Skip to contentSkip to conversation

pds_vfio driver

PATCH Superseded Brett Creeley 5 participants 21 messages Apr 4, 2023 – Apr 21, 2023

Subsystems (9)

v8 of 14 Superseded Latest is v14 →

lore
Share & tools
lore

Brett Creeley

author maintainer
#0 lore raw

v8 cover pds_vfio driver

This is a patchset for a new vendor specific VFIO driver (pds_vfio) for use with the AMD/Pensando Distributed Services Card (DSC). This driver makes use of the newly introduced pds_core driver, which the latest version can be referenced at:

https://lore.kernel.org/netdev/[email protected]/

This driver will use the pds_core device's adminq as the VFIO control path to the DSC. In order to make adminq calls, the VFIO instance makes use of functions exported by the pds_core driver.

In order to receive events from pds_core, the pds_vfio driver registers to a private notifier. This is needed for various events that come from the device.

An ASCII diagram of a VFIO instance looks something like this and can be used with the VFIO subsystem to provide the VF device VFIO and live migration support.

                               .------.  .-----------------------.
                               | QEMU |--|  VM  .-------------.  |
                               '......'  |      |   Eth VF    |  |
                                  |      |      .-------------.  |
                                  |      |      |  SR-IOV VF  |  |
                                  |      |      '-------------'  |
                                  |      '------------||---------'
                               .--------------.       ||
                               |/dev/<vfio_fd>|       ||
                               '--------------'       ||
Host Userspace                         |              ||
===================================================   ||
Host Kernel                            |              ||
                                  .--------.          ||
                                  |vfio-pci|          ||
                                  '--------'          ||
       .------------------.           ||              ||
       |   | exported API |<----+     ||              ||
       |   '--------------|     |     ||              ||
       |                  |    .-------------.        ||
       |     pds_core     |--->|   pds_vfio  |        ||
       '------------------' |  '-------------'        ||
               ||           |         ||              ||
             09:00.0     notifier    09:00.1          ||
== PCI ===============================================||=====
               ||                     ||              ||
          .----------.          .----------.          ||
    ,-----|    PF    |----------|    VF    |-------------------,
    |     '----------'          '----------'  |       VF       |
    |                     DSC                 |  data/control  |
    |                                         |      path      |
    -----------------------------------------------------------

The pds_vfio driver is targeted to reside in drivers/vfio/pci/pds. It makes use of and introduces new files in the common include/linux/pds include directory.

Changes:

v8:
- provide default iommufd callbacks for bind_iommufd, unbind_iommufd, and attach_ioas for the VFIO device as suggested by Shameerali Kolothum Thodi.

v7:
https://lore.kernel.org/netdev/[email protected]/
- Disable and clean up dirty page tracking when the VFIO device is closed
- Various improvements suggested by Simon Horman:
- Fix RCT in vfio_combine_iova_ranges()
- Simplify function exit paths by removing unnecessary goto labels
- Cleanup pds_vifo_print_guest_region_info() by adding a goto label for freeing memory, which allowed for reduced indentation on a for loop
- Where possible use C99 style for loops

v6:
https://lore.kernel.org/netdev/[email protected]/
- As suggested by Alex Williamson, use pci_domain_nr() macro to make sure the pds_vfio client's devname is unique
- Remove unnecessary forward declaration and include
- Fix copyright comment to use correct company name
- Remove "." from struct documentation for consistency

v5:
https://lore.kernel.org/netdev/[email protected]/
- Fix SPDX comments in .h files
- Remove adminqcq argument from pdsc_post_adminq() uses
- Unregister client on vfio_pci_core_register_device() failure
- Other minor checkpatch issues

v4:
https://lore.kernel.org/netdev/[email protected]/
- Update cover letter ASCII diagram to reflect new driver architecture
- Remove auxiliary driver implementation
- Use pds_core's exported functions to communicate with the device
- Implement and register notifier for events from the device/pds_core
- Use module_pci_driver() macro since auxiliary driver configuration is no longer needed in __init/__exit

v3:
https://lore.kernel.org/netdev/[email protected]/
- Update copyright year to 2023 and use "Advanced Micro Devices, Inc." for the company name
- Clarify the fact that AMD/Pensando's VFIO solution is device type agnostic, which aligns with other current VFIO solutions
- Add line in drivers/vfio/pci/Makefile to build pds_vfio
- Move documentation to amd sub-directory
- Remove some dead code due to the pds_core implementation of listening to BIND/UNBIND events
- Move a dev_dbg() to a previous patch in the series
- Add implementation for vfio_migration_ops.migration_get_data_size to return the maximum possible device state size

