Skip to contentSkip to conversation

Add ACPI CPER firmware first error injection on ARM emulation

PATCH Superseded Mauro Carvalho Chehab 3 participants 34 messages Aug 16, 2024 – Sep 14, 2024

Subsystems (2)

v8 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

v8 cover Add ACPI CPER firmware first error injection on ARM emulation

This series add support for injecting generic CPER records. Such records are generated outside QEMU via a provided script.

On this version, I added two optional patches at the end:
- acpi/ghes: cleanup generic error data logic

It drops some obvious comments from some already-existing code. As we're already doing lots of changes at the code, it sounded reasonable to me to have such cleanup here;

- acpi/ghes: check if the BIOS pointers for HEST are correct

QEMU has two ways to navigate to a CPER start data: via its memory address or indirectly following 2 BIOS pointers. OS only have the latter one. This patch validates if the BIOS links used by the OS were properly produced, comparing to the actual location of the CPER record.

---

v8:
- Fix one of the BIOS links that were incorrect;
- Changed mem error internal injection to use a common code;
- No more hardcoded values for CPER: instead of using just the payload at the QAPI, it now has the full raw CPER there;
- Error injection script now supports changing fields at the Generic Error Data section of the CPER;
- Several minor cleanups.

v7:
- Change the way offsets are calculated and used on HEST table. Now, it is compatible with migrations as all offsets are relative to the HEST table;
- GHES interface is now more generic: the entire CPER is sent via QMP, instead of just the payload;
- Some code cleanups to make the code more robust;
- The python script now uses QEMUMonitorProtocol class.

v6:
- PNP0C33 device creation moved to aml-build.c;
- acpi_ghes record functions now use ACPI notify parameter, instead of source ID;
- the number of source IDs is now automatically calculated;
- some code cleanups and function/var renames;
- some fixes and cleanups at the error injection script;
- ghes cper stub now produces an error if cper JSON is not compiled;
- Offset calculation logic for GHES was refactored;
- Updated documentation to reflect the GHES allocated size;
- Added a x-mpidr object for QOM usage;
- Added a patch making usage of x-mpidr field at ARM injection script;

v5:
- CPER guid is now passing as string;
- raw-data is now passed with base64 encode;
- Removed several GPIO left-overs from arm/virt.c changes;
- Lots of cleanups and improvements at the error injection script. It now better handles QMP dialog and doesn't print debug messages. Also, code was split on two modules, to make easier to add more error injection commands.

v4:
- CPER generation moved to happen outside QEMU;
- One patch adding support for mpidr query was removed.

v3:
- patch 1 cleanups with some comment changes and adding another place where the poweroff GPIO define should be used. No changes on other patches (except due to conflict resolution).

v2:
- added a new patch using a define for GPIO power pin;
- patch 2 changed to also use a define for generic error GPIO pin;
- a couple cleanups at patch 2 removing uneeded else clauses.

Example of generating a CPER record:

$ scripts/ghes_inject.py -d arm -p 0xdeadbeef
GUID: e19e3d16-bc11-11e4-9caa-c2051d5d46b0
Generic Error Status Block (20 bytes):
      00000000  01 00 00 00 00 00 00 00 00 00 00 00 90 00 00 00   ................
      00000010  00 00 00 00                                       ....

Generic Error Data Entry (72 bytes):

      00000000  16 3d 9e e1 11 bc e4 11 9c aa c2 05 1d 5d 46 b0   .=...........]F.
      00000010  00 00 00 00 00 03 00 00 48 00 00 00 00 00 00 00   ........H.......
      00000020  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00   ................
      00000030  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00   ................
      00000040  00 00 00 00 00 00 00 00                           ........

Payload (72 bytes):

      00000000  05 00 00 00 01 00 00 00 48 00 00 00 00 00 00 00   ........H.......
      00000010  00 00 00 80 00 00 00 00 10 05 0f 00 00 00 00 00   ................
      00000020  00 00 00 00 00 00 00 00 00 20 14 00 02 01 00 03   ......... ......
      00000030  0f 00 91 00 00 00 00 00 ef be ad de 00 00 00 00   ................
      00000040  ef be ad de 00 00 00 00                           ........

Error injected.

[    9.358364] {1}[Hardware Error]: Hardware error from APEI Generic Hardware Error Source: 1
[    9.359027] {1}[Hardware Error]: event severity: recoverable
[    9.359586] {1}[Hardware Error]:  Error 0, type: recoverable
[    9.360124] {1}[Hardware Error]:   section_type: ARM processor error
[    9.360561] {1}[Hardware Error]:   MIDR: 0x00000000000f0510
[    9.361160] {1}[Hardware Error]:   Multiprocessor Affinity Register (MPIDR): 0x0000000080000000
[    9.361643] {1}[Hardware Error]:   running state: 0x0
[    9.362142] {1}[Hardware Error]:   Power State Coordination Interface state: 0
[    9.362682] {1}[Hardware Error]:   Error info structure 0:
[    9.363030] {1}[Hardware Error]:   num errors: 2
[    9.363656] {1}[Hardware Error]:    error_type: 0x02: cache error
[    9.364163] {1}[Hardware Error]:    error_info: 0x000000000091000f
[    9.364834] {1}[Hardware Error]:     transaction type: Data Access
[    9.365599] {1}[Hardware Error]:     cache error, operation type: Data write
[    9.366441] {1}[Hardware Error]:     cache level: 2
[    9.367005] {1}[Hardware Error]:     processor context not corrupted
[    9.367753] {1}[Hardware Error]:    physical fault address: 0x00000000deadbeef
[    9.374267] Memory failure: 0xdeadb: recovery action for free buddy page: Recovered

Such script currently supports arm processor error CPER, but can easily be extended to other GHES notification types.

Jonathan Cameron (1):
acpi/ghes: Add support for GED error device

Mauro Carvalho Chehab (12):
acpi/generic_event_device: add an APEI error device arm/virt: Wire up a GED error device for ACPI / GHES qapi/acpi-hest: add an interface to do generic CPER error injection acpi/ghes: rework the logic to handle HEST source ID acpi/ghes: add support for generic error injection via QAPI acpi/ghes: cleanup the memory error code logic
docs: acpi_hest_ghes: fix documentation for CPER size scripts/ghes_inject: add a script to generate GHES error inject target/arm: add an experimental mpidr arm cpu property object scripts/arm_processor_error.py: retrieve mpidr if not filled acpi/ghes: cleanup generic error data logic
acpi/ghes: check if the BIOS pointers for HEST are correct

 MAINTAINERS                            |  10 +
 docs/specs/acpi_hest_ghes.rst          |   6 +-
 hw/acpi/Kconfig                        |   5 +
 hw/acpi/aml-build.c                    |  10 +
 hw/acpi/generic_event_device.c         |   8 +
 hw/acpi/ghes-stub.c                    |   3 +-
 hw/acpi/ghes.c                         | 362 ++++++++-----
 hw/acpi/ghes_cper.c                    |  33 ++
 hw/acpi/ghes_cper_stub.c               |  19 +
 hw/acpi/meson.build                    |   2 +
 hw/arm/Kconfig                         |   5 +
 hw/arm/virt-acpi-build.c               |   6 +-
 hw/arm/virt.c                          |  12 +-
 include/hw/acpi/acpi_dev_interface.h   |   1 +
 include/hw/acpi/aml-build.h            |   2 +
 include/hw/acpi/generic_event_device.h |   1 +
 include/hw/acpi/ghes.h                 |  24 +-
 include/hw/arm/virt.h                  |   1 +
 qapi/acpi-hest.json                    |  36 ++
 qapi/meson.build                       |   1 +
 qapi/qapi-schema.json                  |   1 +
 scripts/arm_processor_error.py         | 388 ++++++++++++++
 scripts/ghes_inject.py                 |  51 ++
 scripts/qmp_helper.py                  | 702 +++++++++++++++++++++++++
 target/arm/cpu.c                       |   1 +
 target/arm/cpu.h                       |   1 +
 target/arm/helper.c                    |  10 +-
 target/arm/kvm.c                       |   2 +-
 28 files changed, 1551 insertions(+), 152 deletions(-)
 create mode 100644 hw/acpi/ghes_cper.c
 create mode 100644 hw/acpi/ghes_cper_stub.c
 create mode 100644 qapi/acpi-hest.json
 create mode 100644 scripts/arm_processor_error.py
 create mode 100755 scripts/ghes_inject.py
 create mode 100644 scripts/qmp_helper.py
Signature · 1 line
2.46.0
#1 lore raw

PATCH v8 1/13 acpi/generic_event_device: add an APEI error device

Adds a generic error device to handle generic hardware error events as specified at ACPI 6.5 specification at 18.3.2.7.2:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#event-notification-for-generic-error-sources
using HID PNP0C33.

The PNP0C33 device is used to report hardware errors to the guest via ACPI APEI Generic Hardware Error Source (GHES).

---
 hw/acpi/aml-build.c                    | 10 ++++++++++
 hw/acpi/generic_event_device.c         |  8 ++++++++
 include/hw/acpi/acpi_dev_interface.h   |  1 +
 include/hw/acpi/aml-build.h            |  2 ++
 include/hw/acpi/generic_event_device.h |  1 +
 5 files changed, 22 insertions(+)
hw/acpi/aml-build.c +10 -0
index 6d4517cfbe3d..cb167523859f 100644
@@ -2520,3 +2520,13 @@ Aml *aml_i2c_serial_bus_device(uint16_t address, const char *resource_source)
 
     return var;
 }
+
+/* ACPI 5.0: 18.3.2.6.2 Event Notification For Generic Error Sources */
+Aml *aml_error_device(void)
+{
+    Aml *dev = aml_device(ACPI_APEI_ERROR_DEVICE);
+    aml_append(dev, aml_name_decl("_HID", aml_string("PNP0C33")));
+    aml_append(dev, aml_name_decl("_UID", aml_int(0)));
+
+    return dev;
+}
hw/acpi/generic_event_device.c +8 -0
index 15b4c3ebbf24..b4c83a089a02 100644
@@ -26,6 +26,7 @@ static const uint32_t ged_supported_events[] = {
     ACPI_GED_PWR_DOWN_EVT,
     ACPI_GED_NVDIMM_HOTPLUG_EVT,
     ACPI_GED_CPU_HOTPLUG_EVT,
+    ACPI_GED_ERROR_EVT,
 };
 
 /*
@@ -116,6 +117,11 @@ void build_ged_aml(Aml *table, const char *name, HotplugHandler *hotplug_dev,
                            aml_notify(aml_name(ACPI_POWER_BUTTON_DEVICE),
                                       aml_int(0x80)));
                 break;
+            case ACPI_GED_ERROR_EVT:
+                aml_append(if_ctx,
+                           aml_notify(aml_name(ACPI_APEI_ERROR_DEVICE),
+                                      aml_int(0x80)));
+                break;
             case ACPI_GED_NVDIMM_HOTPLUG_EVT:
                 aml_append(if_ctx,
                            aml_notify(aml_name("\\_SB.NVDR"),
@@ -295,6 +301,8 @@ static void acpi_ged_send_event(AcpiDeviceIf *adev, AcpiEventStatusBits ev)
         sel = ACPI_GED_MEM_HOTPLUG_EVT;
     } else if (ev & ACPI_POWER_DOWN_STATUS) {
         sel = ACPI_GED_PWR_DOWN_EVT;
+    } else if (ev & ACPI_GENERIC_ERROR) {
+        sel = ACPI_GED_ERROR_EVT;
     } else if (ev & ACPI_NVDIMM_HOTPLUG_STATUS) {
         sel = ACPI_GED_NVDIMM_HOTPLUG_EVT;
     } else if (ev & ACPI_CPU_HOTPLUG_STATUS) {
include/hw/acpi/acpi_dev_interface.h +1 -0
index 68d9d15f50aa..8294f8f0ccca 100644
@@ -13,6 +13,7 @@ typedef enum {
     ACPI_NVDIMM_HOTPLUG_STATUS = 16,
     ACPI_VMGENID_CHANGE_STATUS = 32,
     ACPI_POWER_DOWN_STATUS = 64,
+    ACPI_GENERIC_ERROR = 128,
 } AcpiEventStatusBits;
 
 #define TYPE_ACPI_DEVICE_IF "acpi-device-interface"
include/hw/acpi/aml-build.h +2 -0
index a3784155cb33..44d1a6af0c69 100644
@@ -252,6 +252,7 @@ struct CrsRangeSet {
 /* Consumer/Producer */
 #define AML_SERIAL_BUS_FLAG_CONSUME_ONLY        (1 << 1)
 
+#define ACPI_APEI_ERROR_DEVICE   "GEDD"
 /**
  * init_aml_allocator:
  *
@@ -382,6 +383,7 @@ Aml *aml_dma(AmlDmaType typ, AmlDmaBusMaster bm, AmlTransferSize sz,
              uint8_t channel);
 Aml *aml_sleep(uint64_t msec);
 Aml *aml_i2c_serial_bus_device(uint16_t address, const char *resource_source);
+Aml *aml_error_device(void);
 
 /* Block AML object primitives */
 Aml *aml_scope(const char *name_format, ...) G_GNUC_PRINTF(1, 2);
include/hw/acpi/generic_event_device.h +1 -0
index 40af3550b56d..9ace8fe70328 100644
@@ -98,6 +98,7 @@ OBJECT_DECLARE_SIMPLE_TYPE(AcpiGedState, ACPI_GED)
 #define ACPI_GED_PWR_DOWN_EVT      0x2
 #define ACPI_GED_NVDIMM_HOTPLUG_EVT 0x4
 #define ACPI_GED_CPU_HOTPLUG_EVT    0x8
+#define ACPI_GED_ERROR_EVT          0x10
 
 typedef struct GEDState {
     MemoryRegion evt;
Signature · 1 line
2.46.0
#2 lore raw

Re: [PATCH v8 01/13] acpi/generic_event_device: add an APEI error device

Quoted 3 lines · On Fri, 16 Aug 2024 09:37:33 +0200 Mauro Carvalho Chehab…

Adds a generic error device to handle generic hardware error events as specified at ACPI 6.5 specification at 18.3.2.7.2:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#event-notification-for-generic-error-sources

make it match comment in the code for consistency
(i.e reference to 5.0b spec incl chapter/name)

Quoted 23 lines

using HID PNP0C33.

The PNP0C33 device is used to report hardware errors to the guest via ACPI APEI Generic Hardware Error Source (GHES).

---
 hw/acpi/aml-build.c                    | 10 ++++++++++
 hw/acpi/generic_event_device.c         |  8 ++++++++
 include/hw/acpi/acpi_dev_interface.h   |  1 +
 include/hw/acpi/aml-build.h            |  2 ++
 include/hw/acpi/generic_event_device.h |  1 +
 5 files changed, 22 insertions(+)
hw/acpi/aml-build.c +0 -0
index 6d4517cfbe3d..cb167523859f 100644

hw/acpi/aml-build.c Aml *aml_i2c_serial_bus_device(uint16_t address, const char *resource_source) · line 2524

Quoted 5 lines
hw/acpi/aml-build.c +2 -0
@@ -2520,3 +2520,13 @@ Aml *aml_i2c_serial_bus_device(uint16_t address, const char *resource_source)
 
     return var;
 }
+
+/* ACPI 5.0: 18.3.2.6.2 Event Notification For Generic Error Sources */

should be ACPI 5.0b

Quoted 85 lines
hw/acpi/aml-build.c +8 -0
+Aml *aml_error_device(void)
+{
+    Aml *dev = aml_device(ACPI_APEI_ERROR_DEVICE);
+    aml_append(dev, aml_name_decl("_HID", aml_string("PNP0C33")));
+    aml_append(dev, aml_name_decl("_UID", aml_int(0)));
+
+    return dev;
+}
hw/acpi/generic_event_device.c +8 -0
index 15b4c3ebbf24..b4c83a089a02 100644
@@ -26,6 +26,7 @@ static const uint32_t ged_supported_events[] = {
     ACPI_GED_PWR_DOWN_EVT,
     ACPI_GED_NVDIMM_HOTPLUG_EVT,
     ACPI_GED_CPU_HOTPLUG_EVT,
+    ACPI_GED_ERROR_EVT,
 };
 
 /*
@@ -116,6 +117,11 @@ void build_ged_aml(Aml *table, const char *name, HotplugHandler *hotplug_dev,
                            aml_notify(aml_name(ACPI_POWER_BUTTON_DEVICE),
                                       aml_int(0x80)));
                 break;
+            case ACPI_GED_ERROR_EVT:
+                aml_append(if_ctx,
+                           aml_notify(aml_name(ACPI_APEI_ERROR_DEVICE),
+                                      aml_int(0x80)));
+                break;
             case ACPI_GED_NVDIMM_HOTPLUG_EVT:
                 aml_append(if_ctx,
                            aml_notify(aml_name("\\_SB.NVDR"),
@@ -295,6 +301,8 @@ static void acpi_ged_send_event(AcpiDeviceIf *adev, AcpiEventStatusBits ev)
         sel = ACPI_GED_MEM_HOTPLUG_EVT;
     } else if (ev & ACPI_POWER_DOWN_STATUS) {
         sel = ACPI_GED_PWR_DOWN_EVT;
+    } else if (ev & ACPI_GENERIC_ERROR) {
+        sel = ACPI_GED_ERROR_EVT;
     } else if (ev & ACPI_NVDIMM_HOTPLUG_STATUS) {
         sel = ACPI_GED_NVDIMM_HOTPLUG_EVT;
     } else if (ev & ACPI_CPU_HOTPLUG_STATUS) {
include/hw/acpi/acpi_dev_interface.h +1 -0
index 68d9d15f50aa..8294f8f0ccca 100644
@@ -13,6 +13,7 @@ typedef enum {
     ACPI_NVDIMM_HOTPLUG_STATUS = 16,
     ACPI_VMGENID_CHANGE_STATUS = 32,
     ACPI_POWER_DOWN_STATUS = 64,
+    ACPI_GENERIC_ERROR = 128,
 } AcpiEventStatusBits;
 
 #define TYPE_ACPI_DEVICE_IF "acpi-device-interface"
include/hw/acpi/aml-build.h +2 -0
index a3784155cb33..44d1a6af0c69 100644
@@ -252,6 +252,7 @@ struct CrsRangeSet {
 /* Consumer/Producer */
 #define AML_SERIAL_BUS_FLAG_CONSUME_ONLY        (1 << 1)
 
+#define ACPI_APEI_ERROR_DEVICE   "GEDD"
 /**
  * init_aml_allocator:
  *
@@ -382,6 +383,7 @@ Aml *aml_dma(AmlDmaType typ, AmlDmaBusMaster bm, AmlTransferSize sz,
              uint8_t channel);
 Aml *aml_sleep(uint64_t msec);
 Aml *aml_i2c_serial_bus_device(uint16_t address, const char *resource_source);
+Aml *aml_error_device(void);
 
 /* Block AML object primitives */
 Aml *aml_scope(const char *name_format, ...) G_GNUC_PRINTF(1, 2);
include/hw/acpi/generic_event_device.h +1 -0
index 40af3550b56d..9ace8fe70328 100644
@@ -98,6 +98,7 @@ OBJECT_DECLARE_SIMPLE_TYPE(AcpiGedState, ACPI_GED)
 #define ACPI_GED_PWR_DOWN_EVT      0x2
 #define ACPI_GED_NVDIMM_HOTPLUG_EVT 0x4
 #define ACPI_GED_CPU_HOTPLUG_EVT    0x8
+#define ACPI_GED_ERROR_EVT          0x10
 
 typedef struct GEDState {
     MemoryRegion evt;
#3 lore raw

PATCH v8 2/13 arm/virt: Wire up a GED error device for ACPI / GHES

In reply to Mauro Carvalho Chehab

Adds support to ARM virtualization to allow handling generic error ACPI Event via GED & error source device.

