include/hw/cxl/cxl.h +2 -0
@@ -14,5 +14,8 @@
 #include "cxl_component.h"
 #include "cxl_device.h"
 
+#define COMPONENT_REG_BAR_IDX 0
+#define DEVICE_REG_BAR_IDX 2  

I'd argue for prefixing all defines

CXL_COMPONENT_REG_BAR_IDX etc

Will make it clear they are generic CXL related things.

Quoted 10 lines
include/hw/cxl/cxl.h +1 -0
+
 #endif
 
include/hw/cxl/cxl_device.h +0 -0
index f3bcf19410..af91bec10c 100644
@@ -80,16 +80,27 @@ typedef struct cxl_device_state {
     MemoryRegion device_registers;
 

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 90

Quoted 10 lines
include/hw/cxl/cxl_device.h +1 -3
     /* mmio for device capabilities array - 8.2.8.2 */
+    MemoryRegion device;
     struct {
         MemoryRegion caps;
         uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
     };
 
-    /* mmio for the device status registers 8.2.8.3 */
-    MemoryRegion device;
-  

Move this block back to where it was originally introduced rather than introduce it then move it later?

Quoted 6 lines
include/hw/cxl/cxl_device.h +4 -1
     /* mmio for the mailbox registers 8.2.8.4 */
-    MemoryRegion mailbox;
+    struct {
+        MemoryRegion mailbox;
+        uint16_t payload_size;
+        union {

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 103

Quoted 10 lines
include/hw/cxl/cxl_device.h +10 -0
+            uint8_t mbox_reg_state[CXL_MAILBOX_REGISTERS_LENGTH];
+            uint32_t mbox_reg_state32[CXL_MAILBOX_REGISTERS_LENGTH / 4];
+            uint64_t mbox_reg_state64[CXL_MAILBOX_REGISTERS_LENGTH / 8];
+        };
+        struct {
+            uint16_t opcode;
+            uint16_t effect;
+        } cel_log[1 << 16];
+        size_t cel_size;
+    };  

If the structure is unnamed, chances of a naming clash seem rather high if you don't prefix all the elements with mbx_ or something like that.

Quoted 14 lines
include/hw/cxl/cxl_device.h +3 -0
     /* memory region for persistent memory, HDM */
     MemoryRegion *pmem;
@@ -135,6 +146,9 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate);
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate);
+
 #define cxl_device_cap_init(dstate, reg, cap_id)                                   \
     do {                                                                           \
         uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \

include/hw/cxl/cxl_device.h REG32(CXL_DEV_MAILBOX_CTRL, 4) · line 184

Quoted 9 lines
include/hw/cxl/cxl_device.h +6 -0
@@ -162,6 +176,12 @@ REG32(CXL_DEV_MAILBOX_CTRL, 4)
     FIELD(CXL_DEV_MAILBOX_CTRL, INT_EN, 1, 1)
     FIELD(CXL_DEV_MAILBOX_CTRL, BG_INT_EN, 2, 1)
 
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_MAILBOX_CMD, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND, 0, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND_SET, 8, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, LENGTH, 16, 20)
+  

Ah. this is where this definition went. Perhaps pull it back into patch 3? That patch defines plenty of other things that aren't used until later patches I think, so one more won't hurt and will save me asking why you skipped it:)

Quoted 3 lines
include/hw/cxl/cxl_device.h +0 -0
 /* XXX: actually a 64b register */
 REG32(CXL_DEV_MAILBOX_STS, 0x10)
     FIELD(CXL_DEV_MAILBOX_STS, BG_OP, 0, 1)  
#15 lore raw

Re: [RFC PATCH v3 05/31] hw/cxl/device: Implement basic mailbox (8.2.8.4)

Quoted 51 lines · Jonathan Cameron
Quoted 33 lines · On Tue, 2 Feb 2021 14:58:30 +0000 Jonathan Cameron…
Quoted 30 lines · On Mon, 1 Feb 2021 16:59:22 -0800 Ben Widawsky <[email protected]> wrote:

This is the beginning of implementing mailbox support for CXL 2.0 devices. The implementation recognizes when the doorbell is rung, handles the command/payload, clears the doorbell while returning error codes and data.

Generally the mailbox mechanism is designed to permit communication between the host OS and the firmware running on the device. For our purposes, we emulate both the firmware, implemented primarily in cxl-mailbox-utils.c, and the hardware.

No commands are implemented yet.

---
 hw/cxl/cxl-device-utils.c   | 125 ++++++++++++++++++++++-
 hw/cxl/cxl-mailbox-utils.c  | 197 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl.h        |   3 +
 include/hw/cxl/cxl_device.h |  28 ++++-
 5 files changed, 349 insertions(+), 5 deletions(-)
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
hw/cxl/cxl-device-utils.c +0 -0
index bb15ad9a0f..6602606f3d 100644
@@ -40,6 +40,111 @@ static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
     return 0;
 }
 
Quoted 5 lines

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 47

Quoted 5 lines
hw/cxl/cxl-device-utils.c +5 -0
+static uint64_t mailbox_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    switch (size) {

As per the thread on the linux driver and the infinite loop I saw there as a result of doing 1 byte reads.

With the current setup of min_access_size = 4 and this function QEMU will helpfully issue a series of unaligned 4 byte reads to this function. It will then mask those down to 1 byte each and combine them. Given the integer division that results in the bottom byte of offset / 4 being returned up to 4 times.

Quoted 6 lines

To handle 2 and 1 byte reads we need explicit support in here and the MemoryRegionOps need to reflect that as well.

All the similar cases where such reads are allowed need to do the same.

From the documentation (memory.rst)
- .impl.min_access_size, .impl.max_access_size define the access sizes (in bytes) supported by the *implementation*; other access sizes will be emulated using the ones available. For example a 4-byte write will be emulated using four 1-byte writes, if .impl.max_access_size = 1.

I'm intentionally not looking at the implementation. As I read this, the behavior you describe is either a QEMU bug, or poor documentation.

Am I missing something?

Quoted 478 lines
Quoted 476 lines

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 51

Quoted 4 lines
hw/cxl/cxl-device-utils.c +4 -0
+    case 8:
+        return cxl_dstate->mbox_reg_state64[offset / 8];
+    case 4:
+        return cxl_dstate->mbox_reg_state32[offset / 4];  

Numeric order seems more natural and I can't see a reason not to do it.
+ you do them in that order below.

Quoted 216 lines
hw/cxl/cxl-device-utils.c +115 -1
+    default:
+        g_assert_not_reached();
+    }
+}
+
+static void mailbox_mem_writel(uint32_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CTRL:
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_CAP:
+        /* RO register */
+        break;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 32-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        break;
+    }
+
+    reg_state[offset / 4] = value;
+}
+
+static void mailbox_mem_writeq(uint64_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CMD:
+        break;
+    case A_CXL_DEV_BG_CMD_STS:
+        /* BG not supported */
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_STS:
+        /* Read only register, will get updated by the state machine */
+        return;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 64-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        return;
+    }
+
+
+    reg_state[offset / 8] = value;
+}
+
+static void mailbox_reg_write(void *opaque, hwaddr offset, uint64_t value,
+                              unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    if (offset >= A_CXL_DEV_CMD_PAYLOAD) {
+        memcpy(cxl_dstate->mbox_reg_state + offset, &value, size);
+        return;
+    }
+
+    /*
+     * Lock is needed to prevent concurrent writes as well as to prevent writes
+     * coming in while the firmware is processing. Without background commands
+     * or the second mailbox implemented, this serves no purpose since the
+     * memory access is synchronized at a higher level (per memory region).
+     */
+    RCU_READ_LOCK_GUARD();
+
+    switch (size) {
+    case 4:
+        mailbox_mem_writel(cxl_dstate->mbox_reg_state32, offset, value);
+        break;
+    case 8:
+        mailbox_mem_writeq(cxl_dstate->mbox_reg_state64, offset, value);
+        break;
+    default:
+        g_assert_not_reached();
+    }
+
+    if (ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                         DOORBELL))
+        cxl_process_mailbox(cxl_dstate);
+}
+
+static const MemoryRegionOps mailbox_ops = {
+    .read = mailbox_reg_read,
+    .write = mailbox_reg_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 8,
+    },
+};
+
 static const MemoryRegionOps dev_ops = {
     .read = dev_reg_read,
     .write = NULL, /* status register is read only */
@@ -80,20 +185,33 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
                           "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);
     memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
                           "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+    memory_region_init_io(&cxl_dstate->mailbox, obj, &mailbox_ops, cxl_dstate,
+                          "mailbox", CXL_MAILBOX_REGISTERS_LENGTH);
 
     memory_region_add_subregion(&cxl_dstate->device_registers, 0,
                                 &cxl_dstate->caps);
     memory_region_add_subregion(&cxl_dstate->device_registers,
                                 CXL_DEVICE_REGISTERS_OFFSET,
                                 &cxl_dstate->device);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_MAILBOX_REGISTERS_OFFSET,
+                                &cxl_dstate->mailbox);
 }
 
 static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
 
+static void mailbox_reg_init_common(CXLDeviceState *cxl_dstate)
+{
+    /* 2048 payload size, with no interrupt or background support */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CAP,
+                     PAYLOAD_SIZE, CXL_MAILBOX_PAYLOAD_SHIFT);
+    cxl_dstate->payload_size = CXL_MAILBOX_MAX_PAYLOAD_SIZE;
+}
+
 void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 {
     uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
-    const int cap_count = 1;
+    const int cap_count = 2;
 
     /* CXL Device Capabilities Array Register */
     ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
@@ -102,4 +220,9 @@ void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 
     cxl_device_cap_init(cxl_dstate, DEVICE, 1);
     device_reg_init_common(cxl_dstate);
+
+    cxl_device_cap_init(cxl_dstate, MAILBOX, 2);
+    mailbox_reg_init_common(cxl_dstate);
+
+    assert(cxl_initialize_mailbox(cxl_dstate) == 0);
 }
hw/cxl/cxl-mailbox-utils.c new file +66 -0
new file mode 100644
index 0000000000..466055b01a
@@ -0,0 +1,197 @@
+/*
+ * CXL Utility library for mailbox interface
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "hw/cxl/cxl.h"
+#include "hw/pci/pci.h"
+#include "qemu/log.h"
+#include "qemu/uuid.h"
+
+/*
+ * How to add a new command, example. The command set FOO, with cmd BAR.
+ *  1. Add the command set and cmd to the enum.
+ *     FOO    = 0x7f,
+ *          #define BAR 0
+ *  2. Forward declare the handler.
+ *     declare_mailbox_handler(FOO_BAR);
+ *  3. Add the command to the cxl_cmd_set[][]
+ *     CXL_CMD(FOO, BAR, 0, 0),
+ *  4. Implement your handler
+ *     define_mailbox_handler(FOO_BAR) { ... return CXL_MBOX_SUCCESS; }
+ *
+ *
+ *  Writing the handler:
+ *    The handler will provide the &struct cxl_cmd, the &CXLDeviceState, and the
+ *    in/out length of the payload. The handler is responsible for consuming the
+ *    payload from cmd->payload and operating upon it as necessary. It must then
+ *    fill the output data into cmd->payload (overwriting what was there),
+ *    setting the length, and returning a valid return code.
+ *
+ *  XXX: The handler need not worry about endianess. The payload is read out of
+ *  a register interface that already deals with it.
+ */
+
+/* 8.2.8.4.5.1 Command Return Codes */
+typedef enum {
+    CXL_MBOX_SUCCESS = 0x0,
+    CXL_MBOX_BG_STARTED = 0x1,
+    CXL_MBOX_INVALID_INPUT = 0x2,
+    CXL_MBOX_UNSUPPORTED = 0x3,
+    CXL_MBOX_INTERNAL_ERROR = 0x4,
+    CXL_MBOX_RETRY_REQUIRED = 0x5,
+    CXL_MBOX_BUSY = 0x6,
+    CXL_MBOX_MEDIA_DISABLED = 0x7,
+    CXL_MBOX_FW_XFER_IN_PROGRESS = 0x8,
+    CXL_MBOX_FW_XFER_OUT_OF_ORDER = 0x9,
+    CXL_MBOX_FW_AUTH_FAILED = 0xa,
+    CXL_MBOX_FW_INVALID_SLOT = 0xb,
+    CXL_MBOX_FW_ROLLEDBACK = 0xc,
+    CXL_MBOX_FW_REST_REQD = 0xd,
+    CXL_MBOX_INVALID_HANDLE = 0xe,
+    CXL_MBOX_INVALID_PA = 0xf,
+    CXL_MBOX_INJECT_POISON_LIMIT = 0x10,
+    CXL_MBOX_PERMANENT_MEDIA_FAILURE = 0x11,
+    CXL_MBOX_ABORTED = 0x12,
+    CXL_MBOX_INVALID_SECURITY_STATE = 0x13,
+    CXL_MBOX_INCORRECT_PASSPHRASE = 0x14,
+    CXL_MBOX_UNSUPPORTED_MAILBOX = 0x15,
+    CXL_MBOX_INVALID_PAYLOAD_LENGTH = 0x16,
+    CXL_MBOX_MAX = 0x17
+} ret_code;

hw/cxl/cxl-mailbox-utils.c line 76

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+
+struct cxl_cmd;
+typedef ret_code (*opcode_handler)(struct cxl_cmd *cmd,
+                                   CXLDeviceState *cxl_dstate, uint16_t *len);
+struct cxl_cmd {
+    const char *name;
+    opcode_handler handler;
+    ssize_t in;
+    uint16_t effect; /* Reported in CEL */
+    uint8_t *payload;  

This payload pointer feels somewhat out of place. Perhaps it should be a parameter passed to the opcode_handler()? The address of the payload doesn't feel like part of the command as such so you are justing using it as somewhere to stash the address when passing to the handler.

Quoted 59 lines
hw/cxl/cxl-mailbox-utils.c new file +59 -0
+};
+
+#define define_mailbox_handler(name)                \
+    static ret_code cmd_##name(struct cxl_cmd *cmd, \
+                               CXLDeviceState *cxl_dstate, uint16_t *len)
+#define declare_mailbox_handler(name) define_mailbox_handler(name)
+
+#define define_mailbox_handler_zeroed(name, size)                         \
+    uint16_t __zero##name = size;                                         \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = __zero##name;                                              \
+        memset(cmd->payload, 0, *len);                                    \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_const(name, data)                          \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = sizeof(data);                                              \
+        memcpy(cmd->payload, data, *len);                                 \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_nop(name)                                  \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+
+#define CXL_CMD(s, c, in, cel_effect) \
+    [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
+
+static struct cxl_cmd cxl_cmd_set[256][256] = {};
+
+#undef CXL_CMD
+
+QemuUUID cel_uuid;
+
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
+{
+    uint16_t ret = CXL_MBOX_SUCCESS;
+    struct cxl_cmd *cxl_cmd;
+    uint64_t status_reg;
+    opcode_handler h;
+
+    /*
+     * current state of mailbox interface
+     *  mbox_cap_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CAP];
+     *  mbox_ctrl_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CTRL];
+     *  status_reg = *(uint64_t *)&cxl_dstate->reg_state[A_CXL_DEV_MAILBOX_STS];
+     */
+    uint64_t command_reg =
+        *(uint64_t *)&cxl_dstate->mbox_reg_state[A_CXL_DEV_MAILBOX_CMD];
+
+    /* Check if we have to do anything */
+    if (!ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                          DOORBELL)) {

hw/cxl/cxl-mailbox-utils.c line 145

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+        qemu_log_mask(LOG_UNIMP, "Corrupt internal state for firmware\n");
+        return;
+    }
+
+    uint8_t set = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET);
+    uint8_t cmd = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND);
+    uint16_t len = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH);
+    cxl_cmd = &cxl_cmd_set[set][cmd];
+    h = cxl_cmd->handler;
+    if (!h) {  

This path seems to not convey information it perhaps should. Maybe some feedback that a command was requested that isn't registered?

Quoted 29 lines
hw/cxl/cxl-mailbox-utils.c new file +29 -0
+        goto handled;
+    }
+
+    if (len != cxl_cmd->in) {
+        ret = CXL_MBOX_INVALID_PAYLOAD_LENGTH;
+    }
+
+    cxl_cmd->payload = cxl_dstate->mbox_reg_state + A_CXL_DEV_CMD_PAYLOAD;
+    ret = (*h)(cxl_cmd, cxl_dstate, &len);
+    assert(len <= cxl_dstate->payload_size);
+
+handled:
+    /*
+     * Set the return code
+     * XXX: it's a 64b register, but we're not setting the vendor, so we can get
+     * away with this
+     */
+    status_reg = FIELD_DP64(0, CXL_DEV_MAILBOX_STS, ERRNO, ret);
+
+    /*
+     * Set the return length
+     */
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH, len);
+
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_CMD / 8] = command_reg;
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_STS / 8] = status_reg;
+

hw/cxl/cxl-mailbox-utils.c line 184

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+    /* Tell the host we're done */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                     DOORBELL, 0);
+}
+
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate)
+{
+    const char *cel_uuidstr = "0da9c0b5-bf41-4b78-8f79-96b1623b3f17";
+
+    for (int i = 0; i < 256; i++) {  

Instead of indexing with i and j, perhaps this would be more consistent using the naming you have above cmd and set?

Quoted 28 lines
hw/cxl/cxl-mailbox-utils.c new file +13 -0
+        for (int j = 0; j < 256; j++) {
+            if (cxl_cmd_set[i][j].handler) {
+                struct cxl_cmd *c = &cxl_cmd_set[i][j];
+
+                cxl_dstate->cel_log[cxl_dstate->cel_size].opcode = (i << 8) | j;
+                cxl_dstate->cel_log[cxl_dstate->cel_size].effect = c->effect;
+                cxl_dstate->cel_size++;
+            }
+        }
+    }
+
+    return qemu_uuid_parse(cel_uuidstr, &cel_uuid);
+}
hw/cxl/meson.build +1 -0
index 47154d6850..0eca715d10 100644
@@ -1,4 +1,5 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
   'cxl-device-utils.c',
+  'cxl-mailbox-utils.c',
 ))
include/hw/cxl/cxl.h +0 -0
index 23f52c4cf9..362cda40de 100644

include/hw/cxl/cxl.h line 18

Quoted 5 lines
include/hw/cxl/cxl.h +2 -0
@@ -14,5 +14,8 @@
 #include "cxl_component.h"
 #include "cxl_device.h"
 
+#define COMPONENT_REG_BAR_IDX 0
+#define DEVICE_REG_BAR_IDX 2  

I'd argue for prefixing all defines

CXL_COMPONENT_REG_BAR_IDX etc

Will make it clear they are generic CXL related things.

Quoted 10 lines
include/hw/cxl/cxl.h +1 -0
+
 #endif
 
include/hw/cxl/cxl_device.h +0 -0
index f3bcf19410..af91bec10c 100644
@@ -80,16 +80,27 @@ typedef struct cxl_device_state {
     MemoryRegion device_registers;
 

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 90

Quoted 10 lines
include/hw/cxl/cxl_device.h +1 -3
     /* mmio for device capabilities array - 8.2.8.2 */
+    MemoryRegion device;
     struct {
         MemoryRegion caps;
         uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
     };
 
-    /* mmio for the device status registers 8.2.8.3 */
-    MemoryRegion device;
-  

Move this block back to where it was originally introduced rather than introduce it then move it later?

Quoted 6 lines
include/hw/cxl/cxl_device.h +4 -1
     /* mmio for the mailbox registers 8.2.8.4 */
-    MemoryRegion mailbox;
+    struct {
+        MemoryRegion mailbox;
+        uint16_t payload_size;
+        union {

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 103

Quoted 10 lines
include/hw/cxl/cxl_device.h +10 -0
+            uint8_t mbox_reg_state[CXL_MAILBOX_REGISTERS_LENGTH];
+            uint32_t mbox_reg_state32[CXL_MAILBOX_REGISTERS_LENGTH / 4];
+            uint64_t mbox_reg_state64[CXL_MAILBOX_REGISTERS_LENGTH / 8];
+        };
+        struct {
+            uint16_t opcode;
+            uint16_t effect;
+        } cel_log[1 << 16];
+        size_t cel_size;
+    };  

If the structure is unnamed, chances of a naming clash seem rather high if you don't prefix all the elements with mbx_ or something like that.

Quoted 14 lines
include/hw/cxl/cxl_device.h +3 -0
     /* memory region for persistent memory, HDM */
     MemoryRegion *pmem;
@@ -135,6 +146,9 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate);
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate);
+
 #define cxl_device_cap_init(dstate, reg, cap_id)                                   \
     do {                                                                           \
         uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \

include/hw/cxl/cxl_device.h REG32(CXL_DEV_MAILBOX_CTRL, 4) · line 184

Quoted 9 lines
include/hw/cxl/cxl_device.h +6 -0
@@ -162,6 +176,12 @@ REG32(CXL_DEV_MAILBOX_CTRL, 4)
     FIELD(CXL_DEV_MAILBOX_CTRL, INT_EN, 1, 1)
     FIELD(CXL_DEV_MAILBOX_CTRL, BG_INT_EN, 2, 1)
 
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_MAILBOX_CMD, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND, 0, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND_SET, 8, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, LENGTH, 16, 20)
+  

Ah. this is where this definition went. Perhaps pull it back into patch 3? That patch defines plenty of other things that aren't used until later patches I think, so one more won't hurt and will save me asking why you skipped it:)

Quoted 3 lines
include/hw/cxl/cxl_device.h +0 -0
 /* XXX: actually a 64b register */
 REG32(CXL_DEV_MAILBOX_STS, 0x10)
     FIELD(CXL_DEV_MAILBOX_STS, BG_OP, 0, 1)  
#16 lore raw

Re: [RFC PATCH v3 05/31] hw/cxl/device: Implement basic mailbox (8.2.8.4)

Quoted 65 lines · On Wed, 17 Feb 2021 16:55:14 -0800 Ben Widawsky <[email protected]> wrote:
Quoted 51 lines · Jonathan Cameron
Quoted 33 lines · On Tue, 2 Feb 2021 14:58:30 +0000 Jonathan Cameron…
Quoted 30 lines · On Mon, 1 Feb 2021 16:59:22 -0800 Ben Widawsky <[email protected]> wrote:

This is the beginning of implementing mailbox support for CXL 2.0 devices. The implementation recognizes when the doorbell is rung, handles the command/payload, clears the doorbell while returning error codes and data.

Generally the mailbox mechanism is designed to permit communication between the host OS and the firmware running on the device. For our purposes, we emulate both the firmware, implemented primarily in cxl-mailbox-utils.c, and the hardware.

No commands are implemented yet.

---
 hw/cxl/cxl-device-utils.c   | 125 ++++++++++++++++++++++-
 hw/cxl/cxl-mailbox-utils.c  | 197 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl.h        |   3 +
 include/hw/cxl/cxl_device.h |  28 ++++-
 5 files changed, 349 insertions(+), 5 deletions(-)
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
hw/cxl/cxl-device-utils.c +0 -0
index bb15ad9a0f..6602606f3d 100644
@@ -40,6 +40,111 @@ static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
     return 0;
 }
 
Quoted 5 lines

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 47

Quoted 5 lines
hw/cxl/cxl-device-utils.c +5 -0
+static uint64_t mailbox_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    switch (size) {  

As per the thread on the linux driver and the infinite loop I saw there as a result of doing 1 byte reads.

With the current setup of min_access_size = 4 and this function QEMU will helpfully issue a series of unaligned 4 byte reads to this function. It will then mask those down to 1 byte each and combine them. Given the integer division that results in the bottom byte of offset / 4 being returned up to 4 times.

Quoted 6 lines

To handle 2 and 1 byte reads we need explicit support in here and the MemoryRegionOps need to reflect that as well.

All the similar cases where such reads are allowed need to do the same.

From the documentation (memory.rst)
- .impl.min_access_size, .impl.max_access_size define the access sizes (in bytes) supported by the *implementation*; other access sizes will be emulated using the ones available. For example a 4-byte write will be emulated using four 1-byte writes, if .impl.max_access_size = 1.

Quoted 4 lines

I'm intentionally not looking at the implementation. As I read this, the behavior you describe is either a QEMU bug, or poor documentation.

Am I missing something?

Good question...

For the case of emulating a larger access than supported it seems to be fine

Code is implemented in softmmu/memory.c as something like.

for (i = 0; i < size; i+= access_size) {
    r |= access_fn(mr, addr + i, value, access_size,
		   i * 8, access_mask, attrs);
}

For smaller read than supported, memory_region_shift_read_access() looks like it would do the masking but the mask and shift passed to the function are wrong. There is a fixme in the code referring to unaligned accesses, but these are aligned.

It's fairly easy to make this work, but unexpected side effects worry me a little + my stab at big endian support may be completely wrong.

I'll mess around with this a bit more then send out something to see how people would like to fix this issue
(either the docs, or the code).

Jonathan

Quoted 479 lines
Quoted 479 lines
Quoted 477 lines

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 51

Quoted 5 lines
hw/cxl/cxl-device-utils.c +4 -0
+    case 8:
+        return cxl_dstate->mbox_reg_state64[offset / 8];
+    case 4:
+        return cxl_dstate->mbox_reg_state32[offset / 4];    

Numeric order seems more natural and I can't see a reason not to do it.
+ you do them in that order below.

Quoted 216 lines
hw/cxl/cxl-device-utils.c +115 -1
+    default:
+        g_assert_not_reached();
+    }
+}
+
+static void mailbox_mem_writel(uint32_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CTRL:
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_CAP:
+        /* RO register */
+        break;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 32-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        break;
+    }
+
+    reg_state[offset / 4] = value;
+}
+
+static void mailbox_mem_writeq(uint64_t *reg_state, hwaddr offset,
+                               uint64_t value)
+{
+    switch (offset) {
+    case A_CXL_DEV_MAILBOX_CMD:
+        break;
+    case A_CXL_DEV_BG_CMD_STS:
+        /* BG not supported */
+        /* fallthrough */
+    case A_CXL_DEV_MAILBOX_STS:
+        /* Read only register, will get updated by the state machine */
+        return;
+    default:
+        qemu_log_mask(LOG_UNIMP,
+                      "%s Unexpected 64-bit access to 0x%" PRIx64 " (WI)\n",
+                      __func__, offset);
+        return;
+    }
+
+
+    reg_state[offset / 8] = value;
+}
+
+static void mailbox_reg_write(void *opaque, hwaddr offset, uint64_t value,
+                              unsigned size)
+{
+    CXLDeviceState *cxl_dstate = opaque;
+
+    if (offset >= A_CXL_DEV_CMD_PAYLOAD) {
+        memcpy(cxl_dstate->mbox_reg_state + offset, &value, size);
+        return;
+    }
+
+    /*
+     * Lock is needed to prevent concurrent writes as well as to prevent writes
+     * coming in while the firmware is processing. Without background commands
+     * or the second mailbox implemented, this serves no purpose since the
+     * memory access is synchronized at a higher level (per memory region).
+     */
+    RCU_READ_LOCK_GUARD();
+
+    switch (size) {
+    case 4:
+        mailbox_mem_writel(cxl_dstate->mbox_reg_state32, offset, value);
+        break;
+    case 8:
+        mailbox_mem_writeq(cxl_dstate->mbox_reg_state64, offset, value);
+        break;
+    default:
+        g_assert_not_reached();
+    }
+
+    if (ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                         DOORBELL))
+        cxl_process_mailbox(cxl_dstate);
+}
+
+static const MemoryRegionOps mailbox_ops = {
+    .read = mailbox_reg_read,
+    .write = mailbox_reg_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 4,
+        .max_access_size = 8,
+    },
+};
+
 static const MemoryRegionOps dev_ops = {
     .read = dev_reg_read,
     .write = NULL, /* status register is read only */
@@ -80,20 +185,33 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
                           "cap-array", CXL_DEVICE_REGISTERS_OFFSET - 0);
     memory_region_init_io(&cxl_dstate->device, obj, &dev_ops, cxl_dstate,
                           "device-status", CXL_DEVICE_REGISTERS_LENGTH);
+    memory_region_init_io(&cxl_dstate->mailbox, obj, &mailbox_ops, cxl_dstate,
+                          "mailbox", CXL_MAILBOX_REGISTERS_LENGTH);
 
     memory_region_add_subregion(&cxl_dstate->device_registers, 0,
                                 &cxl_dstate->caps);
     memory_region_add_subregion(&cxl_dstate->device_registers,
                                 CXL_DEVICE_REGISTERS_OFFSET,
                                 &cxl_dstate->device);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_MAILBOX_REGISTERS_OFFSET,
+                                &cxl_dstate->mailbox);
 }
 
 static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
 
+static void mailbox_reg_init_common(CXLDeviceState *cxl_dstate)
+{
+    /* 2048 payload size, with no interrupt or background support */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CAP,
+                     PAYLOAD_SIZE, CXL_MAILBOX_PAYLOAD_SHIFT);
+    cxl_dstate->payload_size = CXL_MAILBOX_MAX_PAYLOAD_SIZE;
+}
+
 void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 {
     uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
-    const int cap_count = 1;
+    const int cap_count = 2;
 
     /* CXL Device Capabilities Array Register */
     ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
@@ -102,4 +220,9 @@ void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 
     cxl_device_cap_init(cxl_dstate, DEVICE, 1);
     device_reg_init_common(cxl_dstate);
+
+    cxl_device_cap_init(cxl_dstate, MAILBOX, 2);
+    mailbox_reg_init_common(cxl_dstate);
+
+    assert(cxl_initialize_mailbox(cxl_dstate) == 0);
 }
hw/cxl/cxl-mailbox-utils.c new file +66 -0
new file mode 100644
index 0000000000..466055b01a
@@ -0,0 +1,197 @@
+/*
+ * CXL Utility library for mailbox interface
+ *
+ * Copyright(C) 2020 Intel Corporation.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2. See the
+ * COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "hw/cxl/cxl.h"
+#include "hw/pci/pci.h"
+#include "qemu/log.h"
+#include "qemu/uuid.h"
+
+/*
+ * How to add a new command, example. The command set FOO, with cmd BAR.
+ *  1. Add the command set and cmd to the enum.
+ *     FOO    = 0x7f,
+ *          #define BAR 0
+ *  2. Forward declare the handler.
+ *     declare_mailbox_handler(FOO_BAR);
+ *  3. Add the command to the cxl_cmd_set[][]
+ *     CXL_CMD(FOO, BAR, 0, 0),
+ *  4. Implement your handler
+ *     define_mailbox_handler(FOO_BAR) { ... return CXL_MBOX_SUCCESS; }
+ *
+ *
+ *  Writing the handler:
+ *    The handler will provide the &struct cxl_cmd, the &CXLDeviceState, and the
+ *    in/out length of the payload. The handler is responsible for consuming the
+ *    payload from cmd->payload and operating upon it as necessary. It must then
+ *    fill the output data into cmd->payload (overwriting what was there),
+ *    setting the length, and returning a valid return code.
+ *
+ *  XXX: The handler need not worry about endianess. The payload is read out of
+ *  a register interface that already deals with it.
+ */
+
+/* 8.2.8.4.5.1 Command Return Codes */
+typedef enum {
+    CXL_MBOX_SUCCESS = 0x0,
+    CXL_MBOX_BG_STARTED = 0x1,
+    CXL_MBOX_INVALID_INPUT = 0x2,
+    CXL_MBOX_UNSUPPORTED = 0x3,
+    CXL_MBOX_INTERNAL_ERROR = 0x4,
+    CXL_MBOX_RETRY_REQUIRED = 0x5,
+    CXL_MBOX_BUSY = 0x6,
+    CXL_MBOX_MEDIA_DISABLED = 0x7,
+    CXL_MBOX_FW_XFER_IN_PROGRESS = 0x8,
+    CXL_MBOX_FW_XFER_OUT_OF_ORDER = 0x9,
+    CXL_MBOX_FW_AUTH_FAILED = 0xa,
+    CXL_MBOX_FW_INVALID_SLOT = 0xb,
+    CXL_MBOX_FW_ROLLEDBACK = 0xc,
+    CXL_MBOX_FW_REST_REQD = 0xd,
+    CXL_MBOX_INVALID_HANDLE = 0xe,
+    CXL_MBOX_INVALID_PA = 0xf,
+    CXL_MBOX_INJECT_POISON_LIMIT = 0x10,
+    CXL_MBOX_PERMANENT_MEDIA_FAILURE = 0x11,
+    CXL_MBOX_ABORTED = 0x12,
+    CXL_MBOX_INVALID_SECURITY_STATE = 0x13,
+    CXL_MBOX_INCORRECT_PASSPHRASE = 0x14,
+    CXL_MBOX_UNSUPPORTED_MAILBOX = 0x15,
+    CXL_MBOX_INVALID_PAYLOAD_LENGTH = 0x16,
+    CXL_MBOX_MAX = 0x17
+} ret_code;

hw/cxl/cxl-mailbox-utils.c line 76

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+
+struct cxl_cmd;
+typedef ret_code (*opcode_handler)(struct cxl_cmd *cmd,
+                                   CXLDeviceState *cxl_dstate, uint16_t *len);
+struct cxl_cmd {
+    const char *name;
+    opcode_handler handler;
+    ssize_t in;
+    uint16_t effect; /* Reported in CEL */
+    uint8_t *payload;    

This payload pointer feels somewhat out of place. Perhaps it should be a parameter passed to the opcode_handler()? The address of the payload doesn't feel like part of the command as such so you are justing using it as somewhere to stash the address when passing to the handler.

Quoted 59 lines
hw/cxl/cxl-mailbox-utils.c new file +59 -0
+};
+
+#define define_mailbox_handler(name)                \
+    static ret_code cmd_##name(struct cxl_cmd *cmd, \
+                               CXLDeviceState *cxl_dstate, uint16_t *len)
+#define declare_mailbox_handler(name) define_mailbox_handler(name)
+
+#define define_mailbox_handler_zeroed(name, size)                         \
+    uint16_t __zero##name = size;                                         \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = __zero##name;                                              \
+        memset(cmd->payload, 0, *len);                                    \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_const(name, data)                          \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = sizeof(data);                                              \
+        memcpy(cmd->payload, data, *len);                                 \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_nop(name)                                  \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+
+#define CXL_CMD(s, c, in, cel_effect) \
+    [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
+
+static struct cxl_cmd cxl_cmd_set[256][256] = {};
+
+#undef CXL_CMD
+
+QemuUUID cel_uuid;
+
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
+{
+    uint16_t ret = CXL_MBOX_SUCCESS;
+    struct cxl_cmd *cxl_cmd;
+    uint64_t status_reg;
+    opcode_handler h;
+
+    /*
+     * current state of mailbox interface
+     *  mbox_cap_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CAP];
+     *  mbox_ctrl_reg = cxl_dstate->reg_state32[R_CXL_DEV_MAILBOX_CTRL];
+     *  status_reg = *(uint64_t *)&cxl_dstate->reg_state[A_CXL_DEV_MAILBOX_STS];
+     */
+    uint64_t command_reg =
+        *(uint64_t *)&cxl_dstate->mbox_reg_state[A_CXL_DEV_MAILBOX_CMD];
+
+    /* Check if we have to do anything */
+    if (!ARRAY_FIELD_EX32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                          DOORBELL)) {

hw/cxl/cxl-mailbox-utils.c line 145

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+        qemu_log_mask(LOG_UNIMP, "Corrupt internal state for firmware\n");
+        return;
+    }
+
+    uint8_t set = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET);
+    uint8_t cmd = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND);
+    uint16_t len = FIELD_EX64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH);
+    cxl_cmd = &cxl_cmd_set[set][cmd];
+    h = cxl_cmd->handler;
+    if (!h) {    

This path seems to not convey information it perhaps should. Maybe some feedback that a command was requested that isn't registered?

Quoted 29 lines
hw/cxl/cxl-mailbox-utils.c new file +29 -0
+        goto handled;
+    }
+
+    if (len != cxl_cmd->in) {
+        ret = CXL_MBOX_INVALID_PAYLOAD_LENGTH;
+    }
+
+    cxl_cmd->payload = cxl_dstate->mbox_reg_state + A_CXL_DEV_CMD_PAYLOAD;
+    ret = (*h)(cxl_cmd, cxl_dstate, &len);
+    assert(len <= cxl_dstate->payload_size);
+
+handled:
+    /*
+     * Set the return code
+     * XXX: it's a 64b register, but we're not setting the vendor, so we can get
+     * away with this
+     */
+    status_reg = FIELD_DP64(0, CXL_DEV_MAILBOX_STS, ERRNO, ret);
+
+    /*
+     * Set the return length
+     */
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND_SET, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, COMMAND, 0);
+    command_reg = FIELD_DP64(command_reg, CXL_DEV_MAILBOX_CMD, LENGTH, len);
+
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_CMD / 8] = command_reg;
+    cxl_dstate->mbox_reg_state64[A_CXL_DEV_MAILBOX_STS / 8] = status_reg;
+

hw/cxl/cxl-mailbox-utils.c line 184

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c new file +10 -0
+    /* Tell the host we're done */
+    ARRAY_FIELD_DP32(cxl_dstate->mbox_reg_state32, CXL_DEV_MAILBOX_CTRL,
+                     DOORBELL, 0);
+}
+
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate)
+{
+    const char *cel_uuidstr = "0da9c0b5-bf41-4b78-8f79-96b1623b3f17";
+
+    for (int i = 0; i < 256; i++) {    

Instead of indexing with i and j, perhaps this would be more consistent using the naming you have above cmd and set?