RFC to v2:
https://lore.kernel.org/all/[email protected]/
- Implement state transitions for VFIO_MIGRATION_P2P flag
- Improve auxiliary driver probe by returning EPROBE_DEFER when the PCI driver is not set up correctly
- Add pointer to docs in Documentation/networking/device_drivers/ethernet/index.rst

RFC:
https://lore.kernel.org/all/[email protected]/

Brett Creeley (7):
vfio: Commonize combine_ranges for use in other VFIO drivers vfio/pds: Initial support for pds_vfio VFIO driver vfio/pds: register with the pds_core PF
vfio/pds: Add VFIO live migration support vfio/pds: Add support for dirty page tracking vfio/pds: Add support for firmware recovery vfio/pds: Add Kconfig and documentation

 .../device_drivers/ethernet/amd/pds_vfio.rst  |  79 +++
 .../device_drivers/ethernet/index.rst         |   1 +
 MAINTAINERS                                   |   7 +
 drivers/vfio/pci/Kconfig                      |   2 +
 drivers/vfio/pci/Makefile                     |   2 +
 drivers/vfio/pci/mlx5/cmd.c                   |  48 +-
 drivers/vfio/pci/pds/Kconfig                  |  20 +
 drivers/vfio/pci/pds/Makefile                 |  11 +
 drivers/vfio/pci/pds/cmds.c                   | 530 +++++++++++++++++
 drivers/vfio/pci/pds/cmds.h                   |  25 +
 drivers/vfio/pci/pds/dirty.c                  | 536 ++++++++++++++++++
 drivers/vfio/pci/pds/dirty.h                  |  45 ++
 drivers/vfio/pci/pds/lm.c                     | 478 ++++++++++++++++
 drivers/vfio/pci/pds/lm.h                     |  52 ++
 drivers/vfio/pci/pds/pci_drv.c                | 212 +++++++
 drivers/vfio/pci/pds/pci_drv.h                |   9 +
 drivers/vfio/pci/pds/vfio_dev.c               | 239 ++++++++
 drivers/vfio/pci/pds/vfio_dev.h               |  40 ++
 drivers/vfio/vfio_main.c                      |  47 ++
 include/linux/pds/pds_lm.h                    | 389 +++++++++++++
 include/linux/vfio.h                          |   3 +
 21 files changed, 2728 insertions(+), 47 deletions(-)
 create mode 100644 Documentation/networking/device_drivers/ethernet/amd/pds_vfio.rst
 create mode 100644 drivers/vfio/pci/pds/Kconfig
 create mode 100644 drivers/vfio/pci/pds/Makefile
 create mode 100644 drivers/vfio/pci/pds/cmds.c
 create mode 100644 drivers/vfio/pci/pds/cmds.h
 create mode 100644 drivers/vfio/pci/pds/dirty.c
 create mode 100644 drivers/vfio/pci/pds/dirty.h
 create mode 100644 drivers/vfio/pci/pds/lm.c
 create mode 100644 drivers/vfio/pci/pds/lm.h
 create mode 100644 drivers/vfio/pci/pds/pci_drv.c
 create mode 100644 drivers/vfio/pci/pds/pci_drv.h
 create mode 100644 drivers/vfio/pci/pds/vfio_dev.c
 create mode 100644 drivers/vfio/pci/pds/vfio_dev.h
 create mode 100644 include/linux/pds/pds_lm.h
Signature · 1 line
2.17.1
author maintainer
#1 lore raw

PATCH v8 1/7 vfio: Commonize combine_ranges for use in other VFIO drivers

Currently only Mellanox uses the combine_ranges function. The new pds_vfio driver also needs this function. So, move it to a common location for other vendor drivers to use.

Also, Simon Horman noticed that RCT ordering was not followed for vfio_combin_iova_ranges(), so fix that.