It is aligned with Linux Kernel patch:
https://lore.kernel.org/lkml/[email protected]/

---
 hw/acpi/ghes.c           |  3 +++
 hw/arm/virt-acpi-build.c |  1 +
 hw/arm/virt.c            | 12 +++++++++++-
 include/hw/acpi/ghes.h   |  3 +++
 include/hw/arm/virt.h    |  1 +
 5 files changed, 19 insertions(+), 1 deletion(-)
hw/acpi/ghes.c +3 -0
index e9511d9b8f71..13b105c5d02d 100644
@@ -444,6 +444,9 @@ int acpi_ghes_record_errors(uint8_t source_id, uint64_t physical_address)
     return ret;
 }
 
+NotifierList acpi_generic_error_notifiers =
+    NOTIFIER_LIST_INITIALIZER(error_device_notifiers);
+
 bool acpi_ghes_present(void)
 {
     AcpiGedState *acpi_ged_state;
hw/arm/virt-acpi-build.c +1 -0
index f76fb117adff..1769467d23b2 100644
@@ -858,6 +858,7 @@ build_dsdt(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms)
     }
 
     acpi_dsdt_add_power_button(scope);
+    aml_append(scope, aml_error_device());
 #ifdef CONFIG_TPM
     acpi_dsdt_add_tpm(scope, vms);
 #endif
hw/arm/virt.c +11 -1
index 687fe0bb8bc9..22448e5c5b73 100644
@@ -677,7 +677,7 @@ static inline DeviceState *create_acpi_ged(VirtMachineState *vms)
     DeviceState *dev;
     MachineState *ms = MACHINE(vms);
     int irq = vms->irqmap[VIRT_ACPI_GED];
-    uint32_t event = ACPI_GED_PWR_DOWN_EVT;
+    uint32_t event = ACPI_GED_PWR_DOWN_EVT | ACPI_GED_ERROR_EVT;
 
     if (ms->ram_slots) {
         event |= ACPI_GED_MEM_HOTPLUG_EVT;
@@ -1009,6 +1009,13 @@ static void virt_powerdown_req(Notifier *n, void *opaque)
     }
 }
 
+static void virt_generic_error_req(Notifier *n, void *opaque)
+{
+    VirtMachineState *s = container_of(n, VirtMachineState, generic_error_notifier);
+
+    acpi_send_event(s->acpi_dev, ACPI_GENERIC_ERROR);
+}
+
 static void create_gpio_keys(char *fdt, DeviceState *pl061_dev,
                              uint32_t phandle)
 {
@@ -2385,6 +2392,9 @@ static void machvirt_init(MachineState *machine)
 
     if (has_ged && aarch64 && firmware_loaded && virt_is_acpi_enabled(vms)) {
         vms->acpi_dev = create_acpi_ged(vms);
+        vms->generic_error_notifier.notify = virt_generic_error_req;
+        notifier_list_add(&acpi_generic_error_notifiers,
+                          &vms->generic_error_notifier);
     } else {
         create_gpio_devices(vms, VIRT_GPIO, sysmem);
     }
include/hw/acpi/ghes.h +3 -0
index 674f6958e905..fb80897e7eac 100644
@@ -23,6 +23,9 @@
 #define ACPI_GHES_H
 
 #include "hw/acpi/bios-linker-loader.h"
+#include "qemu/notify.h"
+
+extern NotifierList acpi_generic_error_notifiers;
 
 /*
  * Values for Hardware Error Notification Type field
include/hw/arm/virt.h +1 -0
index a4d937ed45ac..ad9f6e94dcc5 100644
@@ -175,6 +175,7 @@ struct VirtMachineState {
     DeviceState *gic;
     DeviceState *acpi_dev;
     Notifier powerdown_notifier;
+    Notifier generic_error_notifier;
     PCIBus *bus;
     char *oem_id;
     char *oem_table_id;
Signature · 1 line
2.46.0
#4 lore raw

PATCH v8 3/13 acpi/ghes: Add support for GED error device

In reply to Mauro Carvalho Chehab

From: Jonathan Cameron <[email protected]>

As a GED error device is now defined, add another type of notification.

Add error notification to GHES v2 using a GED error device GED triggered via interrupt.

[mchehab: do some cleanups at ACPI_HEST_SRC_ID_* checks and

 rename HEST event to better identify GED interrupt OSPM]
---
 hw/acpi/ghes.c         | 11 +++++++++--
 include/hw/acpi/ghes.h |  3 ++-
 2 files changed, 11 insertions(+), 3 deletions(-)
hw/acpi/ghes.c +9 -2
index 13b105c5d02d..df59fd35568c 100644
@@ -34,8 +34,8 @@
 /* The max size in bytes for one error block */
 #define ACPI_GHES_MAX_RAW_DATA_LENGTH   (1 * KiB)
 
-/* Now only support ARMv8 SEA notification type error source */
-#define ACPI_GHES_ERROR_SOURCE_COUNT        1
+/* Support ARMv8 SEA notification type error source and GPIO interrupt. */
+#define ACPI_GHES_ERROR_SOURCE_COUNT        2
 
 /* Generic Hardware Error Source version 2 */
 #define ACPI_GHES_SOURCE_GENERIC_ERROR_V2   10
@@ -290,6 +290,9 @@ void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
 static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
 {
     uint64_t address_offset;
+
+    assert(source_id < ACPI_HEST_SRC_ID_RESERVED);
+
     /*
      * Type:
      * Generic Hardware Error Source version 2(GHESv2 - Type 10)
@@ -327,6 +330,9 @@ static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
          */
         build_ghes_hw_error_notification(table_data, ACPI_GHES_NOTIFY_SEA);
         break;
+    case ACPI_HEST_SRC_ID_GED:
+        build_ghes_hw_error_notification(table_data, ACPI_GHES_NOTIFY_GPIO);
+        break;
     default:
         error_report("Not support this error source");
         abort();
@@ -370,6 +376,7 @@ void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
     /* Error Source Count */
     build_append_int_noprefix(table_data, ACPI_GHES_ERROR_SOURCE_COUNT, 4);
     build_ghes_v2(table_data, ACPI_HEST_SRC_ID_SEA, linker);
+    build_ghes_v2(table_data, ACPI_HEST_SRC_ID_GED, linker);
 
     acpi_table_end(linker, &table);
 }
include/hw/acpi/ghes.h +2 -1
index fb80897e7eac..419a97d5cbd9 100644
@@ -59,9 +59,10 @@ enum AcpiGhesNotifyType {
     ACPI_GHES_NOTIFY_RESERVED = 12
 };
 
+/* Those are used as table indexes when building GHES tables */
 enum {
     ACPI_HEST_SRC_ID_SEA = 0,
-    /* future ids go here */
+    ACPI_HEST_SRC_ID_GED,
     ACPI_HEST_SRC_ID_RESERVED,
 };
 
Signature · 1 line
2.46.0
#5 lore raw

Re: [PATCH v8 03/13] acpi/ghes: Add support for GED error device

Quoted 7 lines · On Fri, 16 Aug 2024 09:37:35 +0200 Mauro Carvalho Chehab…

From: Jonathan Cameron <[email protected]>

As a GED error device is now defined, add another type of notification.

Add error notification to GHES v2 using
a GED error device GED triggered via interrupt.

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This is hard to parse, perhaps update so it would be more clear what does what

Quoted 8 lines

[mchehab: do some cleanups at ACPI_HEST_SRC_ID_* checks and

 rename HEST event to better identify GED interrupt OSPM]

---

in addition to change log in cover letter,
I'd suggest to keep per patch change log as well (after ---), it helps reviewer to notice intended changes.

[...]

hw/acpi/ghes.c static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker) · line 334

Quoted 2 lines
hw/acpi/ghes.c +2 -0
+    case ACPI_HEST_SRC_ID_GED:
+        build_ghes_hw_error_notification(table_data, ACPI_GHES_NOTIFY_GPIO);

While GPIO works for arm, it's not the case for other machines. I recall a suggestion to use ACPI_GHES_NOTIFY_EXTERNAL instead of GPIO one, but that got lost somewhere...

Quoted 28 lines
hw/acpi/ghes.c +2 -0
+        break;
     default:
         error_report("Not support this error source");
         abort();
@@ -370,6 +376,7 @@ void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
     /* Error Source Count */
     build_append_int_noprefix(table_data, ACPI_GHES_ERROR_SOURCE_COUNT, 4);
     build_ghes_v2(table_data, ACPI_HEST_SRC_ID_SEA, linker);
+    build_ghes_v2(table_data, ACPI_HEST_SRC_ID_GED, linker);
 
     acpi_table_end(linker, &table);
 }
include/hw/acpi/ghes.h +2 -1
index fb80897e7eac..419a97d5cbd9 100644
@@ -59,9 +59,10 @@ enum AcpiGhesNotifyType {
     ACPI_GHES_NOTIFY_RESERVED = 12
 };
 
+/* Those are used as table indexes when building GHES tables */
 enum {
     ACPI_HEST_SRC_ID_SEA = 0,
-    /* future ids go here */
+    ACPI_HEST_SRC_ID_GED,
     ACPI_HEST_SRC_ID_RESERVED,
 };
 
#6 lore raw

Re: [PATCH v8 03/13] acpi/ghes: Add support for GED error device

Em Mon, 19 Aug 2024 13:43:04 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 28 lines
Quoted 7 lines · On Fri, 16 Aug 2024 09:37:35 +0200 Mauro Carvalho Chehab…

From: Jonathan Cameron <[email protected]>

As a GED error device is now defined, add another type of notification.

Add error notification to GHES v2 using
a GED error device GED triggered via interrupt.

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
This is hard to parse, perhaps update so it would be more clear what does what

Quoted 8 lines

[mchehab: do some cleanups at ACPI_HEST_SRC_ID_* checks and

 rename HEST event to better identify GED interrupt OSPM]

---

in addition to change log in cover letter,
I'd suggest to keep per patch change log as well (after ---), it helps reviewer to notice intended changes.

Quoted 6 lines

[...]

hw/acpi/ghes.c static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker) · line 334

Quoted 2 lines
hw/acpi/ghes.c +2 -0
+    case ACPI_HEST_SRC_ID_GED:
+        build_ghes_hw_error_notification(table_data, ACPI_GHES_NOTIFY_GPIO);  

While GPIO works for arm, it's not the case for other machines. I recall a suggestion to use ACPI_GHES_NOTIFY_EXTERNAL instead of GPIO one, but that got lost somewhere...

True, but the same also applies to SEA, which is ARMv8+. After having everything in place, I confined the source ID into this code inside ghes.c:

	enum AcpiHestSourceId {
	    ACPI_HEST_SRC_ID_SEA,
	    ACPI_HEST_SRC_ID_GED,

	    /* Shall be the last one */
	    ACPI_HEST_SRC_ID_COUNT
	} AcpiHestSourceId;

	static bool ghes_notify_to_source_id(enum AcpiGhesNotifyType notify,
	                                     enum AcpiHestSourceId *source_id)
	{
	    switch (notify) {
	    case ACPI_GHES_NOTIFY_SEA:             /* ARMv8 */
	        *source_id = ACPI_HEST_SRC_ID_SEA;
	        return false;
	    case ACPI_GHES_NOTIFY_GPIO:
	        *source_id = ACPI_HEST_SRC_ID_GED;
	        return false;
	    default:
	        /* Unsupported notification types */
	        return true;
	    }
	}

The only place where the source ID number is used is at ghes_notify_to_source_id() - still we use ACPI_HEST_SRC_ID_COUNT on other places to initialize and fill in the HEST table and its error source structures.

On other words, the source ID field is filled from the notification types as defined at include/hw/acpi/ghes.h:

ACPI_GHES_NOTIFY_POLLED = 0,
ACPI_GHES_NOTIFY_EXTERNAL = 1,
ACPI_GHES_NOTIFY_LOCAL = 2,
ACPI_GHES_NOTIFY_SCI = 3,
ACPI_GHES_NOTIFY_NMI = 4,
ACPI_GHES_NOTIFY_CMCI = 5,
ACPI_GHES_NOTIFY_MCE = 6,
ACPI_GHES_NOTIFY_GPIO = 7,
ACPI_GHES_NOTIFY_SEA = 8,
ACPI_GHES_NOTIFY_SEI = 9,
ACPI_GHES_NOTIFY_GSIV = 10,
ACPI_GHES_NOTIFY_SDEI = 11,

(please notice that ACPI already defines "EXTERNAL" as being something else)

Now, if we want to add support for x86, we could either add some ifdefs inside ghes.c, e. g. something like:

	enum AcpiHestSourceId {
	#ifdef TARGET_ARM
	    ACPI_HEST_SRC_ID_SEA,
	    ACPI_HEST_SRC_ID_GED,
	#endif
	#ifdef TARGET_I386
	   ACPI_HEST_SRC_ID_MCE,
        #endif

	    /* Shall be the last one */
	    ACPI_HEST_SRC_ID_COUNT
	} AcpiHestSourceId;

and something similar at ghes_notify_to_source_id():

	static bool ghes_notify_to_source_id(enum AcpiGhesNotifyType notify,
	                                     enum AcpiHestSourceId *source_id)
	{
	    switch (notify) {
	#ifdef TARGET_ARM
	    case ACPI_GHES_NOTIFY_SEA:             /* ARMv8 */
	        *source_id = ACPI_HEST_SRC_ID_SEA;
	        return false;
	    case ACPI_GHES_NOTIFY_GPIO:
	        *source_id = ACPI_HEST_SRC_ID_GED;
	        return false;
	#endif
	#ifdef TARGET_I386
	    case ACPI_GHES_NOTIFY_MCE:
	        *source_id = ACPI_HEST_SRC_ID_MCE;
	        return false;
	#endif
	    default:
	        /* Unsupported notification types */
	        return true;
	    }
	}

An alternative would be to move source id/notification code out, placing them at hw/arm, hw/i386, but a more complex binding logic will be needed.

If we're willing to do something like that, I would prefer to not do such redesign now. Better to do such change when we'll be ready to add some notification support that works on x86 (MCE? SCI? NMI?).

Regards,
Mauro

#7 lore raw

PATCH v8 4/13 qapi/acpi-hest: add an interface to do generic CPER error injection

In reply to Mauro Carvalho Chehab

Creates a QMP command to be used for generic ACPI APEI hardware error injection (HEST) via GHESv2.

The actual GHES code will be added at the followup patch.

---
 MAINTAINERS              |  7 +++++++
 hw/acpi/Kconfig          |  5 +++++
 hw/acpi/ghes_cper.c      | 33 +++++++++++++++++++++++++++++++++
 hw/acpi/ghes_cper_stub.c | 19 +++++++++++++++++++
 hw/acpi/meson.build      |  2 ++
 hw/arm/Kconfig           |  5 +++++
 include/hw/acpi/ghes.h   |  3 +++
 qapi/acpi-hest.json      | 36 ++++++++++++++++++++++++++++++++++++
 qapi/meson.build         |  1 +
 qapi/qapi-schema.json    |  1 +
 10 files changed, 112 insertions(+)
 create mode 100644 hw/acpi/ghes_cper.c
 create mode 100644 hw/acpi/ghes_cper_stub.c
 create mode 100644 qapi/acpi-hest.json
MAINTAINERS +7 -0
index 3584d6a6c6da..1d8091818899 100644
@@ -2077,6 +2077,13 @@ F: hw/acpi/ghes.c
 F: include/hw/acpi/ghes.h
 F: docs/specs/acpi_hest_ghes.rst
 
+ACPI/HEST/GHES/ARM processor CPER
+R: Mauro Carvalho Chehab <[email protected]>
+S: Maintained
+F: hw/arm/ghes_cper.c
+F: hw/acpi/ghes_cper_stub.c
+F: qapi/acpi-hest.json
+
 ppc4xx
 L: [email protected]
 S: Orphan
hw/acpi/Kconfig +5 -0
index e07d3204eb36..73ffbb82c150 100644
@@ -51,6 +51,11 @@ config ACPI_APEI
     bool
     depends on ACPI
 
+config GHES_CPER
+    bool
+    depends on ACPI_APEI
+    default y
+
 config ACPI_PCI
     bool
     depends on ACPI && PCI
hw/acpi/ghes_cper.c new file +33 -0
new file mode 100644
index 000000000000..92ca84d738de
@@ -0,0 +1,33 @@
+/*
+ * CPER payload parser for error injection
+ *
+ * Copyright(C) 2024 Huawei LTD.
+ *
+ * This code is licensed under the GPL version 2 or later. See the
+ * COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+
+#include "qemu/base64.h"
+#include "qemu/error-report.h"
+#include "qemu/uuid.h"
+#include "qapi/qapi-commands-acpi-hest.h"
+#include "hw/acpi/ghes.h"
+
+void qmp_ghes_cper(const char *qmp_cper,
+                   Error **errp)
+{
+
+    uint8_t *cper;
+    size_t  len;
+
+    cper = qbase64_decode(qmp_cper, -1, &len, errp);
+    if (!cper) {
+        error_setg(errp, "missing GHES CPER payload");
+        return;
+    }
+
+    /* TODO: call a function at ghes */
+}
hw/acpi/ghes_cper_stub.c new file +19 -0
new file mode 100644
index 000000000000..36138c462ac9
@@ -0,0 +1,19 @@
+/*
+ * Stub interface for CPER payload parser for error injection
+ *
+ * Copyright(C) 2024 Huawei LTD.
+ *
+ * This code is licensed under the GPL version 2 or later. See the
+ * COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "qapi/error.h"
+#include "qapi/qapi-commands-acpi-hest.h"
+#include "hw/acpi/ghes.h"
+
+void qmp_ghes_cper(const char *cper, Error **errp)
+{
+    error_setg(errp, "GHES QMP error inject is not compiled in");
+}
hw/acpi/meson.build +2 -0
index fa5c07db9068..6cbf430eb66d 100644
@@ -34,4 +34,6 @@ endif
 system_ss.add(when: 'CONFIG_ACPI', if_false: files('acpi-stub.c', 'aml-build-stub.c', 'ghes-stub.c', 'acpi_interface.c'))
 system_ss.add(when: 'CONFIG_ACPI_PCI_BRIDGE', if_false: files('pci-bridge-stub.c'))
 system_ss.add_all(when: 'CONFIG_ACPI', if_true: acpi_ss)
+system_ss.add(when: 'CONFIG_GHES_CPER', if_true: files('ghes_cper.c'))
+system_ss.add(when: 'CONFIG_GHES_CPER', if_false: files('ghes_cper_stub.c'))
 system_ss.add(files('acpi-qmp-cmds.c'))
hw/arm/Kconfig +5 -0
index 1ad60da7aa2d..bed6ba27d715 100644
@@ -712,3 +712,8 @@ config ARMSSE
     select UNIMP
     select SSE_COUNTER
     select SSE_TIMER
+
+config GHES_CPER
+    bool
+    depends on ARM
+    default y if AARCH64
include/hw/acpi/ghes.h +3 -0
index 419a97d5cbd9..b977d65564ba 100644
@@ -23,6 +23,7 @@
 #define ACPI_GHES_H
 