Quoted 28 lines
hw/cxl/cxl-mailbox-utils.c new file +13 -0
+        for (int j = 0; j < 256; j++) {
+            if (cxl_cmd_set[i][j].handler) {
+                struct cxl_cmd *c = &cxl_cmd_set[i][j];
+
+                cxl_dstate->cel_log[cxl_dstate->cel_size].opcode = (i << 8) | j;
+                cxl_dstate->cel_log[cxl_dstate->cel_size].effect = c->effect;
+                cxl_dstate->cel_size++;
+            }
+        }
+    }
+
+    return qemu_uuid_parse(cel_uuidstr, &cel_uuid);
+}
hw/cxl/meson.build +1 -0
index 47154d6850..0eca715d10 100644
@@ -1,4 +1,5 @@
 softmmu_ss.add(when: 'CONFIG_CXL', if_true: files(
   'cxl-component-utils.c',
   'cxl-device-utils.c',
+  'cxl-mailbox-utils.c',
 ))
include/hw/cxl/cxl.h +0 -0
index 23f52c4cf9..362cda40de 100644

include/hw/cxl/cxl.h line 18

Quoted 5 lines
include/hw/cxl/cxl.h +2 -0
@@ -14,5 +14,8 @@
 #include "cxl_component.h"
 #include "cxl_device.h"
 
+#define COMPONENT_REG_BAR_IDX 0
+#define DEVICE_REG_BAR_IDX 2    

I'd argue for prefixing all defines

CXL_COMPONENT_REG_BAR_IDX etc

Will make it clear they are generic CXL related things.

Quoted 10 lines
include/hw/cxl/cxl.h +1 -0
+
 #endif
 
include/hw/cxl/cxl_device.h +0 -0
index f3bcf19410..af91bec10c 100644
@@ -80,16 +80,27 @@ typedef struct cxl_device_state {
     MemoryRegion device_registers;
 

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 90

Quoted 10 lines
include/hw/cxl/cxl_device.h +1 -3
     /* mmio for device capabilities array - 8.2.8.2 */
+    MemoryRegion device;
     struct {
         MemoryRegion caps;
         uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
     };
 
-    /* mmio for the device status registers 8.2.8.3 */
-    MemoryRegion device;
-    

Move this block back to where it was originally introduced rather than introduce it then move it later?

Quoted 6 lines
include/hw/cxl/cxl_device.h +4 -1
     /* mmio for the mailbox registers 8.2.8.4 */
-    MemoryRegion mailbox;
+    struct {
+        MemoryRegion mailbox;
+        uint16_t payload_size;
+        union {

include/hw/cxl/cxl_device.h typedef struct cxl_device_state { · line 103

Quoted 10 lines
include/hw/cxl/cxl_device.h +10 -0
+            uint8_t mbox_reg_state[CXL_MAILBOX_REGISTERS_LENGTH];
+            uint32_t mbox_reg_state32[CXL_MAILBOX_REGISTERS_LENGTH / 4];
+            uint64_t mbox_reg_state64[CXL_MAILBOX_REGISTERS_LENGTH / 8];
+        };
+        struct {
+            uint16_t opcode;
+            uint16_t effect;
+        } cel_log[1 << 16];
+        size_t cel_size;
+    };    

If the structure is unnamed, chances of a naming clash seem rather high if you don't prefix all the elements with mbx_ or something like that.

Quoted 14 lines
include/hw/cxl/cxl_device.h +3 -0
     /* memory region for persistent memory, HDM */
     MemoryRegion *pmem;
@@ -135,6 +146,9 @@ CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
 
+int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate);
+void cxl_process_mailbox(CXLDeviceState *cxl_dstate);
+
 #define cxl_device_cap_init(dstate, reg, cap_id)                                   \
     do {                                                                           \
         uint32_t *cap_hdrs = dstate->caps_reg_state32;                             \

include/hw/cxl/cxl_device.h REG32(CXL_DEV_MAILBOX_CTRL, 4) · line 184

Quoted 9 lines
include/hw/cxl/cxl_device.h +6 -0
@@ -162,6 +176,12 @@ REG32(CXL_DEV_MAILBOX_CTRL, 4)
     FIELD(CXL_DEV_MAILBOX_CTRL, INT_EN, 1, 1)
     FIELD(CXL_DEV_MAILBOX_CTRL, BG_INT_EN, 2, 1)
 
+/* XXX: actually a 64b register */
+REG32(CXL_DEV_MAILBOX_CMD, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND, 0, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, COMMAND_SET, 8, 8)
+    FIELD(CXL_DEV_MAILBOX_CMD, LENGTH, 16, 20)
+    

Ah. this is where this definition went. Perhaps pull it back into patch 3? That patch defines plenty of other things that aren't used until later patches I think, so one more won't hurt and will save me asking why you skipped it:)

Quoted 3 lines
include/hw/cxl/cxl_device.h +0 -0
 /* XXX: actually a 64b register */
 REG32(CXL_DEV_MAILBOX_STS, 0x10)
     FIELD(CXL_DEV_MAILBOX_STS, BG_OP, 0, 1)    
#17 lore raw

Re: [RFC PATCH v3 05/31] hw/cxl/device: Implement basic mailbox (8.2.8.4)

In reply to Ben Widawsky
Quoted 5 lines · On Mon, 1 Feb 2021 16:59:22 -0800 Ben Widawsky <[email protected]> wrote:

This is the beginning of implementing mailbox support for CXL 2.0 devices. The implementation recognizes when the doorbell is rung, handles the command/payload, clears the doorbell while returning error codes and data.

Quoted 8 lines

Generally the mailbox mechanism is designed to permit communication between the host OS and the firmware running on the device. For our purposes, we emulate both the firmware, implemented primarily in cxl-mailbox-utils.c, and the hardware.

No commands are implemented yet.

Sorry review is a little incoherent. It's a lot of patches so I've ended up looking at your tree then trying to figure out which patch a given comment belongs alongside.

Quoted 16 lines
---
 hw/cxl/cxl-device-utils.c   | 125 ++++++++++++++++++++++-
 hw/cxl/cxl-mailbox-utils.c  | 197 ++++++++++++++++++++++++++++++++++++
 hw/cxl/meson.build          |   1 +
 include/hw/cxl/cxl.h        |   3 +
 include/hw/cxl/cxl_device.h |  28 ++++-
 5 files changed, 349 insertions(+), 5 deletions(-)
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
hw/cxl/cxl-device-utils.c +0 -0
index bb15ad9a0f..6602606f3d 100644
@@ -40,6 +40,111 @@ static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size)
     return 0;
 }

hw/cxl/cxl-device-utils.c static uint64_t dev_reg_read(void *opaque, hwaddr offset, unsigned size) · line 51

Quoted 10 lines
hw/cxl/cxl-device-utils.c +7 -0
 
 
 
+
+#define define_mailbox_handler_zeroed(name, size)                         \
+    uint16_t __zero##name = size;                                         \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = __zero##name;                                              \

Why not just use the value of size here?

__zero##name isn't used anywhere else.

Quoted 18 lines
hw/cxl/cxl-device-utils.c +18 -0
+        memset(cmd->payload, 0, *len);                                    \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_const(name, data)                          \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        *len = sizeof(data);                                              \
+        memcpy(cmd->payload, data, *len);                                 \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+#define define_mailbox_handler_nop(name)                                  \
+    static ret_code cmd_##name(struct cxl_cmd *cmd,                       \
+                               CXLDeviceState *cxl_dstate, uint16_t *len) \
+    {                                                                     \
+        return CXL_MBOX_SUCCESS;                                          \
+    }
+
#18 lore raw

RFC PATCH v3 6/31 hw/cxl/device: Add memory device utilities

In reply to Ben Widawsky

Memory devices implement extra capabilities on top of CXL devices. This adds support for that.

A large part of memory devices is the mailbox/command interface. All of the mailbox handling is done in the mailbox-utils library. Longer term, new CXL devices that are being emulated may want to handle commands differently, and therefore would need a mechanism to opt in/out of the specific generic handlers. As such, this is considered sufficient for now, but may need more depth in the future.

---
 hw/cxl/cxl-device-utils.c   | 38 ++++++++++++++++++++++++++++++++++++-
 include/hw/cxl/cxl_device.h | 18 +++++++++++++++++-
 2 files changed, 54 insertions(+), 2 deletions(-)
hw/cxl/cxl-device-utils.c +37 -1
index 6602606f3d..639ace523d 100644
@@ -130,6 +130,31 @@ static void mailbox_reg_write(void *opaque, hwaddr offset, uint64_t value,
         cxl_process_mailbox(cxl_dstate);
 }
 
+static uint64_t mdev_reg_read(void *opaque, hwaddr offset, unsigned size)
+{
+    uint64_t retval = 0;
+
+    retval = FIELD_DP64(retval, CXL_MEM_DEV_STS, MEDIA_STATUS, 1);
+    retval = FIELD_DP64(retval, CXL_MEM_DEV_STS, MBOX_READY, 1);
+
+    return retval;
+}
+
+static const MemoryRegionOps mdev_ops = {
+    .read = mdev_reg_read,
+    .write = NULL, /* memory device register is read only */
+    .endianness = DEVICE_LITTLE_ENDIAN,
+    .valid = {
+        .min_access_size = 1,
+        .max_access_size = 8,
+        .unaligned = false,
+    },
+    .impl = {
+        .min_access_size = 8,
+        .max_access_size = 8,
+    },
+};
+
 static const MemoryRegionOps mailbox_ops = {
     .read = mailbox_reg_read,
     .write = mailbox_reg_write,
@@ -187,6 +212,9 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
                           "device-status", CXL_DEVICE_REGISTERS_LENGTH);
     memory_region_init_io(&cxl_dstate->mailbox, obj, &mailbox_ops, cxl_dstate,
                           "mailbox", CXL_MAILBOX_REGISTERS_LENGTH);
+    memory_region_init_io(&cxl_dstate->memory_device, obj, &mdev_ops,
+                          cxl_dstate, "memory device caps",
+                          CXL_MEMORY_DEVICE_REGISTERS_LENGTH);
 
     memory_region_add_subregion(&cxl_dstate->device_registers, 0,
                                 &cxl_dstate->caps);
@@ -196,6 +224,9 @@ void cxl_device_register_block_init(Object *obj, CXLDeviceState *cxl_dstate)
     memory_region_add_subregion(&cxl_dstate->device_registers,
                                 CXL_MAILBOX_REGISTERS_OFFSET,
                                 &cxl_dstate->mailbox);
+    memory_region_add_subregion(&cxl_dstate->device_registers,
+                                CXL_MEMORY_DEVICE_REGISTERS_OFFSET,
+                                &cxl_dstate->memory_device);
 }
 
 static void device_reg_init_common(CXLDeviceState *cxl_dstate) { }
@@ -208,10 +239,12 @@ static void mailbox_reg_init_common(CXLDeviceState *cxl_dstate)
     cxl_dstate->payload_size = CXL_MAILBOX_MAX_PAYLOAD_SIZE;
 }
 
+static void memdev_reg_init_common(CXLDeviceState *cxl_dstate) { }
+
 void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
 {
     uint32_t *cap_hdrs = cxl_dstate->caps_reg_state32;
-    const int cap_count = 2;
+    const int cap_count = 3;
 
     /* CXL Device Capabilities Array Register */
     ARRAY_FIELD_DP32(cap_hdrs, CXL_DEV_CAP_ARRAY, CAP_ID, 0);
@@ -224,5 +257,8 @@ void cxl_device_register_init_common(CXLDeviceState *cxl_dstate)
     cxl_device_cap_init(cxl_dstate, MAILBOX, 2);
     mailbox_reg_init_common(cxl_dstate);
 
+    cxl_device_cap_init(cxl_dstate, MEMORY_DEVICE, 0x4000);
+    memdev_reg_init_common(cxl_dstate);
+
     assert(cxl_initialize_mailbox(cxl_dstate) == 0);
 }
include/hw/cxl/cxl_device.h +17 -1
index af91bec10c..0cc5354ba4 100644
@@ -72,15 +72,20 @@
 #define CXL_MAILBOX_REGISTERS_LENGTH \
     (CXL_MAILBOX_REGISTERS_SIZE + CXL_MAILBOX_MAX_PAYLOAD_SIZE)
 
+#define CXL_MEMORY_DEVICE_REGISTERS_OFFSET \
+    (CXL_MAILBOX_REGISTERS_OFFSET + CXL_MAILBOX_REGISTERS_LENGTH)
+#define CXL_MEMORY_DEVICE_REGISTERS_LENGTH 0x8
+
 #define CXL_MMIO_SIZE                                       \
     CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_REGISTERS_LENGTH + \
-        CXL_MAILBOX_REGISTERS_LENGTH
+        CXL_MAILBOX_REGISTERS_LENGTH + CXL_MEMORY_DEVICE_REGISTERS_LENGTH
 
 typedef struct cxl_device_state {
     MemoryRegion device_registers;
 
     /* mmio for device capabilities array - 8.2.8.2 */
     MemoryRegion device;
+    MemoryRegion memory_device;
     struct {
         MemoryRegion caps;
         uint32_t caps_reg_state32[CXL_CAPS_SIZE / 4];
@@ -145,6 +150,9 @@ REG32(CXL_DEV_CAP_ARRAY2, 4) /* We're going to pretend it's 64b */
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(DEVICE, CXL_DEVICE_CAP_HDR1_OFFSET)
 CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MAILBOX, CXL_DEVICE_CAP_HDR1_OFFSET + \
                                                CXL_DEVICE_CAP_REG_SIZE)
+CXL_DEVICE_CAPABILITY_HEADER_REGISTER(MEMORY_DEVICE,
+                                      CXL_DEVICE_CAP_HDR1_OFFSET +
+                                          CXL_DEVICE_CAP_REG_SIZE * 2)
 
 int cxl_initialize_mailbox(CXLDeviceState *cxl_dstate);
 void cxl_process_mailbox(CXLDeviceState *cxl_dstate);
@@ -197,4 +205,12 @@ REG32(CXL_DEV_BG_CMD_STS, 0x18)
 
 REG32(CXL_DEV_CMD_PAYLOAD, 0x20)
 
+/* XXX: actually a 64b registers */
+REG32(CXL_MEM_DEV_STS, 0)
+    FIELD(CXL_MEM_DEV_STS, FATAL, 0, 1)
+    FIELD(CXL_MEM_DEV_STS, FW_HALT, 1, 1)
+    FIELD(CXL_MEM_DEV_STS, MEDIA_STATUS, 2, 2)
+    FIELD(CXL_MEM_DEV_STS, MBOX_READY, 4, 1)
+    FIELD(CXL_MEM_DEV_STS, RESET_NEEDED, 5, 3)
+
 #endif
Signature · 1 line
2.30.0
#19 lore raw

RFC PATCH v3 7/31 hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1)

In reply to Ben Widawsky

Using the previously implemented stubbed helpers, it is now possible to easily add the missing, required commands to the implementation.

---
 hw/cxl/cxl-mailbox-utils.c | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)
hw/cxl/cxl-mailbox-utils.c +22 -1
index 466055b01a..7c939a1851 100644
@@ -37,6 +37,14 @@
  *  a register interface that already deals with it.
  */
 
+enum {
+    EVENTS      = 0x01,
+        #define GET_RECORDS   0x0
+        #define CLEAR_RECORDS   0x1
+        #define GET_INTERRUPT_POLICY   0x2
+        #define SET_INTERRUPT_POLICY   0x3
+};
+
 /* 8.2.8.4.5.1 Command Return Codes */
 typedef enum {
     CXL_MBOX_SUCCESS = 0x0,
@@ -105,10 +113,23 @@ struct cxl_cmd {
         return CXL_MBOX_SUCCESS;                                          \
     }
 
+define_mailbox_handler_zeroed(EVENTS_GET_RECORDS, 0x20);
+define_mailbox_handler_nop(EVENTS_CLEAR_RECORDS);
+define_mailbox_handler_zeroed(EVENTS_GET_INTERRUPT_POLICY, 4);
+define_mailbox_handler_nop(EVENTS_SET_INTERRUPT_POLICY);
+
+#define IMMEDIATE_CONFIG_CHANGE (1 << 1)
+#define IMMEDIATE_LOG_CHANGE (1 << 4)
+
 #define CXL_CMD(s, c, in, cel_effect) \
     [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
 
-static struct cxl_cmd cxl_cmd_set[256][256] = {};
+static struct cxl_cmd cxl_cmd_set[256][256] = {
+    CXL_CMD(EVENTS, GET_RECORDS, 1, 0),
+    CXL_CMD(EVENTS, CLEAR_RECORDS, ~0, IMMEDIATE_LOG_CHANGE),
+    CXL_CMD(EVENTS, GET_INTERRUPT_POLICY, 0, 0),
+    CXL_CMD(EVENTS, SET_INTERRUPT_POLICY, 4, IMMEDIATE_CONFIG_CHANGE),
+};
 
 #undef CXL_CMD
 
Signature · 1 line
2.30.0
#20 lore raw

Re: [RFC PATCH v3 07/31] hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1)

Quoted 4 lines · On Mon, 1 Feb 2021 16:59:24 -0800 Ben Widawsky <[email protected]> wrote:

Using the previously implemented stubbed helpers, it is now possible to easily add the missing, required commands to the implementation.

comment inline. Otherwise LGTM.

Quoted 24 lines
---
 hw/cxl/cxl-mailbox-utils.c | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)
hw/cxl/cxl-mailbox-utils.c +8 -0
index 466055b01a..7c939a1851 100644
@@ -37,6 +37,14 @@
  *  a register interface that already deals with it.
  */
 
+enum {
+    EVENTS      = 0x01,
+        #define GET_RECORDS   0x0
+        #define CLEAR_RECORDS   0x1
+        #define GET_INTERRUPT_POLICY   0x2
+        #define SET_INTERRUPT_POLICY   0x3
+};
+
 /* 8.2.8.4.5.1 Command Return Codes */
 typedef enum {
     CXL_MBOX_SUCCESS = 0x0,

hw/cxl/cxl-mailbox-utils.c struct cxl_cmd { · line 118

Quoted 6 lines
hw/cxl/cxl-mailbox-utils.c +3 -0
@@ -105,10 +113,23 @@ struct cxl_cmd {
         return CXL_MBOX_SUCCESS;                                          \
     }
 
+define_mailbox_handler_zeroed(EVENTS_GET_RECORDS, 0x20);
+define_mailbox_handler_nop(EVENTS_CLEAR_RECORDS);
+define_mailbox_handler_zeroed(EVENTS_GET_INTERRUPT_POLICY, 4);

Ideally add section reference comments for these. Makes my life a tiny bit easier as a reviewer!

Quoted 18 lines
hw/cxl/cxl-mailbox-utils.c +11 -1
+define_mailbox_handler_nop(EVENTS_SET_INTERRUPT_POLICY);
+
+#define IMMEDIATE_CONFIG_CHANGE (1 << 1)
+#define IMMEDIATE_LOG_CHANGE (1 << 4)
+
 #define CXL_CMD(s, c, in, cel_effect) \
     [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
 
-static struct cxl_cmd cxl_cmd_set[256][256] = {};
+static struct cxl_cmd cxl_cmd_set[256][256] = {
+    CXL_CMD(EVENTS, GET_RECORDS, 1, 0),
+    CXL_CMD(EVENTS, CLEAR_RECORDS, ~0, IMMEDIATE_LOG_CHANGE),
+    CXL_CMD(EVENTS, GET_INTERRUPT_POLICY, 0, 0),
+    CXL_CMD(EVENTS, SET_INTERRUPT_POLICY, 4, IMMEDIATE_CONFIG_CHANGE),
+};
 
 #undef CXL_CMD
 
#21 lore raw

Re: [RFC PATCH v3 07/31] hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1)

In reply to Ben Widawsky
Quoted 38 lines · On Mon, 1 Feb 2021 16:59:24 -0800 Ben Widawsky <[email protected]> wrote:

Using the previously implemented stubbed helpers, it is now possible to easily add the missing, required commands to the implementation.

---
 hw/cxl/cxl-mailbox-utils.c | 23 ++++++++++++++++++++++-
 1 file changed, 22 insertions(+), 1 deletion(-)
hw/cxl/cxl-mailbox-utils.c +15 -0
index 466055b01a..7c939a1851 100644
@@ -37,6 +37,14 @@
  *  a register interface that already deals with it.
  */
 
+enum {
+    EVENTS      = 0x01,
+        #define GET_RECORDS   0x0
+        #define CLEAR_RECORDS   0x1
+        #define GET_INTERRUPT_POLICY   0x2
+        #define SET_INTERRUPT_POLICY   0x3
+};
+
 /* 8.2.8.4.5.1 Command Return Codes */
 typedef enum {
     CXL_MBOX_SUCCESS = 0x0,
@@ -105,10 +113,23 @@ struct cxl_cmd {
         return CXL_MBOX_SUCCESS;                                          \
     }
 
+define_mailbox_handler_zeroed(EVENTS_GET_RECORDS, 0x20);
+define_mailbox_handler_nop(EVENTS_CLEAR_RECORDS);
+define_mailbox_handler_zeroed(EVENTS_GET_INTERRUPT_POLICY, 4);
+define_mailbox_handler_nop(EVENTS_SET_INTERRUPT_POLICY);
+
+#define IMMEDIATE_CONFIG_CHANGE (1 << 1)
+#define IMMEDIATE_LOG_CHANGE (1 << 4)

hw/cxl/cxl-mailbox-utils.c struct cxl_cmd { · line 131

Quoted 10 lines
hw/cxl/cxl-mailbox-utils.c +6 -1
+
 #define CXL_CMD(s, c, in, cel_effect) \
     [s][c] = { stringify(s##_##c), cmd_##s##_##c, in, cel_effect }
 
-static struct cxl_cmd cxl_cmd_set[256][256] = {};
+static struct cxl_cmd cxl_cmd_set[256][256] = {
+    CXL_CMD(EVENTS, GET_RECORDS, 1, 0),
+    CXL_CMD(EVENTS, CLEAR_RECORDS, ~0, IMMEDIATE_LOG_CHANGE),
+    CXL_CMD(EVENTS, GET_INTERRUPT_POLICY, 0, 0),
+    CXL_CMD(EVENTS, SET_INTERRUPT_POLICY, 4, IMMEDIATE_CONFIG_CHANGE),

CXL 2.0 spec says IMMEDIATE_POLICY_CHANGE for this rather than IMMEDIATE_CONFIG_CHANGE.

Quoted 4 lines
hw/cxl/cxl-mailbox-utils.c +1 -0
+};
 
 #undef CXL_CMD
 
#22 lore raw

RFC PATCH v3 8/31 hw/cxl/device: Timestamp implementation (8.2.9.3)

In reply to Ben Widawsky

Per spec, timestamp appears to be a free-running counter from a value set by the host via the Set Timestamp command (0301h). There are references to the epoch, which seem like a red herring. Therefore, the implementation implements the timestamp as freerunning counter from the last value that was issued by the Set Timestamp command.

---
 hw/cxl/cxl-mailbox-utils.c  | 53 +++++++++++++++++++++++++++++++++++++
 include/hw/cxl/cxl_device.h |  6 +++++
 2 files changed, 59 insertions(+)
hw/cxl/cxl-mailbox-utils.c +53 -0
index 7c939a1851..3d36614c0c 100644
@@ -43,6 +43,9 @@ enum {
         #define CLEAR_RECORDS   0x1
         #define GET_INTERRUPT_POLICY   0x2
         #define SET_INTERRUPT_POLICY   0x3
+    TIMESTAMP   = 0x03,
+        #define GET           0x0
+        #define SET           0x1
 };
 
 /* 8.2.8.4.5.1 Command Return Codes */
@@ -117,8 +120,11 @@ define_mailbox_handler_zeroed(EVENTS_GET_RECORDS, 0x20);
 define_mailbox_handler_nop(EVENTS_CLEAR_RECORDS);
 define_mailbox_handler_zeroed(EVENTS_GET_INTERRUPT_POLICY, 4);
 define_mailbox_handler_nop(EVENTS_SET_INTERRUPT_POLICY);
+declare_mailbox_handler(TIMESTAMP_GET);
+declare_mailbox_handler(TIMESTAMP_SET);
 
 #define IMMEDIATE_CONFIG_CHANGE (1 << 1)
+#define IMMEDIATE_POLICY_CHANGE (1 << 3)
 #define IMMEDIATE_LOG_CHANGE (1 << 4)
 
 #define CXL_CMD(s, c, in, cel_effect) \
@@ -129,10 +135,57 @@ static struct cxl_cmd cxl_cmd_set[256][256] = {
     CXL_CMD(EVENTS, CLEAR_RECORDS, ~0, IMMEDIATE_LOG_CHANGE),
     CXL_CMD(EVENTS, GET_INTERRUPT_POLICY, 0, 0),
     CXL_CMD(EVENTS, SET_INTERRUPT_POLICY, 4, IMMEDIATE_CONFIG_CHANGE),
+    CXL_CMD(TIMESTAMP, GET, 0, 0),
+    CXL_CMD(TIMESTAMP, SET, 8, IMMEDIATE_POLICY_CHANGE),
 };
 
 #undef CXL_CMD
 
+/*
+ * 8.2.9.3.1
+ */
+define_mailbox_handler(TIMESTAMP_GET)
+{
+    struct timespec ts;
+    uint64_t delta;
+
+    if (!cxl_dstate->timestamp.set) {
+        *(uint64_t *)cmd->payload = 0;
+        goto done;
+    }
+
+    /* First find the delta from the last time the host set the time. */
+    clock_gettime(CLOCK_REALTIME, &ts);
+    delta = (ts.tv_sec * NANOSECONDS_PER_SECOND + ts.tv_nsec) -
+            cxl_dstate->timestamp.last_set;
+
+    /* Then adjust the actual time */
+    stq_le_p(cmd->payload, cxl_dstate->timestamp.host_set + delta);
+
+done:
+    *len = 8;
+    return CXL_MBOX_SUCCESS;
+}
+
+/*
+ * 8.2.9.3.2
+ */
+define_mailbox_handler(TIMESTAMP_SET)
+{
+    struct timespec ts;
+
+    clock_gettime(CLOCK_REALTIME, &ts);
+
+    cxl_dstate->timestamp.set = true;
+    cxl_dstate->timestamp.last_set =
+        ts.tv_sec * NANOSECONDS_PER_SECOND + ts.tv_nsec;
+
+    cxl_dstate->timestamp.host_set = le64_to_cpu(*(uint64_t *)cmd->payload);
+
+    *len = 0;
+    return CXL_MBOX_SUCCESS;
+}
+
 QemuUUID cel_uuid;
 
 void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
include/hw/cxl/cxl_device.h +6 -0
index 0cc5354ba4..ca5328a581 100644
@@ -107,6 +107,12 @@ typedef struct cxl_device_state {
         size_t cel_size;
     };
 
+    struct {
+        bool set;
+        uint64_t last_set;
+        uint64_t host_set;
+    } timestamp;
+
     /* memory region for persistent memory, HDM */
     MemoryRegion *pmem;
 
Signature · 1 line
2.30.0
#23 lore raw

RFC PATCH v3 9/31 hw/cxl/device: Add log commands (8.2.9.4) + CEL

In reply to Ben Widawsky

CXL specification provides for the ability to obtain logs from the device. Logs are either spec defined, like the "Command Effects Log" (CEL), or vendor specific. UUIDs are defined for all log types.

The CEL is a mechanism to provide information to the host about which commands are supported. It is useful both to determine which spec'd optional commands are supported, as well as provide a list of vendor specified commands that might be used. The CEL is already created as part of mailbox initialization, but here it is now exported to hosts that use these log commands.

---
 hw/cxl/cxl-mailbox-utils.c | 67 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 67 insertions(+)
hw/cxl/cxl-mailbox-utils.c +67 -0
index 3d36614c0c..3f0ae8b9e5 100644
@@ -46,6 +46,9 @@ enum {
     TIMESTAMP   = 0x03,
         #define GET           0x0
         #define SET           0x1
+    LOGS        = 0x04,
+        #define GET_SUPPORTED 0x0
+        #define GET_LOG       0x1
 };
 
 /* 8.2.8.4.5.1 Command Return Codes */
@@ -122,6 +125,8 @@ define_mailbox_handler_zeroed(EVENTS_GET_INTERRUPT_POLICY, 4);
 define_mailbox_handler_nop(EVENTS_SET_INTERRUPT_POLICY);
 declare_mailbox_handler(TIMESTAMP_GET);
 declare_mailbox_handler(TIMESTAMP_SET);
+declare_mailbox_handler(LOGS_GET_SUPPORTED);
+declare_mailbox_handler(LOGS_GET_LOG);
 
 #define IMMEDIATE_CONFIG_CHANGE (1 << 1)
 #define IMMEDIATE_POLICY_CHANGE (1 << 3)
@@ -137,6 +142,8 @@ static struct cxl_cmd cxl_cmd_set[256][256] = {
     CXL_CMD(EVENTS, SET_INTERRUPT_POLICY, 4, IMMEDIATE_CONFIG_CHANGE),
     CXL_CMD(TIMESTAMP, GET, 0, 0),
     CXL_CMD(TIMESTAMP, SET, 8, IMMEDIATE_POLICY_CHANGE),
+    CXL_CMD(LOGS, GET_SUPPORTED, 0, 0),
+    CXL_CMD(LOGS, GET_LOG, 0x18, 0),
 };
 
 #undef CXL_CMD
@@ -188,6 +195,66 @@ define_mailbox_handler(TIMESTAMP_SET)
 
 QemuUUID cel_uuid;
 
+/* 8.2.9.4.1 */
+define_mailbox_handler(LOGS_GET_SUPPORTED)
+{
+    struct {
+        uint16_t entries;
+        uint8_t rsvd[6];
+        struct {
+            QemuUUID uuid;
+            uint32_t size;
+        } log_entries[1];
+    } __attribute__((packed)) *supported_logs = (void *)cmd->payload;
+    _Static_assert(sizeof(*supported_logs) == 0x1c, "Bad supported log size");
+
+    supported_logs->entries = 1;
+    supported_logs->log_entries[0].uuid = cel_uuid;
+    supported_logs->log_entries[0].size = 4 * cxl_dstate->cel_size;
+
+    *len = sizeof(*supported_logs);
+    return CXL_MBOX_SUCCESS;
+}
+
+/* 8.2.9.4.2 */
+define_mailbox_handler(LOGS_GET_LOG)
+{
+    struct {
+        QemuUUID uuid;
+        uint32_t offset;
+        uint32_t length;
+    } __attribute__((packed, __aligned__(16))) *get_log = (void *)cmd->payload;
+
+    /*
+     * 8.2.9.4.2
+     *   The device shall return Invalid Parameter if the Offset or Length
+     *   fields attempt to access beyond the size of the log as reported by Get
+     *   Supported Logs.
+     *
+     * XXX: Spec is wrong, "Invalid Parameter" isn't a thing.
+     * XXX: Spec doesn't address incorrect UUID incorrectness.
+     *
+     * The CEL buffer is large enough to fit all commands in the emulation, so
+     * the only possible failure would be if the mailbox itself isn't big
+     * enough.
+     */
+    if (get_log->offset + get_log->length > cxl_dstate->payload_size) {
+        return CXL_MBOX_INVALID_INPUT;
+    }
+
+    if (!qemu_uuid_is_equal(&get_log->uuid, &cel_uuid)) {
+        return CXL_MBOX_UNSUPPORTED;
+    }
+
+    /* Store off everything to local variables so we can wipe out the payload */
+    *len = get_log->length;
+
+    memmove(cmd->payload, cxl_dstate->cel_log + get_log->offset,
+           get_log->length);
+
+    return CXL_MBOX_SUCCESS;
+}
+
 void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
 {
     uint16_t ret = CXL_MBOX_SUCCESS;
Signature · 1 line
2.30.0
#24 lore raw

RFC PATCH v3 10/31 hw/pxb: Use a type for realizing expanders

In reply to Ben Widawsky

This opens up the possibility for more types of expanders (other than PCI and PCIe). We'll need this to create a CXL expander.

---
 hw/pci-bridge/pci_expander_bridge.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
hw/pci-bridge/pci_expander_bridge.c +7 -4
index aedded1064..232b7ce305 100644
@@ -24,6 +24,8 @@
 #include "hw/boards.h"
 #include "qom/object.h"
 
+enum BusType { PCI, PCIE };
+
 #define TYPE_PXB_BUS "pxb-bus"
 typedef struct PXBBus PXBBus;
 DECLARE_INSTANCE_CHECKER(PXBBus, PXB_BUS,
@@ -214,7 +216,8 @@ static gint pxb_compare(gconstpointer a, gconstpointer b)
            0;
 }
 
-static void pxb_dev_realize_common(PCIDevice *dev, bool pcie, Error **errp)
+static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
+                                   Error **errp)
 {
     PXBDev *pxb = convert_to_pxb(dev);
     DeviceState *ds, *bds = NULL;
@@ -239,7 +242,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, bool pcie, Error **errp)
     }
 
     ds = qdev_new(TYPE_PXB_HOST);
-    if (pcie) {
+    if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else {
         bus = pci_root_bus_new(ds, "pxb-internal", NULL, NULL, 0, TYPE_PXB_BUS);
@@ -287,7 +290,7 @@ static void pxb_dev_realize(PCIDevice *dev, Error **errp)
         return;
     }
 
-    pxb_dev_realize_common(dev, false, errp);
+    pxb_dev_realize_common(dev, PCI, errp);
 }
 
 static void pxb_dev_exitfn(PCIDevice *pci_dev)
@@ -339,7 +342,7 @@ static void pxb_pcie_dev_realize(PCIDevice *dev, Error **errp)
         return;
     }
 
-    pxb_dev_realize_common(dev, true, errp);
+    pxb_dev_realize_common(dev, PCIE, errp);
 }
 
 static void pxb_pcie_dev_class_init(ObjectClass *klass, void *data)
Signature · 1 line
2.30.0
#25 lore raw

Re: [RFC PATCH v3 10/31] hw/pxb: Use a type for realizing expanders

Quoted 4 lines · On Mon, 1 Feb 2021 16:59:27 -0800 Ben Widawsky <[email protected]> wrote:

This opens up the possibility for more types of expanders (other than PCI and PCIe). We'll need this to create a CXL expander.

Minor suggestion inline but nothing important if you don't want to change it.

Jonathan

Quoted 28 lines
---
 hw/pci-bridge/pci_expander_bridge.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
hw/pci-bridge/pci_expander_bridge.c +4 -1
index aedded1064..232b7ce305 100644
@@ -24,6 +24,8 @@
 #include "hw/boards.h"
 #include "qom/object.h"
 
+enum BusType { PCI, PCIE };
+
 #define TYPE_PXB_BUS "pxb-bus"
 typedef struct PXBBus PXBBus;
 DECLARE_INSTANCE_CHECKER(PXBBus, PXB_BUS,
@@ -214,7 +216,8 @@ static gint pxb_compare(gconstpointer a, gconstpointer b)
            0;
 }
 
-static void pxb_dev_realize_common(PCIDevice *dev, bool pcie, Error **errp)
+static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
+                                   Error **errp)
 {
     PXBDev *pxb = convert_to_pxb(dev);
     DeviceState *ds, *bds = NULL;

hw/pci-bridge/pci_expander_bridge.c static void pxb_dev_realize_common(PCIDevice *dev, bool pcie, Error **errp) · line 245

Quoted 5 lines
hw/pci-bridge/pci_expander_bridge.c +1 -1
@@ -239,7 +242,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, bool pcie, Error **errp)
     }
 
     ds = qdev_new(TYPE_PXB_HOST);
-    if (pcie) {
+    if (type == PCIE) {

I'd make this a switch statement now given we are about to the 3 entries and may well get more in the future.

Quoted 21 lines
hw/pci-bridge/pci_expander_bridge.c +2 -2
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else {
         bus = pci_root_bus_new(ds, "pxb-internal", NULL, NULL, 0, TYPE_PXB_BUS);
@@ -287,7 +290,7 @@ static void pxb_dev_realize(PCIDevice *dev, Error **errp)
         return;
     }
 
-    pxb_dev_realize_common(dev, false, errp);
+    pxb_dev_realize_common(dev, PCI, errp);
 }
 
 static void pxb_dev_exitfn(PCIDevice *pci_dev)
@@ -339,7 +342,7 @@ static void pxb_pcie_dev_realize(PCIDevice *dev, Error **errp)
         return;
     }
 
-    pxb_dev_realize_common(dev, true, errp);
+    pxb_dev_realize_common(dev, PCIE, errp);
 }
 
 static void pxb_pcie_dev_class_init(ObjectClass *klass, void *data)
#26 lore raw

RFC PATCH v3 11/31 hw/pci/cxl: Create a CXL bus type

In reply to Ben Widawsky

The easiest way to differentiate a CXL bus, and a PCIE bus is using a flag. A CXL bus, in hardware, is backward compatible with PCIE, and therefore the code tries pretty hard to keep them in sync as much as possible.

The other way to implement this would be to try to cast the bus to the correct type. This is less code and useful for debugging via simply looking at the flags.