---
 drivers/vfio/pci/mlx5/cmd.c | 48 +------------------------------------
 drivers/vfio/vfio_main.c    | 47 ++++++++++++++++++++++++++++++++++++
 include/linux/vfio.h        |  3 +++
 3 files changed, 51 insertions(+), 47 deletions(-)
drivers/vfio/pci/mlx5/cmd.c +1 -47
index deed156e6165..7f6c51992a15 100644
@@ -732,52 +732,6 @@ void mlx5fv_cmd_clean_migf_resources(struct mlx5_vf_migration_file *migf)
 	mlx5vf_cmd_dealloc_pd(migf);
 }
 
-static void combine_ranges(struct rb_root_cached *root, u32 cur_nodes,
-			   u32 req_nodes)
-{
-	struct interval_tree_node *prev, *curr, *comb_start, *comb_end;
-	unsigned long min_gap;
-	unsigned long curr_gap;
-
-	/* Special shortcut when a single range is required */
-	if (req_nodes == 1) {
-		unsigned long last;
-
-		curr = comb_start = interval_tree_iter_first(root, 0, ULONG_MAX);
-		while (curr) {
-			last = curr->last;
-			prev = curr;
-			curr = interval_tree_iter_next(curr, 0, ULONG_MAX);
-			if (prev != comb_start)
-				interval_tree_remove(prev, root);
-		}
-		comb_start->last = last;
-		return;
-	}
-
-	/* Combine ranges which have the smallest gap */
-	while (cur_nodes > req_nodes) {
-		prev = NULL;
-		min_gap = ULONG_MAX;
-		curr = interval_tree_iter_first(root, 0, ULONG_MAX);
-		while (curr) {
-			if (prev) {
-				curr_gap = curr->start - prev->last;
-				if (curr_gap < min_gap) {
-					min_gap = curr_gap;
-					comb_start = prev;
-					comb_end = curr;
-				}
-			}
-			prev = curr;
-			curr = interval_tree_iter_next(curr, 0, ULONG_MAX);
-		}
-		comb_start->last = comb_end->last;
-		interval_tree_remove(comb_end, root);
-		cur_nodes--;
-	}
-}
-
 static int mlx5vf_create_tracker(struct mlx5_core_dev *mdev,
 				 struct mlx5vf_pci_core_device *mvdev,
 				 struct rb_root_cached *ranges, u32 nnodes)
@@ -800,7 +754,7 @@ static int mlx5vf_create_tracker(struct mlx5_core_dev *mdev,
 	int i;
 
 	if (num_ranges > max_num_range) {
-		combine_ranges(ranges, nnodes, max_num_range);
+		vfio_combine_iova_ranges(ranges, nnodes, max_num_range);
 		num_ranges = max_num_range;
 	}
 
drivers/vfio/vfio_main.c +47 -0

53 lines not on this page · raw message

include/linux/vfio.h +3 -0
index 93134b023968..10db5aa3e985 100644
@@ -241,6 +241,9 @@ int vfio_mig_get_next_state(struct vfio_device *device,
 			    enum vfio_device_mig_state new_fsm,
 			    enum vfio_device_mig_state *next_fsm);
 
+void vfio_combine_iova_ranges(struct rb_root_cached *root, u32 cur_nodes,
+			      u32 req_nodes);
+
 /*
  * External user API
  */
Signature · 1 line
2.17.1
reviewer Reviewed-by on v8 1/7
#2 lore raw

Re: [PATCH v8 vfio 1/7] vfio: Commonize combine_ranges for use in other VFIO drivers

Quoted 11 lines · Brett Creeley

Currently only Mellanox uses the combine_ranges function. The new pds_vfio driver also needs this function. So, move it to a common location for other vendor drivers to use.

Also, Simon Horman noticed that RCT ordering was not followed for vfio_combin_iova_ranges(), so fix that.

Quoted 5 lines
---
 drivers/vfio/pci/mlx5/cmd.c | 48 +------------------------------------
 drivers/vfio/vfio_main.c    | 47 ++++++++++++++++++++++++++++++++++++
 include/linux/vfio.h        |  3 +++
 3 files changed, 51 insertions(+), 47 deletions(-)

Jason

author maintainer
#3 lore raw

PATCH v8 2/7 vfio/pds: Initial support for pds_vfio VFIO driver

In reply to Brett Creeley

This is the initial framework for the new pds_vfio device driver. This does the very basics of registering the PDS PCI device and configuring it as a VFIO PCI device.