 #include "hw/acpi/bios-linker-loader.h"
+#include "qapi/error.h"
 #include "qemu/notify.h"
 
 extern NotifierList acpi_generic_error_notifiers;
@@ -77,6 +78,8 @@ void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
 void acpi_ghes_add_fw_cfg(AcpiGhesState *vms, FWCfgState *s,
                           GArray *hardware_errors);
 int acpi_ghes_record_errors(uint8_t notify, uint64_t error_physical_addr);
+void ghes_record_cper_errors(const void *cper, size_t len,
+                             enum AcpiGhesNotifyType notify, Error **errp);
 
 /**
  * acpi_ghes_present: Report whether ACPI GHES table is present
qapi/acpi-hest.json new file +36 -0
new file mode 100644
index 000000000000..91296755d285
@@ -0,0 +1,36 @@
+# -*- Mode: Python -*-
+# vim: filetype=python
+
+##
+# = GHESv2 CPER Error Injection
+#
+# Defined since ACPI Specification 6.2,
+# section 18.3.2.8 Generic Hardware Error Source version 2. See:
+#
+# https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#generic-hardware-error-source-version-2-ghesv2-type-10
+##
+
+
+##
+# @ghes-cper:
+#
+# Inject a CPER error data to be filled according to ACPI 6.2
+# spec via GHESv2.
+#
+# @cper: contains a base64 encoded string with raw data for a single CPER
+#     record with Generic Error Status Block, Generic Error Data Entry and
+#     generic error data payload, as described at
+#     https://uefi.org/specs/UEFI/2.10/Apx_N_Common_Platform_Error_Record.html#format
+#
+# Features:
+#
+# @unstable: This command is experimental.
+#
+# Since: 9.2
+##
+{ 'command': 'ghes-cper',
+  'data': {
+    'cper': 'str'
+  },
+  'features': [ 'unstable' ]
+}
qapi/meson.build +1 -0
index e7bc54e5d047..35cea6147262 100644
@@ -59,6 +59,7 @@ qapi_all_modules = [
 if have_system
   qapi_all_modules += [
     'acpi',
+    'acpi-hest',
     'audio',
     'cryptodev',
     'qdev',
qapi/qapi-schema.json +1 -0
index b1581988e4eb..baf19ab73afe 100644
@@ -75,6 +75,7 @@
 { 'include': 'misc-target.json' }
 { 'include': 'audio.json' }
 { 'include': 'acpi.json' }
+{ 'include': 'acpi-hest.json' }
 { 'include': 'pci.json' }
 { 'include': 'stats.json' }
 { 'include': 'virtio.json' }
Signature · 1 line
2.46.0
#8 lore raw

Re: [PATCH v8 04/13] qapi/acpi-hest: add an interface to do generic CPER error injection

Quoted 4 lines · On Fri, 16 Aug 2024 09:37:36 +0200 Mauro Carvalho Chehab…

Creates a QMP command to be used for generic ACPI APEI hardware error injection (HEST) via GHESv2.

The actual GHES code will be added at the followup patch.

modulo inconsistency in comments (see below), LGTM

Quoted 169 lines
---
 MAINTAINERS              |  7 +++++++
 hw/acpi/Kconfig          |  5 +++++
 hw/acpi/ghes_cper.c      | 33 +++++++++++++++++++++++++++++++++
 hw/acpi/ghes_cper_stub.c | 19 +++++++++++++++++++
 hw/acpi/meson.build      |  2 ++
 hw/arm/Kconfig           |  5 +++++
 include/hw/acpi/ghes.h   |  3 +++
 qapi/acpi-hest.json      | 36 ++++++++++++++++++++++++++++++++++++
 qapi/meson.build         |  1 +
 qapi/qapi-schema.json    |  1 +
 10 files changed, 112 insertions(+)
 create mode 100644 hw/acpi/ghes_cper.c
 create mode 100644 hw/acpi/ghes_cper_stub.c
 create mode 100644 qapi/acpi-hest.json
MAINTAINERS +7 -0
index 3584d6a6c6da..1d8091818899 100644
@@ -2077,6 +2077,13 @@ F: hw/acpi/ghes.c
 F: include/hw/acpi/ghes.h
 F: docs/specs/acpi_hest_ghes.rst
 
+ACPI/HEST/GHES/ARM processor CPER
+R: Mauro Carvalho Chehab <[email protected]>
+S: Maintained
+F: hw/arm/ghes_cper.c
+F: hw/acpi/ghes_cper_stub.c
+F: qapi/acpi-hest.json
+
 ppc4xx
 L: [email protected]
 S: Orphan
hw/acpi/Kconfig +5 -0
index e07d3204eb36..73ffbb82c150 100644
@@ -51,6 +51,11 @@ config ACPI_APEI
     bool
     depends on ACPI
 
+config GHES_CPER
+    bool
+    depends on ACPI_APEI
+    default y
+
 config ACPI_PCI
     bool
     depends on ACPI && PCI
hw/acpi/ghes_cper.c new file +33 -0
new file mode 100644
index 000000000000..92ca84d738de
@@ -0,0 +1,33 @@
+/*
+ * CPER payload parser for error injection
+ *
+ * Copyright(C) 2024 Huawei LTD.
+ *
+ * This code is licensed under the GPL version 2 or later. See the
+ * COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+
+#include "qemu/base64.h"
+#include "qemu/error-report.h"
+#include "qemu/uuid.h"
+#include "qapi/qapi-commands-acpi-hest.h"
+#include "hw/acpi/ghes.h"
+
+void qmp_ghes_cper(const char *qmp_cper,
+                   Error **errp)
+{
+
+    uint8_t *cper;
+    size_t  len;
+
+    cper = qbase64_decode(qmp_cper, -1, &len, errp);
+    if (!cper) {
+        error_setg(errp, "missing GHES CPER payload");
+        return;
+    }
+
+    /* TODO: call a function at ghes */
+}
hw/acpi/ghes_cper_stub.c new file +19 -0
new file mode 100644
index 000000000000..36138c462ac9
@@ -0,0 +1,19 @@
+/*
+ * Stub interface for CPER payload parser for error injection
+ *
+ * Copyright(C) 2024 Huawei LTD.
+ *
+ * This code is licensed under the GPL version 2 or later. See the
+ * COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+#include "qapi/error.h"
+#include "qapi/qapi-commands-acpi-hest.h"
+#include "hw/acpi/ghes.h"
+
+void qmp_ghes_cper(const char *cper, Error **errp)
+{
+    error_setg(errp, "GHES QMP error inject is not compiled in");
+}
hw/acpi/meson.build +2 -0
index fa5c07db9068..6cbf430eb66d 100644
@@ -34,4 +34,6 @@ endif
 system_ss.add(when: 'CONFIG_ACPI', if_false: files('acpi-stub.c', 'aml-build-stub.c', 'ghes-stub.c', 'acpi_interface.c'))
 system_ss.add(when: 'CONFIG_ACPI_PCI_BRIDGE', if_false: files('pci-bridge-stub.c'))
 system_ss.add_all(when: 'CONFIG_ACPI', if_true: acpi_ss)
+system_ss.add(when: 'CONFIG_GHES_CPER', if_true: files('ghes_cper.c'))
+system_ss.add(when: 'CONFIG_GHES_CPER', if_false: files('ghes_cper_stub.c'))
 system_ss.add(files('acpi-qmp-cmds.c'))
hw/arm/Kconfig +5 -0
index 1ad60da7aa2d..bed6ba27d715 100644
@@ -712,3 +712,8 @@ config ARMSSE
     select UNIMP
     select SSE_COUNTER
     select SSE_TIMER
+
+config GHES_CPER
+    bool
+    depends on ARM
+    default y if AARCH64
include/hw/acpi/ghes.h +3 -0
index 419a97d5cbd9..b977d65564ba 100644
@@ -23,6 +23,7 @@
 #define ACPI_GHES_H
 
 #include "hw/acpi/bios-linker-loader.h"
+#include "qapi/error.h"
 #include "qemu/notify.h"
 
 extern NotifierList acpi_generic_error_notifiers;
@@ -77,6 +78,8 @@ void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
 void acpi_ghes_add_fw_cfg(AcpiGhesState *vms, FWCfgState *s,
                           GArray *hardware_errors);
 int acpi_ghes_record_errors(uint8_t notify, uint64_t error_physical_addr);
+void ghes_record_cper_errors(const void *cper, size_t len,
+                             enum AcpiGhesNotifyType notify, Error **errp);
 
 /**
  * acpi_ghes_present: Report whether ACPI GHES table is present
qapi/acpi-hest.json new file +0 -0
new file mode 100644
index 000000000000..91296755d285

qapi/acpi-hest.json line 8

Quoted 8 lines
qapi/acpi-hest.json new file +8 -0
@@ -0,0 +1,36 @@
+# -*- Mode: Python -*-
+# vim: filetype=python
+
+##
+# = GHESv2 CPER Error Injection
+#
+# Defined since ACPI Specification 6.2,
+# section 18.3.2.8 Generic Hardware Error Source version 2. See:

earliest definition, I've found was 6.1, so this should be fixed up here

qapi/acpi-hest.json line 10

Quoted 2 lines
qapi/acpi-hest.json new file +2 -0
+#
+# https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#generic-hardware-error-source-version-2-ghesv2-type-10

make it consistent with above spec version

qapi/acpi-hest.json line 17

Quoted 7 lines
qapi/acpi-hest.json new file +7 -0
+##
+
+
+##
+# @ghes-cper:
+#
+# Inject a CPER error data to be filled according to ACPI 6.2

ditto

Quoted 43 lines
qapi/acpi-hest.json new file +19 -0
+# spec via GHESv2.
+#
+# @cper: contains a base64 encoded string with raw data for a single CPER
+#     record with Generic Error Status Block, Generic Error Data Entry and
+#     generic error data payload, as described at
+#     https://uefi.org/specs/UEFI/2.10/Apx_N_Common_Platform_Error_Record.html#format
+#
+# Features:
+#
+# @unstable: This command is experimental.
+#
+# Since: 9.2
+##
+{ 'command': 'ghes-cper',
+  'data': {
+    'cper': 'str'
+  },
+  'features': [ 'unstable' ]
+}
qapi/meson.build +1 -0
index e7bc54e5d047..35cea6147262 100644
@@ -59,6 +59,7 @@ qapi_all_modules = [
 if have_system
   qapi_all_modules += [
     'acpi',
+    'acpi-hest',
     'audio',
     'cryptodev',
     'qdev',
qapi/qapi-schema.json +1 -0
index b1581988e4eb..baf19ab73afe 100644
@@ -75,6 +75,7 @@
 { 'include': 'misc-target.json' }
 { 'include': 'audio.json' }
 { 'include': 'acpi.json' }
+{ 'include': 'acpi-hest.json' }
 { 'include': 'pci.json' }
 { 'include': 'stats.json' }
 { 'include': 'virtio.json' }
#9 lore raw

PATCH v8 5/13 acpi/ghes: rework the logic to handle HEST source ID

In reply to Mauro Carvalho Chehab

The current logic is based on a lot of duct tape, with offsets calculated based on one define with the number of source IDs and an enum.

Rewrite the logic in a way that it would be more resilient of code changes, by moving the source ID count to an enum and make the offset calculus more explicit.

Such change was inspired on a patch from Jonathan Cameron splitting the logic to get the CPER address on a separate function, as this will be needed to support generic error injection.

---
 hw/acpi/ghes-stub.c      |   3 +-
 hw/acpi/ghes.c           | 210 ++++++++++++++++++++++++---------------
 hw/arm/virt-acpi-build.c |   5 +-
 include/hw/acpi/ghes.h   |  17 ++--
 4 files changed, 138 insertions(+), 97 deletions(-)
hw/acpi/ghes-stub.c +2 -1
index c315de1802d6..8762449870b5 100644
@@ -11,7 +11,8 @@
 #include "qemu/osdep.h"
 #include "hw/acpi/ghes.h"
 
-int acpi_ghes_record_errors(uint8_t source_id, uint64_t physical_address)
+int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+                            uint64_t physical_address)
 {
     return -1;
 }
hw/acpi/ghes.c +128 -82

324 lines not on this page · raw message

hw/arm/virt-acpi-build.c +2 -3
index 1769467d23b2..79635bc7a0a8 100644
@@ -944,10 +944,9 @@ void virt_acpi_build(VirtMachineState *vms, AcpiBuildTables *tables)
     build_dbg2(tables_blob, tables->linker, vms);
 
     if (vms->ras) {
-        build_ghes_error_table(tables->hardware_errors, tables->linker);
         acpi_add_table(table_offsets, tables_blob);
-        acpi_build_hest(tables_blob, tables->linker, vms->oem_id,
-                        vms->oem_table_id);
+        acpi_build_hest(tables_blob, tables->hardware_errors, tables->linker,
+                        vms->oem_id, vms->oem_table_id);
     }
 
     if (ms->numa_state->num_nodes > 0) {
include/hw/acpi/ghes.h +6 -11
index b977d65564ba..6e349264fd8b 100644
@@ -29,7 +29,7 @@
 extern NotifierList acpi_generic_error_notifiers;
 
 /*
- * Values for Hardware Error Notification Type field
+ * ACPI spec values for Hardware Error Notification Type field
  */
 enum AcpiGhesNotifyType {
     /* Polled */
@@ -60,24 +60,19 @@ enum AcpiGhesNotifyType {
     ACPI_GHES_NOTIFY_RESERVED = 12
 };
 
-/* Those are used as table indexes when building GHES tables */
-enum {
-    ACPI_HEST_SRC_ID_SEA = 0,
-    ACPI_HEST_SRC_ID_GED,
-    ACPI_HEST_SRC_ID_RESERVED,
-};
-
 typedef struct AcpiGhesState {
+    uint64_t hest_addr_le;
     uint64_t ghes_addr_le;
     bool present; /* True if GHES is present at all on this board */
 } AcpiGhesState;
 
-void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker);
-void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
+void acpi_build_hest(GArray *table_data, GArray *hardware_errors,
+                     BIOSLinker *linker,
                      const char *oem_id, const char *oem_table_id);
 void acpi_ghes_add_fw_cfg(AcpiGhesState *vms, FWCfgState *s,
                           GArray *hardware_errors);
-int acpi_ghes_record_errors(uint8_t notify, uint64_t error_physical_addr);
+int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+                            uint64_t error_physical_addr);
 void ghes_record_cper_errors(const void *cper, size_t len,
                              enum AcpiGhesNotifyType notify, Error **errp);
 
Signature · 1 line
2.46.0
#10 lore raw

Re: [PATCH v8 05/13] acpi/ghes: rework the logic to handle HEST source ID

Quoted 4 lines · On Fri, 16 Aug 2024 09:37:37 +0200 Mauro Carvalho Chehab…

The current logic is based on a lot of duct tape, with offsets calculated based on one define with the number of source IDs and an enum.

Quoted 8 lines

Rewrite the logic in a way that it would be more resilient of code changes, by moving the source ID count to an enum and make the offset calculus more explicit.

Such change was inspired on a patch from Jonathan Cameron splitting the logic to get the CPER address on a separate function, as this will be needed to support generic error injection.

patch does too many things, that it's hard to review. Please split it up on smaller distinct parts, with more specific commit messages. (see some comments below)

Quoted 27 lines
---
 hw/acpi/ghes-stub.c      |   3 +-
 hw/acpi/ghes.c           | 210 ++++++++++++++++++++++++---------------
 hw/arm/virt-acpi-build.c |   5 +-
 include/hw/acpi/ghes.h   |  17 ++--
 4 files changed, 138 insertions(+), 97 deletions(-)
hw/acpi/ghes-stub.c +2 -1
index c315de1802d6..8762449870b5 100644
@@ -11,7 +11,8 @@
 #include "qemu/osdep.h"
 #include "hw/acpi/ghes.h"
 
-int acpi_ghes_record_errors(uint8_t source_id, uint64_t physical_address)
+int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+                            uint64_t physical_address)
 {
     return -1;
 }
hw/acpi/ghes.c +0 -0
index df59fd35568c..7870f51e2a9e 100644

hw/acpi/ghes.c line 32

Quoted 7 lines
hw/acpi/ghes.c +2 -2
@@ -28,14 +28,23 @@
 #include "hw/nvram/fw_cfg.h"
 #include "qemu/uuid.h"
 
-#define ACPI_GHES_ERRORS_FW_CFG_FILE        "etc/hardware_errors"
-#define ACPI_GHES_DATA_ADDR_FW_CFG_FILE     "etc/hardware_errors_addr"
+#define ACPI_HW_ERROR_FW_CFG_FILE           "etc/hardware_errors"
+#define ACPI_HW_ERROR_ADDR_FW_CFG_FILE      "etc/hardware_errors_addr"

split out renaming part into a presiding separate patch, so it won't mask a new code

Quoted 10 lines
hw/acpi/ghes.c +2 -2
+#define ACPI_HEST_ADDR_FW_CFG_FILE          "etc/acpi_table_hest_addr"
 
 /* The max size in bytes for one error block */
 #define ACPI_GHES_MAX_RAW_DATA_LENGTH   (1 * KiB)
 
 
 
-/* Support ARMv8 SEA notification type error source and GPIO interrupt. */
-#define ACPI_GHES_ERROR_SOURCE_COUNT        2
+/*

hw/acpi/ghes.c line 48

Quoted 10 lines
hw/acpi/ghes.c +9 -0
+ * ID numbers used to fill HEST source ID field
+ */
+enum AcpiHestSourceId {
+    ACPI_HEST_SRC_ID_SEA,
+    ACPI_HEST_SRC_ID_GED,
+
+    /* Shall be the last one */
+    ACPI_HEST_SRC_ID_COUNT
+} AcpiHestSourceId;
 

this rename also should go into its own separate patch.