---
 hw/pci-bridge/pci_expander_bridge.c | 9 ++++++++-
 include/hw/pci/pci_bus.h            | 7 +++++++
 2 files changed, 15 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +8 -1
index 232b7ce305..88c45dc3b5 100644
@@ -24,7 +24,7 @@
 #include "hw/boards.h"
 #include "qom/object.h"
 
-enum BusType { PCI, PCIE };
+enum BusType { PCI, PCIE, CXL };
 
 #define TYPE_PXB_BUS "pxb-bus"
 typedef struct PXBBus PXBBus;
@@ -35,6 +35,10 @@ DECLARE_INSTANCE_CHECKER(PXBBus, PXB_BUS,
 DECLARE_INSTANCE_CHECKER(PXBBus, PXB_PCIE_BUS,
                          TYPE_PXB_PCIE_BUS)
 
+#define TYPE_PXB_CXL_BUS "pxb-cxl-bus"
+DECLARE_INSTANCE_CHECKER(PXBBus, PXB_CXL_BUS,
+                         TYPE_PXB_CXL_BUS)
+
 struct PXBBus {
     /*< private >*/
     PCIBus parent_obj;
@@ -244,6 +248,9 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
     ds = qdev_new(TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
+    } else if (type == CXL) {
+        bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_CXL_BUS);
+        bus->flags |= PCI_BUS_CXL;
     } else {
         bus = pci_root_bus_new(ds, "pxb-internal", NULL, NULL, 0, TYPE_PXB_BUS);
         bds = qdev_new("pci-bridge");
include/hw/pci/pci_bus.h +7 -0
index 347440d42c..eb94e7e85c 100644
@@ -24,6 +24,8 @@ enum PCIBusFlags {
     PCI_BUS_IS_ROOT                                         = 0x0001,
     /* PCIe extended configuration space is accessible on this bus */
     PCI_BUS_EXTENDED_CONFIG_SPACE                           = 0x0002,
+    /* This is a CXL Type BUS */
+    PCI_BUS_CXL                                             = 0x0004,
 };
 
 struct PCIBus {
@@ -53,6 +55,11 @@ struct PCIBus {
     Notifier machine_done;
 };
 
+static inline bool pci_bus_is_cxl(PCIBus *bus)
+{
+    return !!(bus->flags & PCI_BUS_CXL);
+}
+
 static inline bool pci_bus_is_root(PCIBus *bus)
 {
     return !!(bus->flags & PCI_BUS_IS_ROOT);
Signature · 1 line
2.30.0
#27 lore raw

RFC PATCH v3 12/31 hw/pxb: Allow creation of a CXL PXB (host bridge)

In reply to Ben Widawsky

This works like adding a typical pxb device, except the name is 'pxb-cxl' instead of 'pxb-pcie'. An example command line would be as follows:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1

A CXL PXB is backward compatible with PCIe. What this means in practice is that an operating system that is unaware of CXL should still be able to enumerate this topology as if it were PCIe.

One can create multiple CXL PXB host bridges, but a host bridge can only be connected to the main root bus. Host bridges cannot appear elsewhere in the topology.

Note that as of this patch, the ACPI tables needed for the host bridge (specifically, an ACPI object in _SB named ACPI0016 and the CEDT) aren't created. So while this patch internally creates it, it cannot be properly used by an operating system or other system software.

Upcoming patches will allow creating multiple host bridges.

v2: Remove vendor and device ID (Ben)

---
 hw/pci-bridge/pci_expander_bridge.c | 67 ++++++++++++++++++++++++++++-
 hw/pci/pci.c                        |  7 +++
 include/hw/pci/pci.h                |  6 +++
 3 files changed, 78 insertions(+), 2 deletions(-)
hw/pci-bridge/pci_expander_bridge.c +65 -2
index 88c45dc3b5..b42592e1ff 100644
@@ -56,6 +56,10 @@ DECLARE_INSTANCE_CHECKER(PXBDev, PXB_DEV,
 DECLARE_INSTANCE_CHECKER(PXBDev, PXB_PCIE_DEV,
                          TYPE_PXB_PCIE_DEVICE)
 
+#define TYPE_PXB_CXL_DEVICE "pxb-cxl"
+DECLARE_INSTANCE_CHECKER(PXBDev, PXB_CXL_DEV,
+                         TYPE_PXB_CXL_DEVICE)
+
 struct PXBDev {
     /*< private >*/
     PCIDevice parent_obj;
@@ -67,6 +71,11 @@ struct PXBDev {
 
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
+    /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
+    if (object_dynamic_cast(OBJECT(dev), TYPE_PXB_CXL_DEVICE)) {
+        return PXB_CXL_DEV(dev);
+    }
+
     return pci_bus_is_express(pci_get_bus(dev))
         ? PXB_PCIE_DEV(dev) : PXB_DEV(dev);
 }
@@ -111,11 +120,20 @@ static const TypeInfo pxb_pcie_bus_info = {
     .class_init    = pxb_bus_class_init,
 };
 
+static const TypeInfo pxb_cxl_bus_info = {
+    .name          = TYPE_PXB_CXL_BUS,
+    .parent        = TYPE_CXL_BUS,
+    .instance_size = sizeof(PXBBus),
+    .class_init    = pxb_bus_class_init,
+};
+
 static const char *pxb_host_root_bus_path(PCIHostState *host_bridge,
                                           PCIBus *rootbus)
 {
-    PXBBus *bus = pci_bus_is_express(rootbus) ?
-                  PXB_PCIE_BUS(rootbus) : PXB_BUS(rootbus);
+    PXBBus *bus = pci_bus_is_cxl(rootbus) ?
+                      PXB_CXL_BUS(rootbus) :
+                      pci_bus_is_express(rootbus) ? PXB_PCIE_BUS(rootbus) :
+                                                    PXB_BUS(rootbus);
 
     snprintf(bus->bus_path, 8, "0000:%02x", pxb_bus_num(rootbus));
     return bus->bus_path;
@@ -380,13 +398,58 @@ static const TypeInfo pxb_pcie_dev_info = {
     },
 };
 
+static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
+{
+    /* A CXL PXB's parent bus is still PCIe */
+    if (!pci_bus_is_express(pci_get_bus(dev))) {
+        error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
+        return;
+    }
+
+    pxb_dev_realize_common(dev, CXL, errp);
+}
+
+static void pxb_cxl_dev_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc   = DEVICE_CLASS(klass);
+    PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
+
+    k->realize             = pxb_cxl_dev_realize;
+    k->exit                = pxb_dev_exitfn;
+    /*
+     * XXX: These types of bridges don't actually show up in the hierarchy so
+     * vendor, device, class, etc. ids are intentionally left out.
+     */
+
+    dc->desc = "CXL Host Bridge";
+    device_class_set_props(dc, pxb_dev_properties);
+    set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories);
+
+    /* Host bridges aren't hotpluggable. FIXME: spec reference */
+    dc->hotpluggable = false;
+}
+
+static const TypeInfo pxb_cxl_dev_info = {
+    .name          = TYPE_PXB_CXL_DEVICE,
+    .parent        = TYPE_PCI_DEVICE,
+    .instance_size = sizeof(PXBDev),
+    .class_init    = pxb_cxl_dev_class_init,
+    .interfaces =
+        (InterfaceInfo[]){
+            { INTERFACE_CONVENTIONAL_PCI_DEVICE },
+            {},
+        },
+};
+
 static void pxb_register_types(void)
 {
     type_register_static(&pxb_bus_info);
     type_register_static(&pxb_pcie_bus_info);
+    type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
+    type_register_static(&pxb_cxl_dev_info);
 }
 
 type_init(pxb_register_types)
hw/pci/pci.c +7 -0
index a45ca326ed..adbe8aa260 100644
@@ -222,6 +222,12 @@ static const TypeInfo pcie_bus_info = {
     .class_init = pcie_bus_class_init,
 };
 
+static const TypeInfo cxl_bus_info = {
+    .name       = TYPE_CXL_BUS,
+    .parent     = TYPE_PCIE_BUS,
+    .class_init = pcie_bus_class_init,
+};
+
 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
 static void pci_update_mappings(PCIDevice *d);
 static void pci_irq_handler(void *opaque, int irq_num, int level);
@@ -2825,6 +2831,7 @@ static void pci_register_types(void)
 {
     type_register_static(&pci_bus_info);
     type_register_static(&pcie_bus_info);
+    type_register_static(&cxl_bus_info);
     type_register_static(&conventional_pci_interface_info);
     type_register_static(&cxl_interface_info);
     type_register_static(&pcie_interface_info);
include/hw/pci/pci.h +6 -0
index 528cef341c..bde3697bee 100644
@@ -406,6 +406,7 @@ typedef PCIINTxRoute (*pci_route_irq_fn)(void *opaque, int pin);
 #define TYPE_PCI_BUS "PCI"
 OBJECT_DECLARE_TYPE(PCIBus, PCIBusClass, PCI_BUS)
 #define TYPE_PCIE_BUS "PCIE"
+#define TYPE_CXL_BUS "CXL"
 
 bool pci_bus_is_express(PCIBus *bus);
 
@@ -754,6 +755,11 @@ static inline void pci_irq_pulse(PCIDevice *pci_dev)
     pci_irq_deassert(pci_dev);
 }
 
+static inline int pci_is_cxl(const PCIDevice *d)
+{
+    return d->cap_present & QEMU_PCIE_CAP_CXL;
+}
+
 static inline int pci_is_express(const PCIDevice *d)
 {
     return d->cap_present & QEMU_PCI_CAP_EXPRESS;
Signature · 1 line
2.30.0
#28 lore raw

RFC PATCH v3 13/31 qtest: allow DSDT acpi table changes

In reply to Ben Widawsky
---
 tests/qtest/bios-tables-test-allowed-diff.h | 21 +++++++++++++++++++++
 1 file changed, 21 insertions(+)
tests/qtest/bios-tables-test-allowed-diff.h +21 -0
index dfb8523c8b..5c695cdf37 100644
@@ -1 +1,22 @@
 /* List of comma-separated changed AML files to ignore */
+"tests/data/acpi/pc/DSDT",
+"tests/data/acpi/pc/DSDT.acpihmat",
+"tests/data/acpi/pc/DSDT.bridge",
+"tests/data/acpi/pc/DSDT.cphp",
+"tests/data/acpi/pc/DSDT.dimmpxm",
+"tests/data/acpi/pc/DSDT.hpbridge",
+"tests/data/acpi/pc/DSDT.hpbrroot",
+"tests/data/acpi/pc/DSDT.ipmikcs",
+"tests/data/acpi/pc/DSDT.memhp",
+"tests/data/acpi/pc/DSDT.numamem",
+"tests/data/acpi/pc/DSDT.roothp",
+"tests/data/acpi/q35/DSDT",
+"tests/data/acpi/q35/DSDT.acpihmat",
+"tests/data/acpi/q35/DSDT.bridge",
+"tests/data/acpi/q35/DSDT.cphp",
+"tests/data/acpi/q35/DSDT.dimmpxm",
+"tests/data/acpi/q35/DSDT.ipmibt",
+"tests/data/acpi/q35/DSDT.memhp",
+"tests/data/acpi/q35/DSDT.mmio64",
+"tests/data/acpi/q35/DSDT.numamem",
+"tests/data/acpi/q35/DSDT.tis",
Signature · 1 line
2.30.0
#29 lore raw

RFC PATCH v3 14/31 acpi/pci: Consolidate host bridge setup

In reply to Ben Widawsky

This cleanup will make it easier to add support for CXL to the mix.

---
 hw/i386/acpi-build.c | 31 +++++++++++++++++--------------
 1 file changed, 17 insertions(+), 14 deletions(-)
hw/i386/acpi-build.c +17 -14
index f56d699c7f..cf6eb54c22 100644
@@ -1194,6 +1194,20 @@ static void build_smb0(Aml *table, I2CBus *smbus, int devnr, int func)
     aml_append(table, scope);
 }
 
+enum { PCI, PCIE };
+static void init_pci_acpi(Aml *dev, int uid, int type)
+{
+    if (type == PCI) {
+        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+    } else {
+        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
+        aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+        aml_append(dev, build_q35_osc_method());
+    }
+}
+
 static void
 build_dsdt(GArray *table_data, BIOSLinker *linker,
            AcpiPmInfo *pm, AcpiMiscInfo *misc,
@@ -1222,9 +1236,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     if (misc->is_piix4) {
         sb_scope = aml_scope("_SB");
         dev = aml_device("PCI0");
-        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
+        init_pci_acpi(dev, 0, PCI);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
-        aml_append(dev, aml_name_decl("_UID", aml_int(0)));
         aml_append(sb_scope, dev);
         aml_append(dsdt, sb_scope);
 
@@ -1238,11 +1251,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     } else {
         sb_scope = aml_scope("_SB");
         dev = aml_device("PCI0");
-        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
-        aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
+        init_pci_acpi(dev, 0, PCIE);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
-        aml_append(dev, aml_name_decl("_UID", aml_int(0)));
-        aml_append(dev, build_q35_osc_method());
         aml_append(sb_scope, dev);
 
         if (pm->smi_on_cpuhp) {
@@ -1345,15 +1355,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
 
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
-            aml_append(dev, aml_name_decl("_UID", aml_int(bus_num)));
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            if (pci_bus_is_express(bus)) {
-                aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
-                aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
-                aml_append(dev, build_q35_osc_method());
-            } else {
-                aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
-            }
+            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
Signature · 1 line
2.30.0
#30 lore raw

Re: [RFC PATCH v3 14/31] acpi/pci: Consolidate host bridge setup

Quoted 3 lines · On Mon, 1 Feb 2021 16:59:31 -0800 Ben Widawsky <[email protected]> wrote:

This cleanup will make it easier to add support for CXL to the mix.

One trivial but up to you on whether you think it's worth changing.

Quoted 14 lines
---
 hw/i386/acpi-build.c | 31 +++++++++++++++++--------------
 1 file changed, 17 insertions(+), 14 deletions(-)
hw/i386/acpi-build.c +2 -0
index f56d699c7f..cf6eb54c22 100644
@@ -1194,6 +1194,20 @@ static void build_smb0(Aml *table, I2CBus *smbus, int devnr, int func)
     aml_append(table, scope);
 }
 
+enum { PCI, PCIE };
+static void init_pci_acpi(Aml *dev, int uid, int type)

hw/i386/acpi-build.c static void build_smb0(Aml *table, I2CBus *smbus, int devnr, int func) · line 1208

Quoted 10 lines
hw/i386/acpi-build.c +10 -0
+{
+    if (type == PCI) {
+        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+    } else {
+        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
+        aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+        aml_append(dev, build_q35_osc_method());
+    }

Gut feeling would be this will end up nicer as a switch. I suspect this isn't the last time this will get extended!

Quoted 46 lines
hw/i386/acpi-build.c +5 -14
+}
+
 static void
 build_dsdt(GArray *table_data, BIOSLinker *linker,
            AcpiPmInfo *pm, AcpiMiscInfo *misc,
@@ -1222,9 +1236,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     if (misc->is_piix4) {
         sb_scope = aml_scope("_SB");
         dev = aml_device("PCI0");
-        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
+        init_pci_acpi(dev, 0, PCI);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
-        aml_append(dev, aml_name_decl("_UID", aml_int(0)));
         aml_append(sb_scope, dev);
         aml_append(dsdt, sb_scope);
 
@@ -1238,11 +1251,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     } else {
         sb_scope = aml_scope("_SB");
         dev = aml_device("PCI0");
-        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
-        aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
+        init_pci_acpi(dev, 0, PCIE);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
-        aml_append(dev, aml_name_decl("_UID", aml_int(0)));
-        aml_append(dev, build_q35_osc_method());
         aml_append(sb_scope, dev);
 
         if (pm->smi_on_cpuhp) {
@@ -1345,15 +1355,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
 
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
-            aml_append(dev, aml_name_decl("_UID", aml_int(bus_num)));
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            if (pci_bus_is_express(bus)) {
-                aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
-                aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
-                aml_append(dev, build_q35_osc_method());
-            } else {
-                aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
-            }
+            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
#31 lore raw

Re: [RFC PATCH v3 14/31] acpi/pci: Consolidate host bridge setup

In reply to Ben Widawsky
Quoted 3 lines · On Mon, 1 Feb 2021 16:59:31 -0800 Ben Widawsky <[email protected]> wrote:

This cleanup will make it easier to add support for CXL to the mix.

Hi Ben / all (particularly PCI experts!)

So... I was looking at the large impact this has on needing to update ACPI tables for the tests and that got me wondering.
The issue is it reorders things a bit rather than making any functional changes.

Why does the pxb expander ACPI not have ADR set and all other PNP0A03 / PNP0A08 root bridge acpi chunks do?

I've messed around with the values provided and dug around in Linux and other than them being exposed in the sysfs for firmware blobs associated with /sys/class/pci_bus/devices these particular _ADR entries don't actually seem to be used by Linux.

So it becomes a question of what the specification says...

As ever clear as mud :)

PCI firmware spec doesn't say, but has an example with non _ADR entry. 4.5.3 in the PCI Firmware Specification Revisions 3.3
The ACPI spec has an example with _ADR when describing _SEG. 6.5.6 in ACPI 6.4

_ADR description is the address of the device on it's parent bus. I'm not sure a root bridge parent bus (which is probably the SB, has any such concept of an address (which probably explains why I've never seen it set to anything other than 0).

Checking where the _ADR 0 entries came from, they go back to Michael importing tables from seabios in 2013.

My current feeling is we don't want to risk breaking some user of these values so perhaps the simple option is just to add _ADR to the PXB ACPI block as well?

That way I think we will cause less churn in the ACPI tables needed for tests and end up at least consistent in what QEMU presents.

Note I'd ideally like to separate this as a precursor to the CXL series rather than burying it in the middle of that!

Jonathan

Quoted 70 lines
---
 hw/i386/acpi-build.c | 31 +++++++++++++++++--------------
 1 file changed, 17 insertions(+), 14 deletions(-)
hw/i386/acpi-build.c +17 -14
index f56d699c7f..cf6eb54c22 100644
@@ -1194,6 +1194,20 @@ static void build_smb0(Aml *table, I2CBus *smbus, int devnr, int func)
     aml_append(table, scope);
 }
 
+enum { PCI, PCIE };
+static void init_pci_acpi(Aml *dev, int uid, int type)
+{
+    if (type == PCI) {
+        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+    } else {
+        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
+        aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+        aml_append(dev, build_q35_osc_method());
+    }
+}
+
 static void
 build_dsdt(GArray *table_data, BIOSLinker *linker,
            AcpiPmInfo *pm, AcpiMiscInfo *misc,
@@ -1222,9 +1236,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     if (misc->is_piix4) {
         sb_scope = aml_scope("_SB");
         dev = aml_device("PCI0");
-        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
+        init_pci_acpi(dev, 0, PCI);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
-        aml_append(dev, aml_name_decl("_UID", aml_int(0)));
         aml_append(sb_scope, dev);
         aml_append(dsdt, sb_scope);
 
@@ -1238,11 +1251,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     } else {
         sb_scope = aml_scope("_SB");
         dev = aml_device("PCI0");
-        aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
-        aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
+        init_pci_acpi(dev, 0, PCIE);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
-        aml_append(dev, aml_name_decl("_UID", aml_int(0)));
-        aml_append(dev, build_q35_osc_method());
         aml_append(sb_scope, dev);
 
         if (pm->smi_on_cpuhp) {
@@ -1345,15 +1355,8 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
 
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
-            aml_append(dev, aml_name_decl("_UID", aml_int(bus_num)));
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            if (pci_bus_is_express(bus)) {
-                aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
-                aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
-                aml_append(dev, build_q35_osc_method());
-            } else {
-                aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
-            }
+            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
#32 lore raw

RFC PATCH v3 15/31 tests/acpi: remove stale allowed tables

In reply to Ben Widawsky
     Scope (_SB)
     {
         Device (PCI0)
         {
             Name (_HID, EisaId ("PNP0A03") /* PCI Bus */)  // _HID: Hardware ID
-            Name (_ADR, Zero)  // _ADR: Address
             Name (_UID, Zero)  // _UID: Unique ID
+            Name (_ADR, Zero)  // _ADR: Address
---
 tests/data/acpi/pc/DSDT                     | Bin 5065 -> 5065 bytes
 tests/data/acpi/pc/DSDT.acpihmat            | Bin 6390 -> 6390 bytes
 tests/data/acpi/pc/DSDT.bridge              | Bin 6924 -> 6924 bytes
 tests/data/acpi/pc/DSDT.cphp                | Bin 5529 -> 5529 bytes
 tests/data/acpi/pc/DSDT.dimmpxm             | Bin 6719 -> 6719 bytes
 tests/data/acpi/pc/DSDT.hpbridge            | Bin 5026 -> 5026 bytes
 tests/data/acpi/pc/DSDT.hpbrroot            | Bin 3084 -> 3084 bytes
 tests/data/acpi/pc/DSDT.ipmikcs             | Bin 5137 -> 5137 bytes
 tests/data/acpi/pc/DSDT.memhp               | Bin 6424 -> 6424 bytes
 tests/data/acpi/pc/DSDT.numamem             | Bin 5071 -> 5071 bytes
 tests/data/acpi/pc/DSDT.roothp              | Bin 5261 -> 5261 bytes
 tests/data/acpi/q35/DSDT                    | Bin 7801 -> 7801 bytes
 tests/data/acpi/q35/DSDT.acpihmat           | Bin 9126 -> 9126 bytes
 tests/data/acpi/q35/DSDT.bridge             | Bin 7819 -> 7819 bytes
 tests/data/acpi/q35/DSDT.cphp               | Bin 8265 -> 8265 bytes
 tests/data/acpi/q35/DSDT.dimmpxm            | Bin 9455 -> 9455 bytes
 tests/data/acpi/q35/DSDT.ipmibt             | Bin 7876 -> 7876 bytes
 tests/data/acpi/q35/DSDT.memhp              | Bin 9160 -> 9160 bytes
 tests/data/acpi/q35/DSDT.mmio64             | Bin 8932 -> 8932 bytes
 tests/data/acpi/q35/DSDT.numamem            | Bin 7807 -> 7807 bytes
 tests/qtest/bios-tables-test-allowed-diff.h |  21 --------------------
 21 files changed, 21 deletions(-)
tests/data/acpi/pc/DSDT binary +0 -0
index f6173df1d598767a79aa34ad7585ad7d45c5d4f3..b516745128e3f1a297b6327e9057026a2d16229c 100644
GIT binary patch
tests/data/acpi/pc/DSDT.acpihmat binary +0 -0
index 67f3f7249eaaa9404ebf0f2d0a324b8c8e3bd445..aeae285c6434ae6cf3c53660e34425727a497871 100644
GIT binary patch
tests/data/acpi/pc/DSDT.bridge binary +0 -0
index 643390f4c4138b37fc481656d3f555d0eeedcb02..4cd26a87dd11d96e10bf6de786b9d56ebfe0a4f9 100644
GIT binary patch
tests/data/acpi/pc/DSDT.cphp binary +0 -0
index 1ddcf7d8812f5d8d4d38fe7e7b35fd5885806046..fecb784812cbb2308ef58acf4a2c580f56d35c39 100644
GIT binary patch
tests/data/acpi/pc/DSDT.dimmpxm binary +0 -0
index c44385cc01879324738ffb7f997b8cdd762cbf97..f2c31e150ead16e4931367a6dab42704950a21e9 100644
GIT binary patch
tests/data/acpi/pc/DSDT.hpbridge binary +0 -0
index 4ecf1eb13bf49499f729b53a6d0114672a76e28d..7a8955cdbc52c025a2fd8f160cf8aff9442c985b 100644
GIT binary patch
tests/data/acpi/pc/DSDT.hpbrroot binary +0 -0
index a3046226ec1dcb234b726029b3790dfedb3b9221..88d23fca4743c2ee57493e7d77d6297a60964d3c 100644
GIT binary patch
tests/data/acpi/pc/DSDT.ipmikcs binary +0 -0
index f1638c5d079a9442c09390426a913010df6efd8d..d670ae793b5778c095a7f8c79ff1a046889d1a56 100644
GIT binary patch
tests/data/acpi/pc/DSDT.memhp binary +0 -0
index 4c19e45e66918c61674785c99e4474e58866f125..a7de3d9fd94e62e8fc357fe3093bf7f394a39219 100644
GIT binary patch
tests/data/acpi/pc/DSDT.numamem binary +0 -0
index 40cfd933259af05ac2aee07fca32f22122255211..57958b6cec216c1fb8731f4ed2da67f0fad7484a 100644
GIT binary patch
tests/data/acpi/pc/DSDT.roothp binary +0 -0
index 078fc8031b479cc77b6527a2b7b4bd576b6e6028..624d0e367693fe267a4237a5fc97295cee2ebd60 100644
GIT binary patch
tests/data/acpi/q35/DSDT binary +0 -0
index d25cd7072932886d6967f4023faac1e1fa6e836c..17e2aebde98e0a3161d93e9b2e200737b13699ac 100644
GIT binary patch
tests/data/acpi/q35/DSDT.acpihmat binary +0 -0
index 722e06af83abcde203a2b96a8ec81fd3bab9fc98..7b3d659352a0923822f6a5db1dbd0a6ad853c446 100644
GIT binary patch
tests/data/acpi/q35/DSDT.bridge binary +0 -0
index 06bac139d668ddfc7914e258b471a303c9dbd192..5961b55b1067c3090b2f1f4cd3386d71efee241d 100644
GIT binary patch
tests/data/acpi/q35/DSDT.cphp binary +0 -0
index 2b933ac482e6883efccbd7d6c96089602f2c0b4d..09c92d52f92bb346ed807945b9638cad958446f8 100644
GIT binary patch
tests/data/acpi/q35/DSDT.dimmpxm binary +0 -0
index bd8f8305b028ef20f9b6d1a0c69ac428d027e3d1..1da97afb32dddafefe7f27934acbcb7d56a67489 100644
GIT binary patch
tests/data/acpi/q35/DSDT.ipmibt binary +0 -0
index a8f868e23c25688ab1c0371016c071f23e9d732f..c7e68432b66e7b4d03284c882c65bbf3066825dc 100644
GIT binary patch
tests/data/acpi/q35/DSDT.memhp binary +0 -0
index 9a802e4c67022386442976d5cb997ea3fc57b58f..3af457dd550461b2d2ea85aa85d7740452913b34 100644
GIT binary patch
tests/data/acpi/q35/DSDT.mmio64 binary +0 -0
index 948c2dc7264c31932b490ca00691a7c4d9aefdb0..a4d20f676ac173e6846dcd4e076220d512215963 100644
GIT binary patch
tests/data/acpi/q35/DSDT.numamem binary +0 -0
index 44ec1b0af400da6d298284aa959aa38add7e6dd5..bbab0d10a2a064528519fa69e90c799430129b75 100644
GIT binary patch
tests/qtest/bios-tables-test-allowed-diff.h +0 -21
index 5c695cdf37..dfb8523c8b 100644
@@ -1,22 +1 @@
 /* List of comma-separated changed AML files to ignore */
-"tests/data/acpi/pc/DSDT",
-"tests/data/acpi/pc/DSDT.acpihmat",
-"tests/data/acpi/pc/DSDT.bridge",
-"tests/data/acpi/pc/DSDT.cphp",
-"tests/data/acpi/pc/DSDT.dimmpxm",
-"tests/data/acpi/pc/DSDT.hpbridge",
-"tests/data/acpi/pc/DSDT.hpbrroot",
-"tests/data/acpi/pc/DSDT.ipmikcs",
-"tests/data/acpi/pc/DSDT.memhp",
-"tests/data/acpi/pc/DSDT.numamem",
-"tests/data/acpi/pc/DSDT.roothp",
-"tests/data/acpi/q35/DSDT",
-"tests/data/acpi/q35/DSDT.acpihmat",
-"tests/data/acpi/q35/DSDT.bridge",
-"tests/data/acpi/q35/DSDT.cphp",
-"tests/data/acpi/q35/DSDT.dimmpxm",
-"tests/data/acpi/q35/DSDT.ipmibt",
-"tests/data/acpi/q35/DSDT.memhp",
-"tests/data/acpi/q35/DSDT.mmio64",
-"tests/data/acpi/q35/DSDT.numamem",
-"tests/data/acpi/q35/DSDT.tis",
Signature · 1 line
2.30.0
#33 lore raw

RFC PATCH v3 16/31 hw/pci: Plumb _UID through host bridges

In reply to Ben Widawsky

Currently, QEMU makes _UID equivalent to the bus number (_BBN). While there is nothing wrong with doing it this way, CXL spec has a heavy reliance on _UID to identify host bridges and there is no link to the bus number. Having a distinct UID solves two problems. The first is it gets us around the limitation of 256 (current max bus number). The second is it allows us to replicate hardware configurations where bus number and uid aren't equivalent. The latter has benefits for our development and debugging using QEMU.

The other way to do this would be to implement the expanded bus numbering, but having an explicit uid makes more sense when trying to replicate real hardware configurations.

The QEMU commandline to utilize this would be:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1,uid=x

--

I'm guessing this patch will be somewhat controversial. For early CXL work, this can be dropped without too much heartache.

---
 hw/i386/acpi-build.c                |  3 ++-
 hw/pci-bridge/pci_expander_bridge.c | 19 +++++++++++++++++++
 hw/pci/pci.c                        | 11 +++++++++++
 include/hw/pci/pci.h                |  1 +
 include/hw/pci/pci_bus.h            |  1 +
 5 files changed, 34 insertions(+), 1 deletion(-)
hw/i386/acpi-build.c +2 -1
index cf6eb54c22..145a503e92 100644
@@ -1343,6 +1343,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         QLIST_FOREACH(bus, &bus->child, sibling) {
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
+            int32_t uid = pci_bus_uid(bus);
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
@@ -1356,7 +1357,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
hw/pci-bridge/pci_expander_bridge.c +19 -0
index b42592e1ff..5021b60435 100644
@@ -67,6 +67,7 @@ struct PXBDev {
 
     uint8_t bus_nr;
     uint16_t numa_node;
+    int32_t uid;
 };
 
 static PXBDev *convert_to_pxb(PCIDevice *dev)
@@ -98,12 +99,20 @@ static uint16_t pxb_bus_numa_node(PCIBus *bus)
     return pxb->numa_node;
 }
 
+static int32_t pxb_bus_uid(PCIBus *bus)
+{
+    PXBDev *pxb = convert_to_pxb(bus->parent_dev);
+
+    return pxb->uid;
+}
+
 static void pxb_bus_class_init(ObjectClass *class, void *data)
 {
     PCIBusClass *pbc = PCI_BUS_CLASS(class);
 
     pbc->bus_num = pxb_bus_num;
     pbc->numa_node = pxb_bus_numa_node;
+    pbc->uid = pxb_bus_uid;
 }
 
 static const TypeInfo pxb_bus_info = {
@@ -329,6 +338,7 @@ static Property pxb_dev_properties[] = {
     /* Note: 0 is not a legal PXB bus number. */
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
+    DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -400,12 +410,21 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
+    PXBDev *pxb = convert_to_pxb(dev);
+
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
         error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
         return;
     }
 
+    if (pxb->uid < 0) {
+        error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");
+        return;
+    }
+
+    /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
+
     pxb_dev_realize_common(dev, CXL, errp);
 }
 
hw/pci/pci.c +11 -0
index adbe8aa260..bf019d91a0 100644
@@ -170,6 +170,11 @@ static uint16_t pcibus_numa_node(PCIBus *bus)
     return NUMA_NODE_UNASSIGNED;
 }
 
+static int32_t pcibus_uid(PCIBus *bus)
+{
+    return -1;
+}
+
 static void pci_bus_class_init(ObjectClass *klass, void *data)
 {
     BusClass *k = BUS_CLASS(klass);
@@ -184,6 +189,7 @@ static void pci_bus_class_init(ObjectClass *klass, void *data)
 
     pbc->bus_num = pcibus_num;
     pbc->numa_node = pcibus_numa_node;
+    pbc->uid = pcibus_uid;
 }
 
 static const TypeInfo pci_bus_info = {
@@ -530,6 +536,11 @@ int pci_bus_numa_node(PCIBus *bus)
     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
 }
 
+int pci_bus_uid(PCIBus *bus)
+{
+    return PCI_BUS_GET_CLASS(bus)->uid(bus);
+}
+
 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
                                  const VMStateField *field)
 {
include/hw/pci/pci.h +1 -0
index bde3697bee..a46de48ccd 100644
@@ -463,6 +463,7 @@ static inline int pci_dev_bus_num(const PCIDevice *dev)
 }
 
 int pci_bus_numa_node(PCIBus *bus);
+int pci_bus_uid(PCIBus *bus);
 void pci_for_each_device(PCIBus *bus, int bus_num,
                          void (*fn)(PCIBus *bus, PCIDevice *d, void *opaque),
                          void *opaque);
include/hw/pci/pci_bus.h +1 -0
index eb94e7e85c..3c9fbc55bb 100644
@@ -17,6 +17,7 @@ struct PCIBusClass {
 
     int (*bus_num)(PCIBus *bus);
     uint16_t (*numa_node)(PCIBus *bus);
+    int32_t (*uid)(PCIBus *bus);
 };
 
 enum PCIBusFlags {
Signature · 1 line
2.30.0
#34 lore raw

Re: [RFC PATCH v3 16/31] hw/pci: Plumb _UID through host bridges

Quoted 16 lines · On Mon, 1 Feb 2021 16:59:33 -0800 Ben Widawsky <[email protected]> wrote:

Currently, QEMU makes _UID equivalent to the bus number (_BBN). While there is nothing wrong with doing it this way, CXL spec has a heavy reliance on _UID to identify host bridges and there is no link to the bus number. Having a distinct UID solves two problems. The first is it gets us around the limitation of 256 (current max bus number). The second is it allows us to replicate hardware configurations where bus number and uid aren't equivalent. The latter has benefits for our development and debugging using QEMU.

The other way to do this would be to implement the expanded bus numbering, but having an explicit uid makes more sense when trying to replicate real hardware configurations.

The QEMU commandline to utilize this would be:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1,uid=x

Quoted 6 lines

--

I'm guessing this patch will be somewhat controversial. For early CXL work, this can be dropped without too much heartache.

Whilst I'm not personally against, this maybe best to drop for now as you say.

Quoted 152 lines
---
 hw/i386/acpi-build.c                |  3 ++-
 hw/pci-bridge/pci_expander_bridge.c | 19 +++++++++++++++++++
 hw/pci/pci.c                        | 11 +++++++++++
 include/hw/pci/pci.h                |  1 +
 include/hw/pci/pci_bus.h            |  1 +
 5 files changed, 34 insertions(+), 1 deletion(-)
hw/i386/acpi-build.c +2 -1
index cf6eb54c22..145a503e92 100644
@@ -1343,6 +1343,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         QLIST_FOREACH(bus, &bus->child, sibling) {
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
+            int32_t uid = pci_bus_uid(bus);
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
@@ -1356,7 +1357,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
hw/pci-bridge/pci_expander_bridge.c +19 -0
index b42592e1ff..5021b60435 100644
@@ -67,6 +67,7 @@ struct PXBDev {
 
     uint8_t bus_nr;
     uint16_t numa_node;
+    int32_t uid;
 };
 
 static PXBDev *convert_to_pxb(PCIDevice *dev)
@@ -98,12 +99,20 @@ static uint16_t pxb_bus_numa_node(PCIBus *bus)
     return pxb->numa_node;
 }
 
+static int32_t pxb_bus_uid(PCIBus *bus)
+{
+    PXBDev *pxb = convert_to_pxb(bus->parent_dev);
+
+    return pxb->uid;
+}
+
 static void pxb_bus_class_init(ObjectClass *class, void *data)
 {
     PCIBusClass *pbc = PCI_BUS_CLASS(class);
 
     pbc->bus_num = pxb_bus_num;
     pbc->numa_node = pxb_bus_numa_node;
+    pbc->uid = pxb_bus_uid;
 }
 
 static const TypeInfo pxb_bus_info = {
@@ -329,6 +338,7 @@ static Property pxb_dev_properties[] = {
     /* Note: 0 is not a legal PXB bus number. */
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
+    DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -400,12 +410,21 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
+    PXBDev *pxb = convert_to_pxb(dev);
+
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
         error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
         return;
     }
 
+    if (pxb->uid < 0) {
+        error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");
+        return;
+    }
+
+    /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
+
     pxb_dev_realize_common(dev, CXL, errp);
 }
 
hw/pci/pci.c +11 -0
index adbe8aa260..bf019d91a0 100644
@@ -170,6 +170,11 @@ static uint16_t pcibus_numa_node(PCIBus *bus)
     return NUMA_NODE_UNASSIGNED;
 }
 
+static int32_t pcibus_uid(PCIBus *bus)
+{
+    return -1;
+}
+
 static void pci_bus_class_init(ObjectClass *klass, void *data)
 {
     BusClass *k = BUS_CLASS(klass);
@@ -184,6 +189,7 @@ static void pci_bus_class_init(ObjectClass *klass, void *data)
 
     pbc->bus_num = pcibus_num;
     pbc->numa_node = pcibus_numa_node;
+    pbc->uid = pcibus_uid;
 }
 
 static const TypeInfo pci_bus_info = {
@@ -530,6 +536,11 @@ int pci_bus_numa_node(PCIBus *bus)
     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
 }
 
+int pci_bus_uid(PCIBus *bus)
+{
+    return PCI_BUS_GET_CLASS(bus)->uid(bus);
+}
+
 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
                                  const VMStateField *field)
 {
include/hw/pci/pci.h +1 -0
index bde3697bee..a46de48ccd 100644
@@ -463,6 +463,7 @@ static inline int pci_dev_bus_num(const PCIDevice *dev)
 }
 
 int pci_bus_numa_node(PCIBus *bus);
+int pci_bus_uid(PCIBus *bus);
 void pci_for_each_device(PCIBus *bus, int bus_num,
                          void (*fn)(PCIBus *bus, PCIDevice *d, void *opaque),
                          void *opaque);
include/hw/pci/pci_bus.h +1 -0
index eb94e7e85c..3c9fbc55bb 100644
@@ -17,6 +17,7 @@ struct PCIBusClass {
 
     int (*bus_num)(PCIBus *bus);
     uint16_t (*numa_node)(PCIBus *bus);
+    int32_t (*uid)(PCIBus *bus);
 };
 
 enum PCIBusFlags {
#35 lore raw

Re: [RFC PATCH v3 16/31] hw/pci: Plumb _UID through host bridges

Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Mon, 1 Feb 2021 16:59:33 -0800 Ben Widawsky <[email protected]> wrote:

Currently, QEMU makes _UID equivalent to the bus number (_BBN). While there is nothing wrong with doing it this way, CXL spec has a heavy reliance on _UID to identify host bridges and there is no link to the bus number. Having a distinct UID solves two problems. The first is it gets us around the limitation of 256 (current max bus number).