With this change, the VF device can be bound to the pds_vfio driver on the host and presented to the VM as the VF's device type.

---
 drivers/vfio/pci/Makefile       |  2 +
 drivers/vfio/pci/pds/Makefile   |  8 ++++
 drivers/vfio/pci/pds/pci_drv.c  | 74 +++++++++++++++++++++++++++++++
 drivers/vfio/pci/pds/vfio_dev.c | 77 +++++++++++++++++++++++++++++++++
 drivers/vfio/pci/pds/vfio_dev.h | 21 +++++++++
 5 files changed, 182 insertions(+)
 create mode 100644 drivers/vfio/pci/pds/Makefile
 create mode 100644 drivers/vfio/pci/pds/pci_drv.c
 create mode 100644 drivers/vfio/pci/pds/vfio_dev.c
 create mode 100644 drivers/vfio/pci/pds/vfio_dev.h
drivers/vfio/pci/Makefile +2 -0
index 24c524224da5..45167be462d8 100644
@@ -11,3 +11,5 @@ obj-$(CONFIG_VFIO_PCI) += vfio-pci.o
 obj-$(CONFIG_MLX5_VFIO_PCI)           += mlx5/
 
 obj-$(CONFIG_HISI_ACC_VFIO_PCI) += hisilicon/
+
+obj-$(CONFIG_PDS_VFIO_PCI) += pds/
drivers/vfio/pci/pds/Makefile new file +8 -0
new file mode 100644
index 000000000000..e1a55ae0f079
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2023 Advanced Micro Devices, Inc.
+
+obj-$(CONFIG_PDS_VFIO_PCI) += pds_vfio.o
+
+pds_vfio-y := \
+	pci_drv.o	\
+	vfio_dev.o
drivers/vfio/pci/pds/pci_drv.c new file +74 -0

74 lines not on this page · raw message

drivers/vfio/pci/pds/vfio_dev.c new file +77 -0

77 lines not on this page · raw message

drivers/vfio/pci/pds/vfio_dev.h new file +21 -0
new file mode 100644
index 000000000000..a66f8069b88c
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _VFIO_DEV_H_
+#define _VFIO_DEV_H_
+
+#include <linux/pci.h>
+#include <linux/vfio_pci_core.h>
+
+struct pds_vfio_pci_device {
+	struct vfio_pci_core_device vfio_coredev;
+	struct pci_dev *pdev;
+
+	int vf_id;
+	int pci_id;
+};
+
+const struct vfio_device_ops *pds_vfio_ops_info(void);
+struct pds_vfio_pci_device *pds_vfio_pci_drvdata(struct pci_dev *pdev);
+
+#endif /* _VFIO_DEV_H_ */
Signature · 1 line
2.17.1
author maintainer
#4 lore raw

PATCH v8 3/7 vfio/pds: register with the pds_core PF

In reply to Brett Creeley

The pds_core driver will supply adminq services, so find the PF and register with the DSC services.

Use the following commands to enable a VF:
echo 1 > /sys/bus/pci/drivers/pds_core/$PF_BDF/sriov_numvfs

---
 drivers/vfio/pci/pds/Makefile   |  1 +
 drivers/vfio/pci/pds/cmds.c     | 69 +++++++++++++++++++++++++++++++++
 drivers/vfio/pci/pds/cmds.h     | 10 +++++
 drivers/vfio/pci/pds/pci_drv.c  | 16 +++++++-
 drivers/vfio/pci/pds/pci_drv.h  |  9 +++++
 drivers/vfio/pci/pds/vfio_dev.c |  5 +++
 drivers/vfio/pci/pds/vfio_dev.h |  2 +
 include/linux/pds/pds_lm.h      | 12 ++++++
 8 files changed, 123 insertions(+), 1 deletion(-)
 create mode 100644 drivers/vfio/pci/pds/cmds.c
 create mode 100644 drivers/vfio/pci/pds/cmds.h
 create mode 100644 drivers/vfio/pci/pds/pci_drv.h
 create mode 100644 include/linux/pds/pds_lm.h
drivers/vfio/pci/pds/Makefile +1 -0
index e1a55ae0f079..87581111fa17 100644
@@ -4,5 +4,6 @@
 obj-$(CONFIG_PDS_VFIO_PCI) += pds_vfio.o
 
 pds_vfio-y := \
+	cmds.o		\
 	pci_drv.o	\
 	vfio_dev.o
drivers/vfio/pci/pds/cmds.c new file +69 -0