Quoted 91 lines

/* Generic Hardware Error Source version 2 */
#define ACPI_GHES_SOURCE_GENERIC_ERROR_V2 10

hw/acpi/ghes.c +37 -10
@@ -63,6 +72,19 @@
  */
 #define ACPI_GHES_GESB_SIZE                 20
 
+/*
+ * Offsets with regards to the start of the HEST table stored at
+ * ags->hest_addr_le, according with the memory layout map at
+ * docs/specs/acpi_hest_ghes.rst.
+ */
+
+/* ACPI 4.0: 17.3.2 ACPI Error Source */
+#define ACPI_HEST_HEADER_SIZE    40
+
+/* ACPI 6.2: 18.3.2.8 Generic Hardware Error Source version 2 */
+#define HEST_GHES_V2_TABLE_SIZE  92
+#define GHES_ACK_OFFSET          (64 + GAS_ADDR_OFFSET + ACPI_HEST_HEADER_SIZE)
+
 /*
  * Values for error_severity field
  */
@@ -236,17 +258,17 @@ static int acpi_ghes_record_mem_error(uint64_t error_block_address,
  * Initialize "etc/hardware_errors" and "etc/hardware_errors_addr" fw_cfg blobs.
  * See docs/specs/acpi_hest_ghes.rst for blobs format.
  */
-void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
+static void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
 {
     int i, error_status_block_offset;
 
     /* Build error_block_address */
-    for (i = 0; i < ACPI_GHES_ERROR_SOURCE_COUNT; i++) {
+    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
         build_append_int_noprefix(hardware_errors, 0, sizeof(uint64_t));
     }
 
     /* Build read_ack_register */
-    for (i = 0; i < ACPI_GHES_ERROR_SOURCE_COUNT; i++) {
+    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
         /*
          * Initialize the value of read_ack_register to 1, so GHES can be
          * writable after (re)boot.
@@ -261,20 +283,20 @@ void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
 
     /* Reserve space for Error Status Data Block */
     acpi_data_push(hardware_errors,
-        ACPI_GHES_MAX_RAW_DATA_LENGTH * ACPI_GHES_ERROR_SOURCE_COUNT);
+        ACPI_GHES_MAX_RAW_DATA_LENGTH * ACPI_HEST_SRC_ID_COUNT);
 
     /* Tell guest firmware to place hardware_errors blob into RAM */
-    bios_linker_loader_alloc(linker, ACPI_GHES_ERRORS_FW_CFG_FILE,
+    bios_linker_loader_alloc(linker, ACPI_HW_ERROR_FW_CFG_FILE,
                              hardware_errors, sizeof(uint64_t), false);
 
-    for (i = 0; i < ACPI_GHES_ERROR_SOURCE_COUNT; i++) {
+    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
         /*
          * Tell firmware to patch error_block_address entries to point to
          * corresponding "Generic Error Status Block"
          */
         bios_linker_loader_add_pointer(linker,
-            ACPI_GHES_ERRORS_FW_CFG_FILE, sizeof(uint64_t) * i,
-            sizeof(uint64_t), ACPI_GHES_ERRORS_FW_CFG_FILE,
+            ACPI_HW_ERROR_FW_CFG_FILE, sizeof(uint64_t) * i,
+            sizeof(uint64_t), ACPI_HW_ERROR_FW_CFG_FILE,
             error_status_block_offset + i * ACPI_GHES_MAX_RAW_DATA_LENGTH);
     }
 
@@ -282,16 +304,39 @@ void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
      * tell firmware to write hardware_errors GPA into
      * hardware_errors_addr fw_cfg, once the former has been initialized.
      */
-    bios_linker_loader_write_pointer(linker, ACPI_GHES_DATA_ADDR_FW_CFG_FILE,
-        0, sizeof(uint64_t), ACPI_GHES_ERRORS_FW_CFG_FILE, 0);
+    bios_linker_loader_write_pointer(linker, ACPI_HW_ERROR_ADDR_FW_CFG_FILE, 0,
+                                     sizeof(uint64_t),
+                                     ACPI_HW_ERROR_FW_CFG_FILE, 0);
+}
+
+static bool ghes_notify_to_source_id(enum AcpiGhesNotifyType notify,
+                                     enum AcpiHestSourceId *source_id)
+{
+    switch (notify) {
+    case ACPI_GHES_NOTIFY_SEA:             /* ARMv8 */
+        *source_id = ACPI_HEST_SRC_ID_SEA;
+        return false;
+    case ACPI_GHES_NOTIFY_GPIO:
+        *source_id = ACPI_HEST_SRC_ID_GED;
+        return false;
+    default:

hw/acpi/ghes.c void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker) · line 331

Quoted 10 lines
hw/acpi/ghes.c +6 -1
+        /* Unsupported notification types */
+        return true;
+    }
 }
 
 /* Build Generic Hardware Error Source version 2 (GHESv2) */
-static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
+static void build_ghes_v2(GArray *table_data,
+                          enum AcpiGhesNotifyType notify,
+                          BIOSLinker *linker)

I'd suggest to drop this change as AcpiGhesNotifyType is not unique ID, in fact one can easily have several ID with the same type.

Quoted 265 lines
hw/acpi/ghes.c +72 -67
 {
     uint64_t address_offset;
+    enum AcpiHestSourceId source_id;
 
-    assert(source_id < ACPI_HEST_SRC_ID_RESERVED);
+    if (ghes_notify_to_source_id(notify, &source_id)) {
+        error_report("Error: notify %d not supported", notify);
+        abort();
+    }
 
     /*
      * Type:
@@ -319,24 +364,13 @@ static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
     build_append_gas(table_data, AML_AS_SYSTEM_MEMORY, 0x40, 0,
                      4 /* QWord access */, 0);
     bios_linker_loader_add_pointer(linker, ACPI_BUILD_TABLE_FILE,
-        address_offset + GAS_ADDR_OFFSET, sizeof(uint64_t),
-        ACPI_GHES_ERRORS_FW_CFG_FILE, source_id * sizeof(uint64_t));
+                                   address_offset + GAS_ADDR_OFFSET,
+                                   sizeof(uint64_t),
+                                   ACPI_HW_ERROR_FW_CFG_FILE,
+                                   source_id * sizeof(uint64_t));
 
-    switch (source_id) {
-    case ACPI_HEST_SRC_ID_SEA:
-        /*
-         * Notification Structure
-         * Now only enable ARMv8 SEA notification type
-         */
-        build_ghes_hw_error_notification(table_data, ACPI_GHES_NOTIFY_SEA);
-        break;
-    case ACPI_HEST_SRC_ID_GED:
-        build_ghes_hw_error_notification(table_data, ACPI_GHES_NOTIFY_GPIO);
-        break;
-    default:
-        error_report("Not support this error source");
-        abort();
-    }
+    /* Notification Structure */
+    build_ghes_hw_error_notification(table_data, notify);
 
     /* Error Status Block Length */
     build_append_int_noprefix(table_data, ACPI_GHES_MAX_RAW_DATA_LENGTH, 4);
@@ -350,9 +384,11 @@ static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
     build_append_gas(table_data, AML_AS_SYSTEM_MEMORY, 0x40, 0,
                      4 /* QWord access */, 0);
     bios_linker_loader_add_pointer(linker, ACPI_BUILD_TABLE_FILE,
-        address_offset + GAS_ADDR_OFFSET,
-        sizeof(uint64_t), ACPI_GHES_ERRORS_FW_CFG_FILE,
-        (ACPI_GHES_ERROR_SOURCE_COUNT + source_id) * sizeof(uint64_t));
+                                   address_offset + GAS_ADDR_OFFSET,
+                                   sizeof(uint64_t),
+                                   ACPI_HW_ERROR_FW_CFG_FILE,
+                                   (ACPI_HEST_SRC_ID_COUNT + source_id) *
+                                   sizeof(uint64_t));
 
     /*
      * Read Ack Preserve field
@@ -365,90 +401,100 @@ static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
 }
 
 /* Build Hardware Error Source Table */
-void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
+void acpi_build_hest(GArray *table_data, GArray *hardware_errors,
+                     BIOSLinker *linker,
                      const char *oem_id, const char *oem_table_id)
 {
     AcpiTable table = { .sig = "HEST", .rev = 1,
                         .oem_id = oem_id, .oem_table_id = oem_table_id };
 
+    build_ghes_error_table(hardware_errors, linker);
+
+    int hest_offset = table_data->len;
+
     acpi_table_begin(&table, table_data);
 
     /* Error Source Count */
-    build_append_int_noprefix(table_data, ACPI_GHES_ERROR_SOURCE_COUNT, 4);
-    build_ghes_v2(table_data, ACPI_HEST_SRC_ID_SEA, linker);
-    build_ghes_v2(table_data, ACPI_HEST_SRC_ID_GED, linker);
+    build_append_int_noprefix(table_data, ACPI_HEST_SRC_ID_COUNT, 4);
+    build_ghes_v2(table_data, ACPI_GHES_NOTIFY_SEA, linker);
+    build_ghes_v2(table_data, ACPI_GHES_NOTIFY_GPIO, linker);
 
     acpi_table_end(linker, &table);
+
+    /*
+     * tell firmware to write into GPA the address of HEST via fw_cfg,
+     * once initialized.
+     */
+    bios_linker_loader_write_pointer(linker,
+                                     ACPI_HEST_ADDR_FW_CFG_FILE, 0,
+                                     sizeof(uint64_t),
+                                     ACPI_BUILD_TABLE_FILE, hest_offset);
 }
 
 void acpi_ghes_add_fw_cfg(AcpiGhesState *ags, FWCfgState *s,
                           GArray *hardware_error)
 {
     /* Create a read-only fw_cfg file for GHES */
-    fw_cfg_add_file(s, ACPI_GHES_ERRORS_FW_CFG_FILE, hardware_error->data,
+    fw_cfg_add_file(s, ACPI_HW_ERROR_FW_CFG_FILE, hardware_error->data,
                     hardware_error->len);
 
     /* Create a read-write fw_cfg file for Address */
-    fw_cfg_add_file_callback(s, ACPI_GHES_DATA_ADDR_FW_CFG_FILE, NULL, NULL,
+    fw_cfg_add_file_callback(s, ACPI_HW_ERROR_ADDR_FW_CFG_FILE, NULL, NULL,
         NULL, &(ags->ghes_addr_le), sizeof(ags->ghes_addr_le), false);
 
+    fw_cfg_add_file_callback(s, ACPI_HEST_ADDR_FW_CFG_FILE, NULL, NULL,
+        NULL, &(ags->hest_addr_le), sizeof(ags->hest_addr_le), false);
+
     ags->present = true;
 }
 
-int acpi_ghes_record_errors(uint8_t source_id, uint64_t physical_address)
+int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+                            uint64_t physical_address)
 {
-    uint64_t error_block_addr, read_ack_register_addr, read_ack_register = 0;
-    uint64_t start_addr;
-    bool ret = -1;
+    uint64_t cper_addr, read_ack_register = 0;
+    uint64_t read_ack_start_addr;
+    enum AcpiHestSourceId source;
     AcpiGedState *acpi_ged_state;
     AcpiGhesState *ags;
 
-    assert(source_id < ACPI_HEST_SRC_ID_RESERVED);
+    if (ghes_notify_to_source_id(ACPI_HEST_SRC_ID_SEA, &source)) {
+        error_report("GHES: Invalid error block/ack address(es) for notify %d",
+                     notify);
+        return -1;
+    }
 
     acpi_ged_state = ACPI_GED(object_resolve_path_type("", TYPE_ACPI_GED,
                                                        NULL));
     g_assert(acpi_ged_state);
     ags = &acpi_ged_state->ghes_state;
 
-    start_addr = le64_to_cpu(ags->ghes_addr_le);
+    cper_addr = le64_to_cpu(ags->ghes_addr_le);
+    cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
+    read_ack_start_addr = cper_addr + source * sizeof(uint64_t);
 
-    if (physical_address) {
+    cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
+    cper_addr += source * ACPI_GHES_MAX_RAW_DATA_LENGTH;
 
-        if (source_id < ACPI_HEST_SRC_ID_RESERVED) {
-            start_addr += source_id * sizeof(uint64_t);
-        }
-
-        cpu_physical_memory_read(start_addr, &error_block_addr,
-                                 sizeof(error_block_addr));
-
-        error_block_addr = le64_to_cpu(error_block_addr);
-
-        read_ack_register_addr = start_addr +
-            ACPI_GHES_ERROR_SOURCE_COUNT * sizeof(uint64_t);
-
-        cpu_physical_memory_read(read_ack_register_addr,
-                                 &read_ack_register, sizeof(read_ack_register));
-
-        /* zero means OSPM does not acknowledge the error */
-        if (!read_ack_register) {
-            error_report("OSPM does not acknowledge previous error,"
-                " so can not record CPER for current error anymore");
-        } else if (error_block_addr) {
-            read_ack_register = cpu_to_le64(0);
-            /*
-             * Clear the Read Ack Register, OSPM will write it to 1 when
-             * it acknowledges this error.
-             */
-            cpu_physical_memory_write(read_ack_register_addr,
-                &read_ack_register, sizeof(uint64_t));
-
-            ret = acpi_ghes_record_mem_error(error_block_addr,
-                                             physical_address);
-        } else
-            error_report("can not find Generic Error Status Block");
+    if (!physical_address) {
+        error_report("can not find Generic Error Status Block for notify %d",
+                     notify);
+        return -1;
     }
 
-    return ret;
+    cpu_physical_memory_read(read_ack_start_addr,
+                             &read_ack_register, sizeof(read_ack_register));
+
+    /* zero means OSPM does not acknowledge the error */
+
+    read_ack_register = cpu_to_le64(0);
+    /*
+     * Clear the Read Ack Register, OSPM will write it to 1 when
+     * it acknowledges this error.
+     */
+    cpu_physical_memory_write(read_ack_start_addr,
+                              &read_ack_register, sizeof(uint64_t));
+
+    return acpi_ghes_record_mem_error(cper_addr, physical_address);
 }
 
 NotifierList acpi_generic_error_notifiers =
hw/arm/virt-acpi-build.c +2 -3
index 1769467d23b2..79635bc7a0a8 100644
@@ -944,10 +944,9 @@ void virt_acpi_build(VirtMachineState *vms, AcpiBuildTables *tables)
     build_dbg2(tables_blob, tables->linker, vms);
 
     if (vms->ras) {
-        build_ghes_error_table(tables->hardware_errors, tables->linker);
         acpi_add_table(table_offsets, tables_blob);
-        acpi_build_hest(tables_blob, tables->linker, vms->oem_id,
-                        vms->oem_table_id);
+        acpi_build_hest(tables_blob, tables->hardware_errors, tables->linker,
+                        vms->oem_id, vms->oem_table_id);
     }
 
     if (ms->numa_state->num_nodes > 0) {
include/hw/acpi/ghes.h +6 -11
index b977d65564ba..6e349264fd8b 100644
@@ -29,7 +29,7 @@
 extern NotifierList acpi_generic_error_notifiers;
 
 /*
- * Values for Hardware Error Notification Type field
+ * ACPI spec values for Hardware Error Notification Type field
  */
 enum AcpiGhesNotifyType {
     /* Polled */
@@ -60,24 +60,19 @@ enum AcpiGhesNotifyType {
     ACPI_GHES_NOTIFY_RESERVED = 12
 };
 
-/* Those are used as table indexes when building GHES tables */
-enum {
-    ACPI_HEST_SRC_ID_SEA = 0,
-    ACPI_HEST_SRC_ID_GED,
-    ACPI_HEST_SRC_ID_RESERVED,
-};
-
 typedef struct AcpiGhesState {
+    uint64_t hest_addr_le;
     uint64_t ghes_addr_le;
     bool present; /* True if GHES is present at all on this board */
 } AcpiGhesState;
 
-void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker);
-void acpi_build_hest(GArray *table_data, BIOSLinker *linker,
+void acpi_build_hest(GArray *table_data, GArray *hardware_errors,
+                     BIOSLinker *linker,
                      const char *oem_id, const char *oem_table_id);
 void acpi_ghes_add_fw_cfg(AcpiGhesState *vms, FWCfgState *s,
                           GArray *hardware_errors);
-int acpi_ghes_record_errors(uint8_t notify, uint64_t error_physical_addr);
+int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+                            uint64_t error_physical_addr);
 void ghes_record_cper_errors(const void *cper, size_t len,
                              enum AcpiGhesNotifyType notify, Error **errp);
 
#11 lore raw

Re: [PATCH v8 05/13] acpi/ghes: rework the logic to handle HEST source ID

Em Mon, 19 Aug 2024 14:10:37 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 11 lines
Quoted 8 lines · On Fri, 16 Aug 2024 09:37:37 +0200 Mauro Carvalho Chehab…

The current logic is based on a lot of duct tape, with offsets calculated based on one define with the number of source IDs and an enum.

Rewrite the logic in a way that it would be more resilient of code changes, by moving the source ID count to an enum and make the offset calculus more explicit.

Quoted 8 lines
Quoted 4 lines

Such change was inspired on a patch from Jonathan Cameron splitting the logic to get the CPER address on a separate function, as this will be needed to support generic error injection.

patch does too many things, that it's hard to review. Please split it up on smaller distinct parts, with more specific commit messages. (see some comments below)

True, but there's not much that can be done when doing it and still keeping the code working. I'll split the renames.

Quoted 46 lines
Quoted 29 lines
---
 hw/acpi/ghes-stub.c      |   3 +-
 hw/acpi/ghes.c           | 210 ++++++++++++++++++++++++---------------
 hw/arm/virt-acpi-build.c |   5 +-
 include/hw/acpi/ghes.h   |  17 ++--
 4 files changed, 138 insertions(+), 97 deletions(-)
hw/acpi/ghes-stub.c +2 -1
index c315de1802d6..8762449870b5 100644
@@ -11,7 +11,8 @@
 #include "qemu/osdep.h"
 #include "hw/acpi/ghes.h"
 
-int acpi_ghes_record_errors(uint8_t source_id, uint64_t physical_address)
+int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+                            uint64_t physical_address)
 {
     return -1;
 }
hw/acpi/ghes.c +0 -0
index df59fd35568c..7870f51e2a9e 100644

hw/acpi/ghes.c line 32

Quoted 7 lines
hw/acpi/ghes.c +2 -2
@@ -28,14 +28,23 @@
 #include "hw/nvram/fw_cfg.h"
 #include "qemu/uuid.h"
 
-#define ACPI_GHES_ERRORS_FW_CFG_FILE        "etc/hardware_errors"
-#define ACPI_GHES_DATA_ADDR_FW_CFG_FILE     "etc/hardware_errors_addr"
+#define ACPI_HW_ERROR_FW_CFG_FILE           "etc/hardware_errors"
+#define ACPI_HW_ERROR_ADDR_FW_CFG_FILE      "etc/hardware_errors_addr"  

split out renaming part into a presiding separate patch, so it won't mask a new code

Quoted 5 lines
hw/acpi/ghes.c +1 -0
+#define ACPI_HEST_ADDR_FW_CFG_FILE          "etc/acpi_table_hest_addr"
 
 /* The max size in bytes for one error block */
 #define ACPI_GHES_MAX_RAW_DATA_LENGTH   (1 * KiB)
 
Quoted 14 lines

hw/acpi/ghes.c line 48

Quoted 13 lines
hw/acpi/ghes.c +10 -2
-/* Support ARMv8 SEA notification type error source and GPIO interrupt. */
-#define ACPI_GHES_ERROR_SOURCE_COUNT        2
+/*
+ * ID numbers used to fill HEST source ID field
+ */
+enum AcpiHestSourceId {
+    ACPI_HEST_SRC_ID_SEA,
+    ACPI_HEST_SRC_ID_GED,
+
+    /* Shall be the last one */
+    ACPI_HEST_SRC_ID_COUNT
+} AcpiHestSourceId;
 

this rename also should go into its own separate patch.

I opted to remove this completely and move it to arm/virt, as this specific set of sources is for ARM.

On such split, I ended placing the QMP error injection as the first one, as this is probably the first one that we'll be mapping on x86 and other architectures.