Not sure I understand. You want more than 256 host bridges?

Quoted 3 lines

The

Quoted 2 lines

second is it allows us to replicate hardware configurations where bus number and uid aren't equivalent.

A bit more data on when this needs to be the case?

Quoted 11 lines

The latter has benefits for our

Quoted 10 lines

development and debugging using QEMU.

The other way to do this would be to implement the expanded bus numbering, but having an explicit uid makes more sense when trying to replicate real hardware configurations.

The QEMU commandline to utilize this would be:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1,uid=x

However, if doing this how do we ensure UID is still unique? What do we do for cases where UID was not specified? One idea is to generate a string UID and just stick the bus # in there.

Quoted 43 lines
Quoted 4 lines

--

I'm guessing this patch will be somewhat controversial. For early CXL work, this can be dropped without too much heartache.

Whilst I'm not personally against, this maybe best to drop for now as you say.

Quoted 35 lines
---
 hw/i386/acpi-build.c                |  3 ++-
 hw/pci-bridge/pci_expander_bridge.c | 19 +++++++++++++++++++
 hw/pci/pci.c                        | 11 +++++++++++
 include/hw/pci/pci.h                |  1 +
 include/hw/pci/pci_bus.h            |  1 +
 5 files changed, 34 insertions(+), 1 deletion(-)
hw/i386/acpi-build.c +2 -1
index cf6eb54c22..145a503e92 100644
@@ -1343,6 +1343,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         QLIST_FOREACH(bus, &bus->child, sibling) {
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
+            int32_t uid = pci_bus_uid(bus);
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
@@ -1356,7 +1357,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
hw/pci-bridge/pci_expander_bridge.c +0 -0
index b42592e1ff..5021b60435 100644
@@ -67,6 +67,7 @@ struct PXBDev {
 
Quoted 6 lines

hw/pci-bridge/pci_expander_bridge.c struct PXBDev { · line 73

Quoted 6 lines
hw/pci-bridge/pci_expander_bridge.c +1 -0
     uint8_t bus_nr;
     uint16_t numa_node;
+    int32_t uid;
 };
 
 static PXBDev *convert_to_pxb(PCIDevice *dev)

As long as we are doing this, do we want to support a string uid too? How about a 64 bit uid? Why not?

Quoted 111 lines
Quoted 111 lines
hw/pci-bridge/pci_expander_bridge.c +18 -0
@@ -98,12 +99,20 @@ static uint16_t pxb_bus_numa_node(PCIBus *bus)
     return pxb->numa_node;
 }
 
+static int32_t pxb_bus_uid(PCIBus *bus)
+{
+    PXBDev *pxb = convert_to_pxb(bus->parent_dev);
+
+    return pxb->uid;
+}
+
 static void pxb_bus_class_init(ObjectClass *class, void *data)
 {
     PCIBusClass *pbc = PCI_BUS_CLASS(class);
 
     pbc->bus_num = pxb_bus_num;
     pbc->numa_node = pxb_bus_numa_node;
+    pbc->uid = pxb_bus_uid;
 }
 
 static const TypeInfo pxb_bus_info = {
@@ -329,6 +338,7 @@ static Property pxb_dev_properties[] = {
     /* Note: 0 is not a legal PXB bus number. */
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
+    DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -400,12 +410,21 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
+    PXBDev *pxb = convert_to_pxb(dev);
+
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
         error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
         return;
     }
 
+    if (pxb->uid < 0) {
+        error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");
+        return;
+    }
+
+    /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
+
     pxb_dev_realize_common(dev, CXL, errp);
 }
 
hw/pci/pci.c +11 -0
index adbe8aa260..bf019d91a0 100644
@@ -170,6 +170,11 @@ static uint16_t pcibus_numa_node(PCIBus *bus)
     return NUMA_NODE_UNASSIGNED;
 }
 
+static int32_t pcibus_uid(PCIBus *bus)
+{
+    return -1;
+}
+
 static void pci_bus_class_init(ObjectClass *klass, void *data)
 {
     BusClass *k = BUS_CLASS(klass);
@@ -184,6 +189,7 @@ static void pci_bus_class_init(ObjectClass *klass, void *data)
 
     pbc->bus_num = pcibus_num;
     pbc->numa_node = pcibus_numa_node;
+    pbc->uid = pcibus_uid;
 }
 
 static const TypeInfo pci_bus_info = {
@@ -530,6 +536,11 @@ int pci_bus_numa_node(PCIBus *bus)
     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
 }
 
+int pci_bus_uid(PCIBus *bus)
+{
+    return PCI_BUS_GET_CLASS(bus)->uid(bus);
+}
+
 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
                                  const VMStateField *field)
 {
include/hw/pci/pci.h +1 -0
index bde3697bee..a46de48ccd 100644
@@ -463,6 +463,7 @@ static inline int pci_dev_bus_num(const PCIDevice *dev)
 }
 
 int pci_bus_numa_node(PCIBus *bus);
+int pci_bus_uid(PCIBus *bus);
 void pci_for_each_device(PCIBus *bus, int bus_num,
                          void (*fn)(PCIBus *bus, PCIDevice *d, void *opaque),
                          void *opaque);
include/hw/pci/pci_bus.h +1 -0
index eb94e7e85c..3c9fbc55bb 100644
@@ -17,6 +17,7 @@ struct PCIBusClass {
 
     int (*bus_num)(PCIBus *bus);
     uint16_t (*numa_node)(PCIBus *bus);
+    int32_t (*uid)(PCIBus *bus);
 };
 
 enum PCIBusFlags {
#36 lore raw

Re: [RFC PATCH v3 16/31] hw/pci: Plumb _UID through host bridges

Thanks for looking! Mixing comments to Jonathan and Michael..

Quoted 4 lines · Michael S. Tsirkin
Quoted 3 lines · Jonathan Cameron

On Mon, 1 Feb 2021 16:59:33 -0800
Ben Widawsky <[email protected]> wrote:

Quoted 8 lines
Quoted 5 lines
Quoted 5 lines

Currently, QEMU makes _UID equivalent to the bus number (_BBN). While there is nothing wrong with doing it this way, CXL spec has a heavy reliance on _UID to identify host bridges and there is no link to the bus number. Having a distinct UID solves two problems. The first is it gets us around the limitation of 256 (current max bus number).

Not sure I understand. You want more than 256 host bridges?

I don't want more than 256 host bridges, but I want the ability to disaggregate _UID and bus (_BBN). The reasoning is just to align with the spec where _UID is used to identify a CXL host bridge which is unrelated (perhaps) to the bus number.

Quoted 6 lines
Quoted 3 lines

The

Quoted 2 lines

second is it allows us to replicate hardware configurations where bus number and uid aren't equivalent.

A bit more data on when this needs to be the case?

Doesn't *need* to be the case. I was making a concerted effort to allow full spec flexibility, but I don't believe it to be necessary unless we want to accurately emulate a real platform.

Quoted 10 lines
Quoted 10 lines

The latter has benefits for our

Quoted 9 lines

development and debugging using QEMU.

The other way to do this would be to implement the expanded bus numbering, but having an explicit uid makes more sense when trying to replicate real hardware configurations.

The QEMU commandline to utilize this would be:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1,uid=x

Quoted 6 lines
Quoted 1 line
Quoted 1 line

However, if doing this how do we ensure UID is still unique? What do we do for cases where UID was not specified? One idea is to generate a string UID and just stick the bus # in there.

This is totally mishandled in the code currently. I like your idea though.

Quoted 10 lines
Quoted 10 lines
Quoted 6 lines

--

I'm guessing this patch will be somewhat controversial. For early CXL work, this can be dropped without too much heartache.

Whilst I'm not personally against, this maybe best to drop for now as you say.

I think it'd be good to understand from the PCIe experts if CXL matches in this regard. If PCIe generally allows (and does in practice) _UID not matching _BBN, perhaps this is an overall improvement to the code.

Quoted 40 lines
Quoted 40 lines
Quoted 40 lines
---
 hw/i386/acpi-build.c                |  3 ++-
 hw/pci-bridge/pci_expander_bridge.c | 19 +++++++++++++++++++
 hw/pci/pci.c                        | 11 +++++++++++
 include/hw/pci/pci.h                |  1 +
 include/hw/pci/pci_bus.h            |  1 +
 5 files changed, 34 insertions(+), 1 deletion(-)
hw/i386/acpi-build.c +2 -1
index cf6eb54c22..145a503e92 100644
@@ -1343,6 +1343,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         QLIST_FOREACH(bus, &bus->child, sibling) {
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
+            int32_t uid = pci_bus_uid(bus);
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
@@ -1356,7 +1357,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
hw/pci-bridge/pci_expander_bridge.c +1 -0
index b42592e1ff..5021b60435 100644
@@ -67,6 +67,7 @@ struct PXBDev {
 
     uint8_t bus_nr;
     uint16_t numa_node;
+    int32_t uid;
 };
 
Quoted 4 lines
Quoted 1 line

hw/pci-bridge/pci_expander_bridge.c struct PXBDev { · line 73

Quoted 1 line
hw/pci-bridge/pci_expander_bridge.c +0 -0
 static PXBDev *convert_to_pxb(PCIDevice *dev)

As long as we are doing this, do we want to support a string uid too? How about a 64 bit uid? Why not?

If generally the idea of the patch is welcome, I am happy to change it.

Quoted 114 lines
Quoted 113 lines
Quoted 113 lines
hw/pci-bridge/pci_expander_bridge.c +18 -0
@@ -98,12 +99,20 @@ static uint16_t pxb_bus_numa_node(PCIBus *bus)
     return pxb->numa_node;
 }
 
+static int32_t pxb_bus_uid(PCIBus *bus)
+{
+    PXBDev *pxb = convert_to_pxb(bus->parent_dev);
+
+    return pxb->uid;
+}
+
 static void pxb_bus_class_init(ObjectClass *class, void *data)
 {
     PCIBusClass *pbc = PCI_BUS_CLASS(class);
 
     pbc->bus_num = pxb_bus_num;
     pbc->numa_node = pxb_bus_numa_node;
+    pbc->uid = pxb_bus_uid;
 }
 
 static const TypeInfo pxb_bus_info = {
@@ -329,6 +338,7 @@ static Property pxb_dev_properties[] = {
     /* Note: 0 is not a legal PXB bus number. */
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
+    DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -400,12 +410,21 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
+    PXBDev *pxb = convert_to_pxb(dev);
+
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
         error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
         return;
     }
 
+    if (pxb->uid < 0) {
+        error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");
+        return;
+    }
+
+    /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
+
     pxb_dev_realize_common(dev, CXL, errp);
 }
 
hw/pci/pci.c +11 -0
index adbe8aa260..bf019d91a0 100644
@@ -170,6 +170,11 @@ static uint16_t pcibus_numa_node(PCIBus *bus)
     return NUMA_NODE_UNASSIGNED;
 }
 
+static int32_t pcibus_uid(PCIBus *bus)
+{
+    return -1;
+}
+
 static void pci_bus_class_init(ObjectClass *klass, void *data)
 {
     BusClass *k = BUS_CLASS(klass);
@@ -184,6 +189,7 @@ static void pci_bus_class_init(ObjectClass *klass, void *data)
 
     pbc->bus_num = pcibus_num;
     pbc->numa_node = pcibus_numa_node;
+    pbc->uid = pcibus_uid;
 }
 
 static const TypeInfo pci_bus_info = {
@@ -530,6 +536,11 @@ int pci_bus_numa_node(PCIBus *bus)
     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
 }
 
+int pci_bus_uid(PCIBus *bus)
+{
+    return PCI_BUS_GET_CLASS(bus)->uid(bus);
+}
+
 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
                                  const VMStateField *field)
 {
include/hw/pci/pci.h +1 -0
index bde3697bee..a46de48ccd 100644
@@ -463,6 +463,7 @@ static inline int pci_dev_bus_num(const PCIDevice *dev)
 }
 
 int pci_bus_numa_node(PCIBus *bus);
+int pci_bus_uid(PCIBus *bus);
 void pci_for_each_device(PCIBus *bus, int bus_num,
                          void (*fn)(PCIBus *bus, PCIDevice *d, void *opaque),
                          void *opaque);
include/hw/pci/pci_bus.h +1 -0
index eb94e7e85c..3c9fbc55bb 100644
@@ -17,6 +17,7 @@ struct PCIBusClass {
 
     int (*bus_num)(PCIBus *bus);
     uint16_t (*numa_node)(PCIBus *bus);
+    int32_t (*uid)(PCIBus *bus);
 };
 
 enum PCIBusFlags {
#37 lore raw

Re: [RFC PATCH v3 16/31] hw/pci: Plumb _UID through host bridges

Quoted 13 lines · Ben Widawsky

Thanks for looking! Mixing comments to Jonathan and Michael..

Quoted 10 lines · Michael S. Tsirkin
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Mon, 1 Feb 2021 16:59:33 -0800 Ben Widawsky <[email protected]> wrote:

Currently, QEMU makes _UID equivalent to the bus number (_BBN). While there is nothing wrong with doing it this way, CXL spec has a heavy reliance on _UID to identify host bridges and there is no link to the bus number. Having a distinct UID solves two problems. The first is it gets us around the limitation of 256 (current max bus number).

Quoted 7 lines
Quoted 2 lines

Not sure I understand. You want more than 256 host bridges?

I don't want more than 256 host bridges, but I want the ability to disaggregate _UID and bus (_BBN). The reasoning is just to align with the spec where _UID is used to identify a CXL host bridge which is unrelated (perhaps) to the bus number.

Which spec is that?

Quoted 37 lines
Quoted 6 lines
Quoted 3 lines

The

Quoted 2 lines

second is it allows us to replicate hardware configurations where bus number and uid aren't equivalent.

A bit more data on when this needs to be the case?

Doesn't *need* to be the case. I was making a concerted effort to allow full spec flexibility, but I don't believe it to be necessary unless we want to accurately emulate a real platform.

Quoted 10 lines
Quoted 10 lines

The latter has benefits for our

Quoted 9 lines

development and debugging using QEMU.

The other way to do this would be to implement the expanded bus numbering, but having an explicit uid makes more sense when trying to replicate real hardware configurations.

The QEMU commandline to utilize this would be:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1,uid=x

Quoted 6 lines
Quoted 1 line
Quoted 1 line

However, if doing this how do we ensure UID is still unique? What do we do for cases where UID was not specified? One idea is to generate a string UID and just stick the bus # in there.

This is totally mishandled in the code currently. I like your idea though.

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines

--

I'm guessing this patch will be somewhat controversial. For early CXL work, this can be dropped without too much heartache.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines

Whilst I'm not personally against, this maybe best to drop for now as you say.

I think it'd be good to understand from the PCIe experts if CXL matches in this regard. If PCIe generally allows (and does in practice) _UID not matching _BBN, perhaps this is an overall improvement to the code.

Well

6.1.12 _UID (Unique ID)
This object provides OSPM with a logical device ID that does not change across reboots. This object is optional, but is required when the device has no other way to report a persistent unique device ID. The _UID must be unique across all devices with either a common _HID or _CID. This is because a device needs to be uniquely identified to the OSPM, which may match on either a _HID or a _CID to identify the device. The uniqueness match must be true regardless of whether the OSPM uses the _HID or the _CID. OSPM typically uses the unique device ID to ensure that the device-specific information, such as network protocol binding information, is remembered for the device even if its relative location changes. For most integrated devices, this object contains a unique identifier. A _UID object evaluates to either a numeric value or a string.

IOW _UID is there so guest can tell devices with identical HID/CID apart.

Quoted 40 lines
Quoted 40 lines
Quoted 40 lines
Quoted 40 lines
---
 hw/i386/acpi-build.c                |  3 ++-
 hw/pci-bridge/pci_expander_bridge.c | 19 +++++++++++++++++++
 hw/pci/pci.c                        | 11 +++++++++++
 include/hw/pci/pci.h                |  1 +
 include/hw/pci/pci_bus.h            |  1 +
 5 files changed, 34 insertions(+), 1 deletion(-)
hw/i386/acpi-build.c +2 -1
index cf6eb54c22..145a503e92 100644
@@ -1343,6 +1343,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         QLIST_FOREACH(bus, &bus->child, sibling) {
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
+            int32_t uid = pci_bus_uid(bus);
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
@@ -1356,7 +1357,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
hw/pci-bridge/pci_expander_bridge.c +1 -0
index b42592e1ff..5021b60435 100644
@@ -67,6 +67,7 @@ struct PXBDev {
 
     uint8_t bus_nr;
     uint16_t numa_node;
+    int32_t uid;
 };
 
Quoted 6 lines
Quoted 4 lines
Quoted 1 line

hw/pci-bridge/pci_expander_bridge.c struct PXBDev { · line 73

Quoted 1 line
hw/pci-bridge/pci_expander_bridge.c +0 -0
 static PXBDev *convert_to_pxb(PCIDevice *dev)

As long as we are doing this, do we want to support a string uid too? How about a 64 bit uid? Why not?

If generally the idea of the patch is welcome, I am happy to change it.

I am still not sure, let's figure out the motivation.

For example, grep for UID and you will find more cases like this. E.g. hw/pci-host/gpex-acpi.c

Do we care?

Quoted 114 lines
Quoted 114 lines
Quoted 113 lines
Quoted 113 lines
hw/pci-bridge/pci_expander_bridge.c +18 -0
@@ -98,12 +99,20 @@ static uint16_t pxb_bus_numa_node(PCIBus *bus)
     return pxb->numa_node;
 }
 
+static int32_t pxb_bus_uid(PCIBus *bus)
+{
+    PXBDev *pxb = convert_to_pxb(bus->parent_dev);
+
+    return pxb->uid;
+}
+
 static void pxb_bus_class_init(ObjectClass *class, void *data)
 {
     PCIBusClass *pbc = PCI_BUS_CLASS(class);
 
     pbc->bus_num = pxb_bus_num;
     pbc->numa_node = pxb_bus_numa_node;
+    pbc->uid = pxb_bus_uid;
 }
 
 static const TypeInfo pxb_bus_info = {
@@ -329,6 +338,7 @@ static Property pxb_dev_properties[] = {
     /* Note: 0 is not a legal PXB bus number. */
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
+    DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -400,12 +410,21 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
+    PXBDev *pxb = convert_to_pxb(dev);
+
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
         error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
         return;
     }
 
+    if (pxb->uid < 0) {
+        error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");
+        return;
+    }
+
+    /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
+
     pxb_dev_realize_common(dev, CXL, errp);
 }
 
hw/pci/pci.c +11 -0
index adbe8aa260..bf019d91a0 100644
@@ -170,6 +170,11 @@ static uint16_t pcibus_numa_node(PCIBus *bus)
     return NUMA_NODE_UNASSIGNED;
 }
 
+static int32_t pcibus_uid(PCIBus *bus)
+{
+    return -1;
+}
+
 static void pci_bus_class_init(ObjectClass *klass, void *data)
 {
     BusClass *k = BUS_CLASS(klass);
@@ -184,6 +189,7 @@ static void pci_bus_class_init(ObjectClass *klass, void *data)
 
     pbc->bus_num = pcibus_num;
     pbc->numa_node = pcibus_numa_node;
+    pbc->uid = pcibus_uid;
 }
 
 static const TypeInfo pci_bus_info = {
@@ -530,6 +536,11 @@ int pci_bus_numa_node(PCIBus *bus)
     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
 }
 
+int pci_bus_uid(PCIBus *bus)
+{
+    return PCI_BUS_GET_CLASS(bus)->uid(bus);
+}
+
 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
                                  const VMStateField *field)
 {
include/hw/pci/pci.h +1 -0
index bde3697bee..a46de48ccd 100644
@@ -463,6 +463,7 @@ static inline int pci_dev_bus_num(const PCIDevice *dev)
 }
 
 int pci_bus_numa_node(PCIBus *bus);
+int pci_bus_uid(PCIBus *bus);
 void pci_for_each_device(PCIBus *bus, int bus_num,
                          void (*fn)(PCIBus *bus, PCIDevice *d, void *opaque),
                          void *opaque);
include/hw/pci/pci_bus.h +1 -0
index eb94e7e85c..3c9fbc55bb 100644
@@ -17,6 +17,7 @@ struct PCIBusClass {
 
     int (*bus_num)(PCIBus *bus);
     uint16_t (*numa_node)(PCIBus *bus);
+    int32_t (*uid)(PCIBus *bus);
 };
 
 enum PCIBusFlags {
#38 lore raw

Re: [RFC PATCH v3 16/31] hw/pci: Plumb _UID through host bridges

Quoted 17 lines · Michael S. Tsirkin
Quoted 16 lines · Ben Widawsky

Thanks for looking! Mixing comments to Jonathan and Michael..

Quoted 12 lines · Michael S. Tsirkin
Quoted 8 lines · Jonathan Cameron
Quoted 5 lines · On Mon, 1 Feb 2021 16:59:33 -0800 Ben Widawsky <[email protected]> wrote:

Currently, QEMU makes _UID equivalent to the bus number (_BBN). While there is nothing wrong with doing it this way, CXL spec has a heavy reliance on _UID to identify host bridges and there is no link to the bus number. Having a distinct UID solves two problems. The first is it gets us around the limitation of 256 (current max bus number).

Not sure I understand. You want more than 256 host bridges?

Quoted 7 lines
Quoted 4 lines

I don't want more than 256 host bridges, but I want the ability to disaggregate _UID and bus (_BBN). The reasoning is just to align with the spec where _UID is used to identify a CXL host bridge which is unrelated (perhaps) to the bus number.

Which spec is that?

CXL spec
https://www.computeexpresslink.org/download-the-specification

The spec introduces a new ACPI table which links information about a host bridge based on the _UID. The _UID is 4 bytes.

"In an ACPI compliant system, there shall be one instance of CXL Host Bridge Device object in ACPI namespace (HID=”ACPI0016”) for every CHBS entry. The _UID object under a CXL Host Bridge object, when evaluated, shall match the UID field in the associated CHBS entry."

"CXL Host Bridge Unique ID. Used to associate a CHBS instance with CXL Host Bridge instance. The value of this field shall match the output of _UID under the associated CXL Host Bridge in ACPI namespace"

Quoted 59 lines
Quoted 37 lines
Quoted 6 lines
Quoted 3 lines

The

Quoted 2 lines

second is it allows us to replicate hardware configurations where bus number and uid aren't equivalent.

A bit more data on when this needs to be the case?

Doesn't *need* to be the case. I was making a concerted effort to allow full spec flexibility, but I don't believe it to be necessary unless we want to accurately emulate a real platform.

Quoted 10 lines
Quoted 10 lines

The latter has benefits for our

Quoted 9 lines

development and debugging using QEMU.

The other way to do this would be to implement the expanded bus numbering, but having an explicit uid makes more sense when trying to replicate real hardware configurations.

The QEMU commandline to utilize this would be:
-device pxb-cxl,id=cxl.0,bus="pcie.0",bus_nr=1,uid=x

Quoted 6 lines
Quoted 1 line
Quoted 1 line

However, if doing this how do we ensure UID is still unique? What do we do for cases where UID was not specified? One idea is to generate a string UID and just stick the bus # in there.

This is totally mishandled in the code currently. I like your idea though.

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines

--

I'm guessing this patch will be somewhat controversial. For early CXL work, this can be dropped without too much heartache.

Quoted 7 lines
Quoted 3 lines
Quoted 3 lines

Whilst I'm not personally against, this maybe best to drop for now as you say.

I think it'd be good to understand from the PCIe experts if CXL matches in this regard. If PCIe generally allows (and does in practice) _UID not matching _BBN, perhaps this is an overall improvement to the code.

Well

6.1.12 _UID (Unique ID)
This object provides OSPM with a logical device ID that does not change across reboots. This object is optional, but is required when the device has no other way to report a persistent unique device ID. The _UID must be unique across all devices with either a common _HID or _CID. This is because a device needs to be uniquely identified to the OSPM, which may match on either a _HID or a _CID to identify the device. The uniqueness match must be true regardless of whether the OSPM uses the _HID or the _CID. OSPM typically uses the unique device ID to ensure that the device-specific information, such as network protocol binding information, is remembered for the device even if its relative location changes. For most integrated devices, this object contains a unique identifier. A _UID object evaluates to either a numeric value or a string.

Quoted 2 lines

IOW _UID is there so guest can tell devices with identical HID/CID apart.

Right, so it's pretty much the same thing. There's ancillary data about the CXL host bridge that is identified by the _UID.

Quoted 48 lines
Quoted 48 lines
Quoted 47 lines
Quoted 38 lines
Quoted 38 lines


---
 hw/i386/acpi-build.c                |  3 ++-
 hw/pci-bridge/pci_expander_bridge.c | 19 +++++++++++++++++++
 hw/pci/pci.c                        | 11 +++++++++++
 include/hw/pci/pci.h                |  1 +
 include/hw/pci/pci_bus.h            |  1 +
 5 files changed, 34 insertions(+), 1 deletion(-)
hw/i386/acpi-build.c +2 -1
index cf6eb54c22..145a503e92 100644
@@ -1343,6 +1343,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         QLIST_FOREACH(bus, &bus->child, sibling) {
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
+            int32_t uid = pci_bus_uid(bus);
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
@@ -1356,7 +1357,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             scope = aml_scope("\\_SB");
             dev = aml_device("PC%.02X", bus_num);
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, bus_num, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
hw/pci-bridge/pci_expander_bridge.c +0 -0
index b42592e1ff..5021b60435 100644
@@ -67,6 +67,7 @@ struct PXBDev {
 
Quoted 6 lines

hw/pci-bridge/pci_expander_bridge.c struct PXBDev { · line 73

Quoted 6 lines
hw/pci-bridge/pci_expander_bridge.c +1 -0
     uint8_t bus_nr;
     uint16_t numa_node;
+    int32_t uid;
 };
 
 static PXBDev *convert_to_pxb(PCIDevice *dev)

As long as we are doing this, do we want to support a string uid too? How about a 64 bit uid? Why not?

Quoted 8 lines
Quoted 1 line

If generally the idea of the patch is welcome, I am happy to change it.

I am still not sure, let's figure out the motivation.

For example, grep for UID and you will find more cases like this. E.g. hw/pci-host/gpex-acpi.c

Do we care?

For my purposes, I do not.

Quoted 118 lines
Quoted 117 lines
Quoted 117 lines
Quoted 116 lines
Quoted 116 lines
hw/pci-bridge/pci_expander_bridge.c +18 -0
@@ -98,12 +99,20 @@ static uint16_t pxb_bus_numa_node(PCIBus *bus)
     return pxb->numa_node;
 }
 
+static int32_t pxb_bus_uid(PCIBus *bus)
+{
+    PXBDev *pxb = convert_to_pxb(bus->parent_dev);
+
+    return pxb->uid;
+}
+
 static void pxb_bus_class_init(ObjectClass *class, void *data)
 {
     PCIBusClass *pbc = PCI_BUS_CLASS(class);
 
     pbc->bus_num = pxb_bus_num;
     pbc->numa_node = pxb_bus_numa_node;
+    pbc->uid = pxb_bus_uid;
 }
 
 static const TypeInfo pxb_bus_info = {
@@ -329,6 +338,7 @@ static Property pxb_dev_properties[] = {
     /* Note: 0 is not a legal PXB bus number. */
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
+    DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -400,12 +410,21 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
+    PXBDev *pxb = convert_to_pxb(dev);
+
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
         error_setg(errp, "pxb-cxl devices cannot reside on a PCI bus");
         return;
     }
 
+    if (pxb->uid < 0) {
+        error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");
+        return;
+    }
+
+    /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
+
     pxb_dev_realize_common(dev, CXL, errp);
 }
 
hw/pci/pci.c +11 -0
index adbe8aa260..bf019d91a0 100644
@@ -170,6 +170,11 @@ static uint16_t pcibus_numa_node(PCIBus *bus)
     return NUMA_NODE_UNASSIGNED;
 }
 
+static int32_t pcibus_uid(PCIBus *bus)
+{
+    return -1;
+}
+
 static void pci_bus_class_init(ObjectClass *klass, void *data)
 {
     BusClass *k = BUS_CLASS(klass);
@@ -184,6 +189,7 @@ static void pci_bus_class_init(ObjectClass *klass, void *data)
 
     pbc->bus_num = pcibus_num;
     pbc->numa_node = pcibus_numa_node;
+    pbc->uid = pcibus_uid;
 }
 
 static const TypeInfo pci_bus_info = {
@@ -530,6 +536,11 @@ int pci_bus_numa_node(PCIBus *bus)
     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
 }
 
+int pci_bus_uid(PCIBus *bus)
+{
+    return PCI_BUS_GET_CLASS(bus)->uid(bus);
+}
+
 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
                                  const VMStateField *field)
 {
include/hw/pci/pci.h +1 -0
index bde3697bee..a46de48ccd 100644
@@ -463,6 +463,7 @@ static inline int pci_dev_bus_num(const PCIDevice *dev)
 }
 
 int pci_bus_numa_node(PCIBus *bus);
+int pci_bus_uid(PCIBus *bus);
 void pci_for_each_device(PCIBus *bus, int bus_num,
                          void (*fn)(PCIBus *bus, PCIDevice *d, void *opaque),
                          void *opaque);
include/hw/pci/pci_bus.h +1 -0
index eb94e7e85c..3c9fbc55bb 100644
@@ -17,6 +17,7 @@ struct PCIBusClass {
 
     int (*bus_num)(PCIBus *bus);
     uint16_t (*numa_node)(PCIBus *bus);
+    int32_t (*uid)(PCIBus *bus);
 };
 
 enum PCIBusFlags {
#39 lore raw

RFC PATCH v3 17/31 hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)

In reply to Ben Widawsky

CXL host bridges themselves may have MMIO. Since host bridges don't have a BAR they are treated as special for MMIO.

--

It's arbitrarily chosen here to pick 0xD0000000 as the base for the host bridge MMIO. I'm not sure what the right way to find free space for platform hardcoded things like this is.

---
 hw/pci-bridge/pci_expander_bridge.c | 53 ++++++++++++++++++++++++++++-
 include/hw/cxl/cxl.h                |  2 ++
 2 files changed, 54 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +52 -1
index 5021b60435..226a8a5fff 100644
@@ -17,6 +17,7 @@
 #include "hw/pci/pci_host.h"
 #include "hw/qdev-properties.h"
 #include "hw/pci/pci_bridge.h"
+#include "hw/cxl/cxl.h"
 #include "qemu/range.h"
 #include "qemu/error-report.h"
 #include "qemu/module.h"
@@ -70,6 +71,12 @@ struct PXBDev {
     int32_t uid;
 };
 
+typedef struct CXLHost {
+    PCIHostState parent_obj;
+
+    CXLComponentState cxl_cstate;
+} CXLHost;
+
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
     /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
@@ -85,6 +92,9 @@ static GList *pxb_dev_list;
 
 #define TYPE_PXB_HOST "pxb-host"
 
+#define TYPE_PXB_CXL_HOST "pxb-cxl-host"
+#define PXB_CXL_HOST(obj) OBJECT_CHECK(CXLHost, (obj), TYPE_PXB_CXL_HOST)
+
 static int pxb_bus_num(PCIBus *bus)
 {
     PXBDev *pxb = convert_to_pxb(bus->parent_dev);
@@ -198,6 +208,46 @@ static const TypeInfo pxb_host_info = {
     .class_init    = pxb_host_class_init,
 };
 
+static void pxb_cxl_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    PCIHostState *phb = PCI_HOST_BRIDGE(dev);
+    CXLHost *cxl = PXB_CXL_HOST(dev);
+    CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
+    struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+
+    cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
+                                      TYPE_PXB_CXL_HOST);
+    sysbus_init_mmio(sbd, mr);
+
+    /* FIXME: support multiple host bridges. */
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
+                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+}
+
+static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(class);
+    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(class);
+
+    hc->root_bus_path = pxb_host_root_bus_path;
+    dc->fw_name = "cxl";
+    dc->realize = pxb_cxl_realize;
+    /* Reason: Internal part of the pxb/pxb-pcie device, not usable by itself */
+    dc->user_creatable = false;
+}
+
+/*
+ * This is a device to handle the MMIO for a CXL host bridge. It does nothing
+ * else.
+ */
+static const TypeInfo cxl_host_info = {
+    .name          = TYPE_PXB_CXL_HOST,
+    .parent        = TYPE_PCI_HOST_BRIDGE,
+    .instance_size = sizeof(CXLHost),
+    .class_init    = pxb_cxl_host_class_init,
+};
+
 /*
  * Registers the PXB bus as a child of pci host root bus.
  */
@@ -272,7 +322,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         dev_name = dev->qdev.id;
     }
 
-    ds = qdev_new(TYPE_PXB_HOST);
+    ds = qdev_new(type == CXL ? TYPE_PXB_CXL_HOST : TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else if (type == CXL) {
@@ -466,6 +516,7 @@ static void pxb_register_types(void)
     type_register_static(&pxb_pcie_bus_info);
     type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
+    type_register_static(&cxl_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
     type_register_static(&pxb_cxl_dev_info);
include/hw/cxl/cxl.h +2 -0
index 362cda40de..6bc344f205 100644
@@ -17,5 +17,7 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define CXL_HOST_BASE 0xD0000000
+
 #endif
 
Signature · 1 line
2.30.0
#40 lore raw

Re: [RFC PATCH v3 17/31] hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)

Quoted 10 lines · On Mon, 1 Feb 2021 16:59:34 -0800 Ben Widawsky <[email protected]> wrote:

CXL host bridges themselves may have MMIO. Since host bridges don't have a BAR they are treated as special for MMIO.

--

It's arbitrarily chosen here to pick 0xD0000000 as the base for the host bridge MMIO. I'm not sure what the right way to find free space for platform hardcoded things like this is.

Seems like this needs to come from the machine definition.
This is fairly easy for arm/virt, where there is a clearly laid out memory map. For hw/i386 I'm less sure on how to do it.

Having said that, for this particular magic device, we do have a PCI EP associated with it. How about putting all the host bridge MMIO into a BAR of that rather than having it separate.
That has the added advantage of making it discoverable from firmware.

Any normal system is going to have this is impdef for discovery anyway.

That would then let you drop the separate definition of CXLHost structure though it needs a bit of figuring out what to do with the memory window setup etc.

I tried hacking it together, but not gotten it working yet.

Quoted 116 lines
---
 hw/pci-bridge/pci_expander_bridge.c | 53 ++++++++++++++++++++++++++++-
 include/hw/cxl/cxl.h                |  2 ++
 2 files changed, 54 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +52 -1
index 5021b60435..226a8a5fff 100644
@@ -17,6 +17,7 @@
 #include "hw/pci/pci_host.h"
 #include "hw/qdev-properties.h"
 #include "hw/pci/pci_bridge.h"
+#include "hw/cxl/cxl.h"
 #include "qemu/range.h"
 #include "qemu/error-report.h"
 #include "qemu/module.h"
@@ -70,6 +71,12 @@ struct PXBDev {
     int32_t uid;
 };
 
+typedef struct CXLHost {
+    PCIHostState parent_obj;
+
+    CXLComponentState cxl_cstate;
+} CXLHost;
+
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
     /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
@@ -85,6 +92,9 @@ static GList *pxb_dev_list;
 
 #define TYPE_PXB_HOST "pxb-host"
 
+#define TYPE_PXB_CXL_HOST "pxb-cxl-host"
+#define PXB_CXL_HOST(obj) OBJECT_CHECK(CXLHost, (obj), TYPE_PXB_CXL_HOST)
+
 static int pxb_bus_num(PCIBus *bus)
 {
     PXBDev *pxb = convert_to_pxb(bus->parent_dev);
@@ -198,6 +208,46 @@ static const TypeInfo pxb_host_info = {
     .class_init    = pxb_host_class_init,
 };
 
+static void pxb_cxl_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    PCIHostState *phb = PCI_HOST_BRIDGE(dev);
+    CXLHost *cxl = PXB_CXL_HOST(dev);
+    CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
+    struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+
+    cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
+                                      TYPE_PXB_CXL_HOST);
+    sysbus_init_mmio(sbd, mr);
+
+    /* FIXME: support multiple host bridges. */
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
+                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+}
+
+static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(class);
+    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(class);
+
+    hc->root_bus_path = pxb_host_root_bus_path;
+    dc->fw_name = "cxl";
+    dc->realize = pxb_cxl_realize;
+    /* Reason: Internal part of the pxb/pxb-pcie device, not usable by itself */
+    dc->user_creatable = false;
+}
+
+/*
+ * This is a device to handle the MMIO for a CXL host bridge. It does nothing
+ * else.
+ */
+static const TypeInfo cxl_host_info = {
+    .name          = TYPE_PXB_CXL_HOST,
+    .parent        = TYPE_PCI_HOST_BRIDGE,
+    .instance_size = sizeof(CXLHost),
+    .class_init    = pxb_cxl_host_class_init,
+};
+
 /*
  * Registers the PXB bus as a child of pci host root bus.
  */
@@ -272,7 +322,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         dev_name = dev->qdev.id;
     }
 
-    ds = qdev_new(TYPE_PXB_HOST);
+    ds = qdev_new(type == CXL ? TYPE_PXB_CXL_HOST : TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else if (type == CXL) {
@@ -466,6 +516,7 @@ static void pxb_register_types(void)
     type_register_static(&pxb_pcie_bus_info);
     type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
+    type_register_static(&cxl_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
     type_register_static(&pxb_cxl_dev_info);
include/hw/cxl/cxl.h +2 -0
index 362cda40de..6bc344f205 100644
@@ -17,5 +17,7 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define CXL_HOST_BASE 0xD0000000
+
 #endif
 
#41 lore raw

Re: [RFC PATCH v3 17/31] hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)

Quoted 10 lines · Jonathan Cameron
Quoted 7 lines · On Mon, 1 Feb 2021 16:59:34 -0800 Ben Widawsky <[email protected]> wrote:

CXL host bridges themselves may have MMIO. Since host bridges don't have a BAR they are treated as special for MMIO.