This way, the code at ghes.c won't rely on any hardcoded values. They'll be passed at target ACPI table preparation using this function:

void acpi_build_hest(GArray *table_data, GArray *hardware_errors,
BIOSLinker *linker,
const uint16_t * const notify,
int num_sources,
const char *oem_id, const char *oem_table_id)

On arm (at the rework patch, before adding GPIO method), the call to the HEST build table (and etc/hardware_errors init) is done via:

    static const uint16_t hest_ghes_notify[] = {
        [ARM_ACPI_HEST_SRC_ID_SEA] = ACPI_GHES_NOTIFY_SEA,
    };

    void virt_acpi_build(VirtMachineState *vms, AcpiBuildTables *tables)
    {
	...
	if (vms->ras) {
        acpi_add_table(table_offsets, tables_blob);
        acpi_build_hest(tables_blob, tables->hardware_errors, tables->linker,
                        hest_ghes_notify, sizeof(hest_ghes_notify),
                        vms->oem_id, vms->oem_table_id);
        }
    ...

This way, adding support for a new notification type at the arch-specific code is as easy as adding a new entry to hest_ghes_notify[].

Quoted 97 lines
Quoted 97 lines
hw/acpi/ghes.c +40 -10
 /* Generic Hardware Error Source version 2 */
 #define ACPI_GHES_SOURCE_GENERIC_ERROR_V2   10
@@ -63,6 +72,19 @@
  */
 #define ACPI_GHES_GESB_SIZE                 20
 
+/*
+ * Offsets with regards to the start of the HEST table stored at
+ * ags->hest_addr_le, according with the memory layout map at
+ * docs/specs/acpi_hest_ghes.rst.
+ */
+
+/* ACPI 4.0: 17.3.2 ACPI Error Source */
+#define ACPI_HEST_HEADER_SIZE    40
+
+/* ACPI 6.2: 18.3.2.8 Generic Hardware Error Source version 2 */
+#define HEST_GHES_V2_TABLE_SIZE  92
+#define GHES_ACK_OFFSET          (64 + GAS_ADDR_OFFSET + ACPI_HEST_HEADER_SIZE)
+
 /*
  * Values for error_severity field
  */
@@ -236,17 +258,17 @@ static int acpi_ghes_record_mem_error(uint64_t error_block_address,
  * Initialize "etc/hardware_errors" and "etc/hardware_errors_addr" fw_cfg blobs.
  * See docs/specs/acpi_hest_ghes.rst for blobs format.
  */
-void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
+static void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
 {
     int i, error_status_block_offset;
 
     /* Build error_block_address */
-    for (i = 0; i < ACPI_GHES_ERROR_SOURCE_COUNT; i++) {
+    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
         build_append_int_noprefix(hardware_errors, 0, sizeof(uint64_t));
     }
 
     /* Build read_ack_register */
-    for (i = 0; i < ACPI_GHES_ERROR_SOURCE_COUNT; i++) {
+    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
         /*
          * Initialize the value of read_ack_register to 1, so GHES can be
          * writable after (re)boot.
@@ -261,20 +283,20 @@ void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
 
     /* Reserve space for Error Status Data Block */
     acpi_data_push(hardware_errors,
-        ACPI_GHES_MAX_RAW_DATA_LENGTH * ACPI_GHES_ERROR_SOURCE_COUNT);
+        ACPI_GHES_MAX_RAW_DATA_LENGTH * ACPI_HEST_SRC_ID_COUNT);
 
     /* Tell guest firmware to place hardware_errors blob into RAM */
-    bios_linker_loader_alloc(linker, ACPI_GHES_ERRORS_FW_CFG_FILE,
+    bios_linker_loader_alloc(linker, ACPI_HW_ERROR_FW_CFG_FILE,
                              hardware_errors, sizeof(uint64_t), false);
 
-    for (i = 0; i < ACPI_GHES_ERROR_SOURCE_COUNT; i++) {
+    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
         /*
          * Tell firmware to patch error_block_address entries to point to
          * corresponding "Generic Error Status Block"
          */
         bios_linker_loader_add_pointer(linker,
-            ACPI_GHES_ERRORS_FW_CFG_FILE, sizeof(uint64_t) * i,
-            sizeof(uint64_t), ACPI_GHES_ERRORS_FW_CFG_FILE,
+            ACPI_HW_ERROR_FW_CFG_FILE, sizeof(uint64_t) * i,
+            sizeof(uint64_t), ACPI_HW_ERROR_FW_CFG_FILE,
             error_status_block_offset + i * ACPI_GHES_MAX_RAW_DATA_LENGTH);
     }
 
@@ -282,16 +304,39 @@ void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker)
      * tell firmware to write hardware_errors GPA into
      * hardware_errors_addr fw_cfg, once the former has been initialized.
      */
-    bios_linker_loader_write_pointer(linker, ACPI_GHES_DATA_ADDR_FW_CFG_FILE,
-        0, sizeof(uint64_t), ACPI_GHES_ERRORS_FW_CFG_FILE, 0);
+    bios_linker_loader_write_pointer(linker, ACPI_HW_ERROR_ADDR_FW_CFG_FILE, 0,
+                                     sizeof(uint64_t),
+                                     ACPI_HW_ERROR_FW_CFG_FILE, 0);
+}
+
+static bool ghes_notify_to_source_id(enum AcpiGhesNotifyType notify,
+                                     enum AcpiHestSourceId *source_id)
+{
+    switch (notify) {
+    case ACPI_GHES_NOTIFY_SEA:             /* ARMv8 */
+        *source_id = ACPI_HEST_SRC_ID_SEA;
+        return false;
+    case ACPI_GHES_NOTIFY_GPIO:
+        *source_id = ACPI_HEST_SRC_ID_GED;
+        return false;
+    default:
+        /* Unsupported notification types */
+        return true;
+    }
 }
 
Quoted 8 lines

hw/acpi/ghes.c void build_ghes_error_table(GArray *hardware_errors, BIOSLinker *linker) · line 331

Quoted 5 lines
hw/acpi/ghes.c +3 -1
 /* Build Generic Hardware Error Source version 2 (GHESv2) */
-static void build_ghes_v2(GArray *table_data, int source_id, BIOSLinker *linker)
+static void build_ghes_v2(GArray *table_data,
+                          enum AcpiGhesNotifyType notify,
+                          BIOSLinker *linker)  

I'd suggest to drop this change as AcpiGhesNotifyType is not unique ID, in fact one can easily have several ID with the same type.

IMO, the best is to have different IDs for each type of notification on a given guest type. The way I see, each supported notification type will require a separate source ID, as different notification mechanisms will use different ways to synchronize between QEMU, BIOS and OS.

So, with the new version I'll be sending, this will also use whatever table with a notification/source_ID association is passed by target acpi init code, so the way I'm designing it is that different archs may reuse the same id numbers like:

    /* ARM acpi build logic - currently the only one implemented */
    static const uint16_t hest_ghes_notify[] = {
        [0] = ACPI_GHES_NOTIFY_GPIO,
        [1] = ACPI_GHES_NOTIFY_SEA,
    };

    /* x86 acpi build logic */
    static const uint16_t hest_ghes_notify[] = {
        [0] = ACPI_GHES_NOTIFY_SCI,
        [1] = ACPI_GHES_NOTIFY_MCE,
        [2] = ACPI_GHES_NOTIFY_NMI,
    };

    /* PPC acpi build logic */
    static const uint16_t hest_ghes_notify[] = {
        [0] = ACPI_GHES_NOTIFY_GPIO,
    }

...

Btw, I'm reserving 0 for QMP based event injection, as we don't know in advance how much events each architecture will implement. So, basically this will be added to ghes.h:

    /* Source ID associated with qapi/acpi-hest.json QMP error injection */
    #define ACPI_HEST_SRC_ID_QMP   0

And the actual binding on arm is done as:

    enum {
        ARM_ACPI_HEST_SRC_ID_GPIO = ACPI_HEST_SRC_ID_QMP,
        ARM_ACPI_HEST_SRC_ID_SEA,
    };

    static const uint16_t hest_ghes_notify[] = {
        [ARM_ACPI_HEST_SRC_ID_SEA] = ACPI_GHES_NOTIFY_SEA,
        [ARM_ACPI_HEST_SRC_ID_GPIO] = ACPI_GHES_NOTIFY_GPIO,
    };

Regards,
Mauro

#12 lore raw

PATCH v8 6/13 acpi/ghes: add support for generic error injection via QAPI

In reply to Mauro Carvalho Chehab

Provide a generic interface for error injection via GHESv2.

This patch is co-authored:
- original ghes logic to inject a simple ARM record by Shiju Jose;
- generic logic to handle block addresses by Jonathan Cameron;
- generic GHESv2 error inject by Mauro Carvalho Chehab;

---
 hw/acpi/ghes.c      | 57 +++++++++++++++++++++++++++++++++++++++++++++
 hw/acpi/ghes_cper.c |  2 +-
 2 files changed, 58 insertions(+), 1 deletion(-)
hw/acpi/ghes.c +57 -0
index 7870f51e2a9e..a3ae710dcf81 100644
@@ -500,6 +500,63 @@ int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
 NotifierList acpi_generic_error_notifiers =
     NOTIFIER_LIST_INITIALIZER(error_device_notifiers);
 
+void ghes_record_cper_errors(uint8_t *cper, size_t len,
+                             enum AcpiGhesNotifyType notify, Error **errp)
+{
+    uint64_t cper_addr, read_ack_start_addr;
+    enum AcpiHestSourceId source;
+    AcpiGedState *acpi_ged_state;
+    AcpiGhesState *ags;
+    uint64_t read_ack;
+
+    if (ghes_notify_to_source_id(notify, &source)) {
+        error_setg(errp,
+                   "GHES: Invalid error block/ack address(es) for notify %d",
+                   notify);
+        return;
+    }
+
+    acpi_ged_state = ACPI_GED(object_resolve_path_type("", TYPE_ACPI_GED,
+                                                       NULL));
+    g_assert(acpi_ged_state);
+    ags = &acpi_ged_state->ghes_state;
+
+    cper_addr = le64_to_cpu(ags->ghes_addr_le);
+    cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
+    read_ack_start_addr = cper_addr + source * sizeof(uint64_t);
+
+    cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
+    cper_addr += source * ACPI_GHES_MAX_RAW_DATA_LENGTH;
+
+    cpu_physical_memory_read(read_ack_start_addr,
+                             &read_ack, sizeof(uint64_t));
+
+    /* zero means OSPM does not acknowledge the error */
+    if (!read_ack) {
+        error_setg(errp,
+                   "Last CPER record was not acknowledged yet");
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, sizeof(uint64_t));
+        return;
+    }
+
+    read_ack = cpu_to_le64(0);
+    cpu_physical_memory_write(read_ack_start_addr,
+                              &read_ack, sizeof(uint64_t));
+
+    /* Build CPER record */
+
+    if (len > ACPI_GHES_MAX_RAW_DATA_LENGTH) {
+        error_setg(errp, "GHES CPER record is too big: %ld", len);
+    }
+
+    /* Write the generic error data entry into guest memory */
+    cpu_physical_memory_write(cper_addr, cper, len);
+
+    notifier_list_notify(&acpi_generic_error_notifiers, NULL);
+}
+
 bool acpi_ghes_present(void)
 {
     AcpiGedState *acpi_ged_state;
hw/acpi/ghes_cper.c +1 -1
index 92ca84d738de..2328dbff7012 100644
@@ -29,5 +29,5 @@ void qmp_ghes_cper(const char *qmp_cper,
         return;
     }
 
-    /* TODO: call a function at ghes */
+    ghes_record_cper_errors(cper, len, ACPI_GHES_NOTIFY_GPIO, errp);
 }
Signature · 1 line
2.46.0
#13 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Quoted 38 lines · On Fri, 16 Aug 2024 09:37:38 +0200 Mauro Carvalho Chehab…

Provide a generic interface for error injection via GHESv2.

This patch is co-authored:
- original ghes logic to inject a simple ARM record by Shiju Jose;
- generic logic to handle block addresses by Jonathan Cameron;
- generic GHESv2 error inject by Mauro Carvalho Chehab;

---
 hw/acpi/ghes.c      | 57 +++++++++++++++++++++++++++++++++++++++++++++
 hw/acpi/ghes_cper.c |  2 +-
 2 files changed, 58 insertions(+), 1 deletion(-)
hw/acpi/ghes.c +12 -0
index 7870f51e2a9e..a3ae710dcf81 100644
@@ -500,6 +500,63 @@ int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
 NotifierList acpi_generic_error_notifiers =
     NOTIFIER_LIST_INITIALIZER(error_device_notifiers);
 
+void ghes_record_cper_errors(uint8_t *cper, size_t len,
+                             enum AcpiGhesNotifyType notify, Error **errp)
+{
+    uint64_t cper_addr, read_ack_start_addr;
+    enum AcpiHestSourceId source;
+    AcpiGedState *acpi_ged_state;
+    AcpiGhesState *ags;
+    uint64_t read_ack;
+
+    if (ghes_notify_to_source_id(notify, &source)) {
+        error_setg(errp,
+                   "GHES: Invalid error block/ack address(es) for notify %d",

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify, · line 524

Quoted 10 lines
hw/acpi/ghes.c +10 -0
+                   notify);
+        return;
+    }
+
+    acpi_ged_state = ACPI_GED(object_resolve_path_type("", TYPE_ACPI_GED,
+                                                       NULL));
+    g_assert(acpi_ged_state);
+    ags = &acpi_ged_state->ghes_state;
+
+    cper_addr = le64_to_cpu(ags->ghes_addr_le);

^^^ suggest to rename to error_block_address that way reader can easily match it with spec.

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify, · line 525

Quoted 1 line
hw/acpi/ghes.c +1 -0
+    cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);

and it would be better to merge this with previous line to be more clear
+ to avoid shifting meaning of variable between lines.

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify, · line 529

Quoted 4 lines
hw/acpi/ghes.c +3 -0
+    read_ack_start_addr = cper_addr + source * sizeof(uint64_t);
 
+    cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
+    cper_addr += source * ACPI_GHES_MAX_RAW_DATA_LENGTH;

I'd avoid changing meaning of variable, it adds up to confusion. Anyway, what the point of of above math?

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify, · line 532

Quoted 3 lines
hw/acpi/ghes.c +3 -0
+
+    cpu_physical_memory_read(read_ack_start_addr,
+                             &read_ack, sizeof(uint64_t));

s/sizeof(uint64_t)/sizeof(read_ack)/
ditto elsewhere

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,

Quoted 9 lines
hw/acpi/ghes.c +8 -0
+
+    /* zero means OSPM does not acknowledge the error */
+    if (!read_ack) {
+        error_setg(errp,
+                   "Last CPER record was not acknowledged yet");
 
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));

we don't do this for SEV so, why are you setting it to 1 here?

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify, · line 552

Quoted 12 lines
hw/acpi/ghes.c +12 -0
+        return;
+    }
+
+    read_ack = cpu_to_le64(0);
+    cpu_physical_memory_write(read_ack_start_addr,
+                              &read_ack, sizeof(uint64_t));
+
+    /* Build CPER record */
+
+    if (len > ACPI_GHES_MAX_RAW_DATA_LENGTH) {
+        error_setg(errp, "GHES CPER record is too big: %ld", len);
+    }

move check at start of function?

Quoted 21 lines
hw/acpi/ghes.c +7 -0
+
+    /* Write the generic error data entry into guest memory */
+    cpu_physical_memory_write(cper_addr, cper, len);
+
+    notifier_list_notify(&acpi_generic_error_notifiers, NULL);
+}
+
 bool acpi_ghes_present(void)
 {
     AcpiGedState *acpi_ged_state;
hw/acpi/ghes_cper.c +1 -1
index 92ca84d738de..2328dbff7012 100644
@@ -29,5 +29,5 @@ void qmp_ghes_cper(const char *qmp_cper,
         return;
     }
 
-    /* TODO: call a function at ghes */
+    ghes_record_cper_errors(cper, len, ACPI_GHES_NOTIFY_GPIO, errp);
 }
#14 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,

Quoted 3 lines
hw/acpi/ghes.c +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));  

we don't do this for SEV so, why are you setting it to 1 here?

According with:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#generic-hardware-error-source-version-2-ghesv2-type-10

"These are the steps the OS must take once detecting an error from a particular GHESv2 error source:

OSPM detects error (via interrupt/exception or polling the block status)

OSPM copies the error status block

OSPM clears the block status field of the error status block

OSPM acknowledges the error via Read Ack register. For example:

OSPM reads the Read Ack register –> X

OSPM writes –> (( X & ReadAckPreserve) | ReadAckWrite)"

So, basically the guest OS takes some time to detect that an error is raised. When it detects, it needs to mark that the error was handled.

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

#15 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Quoted 21 lines · On Sun, 25 Aug 2024 05:29:23 +0200 Mauro Carvalho Chehab…

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,

Quoted 3 lines
hw/acpi/ghes.c +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));    

we don't do this for SEV so, why are you setting it to 1 here?

According with:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#generic-hardware-error-source-version-2-ghesv2-type-10

"These are the steps the OS must take once detecting an error from a particular GHESv2 error source:

OSPM detects error (via interrupt/exception or polling the block status)

OSPM copies the error status block

OSPM clears the block status field of the error status block

OSPM acknowledges the error via Read Ack register. For example:

Quoted 8 lines

OSPM reads the Read Ack register –> X

OSPM writes –> (( X & ReadAckPreserve) | ReadAckWrite)"

So, basically the guest OS takes some time to detect that an error is raised. When it detects, it needs to mark that the error was handled.

what you are doing here by setting read_ack = 1,
is making ack on behalf of OSPM when OSPM haven't handled existing error yet.

Essentially making HW/FW do the job of OSPM. That looks wrong to me. From HW/FW side read_ack register should be thought as read-only.

Quoted 6 lines

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

#16 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Quoted 33 lines · On Wed, 11 Sep 2024 15:21:32 +0200 Igor Mammedov <[email protected]> wrote:
Quoted 21 lines · On Sun, 25 Aug 2024 05:29:23 +0200 Mauro Carvalho Chehab…

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,

Quoted 3 lines
hw/acpi/ghes.c +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));      

we don't do this for SEV so, why are you setting it to 1 here?

According with:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#generic-hardware-error-source-version-2-ghesv2-type-10

"These are the steps the OS must take once detecting an error from a particular GHESv2 error source:

OSPM detects error (via interrupt/exception or polling the block status)

OSPM copies the error status block

OSPM clears the block status field of the error status block

OSPM acknowledges the error via Read Ack register. For example:

Quoted 8 lines

OSPM reads the Read Ack register –> X

OSPM writes –> (( X & ReadAckPreserve) | ReadAckWrite)"

So, basically the guest OS takes some time to detect that an error is raised. When it detects, it needs to mark that the error was handled.

Quoted 5 lines

what you are doing here by setting read_ack = 1,
is making ack on behalf of OSPM when OSPM haven't handled existing error yet.

Essentially making HW/FW do the job of OSPM. That looks wrong to me. From HW/FW side read_ack register should be thought as read-only.

It's not read-only because HW/FW has to clear it so that HW/FW can detect when the OSPM next writes it.

Agreed this write to 1 looks wrong, but the one a few lines further down (to zero it) is correct.

My bug a long time back I think.

Jonathan

Quoted 9 lines
Quoted 7 lines

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

#17 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Quoted 36 lines · On Wed, 11 Sep 2024 16:34:36 +0100 Jonathan Cameron…
Quoted 33 lines · On Wed, 11 Sep 2024 15:21:32 +0200 Igor Mammedov <[email protected]> wrote:
Quoted 29 lines · On Sun, 25 Aug 2024 05:29:23 +0200 Mauro Carvalho Chehab…

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,

Quoted 3 lines
hw/acpi/ghes.c +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));        

we don't do this for SEV so, why are you setting it to 1 here?