--

Quoted 7 lines
Quoted 3 lines

It's arbitrarily chosen here to pick 0xD0000000 as the base for the host bridge MMIO. I'm not sure what the right way to find free space for platform hardcoded things like this is.

Seems like this needs to come from the machine definition.
This is fairly easy for arm/virt, where there is a clearly laid out memory map. For hw/i386 I'm less sure on how to do it.

I think this is how to do it :D

Quoted 8 lines

Having said that, for this particular magic device, we do have a PCI EP associated with it. How about putting all the host bridge MMIO into a BAR of that rather than having it separate.
That has the added advantage of making it discoverable from firmware.

Any normal system is going to have this is impdef for discovery anyway.

This is not how it's expected to work for Intel at least. If the device was discoverable you wouldn't need CEDT/CHBS. The magic host bridges are only advertised via the CEDT.

When I build and run QEMU for x86_64, I do not see the host bridge in the pci topology, do you (it's meant to not be there)?

00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
...
34:00.0 PCI bridge: Intel Corporation Device 7075
35:00.0 Memory controller [0502]: Intel Corporation Device 0d93 (rev 01)

That's Q35, Root Port, and Type 3 device respectively.

Quoted 124 lines

That would then let you drop the separate definition of CXLHost structure though it needs a bit of figuring out what to do with the memory window setup etc.

I tried hacking it together, but not gotten it working yet.

Quoted 116 lines
---
 hw/pci-bridge/pci_expander_bridge.c | 53 ++++++++++++++++++++++++++++-
 include/hw/cxl/cxl.h                |  2 ++
 2 files changed, 54 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +52 -1
index 5021b60435..226a8a5fff 100644
@@ -17,6 +17,7 @@
 #include "hw/pci/pci_host.h"
 #include "hw/qdev-properties.h"
 #include "hw/pci/pci_bridge.h"
+#include "hw/cxl/cxl.h"
 #include "qemu/range.h"
 #include "qemu/error-report.h"
 #include "qemu/module.h"
@@ -70,6 +71,12 @@ struct PXBDev {
     int32_t uid;
 };
 
+typedef struct CXLHost {
+    PCIHostState parent_obj;
+
+    CXLComponentState cxl_cstate;
+} CXLHost;
+
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
     /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
@@ -85,6 +92,9 @@ static GList *pxb_dev_list;
 
 #define TYPE_PXB_HOST "pxb-host"
 
+#define TYPE_PXB_CXL_HOST "pxb-cxl-host"
+#define PXB_CXL_HOST(obj) OBJECT_CHECK(CXLHost, (obj), TYPE_PXB_CXL_HOST)
+
 static int pxb_bus_num(PCIBus *bus)
 {
     PXBDev *pxb = convert_to_pxb(bus->parent_dev);
@@ -198,6 +208,46 @@ static const TypeInfo pxb_host_info = {
     .class_init    = pxb_host_class_init,
 };
 
+static void pxb_cxl_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    PCIHostState *phb = PCI_HOST_BRIDGE(dev);
+    CXLHost *cxl = PXB_CXL_HOST(dev);
+    CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
+    struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+
+    cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
+                                      TYPE_PXB_CXL_HOST);
+    sysbus_init_mmio(sbd, mr);
+
+    /* FIXME: support multiple host bridges. */
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
+                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+}
+
+static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(class);
+    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(class);
+
+    hc->root_bus_path = pxb_host_root_bus_path;
+    dc->fw_name = "cxl";
+    dc->realize = pxb_cxl_realize;
+    /* Reason: Internal part of the pxb/pxb-pcie device, not usable by itself */
+    dc->user_creatable = false;
+}
+
+/*
+ * This is a device to handle the MMIO for a CXL host bridge. It does nothing
+ * else.
+ */
+static const TypeInfo cxl_host_info = {
+    .name          = TYPE_PXB_CXL_HOST,
+    .parent        = TYPE_PCI_HOST_BRIDGE,
+    .instance_size = sizeof(CXLHost),
+    .class_init    = pxb_cxl_host_class_init,
+};
+
 /*
  * Registers the PXB bus as a child of pci host root bus.
  */
@@ -272,7 +322,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         dev_name = dev->qdev.id;
     }
 
-    ds = qdev_new(TYPE_PXB_HOST);
+    ds = qdev_new(type == CXL ? TYPE_PXB_CXL_HOST : TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else if (type == CXL) {
@@ -466,6 +516,7 @@ static void pxb_register_types(void)
     type_register_static(&pxb_pcie_bus_info);
     type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
+    type_register_static(&cxl_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
     type_register_static(&pxb_cxl_dev_info);
include/hw/cxl/cxl.h +2 -0
index 362cda40de..6bc344f205 100644
@@ -17,5 +17,7 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define CXL_HOST_BASE 0xD0000000
+
 #endif
 
#42 lore raw

Re: [RFC PATCH v3 17/31] hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)

Quoted 15 lines · On Tue, 2 Feb 2021 11:45:05 -0800 Ben Widawsky <[email protected]> wrote:
Quoted 14 lines · Jonathan Cameron
Quoted 10 lines · On Mon, 1 Feb 2021 16:59:34 -0800 Ben Widawsky <[email protected]> wrote:

CXL host bridges themselves may have MMIO. Since host bridges don't have a BAR they are treated as special for MMIO.

--

It's arbitrarily chosen here to pick 0xD0000000 as the base for the host bridge MMIO. I'm not sure what the right way to find free space for platform hardcoded things like this is.

Quoted 5 lines
Quoted 3 lines

Seems like this needs to come from the machine definition.
This is fairly easy for arm/virt, where there is a clearly laid out memory map. For hw/i386 I'm less sure on how to do it.

I think this is how to do it :D

It may well be, but they you'll need to find a suitable region and document it and ensure no one else ever tramples on it. Easy to do on a physical system, bit trickier in emulation.

Quoted 7 lines
Quoted 7 lines

Having said that, for this particular magic device, we do have a PCI EP associated with it. How about putting all the host bridge MMIO into a BAR of that rather than having it separate.
That has the added advantage of making it discoverable from firmware.

Quoted 6 lines
Quoted 2 lines

Any normal system is going to have this is impdef for discovery anyway.

This is not how it's expected to work for Intel at least. If the device was discoverable you wouldn't need CEDT/CHBS. The magic host bridges are only advertised via the CEDT.

I agree on a normal system (i.e. a real one) this doesn't work, but then a normal system doesn't involve a magic PCI RCiEP that also happens to instantiate an extra host bridge. This is what pxb_pcie is doing and what your pxb_cxl is almost doing.

Quoted 10 lines

When I build and run QEMU for x86_64, I do not see the host bridge in the pci topology, do you (it's meant to not be there)?

00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
...
34:00.0 PCI bridge: Intel Corporation Device 7075
35:00.0 Memory controller [0502]: Intel Corporation Device 0d93 (rev 01)

That's Q35, Root Port, and Type 3 device respectively.

You don't see the host bridge, for pxb_cxl, but for pxb_pcie you do. 00:06.0 Host bridge: Red Hat, Inc QEMU PCIe Expander bridge.
If you have another device after your pxb-cxl you'll also notice that there is a hole punched in the list where you'd expect pxb-cxl to be (device number skipped). (that had me confused earlier).

This seems to be because no VID etc (unlike pxb-pcie).

I gave vendor and device IDs (and a bar to test that could be done) and it then appears just like pxb_pcie does. Hence handy place to hang our magic memory off so that EDK2 or similar can work with it and indeed construct he CHBS as needed so we can get to this via the same paths as a normal system. It's a bit convoluted but in some ways more elegant.

Jonathan

Quoted 126 lines
Quoted 125 lines

That would then let you drop the separate definition of CXLHost structure though it needs a bit of figuring out what to do with the memory window setup etc.

I tried hacking it together, but not gotten it working yet.

Quoted 116 lines
---
 hw/pci-bridge/pci_expander_bridge.c | 53 ++++++++++++++++++++++++++++-
 include/hw/cxl/cxl.h                |  2 ++
 2 files changed, 54 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +52 -1
index 5021b60435..226a8a5fff 100644
@@ -17,6 +17,7 @@
 #include "hw/pci/pci_host.h"
 #include "hw/qdev-properties.h"
 #include "hw/pci/pci_bridge.h"
+#include "hw/cxl/cxl.h"
 #include "qemu/range.h"
 #include "qemu/error-report.h"
 #include "qemu/module.h"
@@ -70,6 +71,12 @@ struct PXBDev {
     int32_t uid;
 };
 
+typedef struct CXLHost {
+    PCIHostState parent_obj;
+
+    CXLComponentState cxl_cstate;
+} CXLHost;
+
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
     /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
@@ -85,6 +92,9 @@ static GList *pxb_dev_list;
 
 #define TYPE_PXB_HOST "pxb-host"
 
+#define TYPE_PXB_CXL_HOST "pxb-cxl-host"
+#define PXB_CXL_HOST(obj) OBJECT_CHECK(CXLHost, (obj), TYPE_PXB_CXL_HOST)
+
 static int pxb_bus_num(PCIBus *bus)
 {
     PXBDev *pxb = convert_to_pxb(bus->parent_dev);
@@ -198,6 +208,46 @@ static const TypeInfo pxb_host_info = {
     .class_init    = pxb_host_class_init,
 };
 
+static void pxb_cxl_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    PCIHostState *phb = PCI_HOST_BRIDGE(dev);
+    CXLHost *cxl = PXB_CXL_HOST(dev);
+    CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
+    struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+
+    cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
+                                      TYPE_PXB_CXL_HOST);
+    sysbus_init_mmio(sbd, mr);
+
+    /* FIXME: support multiple host bridges. */
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
+                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+}
+
+static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(class);
+    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(class);
+
+    hc->root_bus_path = pxb_host_root_bus_path;
+    dc->fw_name = "cxl";
+    dc->realize = pxb_cxl_realize;
+    /* Reason: Internal part of the pxb/pxb-pcie device, not usable by itself */
+    dc->user_creatable = false;
+}
+
+/*
+ * This is a device to handle the MMIO for a CXL host bridge. It does nothing
+ * else.
+ */
+static const TypeInfo cxl_host_info = {
+    .name          = TYPE_PXB_CXL_HOST,
+    .parent        = TYPE_PCI_HOST_BRIDGE,
+    .instance_size = sizeof(CXLHost),
+    .class_init    = pxb_cxl_host_class_init,
+};
+
 /*
  * Registers the PXB bus as a child of pci host root bus.
  */
@@ -272,7 +322,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         dev_name = dev->qdev.id;
     }
 
-    ds = qdev_new(TYPE_PXB_HOST);
+    ds = qdev_new(type == CXL ? TYPE_PXB_CXL_HOST : TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else if (type == CXL) {
@@ -466,6 +516,7 @@ static void pxb_register_types(void)
     type_register_static(&pxb_pcie_bus_info);
     type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
+    type_register_static(&cxl_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
     type_register_static(&pxb_cxl_dev_info);
include/hw/cxl/cxl.h +2 -0
index 362cda40de..6bc344f205 100644
@@ -17,5 +17,7 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define CXL_HOST_BASE 0xD0000000
+
 #endif
 
#43 lore raw

Re: [RFC PATCH v3 17/31] hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)

Quoted 22 lines · Jonathan Cameron
Quoted 19 lines · On Tue, 2 Feb 2021 11:45:05 -0800 Ben Widawsky <[email protected]> wrote:
Quoted 17 lines · Jonathan Cameron
Quoted 10 lines · On Mon, 1 Feb 2021 16:59:34 -0800 Ben Widawsky <[email protected]> wrote:

CXL host bridges themselves may have MMIO. Since host bridges don't have a BAR they are treated as special for MMIO.

--

It's arbitrarily chosen here to pick 0xD0000000 as the base for the host bridge MMIO. I'm not sure what the right way to find free space for platform hardcoded things like this is.

Seems like this needs to come from the machine definition.
This is fairly easy for arm/virt, where there is a clearly laid out memory map. For hw/i386 I'm less sure on how to do it.

Quoted 6 lines
Quoted 1 line

I think this is how to do it :D

It may well be, but they you'll need to find a suitable region and document it and ensure no one else ever tramples on it. Easy to do on a physical system, bit trickier in emulation.

Maybe? x86 is full of magic physical address holes. As long as it's conveyed to EDK via _CRS, I think it's pretty safe. If something else tries to use the same address, you should get a fairly obvious error.

Document somehow, yes please.

Quoted 30 lines
Quoted 7 lines
Quoted 7 lines

Having said that, for this particular magic device, we do have a PCI EP associated with it. How about putting all the host bridge MMIO into a BAR of that rather than having it separate.
That has the added advantage of making it discoverable from firmware.

Quoted 6 lines
Quoted 2 lines

Any normal system is going to have this is impdef for discovery anyway.

This is not how it's expected to work for Intel at least. If the device was discoverable you wouldn't need CEDT/CHBS. The magic host bridges are only advertised via the CEDT.

I agree on a normal system (i.e. a real one) this doesn't work, but then a normal system doesn't involve a magic PCI RCiEP that also happens to instantiate an extra host bridge. This is what pxb_pcie is doing and what your pxb_cxl is almost doing.

Quoted 10 lines

When I build and run QEMU for x86_64, I do not see the host bridge in the pci topology, do you (it's meant to not be there)?

00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
...
34:00.0 PCI bridge: Intel Corporation Device 7075
35:00.0 Memory controller [0502]: Intel Corporation Device 0d93 (rev 01)

That's Q35, Root Port, and Type 3 device respectively.

Quoted 8 lines

You don't see the host bridge, for pxb_cxl, but for pxb_pcie you do. 00:06.0 Host bridge: Red Hat, Inc QEMU PCIe Expander bridge.
If you have another device after your pxb-cxl you'll also notice that there is a hole punched in the list where you'd expect pxb-cxl to be (device number skipped). (that had me confused earlier).

This seems to be because no VID etc (unlike pxb-pcie).

Right. This was in an earlier version of the series and you made me realize if I got rid of them that it disappears. I really like that this more accurately represents the hardware.

I agree it can be implemented more simply, but why do it if you can accurately model it?

Quoted 6 lines

I gave vendor and device IDs (and a bar to test that could be done) and it then appears just like pxb_pcie does. Hence handy place to hang our magic memory off so that EDK2 or similar can work with it and indeed construct he CHBS as needed so we can get to this via the same paths as a normal system. It's a bit convoluted but in some ways more elegant.

What are you looking to get out of EDK2 or similar? Anything you want to convey should work with _CRS, I think. That was the path I was going down.

Quoted 129 lines

Jonathan

Quoted 126 lines
Quoted 125 lines

That would then let you drop the separate definition of CXLHost structure though it needs a bit of figuring out what to do with the memory window setup etc.

I tried hacking it together, but not gotten it working yet.

Quoted 116 lines
---
 hw/pci-bridge/pci_expander_bridge.c | 53 ++++++++++++++++++++++++++++-
 include/hw/cxl/cxl.h                |  2 ++
 2 files changed, 54 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +52 -1
index 5021b60435..226a8a5fff 100644
@@ -17,6 +17,7 @@
 #include "hw/pci/pci_host.h"
 #include "hw/qdev-properties.h"
 #include "hw/pci/pci_bridge.h"
+#include "hw/cxl/cxl.h"
 #include "qemu/range.h"
 #include "qemu/error-report.h"
 #include "qemu/module.h"
@@ -70,6 +71,12 @@ struct PXBDev {
     int32_t uid;
 };
 
+typedef struct CXLHost {
+    PCIHostState parent_obj;
+
+    CXLComponentState cxl_cstate;
+} CXLHost;
+
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
     /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
@@ -85,6 +92,9 @@ static GList *pxb_dev_list;
 
 #define TYPE_PXB_HOST "pxb-host"
 
+#define TYPE_PXB_CXL_HOST "pxb-cxl-host"
+#define PXB_CXL_HOST(obj) OBJECT_CHECK(CXLHost, (obj), TYPE_PXB_CXL_HOST)
+
 static int pxb_bus_num(PCIBus *bus)
 {
     PXBDev *pxb = convert_to_pxb(bus->parent_dev);
@@ -198,6 +208,46 @@ static const TypeInfo pxb_host_info = {
     .class_init    = pxb_host_class_init,
 };
 
+static void pxb_cxl_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    PCIHostState *phb = PCI_HOST_BRIDGE(dev);
+    CXLHost *cxl = PXB_CXL_HOST(dev);
+    CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
+    struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+
+    cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
+                                      TYPE_PXB_CXL_HOST);
+    sysbus_init_mmio(sbd, mr);
+
+    /* FIXME: support multiple host bridges. */
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
+                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+}
+
+static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(class);
+    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(class);
+
+    hc->root_bus_path = pxb_host_root_bus_path;
+    dc->fw_name = "cxl";
+    dc->realize = pxb_cxl_realize;
+    /* Reason: Internal part of the pxb/pxb-pcie device, not usable by itself */
+    dc->user_creatable = false;
+}
+
+/*
+ * This is a device to handle the MMIO for a CXL host bridge. It does nothing
+ * else.
+ */
+static const TypeInfo cxl_host_info = {
+    .name          = TYPE_PXB_CXL_HOST,
+    .parent        = TYPE_PCI_HOST_BRIDGE,
+    .instance_size = sizeof(CXLHost),
+    .class_init    = pxb_cxl_host_class_init,
+};
+
 /*
  * Registers the PXB bus as a child of pci host root bus.
  */
@@ -272,7 +322,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         dev_name = dev->qdev.id;
     }
 
-    ds = qdev_new(TYPE_PXB_HOST);
+    ds = qdev_new(type == CXL ? TYPE_PXB_CXL_HOST : TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else if (type == CXL) {
@@ -466,6 +516,7 @@ static void pxb_register_types(void)
     type_register_static(&pxb_pcie_bus_info);
     type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
+    type_register_static(&cxl_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
     type_register_static(&pxb_cxl_dev_info);
include/hw/cxl/cxl.h +2 -0
index 362cda40de..6bc344f205 100644
@@ -17,5 +17,7 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define CXL_HOST_BASE 0xD0000000
+
 #endif
 
#44 lore raw

Re: [RFC PATCH v3 17/31] hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)

Quoted 75 lines · On Tue, 2 Feb 2021 13:03:57 -0800 Ben Widawsky <[email protected]> wrote:
Quoted 22 lines · Jonathan Cameron
Quoted 19 lines · On Tue, 2 Feb 2021 11:45:05 -0800 Ben Widawsky <[email protected]> wrote:
Quoted 17 lines · Jonathan Cameron
Quoted 10 lines · On Mon, 1 Feb 2021 16:59:34 -0800 Ben Widawsky <[email protected]> wrote:

CXL host bridges themselves may have MMIO. Since host bridges don't have a BAR they are treated as special for MMIO.

--

It's arbitrarily chosen here to pick 0xD0000000 as the base for the host bridge MMIO. I'm not sure what the right way to find free space for platform hardcoded things like this is.

Seems like this needs to come from the machine definition.
This is fairly easy for arm/virt, where there is a clearly laid out memory map. For hw/i386 I'm less sure on how to do it.

Quoted 6 lines
Quoted 1 line

I think this is how to do it :D

It may well be, but they you'll need to find a suitable region and document it and ensure no one else ever tramples on it. Easy to do on a physical system, bit trickier in emulation.

Maybe? x86 is full of magic physical address holes. As long as it's conveyed to EDK via _CRS, I think it's pretty safe. If something else tries to use the same address, you should get a fairly obvious error.

Document somehow, yes please.

Quoted 30 lines
Quoted 7 lines
Quoted 7 lines

Having said that, for this particular magic device, we do have a PCI EP associated with it. How about putting all the host bridge MMIO into a BAR of that rather than having it separate.
That has the added advantage of making it discoverable from firmware.

Quoted 6 lines
Quoted 2 lines

Any normal system is going to have this is impdef for discovery anyway.

This is not how it's expected to work for Intel at least. If the device was discoverable you wouldn't need CEDT/CHBS. The magic host bridges are only advertised via the CEDT.

I agree on a normal system (i.e. a real one) this doesn't work, but then a normal system doesn't involve a magic PCI RCiEP that also happens to instantiate an extra host bridge. This is what pxb_pcie is doing and what your pxb_cxl is almost doing.

Quoted 10 lines

When I build and run QEMU for x86_64, I do not see the host bridge in the pci topology, do you (it's meant to not be there)?

00:00.0 Host bridge: Intel Corporation 82G33/G31/P35/P31 Express DRAM Controller
...
34:00.0 PCI bridge: Intel Corporation Device 7075
35:00.0 Memory controller [0502]: Intel Corporation Device 0d93 (rev 01)

That's Q35, Root Port, and Type 3 device respectively.

Quoted 8 lines

You don't see the host bridge, for pxb_cxl, but for pxb_pcie you do. 00:06.0 Host bridge: Red Hat, Inc QEMU PCIe Expander bridge.
If you have another device after your pxb-cxl you'll also notice that there is a hole punched in the list where you'd expect pxb-cxl to be (device number skipped). (that had me confused earlier).

This seems to be because no VID etc (unlike pxb-pcie).

Quoted 6 lines

Right. This was in an earlier version of the series and you made me realize if I got rid of them that it disappears. I really like that this more accurately represents the hardware.

I agree it can be implemented more simply, but why do it if you can accurately model it?

I'd like to understand all the reasons pxb_pcie is deliberately exposed and whether any of them affect CXL cases. Right now I'm not entirely sure what they are.

Quoted 10 lines
Quoted 7 lines

I gave vendor and device IDs (and a bar to test that could be done) and it then appears just like pxb_pcie does. Hence handy place to hang our magic memory off so that EDK2 or similar can work with it and indeed construct he CHBS as needed so we can get to this via the same paths as a normal system. It's a bit convoluted but in some ways more elegant.

What are you looking to get out of EDK2 or similar? Anything you want to convey should work with _CRS, I think. That was the path I was going down.

If you go down the fixed route that's fine. Just seemed a bit convenient we had somewhere to hang this that didn't rely on arbitrary fixed regions.

If we went with the BAR we'd need to enumerate first to get bar memory location and then fill in the table. Bit messy so fixed does look like a better option.

Jonathan

Quoted 130 lines
Quoted 130 lines

Jonathan

Quoted 126 lines
Quoted 125 lines

That would then let you drop the separate definition of CXLHost structure though it needs a bit of figuring out what to do with the memory window setup etc.

I tried hacking it together, but not gotten it working yet.

Quoted 116 lines
---
 hw/pci-bridge/pci_expander_bridge.c | 53 ++++++++++++++++++++++++++++-
 include/hw/cxl/cxl.h                |  2 ++
 2 files changed, 54 insertions(+), 1 deletion(-)
hw/pci-bridge/pci_expander_bridge.c +52 -1
index 5021b60435..226a8a5fff 100644
@@ -17,6 +17,7 @@
 #include "hw/pci/pci_host.h"
 #include "hw/qdev-properties.h"
 #include "hw/pci/pci_bridge.h"
+#include "hw/cxl/cxl.h"
 #include "qemu/range.h"
 #include "qemu/error-report.h"
 #include "qemu/module.h"
@@ -70,6 +71,12 @@ struct PXBDev {
     int32_t uid;
 };
 
+typedef struct CXLHost {
+    PCIHostState parent_obj;
+
+    CXLComponentState cxl_cstate;
+} CXLHost;
+
 static PXBDev *convert_to_pxb(PCIDevice *dev)
 {
     /* A CXL PXB's parent bus is PCIe, so the normal check won't work */
@@ -85,6 +92,9 @@ static GList *pxb_dev_list;
 
 #define TYPE_PXB_HOST "pxb-host"
 
+#define TYPE_PXB_CXL_HOST "pxb-cxl-host"
+#define PXB_CXL_HOST(obj) OBJECT_CHECK(CXLHost, (obj), TYPE_PXB_CXL_HOST)
+
 static int pxb_bus_num(PCIBus *bus)
 {
     PXBDev *pxb = convert_to_pxb(bus->parent_dev);
@@ -198,6 +208,46 @@ static const TypeInfo pxb_host_info = {
     .class_init    = pxb_host_class_init,
 };
 
+static void pxb_cxl_realize(DeviceState *dev, Error **errp)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
+    PCIHostState *phb = PCI_HOST_BRIDGE(dev);
+    CXLHost *cxl = PXB_CXL_HOST(dev);
+    CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
+    struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+
+    cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
+                                      TYPE_PXB_CXL_HOST);
+    sysbus_init_mmio(sbd, mr);
+
+    /* FIXME: support multiple host bridges. */
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
+                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+}
+
+static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(class);
+    PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(class);
+
+    hc->root_bus_path = pxb_host_root_bus_path;
+    dc->fw_name = "cxl";
+    dc->realize = pxb_cxl_realize;
+    /* Reason: Internal part of the pxb/pxb-pcie device, not usable by itself */
+    dc->user_creatable = false;
+}
+
+/*
+ * This is a device to handle the MMIO for a CXL host bridge. It does nothing
+ * else.
+ */
+static const TypeInfo cxl_host_info = {
+    .name          = TYPE_PXB_CXL_HOST,
+    .parent        = TYPE_PCI_HOST_BRIDGE,
+    .instance_size = sizeof(CXLHost),
+    .class_init    = pxb_cxl_host_class_init,
+};
+
 /*
  * Registers the PXB bus as a child of pci host root bus.
  */
@@ -272,7 +322,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         dev_name = dev->qdev.id;
     }
 
-    ds = qdev_new(TYPE_PXB_HOST);
+    ds = qdev_new(type == CXL ? TYPE_PXB_CXL_HOST : TYPE_PXB_HOST);
     if (type == PCIE) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_PCIE_BUS);
     } else if (type == CXL) {
@@ -466,6 +516,7 @@ static void pxb_register_types(void)
     type_register_static(&pxb_pcie_bus_info);
     type_register_static(&pxb_cxl_bus_info);
     type_register_static(&pxb_host_info);
+    type_register_static(&cxl_host_info);
     type_register_static(&pxb_dev_info);
     type_register_static(&pxb_pcie_dev_info);
     type_register_static(&pxb_cxl_dev_info);
include/hw/cxl/cxl.h +2 -0
index 362cda40de..6bc344f205 100644
@@ -17,5 +17,7 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define CXL_HOST_BASE 0xD0000000
+
 #endif
 
#45 lore raw

RFC PATCH v3 18/31 acpi/pxb/cxl: Reserve host bridge MMIO

In reply to Ben Widawsky

For all host bridges, reserve MMIO space with _CRS. The MMIO for the host bridge lives in a magically hard coded space in the system's physical address space. The standard mechanism to tell the OS about regions which can't be used for host bridges is _CRS.

---
 hw/i386/acpi-build.c | 22 +++++++++++++++++++---
 1 file changed, 19 insertions(+), 3 deletions(-)
hw/i386/acpi-build.c +19 -3
index 145a503e92..ecdc10b148 100644
@@ -28,6 +28,7 @@
 #include "qemu/bitmap.h"
 #include "qemu/error-report.h"
 #include "hw/pci/pci.h"
+#include "hw/cxl/cxl.h"
 #include "hw/core/cpu.h"
 #include "target/i386/cpu.h"
 #include "hw/misc/pvpanic.h"
@@ -1194,7 +1195,7 @@ static void build_smb0(Aml *table, I2CBus *smbus, int devnr, int func)
     aml_append(table, scope);
 }
 
-enum { PCI, PCIE };
+enum { PCI, PCIE, CXL };
 static void init_pci_acpi(Aml *dev, int uid, int type)
 {
     if (type == PCI) {
@@ -1344,20 +1345,28 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             uint8_t bus_num = pci_bus_num(bus);
             uint8_t numa_node = pci_bus_numa_node(bus);
             int32_t uid = pci_bus_uid(bus);
+            int type;
 
             /* look only for expander root buses */
             if (!pci_bus_is_root(bus)) {
                 continue;
             }
 
+            type = pci_bus_is_cxl(bus) ? CXL :
+                                         pci_bus_is_express(bus) ? PCIE : PCI;
+
             if (bus_num < root_bus_limit) {
                 root_bus_limit = bus_num - 1;
             }
 
             scope = aml_scope("\\_SB");
-            dev = aml_device("PC%.02X", bus_num);
+            if (type == CXL) {
+                dev = aml_device("CXL%.01X", pci_bus_uid(bus));
+            } else {
+                dev = aml_device("PC%.02X", bus_num);
+            }
             aml_append(dev, aml_name_decl("_BBN", aml_int(bus_num)));
-            init_pci_acpi(dev, uid, pci_bus_is_express(bus) ? PCIE : PCI);
+            init_pci_acpi(dev, uid, type);
 
             if (numa_node != NUMA_NODE_UNASSIGNED) {
                 aml_append(dev, aml_name_decl("_PXM", aml_int(numa_node)));
@@ -1369,6 +1378,13 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
             aml_append(dev, aml_name_decl("_CRS", crs));
             aml_append(scope, dev);
             aml_append(dsdt, scope);
+
+            /* Handle the ranges for the PXB expanders */
+            if (type == CXL) {
+                uint64_t base = CXL_HOST_BASE + uid * 0x10000;
+                crs_range_insert(crs_range_set.mem_ranges, base,
+                                 base + 0x10000 - 1);
+            }
         }
     }
 
Signature · 1 line
2.30.0
#46 lore raw

RFC PATCH v3 19/31 hw/pxb/cxl: Add "windows" for host bridges

In reply to Ben Widawsky

In a bare metal CXL capable system, system firmware will program physical address ranges on the host. This is done by programming internal registers that aren't typically known to OS. These address ranges might be contiguous or interleaved across host bridges.

For a QEMU guest a new construct is introduced allowing passing a memory backend to the host bridge for this same purpose. Each memory backend needs to be passed to the host bridge as well as any device that will be emulating that memory (not implemented here).

I'm hopeful the interleaving work in the link can be re-purposed here (see Link).

An example to create a host bridges with a 512M window at 0x4c0000000
-object memory-backend-file,id=cxl-mem1,share,mem-path=cxl-type3,size=512M
-device pxb-cxl,id=cxl.0,bus=pcie.0,bus_nr=52,uid=0,len-memory-base=1,memory-base\[0\]=0x4c0000000,memory\[0\]=cxl-mem1

---
 hw/pci-bridge/pci_expander_bridge.c | 65 +++++++++++++++++++++++++++--
 include/hw/cxl/cxl.h                |  1 +
 2 files changed, 62 insertions(+), 4 deletions(-)
hw/pci-bridge/pci_expander_bridge.c +61 -4
index 226a8a5fff..af1450c69d 100644
@@ -69,12 +69,19 @@ struct PXBDev {
     uint8_t bus_nr;
     uint16_t numa_node;
     int32_t uid;
+    struct cxl_dev {
+        HostMemoryBackend *memory_window[CXL_WINDOW_MAX];
+
+        uint32_t num_windows;
+        hwaddr *window_base[CXL_WINDOW_MAX];
+    } cxl;
 };
 
 typedef struct CXLHost {
     PCIHostState parent_obj;
 
     CXLComponentState cxl_cstate;
+    PXBDev *dev;
 } CXLHost;
 
 static PXBDev *convert_to_pxb(PCIDevice *dev)
@@ -213,16 +220,31 @@ static void pxb_cxl_realize(DeviceState *dev, Error **errp)
     SysBusDevice *sbd = SYS_BUS_DEVICE(dev);
     PCIHostState *phb = PCI_HOST_BRIDGE(dev);
     CXLHost *cxl = PXB_CXL_HOST(dev);
+    struct cxl_dev *cxl_dev = &cxl->dev->cxl;
     CXLComponentState *cxl_cstate = &cxl->cxl_cstate;
     struct MemoryRegion *mr = &cxl_cstate->crb.component_registers;
+    int uid = pci_bus_uid(phb->bus);
 
     cxl_component_register_block_init(OBJECT(dev), cxl_cstate,
                                       TYPE_PXB_CXL_HOST);
     sysbus_init_mmio(sbd, mr);
 
-    /* FIXME: support multiple host bridges. */
-    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE +
-                            memory_region_size(mr) * pci_bus_uid(phb->bus));
+    sysbus_mmio_map(sbd, 0, CXL_HOST_BASE + memory_region_size(mr) * uid);
+
+    /*
+     * A CXL host bridge can exist without a fixed memory window, but it would
+     * only operate in legacy PCIe mode.
+     */
+    if (!cxl_dev->memory_window[uid]) {
+        warn_report(
+            "CXL expander bridge created without window. Consider using %s",
+            "memdev[0]=<memory_backend>");
+        return;
+    }
+
+    mr = host_memory_backend_get_memory(cxl_dev->memory_window[uid]);
+    sysbus_init_mmio(sbd, mr);
+    sysbus_mmio_map(sbd, 1 + uid, *cxl_dev->window_base[uid]);
 }
 
 static void pxb_cxl_host_class_init(ObjectClass *class, void *data)
@@ -328,6 +350,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
     } else if (type == CXL) {
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_CXL_BUS);
         bus->flags |= PCI_BUS_CXL;
+        PXB_CXL_HOST(ds)->dev = PXB_CXL_DEV(dev);
     } else {
         bus = pci_root_bus_new(ds, "pxb-internal", NULL, NULL, 0, TYPE_PXB_BUS);
         bds = qdev_new("pci-bridge");
@@ -389,6 +412,8 @@ static Property pxb_dev_properties[] = {
     DEFINE_PROP_UINT8("bus_nr", PXBDev, bus_nr, 0),
     DEFINE_PROP_UINT16("numa_node", PXBDev, numa_node, NUMA_NODE_UNASSIGNED),
     DEFINE_PROP_INT32("uid", PXBDev, uid, -1),
+    DEFINE_PROP_ARRAY("window-base", PXBDev, cxl.num_windows, cxl.window_base,
+                      qdev_prop_uint64, hwaddr),
     DEFINE_PROP_END_OF_LIST(),
 };
 
@@ -460,7 +485,9 @@ static const TypeInfo pxb_pcie_dev_info = {
 
 static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
 {
-    PXBDev *pxb = convert_to_pxb(dev);
+    PXBDev *pxb = PXB_CXL_DEV(dev);
+    struct cxl_dev *cxl = &pxb->cxl;
+    int count = 0;
 
     /* A CXL PXB's parent bus is still PCIe */
     if (!pci_bus_is_express(pci_get_bus(dev))) {
@@ -476,6 +503,23 @@ static void pxb_cxl_dev_realize(PCIDevice *dev, Error **errp)
     /* FIXME: Check that uid doesn't collide with UIDs of other host bridges */
 
     pxb_dev_realize_common(dev, CXL, errp);
+
+    for (unsigned i = 0; i < CXL_WINDOW_MAX; i++) {
+        if (!cxl->memory_window[i]) {
+            continue;
+        }
+
+        count++;
+    }
+
+    if (!count) {
+        warn_report("memory-windows should be set when creating CXL host bridges");
+    }
+
+    if (count != cxl->num_windows) {
+        error_setg(errp, "window bases count (%d) must match window count (%d)",
+                   cxl->num_windows, count);
+    }
 }
 
 static void pxb_cxl_dev_class_init(ObjectClass *klass, void *data)
@@ -496,6 +540,19 @@ static void pxb_cxl_dev_class_init(ObjectClass *klass, void *data)
 
     /* Host bridges aren't hotpluggable. FIXME: spec reference */
     dc->hotpluggable = false;
+
+    /*
+     * Below is moral equivalent of:
+     *   DEFINE_PROP_ARRAY("memdev", PXBDev, window_count, windows,
+     *                     qdev_prop_memory_device, HostMemoryBackend)
+     */
+    for (unsigned i = 0; i < CXL_WINDOW_MAX; i++) {
+        g_autofree char *name = g_strdup_printf("memdev[%u]", i);
+        object_class_property_add_link(klass, name, TYPE_MEMORY_BACKEND,
+                offsetof(PXBDev, cxl.memory_window[i]),
+                qdev_prop_allow_set_link_before_realize,
+                OBJ_PROP_LINK_STRONG);
+    }
 }
 
 static const TypeInfo pxb_cxl_dev_info = {
include/hw/cxl/cxl.h +1 -0
index 6bc344f205..b1e5f4a8fa 100644
@@ -18,6 +18,7 @@
 #define DEVICE_REG_BAR_IDX 2
 
 #define CXL_HOST_BASE 0xD0000000
+#define CXL_WINDOW_MAX 10
 
 #endif
 
Signature · 1 line
2.30.0
#47 lore raw

RFC PATCH v3 20/31 hw/cxl/rp: Add a root port

In reply to Ben Widawsky

This adds just enough of a root port implementation to be able to enumerate root ports (creating the required DVSEC entries). What's not here yet is the MMIO nor the ability to write some of the DVSEC entries.

This can be added with the qemu commandline by adding a rootport to a specific CXL host bridge. For example:
-device cxl-rp,id=rp0,bus="cxl.0",addr=0.0,chassis=4

Like the host bridge patch, the ACPI tables aren't generated at this point and so system software cannot use it.

---
 hw/pci-bridge/Kconfig          |   5 +
 hw/pci-bridge/cxl_root_port.c  | 231 +++++++++++++++++++++++++++++++++
 hw/pci-bridge/meson.build      |   1 +
 hw/pci-bridge/pcie_root_port.c |   6 +-
 hw/pci/pci.c                   |   4 +-
 5 files changed, 245 insertions(+), 2 deletions(-)
 create mode 100644 hw/pci-bridge/cxl_root_port.c
hw/pci-bridge/Kconfig +5 -0
index f8df4315ba..02614f49aa 100644
@@ -27,3 +27,8 @@ config DEC_PCI
 
 config SIMBA
     bool
+
+config CXL
+    bool
+    default y if PCI_EXPRESS && PXB
+    depends on PCI_EXPRESS && MSI_NONBROKEN && PXB
hw/pci-bridge/cxl_root_port.c new file +231 -0

231 lines not on this page · raw message

hw/pci-bridge/meson.build +1 -0
index daab8acf2a..b6d26a03d5 100644
@@ -5,6 +5,7 @@ pci_ss.add(when: 'CONFIG_IOH3420', if_true: files('ioh3420.c'))
 pci_ss.add(when: 'CONFIG_PCIE_PORT', if_true: files('pcie_root_port.c', 'gen_pcie_root_port.c', 'pcie_pci_bridge.c'))
 pci_ss.add(when: 'CONFIG_PXB', if_true: files('pci_expander_bridge.c'))
 pci_ss.add(when: 'CONFIG_XIO3130', if_true: files('xio3130_upstream.c', 'xio3130_downstream.c'))
+pci_ss.add(when: 'CONFIG_CXL', if_true: files('cxl_root_port.c'))
 
 # NewWorld PowerMac
 pci_ss.add(when: 'CONFIG_DEC_PCI', if_true: files('dec.c'))
hw/pci-bridge/pcie_root_port.c +5 -1
index f1cfe9d14a..460e48269d 100644
@@ -67,7 +67,11 @@ static void rp_realize(PCIDevice *d, Error **errp)
     int rc;
 
     pci_config_set_interrupt_pin(d->config, 1);
-    pci_bridge_initfn(d, TYPE_PCIE_BUS);
+    if (d->cap_present & QEMU_PCIE_CAP_CXL) {
+        pci_bridge_initfn(d, TYPE_CXL_BUS);
+    } else {
+        pci_bridge_initfn(d, TYPE_PCIE_BUS);
+    }
     pcie_port_init_reg(d);
 
     rc = pci_bridge_ssvid_init(d, rpc->ssvid_offset, dc->vendor_id,
hw/pci/pci.c +3 -1
index bf019d91a0..eb325704e7 100644
@@ -2668,7 +2668,9 @@ static void pci_device_class_base_init(ObjectClass *klass, void *data)
             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
         ObjectClass *pcie =
             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
-        assert(conventional || pcie);
+        ObjectClass *cxl =
+            object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
+        assert(conventional || pcie || cxl);
     }
 }
 
Signature · 1 line
2.30.0
#48 lore raw

RFC PATCH v3 21/31 hw/cxl/device: Add a memory device (8.2.8.5)

In reply to Ben Widawsky

A CXL memory device (AKA Type 3) is a CXL component that contains some combination of volatile and persistent memory. It also implements the previously defined mailbox interface as well as the memory device firmware interface.

Although the memory device is configured like a normal PCIe device, the memory traffic is on an entirely separate bus conceptually (using the same physical wires as PCIe, but different protocol).

The guest physical address for the memory device is part of a larger window which is owned by the platform. Currently, this is hardcoded as an object property on host bridge (PXB) creation, but that will need to change for interleaving.

The following example will create a 256M device in a 512M window:
-object "memory-backend-file,id=cxl-mem1,share,mem-path=cxl-type3,size=512M" -device "cxl-type3,bus=rp0,memdev=cxl-mem1,id=cxl-pmem0,size=256M"

---
 hw/core/numa.c             |   3 +
 hw/cxl/cxl-mailbox-utils.c |  41 ++++++
 hw/i386/pc.c               |   1 +
 hw/mem/Kconfig             |   5 +
 hw/mem/cxl_type3.c         | 281 +++++++++++++++++++++++++++++++++++++
 hw/mem/meson.build         |   1 +
 hw/pci/pcie.c              |  30 ++++
 include/hw/cxl/cxl.h       |   2 +
 include/hw/cxl/cxl_pci.h   |  22 +++
 include/hw/pci/pci_ids.h   |   1 +
 monitor/hmp-cmds.c         |  15 ++
 qapi/machine.json          |   1 +
 12 files changed, 403 insertions(+)
 create mode 100644 hw/mem/cxl_type3.c
hw/core/numa.c +3 -0
index 68cee65f61..cd7df371e6 100644
@@ -770,6 +770,9 @@ static void numa_stat_memory_devices(NumaNodeMem node_mem[])
                 node_mem[pcdimm_info->node].node_plugged_mem +=
                     pcdimm_info->size;
                 break;
+            case MEMORY_DEVICE_INFO_KIND_CXL:
+                /* FINISHME */
+                break;
             case MEMORY_DEVICE_INFO_KIND_VIRTIO_PMEM:
                 vpi = value->u.virtio_pmem.data;
                 /* TODO: once we support numa, assign to right node */
hw/cxl/cxl-mailbox-utils.c +41 -0

65 lines not on this page · raw message

hw/i386/pc.c +1 -0
index 5458f61d10..5d41809b37 100644
@@ -79,6 +79,7 @@
 #include "acpi-build.h"
 #include "hw/mem/pc-dimm.h"
 #include "hw/mem/nvdimm.h"
+#include "hw/cxl/cxl.h"
 #include "qapi/error.h"
 #include "qapi/qapi-visit-common.h"
 #include "qapi/visitor.h"
hw/mem/Kconfig +5 -0
index a0ef2cf648..7d9d1ced3e 100644
@@ -10,3 +10,8 @@ config NVDIMM
     default y
     depends on (PC || PSERIES || ARM_VIRT)
     select MEM_DEVICE
+
+config CXL_MEM_DEVICE
+    bool
+    default y if CXL
+    select MEM_DEVICE
hw/mem/cxl_type3.c new file +281 -0

281 lines not on this page · raw message

hw/mem/meson.build +1 -0
index 0d22f2b572..d13c3ed117 100644
@@ -3,5 +3,6 @@ mem_ss.add(files('memory-device.c'))
 mem_ss.add(when: 'CONFIG_DIMM', if_true: files('pc-dimm.c'))
 mem_ss.add(when: 'CONFIG_NPCM7XX', if_true: files('npcm7xx_mc.c'))
 mem_ss.add(when: 'CONFIG_NVDIMM', if_true: files('nvdimm.c'))
+mem_ss.add(when: 'CONFIG_CXL_MEM_DEVICE', if_true: files('cxl_type3.c'))
 
 softmmu_ss.add_all(when: 'CONFIG_MEM_DEVICE', if_true: mem_ss)
hw/pci/pcie.c +30 -0
index d4010cf8f3..1ecf6f6a55 100644
@@ -20,6 +20,7 @@
 
 #include "qemu/osdep.h"
 #include "qapi/error.h"
+#include "hw/mem/memory-device.h"
 #include "hw/pci/pci_bridge.h"
 #include "hw/pci/pcie.h"
 #include "hw/pci/msix.h"
@@ -27,6 +28,8 @@
 #include "hw/pci/pci_bus.h"
 #include "hw/pci/pcie_regs.h"
 #include "hw/pci/pcie_port.h"
+#include "hw/cxl/cxl.h"
+#include "hw/boards.h"
 #include "qemu/range.h"
 
 //#define DEBUG_PCIE
@@ -419,6 +422,28 @@ void pcie_cap_slot_pre_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
     }
 
     pcie_cap_slot_plug_common(PCI_DEVICE(hotplug_dev), dev, errp);
+
+#ifdef CXL_MEM_DEVICE
+    /*
+     * FIXME:
+     * if (object_dynamic_cast(OBJECT(dev), TYPE_CXL_TYPE3_DEV)) {
+     *    HotplugHandler *hotplug_ctrl;
+     *   Error *local_err = NULL;
+     *  hotplug_ctrl = qdev_get_hotplug_handler(dev);
+     *  if (hotplug_ctrl) {
+     *      hotplug_handler_pre_plug(hotplug_ctrl, dev, &local_err);
+     *      if (local_err) {
+     *          error_propagate(errp, local_err);
+     *          return;
+     *      }
+     *  }
+     */
+
+    if (object_dynamic_cast(OBJECT(dev), TYPE_CXL_TYPE3_DEV)) {
+        memory_device_pre_plug(MEMORY_DEVICE(dev), MACHINE(qdev_get_machine()),
+                               NULL, errp);
+    }
+#endif
 }
 
 void pcie_cap_slot_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
@@ -455,6 +480,11 @@ void pcie_cap_slot_plug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
         pcie_cap_slot_event(hotplug_pdev,
                             PCI_EXP_HP_EV_PDC | PCI_EXP_HP_EV_ABP);
     }
+
+#ifdef CXL_MEM_DEVICE
+    if (object_dynamic_cast(OBJECT(dev), TYPE_CXL_TYPE3_DEV))
+        memory_device_plug(MEMORY_DEVICE(dev), MACHINE(qdev_get_machine()));
+#endif
 }
 
 void pcie_cap_slot_unplug_cb(HotplugHandler *hotplug_dev, DeviceState *dev,
include/hw/cxl/cxl.h +2 -0
index b1e5f4a8fa..809ed7de60 100644
@@ -17,6 +17,8 @@
 #define COMPONENT_REG_BAR_IDX 0
 #define DEVICE_REG_BAR_IDX 2
 
+#define TYPE_CXL_TYPE3_DEV "cxl-type3"
+
 #define CXL_HOST_BASE 0xD0000000
 #define CXL_WINDOW_MAX 10
 
include/hw/cxl/cxl_pci.h +22 -0
index a53c2e5ae7..9ec28c9feb 100644
@@ -64,6 +64,28 @@ _Static_assert(sizeof(struct dvsec_header) == 10,
  * CXL 2.0 Downstream Port: 3, 4, 7, 8
  */
 
+/* CXL 2.0 - 8.1.3 (ID 0001) */
+struct dvsec_device {
+    struct dvsec_header hdr;
+    uint16_t cap;
+    uint16_t ctrl;
+    uint16_t status;
+    uint16_t ctrl2;
+    uint16_t status2;
+    uint16_t lock;
+    uint16_t cap2;
+    uint32_t range1_size_hi;
+    uint32_t range1_size_lo;
+    uint32_t range1_base_hi;
+    uint32_t range1_base_lo;
+    uint32_t range2_size_hi;
+    uint32_t range2_size_lo;
+    uint32_t range2_base_hi;
+    uint32_t range2_base_lo;
+};
+_Static_assert(sizeof(struct dvsec_device) == 0x38,
+               "dvsec device size incorrect");
+
 /* CXL 2.0 - 8.1.5 (ID 0003) */
 struct extensions_dvsec_port {
     struct dvsec_header hdr;
include/hw/pci/pci_ids.h +1 -0
index 11f8ab7149..76bf3ed590 100644
@@ -53,6 +53,7 @@
 #define PCI_BASE_CLASS_MEMORY            0x05
 #define PCI_CLASS_MEMORY_RAM             0x0500
 #define PCI_CLASS_MEMORY_FLASH           0x0501
+#define PCI_CLASS_MEMORY_CXL             0x0502
 #define PCI_CLASS_MEMORY_OTHER           0x0580
 
 #define PCI_BASE_CLASS_BRIDGE            0x06
monitor/hmp-cmds.c +15 -0
index a48bc1e904..1c8edfb136 100644
@@ -1884,6 +1884,21 @@ void hmp_info_memory_devices(Monitor *mon, const QDict *qdict)
                 monitor_printf(mon, "  hotpluggable: %s\n",
                                di->hotpluggable ? "true" : "false");
                 break;
+            case MEMORY_DEVICE_INFO_KIND_CXL:
+                di = value->u.cxl.data;
+                monitor_printf(mon, "Memory device [%s]: \"%s\"\n",
+                               MemoryDeviceInfoKind_str(value->type),
+                               di->id ? di->id : "");
+                monitor_printf(mon, "  addr: 0x%" PRIx64 "\n", di->addr);
+                monitor_printf(mon, "  slot: %" PRId64 "\n", di->slot);
+                monitor_printf(mon, "  node: %" PRId64 "\n", di->node);
+                monitor_printf(mon, "  size: %" PRIu64 "\n", di->size);
+                monitor_printf(mon, "  memdev: %s\n", di->memdev);
+                monitor_printf(mon, "  hotplugged: %s\n",
+                               di->hotplugged ? "true" : "false");
+                monitor_printf(mon, "  hotpluggable: %s\n",
+                               di->hotpluggable ? "true" : "false");
+                break;
             case MEMORY_DEVICE_INFO_KIND_VIRTIO_PMEM:
                 vpi = value->u.virtio_pmem.data;
                 monitor_printf(mon, "Memory device [%s]: \"%s\"\n",
qapi/machine.json +1 -0
index 330189efe3..aa96d662bd 100644
@@ -1394,6 +1394,7 @@
 { 'union': 'MemoryDeviceInfo',
   'data': { 'dimm': 'PCDIMMDeviceInfo',
             'nvdimm': 'PCDIMMDeviceInfo',
+            'cxl': 'PCDIMMDeviceInfo',
             'virtio-pmem': 'VirtioPMEMDeviceInfo',
             'virtio-mem': 'VirtioMEMDeviceInfo'
           }
Signature · 1 line
2.30.0
#49 lore raw

Re: [RFC PATCH v3 21/31] hw/cxl/device: Add a memory device (8.2.8.5)

Quoted 23 lines · Ben Widawsky

A CXL memory device (AKA Type 3) is a CXL component that contains some combination of volatile and persistent memory. It also implements the previously defined mailbox interface as well as the memory device firmware interface.

Although the memory device is configured like a normal PCIe device, the memory traffic is on an entirely separate bus conceptually (using the same physical wires as PCIe, but different protocol).

The guest physical address for the memory device is part of a larger window which is owned by the platform. Currently, this is hardcoded as an object property on host bridge (PXB) creation, but that will need to change for interleaving.

The following example will create a 256M device in a 512M window:

(quoted diff) +0 -2
-object "memory-backend-file,id=cxl-mem1,share,mem-path=cxl-type3,size=512M"
-device "cxl-type3,bus=rp0,memdev=cxl-mem1,id=cxl-pmem0,size=256M"

---

+++ b/qapi/machine.json

qapi/machine.json line 1400

Quoted 7 lines
qapi/machine.json +1 -0
@@ -1394,6 +1394,7 @@
 { 'union': 'MemoryDeviceInfo',
   'data': { 'dimm': 'PCDIMMDeviceInfo',
             'nvdimm': 'PCDIMMDeviceInfo',
+            'cxl': 'PCDIMMDeviceInfo',
             'virtio-pmem': 'VirtioPMEMDeviceInfo',
             'virtio-mem': 'VirtioMEMDeviceInfo'
           }

Missing documentation that 'cxl' was introduced in 6.0. Also, is it worth keeping the branches of this union in lexicographic order?

Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#50 lore raw

Re: [RFC PATCH v3 21/31] hw/cxl/device: Add a memory device (8.2.8.5)

Quoted 22 lines · Eric Blake
Quoted 20 lines · Ben Widawsky

A CXL memory device (AKA Type 3) is a CXL component that contains some combination of volatile and persistent memory. It also implements the previously defined mailbox interface as well as the memory device firmware interface.

Although the memory device is configured like a normal PCIe device, the memory traffic is on an entirely separate bus conceptually (using the same physical wires as PCIe, but different protocol).

The guest physical address for the memory device is part of a larger window which is owned by the platform. Currently, this is hardcoded as an object property on host bridge (PXB) creation, but that will need to change for interleaving.

The following example will create a 256M device in a 512M window:

(quoted diff) +0 -2
-object "memory-backend-file,id=cxl-mem1,share,mem-path=cxl-type3,size=512M"
-device "cxl-type3,bus=rp0,memdev=cxl-mem1,id=cxl-pmem0,size=256M"

---

Quoted 13 lines

qapi/machine.json line 1400

Quoted 9 lines

+++ b/qapi/machine.json

qapi/machine.json +1 -0
@@ -1394,6 +1394,7 @@
 { 'union': 'MemoryDeviceInfo',
   'data': { 'dimm': 'PCDIMMDeviceInfo',
             'nvdimm': 'PCDIMMDeviceInfo',
+            'cxl': 'PCDIMMDeviceInfo',
             'virtio-pmem': 'VirtioPMEMDeviceInfo',
             'virtio-mem': 'VirtioMEMDeviceInfo'
           }

Missing documentation that 'cxl' was introduced in 6.0. Also, is it worth keeping the branches of this union in lexicographic order?

Sure.

As discussed on the list previously, I think more thought needs to be put in here, and I could really use some input.

A CXL type3 memory device can have both persistent and volatile capacity. As such a single PCDIMMDeviceInfo I believe is insufficient. The current code supports persistent memory only, so this is fine for now.

I'd guess my best bet is to create a new CXLType3DeviceInfo, but I'm not entirely sure of all the implications that has.

Any advice?

Quoted 6 lines
Signature · 3 lines
Eric Blake, Principal Software Engineer
Red Hat, Inc.           +1-919-301-3226
Virtualization:  qemu.org | libvirt.org
#51 lore raw

RFC PATCH v3 22/31 hw/cxl/device: Implement MMIO HDM decoding (8.2.5.12)

In reply to Ben Widawsky

A device's volatile and persistent memory are known Host Defined Memory (HDM) regions. The mechanism by which the device is programmed to claim the addresses associated with those regions is through dedicated logic known as the HDM decoder. In order to allow the OS to properly program the HDMs, the HDM decoders must be modeled.

There are two ways the HDM decoders can be implemented, the legacy mechanism is through the PCIe DVSEC programming from CXL 1.1 (8.1.3.8), and MMIO is found in 8.2.5.12 of the spec. For now, 8.1.3.8 is not implemented.

Much of CXL device logic is implemented in cxl-utils. The HDM decoder however is implemented directly by the device implementation. The generic cxl-utils probably should be the correct place to put this since HDM decoders aren't unique to a type3 device. It is however easier at the moment, and requires less design consideration to simply implement it in the device, and figure out how to consolidate it later.

---
 hw/mem/cxl_type3.c | 92 ++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 84 insertions(+), 8 deletions(-)
hw/mem/cxl_type3.c +84 -8
index 4e9a016448..fe02c3b63c 100644
@@ -57,6 +57,84 @@ static void build_dvsecs(CXLType3Dev *ct3d)
                                REG_LOC_DVSEC_REVID, dvsec);
 }
 
+static void cxl_set_addr(CXLType3Dev *ct3d, hwaddr addr, Error **errp)
+{
+    MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(ct3d);
+    mdc->set_addr(MEMORY_DEVICE(ct3d), addr, errp);
+}
+
+static void hdm_decoder_commit(CXLType3Dev *ct3d, int which)
+{
+    MemoryRegion *pmem = ct3d->cxl_dstate.pmem;
+    MemoryRegion *mr = host_memory_backend_get_memory(ct3d->hostmem);
+    Range window, device;
+    ComponentRegisters *cregs = &ct3d->cxl_cstate.crb;
+    uint32_t *cache_mem = cregs->cache_mem_registers;
+    uint64_t offset, size;
+    Error *err = NULL;
+
+    assert(which == 0);
+
+    ARRAY_FIELD_DP32(cache_mem, CXL_HDM_DECODER0_CTRL, COMMIT, 0);
+    ARRAY_FIELD_DP32(cache_mem, CXL_HDM_DECODER0_CTRL, ERROR, 0);
+
+    offset = ((uint64_t)cache_mem[R_CXL_HDM_DECODER0_BASE_HI] << 32) |
+             cache_mem[R_CXL_HDM_DECODER0_BASE_LO];
+    size = ((uint64_t)cache_mem[R_CXL_HDM_DECODER0_SIZE_HI] << 32) |
+           cache_mem[R_CXL_HDM_DECODER0_SIZE_LO];
+
+    range_init_nofail(&window, mr->addr, memory_region_size(mr));
+    range_init_nofail(&device, offset, size);
+
+    if (!range_contains_range(&window, &device)) {
+        ARRAY_FIELD_DP32(cache_mem, CXL_HDM_DECODER0_CTRL, ERROR, 1);
+        return;
+    }
+
+    /*
+     * FIXME: Support resizing.
+     * Maybe just memory_region_ram_resize(pmem, size, &err)?
+     */
+    if (size != memory_region_size(pmem)) {
+        ARRAY_FIELD_DP32(cache_mem, CXL_HDM_DECODER0_CTRL, ERROR, 1);
+        return;
+    }
+
+    cxl_set_addr(ct3d, offset, &err);
+    if (err) {
+        ARRAY_FIELD_DP32(cache_mem, CXL_HDM_DECODER0_CTRL, ERROR, 1);
+        return;
+    }
+    memory_region_set_enabled(pmem, true);
+
+    ARRAY_FIELD_DP32(cache_mem, CXL_HDM_DECODER0_CTRL, COMMITTED, 1);
+}
+
+static void ct3d_reg_write(void *opaque, hwaddr offset, uint64_t value, unsigned size)
+{
+    CXLComponentState *cxl_cstate = opaque;
+    ComponentRegisters *cregs = &cxl_cstate->crb;
+    CXLType3Dev *ct3d = container_of(cxl_cstate, CXLType3Dev, cxl_cstate);
+    uint32_t *cache_mem = cregs->cache_mem_registers;
+    bool should_commit = false;
+    int which_hdm = -1;
+
+    assert(size == 4);
+
+    switch (offset) {
+    case A_CXL_HDM_DECODER0_CTRL:
+        should_commit = FIELD_EX32(value, CXL_HDM_DECODER0_CTRL, COMMIT);
+        which_hdm = 0;
+        break;
+    default:
+        break;
+    }
+
+    stl_le_p((uint8_t *)cache_mem + offset, value);
+    if (should_commit)
+        hdm_decoder_commit(ct3d, which_hdm);
+}
+
 static void ct3_instance_init(Object *obj)
 {
     /* MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(obj); */
@@ -65,18 +143,13 @@ static void ct3_instance_init(Object *obj)
 static void ct3_finalize(Object *obj)
 {
     CXLType3Dev *ct3d = CT3(obj);
+    CXLComponentState *cxl_cstate = &ct3d->cxl_cstate;
+    ComponentRegisters *regs = &cxl_cstate->crb;
 
+    g_free((void *)regs->special_ops);
     g_free(ct3d->cxl_dstate.pmem);
 }
 
-#ifdef SET_PMEM_PADDR
-static void cxl_set_addr(CXLType3Dev *ct3d, hwaddr addr, Error **errp)
-{
-    MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(ct3d);
-    mdc->set_addr(MEMORY_DEVICE(ct3d), addr, errp);
-}
-#endif
-
 static void cxl_setup_memory(CXLType3Dev *ct3d, Error **errp)
 {
     MemoryRegionSection mrs;
@@ -160,6 +233,9 @@ static void ct3_realize(PCIDevice *pci_dev, Error **errp)
     ct3d->cxl_cstate.pdev = pci_dev;
     build_dvsecs(ct3d);
 
+    regs->special_ops = g_new0(MemoryRegionOps, 1);
+    regs->special_ops->write = ct3d_reg_write;
+
     cxl_component_register_block_init(OBJECT(pci_dev), cxl_cstate,
                                       TYPE_CXL_TYPE3_DEV);
 
Signature · 1 line
2.30.0
#52 lore raw

RFC PATCH v3 23/31 acpi/cxl: Add _OSC implementation (9.14.2)

In reply to Ben Widawsky

CXL 2.0 specification adds 2 new dwords to the existing _OSC definition from PCIe. The new dwords are accessed with a new uuid. This implementation supports what is in the specification.

We are currently in the process of trying to define a new definition for _OSC. See later work for an explanation.

---
 hw/acpi/Kconfig       |   5 ++
 hw/acpi/cxl.c         | 104 ++++++++++++++++++++++++++++++++++++++++++
 hw/acpi/meson.build   |   1 +
 hw/i386/acpi-build.c  |  12 ++++-
 include/hw/acpi/cxl.h |  23 ++++++++++
 5 files changed, 144 insertions(+), 1 deletion(-)
 create mode 100644 hw/acpi/cxl.c
 create mode 100644 include/hw/acpi/cxl.h
hw/acpi/Kconfig +5 -0
index 1932f66af8..b27907953e 100644
@@ -5,6 +5,7 @@ config ACPI_X86
     bool
     select ACPI
     select ACPI_NVDIMM
+    select ACPI_CXL
     select ACPI_CPU_HOTPLUG
     select ACPI_MEMORY_HOTPLUG
     select ACPI_HMAT
@@ -42,3 +43,7 @@ config ACPI_VMGENID
     depends on PC
 
 config ACPI_HW_REDUCED
+
+config ACPI_CXL
+    bool
+    depends on ACPI
hw/acpi/cxl.c new file +104 -0

104 lines not on this page · raw message

hw/acpi/meson.build +1 -0
index dd69577212..9f5c5ced28 100644
@@ -10,6 +10,7 @@ acpi_ss.add(when: 'CONFIG_ACPI_CPU_HOTPLUG', if_true: files('cpu_hotplug.c'))
 acpi_ss.add(when: 'CONFIG_ACPI_MEMORY_HOTPLUG', if_true: files('memory_hotplug.c'))
 acpi_ss.add(when: 'CONFIG_ACPI_NVDIMM', if_true: files('nvdimm.c'))
 acpi_ss.add(when: 'CONFIG_ACPI_PCI', if_true: files('pci.c'))
+acpi_ss.add(when: 'CONFIG_ACPI_CXL', if_true: files('cxl.c'))
 acpi_ss.add(when: 'CONFIG_ACPI_VMGENID', if_true: files('vmgenid.c'))
 acpi_ss.add(when: 'CONFIG_ACPI_HW_REDUCED', if_true: files('generic_event_device.c'))
 acpi_ss.add(when: 'CONFIG_ACPI_HMAT', if_true: files('hmat.c'))
hw/i386/acpi-build.c +11 -1
index ecdc10b148..2c2293b55f 100644
@@ -66,6 +66,7 @@
 #include "hw/acpi/aml-build.h"
 #include "hw/acpi/utils.h"
 #include "hw/acpi/pci.h"
+#include "hw/acpi/cxl.h"
 
 #include "qom/qom-qobject.h"
 #include "hw/i386/amd_iommu.h"
@@ -1201,11 +1202,20 @@ static void init_pci_acpi(Aml *dev, int uid, int type)
     if (type == PCI) {
         aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A03")));
         aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
-    } else {
+    } else if (type == PCIE) {
         aml_append(dev, aml_name_decl("_HID", aml_eisaid("PNP0A08")));
         aml_append(dev, aml_name_decl("_CID", aml_eisaid("PNP0A03")));
         aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
         aml_append(dev, build_q35_osc_method());
+    } else /* CXL */ {
+        struct Aml *pkg = aml_package(2);
+
+        aml_append(dev, aml_name_decl("_HID", aml_string("ACPI0016")));
+        aml_append(pkg, aml_eisaid("PNP0A08"));
+        aml_append(pkg, aml_eisaid("PNP0A03"));
+        aml_append(dev, aml_name_decl("_CID", pkg));
+        aml_append(dev, aml_name_decl("_UID", aml_int(uid)));
+        build_cxl_osc_method(dev);
     }
 }
 
include/hw/acpi/cxl.h new file +23 -0
new file mode 100644
index 0000000000..7b8f3b8a2e
@@ -0,0 +1,23 @@
+/*
+ * Copyright (C) 2020 Intel Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, see <http://www.gnu.org/licenses/>.
+ */
+
+#ifndef HW_ACPI_CXL_H
+#define HW_ACPI_CXL_H
+
+void build_cxl_osc_method(Aml *dev);
+
+#endif
Signature · 1 line
2.30.0
#53 lore raw

RFC PATCH v3 24/31 tests/acpi: allow CEDT table addition

In reply to Ben Widawsky
---
 tests/data/acpi/pc/CEDT                     | 0
 tests/data/acpi/q35/CEDT                    | 0
 tests/qtest/bios-tables-test-allowed-diff.h | 2 ++
 3 files changed, 2 insertions(+)
 create mode 100644 tests/data/acpi/pc/CEDT
 create mode 100644 tests/data/acpi/q35/CEDT
tests/data/acpi/pc/CEDT new file +0 -0
new file mode 100644
index 0000000000..e69de29bb2
tests/data/acpi/q35/CEDT new file +0 -0
new file mode 100644
index 0000000000..e69de29bb2
tests/qtest/bios-tables-test-allowed-diff.h +2 -0
index dfb8523c8b..9b07f1e1ff 100644
@@ -1 +1,3 @@
 /* List of comma-separated changed AML files to ignore */
+"tests/data/acpi/pc/CEDT",
+"tests/data/acpi/q35/CEDT",
Signature · 1 line
2.30.0
#54 lore raw

RFC PATCH v3 25/31 acpi/cxl: Create the CEDT (9.14.1)

In reply to Ben Widawsky

The CXL Early Discovery Table is defined in the CXL 2.0 specification as a way for the OS to get CXL specific information from the system firmware.

CXL 2.0 specification adds an _HID, ACPI0016, for CXL capable host bridges, with a _CID of PNP0A08 (PCIe host bridge). CXL aware software is able to use this initiate the proper _OSC method, and get the _UID which is referenced by the CEDT. Therefore the existence of an ACPI0016 device allows a CXL aware driver perform the necessary actions. For a CXL capable OS, this works. For a CXL unaware OS, this works.

CEDT awaremess requires more. The motivation for ACPI0017 is to provide the possibility of having a Linux CXL module that can work on a legacy Linux kernel. Linux core PCI/ACPI which won't be built as a module, will see the _CID of PNP0A08 and bind a driver to it. If we later loaded a driver for ACPI0016, Linux won't be able to bind it to the hardware because it has already bound the PNP0A08 driver. The ACPI0017 device is an opportunity to have an object to bind a driver will be used by a Linux driver to walk the CXL topology and do everything that we would have preferred to do with ACPI0016.

There is another motivation for an ACPI0017 device which isn't implemented here. An operating system needs an attach point for a non-volatile region provider that understands cross-hostbridge interleaving. Since QEMU emulation doesn't support interleaving yet, this is more important on the OS side, for now.

As of CXL 2.0 spec, only 1 sub structure is defined, the CXL Host Bridge Structure (CHBS) which is primarily useful for telling the OS exactly where the MMIO for the host bridge is.

v2: Update CHBS to spec released definition
v3: squash ACPI0017 in now that it's ratified.

---
 hw/acpi/cxl.c                       | 69 +++++++++++++++++++++++++++++
 hw/i386/acpi-build.c                | 25 ++++++++++-
 hw/pci-bridge/pci_expander_bridge.c | 21 +--------
 include/hw/acpi/cxl.h               |  4 ++
 include/hw/pci/pci_bridge.h         | 25 +++++++++++
 5 files changed, 123 insertions(+), 21 deletions(-)
hw/acpi/cxl.c +69 -0
index 7124d5a1a3..68db0fe3a8 100644
@@ -18,14 +18,83 @@
  */
 
 #include "qemu/osdep.h"
+#include "hw/sysbus.h"
+#include "hw/pci/pci_bridge.h"
+#include "hw/pci/pci_host.h"
 #include "hw/cxl/cxl.h"
+#include "hw/mem/memory-device.h"
 #include "hw/acpi/acpi.h"
 #include "hw/acpi/aml-build.h"
 #include "hw/acpi/bios-linker-loader.h"
 #include "hw/acpi/cxl.h"
+#include "hw/acpi/cxl.h"
 #include "qapi/error.h"
 #include "qemu/uuid.h"
 
+static void cedt_build_chbs(GArray *table_data, PXBDev *cxl)
+{
+    SysBusDevice *sbd = SYS_BUS_DEVICE(cxl->cxl.cxl_host_bridge);
+    struct MemoryRegion *mr = sbd->mmio[0].memory;
+
+    /* Type */
+    build_append_int_noprefix(table_data, 0, 1);
+
+    /* Reserved */
+    build_append_int_noprefix(table_data, 0, 1);
+
+    /* Record Length */
+    build_append_int_noprefix(table_data, 32, 2);
+
+    /* UID */
+    build_append_int_noprefix(table_data, cxl->uid, 4);
+
+    /* Version */
+    build_append_int_noprefix(table_data, 1, 4);
+
+    /* Reserved */
+    build_append_int_noprefix(table_data, 0, 4);
+
+    /* Base */
+    build_append_int_noprefix(table_data, mr->addr, 8);
+
+    /* Length */
+    build_append_int_noprefix(table_data, memory_region_size(mr), 8);
+}
+
+static int cxl_foreach_pxb_hb(Object *obj, void *opaque)
+{
+    Aml *cedt = opaque;
+
+    if (object_dynamic_cast(obj, TYPE_PXB_CXL_DEVICE)) {
+        PXBDev *pxb = PXB_CXL_DEV(obj);
+
+        cedt_build_chbs(cedt->buf, pxb);
+    }
+
+    return 0;
+}
+
+void cxl_build_cedt(GArray *table_offsets, GArray *table_data,
+                    BIOSLinker *linker)
+{
+    const int cedt_start = table_data->len;
+    Aml *cedt;
+
+    cedt = init_aml_allocator();
+
+    /* reserve space for CEDT header */
+    acpi_add_table(table_offsets, table_data);
+    acpi_data_push(cedt->buf, sizeof(AcpiTableHeader));
+
+    object_child_foreach_recursive(object_get_root(), cxl_foreach_pxb_hb, cedt);
+
+    /* copy AML table into ACPI tables blob and patch header there */
+    g_array_append_vals(table_data, cedt->buf->data, cedt->buf->len);
+    build_header(linker, table_data, (void *)(table_data->data + cedt_start),
+                 "CEDT", table_data->len - cedt_start, 1, NULL, NULL);
+    free_aml_allocator();
+}
+
 static Aml *__build_cxl_osc_method(void)
 {
     Aml *method, *if_uuid, *else_uuid, *if_arg1_not_1, *if_cxl, *if_caps_masked;
hw/i386/acpi-build.c +24 -1
index 2c2293b55f..7706856c49 100644
@@ -75,6 +75,8 @@
 #include "hw/acpi/ipmi.h"
 #include "hw/acpi/hmat.h"
 
+#include "hw/acpi/cxl.h"
+
 /* These are used to size the ACPI tables for -M pc-i440fx-1.7 and
  * -M pc-i440fx-2.0.  Even if the actual amount of AML generated grows
  * a little bit, there should be plenty of free space since the DSDT
@@ -1219,6 +1221,19 @@ static void init_pci_acpi(Aml *dev, int uid, int type)
     }
 }
 
+static void build_acpi0017(Aml *table)
+{
+    Aml *dev;
+    Aml *scope;
+
+    scope =  aml_scope("_SB");
+    dev = aml_device("CXLM");
+    aml_append(dev, aml_name_decl("_HID", aml_string("ACPI0017")));
+
+    aml_append(scope, dev);
+    aml_append(table, scope);
+}
+
 static void
 build_dsdt(GArray *table_data, BIOSLinker *linker,
            AcpiPmInfo *pm, AcpiMiscInfo *misc,
@@ -1235,6 +1250,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
     int root_bus_limit = 0xFF;
     PCIBus *bus = NULL;
     TPMIf *tpm = tpm_find();
+    bool cxl_present = false;
     int i;
     VMBusBridge *vmbus_bridge = vmbus_bridge_find();
 
@@ -1371,7 +1387,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
 
             scope = aml_scope("\\_SB");
             if (type == CXL) {
-                dev = aml_device("CXL%.01X", pci_bus_uid(bus));
+                dev = aml_device("CXL%.01X", uid);
             } else {
                 dev = aml_device("PC%.02X", bus_num);
             }
@@ -1391,6 +1407,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
 
             /* Handle the ranges for the PXB expanders */
             if (type == CXL) {
+                cxl_present = true;
                 uint64_t base = CXL_HOST_BASE + uid * 0x10000;
                 crs_range_insert(crs_range_set.mem_ranges, base,
                                  base + 0x10000 - 1);
@@ -1398,6 +1415,10 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         }
     }
 
+    if (cxl_present) {
+        build_acpi0017(dsdt);
+    }
+
     /*
      * At this point crs_range_set has all the ranges used by pci
      * busses *other* than PCI0.  These ranges will be excluded from
@@ -2278,6 +2299,8 @@ void acpi_build(AcpiBuildTables *tables, MachineState *machine)
                           machine->nvdimms_state, machine->ram_slots);
     }
 
+    cxl_build_cedt(table_offsets, tables_blob, tables->linker);
+
     acpi_add_table(table_offsets, tables_blob);
     build_waet(tables_blob, tables->linker);
 
hw/pci-bridge/pci_expander_bridge.c +1 -20
index af1450c69d..6458d5b76e 100644
@@ -57,26 +57,6 @@ DECLARE_INSTANCE_CHECKER(PXBDev, PXB_DEV,
 DECLARE_INSTANCE_CHECKER(PXBDev, PXB_PCIE_DEV,
                          TYPE_PXB_PCIE_DEVICE)
 
-#define TYPE_PXB_CXL_DEVICE "pxb-cxl"
-DECLARE_INSTANCE_CHECKER(PXBDev, PXB_CXL_DEV,
-                         TYPE_PXB_CXL_DEVICE)
-
-struct PXBDev {
-    /*< private >*/
-    PCIDevice parent_obj;
-    /*< public >*/
-
-    uint8_t bus_nr;
-    uint16_t numa_node;
-    int32_t uid;
-    struct cxl_dev {
-        HostMemoryBackend *memory_window[CXL_WINDOW_MAX];
-
-        uint32_t num_windows;
-        hwaddr *window_base[CXL_WINDOW_MAX];
-    } cxl;
-};
-
 typedef struct CXLHost {
     PCIHostState parent_obj;
 
@@ -351,6 +331,7 @@ static void pxb_dev_realize_common(PCIDevice *dev, enum BusType type,
         bus = pci_root_bus_new(ds, dev_name, NULL, NULL, 0, TYPE_PXB_CXL_BUS);
         bus->flags |= PCI_BUS_CXL;
         PXB_CXL_HOST(ds)->dev = PXB_CXL_DEV(dev);
+        PXB_CXL_DEV(dev)->cxl.cxl_host_bridge = ds;
     } else {
         bus = pci_root_bus_new(ds, "pxb-internal", NULL, NULL, 0, TYPE_PXB_BUS);
         bds = qdev_new("pci-bridge");
include/hw/acpi/cxl.h +4 -0
index 7b8f3b8a2e..db2063f8c9 100644
@@ -18,6 +18,10 @@
 #ifndef HW_ACPI_CXL_H
 #define HW_ACPI_CXL_H
 
+#include "hw/acpi/bios-linker-loader.h"
+
+void cxl_build_cedt(GArray *table_offsets, GArray *table_data,
+                    BIOSLinker *linker);
 void build_cxl_osc_method(Aml *dev);
 
 #endif
include/hw/pci/pci_bridge.h +25 -0
index a94d350034..50dd7fdf33 100644
@@ -28,6 +28,7 @@
 
 #include "hw/pci/pci.h"
 #include "hw/pci/pci_bus.h"
+#include "hw/cxl/cxl.h"
 #include "qom/object.h"
 
 typedef struct PCIBridgeWindows PCIBridgeWindows;
@@ -81,6 +82,30 @@ struct PCIBridge {
 #define PCI_BRIDGE_DEV_PROP_MSI        "msi"
 #define PCI_BRIDGE_DEV_PROP_SHPC       "shpc"
 
+struct PXBDev {
+    /*< private >*/
+    PCIDevice parent_obj;
+    /*< public >*/
+
+    uint8_t bus_nr;
+    uint16_t numa_node;
+    int32_t uid;
+
+    struct cxl_dev {
+        HostMemoryBackend *memory_window[CXL_WINDOW_MAX];
+
+        uint32_t num_windows;
+        hwaddr *window_base[CXL_WINDOW_MAX];
+
+        void *cxl_host_bridge; /* Pointer to a CXLHost */
+    } cxl;
+};
+
+typedef struct PXBDev PXBDev;
+#define TYPE_PXB_CXL_DEVICE "pxb-cxl"
+DECLARE_INSTANCE_CHECKER(PXBDev, PXB_CXL_DEV,
+                         TYPE_PXB_CXL_DEVICE)
+
 int pci_bridge_ssvid_init(PCIDevice *dev, uint8_t offset,
                           uint16_t svid, uint16_t ssid,
                           Error **errp);
Signature · 1 line
2.30.0
#55 lore raw

RFC PATCH v3 26/31 tests/acpi: Add new CEDT files

In reply to Ben Widawsky
---
 tests/data/acpi/pc/CEDT                     | Bin 0 -> 36 bytes
 tests/data/acpi/q35/CEDT                    | Bin 0 -> 36 bytes
 tests/qtest/bios-tables-test-allowed-diff.h |   2 --
 3 files changed, 2 deletions(-)
tests/data/acpi/pc/CEDT binary +0 -0
index e69de29bb2d1d6434b8b29ae775ad8c2e48c5391..ebf9b54b0b27d9efca53359c3c2e560511f0e165 100644
GIT binary patch
tests/data/acpi/q35/CEDT binary +0 -0
index e69de29bb2d1d6434b8b29ae775ad8c2e48c5391..ebf9b54b0b27d9efca53359c3c2e560511f0e165 100644
GIT binary patch
tests/qtest/bios-tables-test-allowed-diff.h +0 -2
index 9b07f1e1ff..dfb8523c8b 100644
@@ -1,3 +1 @@
 /* List of comma-separated changed AML files to ignore */
-"tests/data/acpi/pc/CEDT",
-"tests/data/acpi/q35/CEDT",
Signature · 1 line
2.30.0
#56 lore raw

RFC PATCH v3 27/31 hw/cxl/device: Add some trivial commands

In reply to Ben Widawsky

GET_FW_INFO and GET_PARTITION_INFO, for this emulation, is equivalent to info already returned in the IDENTIFY command. To have a more robust implementation, add those.