According with:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#generic-hardware-error-source-version-2-ghesv2-type-10

"These are the steps the OS must take once detecting an error from a particular GHESv2 error source:

OSPM detects error (via interrupt/exception or polling the block status)

OSPM copies the error status block

OSPM clears the block status field of the error status block

OSPM acknowledges the error via Read Ack register. For example:

OSPM reads the Read Ack register –> X

OSPM writes –> (( X & ReadAckPreserve) | ReadAckWrite)"

So, basically the guest OS takes some time to detect that an error is raised. When it detects, it needs to mark that the error was handled.

Quoted 8 lines
Quoted 5 lines

what you are doing here by setting read_ack = 1,
is making ack on behalf of OSPM when OSPM haven't handled existing error yet.

Essentially making HW/FW do the job of OSPM. That looks wrong to me. From HW/FW side read_ack register should be thought as read-only.

It's not read-only because HW/FW has to clear it so that HW/FW can detect when the OSPM next writes it.

By readonly, I've meant that hw shall not do above mentioned write (bad phrasing on my side).

Quoted 3 lines

Agreed this write to 1 looks wrong, but the one a few lines further down (to zero it) is correct.

yep, hw should clear register.
It would be better to so on OSPM ACK, but alas we can't intercept that, so the next option would be to do that at the time when we add a new error block

Quoted 15 lines

My bug a long time back I think.

Jonathan

Quoted 9 lines
Quoted 7 lines

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

#18 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Em Thu, 12 Sep 2024 14:42:33 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 9 lines
Quoted 6 lines · On Wed, 11 Sep 2024 16:34:36 +0100 Jonathan Cameron…
Quoted 3 lines · On Wed, 11 Sep 2024 15:21:32 +0200 Igor Mammedov <[email protected]> wrote:

On Sun, 25 Aug 2024 05:29:23 +0200
Mauro Carvalho Chehab <[email protected]> wrote:

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

hw/acpi/ghes.c int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,

Quoted 3 lines
hw/acpi/ghes.c +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));        

we don't do this for SEV so, why are you setting it to 1 here?

The diffstat doesn't really help here. The full code is:

    /* zero means OSPM does not acknowledge the error */
    if (!read_ack) {
        error_setg(errp,
                   "Last CPER record was not acknowledged yet");
        read_ack = 1;
        cpu_physical_memory_write(read_ack_start_addr,
                                  &read_ack, sizeof(read_ack));
        return;
    }
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines

what you are doing here by setting read_ack = 1,
is making ack on behalf of OSPM when OSPM haven't handled existing error yet.

Essentially making HW/FW do the job of OSPM. That looks wrong to me. From HW/FW side read_ack register should be thought as read-only.

It's not read-only because HW/FW has to clear it so that HW/FW can detect when the OSPM next writes it.

By readonly, I've meant that hw shall not do above mentioned write (bad phrasing on my side).

The above code is actually an error handling condition: if for some reason errors are triggered too fast, there's a bug on QEMU or there is a bug at the OSPM, an error message is raised and the logic resets the record to a sane state. So, on a next error, OSPM will get it.

As described at https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html?highlight=asynchronous#generic-hardware-error-source:

"Some platforms may describe multiple Generic Hardware Error Source
structures with different notification types, as defined in Table 18.10. For example, a platform may describe one error source for the handling of synchronous errors (e.g. MCE or SEA), and a second source for handling asynchronous errors (e.g. SCI or External Interrupt)."

Basically, the error logic there seems to fit for the asynchronous case, detecting if another error happened before OSPM handles the first one.

IMO, there are a couple of alternatives to handle such case:

1. Keep the code as-is: if this ever happens, an error message will be issued. If SEA/MCE gets implemented synchronously on HW/FW/OSPM, the above code will never be called;
2. Change the logic to do that only for asynchronous sources (currently, only if source ID is QMP);
3. Add a special QMP message to reset the notification ack. Probably would use Notification type as an input parameter;
4. Have a much more complex code to implement asynchronous notifications, with a queue to receive HEST errors and a separate thread to deliver errors to OSPM asynchronously. If we go this way, QMP would be returning the number of error messages queued, allowing error injection code to know if OSPM has troubles delivering errors;
5. Just return an error code without doing any resets. To me, this is the worse scenario.

I don't like (5), as if something bad happens, there's nothing to be done.

For QMP error injection (4) seems is overkill. It may be needed in the future if we end implementing a logic where host OS informs guest about hardware problems, and such errors use asynchronous notifications.

I would also avoid implementing (3) at least for now, as reporting such error via QMP seems enough for the QMP usecase.

So, if ok for you, I'll change the code to (2).

Quoted 23 lines
Quoted 2 lines

Agreed this write to 1 looks wrong, but the one a few lines further down (to zero it) is correct.

yep, hw should clear register.
It would be better to so on OSPM ACK, but alas we can't intercept that, so the next option would be to do that at the time when we add a new error block

Quoted 15 lines

My bug a long time back I think.

Jonathan

Quoted 9 lines
Quoted 7 lines

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

Thanks,
Mauro

#19 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Quoted 58 lines · On Fri, 13 Sep 2024 07:20:25 +0200 Mauro Carvalho Chehab…

Em Thu, 12 Sep 2024 14:42:33 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 9 lines
Quoted 6 lines · On Wed, 11 Sep 2024 16:34:36 +0100 Jonathan Cameron…
Quoted 3 lines · On Wed, 11 Sep 2024 15:21:32 +0200 Igor Mammedov <[email protected]> wrote:

On Sun, 25 Aug 2024 05:29:23 +0200
Mauro Carvalho Chehab <[email protected]> wrote:

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));          

we don't do this for SEV so, why are you setting it to 1 here?

The diffstat doesn't really help here. The full code is:

    /* zero means OSPM does not acknowledge the error */
    if (!read_ack) {
        error_setg(errp,
                   "Last CPER record was not acknowledged yet");
        read_ack = 1;
        cpu_physical_memory_write(read_ack_start_addr,
                                  &read_ack, sizeof(read_ack));
        return;
    }
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines

what you are doing here by setting read_ack = 1,
is making ack on behalf of OSPM when OSPM haven't handled existing error yet.

Essentially making HW/FW do the job of OSPM. That looks wrong to me. From HW/FW side read_ack register should be thought as read-only.

It's not read-only because HW/FW has to clear it so that HW/FW can detect when the OSPM next writes it.

By readonly, I've meant that hw shall not do above mentioned write (bad phrasing on my side).

The above code is actually an error handling condition: if for some reason errors are triggered too fast, there's a bug on QEMU or there is a bug at the OSPM, an error message is raised and the logic resets the record to a sane state. So, on a next error, OSPM will get it.

As described at https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html?highlight=asynchronous#generic-hardware-error-source:

"Some platforms may describe multiple Generic Hardware Error Source
structures with different notification types, as defined in Table 18.10. For example, a platform may describe one error source for the handling of synchronous errors (e.g. MCE or SEA), and a second source for handling asynchronous errors (e.g. SCI or External Interrupt)."

Quoted 3 lines

Basically, the error logic there seems to fit for the asynchronous case, detecting if another error happened before OSPM handles the first one.

Agreed - the error logic to act as backpressure for the tool injecting the error makes sense - it's just hardware acknowledging to paper over slow software that is an issue.

Quoted 15 lines

IMO, there are a couple of alternatives to handle such case:

1. Keep the code as-is: if this ever happens, an error message will be issued. If SEA/MCE gets implemented synchronously on HW/FW/OSPM, the above code will never be called;
2. Change the logic to do that only for asynchronous sources (currently, only if source ID is QMP);
3. Add a special QMP message to reset the notification ack. Probably would use Notification type as an input parameter;
4. Have a much more complex code to implement asynchronous notifications, with a queue to receive HEST errors and a separate thread to deliver errors to OSPM asynchronously. If we go this way, QMP would be returning the number of error messages queued, allowing error injection code to know if OSPM has troubles delivering errors;

Is this not better done in the injection code outside of qemu? So detect the error in that and if it happens back off and try again later? Basically EBUSY done in an inelegant way.

Quoted 5 lines

5. Just return an error code without doing any resets. To me, this is the worse scenario.

I don't like (5), as if something bad happens, there's nothing to be done.

If it happens on a real system nothing is done either. So I'm not sure we need to handle that. Or maybe real hardware reinjects the interrupt if the OSPM hasn't done anything about it for a while.

Quoted 9 lines

For QMP error injection (4) seems is overkill. It may be needed in the future if we end implementing a logic where host OS informs guest about hardware problems, and such errors use asynchronous notifications.

I would also avoid implementing (3) at least for now, as reporting such error via QMP seems enough for the QMP usecase.

So, if ok for you, I'll change the code to (2).

Whilst I don't feel strongly about it, I think 5 is unfortunately the correct option if we aren't going to queue errors in qemu (so make it an injection tool problem).

Quoted 30 lines
Quoted 25 lines
Quoted 4 lines

Agreed this write to 1 looks wrong, but the one a few lines further down (to zero it) is correct.

yep, hw should clear register.
It would be better to so on OSPM ACK, but alas we can't intercept that, so the next option would be to do that at the time when we add a new error block

Quoted 15 lines

My bug a long time back I think.

Jonathan

Quoted 9 lines
Quoted 7 lines

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

Thanks,
Mauro

#20 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Quoted 61 lines · On Fri, 13 Sep 2024 11:13:00 +0100 Jonathan Cameron…
Quoted 58 lines · On Fri, 13 Sep 2024 07:20:25 +0200 Mauro Carvalho Chehab…

Em Thu, 12 Sep 2024 14:42:33 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 9 lines
Quoted 6 lines · On Wed, 11 Sep 2024 16:34:36 +0100 Jonathan Cameron…
Quoted 3 lines · On Wed, 11 Sep 2024 15:21:32 +0200 Igor Mammedov <[email protected]> wrote:

On Sun, 25 Aug 2024 05:29:23 +0200
Mauro Carvalho Chehab <[email protected]> wrote:

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Em Mon, 19 Aug 2024 14:51:36 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 4 lines

Review

Quoted 3 lines
(quoted diff) +3 -0
+        read_ack = 1;
+        cpu_physical_memory_write(read_ack_start_addr,
+                                  &read_ack, (uint64_t));            

we don't do this for SEV so, why are you setting it to 1 here?

The diffstat doesn't really help here. The full code is:

    /* zero means OSPM does not acknowledge the error */
    if (!read_ack) {
        error_setg(errp,
                   "Last CPER record was not acknowledged yet");
        read_ack = 1;
        cpu_physical_memory_write(read_ack_start_addr,
                                  &read_ack, sizeof(read_ack));
        return;
    }
Quoted 11 lines
Quoted 8 lines
Quoted 5 lines

what you are doing here by setting read_ack = 1,
is making ack on behalf of OSPM when OSPM haven't handled existing error yet.

Essentially making HW/FW do the job of OSPM. That looks wrong to me. From HW/FW side read_ack register should be thought as read-only.

It's not read-only because HW/FW has to clear it so that HW/FW can detect when the OSPM next writes it.

By readonly, I've meant that hw shall not do above mentioned write (bad phrasing on my side).

The above code is actually an error handling condition: if for some reason errors are triggered too fast, there's a bug on QEMU or there is a bug at the OSPM, an error message is raised and the logic resets the record to a sane state. So, on a next error, OSPM will get it.

As described at https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html?highlight=asynchronous#generic-hardware-error-source:

"Some platforms may describe multiple Generic Hardware Error Source
structures with different notification types, as defined in Table 18.10. For example, a platform may describe one error source for the handling of synchronous errors (e.g. MCE or SEA), and a second source for handling asynchronous errors (e.g. SCI or External Interrupt)."

Quoted 7 lines
Quoted 3 lines

Basically, the error logic there seems to fit for the asynchronous case, detecting if another error happened before OSPM handles the first one.

Agreed - the error logic to act as backpressure for the tool injecting the error makes sense - it's just hardware acknowledging to paper over slow software that is an issue.

on top of that, read_ack is serving as sync primitive
If one disregards it and starts overwriting error block regardless of ack value, One will be inducing race condition, where OSPM might be accessing error_block while HW is in process of overwriting it.

Quoted 36 lines
Quoted 4 lines

IMO, there are a couple of alternatives to handle such case:

Quoted 12 lines

1. Keep the code as-is: if this ever happens, an error message will be issued. If SEA/MCE gets implemented synchronously on HW/FW/OSPM, the above code will never be called;
2. Change the logic to do that only for asynchronous sources (currently, only if source ID is QMP);
3. Add a special QMP message to reset the notification ack. Probably would use Notification type as an input parameter;
4. Have a much more complex code to implement asynchronous notifications, with a queue to receive HEST errors and a separate thread to deliver errors to OSPM asynchronously. If we go this way, QMP would be returning the number of error messages queued, allowing error injection code to know if OSPM has troubles delivering errors;

Is this not better done in the injection code outside of qemu? So detect the error in that and if it happens back off and try again later? Basically EBUSY done in an inelegant way.

Quoted 5 lines

5. Just return an error code without doing any resets. To me, this is the worse scenario.

I don't like (5), as if something bad happens, there's nothing to be done.

If it happens on a real system nothing is done either. So I'm not sure we need to handle that. Or maybe real hardware reinjects the interrupt if the OSPM hasn't done anything about it for a while.

Quoted 5 lines

For QMP error injection (4) seems is overkill. It may be needed in the future if we end implementing a logic where host OS informs guest about hardware problems, and such errors use asynchronous notifications.

Quoted 8 lines
Quoted 4 lines

I would also avoid implementing (3) at least for now, as reporting such error via QMP seems enough for the QMP usecase.

So, if ok for you, I'll change the code to (2).

Whilst I don't feel strongly about it, I think 5 is unfortunately the correct option if we aren't going to queue errors in qemu (so make it an injection tool problem).

+1 to option (5)

Quoted 31 lines
Quoted 30 lines
Quoted 25 lines
Quoted 4 lines

Agreed this write to 1 looks wrong, but the one a few lines further down (to zero it) is correct.

yep, hw should clear register.
It would be better to so on OSPM ACK, but alas we can't intercept that, so the next option would be to do that at the time when we add a new error block

Quoted 15 lines

My bug a long time back I think.

Jonathan

Quoted 9 lines
Quoted 7 lines

IMO, this is needed, independently of the notification mechanism.

Regards,
Mauro

Thanks,
Mauro

#21 lore raw

Re: [PATCH v8 06/13] acpi/ghes: add support for generic error injection via QAPI

Em Fri, 13 Sep 2024 14:28:02 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 18 lines
Quoted 18 lines
Quoted 5 lines

5. Just return an error code without doing any resets. To me, this is the worse scenario.

I don't like (5), as if something bad happens, there's nothing to be done.

If it happens on a real system nothing is done either. So I'm not sure we need to handle that. Or maybe real hardware reinjects the interrupt if the OSPM hasn't done anything about it for a while.

Quoted 8 lines

For QMP error injection (4) seems is overkill. It may be needed in the future if we end implementing a logic where host OS informs guest about hardware problems, and such errors use asynchronous notifications.

I would also avoid implementing (3) at least for now, as reporting such error via QMP seems enough for the QMP usecase.

Quoted 7 lines
Quoted 5 lines
Quoted 1 line

So, if ok for you, I'll change the code to (2).

Whilst I don't feel strongly about it, I think 5 is unfortunately the correct option if we aren't going to queue errors in qemu (so make it an injection tool problem).

+1 to option (5)

Ok, will do (5) then.

Thanks,
Mauro

#22 lore raw

PATCH v8 7/13 acpi/ghes: cleanup the memory error code logic

In reply to Mauro Carvalho Chehab

Better organize the code of the function, making it to use the raw CPER function, thus removing duplicated code.

While here, rename the function to actually reflect what it does.

---
 hw/acpi/ghes-stub.c    |   2 +-
 hw/acpi/ghes.c         | 125 +++++++++++++++--------------------------
 include/hw/acpi/ghes.h |   4 +-
 target/arm/kvm.c       |   2 +-
 4 files changed, 50 insertions(+), 83 deletions(-)
hw/acpi/ghes-stub.c +1 -1
index 8762449870b5..a60ae07a8e7c 100644
@@ -11,7 +11,7 @@
 #include "qemu/osdep.h"
 #include "hw/acpi/ghes.h"
 
-int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+int acpi_ghes_memory_errors(enum AcpiGhesNotifyType notify,
                             uint64_t physical_address)
 {
     return -1;
hw/acpi/ghes.c +46 -79

162 lines not on this page · raw message

include/hw/acpi/ghes.h +2 -2
index 6e349264fd8b..a24fe8a3bc33 100644
@@ -71,7 +71,7 @@ void acpi_build_hest(GArray *table_data, GArray *hardware_errors,
                      const char *oem_id, const char *oem_table_id);
 void acpi_ghes_add_fw_cfg(AcpiGhesState *vms, FWCfgState *s,
                           GArray *hardware_errors);
-int acpi_ghes_record_errors(enum AcpiGhesNotifyType notify,
+int acpi_ghes_memory_errors(enum AcpiGhesNotifyType notify,
                             uint64_t error_physical_addr);
 void ghes_record_cper_errors(const void *cper, size_t len,
                              enum AcpiGhesNotifyType notify, Error **errp);
@@ -80,7 +80,7 @@ void ghes_record_cper_errors(const void *cper, size_t len,
  * acpi_ghes_present: Report whether ACPI GHES table is present
  *
  * Returns: true if the system has an ACPI GHES table and it is
- * safe to call acpi_ghes_record_errors() to record a memory error.
+ * safe to call acpi_ghes_memory_errors() to record a memory error.
  */
 bool acpi_ghes_present(void);
 #endif
target/arm/kvm.c +1 -1
index 849e2e21b304..915f07376c8f 100644
@@ -2373,7 +2373,7 @@ void kvm_arch_on_sigbus_vcpu(CPUState *c, int code, void *addr)
              */
             if (code == BUS_MCEERR_AR) {
                 kvm_cpu_synchronize_state(c);
-                if (!acpi_ghes_record_errors(ACPI_HEST_SRC_ID_SEA, paddr)) {
+                if (!acpi_ghes_memory_errors(ACPI_HEST_SRC_ID_SEA, paddr)) {
                     kvm_inject_arm_sea(c);
                 } else {
                     error_report("failed to record the error");
Signature · 1 line
2.46.0
#23 lore raw

PATCH v8 8/13 docs: acpi_hest_ghes: fix documentation for CPER size

In reply to Mauro Carvalho Chehab

While the spec defines a CPER size of 4KiB for each record, currently it is set to 1KiB. Fix the documentation and add a pointer to the macro name there, as this may help to keep it updated.