---
 hw/cxl/cxl-mailbox-utils.c | 65 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 65 insertions(+)
hw/cxl/cxl-mailbox-utils.c +65 -0
index f92dfad882..dc8e0eb08e 100644
@@ -43,6 +43,8 @@ enum {
         #define CLEAR_RECORDS   0x1
         #define GET_INTERRUPT_POLICY   0x2
         #define SET_INTERRUPT_POLICY   0x3
+    FIRMWARE_UPDATE = 0x02,
+        #define GET_INFO      0x0
     TIMESTAMP   = 0x03,
         #define GET           0x0
         #define SET           0x1
@@ -51,6 +53,8 @@ enum {
         #define GET_LOG       0x1
     IDENTIFY    = 0x40,
         #define MEMORY_DEVICE 0x0
+    CCLS        = 0x41,
+        #define GET_PARTITION_INFO     0x0
 };
 
 /* 8.2.8.4.5.1 Command Return Codes */
@@ -125,11 +129,13 @@ define_mailbox_handler_zeroed(EVENTS_GET_RECORDS, 0x20);
 define_mailbox_handler_nop(EVENTS_CLEAR_RECORDS);
 define_mailbox_handler_zeroed(EVENTS_GET_INTERRUPT_POLICY, 4);
 define_mailbox_handler_nop(EVENTS_SET_INTERRUPT_POLICY);
+declare_mailbox_handler(FIRMWARE_UPDATE_GET_INFO);
 declare_mailbox_handler(TIMESTAMP_GET);
 declare_mailbox_handler(TIMESTAMP_SET);
 declare_mailbox_handler(LOGS_GET_SUPPORTED);
 declare_mailbox_handler(LOGS_GET_LOG);
 declare_mailbox_handler(IDENTIFY_MEMORY_DEVICE);
+declare_mailbox_handler(CCLS_GET_PARTITION_INFO);
 
 #define IMMEDIATE_CONFIG_CHANGE (1 << 1)
 #define IMMEDIATE_POLICY_CHANGE (1 << 3)
@@ -143,15 +149,50 @@ static struct cxl_cmd cxl_cmd_set[256][256] = {
     CXL_CMD(EVENTS, CLEAR_RECORDS, ~0, IMMEDIATE_LOG_CHANGE),
     CXL_CMD(EVENTS, GET_INTERRUPT_POLICY, 0, 0),
     CXL_CMD(EVENTS, SET_INTERRUPT_POLICY, 4, IMMEDIATE_CONFIG_CHANGE),
+    CXL_CMD(FIRMWARE_UPDATE, GET_INFO, 0, 0),
     CXL_CMD(TIMESTAMP, GET, 0, 0),
     CXL_CMD(TIMESTAMP, SET, 8, IMMEDIATE_POLICY_CHANGE),
     CXL_CMD(LOGS, GET_SUPPORTED, 0, 0),
     CXL_CMD(LOGS, GET_LOG, 0x18, 0),
     CXL_CMD(IDENTIFY, MEMORY_DEVICE, 0, 0),
+    CXL_CMD(CCLS, GET_PARTITION_INFO, 0, 0),
 };
 
 #undef CXL_CMD
 
+/*
+ * 8.2.9.2.1
+ */
+define_mailbox_handler(FIRMWARE_UPDATE_GET_INFO)
+{
+    struct {
+        uint8_t slots_supported;
+        uint8_t slot_info;
+        uint8_t caps;
+        uint8_t rsvd[0xd];
+        char fw_rev1[0x10];
+        char fw_rev2[0x10];
+        char fw_rev3[0x10];
+        char fw_rev4[0x10];
+    } __attribute__((packed)) *fw_info;
+    _Static_assert(sizeof(*fw_info) == 0x50, "Bad firmware info size");
+
+    if (memory_region_size(cxl_dstate->pmem) < (256 << 20)) {
+        return CXL_MBOX_INTERNAL_ERROR;
+    }
+
+    fw_info = (void *)cmd->payload;
+    memset(fw_info, 0, sizeof(*fw_info));
+
+    fw_info->slots_supported = 2;
+    fw_info->slot_info = BIT(0) | BIT(3);
+    fw_info->caps = 0;
+    snprintf(fw_info->fw_rev1, 0x10, "BWFW VERSION %02d", 0);
+
+    *len = sizeof(*fw_info);
+    return CXL_MBOX_SUCCESS;
+}
+
 /*
  * 8.2.9.3.1
  */
@@ -296,6 +337,30 @@ define_mailbox_handler(IDENTIFY_MEMORY_DEVICE)
     return CXL_MBOX_SUCCESS;
 }
 
+define_mailbox_handler(CCLS_GET_PARTITION_INFO)
+{
+    struct {
+        uint64_t active_vmem;
+        uint64_t active_pmem;
+        uint64_t next_vmem;
+        uint64_t next_pmem;
+    } __attribute__((packed)) *part_info = (void *)cmd->payload;
+    _Static_assert(sizeof(*part_info) == 0x20, "Bad get partition info size");
+
+    if (memory_region_size(cxl_dstate->pmem) < (256 << 20)) {
+        return CXL_MBOX_INTERNAL_ERROR;
+    }
+
+    /* PMEM only */
+    part_info->active_vmem = 0;
+    part_info->next_vmem = 0;
+    part_info->active_pmem = memory_region_size(cxl_dstate->pmem);
+    part_info->next_pmem = part_info->active_pmem;
+
+    *len = sizeof(*part_info);
+    return CXL_MBOX_SUCCESS;
+}
+
 void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
 {
     uint16_t ret = CXL_MBOX_SUCCESS;
Signature · 1 line
2.30.0
#57 lore raw

RFC PATCH v3 28/31 hw/cxl/device: Plumb real LSA sizing

In reply to Ben Widawsky

This should introduce no change. Subsequent work will make use of this new class member.

---
 hw/cxl/cxl-mailbox-utils.c  |  4 ++++
 hw/mem/cxl_type3.c          | 24 +++++++++---------------
 include/hw/cxl/cxl.h        |  1 -
 include/hw/cxl/cxl_device.h | 24 ++++++++++++++++++++++++
 4 files changed, 37 insertions(+), 16 deletions(-)
hw/cxl/cxl-mailbox-utils.c +4 -0
index dc8e0eb08e..2637250c7b 100644
@@ -321,6 +321,9 @@ define_mailbox_handler(IDENTIFY_MEMORY_DEVICE)
     } __attribute__((packed)) *id;
     _Static_assert(sizeof(*id) == 0x43, "Bad identify size");
 
+    CXLType3Dev *ct3d = container_of(cxl_dstate, CXLType3Dev, cxl_dstate);
+    CXLType3Class *cvc = CXL_TYPE3_DEV_GET_CLASS(ct3d);
+
     if (memory_region_size(cxl_dstate->pmem) < (256 << 20)) {
         return CXL_MBOX_INTERNAL_ERROR;
     }
@@ -332,6 +335,7 @@ define_mailbox_handler(IDENTIFY_MEMORY_DEVICE)
     snprintf(id->fw_revision, 0x10, "BWFW VERSION %02d", 0);
     id->total_capacity = memory_region_size(cxl_dstate->pmem);
     id->persistent_capacity = memory_region_size(cxl_dstate->pmem);
+    id->lsa_size = cvc->get_lsa_size(ct3d);
 
     *len = sizeof(*id);
     return CXL_MBOX_SUCCESS;
hw/mem/cxl_type3.c +9 -15
index fe02c3b63c..074d1dd41f 100644
@@ -13,21 +13,6 @@
 #include "sysemu/hostmem.h"
 #include "hw/cxl/cxl.h"
 
-typedef struct cxl_type3_dev {
-    /* Private */
-    PCIDevice parent_obj;
-
-    /* Properties */
-    uint64_t size;
-    HostMemoryBackend *hostmem;
-
-    /* State */
-    CXLComponentState cxl_cstate;
-    CXLDeviceState cxl_dstate;
-} CXLType3Dev;
-
-#define CT3(obj) OBJECT_CHECK(CXLType3Dev, (obj), TYPE_CXL_TYPE3_DEV)
-
 static void build_dvsecs(CXLType3Dev *ct3d)
 {
     CXLComponentState *cxl_cstate = &ct3d->cxl_cstate;
@@ -310,11 +295,17 @@ static void pc_dimm_md_fill_device_info(const MemoryDeviceState *md,
     info->type = MEMORY_DEVICE_INFO_KIND_CXL;
 }
 
+static uint64_t get_lsa_size(CXLType3Dev *ct3d)
+{
+    return 0;
+}
+
 static void ct3_class_init(ObjectClass *oc, void *data)
 {
     DeviceClass *dc = DEVICE_CLASS(oc);
     PCIDeviceClass *pc = PCI_DEVICE_CLASS(oc);
     MemoryDeviceClass *mdc = MEMORY_DEVICE_CLASS(oc);
+    CXLType3Class *cvc = CXL_TYPE3_DEV_CLASS(oc);
 
     pc->realize = ct3_realize;
     pc->class_id = PCI_CLASS_STORAGE_EXPRESS;
@@ -332,11 +323,14 @@ static void ct3_class_init(ObjectClass *oc, void *data)
     mdc->fill_device_info = pc_dimm_md_fill_device_info;
     mdc->get_plugged_size = memory_device_get_region_size;
     mdc->set_addr = cxl_md_set_addr;
+
+    cvc->get_lsa_size = get_lsa_size;
 }
 
 static const TypeInfo ct3d_info = {
     .name = TYPE_CXL_TYPE3_DEV,
     .parent = TYPE_PCI_DEVICE,
+    .class_size = sizeof(struct CXLType3Class),
     .class_init = ct3_class_init,
     .instance_size = sizeof(CXLType3Dev),
     .instance_init = ct3_instance_init,
include/hw/cxl/cxl.h +0 -1
index 809ed7de60..c7ca42930f 100644
@@ -23,4 +23,3 @@
 #define CXL_WINDOW_MAX 10
 
 #endif
-
include/hw/cxl/cxl_device.h +24 -0
index ca5328a581..a79a0f106c 100644
@@ -219,4 +219,28 @@ REG32(CXL_MEM_DEV_STS, 0)
     FIELD(CXL_MEM_DEV_STS, MBOX_READY, 4, 1)
     FIELD(CXL_MEM_DEV_STS, RESET_NEEDED, 5, 3)
 
+typedef struct cxl_type3_dev {
+    /* Private */
+    PCIDevice parent_obj;
+
+    /* Properties */
+    uint64_t size;
+    HostMemoryBackend *hostmem;
+    HostMemoryBackend *lsa;
+
+    /* State */
+    CXLComponentState cxl_cstate;
+    CXLDeviceState cxl_dstate;
+} CXLType3Dev;
+
+#define CT3(obj) OBJECT_CHECK(CXLType3Dev, (obj), TYPE_CXL_TYPE3_DEV)
+
+struct CXLType3Class {
+    /* Private */
+    PCIDeviceClass parent_class;
+
+    /* public */
+    uint64_t (*get_lsa_size)(CXLType3Dev *ct3d);
+};
+
 #endif
Signature · 1 line
2.30.0
#58 lore raw

RFC PATCH v3 29/31 hw/cxl/device: Implement get/set LSA

In reply to Ben Widawsky
---
 hw/cxl/cxl-mailbox-utils.c  | 50 +++++++++++++++++++++++++++++++++
 hw/mem/cxl_type3.c          | 56 ++++++++++++++++++++++++++++++++++++-
 include/hw/cxl/cxl_device.h |  9 ++++++
 3 files changed, 114 insertions(+), 1 deletion(-)
hw/cxl/cxl-mailbox-utils.c +50 -0
index 2637250c7b..c133cf0341 100644
@@ -55,6 +55,8 @@ enum {
         #define MEMORY_DEVICE 0x0
     CCLS        = 0x41,
         #define GET_PARTITION_INFO     0x0
+        #define GET_LSA       0x2
+        #define SET_LSA       0x3
 };
 
 /* 8.2.8.4.5.1 Command Return Codes */
@@ -136,8 +138,11 @@ declare_mailbox_handler(LOGS_GET_SUPPORTED);
 declare_mailbox_handler(LOGS_GET_LOG);
 declare_mailbox_handler(IDENTIFY_MEMORY_DEVICE);
 declare_mailbox_handler(CCLS_GET_PARTITION_INFO);
+declare_mailbox_handler(CCLS_GET_LSA);
+declare_mailbox_handler(CCLS_SET_LSA);
 
 #define IMMEDIATE_CONFIG_CHANGE (1 << 1)
+#define IMMEDIATE_DATA_CHANGE (1 << 1)
 #define IMMEDIATE_POLICY_CHANGE (1 << 3)
 #define IMMEDIATE_LOG_CHANGE (1 << 4)
 
@@ -156,6 +161,8 @@ static struct cxl_cmd cxl_cmd_set[256][256] = {
     CXL_CMD(LOGS, GET_LOG, 0x18, 0),
     CXL_CMD(IDENTIFY, MEMORY_DEVICE, 0, 0),
     CXL_CMD(CCLS, GET_PARTITION_INFO, 0, 0),
+    CXL_CMD(CCLS, GET_LSA, 0, 0),
+    CXL_CMD(CCLS, SET_LSA, ~0, IMMEDIATE_CONFIG_CHANGE | IMMEDIATE_DATA_CHANGE),
 };
 
 #undef CXL_CMD
@@ -365,6 +372,49 @@ define_mailbox_handler(CCLS_GET_PARTITION_INFO)
     return CXL_MBOX_SUCCESS;
 }
 
+define_mailbox_handler(CCLS_GET_LSA)
+{
+    struct {
+        uint32_t offset;
+        uint32_t length;
+    } __attribute__((packed, __aligned__(16))) *get_lsa;
+    CXLType3Dev *ct3d = container_of(cxl_dstate, CXLType3Dev, cxl_dstate);
+    CXLType3Class *cvc = CXL_TYPE3_DEV_GET_CLASS(ct3d);
+    uint32_t offset, length;
+
+    get_lsa = (void *)cmd->payload;
+    offset = get_lsa->offset;
+    length = get_lsa->length;
+
+    *len = 0;
+    if (offset + length > cvc->get_lsa_size(ct3d)) {
+        return CXL_MBOX_INVALID_INPUT;
+    }
+
+    *len = cvc->get_lsa(ct3d, get_lsa, length, offset);
+    return CXL_MBOX_SUCCESS;
+}
+
+define_mailbox_handler(CCLS_SET_LSA)
+{
+    struct {
+        uint32_t offset;
+        uint32_t rsvd;
+        void *data;
+    } __attribute__((packed, __aligned__(16))) *set_lsa = (void *)cmd->payload;
+    CXLType3Dev *ct3d = container_of(cxl_dstate, CXLType3Dev, cxl_dstate);
+    CXLType3Class *cvc = CXL_TYPE3_DEV_GET_CLASS(ct3d);
+    uint16_t plen = *len;
+
+    *len = 0;
+    if ((set_lsa->offset + plen) > cvc->get_lsa_size(ct3d)) {
+        return CXL_MBOX_INVALID_INPUT;
+    }
+
+    cvc->set_lsa(ct3d, set_lsa->data, plen, set_lsa->offset);
+    return CXL_MBOX_SUCCESS;
+}
+
 void cxl_process_mailbox(CXLDeviceState *cxl_dstate)
 {
     uint16_t ret = CXL_MBOX_SUCCESS;
hw/mem/cxl_type3.c +55 -1

86 lines not on this page · raw message

include/hw/cxl/cxl_device.h +9 -0
index a79a0f106c..1869876ef6 100644
@@ -233,7 +233,11 @@ typedef struct cxl_type3_dev {
     CXLDeviceState cxl_dstate;
 } CXLType3Dev;
 
+#ifndef TYPE_CXL_TYPE3_DEV
+#define TYPE_CXL_TYPE3_DEV "cxl-type3"
+#endif
 #define CT3(obj) OBJECT_CHECK(CXLType3Dev, (obj), TYPE_CXL_TYPE3_DEV)
+OBJECT_DECLARE_TYPE(CXLType3Device, CXLType3Class, CXL_TYPE3_DEV)
 
 struct CXLType3Class {
     /* Private */
@@ -241,6 +245,11 @@ struct CXLType3Class {
 
     /* public */
     uint64_t (*get_lsa_size)(CXLType3Dev *ct3d);
+
+    uint64_t (*get_lsa)(CXLType3Dev *ct3d, void *buf, uint64_t size,
+                        uint64_t offset);
+    void (*set_lsa)(CXLType3Dev *ct3d, const void *buf, uint64_t size,
+                    uint64_t offset);
 };
 
 #endif
Signature · 1 line
2.30.0
#59 lore raw

RFC PATCH v3 30/31 qtest/cxl: Add very basic sanity tests

In reply to Ben Widawsky
---
 tests/qtest/cxl-test.c  | 93 +++++++++++++++++++++++++++++++++++++++++
 tests/qtest/meson.build |  4 ++
 2 files changed, 97 insertions(+)
 create mode 100644 tests/qtest/cxl-test.c
tests/qtest/cxl-test.c new file +93 -0
new file mode 100644
index 0000000000..00eca14faa
@@ -0,0 +1,93 @@
+/*
+ * QTest testcase for CXL
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ */
+
+#include "qemu/osdep.h"
+#include "libqtest-single.h"
+
+#define QEMU_PXB_CMD "-machine q35 -object memory-backend-file,id=cxl-mem1," \
+                     "share,mem-path=%s,size=512M "                          \
+                     "-device pxb-cxl,id=cxl.0,bus=pcie.0,bus_nr=52,uid=0,"  \
+                     "len-window-base=1,window-base[0]=0x4c0000000,memdev[0]=cxl-mem1"
+#define QEMU_RP "-device cxl-rp,id=rp0,bus=cxl.0,addr=0.0,chassis=0,slot=0"
+
+#define QEMU_T3D "-device cxl-type3,bus=rp0,memdev=cxl-mem1,id=cxl-pmem0,size=256M"
+
+static void cxl_basic_hb(void)
+{
+    qtest_start("-machine q35,cxl");
+    qtest_end();
+}
+
+static void cxl_basic_pxb(void)
+{
+    qtest_start("-machine q35 -device pxb-cxl,bus=pcie.0,uid=0");
+    qtest_end();
+}
+
+static void cxl_pxb_with_window(void)
+{
+    GString *cmdline;
+    char template[] = "/tmp/cxl-test-XXXXXX";
+    const char *tmpfs;
+
+    tmpfs = mkdtemp(template);
+
+    cmdline = g_string_new(NULL);
+    g_string_printf(cmdline, QEMU_PXB_CMD, tmpfs);
+
+    qtest_start(cmdline->str);
+    qtest_end();
+
+    g_string_free(cmdline, TRUE);
+}
+
+static void cxl_root_port(void)
+{
+    GString *cmdline;
+    char template[] = "/tmp/cxl-test-XXXXXX";
+    const char *tmpfs;
+
+    tmpfs = mkdtemp(template);
+
+    cmdline = g_string_new(NULL);
+    g_string_printf(cmdline, QEMU_PXB_CMD " %s", tmpfs, QEMU_RP);
+
+    qtest_start(cmdline->str);
+    qtest_end();
+
+    g_string_free(cmdline, TRUE);
+}
+
+static void cxl_t3d(void)
+{
+    GString *cmdline;
+    char template[] = "/tmp/cxl-test-XXXXXX";
+    const char *tmpfs;
+
+    tmpfs = mkdtemp(template);
+
+    cmdline = g_string_new(NULL);
+    g_string_printf(cmdline, QEMU_PXB_CMD " %s %s", tmpfs, QEMU_RP, QEMU_T3D);
+
+    qtest_start(cmdline->str);
+    qtest_end();
+
+    g_string_free(cmdline, TRUE);
+}
+
+int main(int argc, char **argv)
+{
+    g_test_init(&argc, &argv, NULL);
+
+    qtest_add_func("/pci/cxl/basic_hostbridge", cxl_basic_hb);
+    qtest_add_func("/pci/cxl/basic_pxb", cxl_basic_pxb);
+    qtest_add_func("/pci/cxl/pxb_with_window", cxl_pxb_with_window);
+    qtest_add_func("/pci/cxl/root_port", cxl_root_port);
+    qtest_add_func("/pci/cxl/type3_device", cxl_t3d);
+
+    return g_test_run();
+}
tests/qtest/meson.build +4 -0
index c83bc211b6..554152b7c5 100644
@@ -22,6 +22,9 @@ qtests_pci = \
   (config_all_devices.has_key('CONFIG_VGA') ? ['display-vga-test'] : []) +                  \
   (config_all_devices.has_key('CONFIG_IVSHMEM_DEVICE') ? ['ivshmem-test'] : [])
 
+qtests_cxl = \
+  (config_all_devices.has_key('CONFIG_CXL') ? ['cxl-test'] : [])
+
 qtests_i386 = \
   (slirp.found() ? ['pxe-test', 'test-netfilter'] : []) +             \
   (config_host.has_key('CONFIG_POSIX') ? ['test-filter-mirror'] : []) +                     \
@@ -48,6 +51,7 @@ qtests_i386 = \
   (config_all_devices.has_key('CONFIG_TPM_TIS_ISA') ? ['tpm-tis-swtpm-test'] : []) +        \
   (config_all_devices.has_key('CONFIG_RTL8139_PCI') ? ['rtl8139-test'] : []) +              \
   qtests_pci +                                                                              \
+  qtests_cxl +                                                                              \
   ['fdc-test',
    'ide-test',
    'hd-geo-test',
Signature · 1 line
2.30.0
#60 lore raw

RFC PATCH v3 31/31 WIP: i386/cxl: Initialize a host bridge

In reply to Ben Widawsky

This patch allows initializing the primary host bridge as a CXL capable hostbridge.

--
This patch is WIP.

---
 hw/arm/virt.c        |  1 +
 hw/core/machine.c    | 26 ++++++++++++++++++++++++++
 hw/i386/acpi-build.c |  8 +++++++-
 hw/i386/microvm.c    |  1 +
 hw/i386/pc.c         |  1 +
 hw/ppc/spapr.c       |  2 ++
 include/hw/boards.h  |  2 ++
 include/hw/cxl/cxl.h |  4 ++++
 8 files changed, 44 insertions(+), 1 deletion(-)
hw/arm/virt.c +1 -0
index 399da73454..fd5f5b656c 100644
@@ -2547,6 +2547,7 @@ static void virt_machine_class_init(ObjectClass *oc, void *data)
     hc->unplug_request = virt_machine_device_unplug_request_cb;
     hc->unplug = virt_machine_device_unplug_cb;
     mc->nvdimm_supported = true;
+    mc->cxl_supported = false;
     mc->auto_enable_numa_with_memhp = true;
     mc->auto_enable_numa_with_memdev = true;
     mc->default_ram_id = "mach-virt.ram";
hw/core/machine.c +26 -0
index de3b8f1b31..c739803854 100644
@@ -30,6 +30,7 @@
 #include "sysemu/qtest.h"
 #include "hw/pci/pci.h"
 #include "hw/mem/nvdimm.h"
+#include "hw/cxl/cxl.h"
 #include "migration/global_state.h"
 #include "migration/vmstate.h"
 
@@ -502,6 +503,20 @@ static void machine_set_nvdimm_persistence(Object *obj, const char *value,
     nvdimms_state->persistence_string = g_strdup(value);
 }
 
+static bool machine_get_cxl(Object *obj, Error **errp)
+{
+    MachineState *ms = MACHINE(obj);
+
+    return ms->cxl_devices_state->is_enabled;
+}
+
+static void machine_set_cxl(Object *obj, bool value, Error **errp)
+{
+    MachineState *ms = MACHINE(obj);
+
+    ms->cxl_devices_state->is_enabled = value;
+}
+
 void machine_class_allow_dynamic_sysbus_dev(MachineClass *mc, const char *type)
 {
     QAPI_LIST_PREPEND(mc->allowed_dynamic_sysbus_devices, g_strdup(type));
@@ -903,6 +918,16 @@ static void machine_initfn(Object *obj)
                                         "Valid values are cpu, mem-ctrl");
     }
 
+    if (mc->cxl_supported) {
+        Object *obj = OBJECT(ms);
+
+        ms->cxl_devices_state = g_new0(CXLState, 1);
+        object_property_add_bool(obj, "cxl", machine_get_cxl, machine_set_cxl);
+        object_property_set_description(obj, "cxl",
+                                        "Set on/off to enable/disable "
+                                        "CXL instantiation");
+    }
+
     if (mc->cpu_index_to_instance_props && mc->get_default_cpu_node_id) {
         ms->numa_state = g_new0(NumaState, 1);
         object_property_add_bool(obj, "hmat",
@@ -939,6 +964,7 @@ static void machine_finalize(Object *obj)
     g_free(ms->device_memory);
     g_free(ms->nvdimms_state);
     g_free(ms->numa_state);
+    g_free(ms->cxl_devices_state);
 }
 
 bool machine_usb(MachineState *machine)
hw/i386/acpi-build.c +7 -1
index 7706856c49..2250e6d27b 100644
@@ -53,6 +53,7 @@
 #include "sysemu/numa.h"
 #include "sysemu/reset.h"
 #include "hw/hyperv/vmbus-bridge.h"
+#include "hw/cxl/cxl.h"
 
 /* Supported chipsets: */
 #include "hw/southbridge/piix.h"
@@ -1277,8 +1278,13 @@ build_dsdt(GArray *table_data, BIOSLinker *linker,
         build_piix4_pci0_int(dsdt);
     } else {
         sb_scope = aml_scope("_SB");
+        /*
+         * XXX: CXL spec calls this "CXL0", but that would require lots of
+         * changes throughout and so even for CXL enabled, we call it "PCI0"
+         */
         dev = aml_device("PCI0");
-        init_pci_acpi(dev, 0, PCIE);
+        init_pci_acpi(dev, 0,
+                machine->cxl_devices_state->is_enabled ? CXL : PCIE);
         aml_append(dev, aml_name_decl("_ADR", aml_int(0)));
         aml_append(sb_scope, dev);
 
hw/i386/microvm.c +1 -0
index edf2b0f061..970b299a69 100644
@@ -688,6 +688,7 @@ static void microvm_class_init(ObjectClass *oc, void *data)
     mc->auto_enable_numa_with_memdev = false;
     mc->default_cpu_type = TARGET_DEFAULT_CPU_TYPE;
     mc->nvdimm_supported = false;
+    mc->cxl_supported = false;
     mc->default_ram_id = "microvm.ram";
 
     /* Avoid relying too much on kernel components */
hw/i386/pc.c +1 -0
index 5d41809b37..7350eeea9c 100644
@@ -1725,6 +1725,7 @@ static void pc_machine_class_init(ObjectClass *oc, void *data)
     hc->unplug = pc_machine_device_unplug_cb;
     mc->default_cpu_type = TARGET_DEFAULT_CPU_TYPE;
     mc->nvdimm_supported = true;
+    mc->cxl_supported = true;
     mc->default_ram_id = "pc.ram";
 
     object_class_property_add(oc, PC_MACHINE_MAX_RAM_BELOW_4G, "size",
hw/ppc/spapr.c +2 -0
index 6c47466fc2..9773dbd83c 100644
@@ -4440,6 +4440,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
     mc->has_hotpluggable_cpus = true;
     mc->nvdimm_supported = true;
+    mc->cxl_supported = false;
     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
     fwc->get_dev_path = spapr_get_fw_dev_path;
     nc->nmi_monitor_handler = spapr_nmi;
@@ -4600,6 +4601,7 @@ static void spapr_machine_4_2_class_options(MachineClass *mc)
     smc->default_caps.caps[SPAPR_CAP_FWNMI] = SPAPR_CAP_OFF;
     smc->rma_limit = 16 * GiB;
     mc->nvdimm_supported = false;
+    mc->cxl_supported = false;
 }
 
 DEFINE_SPAPR_MACHINE(4_2, "4.2", false);
include/hw/boards.h +2 -0
index 17b1f3f0b9..808f73e134 100644
@@ -206,6 +206,7 @@ struct MachineClass {
     bool ignore_boot_device_suffixes;
     bool smbus_no_migration_support;
     bool nvdimm_supported;
+    bool cxl_supported;
     bool numa_mem_supported;
     bool auto_enable_numa;
     const char *default_ram_id;
@@ -292,6 +293,7 @@ struct MachineState {
     CPUArchIdList *possible_cpus;
     CpuTopology smp;
     struct NVDIMMState *nvdimms_state;
+    struct CXLState *cxl_devices_state;
     struct NumaState *numa_state;
 };
 
include/hw/cxl/cxl.h +4 -0
index c7ca42930f..784b4f7a04 100644
@@ -22,4 +22,8 @@
 #define CXL_HOST_BASE 0xD0000000
 #define CXL_WINDOW_MAX 10
 
+typedef struct CXLState {
+    bool is_enabled;
+} CXLState;
+
 #endif
Signature · 1 line
2.30.0
#61 lore raw
In reply to Ben Widawsky
Patchew URL: https://patchew.org/QEMU/[email protected]/

Patchew URL: https://patchew.org/QEMU/[email protected]/

Hi,

This series seems to have some coding style problems. See output below for more information:

Type: series
Message-id: [email protected]
Subject: [RFC PATCH v3 00/31] CXL 2.0 Support

=== TEST SCRIPT BEGIN ===
#!/bin/bash
git rev-parse base > /dev/null || exit 0
git config --local diff.renamelimit 0
git config --local diff.renames True
git config --local diff.algorithm histogram
./scripts/checkpatch.pl --mailback base..
=== TEST SCRIPT END ===

Updating 3c8cf5a9c21ff8782164d1def7f44bd888713384
From https://github.com/patchew-project/qemu
* [new tag] patchew/[email protected] -> patchew/[email protected]
Switched to a new branch 'test'
e26ed22 WIP: i386/cxl: Initialize a host bridge
9329c2b qtest/cxl: Add very basic sanity tests
c140fd9 hw/cxl/device: Implement get/set LSA
8ed7755 hw/cxl/device: Plumb real LSA sizing
5f683ab hw/cxl/device: Add some trivial commands
4399501 tests/acpi: Add new CEDT files
6c13c92 acpi/cxl: Create the CEDT (9.14.1)
04a874a tests/acpi: allow CEDT table addition
50f82e6 acpi/cxl: Add _OSC implementation (9.14.2)
7eb8038 hw/cxl/device: Implement MMIO HDM decoding (8.2.5.12) ba80470 hw/cxl/device: Add a memory device (8.2.8.5)
54b9662 hw/cxl/rp: Add a root port
e70de08 hw/pxb/cxl: Add "windows" for host bridges
606831a acpi/pxb/cxl: Reserve host bridge MMIO
29a562b hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142) 32e7bdd hw/pci: Plumb _UID through host bridges
6651f84 tests/acpi: remove stale allowed tables
24837fc acpi/pci: Consolidate host bridge setup
52f548c qtest: allow DSDT acpi table changes
bdcd7d9 hw/pxb: Allow creation of a CXL PXB (host bridge) 5d67d7e hw/pci/cxl: Create a CXL bus type
3b0d310 hw/pxb: Use a type for realizing expanders
5ccf850 hw/cxl/device: Add log commands (8.2.9.4) + CEL 892e722 hw/cxl/device: Timestamp implementation (8.2.9.3) f2444bb hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1) 67fa438 hw/cxl/device: Add memory device utilities
cfa875c hw/cxl/device: Implement basic mailbox (8.2.8.4) bdd7975 hw/cxl/device: Implement the CAP array (8.2.8.1-2) c9e87d1 hw/cxl/device: Introduce a CXL device (8.2.8) 1cc9e2a hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5) 7b0e042 hw/pci/cxl: Add a CXL component type (interface)

=== OUTPUT BEGIN ===
1/31 Checking commit 7b0e042bc22b (hw/pci/cxl: Add a CXL component type (interface)) 2/31 Checking commit 1cc9e2a0a6d5 (hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5))
WARNING: line over 80 characters
#187: FILE: hw/cxl/cxl-component-utils.c:101:
+ reg_state[R_CXL_RAS_ERR_CAP_CTRL] = 0; /* CXL switches and devices must set */

WARNING: line over 80 characters
#193: FILE: hw/cxl/cxl-component-utils.c:107:
+ ARRAY_FIELD_DP32(reg_state, CXL_HDM_DECODER_GLOBAL_CONTROL, HDM_DECODER_ENABLE, 0);

WARNING: line over 80 characters
#406: FILE: include/hw/cxl/cxl_component.h:62:
+#define CXL_RAS_REGISTERS_OFFSET 0x80 /* Give ample space for caps before this */

WARNING: line over 80 characters
#417: FILE: include/hw/cxl/cxl_component.h:73:
+#define CXL_SEC_REGISTERS_OFFSET (CXL_RAS_REGISTERS_OFFSET + CXL_RAS_REGISTERS_SIZE)

WARNING: line over 80 characters
#421: FILE: include/hw/cxl/cxl_component.h:77:
+#define CXL_LINK_REGISTERS_OFFSET (CXL_SEC_REGISTERS_OFFSET + CXL_SEC_REGISTERS_SIZE)

WARNING: line over 80 characters
#465: FILE: include/hw/cxl/cxl_component.h:121:
+#define CXL_EXTSEC_REGISTERS_OFFSET (CXL_HDM_REGISTERS_OFFSET + CXL_HDM_REGISTERS_SIZE)

WARNING: line over 80 characters
#469: FILE: include/hw/cxl/cxl_component.h:125:
+#define CXL_IDE_REGISTERS_OFFSET (CXL_EXTSEC_REGISTERS_OFFSET + CXL_EXTSEC_REGISTERS_SIZE)

WARNING: line over 80 characters
#473: FILE: include/hw/cxl/cxl_component.h:129:
+#define CXL_SNOOP_REGISTERS_OFFSET (CXL_IDE_REGISTERS_OFFSET + CXL_IDE_REGISTERS_SIZE)

total: 0 errors, 8 warnings, 582 lines checked

Patch 2/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
3/31 Checking commit c9e87d150708 (hw/cxl/device: Introduce a CXL device (8.2.8))
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #36:
new file mode 100644

WARNING: line over 80 characters
#156: FILE: include/hw/cxl/cxl_device.h:116:
+#define CXL_DEVICE_CAPABILITY_HEADER_REGISTER(n, offset) \

total: 0 errors, 2 warnings, 162 lines checked

Patch 3/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
4/31 Checking commit bdd7975aa4bc (hw/cxl/device: Implement the CAP array (8.2.8.1-2))
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #23:
new file mode 100644

ERROR: Macros with complex values should be enclosed in parenthesis #159: FILE: include/hw/cxl/cxl_device.h:75:
+#define CXL_MMIO_SIZE \
+ CXL_DEVICE_CAP_REG_SIZE + CXL_DEVICE_REGISTERS_LENGTH + \
+ CXL_MAILBOX_REGISTERS_LENGTH

WARNING: line over 80 characters
#179: FILE: include/hw/cxl/cxl_device.h:138:
+#define cxl_device_cap_init(dstate, reg, cap_id) \

WARNING: line over 80 characters
#180: FILE: include/hw/cxl/cxl_device.h:139:
+ do { \

WARNING: line over 80 characters
#181: FILE: include/hw/cxl/cxl_device.h:140:
+ uint32_t *cap_hdrs = dstate->caps_reg_state32; \

WARNING: line over 80 characters
#182: FILE: include/hw/cxl/cxl_device.h:141:
+ int which = R_CXL_DEV_##reg##_CAP_HDR0; \

WARNING: line over 80 characters
#183: FILE: include/hw/cxl/cxl_device.h:142:
+ cap_hdrs[which] = \

WARNING: line over 80 characters
#184: FILE: include/hw/cxl/cxl_device.h:143:
+ FIELD_DP32(cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_ID, cap_id); \

WARNING: line over 80 characters
#185: FILE: include/hw/cxl/cxl_device.h:144:
+ cap_hdrs[which] = FIELD_DP32( \

WARNING: line over 80 characters
#186: FILE: include/hw/cxl/cxl_device.h:145:
+ cap_hdrs[which], CXL_DEV_##reg##_CAP_HDR0, CAP_VERSION, 1); \

WARNING: line over 80 characters
#187: FILE: include/hw/cxl/cxl_device.h:146:
+ cap_hdrs[which + 1] = \

WARNING: line over 80 characters
#188: FILE: include/hw/cxl/cxl_device.h:147:
+ FIELD_DP32(cap_hdrs[which + 1], CXL_DEV_##reg##_CAP_HDR1, \

WARNING: line over 80 characters
#189: FILE: include/hw/cxl/cxl_device.h:148:
+ CAP_OFFSET, CXL_##reg##_REGISTERS_OFFSET); \

WARNING: line over 80 characters
#190: FILE: include/hw/cxl/cxl_device.h:149:
+ cap_hdrs[which + 2] = \

WARNING: line over 80 characters
#191: FILE: include/hw/cxl/cxl_device.h:150:
+ FIELD_DP32(cap_hdrs[which + 2], CXL_DEV_##reg##_CAP_HDR2, \

WARNING: line over 80 characters
#192: FILE: include/hw/cxl/cxl_device.h:151:
+ CAP_LENGTH, CXL_##reg##_REGISTERS_LENGTH); \

total: 1 errors, 15 warnings, 158 lines checked

Patch 4/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS.