---
 docs/specs/acpi_hest_ghes.rst | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)
docs/specs/acpi_hest_ghes.rst +4 -2
index 68f1fbe0a4af..c3e9f8d9a702 100644
@@ -67,8 +67,10 @@ Design Details
 (3) The address registers table contains N Error Block Address entries
     and N Read Ack Register entries. The size for each entry is 8-byte.
     The Error Status Data Block table contains N Error Status Data Block
-    entries. The size for each entry is 4096(0x1000) bytes. The total size
-    for the "etc/hardware_errors" fw_cfg blob is (N * 8 * 2 + N * 4096) bytes.
+    entries. The size for each entry is defined at the source code as
+    ACPI_GHES_MAX_RAW_DATA_LENGTH (currently 1024 bytes). The total size
+    for the "etc/hardware_errors" fw_cfg blob is
+    (N * 8 * 2 + N * ACPI_GHES_MAX_RAW_DATA_LENGTH) bytes.
     N is the number of the kinds of hardware error sources.
 
 (4) QEMU generates the ACPI linker/loader script for the firmware. The
Signature · 1 line
2.46.0
#24 lore raw

PATCH v8 9/13 scripts/ghes_inject: add a script to generate GHES error inject

In reply to Mauro Carvalho Chehab

Using the QMP GHESv2 API requires preparing a raw data array containing a CPER record.

Add a helper script with subcommands to prepare such data.

Currently, only ARM Processor error CPER record is supported.

---
 MAINTAINERS                    |   3 +
 scripts/arm_processor_error.py | 377 ++++++++++++++++++
 scripts/ghes_inject.py         |  51 +++
 scripts/qmp_helper.py          | 702 +++++++++++++++++++++++++++++++++
 4 files changed, 1133 insertions(+)
 create mode 100644 scripts/arm_processor_error.py
 create mode 100755 scripts/ghes_inject.py
 create mode 100644 scripts/qmp_helper.py
MAINTAINERS +3 -0
index 1d8091818899..249ed2858198 100644
@@ -2083,6 +2083,9 @@ S: Maintained
 F: hw/arm/ghes_cper.c
 F: hw/acpi/ghes_cper_stub.c
 F: qapi/acpi-hest.json
+F: scripts/ghes_inject.py
+F: scripts/arm_processor_error.py
+F: scripts/qmp_helper.py
 
 ppc4xx
 L: [email protected]
scripts/arm_processor_error.py new file +377 -0

377 lines not on this page · raw message

scripts/ghes_inject.py new file +51 -0

51 lines not on this page · raw message

scripts/qmp_helper.py new file +702 -0

702 lines not on this page · raw message

Signature · 1 line
2.46.0
#25 lore raw

PATCH v8 10/13 target/arm: add an experimental mpidr arm cpu property object

In reply to Mauro Carvalho Chehab

Accurately injecting an ARM Processor error ACPI/APEI GHES error record requires the value of the ARM Multiprocessor Affinity Register (mpidr).

While ARM implements it, this is currently not visible.

Add a field at CPU storing it, and place it at arm_cpu_properties as experimental, thus allowing it to be queried via QMP using qom-get function.

---
 target/arm/cpu.c    |  1 +
 target/arm/cpu.h    |  1 +
 target/arm/helper.c | 10 ++++++++--
 3 files changed, 10 insertions(+), 2 deletions(-)
target/arm/cpu.c +1 -0
index 19191c239181..30fcf0a10f46 100644
@@ -2619,6 +2619,7 @@ static ObjectClass *arm_cpu_class_by_name(const char *cpu_model)
 
 static Property arm_cpu_properties[] = {
     DEFINE_PROP_UINT64("midr", ARMCPU, midr, 0),
+    DEFINE_PROP_UINT64("x-mpidr", ARMCPU, mpidr, 0),
     DEFINE_PROP_UINT64("mp-affinity", ARMCPU,
                         mp_affinity, ARM64_AFFINITY_INVALID),
     DEFINE_PROP_INT32("node-id", ARMCPU, node_id, CPU_UNSET_NUMA_NODE_ID),
target/arm/cpu.h +1 -0
index 9a3fd595621f..3ad4de793409 100644
@@ -1033,6 +1033,7 @@ struct ArchCPU {
         uint64_t reset_pmcr_el0;
     } isar;
     uint64_t midr;
+    uint64_t mpidr;
     uint32_t revidr;
     uint32_t reset_fpsid;
     uint64_t ctr;
target/arm/helper.c +8 -2
index 0a582c1cd3b3..d6e7aa069489 100644
@@ -4690,7 +4690,7 @@ static uint64_t mpidr_read_val(CPUARMState *env)
     return mpidr;
 }
 
-static uint64_t mpidr_read(CPUARMState *env, const ARMCPRegInfo *ri)
+static uint64_t mpidr_read(CPUARMState *env)
 {
     unsigned int cur_el = arm_current_el(env);
 
@@ -4700,6 +4700,11 @@ static uint64_t mpidr_read(CPUARMState *env, const ARMCPRegInfo *ri)
     return mpidr_read_val(env);
 }
 
+static uint64_t mpidr_read_ri(CPUARMState *env, const ARMCPRegInfo *ri)
+{
+    return mpidr_read(env);
+}
+
 static const ARMCPRegInfo lpae_cp_reginfo[] = {
     /* NOP AMAIR0/1 */
     { .name = "AMAIR0", .state = ARM_CP_STATE_BOTH,
@@ -9721,7 +9726,7 @@ void register_cp_regs_for_features(ARMCPU *cpu)
             { .name = "MPIDR_EL1", .state = ARM_CP_STATE_BOTH,
               .opc0 = 3, .crn = 0, .crm = 0, .opc1 = 0, .opc2 = 5,
               .fgt = FGT_MPIDR_EL1,
-              .access = PL1_R, .readfn = mpidr_read, .type = ARM_CP_NO_RAW },
+              .access = PL1_R, .readfn = mpidr_read_ri, .type = ARM_CP_NO_RAW },
         };
 #ifdef CONFIG_USER_ONLY
         static const ARMCPRegUserSpaceInfo mpidr_user_cp_reginfo[] = {
@@ -9731,6 +9736,7 @@ void register_cp_regs_for_features(ARMCPU *cpu)
         modify_arm_cp_regs(mpidr_cp_reginfo, mpidr_user_cp_reginfo);
 #endif
         define_arm_cp_regs(cpu, mpidr_cp_reginfo);
+        cpu->mpidr = mpidr_read(env);
     }
 
     if (arm_feature(env, ARM_FEATURE_AUXCR)) {
Signature · 1 line
2.46.0
#26 lore raw

PATCH v8 11/13 scripts/arm_processor_error.py: retrieve mpidr if not filled

In reply to Mauro Carvalho Chehab

Add support to retrieve mpidr value via qom-get.

---
 scripts/arm_processor_error.py | 27 +++++++++++++++++++--------
 1 file changed, 19 insertions(+), 8 deletions(-)
scripts/arm_processor_error.py +19 -8
index 62e0c5662232..0a16d4f0d8b1 100644
@@ -5,12 +5,10 @@
 #
 # Copyright (C) 2024 Mauro Carvalho Chehab <[email protected]>
 
-# TODO: current implementation has dummy defaults.
-#
-# For a better implementation, a QMP addition/call is needed to
-# retrieve some data for ARM Processor Error injection:
-#
-#   - ARM registers: power_state, mpidr.
+# Note: currently it lacks a method to fill the ARM Processor Error CPER
+# psci field from emulation. On a real hardware, this is filled only
+# when a CPU is not running. Implementing support for it to simulate a
+# real hardware is not trivial.
 
 import argparse
 import re
@@ -174,11 +172,24 @@ def send_cper(self, args):
         else:
             cper["running-state"] = 0
 
+        if args.mpidr:
+            cper["mpidr-el1"] = arg["mpidr"]
+        elif cpus:
+            cmd_arg = {
+                'path': cpus[0],
+                'property': "x-mpidr"
+            }
+            ret = qmp_cmd.send_cmd("qom-get", cmd_arg, may_open=True)
+            if isinstance(ret, int):
+                cper["mpidr-el1"] = ret
+            else:
+                cper["mpidr-el1"] = 0
+
         if arm_valid_init:
             if args.affinity:
                 cper["valid"] |= self.arm_valid_bits["affinity"]
 
-            if args.mpidr:
+            if "mpidr-el1" in cper:
                 cper["valid"] |= self.arm_valid_bits["mpidr"]
 
             if "running-state" in cper:
@@ -362,7 +373,7 @@ def send_cper(self, args):
                 if isinstance(ret, int):
                     arg["midr-el1"] = ret
 
-        util.data_add(data, arg.get("mpidr-el1", 0), 8)
+        util.data_add(data, cper["mpidr-el1"], 8)
         util.data_add(data, arg.get("midr-el1", 0), 8)
         util.data_add(data, cper["running-state"], 4)
         util.data_add(data, arg.get("psci-state", 0), 4)
Signature · 1 line
2.46.0
#27 lore raw

PATCH v8 12/13 acpi/ghes: cleanup generic error data logic

In reply to Mauro Carvalho Chehab

Remove comments that are obvious.

No functional changes.

---
 hw/acpi/ghes.c | 38 +++++++++++++++-----------------------
 1 file changed, 15 insertions(+), 23 deletions(-)
hw/acpi/ghes.c +15 -23
index 4f7b6c5ad2b6..a822a5eafaa0 100644
@@ -130,34 +130,28 @@ static void build_ghes_hw_error_notification(GArray *table, const uint8_t type)
  * ACPI 6.1: 18.3.2.7.1 Generic Error Data
  */
 static void acpi_ghes_generic_error_data(GArray *table,
-                const uint8_t *section_type, uint32_t error_severity,
-                uint8_t validation_bits, uint8_t flags,
-                uint32_t error_data_length, QemuUUID fru_id,
-                uint64_t time_stamp)
+                                         const uint8_t *section_type,
+                                         uint32_t error_severity,
+                                         uint8_t validation_bits,
+                                         uint8_t flags,
+                                         uint32_t error_data_length,
+                                         QemuUUID fru_id,
+                                         uint64_t time_stamp)
 {
     const uint8_t fru_text[20] = {0};
 
-    /* Section Type */
     g_array_append_vals(table, section_type, 16);
-
-    /* Error Severity */
     build_append_int_noprefix(table, error_severity, 4);
+
     /* Revision */
     build_append_int_noprefix(table, 0x300, 2);
-    /* Validation Bits */
+
     build_append_int_noprefix(table, validation_bits, 1);
-    /* Flags */
     build_append_int_noprefix(table, flags, 1);
-    /* Error Data Length */
     build_append_int_noprefix(table, error_data_length, 4);
 
-    /* FRU Id */
     g_array_append_vals(table, fru_id.data, ARRAY_SIZE(fru_id.data));
-
-    /* FRU Text */
     g_array_append_vals(table, fru_text, sizeof(fru_text));
-
-    /* Timestamp */
     build_append_int_noprefix(table, time_stamp, 8);
 }
 
@@ -165,19 +159,17 @@ static void acpi_ghes_generic_error_data(GArray *table,
  * Generic Error Status Block
  * ACPI 6.1: 18.3.2.7.1 Generic Error Data
  */
-static void acpi_ghes_generic_error_status(GArray *table, uint32_t block_status,
-                uint32_t raw_data_offset, uint32_t raw_data_length,
-                uint32_t data_length, uint32_t error_severity)
+static void acpi_ghes_generic_error_status(GArray *table,
+                                           uint32_t block_status,
+                                           uint32_t raw_data_offset,
+                                           uint32_t raw_data_length,
+                                           uint32_t data_length,
+                                           uint32_t error_severity)
 {
-    /* Block Status */
     build_append_int_noprefix(table, block_status, 4);
-    /* Raw Data Offset */
     build_append_int_noprefix(table, raw_data_offset, 4);
-    /* Raw Data Length */
     build_append_int_noprefix(table, raw_data_length, 4);
-    /* Data Length */
     build_append_int_noprefix(table, data_length, 4);
-    /* Error Severity */
     build_append_int_noprefix(table, error_severity, 4);
 }
 
Signature · 1 line
2.46.0
#28 lore raw

Re: [PATCH v8 12/13] acpi/ghes: cleanup generic error data logic

Quoted 5 lines · On Fri, 16 Aug 2024 09:37:44 +0200 Mauro Carvalho Chehab…

Remove comments that are obvious.

No functional changes.

these comments help if you don't have spec side by side with code to compare. I'd even say such comments are preferable than no comments when composing an ACPI table.

pls, drop the patch

Quoted 78 lines
---
 hw/acpi/ghes.c | 38 +++++++++++++++-----------------------
 1 file changed, 15 insertions(+), 23 deletions(-)
hw/acpi/ghes.c +15 -23
index 4f7b6c5ad2b6..a822a5eafaa0 100644
@@ -130,34 +130,28 @@ static void build_ghes_hw_error_notification(GArray *table, const uint8_t type)
  * ACPI 6.1: 18.3.2.7.1 Generic Error Data
  */
 static void acpi_ghes_generic_error_data(GArray *table,
-                const uint8_t *section_type, uint32_t error_severity,
-                uint8_t validation_bits, uint8_t flags,
-                uint32_t error_data_length, QemuUUID fru_id,
-                uint64_t time_stamp)
+                                         const uint8_t *section_type,
+                                         uint32_t error_severity,
+                                         uint8_t validation_bits,
+                                         uint8_t flags,
+                                         uint32_t error_data_length,
+                                         QemuUUID fru_id,
+                                         uint64_t time_stamp)
 {
     const uint8_t fru_text[20] = {0};
 
-    /* Section Type */
     g_array_append_vals(table, section_type, 16);
-
-    /* Error Severity */
     build_append_int_noprefix(table, error_severity, 4);
+
     /* Revision */
     build_append_int_noprefix(table, 0x300, 2);
-    /* Validation Bits */
+
     build_append_int_noprefix(table, validation_bits, 1);
-    /* Flags */
     build_append_int_noprefix(table, flags, 1);
-    /* Error Data Length */
     build_append_int_noprefix(table, error_data_length, 4);
 
-    /* FRU Id */
     g_array_append_vals(table, fru_id.data, ARRAY_SIZE(fru_id.data));
-
-    /* FRU Text */
     g_array_append_vals(table, fru_text, sizeof(fru_text));
-
-    /* Timestamp */
     build_append_int_noprefix(table, time_stamp, 8);
 }
 
@@ -165,19 +159,17 @@ static void acpi_ghes_generic_error_data(GArray *table,
  * Generic Error Status Block
  * ACPI 6.1: 18.3.2.7.1 Generic Error Data
  */
-static void acpi_ghes_generic_error_status(GArray *table, uint32_t block_status,
-                uint32_t raw_data_offset, uint32_t raw_data_length,
-                uint32_t data_length, uint32_t error_severity)
+static void acpi_ghes_generic_error_status(GArray *table,
+                                           uint32_t block_status,
+                                           uint32_t raw_data_offset,
+                                           uint32_t raw_data_length,
+                                           uint32_t data_length,
+                                           uint32_t error_severity)
 {
-    /* Block Status */
     build_append_int_noprefix(table, block_status, 4);
-    /* Raw Data Offset */
     build_append_int_noprefix(table, raw_data_offset, 4);
-    /* Raw Data Length */
     build_append_int_noprefix(table, raw_data_length, 4);
-    /* Data Length */
     build_append_int_noprefix(table, data_length, 4);
-    /* Error Severity */
     build_append_int_noprefix(table, error_severity, 4);
 }
 
#29 lore raw

PATCH v8 13/13 acpi/ghes: check if the BIOS pointers for HEST are correct

In reply to Mauro Carvalho Chehab

The OS kernels navigate between HEST, error source struct and CPER by the usage of some pointers. Double-check if such pointers were properly initializing, ensuring that they match the right address for CPER.

---
 hw/acpi/ghes.c | 30 +++++++++++++++++++++++++++++-
 1 file changed, 29 insertions(+), 1 deletion(-)
hw/acpi/ghes.c +29 -1
index a822a5eafaa0..51e2e40e5a9c 100644
@@ -85,6 +85,9 @@ enum AcpiHestSourceId {
 #define HEST_GHES_V2_TABLE_SIZE  92
 #define GHES_ACK_OFFSET          (64 + GAS_ADDR_OFFSET + ACPI_HEST_HEADER_SIZE)
 
+/* ACPI 6.2: 18.3.2.7: Generic Hardware Error Source */
+#define GHES_ERR_ST_ADDR_OFFSET  (20 + GAS_ADDR_OFFSET + ACPI_HEST_HEADER_SIZE)
+
 /*
  * Values for error_severity field
  */
@@ -425,7 +428,10 @@ NotifierList acpi_generic_error_notifiers =
 void ghes_record_cper_errors(const void *cper, size_t len,
                              enum AcpiGhesNotifyType notify, Error **errp)
 {
-    uint64_t cper_addr, read_ack_start_addr;
+    uint64_t hest_read_ack_start_addr, read_ack_start_addr;
+    uint64_t read_ack_start_addr_2, err_source_struct;
+    uint64_t hest_err_block_addr, error_block_addr;
+    uint64_t cper_addr, cper_addr_2;
     enum AcpiHestSourceId source;
     AcpiGedState *acpi_ged_state;
     AcpiGhesState *ags;
@@ -450,6 +456,28 @@ void ghes_record_cper_errors(const void *cper, size_t len,
     cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
     cper_addr += source * ACPI_GHES_MAX_RAW_DATA_LENGTH;
 
+    err_source_struct = le64_to_cpu(ags->hest_addr_le) +
+                        source * HEST_GHES_V2_TABLE_SIZE;
+
+    /* Check if BIOS addr pointers were properly generated */
+
+    hest_err_block_addr = err_source_struct + GHES_ERR_ST_ADDR_OFFSET;
+    hest_read_ack_start_addr = err_source_struct + GHES_ACK_OFFSET;
+
+    cpu_physical_memory_read(hest_err_block_addr, &error_block_addr,
+                             sizeof(error_block_addr));
+
+    cpu_physical_memory_read(error_block_addr, &cper_addr_2,
+                             sizeof(error_block_addr));
+
+    cpu_physical_memory_read(hest_read_ack_start_addr, &read_ack_start_addr_2,
+			     sizeof(read_ack_start_addr_2));
+
+    assert(cper_addr == cper_addr_2);
+    assert(read_ack_start_addr == read_ack_start_addr_2);
+
+    /* Update ACK offset to notify about a new error */
+
     cpu_physical_memory_read(read_ack_start_addr,
                              &read_ack, sizeof(uint64_t));
 
Signature · 1 line
2.46.0
#30 lore raw

Re: [PATCH v8 13/13] acpi/ghes: check if the BIOS pointers for HEST are correct

Quoted 4 lines · On Fri, 16 Aug 2024 09:37:45 +0200 Mauro Carvalho Chehab…

The OS kernels navigate between HEST, error source struct and CPER by the usage of some pointers. Double-check if such pointers were properly initializing, ensuring that they match the right address for CPER.

as QEMU, we don't care about what guest wrote into those addresses (aka it's not hw businesses), even if later qemu will trample on wrong guest memory (it's guest responsibility to do init right).

However this patch introduces usage for hest_addr_le, that I was looking for. See notes below.