5/31 Checking commit cfa875c3c48b (hw/cxl/device: Implement basic mailbox (8.2.8.4))
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #191:
new file mode 100644

ERROR: space prohibited between function name and open parenthesis '(' #264: FILE: hw/cxl/cxl-mailbox-utils.c:69:
+typedef ret_code (*opcode_handler)(struct cxl_cmd *cmd,

total: 1 errors, 1 warnings, 416 lines checked

Patch 5/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS.

6/31 Checking commit 67fa4383326f (hw/cxl/device: Add memory device utilities) 7/31 Checking commit f2444bba9cd7 (hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1)) 8/31 Checking commit 892e722ede17 (hw/cxl/device: Timestamp implementation (8.2.9.3)) 9/31 Checking commit 5ccf850db462 (hw/cxl/device: Add log commands (8.2.9.4) + CEL) 10/31 Checking commit 3b0d3108b26c (hw/pxb: Use a type for realizing expanders) 11/31 Checking commit 5d67d7eb3d82 (hw/pci/cxl: Create a CXL bus type)
12/31 Checking commit bdcd7d995e9f (hw/pxb: Allow creation of a CXL PXB (host bridge)) 13/31 Checking commit 52f548ca385d (qtest: allow DSDT acpi table changes) 14/31 Checking commit 24837fc1bb0e (acpi/pci: Consolidate host bridge setup) 15/31 Checking commit 6651f845de76 (tests/acpi: remove stale allowed tables) 16/31 Checking commit 32e7bdd7607d (hw/pci: Plumb _UID through host bridges)
WARNING: line over 80 characters
#113: FILE: hw/pci-bridge/pci_expander_bridge.c:422:
+ error_setg(errp, "pxb-cxl devices must have a valid uid (0-2147483647)");

total: 0 errors, 1 warnings, 113 lines checked

Patch 16/31 has style problems, please review. If any of these errors
are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
17/31 Checking commit 29a562ba112f (hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142)) 18/31 Checking commit 606831a33eda (acpi/pxb/cxl: Reserve host bridge MMIO)
19/31 Checking commit e70de0847b95 (hw/pxb/cxl: Add "windows" for host bridges)
WARNING: line over 80 characters
#133: FILE: hw/pci-bridge/pci_expander_bridge.c:516:
+ warn_report("memory-windows should be set when creating CXL host bridges");

total: 0 errors, 1 warnings, 127 lines checked

Patch 19/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS.
20/31 Checking commit 54b96623ffd8 (hw/cxl/rp: Add a root port)
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #42:
new file mode 100644

total: 0 errors, 1 warnings, 268 lines checked

Patch 20/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
21/31 Checking commit ba804700c6a6 (hw/cxl/device: Add a memory device (8.2.8.5))
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #155:
new file mode 100644

WARNING: line over 80 characters
#272: FILE: hw/mem/cxl_type3.c:113:
+ "Not enough free space (%zd) required for device (%" PRId64 ")",

total: 0 errors, 2 warnings, 501 lines checked

Patch 21/31 has style problems, please review. If any of these errors
are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
22/31 Checking commit 7eb80384a516 (hw/cxl/device: Implement MMIO HDM decoding (8.2.5.12))
WARNING: line over 80 characters
#92: FILE: hw/mem/cxl_type3.c:113:
+static void ct3d_reg_write(void *opaque, hwaddr offset, uint64_t value, unsigned size)

total: 0 errors, 1 warnings, 114 lines checked

Patch 22/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
23/31 Checking commit 50f82e6dddb1 (acpi/cxl: Add _OSC implementation (9.14.2))
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #45:
new file mode 100644

WARNING: Block comments use a leading /* on a separate line #188: FILE: hw/i386/acpi-build.c:1210:
+ } else /* CXL */ {

total: 0 errors, 2 warnings, 176 lines checked

Patch 23/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS.
24/31 Checking commit 04a874a8d982 (tests/acpi: allow CEDT table addition)
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #16:
new file mode 100644

total: 0 errors, 1 warnings, 3 lines checked

Patch 24/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see
CHECKPATCH in MAINTAINERS.
25/31 Checking commit 6c13c9205746 (acpi/cxl: Create the CEDT (9.14.1)) 26/31 Checking commit 43995014aa28 (tests/acpi: Add new CEDT files) 27/31 Checking commit 5f683ab6ee1a (hw/cxl/device: Add some trivial commands) 28/31 Checking commit 8ed7755c7a36 (hw/cxl/device: Plumb real LSA sizing) 29/31 Checking commit c140fd9d4517 (hw/cxl/device: Implement get/set LSA) 30/31 Checking commit 9329c2b72e7f (qtest/cxl: Add very basic sanity tests)
WARNING: added, moved or deleted file(s), does MAINTAINERS need updating? #15:
new file mode 100644

WARNING: line over 80 characters
#36: FILE: tests/qtest/cxl-test.c:17:
+#define QEMU_T3D "-device cxl-type3,bus=rp0,memdev=cxl-mem1,id=cxl-pmem0,size=256M"

total: 0 errors, 2 warnings, 109 lines checked

Patch 30/31 has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS.
31/31 Checking commit e26ed228d062 (WIP: i386/cxl: Initialize a host bridge) === OUTPUT END ===

Test command exited with code: 1

The full log is available at
http://patchew.org/logs/[email protected]/testing.checkpatch/?type=message.
---
Email generated automatically by Patchew [https://patchew.org/].
Please send your feedback to [email protected]

#62 lore raw
In reply to Ben Widawsky

I've started a barebones project plan:
https://gitlab.com/bwidawsk/qemu/-/snippets/2070304

Jonathan, if you have a moment, perhaps you can send a MR summarizing CDAT/DOE work from you and Chris?

If folks feel priorities are drastically off, we can discuss it in the snippet comments.

As for wider acceptance, if I'm looking at this from the QEMU community perspective, better test cases are really needed. If your fingers are itching for some typing, might I suggest starting with that.

I've opted not to use issue tracker for this because I am hopeful this won't be a long living gitlab project.

Quoted 217 lines · Ben Widawsky

Major changes since v2 [1]:
* Removed all register endian/alignment/size checking. Using core functionality instead. This untested on big endian hosts, but Should Work(tm).
* Fix component capability header generation (off by 1).
* Fixed HDM programming (multiple issues).
* Fixed timestamp command implementations.
* Added commands: GET_FIRMWARE_UPDATE_INFO, GET_PARTITION_INFO, GET_LSA, SET_LSA

Things have remained fairly stable since since v2. The biggest change here is definitely the HDM programming which has received limited (but not 0) testing in the Linux driver.

Jonathan Cameron has gotten this patch series working on ARM [2], and added some much sought after functionality [3].

---

I've started #cxl on OFTC IRC for discussion. Please feel free to use that channel for questions or suggestions in addition to #qemu.

---

Introduce emulation of Compute Express Link 2.0
(https://www.computeexpresslink.org/). Specifically, add support for Type 3 memory expanders with persistent memory.

The emulation has been critical to get the Linux enabling started [4], it would be an ideal place to land regression tests for different topology handling, and there may be applications for this emulation as a way for a guest to manipulate its address space relative to different performance memories.

Three of the five CXL component types are emulated with some level of
functionality: host bridge, root port, and memory device. All components and devices implement basic MMIO. Devices/memory devices implement the mailbo interface. Basic ACPI support is also included. Upstream ports and downstream ports aren't implemented (the two components needed to make up a switch).

CXL 2.0 is built on top of PCIe (see spec for details). As a result, much of the implementation utilizes existing PCI paradigms. To implement the host bridge, I've chosen to use PXB (PCI Expander Bridge). It seemed to be the most natural fit even though it doesn't directly map to how hardware will work. For persistent capacity of the memory device, I utilized the memory subsystem (hw/mem).

We have 3 reasons why this work is valuable:
1. Linux driver feature development benefits from emulation both due to a lack of initial hardware availability, but also, as is seen with NVDIMM/PMEM emulation, there is value in being able to share topologies with system-software developers even after hardware is available.

2. The Linux kernel's unit test suite for NVDIMM/PMEM ended up injecting fake resources via custom modules (nfit_test). In retrospect a QEMU emulation of nfit_test capabilities would have made the test environment more portable, and allowed for easier community contributions of example configurations.

3. This is still being fleshed out, but in short it provides a standardized mechanism for the guest to provide feedback to the host about size and placement needs of the memory. After the host gives the guest a physical window mapping to the CXL device, the emulated HDM decoders allow the guest a way to tell the host how much it wants and where. There are likely simpler ways to do this, but they'd require inventing a new interface and you'd need to have diverging driver code in the guest programming of the HDM decoder vs. the host. Since we've already done this work, why not use it?

There is quite a long list of work to do for full spec compliance, but I don't believe that any of it precludes merging. Off the top of my head:
- Main host bridge support (WIP)
- Interleaving
- Better Tests
- Hot plug support
- Emulating volatile capacity
- CDAT emulation [3]

The flow of the patches in general is to define all the data structures and registers associated with the various components in a top down manner. Host bridge, component, ports, devices. Then, the actual implementation is done in the same order.

The summary is:
1-5: Infrastructure for component and device emulation 6-9: Basic mailbox command implementations
10-19: Implement CXL host bridges as PXB devices 20: Implement a root port
21-22: Implement a memory device
23-26: ACPI bits
27-29: Add some more advanced mailbox command implementations 30: Start working on enabling the main host bridge 31: Basic test case

---

[1]: https://lore.kernel.org/qemu-devel/[email protected]/
[2]: https://lore.kernel.org/qemu-devel/[email protected]/
[3]: https://lore.kernel.org/qemu-devel/[email protected]/
[4]: https://lore.kernel.org/linux-cxl/[email protected]/

---

Ben Widawsky (31):
hw/pci/cxl: Add a CXL component type (interface) hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5) hw/cxl/device: Introduce a CXL device (8.2.8) hw/cxl/device: Implement the CAP array (8.2.8.1-2) hw/cxl/device: Implement basic mailbox (8.2.8.4) hw/cxl/device: Add memory device utilities hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1) hw/cxl/device: Timestamp implementation (8.2.9.3) hw/cxl/device: Add log commands (8.2.9.4) + CEL hw/pxb: Use a type for realizing expanders hw/pci/cxl: Create a CXL bus type
hw/pxb: Allow creation of a CXL PXB (host bridge)
qtest: allow DSDT acpi table changes
acpi/pci: Consolidate host bridge setup
tests/acpi: remove stale allowed tables
hw/pci: Plumb _UID through host bridges hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142) acpi/pxb/cxl: Reserve host bridge MMIO
hw/pxb/cxl: Add "windows" for host bridges
hw/cxl/rp: Add a root port
hw/cxl/device: Add a memory device (8.2.8.5) hw/cxl/device: Implement MMIO HDM decoding (8.2.5.12) acpi/cxl: Add _OSC implementation (9.14.2) tests/acpi: allow CEDT table addition
acpi/cxl: Create the CEDT (9.14.1)
tests/acpi: Add new CEDT files
hw/cxl/device: Add some trivial commands hw/cxl/device: Plumb real LSA sizing
hw/cxl/device: Implement get/set LSA
qtest/cxl: Add very basic sanity tests
WIP: i386/cxl: Initialize a host bridge

 MAINTAINERS                         |   6 +
 hw/Kconfig                          |   1 +
 hw/acpi/Kconfig                     |   5 +
 hw/acpi/cxl.c                       | 173 ++++++++++
 hw/acpi/meson.build                 |   1 +
 hw/arm/virt.c                       |   1 +
 hw/core/machine.c                   |  26 ++
 hw/core/numa.c                      |   3 +
 hw/cxl/Kconfig                      |   3 +
 hw/cxl/cxl-component-utils.c        | 208 ++++++++++++
 hw/cxl/cxl-device-utils.c           | 264 +++++++++++++++
 hw/cxl/cxl-mailbox-utils.c          | 498 ++++++++++++++++++++++++++++
 hw/cxl/meson.build                  |   5 +
 hw/i386/acpi-build.c                |  87 ++++-
 hw/i386/microvm.c                   |   1 +
 hw/i386/pc.c                        |   2 +
 hw/mem/Kconfig                      |   5 +
 hw/mem/cxl_type3.c                  | 405 ++++++++++++++++++++++
 hw/mem/meson.build                  |   1 +
 hw/meson.build                      |   1 +
 hw/pci-bridge/Kconfig               |   5 +
 hw/pci-bridge/cxl_root_port.c       | 231 +++++++++++++
 hw/pci-bridge/meson.build           |   1 +
 hw/pci-bridge/pci_expander_bridge.c | 209 +++++++++++-
 hw/pci-bridge/pcie_root_port.c      |   6 +-
 hw/pci/pci.c                        |  32 +-
 hw/pci/pcie.c                       |  30 ++
 hw/ppc/spapr.c                      |   2 +
 include/hw/acpi/cxl.h               |  27 ++
 include/hw/boards.h                 |   2 +
 include/hw/cxl/cxl.h                |  29 ++
 include/hw/cxl/cxl_component.h      | 187 +++++++++++
 include/hw/cxl/cxl_device.h         | 255 ++++++++++++++
 include/hw/cxl/cxl_pci.h            | 160 +++++++++
 include/hw/pci/pci.h                |  15 +
 include/hw/pci/pci_bridge.h         |  25 ++
 include/hw/pci/pci_bus.h            |   8 +
 include/hw/pci/pci_ids.h            |   1 +
 monitor/hmp-cmds.c                  |  15 +
 qapi/machine.json                   |   1 +
 tests/data/acpi/pc/CEDT             | Bin 0 -> 36 bytes
 tests/data/acpi/pc/DSDT             | Bin 5065 -> 5065 bytes
 tests/data/acpi/pc/DSDT.acpihmat    | Bin 6390 -> 6390 bytes
 tests/data/acpi/pc/DSDT.bridge      | Bin 6924 -> 6924 bytes
 tests/data/acpi/pc/DSDT.cphp        | Bin 5529 -> 5529 bytes
 tests/data/acpi/pc/DSDT.dimmpxm     | Bin 6719 -> 6719 bytes
 tests/data/acpi/pc/DSDT.hpbridge    | Bin 5026 -> 5026 bytes
 tests/data/acpi/pc/DSDT.hpbrroot    | Bin 3084 -> 3084 bytes
 tests/data/acpi/pc/DSDT.ipmikcs     | Bin 5137 -> 5137 bytes
 tests/data/acpi/pc/DSDT.memhp       | Bin 6424 -> 6424 bytes
 tests/data/acpi/pc/DSDT.numamem     | Bin 5071 -> 5071 bytes
 tests/data/acpi/pc/DSDT.roothp      | Bin 5261 -> 5261 bytes
 tests/data/acpi/q35/CEDT            | Bin 0 -> 36 bytes
 tests/data/acpi/q35/DSDT            | Bin 7801 -> 7801 bytes
 tests/data/acpi/q35/DSDT.acpihmat   | Bin 9126 -> 9126 bytes
 tests/data/acpi/q35/DSDT.bridge     | Bin 7819 -> 7819 bytes
 tests/data/acpi/q35/DSDT.cphp       | Bin 8265 -> 8265 bytes
 tests/data/acpi/q35/DSDT.dimmpxm    | Bin 9455 -> 9455 bytes
 tests/data/acpi/q35/DSDT.ipmibt     | Bin 7876 -> 7876 bytes
 tests/data/acpi/q35/DSDT.memhp      | Bin 9160 -> 9160 bytes
 tests/data/acpi/q35/DSDT.mmio64     | Bin 8932 -> 8932 bytes
 tests/data/acpi/q35/DSDT.numamem    | Bin 7807 -> 7807 bytes
 tests/qtest/cxl-test.c              |  93 ++++++
 tests/qtest/meson.build             |   4 +
 64 files changed, 3004 insertions(+), 30 deletions(-)
 create mode 100644 hw/acpi/cxl.c
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/cxl-device-utils.c
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 hw/mem/cxl_type3.c
 create mode 100644 hw/pci-bridge/cxl_root_port.c
 create mode 100644 include/hw/acpi/cxl.h
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_device.h
 create mode 100644 include/hw/cxl/cxl_pci.h
 create mode 100644 tests/data/acpi/pc/CEDT
 create mode 100644 tests/data/acpi/q35/CEDT
 create mode 100644 tests/qtest/cxl-test.c
Signature · 1 line
2.30.0
#63 lore raw
Quoted 2 lines · On Wed, 3 Feb 2021 09:42:16 -0800 Ben Widawsky <[email protected]> wrote:

I've started a barebones project plan:
https://gitlab.com/bwidawsk/qemu/-/snippets/2070304

Great.

Quoted 3 lines

Jonathan, if you have a moment, perhaps you can send a MR summarizing CDAT/DOE work from you and Chris?

I need to catch up with what Chris has posted, but sure after that I'll add to your doc if Chris doesn't get there first. My intent is to let Chris get on with the DOE QEMU support. Plenty of other stuff to do as long as it covers what I need (if not I'll hack stuff on top :) Will catch up with reviewing that in the next few days.

I'll also add some other comments on the plan when I get a chance. Will need to fake at least a partial switch sometime soon for example.

Quoted 10 lines

If folks feel priorities are drastically off, we can discuss it in the snippet comments.

As for wider acceptance, if I'm looking at this from the QEMU community perspective, better test cases are really needed. If your fingers are itching for some typing, might I suggest starting with that.

I've opted not to use issue tracker for this because I am hopeful this won't be a long living gitlab project.

All sounds good.

I've not reviewed that much on the last few patches in here, at least partly because a bunch of them have todo comments so I'm assuming they are very much a work in progress.

One thing I will note is this has become large and complex enough that I'd be tempted to start separating the 'racey cutting edge' parts from bits that have been moderately stable for a while. Hopefully some of that stable part can get wider review without the fun stuff and all the churn related to that.

Jonathan

Quoted 219 lines · Ben Widawsky
Quoted 217 lines

Major changes since v2 [1]:
* Removed all register endian/alignment/size checking. Using core functionality instead. This untested on big endian hosts, but Should Work(tm).
* Fix component capability header generation (off by 1).
* Fixed HDM programming (multiple issues).
* Fixed timestamp command implementations.
* Added commands: GET_FIRMWARE_UPDATE_INFO, GET_PARTITION_INFO, GET_LSA, SET_LSA

Things have remained fairly stable since since v2. The biggest change here is definitely the HDM programming which has received limited (but not 0) testing in the Linux driver.

Jonathan Cameron has gotten this patch series working on ARM [2], and added some much sought after functionality [3].

---

I've started #cxl on OFTC IRC for discussion. Please feel free to use that channel for questions or suggestions in addition to #qemu.

---

Introduce emulation of Compute Express Link 2.0
(https://www.computeexpresslink.org/). Specifically, add support for Type 3 memory expanders with persistent memory.

The emulation has been critical to get the Linux enabling started [4], it would be an ideal place to land regression tests for different topology handling, and there may be applications for this emulation as a way for a guest to manipulate its address space relative to different performance memories.

Three of the five CXL component types are emulated with some level of
functionality: host bridge, root port, and memory device. All components and devices implement basic MMIO. Devices/memory devices implement the mailbo interface. Basic ACPI support is also included. Upstream ports and downstream ports aren't implemented (the two components needed to make up a switch).

CXL 2.0 is built on top of PCIe (see spec for details). As a result, much of the implementation utilizes existing PCI paradigms. To implement the host bridge, I've chosen to use PXB (PCI Expander Bridge). It seemed to be the most natural fit even though it doesn't directly map to how hardware will work. For persistent capacity of the memory device, I utilized the memory subsystem (hw/mem).

We have 3 reasons why this work is valuable:
1. Linux driver feature development benefits from emulation both due to a lack of initial hardware availability, but also, as is seen with NVDIMM/PMEM emulation, there is value in being able to share topologies with system-software developers even after hardware is available.

2. The Linux kernel's unit test suite for NVDIMM/PMEM ended up injecting fake resources via custom modules (nfit_test). In retrospect a QEMU emulation of nfit_test capabilities would have made the test environment more portable, and allowed for easier community contributions of example configurations.

3. This is still being fleshed out, but in short it provides a standardized mechanism for the guest to provide feedback to the host about size and placement needs of the memory. After the host gives the guest a physical window mapping to the CXL device, the emulated HDM decoders allow the guest a way to tell the host how much it wants and where. There are likely simpler ways to do this, but they'd require inventing a new interface and you'd need to have diverging driver code in the guest programming of the HDM decoder vs. the host. Since we've already done this work, why not use it?

There is quite a long list of work to do for full spec compliance, but I don't believe that any of it precludes merging. Off the top of my head:
- Main host bridge support (WIP)
- Interleaving
- Better Tests
- Hot plug support
- Emulating volatile capacity
- CDAT emulation [3]

The flow of the patches in general is to define all the data structures and registers associated with the various components in a top down manner. Host bridge, component, ports, devices. Then, the actual implementation is done in the same order.

The summary is:
1-5: Infrastructure for component and device emulation 6-9: Basic mailbox command implementations
10-19: Implement CXL host bridges as PXB devices 20: Implement a root port
21-22: Implement a memory device
23-26: ACPI bits
27-29: Add some more advanced mailbox command implementations 30: Start working on enabling the main host bridge 31: Basic test case

---

[1]: https://lore.kernel.org/qemu-devel/[email protected]/
[2]: https://lore.kernel.org/qemu-devel/[email protected]/
[3]: https://lore.kernel.org/qemu-devel/[email protected]/
[4]: https://lore.kernel.org/linux-cxl/[email protected]/

---

Ben Widawsky (31):
hw/pci/cxl: Add a CXL component type (interface) hw/cxl/component: Introduce CXL components (8.1.x, 8.2.5) hw/cxl/device: Introduce a CXL device (8.2.8) hw/cxl/device: Implement the CAP array (8.2.8.1-2) hw/cxl/device: Implement basic mailbox (8.2.8.4) hw/cxl/device: Add memory device utilities hw/cxl/device: Add cheap EVENTS implementation (8.2.9.1) hw/cxl/device: Timestamp implementation (8.2.9.3) hw/cxl/device: Add log commands (8.2.9.4) + CEL hw/pxb: Use a type for realizing expanders hw/pci/cxl: Create a CXL bus type
hw/pxb: Allow creation of a CXL PXB (host bridge)
qtest: allow DSDT acpi table changes
acpi/pci: Consolidate host bridge setup
tests/acpi: remove stale allowed tables
hw/pci: Plumb _UID through host bridges hw/cxl/component: Implement host bridge MMIO (8.2.5, table 142) acpi/pxb/cxl: Reserve host bridge MMIO
hw/pxb/cxl: Add "windows" for host bridges
hw/cxl/rp: Add a root port
hw/cxl/device: Add a memory device (8.2.8.5) hw/cxl/device: Implement MMIO HDM decoding (8.2.5.12) acpi/cxl: Add _OSC implementation (9.14.2) tests/acpi: allow CEDT table addition
acpi/cxl: Create the CEDT (9.14.1)
tests/acpi: Add new CEDT files
hw/cxl/device: Add some trivial commands hw/cxl/device: Plumb real LSA sizing
hw/cxl/device: Implement get/set LSA
qtest/cxl: Add very basic sanity tests
WIP: i386/cxl: Initialize a host bridge

 MAINTAINERS                         |   6 +
 hw/Kconfig                          |   1 +
 hw/acpi/Kconfig                     |   5 +
 hw/acpi/cxl.c                       | 173 ++++++++++
 hw/acpi/meson.build                 |   1 +
 hw/arm/virt.c                       |   1 +
 hw/core/machine.c                   |  26 ++
 hw/core/numa.c                      |   3 +
 hw/cxl/Kconfig                      |   3 +
 hw/cxl/cxl-component-utils.c        | 208 ++++++++++++
 hw/cxl/cxl-device-utils.c           | 264 +++++++++++++++
 hw/cxl/cxl-mailbox-utils.c          | 498 ++++++++++++++++++++++++++++
 hw/cxl/meson.build                  |   5 +
 hw/i386/acpi-build.c                |  87 ++++-
 hw/i386/microvm.c                   |   1 +
 hw/i386/pc.c                        |   2 +
 hw/mem/Kconfig                      |   5 +
 hw/mem/cxl_type3.c                  | 405 ++++++++++++++++++++++
 hw/mem/meson.build                  |   1 +
 hw/meson.build                      |   1 +
 hw/pci-bridge/Kconfig               |   5 +
 hw/pci-bridge/cxl_root_port.c       | 231 +++++++++++++
 hw/pci-bridge/meson.build           |   1 +
 hw/pci-bridge/pci_expander_bridge.c | 209 +++++++++++-
 hw/pci-bridge/pcie_root_port.c      |   6 +-
 hw/pci/pci.c                        |  32 +-
 hw/pci/pcie.c                       |  30 ++
 hw/ppc/spapr.c                      |   2 +
 include/hw/acpi/cxl.h               |  27 ++
 include/hw/boards.h                 |   2 +
 include/hw/cxl/cxl.h                |  29 ++
 include/hw/cxl/cxl_component.h      | 187 +++++++++++
 include/hw/cxl/cxl_device.h         | 255 ++++++++++++++
 include/hw/cxl/cxl_pci.h            | 160 +++++++++
 include/hw/pci/pci.h                |  15 +
 include/hw/pci/pci_bridge.h         |  25 ++
 include/hw/pci/pci_bus.h            |   8 +
 include/hw/pci/pci_ids.h            |   1 +
 monitor/hmp-cmds.c                  |  15 +
 qapi/machine.json                   |   1 +
 tests/data/acpi/pc/CEDT             | Bin 0 -> 36 bytes
 tests/data/acpi/pc/DSDT             | Bin 5065 -> 5065 bytes
 tests/data/acpi/pc/DSDT.acpihmat    | Bin 6390 -> 6390 bytes
 tests/data/acpi/pc/DSDT.bridge      | Bin 6924 -> 6924 bytes
 tests/data/acpi/pc/DSDT.cphp        | Bin 5529 -> 5529 bytes
 tests/data/acpi/pc/DSDT.dimmpxm     | Bin 6719 -> 6719 bytes
 tests/data/acpi/pc/DSDT.hpbridge    | Bin 5026 -> 5026 bytes
 tests/data/acpi/pc/DSDT.hpbrroot    | Bin 3084 -> 3084 bytes
 tests/data/acpi/pc/DSDT.ipmikcs     | Bin 5137 -> 5137 bytes
 tests/data/acpi/pc/DSDT.memhp       | Bin 6424 -> 6424 bytes
 tests/data/acpi/pc/DSDT.numamem     | Bin 5071 -> 5071 bytes
 tests/data/acpi/pc/DSDT.roothp      | Bin 5261 -> 5261 bytes
 tests/data/acpi/q35/CEDT            | Bin 0 -> 36 bytes
 tests/data/acpi/q35/DSDT            | Bin 7801 -> 7801 bytes
 tests/data/acpi/q35/DSDT.acpihmat   | Bin 9126 -> 9126 bytes
 tests/data/acpi/q35/DSDT.bridge     | Bin 7819 -> 7819 bytes
 tests/data/acpi/q35/DSDT.cphp       | Bin 8265 -> 8265 bytes
 tests/data/acpi/q35/DSDT.dimmpxm    | Bin 9455 -> 9455 bytes
 tests/data/acpi/q35/DSDT.ipmibt     | Bin 7876 -> 7876 bytes
 tests/data/acpi/q35/DSDT.memhp      | Bin 9160 -> 9160 bytes
 tests/data/acpi/q35/DSDT.mmio64     | Bin 8932 -> 8932 bytes
 tests/data/acpi/q35/DSDT.numamem    | Bin 7807 -> 7807 bytes
 tests/qtest/cxl-test.c              |  93 ++++++
 tests/qtest/meson.build             |   4 +
 64 files changed, 3004 insertions(+), 30 deletions(-)
 create mode 100644 hw/acpi/cxl.c
 create mode 100644 hw/cxl/Kconfig
 create mode 100644 hw/cxl/cxl-component-utils.c
 create mode 100644 hw/cxl/cxl-device-utils.c
 create mode 100644 hw/cxl/cxl-mailbox-utils.c
 create mode 100644 hw/cxl/meson.build
 create mode 100644 hw/mem/cxl_type3.c
 create mode 100644 hw/pci-bridge/cxl_root_port.c
 create mode 100644 include/hw/acpi/cxl.h
 create mode 100644 include/hw/cxl/cxl.h
 create mode 100644 include/hw/cxl/cxl_component.h
 create mode 100644 include/hw/cxl/cxl_device.h
 create mode 100644 include/hw/cxl/cxl_pci.h
 create mode 100644 tests/data/acpi/pc/CEDT
 create mode 100644 tests/data/acpi/q35/CEDT
 create mode 100644 tests/qtest/cxl-test.c
Signature · 1 line
2.30.0
#64 lore raw

RFC PATCH 1/1 hw/mem/cxl_type3: Go back to subregions

In reply to Ben Widawsky

Each device allocates its memory (persistent only for now) out of a container memory that represents a "window" that would be defined by the host bridge. For example, a host bridge may claim all traffic from 0x0 - 0x4000; it might then also direct 0x1000-0x1fff to a specific CXL device. Change the memory region type claiming 0x1000-0x1fff from an alias, to a subregion.

v1 and v2 of the patch series both used a subregion. There were two issues with this and so for v3, an alias was chosen mimicking nvdimm.

The switch to alias left an issue in the implementation. There's logic that check to make sure two memory regions (ie. two devices under the same host bridge) couldn't collide. While hardware doesn't make this guarantee, it's nice for driver debug. There is no clean way to do that with an alias.

More importantly, this change was inspired by implementing support for both volatile and persistent memory. In that case, you may have multiple devices which the BIOS is going to assign address ranges to. Since we are the BIOS in this case, having a way of finding used space in the memory window so that you can allocate the next chunk is easily accomplished here.

With this, I have the following output from info mtree

    0000004c00000000-0000004c1fffffff (prio 0, ram): cxl-mem1
      0000004c00000000-0000004c0fffffff (prio 1, i/o): cxl_type3-memory

And info memory-devices
Memory device [cxl]: "cxl-pmem0"
addr: 0x4c00000000
slot: 0
node: 0
size: 268435456
memdev: /objects/cxl-mem1
hotplugged: false
hotpluggable: true

This functionality has been tested with a WIP linux patch which amounts to this:

       reg = readl(cxlm->regs.hdm_decoder + CXL_HDM_DECODER_CAP_OFFSET);

       dev_err(dev, "decoder cap:\n"
                    "\tcount: %lu\n",
               FIELD_GET(CXL_HDM_DECODER_COUNT_MASK, reg));

       writel(0x4c, cxlm->regs.hdm_decoder + CXL_HDM_DECODER0_BASE_HIGH_OFFSET);
       writel(0, cxlm->regs.hdm_decoder + CXL_HDM_DECODER0_BASE_LOW_OFFSET);
       writel(0, cxlm->regs.hdm_decoder + CXL_HDM_DECODER0_SIZE_HIGH_OFFSET);
       writel(256 << 20, cxlm->regs.hdm_decoder + CXL_HDM_DECODER0_SIZE_LOW_OFFSET);
       writel(BIT(9),  cxlm->regs.hdm_decoder + CXL_HDM_DECODER0_CTRL_OFFSET);

       tmp = ioremap_uc(0x4c00000000, 4096);
       writel(0x20, tmp);
Cc Jonathan Cameron <[email protected]>
---
 hw/mem/cxl_type3.c | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)
hw/mem/cxl_type3.c +11 -8
index bf33ddb915..33991079a6 100644
@@ -46,7 +46,9 @@ static void build_dvsecs(CXLType3Dev *ct3d)
 static void cxl_set_addr(CXLType3Dev *ct3d, hwaddr addr, Error **errp)
 {
     MemoryDeviceClass *mdc = MEMORY_DEVICE_GET_CLASS(ct3d);
-    mdc->set_addr(MEMORY_DEVICE(ct3d), addr, errp);
+    MemoryRegion *mr = host_memory_backend_get_memory(ct3d->hostmem);
+
+    mdc->set_addr(MEMORY_DEVICE(ct3d), addr - mr->addr, errp);
 }
 
 static void hdm_decoder_commit(CXLType3Dev *ct3d, int which)
@@ -180,13 +182,13 @@ static void cxl_setup_memory(CXLType3Dev *ct3d, Error **errp)
 
     memory_region_set_nonvolatile(pmem, true);
     memory_region_set_enabled(pmem, false);
-    memory_region_init_alias(pmem, OBJECT(ct3d), "cxl_type3-memory", mr, 0,
-                             ct3d->size);
+    memory_region_init(pmem, OBJECT(ct3d), "cxl_type3-memory", ct3d->size);
+    memory_region_add_subregion_overlap(mr, 0, pmem, 1);
     ct3d->cxl_dstate.pmem = pmem;
 
 #ifdef SET_PMEM_PADDR
     /* This path will initialize the memory device as if BIOS had done it */
-    cxl_set_addr(ct3d, mr->addr + offset, errp);
+    cxl_set_addr(ct3d, offset, errp);
     memory_region_set_enabled(pmem, true);
 #endif
 }
@@ -246,8 +248,11 @@ static uint64_t cxl_md_get_addr(const MemoryDeviceState *md)
     CXLType3Dev *ct3d = CT3(md);
     MemoryRegion *pmem = ct3d->cxl_dstate.pmem;
 
-    assert(pmem->alias);
-    return pmem->alias_offset;
+    if (pmem) {
+        return pmem->addr + pmem->container->addr;
+    }
+
+    return 0;
 }
 
 static void cxl_md_set_addr(MemoryDeviceState *md, uint64_t addr, Error **errp)
@@ -255,8 +260,6 @@ static void cxl_md_set_addr(MemoryDeviceState *md, uint64_t addr, Error **errp)
     CXLType3Dev *ct3d = CT3(md);
     MemoryRegion *pmem = ct3d->cxl_dstate.pmem;
 
-    assert(pmem->alias);
-    memory_region_set_alias_offset(pmem, addr);
     memory_region_set_address(pmem, addr);
 }
 
Signature · 1 line
2.30.2