Quoted 33 lines
---
 hw/acpi/ghes.c | 30 +++++++++++++++++++++++++++++-
 1 file changed, 29 insertions(+), 1 deletion(-)
hw/acpi/ghes.c +7 -1
index a822a5eafaa0..51e2e40e5a9c 100644
@@ -85,6 +85,9 @@ enum AcpiHestSourceId {
 #define HEST_GHES_V2_TABLE_SIZE  92
 #define GHES_ACK_OFFSET          (64 + GAS_ADDR_OFFSET + ACPI_HEST_HEADER_SIZE)
 
+/* ACPI 6.2: 18.3.2.7: Generic Hardware Error Source */
+#define GHES_ERR_ST_ADDR_OFFSET  (20 + GAS_ADDR_OFFSET + ACPI_HEST_HEADER_SIZE)
+
 /*
  * Values for error_severity field
  */
@@ -425,7 +428,10 @@ NotifierList acpi_generic_error_notifiers =
 void ghes_record_cper_errors(const void *cper, size_t len,
                              enum AcpiGhesNotifyType notify, Error **errp)
 {
-    uint64_t cper_addr, read_ack_start_addr;
+    uint64_t hest_read_ack_start_addr, read_ack_start_addr;
+    uint64_t read_ack_start_addr_2, err_source_struct;
+    uint64_t hest_err_block_addr, error_block_addr;
+    uint64_t cper_addr, cper_addr_2;
     enum AcpiHestSourceId source;
     AcpiGedState *acpi_ged_state;
     AcpiGhesState *ags;

hw/acpi/ghes.c void ghes_record_cper_errors(const void *cper, size_t len, · line 460

Quoted 5 lines
hw/acpi/ghes.c +2 -0
@@ -450,6 +456,28 @@ void ghes_record_cper_errors(const void *cper, size_t len,
     cper_addr += ACPI_HEST_SRC_ID_COUNT * sizeof(uint64_t);
     cper_addr += source * ACPI_GHES_MAX_RAW_DATA_LENGTH;
 
+    err_source_struct = le64_to_cpu(ags->hest_addr_le) +
+                        source * HEST_GHES_V2_TABLE_SIZE;

there is no guaranties that HEST table will contain only GHESv2 sources, and once such is added this place becomes broken.

we need to iterate over HEST taking that into account and find only ghesv2 structure with source id of interest.

This function (and acpi_ghes_record_errors() as well) taking source_id as input should be able to lookup pointers from HEST in guest RAM, very crude idea could look something like this:

typedef struct hest_source_type2len{
uint16_t type
int len
} hest_structure_type2len

hest_structure_type2len supported_hest_sources[] = {
    /* Table 18-344 Generic Hardware Error Source version 2 (GHESv2) Structure */
    {.type = 10, .len = 92},
}
uint64_t find_error_source(src_id) {
    uint32_t struct_offset = hest_header_size;
    uint16_t type, id
    do {
       addr = ags->hest_addr_le + struct_offset
 
       cpu_physical_memory_read(addr, &id)
       if (src_id == id)
         return addr

       cpu_physical_memory_read(addr, &type)
       struct_offset ++= get_len_from_supported_hest_sources(type)
    while(struct_offset < hest_len)
    assert if not found
}

unit64_t get_error_status_block_addr(src_id) {
struct_addr = find_error_source(src_id)
hest_err_block_addr = struct_addr + GHES_ERR_ST_ADDR_OFFSET
// read intermediate pointer to status block addr pointer in hw table cpu_physical_memory_read(hest_err_block_addr, &error_block_addr) // read actual pointer to status block cpu_physical_memory_read(error_block_addr, &error_status_block_addr) return error_status_block_addr
}

ditto for read_ack modulo indirection that we have for error_status_block_addr

This way we can easily map source id to error status block and find needed addresses using pointer info from guest RAM without fragile pointer math and assumptions which might go wrong when new error sources are added and regardless of the order they are being added.

Quoted 22 lines
hw/acpi/ghes.c +19 -0
+    /* Check if BIOS addr pointers were properly generated */
+
+    hest_err_block_addr = err_source_struct + GHES_ERR_ST_ADDR_OFFSET;
+    hest_read_ack_start_addr = err_source_struct + GHES_ACK_OFFSET;
+
+    cpu_physical_memory_read(hest_err_block_addr, &error_block_addr,
+                             sizeof(error_block_addr));
+
+    cpu_physical_memory_read(error_block_addr, &cper_addr_2,
+                             sizeof(error_block_addr));
+
+    cpu_physical_memory_read(hest_read_ack_start_addr, &read_ack_start_addr_2,
+			     sizeof(read_ack_start_addr_2));
+
+    assert(cper_addr == cper_addr_2);
+    assert(read_ack_start_addr == read_ack_start_addr_2);
+
+    /* Update ACK offset to notify about a new error */
+
     cpu_physical_memory_read(read_ack_start_addr,
                              &read_ack, sizeof(uint64_t));
 
#31 lore raw

Re: [PATCH v8 13/13] acpi/ghes: check if the BIOS pointers for HEST are correct

Em Mon, 19 Aug 2024 16:07:33 +0200
Igor Mammedov <[email protected]> escreveu:

Quoted 18 lines

hw/acpi/ghes.c void ghes_record_cper_errors(const void *cper, size_t len, · line 460

Quoted 2 lines
hw/acpi/ghes.c +2 -0
+    err_source_struct = le64_to_cpu(ags->hest_addr_le) +
+                        source * HEST_GHES_V2_TABLE_SIZE;  

there is no guaranties that HEST table will contain only GHESv2 sources, and once such is added this place becomes broken.

we need to iterate over HEST taking that into account and find only ghesv2 structure with source id of interest.

This function (and acpi_ghes_record_errors() as well) taking source_id as input should be able to lookup pointers from HEST in guest RAM, very crude idea could look something like this:

typedef struct hest_source_type2len{
uint16_t type
int len
} hest_structure_type2len

Quoted 4 lines
hest_structure_type2len supported_hest_sources[] = {
    /* Table 18-344 Generic Hardware Error Source version 2 (GHESv2) Structure */
    {.type = 10, .len = 92},
}

Sounds interesting, but IMO it should be done only when other types besides ghes would be added, as:

1. Right now, the file is acpi/ghes.c. Adding non-type 10 HEST structures there would be a little weird. It should likely be renamed to acpi/hest.c when such time comes.

2. ACPI 6.5 has made clear that the above will only work up to type 11, as, from type 12 and above, the length will be added to the error struct, according with:

https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#error-source-structure-header-type-12-onward

3. some types have variable size. Starting from the beginning, type 0, as defined at:
https://uefi.org/specs/ACPI/6.5/18_Platform_Error_Interfaces.html#hardware-errors-and-error-sources

has:

size = 40 + 24 * Number of Hardware banks

So, a simple table like the above with fixed sizes won't work.

The code would need instead a switch if types are <= 11.

Adding proper support for all already defined 12 types sounds lots of work, as the code would need to calculate the size depending on the size, and we don't really initialize the HEST table with other types but GHES.

Ok, we could still do something like this pseudo-code to get the error source offset:

	#define ACPI_HEST_TYPE_GHESV2	11

	err_struct_offset = 0;
	for (i = 0; i < source_id_count; i++) {
		/* NOTE: Other types may have different sizes */
		assert(ghes[i].type == ACPI_HEST_TYPE_GHESV2);
		if (ghes[i].source_id == source_id)
			break;
		err_struct_offset += HEST_GHES_V2_TABLE_SIZE;
	}
	assert (i < source_id_count);

---

That's said, maybe this will just add unwanted complexity, as QEMU is already setting those offsets via bios_linker_loader_add_pointer().

So, an alternative for that is to merge the code on patch 13 with the one on patch 5, dropping the math calcus there and relying that QEMU will always handle properly bios links.

See, the logic which constructs GHESv2 source IDs do this to create the links between HEST ACPI table and etc/hardware_errors:

with:

Per-source ID logic at build_ghes_v2():

    address_offset = table_data->len;
    /* Error Status Address */
    build_append_gas(table_data, AML_AS_SYSTEM_MEMORY, 0x40, 0,
                     4 /* QWord access */, 0);
    bios_linker_loader_add_pointer(linker, ACPI_BUILD_TABLE_FILE,
                                   address_offset + GAS_ADDR_OFFSET,
                                   sizeof(uint64_t),
                                   ACPI_HW_ERROR_FW_CFG_FILE,
                                   source_id * sizeof(uint64_t));
...
    /*
     * Read Ack Register
     * ACPI 6.1: 18.3.2.8 Generic Hardware Error Source
     * version 2 (GHESv2 - Type 10)
     */
    address_offset = table_data->len;
    build_append_gas(table_data, AML_AS_SYSTEM_MEMORY, 0x40, 0,
                     4 /* QWord access */, 0);
    bios_linker_loader_add_pointer(linker, ACPI_BUILD_TABLE_FILE,
                                   address_offset + GAS_ADDR_OFFSET,
                                   sizeof(uint64_t),
                                   ACPI_HW_ERROR_FW_CFG_FILE,
                                   (ACPI_HEST_SRC_ID_COUNT + source_id) *
                                   sizeof(uint64_t));

HEST table creation logic inside build_ghes_error_table():

    for (i = 0; i < ACPI_HEST_SRC_ID_COUNT; i++) {
        /*
         * Tell firmware to patch error_block_address entries to point to
         * corresponding "Generic Error Status Block"
         */
        bios_linker_loader_add_pointer(linker,
            ACPI_HW_ERROR_FW_CFG_FILE, sizeof(uint64_t) * i,
            sizeof(uint64_t), ACPI_HW_ERROR_FW_CFG_FILE,
            error_status_block_offset + i * ACPI_GHES_MAX_RAW_DATA_LENGTH);
    }

Using those, the location of the CPER and ack addresses is easy and won't require any math:

	/* GHESv2 CPER offset */
	cpu_physical_memory_read(hest_err_block_addr, &error_block_addr,
                                 sizeof(error_block_addr));
	cpu_physical_memory_read(error_block_addr, &cper_addr,
                                 sizeof(error_block_addr));

	/* GHESv2 ack offset */
	cpu_physical_memory_read(hest_read_ack_start_addr, &read_ack_start_addr,
			         sizeof(read_ack_start_addr));

Regards,
Mauro

#32 lore raw

Re: [PATCH v8 13/13] acpi/ghes: check if the BIOS pointers for HEST are correct

Em Sat, 24 Aug 2024 02:15:10 +0200
Mauro Carvalho Chehab <[email protected]> escreveu:

Quoted 14 lines

Ok, we could still do something like this pseudo-code to get the error source offset:

	#define ACPI_HEST_TYPE_GHESV2	11

	err_struct_offset = 0;
	for (i = 0; i < source_id_count; i++) {
		/* NOTE: Other types may have different sizes */
		assert(ghes[i].type == ACPI_HEST_TYPE_GHESV2);
		if (ghes[i].source_id == source_id)
			break;
		err_struct_offset += HEST_GHES_V2_TABLE_SIZE;
	}
	assert (i < source_id_count);

This is what I ended implementing on v9.

Regards,
Mauro

#33 lore raw
In reply to Mauro Carvalho Chehab
Quoted 10 lines · On Fri, 16 Aug 2024 09:37:32 +0200 Mauro Carvalho Chehab…

This series add support for injecting generic CPER records. Such records are generated outside QEMU via a provided script.

On this version, I added two optional patches at the end:
- acpi/ghes: cleanup generic error data logic

It drops some obvious comments from some already-existing code. As we're already doing lots of changes at the code, it sounded reasonable to me to have such cleanup here;

Quoted 7 lines

- acpi/ghes: check if the BIOS pointers for HEST are correct

QEMU has two ways to navigate to a CPER start data: via its memory address or indirectly following 2 BIOS pointers. OS only have the latter one. This patch validates if the BIOS links used by the OS were properly produced, comparing to the actual location of the CPER record.

I went over the series,
once suggestion in 13/13 implemented
we can get rid of pointer math that is reshuffled several times in patches here.

I'd suggest to structure series as following:

1: patch that adds hest_addr_le
2: refactoring current code to use address lookup vs pointer math
3. renaming patches
4. patch adding new error source
5. QAPI patch
6. python script for error injection

with that in place we probably would need to
* iron out minor migration compat issues
(I didn't look for them during this review round as much
would change yet)
* make sure that bios tables test is updated

Quoted 159 lines

---

v8:
- Fix one of the BIOS links that were incorrect;
- Changed mem error internal injection to use a common code;
- No more hardcoded values for CPER: instead of using just the payload at the QAPI, it now has the full raw CPER there;
- Error injection script now supports changing fields at the Generic Error Data section of the CPER;
- Several minor cleanups.

v7:
- Change the way offsets are calculated and used on HEST table. Now, it is compatible with migrations as all offsets are relative to the HEST table;
- GHES interface is now more generic: the entire CPER is sent via QMP, instead of just the payload;
- Some code cleanups to make the code more robust;
- The python script now uses QEMUMonitorProtocol class.

v6:
- PNP0C33 device creation moved to aml-build.c;
- acpi_ghes record functions now use ACPI notify parameter, instead of source ID;
- the number of source IDs is now automatically calculated;
- some code cleanups and function/var renames;
- some fixes and cleanups at the error injection script;
- ghes cper stub now produces an error if cper JSON is not compiled;
- Offset calculation logic for GHES was refactored;
- Updated documentation to reflect the GHES allocated size;
- Added a x-mpidr object for QOM usage;
- Added a patch making usage of x-mpidr field at ARM injection script;

v5:
- CPER guid is now passing as string;
- raw-data is now passed with base64 encode;
- Removed several GPIO left-overs from arm/virt.c changes;
- Lots of cleanups and improvements at the error injection script. It now better handles QMP dialog and doesn't print debug messages. Also, code was split on two modules, to make easier to add more error injection commands.

v4:
- CPER generation moved to happen outside QEMU;
- One patch adding support for mpidr query was removed.

v3:
- patch 1 cleanups with some comment changes and adding another place where the poweroff GPIO define should be used. No changes on other patches (except due to conflict resolution).

v2:
- added a new patch using a define for GPIO power pin;
- patch 2 changed to also use a define for generic error GPIO pin;
- a couple cleanups at patch 2 removing uneeded else clauses.

Example of generating a CPER record:

$ scripts/ghes_inject.py -d arm -p 0xdeadbeef
GUID: e19e3d16-bc11-11e4-9caa-c2051d5d46b0
Generic Error Status Block (20 bytes):
      00000000  01 00 00 00 00 00 00 00 00 00 00 00 90 00 00 00   ................
      00000010  00 00 00 00                                       ....

Generic Error Data Entry (72 bytes):

      00000000  16 3d 9e e1 11 bc e4 11 9c aa c2 05 1d 5d 46 b0   .=...........]F.
      00000010  00 00 00 00 00 03 00 00 48 00 00 00 00 00 00 00   ........H.......
      00000020  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00   ................
      00000030  00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00   ................
      00000040  00 00 00 00 00 00 00 00                           ........

Payload (72 bytes):

      00000000  05 00 00 00 01 00 00 00 48 00 00 00 00 00 00 00   ........H.......
      00000010  00 00 00 80 00 00 00 00 10 05 0f 00 00 00 00 00   ................
      00000020  00 00 00 00 00 00 00 00 00 20 14 00 02 01 00 03   ......... ......
      00000030  0f 00 91 00 00 00 00 00 ef be ad de 00 00 00 00   ................
      00000040  ef be ad de 00 00 00 00                           ........

Error injected.

[    9.358364] {1}[Hardware Error]: Hardware error from APEI Generic Hardware Error Source: 1
[    9.359027] {1}[Hardware Error]: event severity: recoverable
[    9.359586] {1}[Hardware Error]:  Error 0, type: recoverable
[    9.360124] {1}[Hardware Error]:   section_type: ARM processor error
[    9.360561] {1}[Hardware Error]:   MIDR: 0x00000000000f0510
[    9.361160] {1}[Hardware Error]:   Multiprocessor Affinity Register (MPIDR): 0x0000000080000000
[    9.361643] {1}[Hardware Error]:   running state: 0x0
[    9.362142] {1}[Hardware Error]:   Power State Coordination Interface state: 0
[    9.362682] {1}[Hardware Error]:   Error info structure 0:
[    9.363030] {1}[Hardware Error]:   num errors: 2
[    9.363656] {1}[Hardware Error]:    error_type: 0x02: cache error
[    9.364163] {1}[Hardware Error]:    error_info: 0x000000000091000f
[    9.364834] {1}[Hardware Error]:     transaction type: Data Access
[    9.365599] {1}[Hardware Error]:     cache error, operation type: Data write
[    9.366441] {1}[Hardware Error]:     cache level: 2
[    9.367005] {1}[Hardware Error]:     processor context not corrupted
[    9.367753] {1}[Hardware Error]:    physical fault address: 0x00000000deadbeef
[    9.374267] Memory failure: 0xdeadb: recovery action for free buddy page: Recovered

Such script currently supports arm processor error CPER, but can easily be extended to other GHES notification types.

Jonathan Cameron (1):
acpi/ghes: Add support for GED error device

Mauro Carvalho Chehab (12):
acpi/generic_event_device: add an APEI error device arm/virt: Wire up a GED error device for ACPI / GHES qapi/acpi-hest: add an interface to do generic CPER error injection acpi/ghes: rework the logic to handle HEST source ID acpi/ghes: add support for generic error injection via QAPI acpi/ghes: cleanup the memory error code logic
docs: acpi_hest_ghes: fix documentation for CPER size scripts/ghes_inject: add a script to generate GHES error inject target/arm: add an experimental mpidr arm cpu property object scripts/arm_processor_error.py: retrieve mpidr if not filled acpi/ghes: cleanup generic error data logic
acpi/ghes: check if the BIOS pointers for HEST are correct

 MAINTAINERS                            |  10 +
 docs/specs/acpi_hest_ghes.rst          |   6 +-
 hw/acpi/Kconfig                        |   5 +
 hw/acpi/aml-build.c                    |  10 +
 hw/acpi/generic_event_device.c         |   8 +
 hw/acpi/ghes-stub.c                    |   3 +-
 hw/acpi/ghes.c                         | 362 ++++++++-----
 hw/acpi/ghes_cper.c                    |  33 ++
 hw/acpi/ghes_cper_stub.c               |  19 +
 hw/acpi/meson.build                    |   2 +
 hw/arm/Kconfig                         |   5 +
 hw/arm/virt-acpi-build.c               |   6 +-
 hw/arm/virt.c                          |  12 +-
 include/hw/acpi/acpi_dev_interface.h   |   1 +
 include/hw/acpi/aml-build.h            |   2 +
 include/hw/acpi/generic_event_device.h |   1 +
 include/hw/acpi/ghes.h                 |  24 +-
 include/hw/arm/virt.h                  |   1 +
 qapi/acpi-hest.json                    |  36 ++
 qapi/meson.build                       |   1 +
 qapi/qapi-schema.json                  |   1 +
 scripts/arm_processor_error.py         | 388 ++++++++++++++
 scripts/ghes_inject.py                 |  51 ++
 scripts/qmp_helper.py                  | 702 +++++++++++++++++++++++++
 target/arm/cpu.c                       |   1 +
 target/arm/cpu.h                       |   1 +
 target/arm/helper.c                    |  10 +-
 target/arm/kvm.c                       |   2 +-
 28 files changed, 1551 insertions(+), 152 deletions(-)
 create mode 100644 hw/acpi/ghes_cper.c
 create mode 100644 hw/acpi/ghes_cper_stub.c
 create mode 100644 qapi/acpi-hest.json
 create mode 100644 scripts/arm_processor_error.py
 create mode 100755 scripts/ghes_inject.py
 create mode 100644 scripts/qmp_helper.py