69 lines not on this page · raw message

drivers/vfio/pci/pds/cmds.h new file +10 -0
new file mode 100644
index 000000000000..4c592afccf89
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _CMDS_H_
+#define _CMDS_H_
+
+int pds_vfio_register_client_cmd(struct pds_vfio_pci_device *pds_vfio);
+void pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio);
+
+#endif /* _CMDS_H_ */
drivers/vfio/pci/pds/pci_drv.c +15 -1
index 5e554420792e..46537afdee2d 100644
@@ -8,9 +8,13 @@
 #include <linux/types.h>
 #include <linux/vfio.h>
 
+#include <linux/pds/pds_common.h>
 #include <linux/pds/pds_core_if.h>
+#include <linux/pds/pds_adminq.h>
 
 #include "vfio_dev.h"
+#include "pci_drv.h"
+#include "cmds.h"
 
 #define PDS_VFIO_DRV_NAME		"pds_vfio"
 #define PDS_VFIO_DRV_DESCRIPTION	"AMD/Pensando VFIO Device Driver"
@@ -30,13 +34,23 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
 	dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
 	pds_vfio->pdev = pdev;
+	pds_vfio->pdsc = pdsc_get_pf_struct(pdev);
+
+	err = pds_vfio_register_client_cmd(pds_vfio);
+	if (err) {
+		dev_err(&pdev->dev, "failed to register as client: %pe\n",
+			ERR_PTR(err));
+		goto out_put_vdev;
+	}
 
 	err = vfio_pci_core_register_device(&pds_vfio->vfio_coredev);
 	if (err)
-		goto out_put_vdev;
+		goto out_unreg_client;
 
 	return 0;
 
+out_unreg_client:
+	pds_vfio_unregister_client_cmd(pds_vfio);
 out_put_vdev:
 	vfio_put_device(&pds_vfio->vfio_coredev.vdev);
 	return err;
drivers/vfio/pci/pds/pci_drv.h new file +9 -0
new file mode 100644
index 000000000000..e79bed12ed14
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _PCI_DRV_H
+#define _PCI_DRV_H
+
+#include <linux/pci.h>
+
+#endif /* _PCI_DRV_H */
drivers/vfio/pci/pds/vfio_dev.c +5 -0
index 0f70efec01e1..056715dea512 100644
@@ -31,6 +31,11 @@ pds_vfio_init_device(struct vfio_device *vdev)
 	pds_vfio->vf_id = pci_iov_vf_id(pdev);
 	pds_vfio->pci_id = PCI_DEVID(pdev->bus->number, pdev->devfn);
 
+	dev_dbg(&pdev->dev, "%s: PF %#04x VF %#04x (%d) vf_id %d domain %d pds_vfio %p\n",
+		__func__, pci_dev_id(pdev->physfn),
+		pds_vfio->pci_id, pds_vfio->pci_id, pds_vfio->vf_id,
+		pci_domain_nr(pdev->bus), pds_vfio);
+
 	return 0;
 }
 
drivers/vfio/pci/pds/vfio_dev.h +2 -0
index a66f8069b88c..10557e8dc829 100644
@@ -10,9 +10,11 @@
 struct pds_vfio_pci_device {
 	struct vfio_pci_core_device vfio_coredev;
 	struct pci_dev *pdev;
+	void *pdsc;
 
 	int vf_id;
 	int pci_id;
+	u16 client_id;
 };
 
 const struct vfio_device_ops *pds_vfio_ops_info(void);
include/linux/pds/pds_lm.h new file +12 -0
new file mode 100644
index 000000000000..2bc2bf79426e
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2022 Pensando Systems, Inc */
+
+#ifndef _PDS_LM_H_
+#define _PDS_LM_H_
+
+#include "pds_common.h"
+
+#define PDS_DEV_TYPE_LM_STR	"LM"
+#define PDS_LM_DEV_NAME		PDS_CORE_DRV_NAME "." PDS_DEV_TYPE_LM_STR
+
+#endif /* _PDS_LM_H_ */
Signature · 1 line
2.17.1
#5 lore raw

Re: [PATCH v8 vfio 3/7] vfio/pds: register with the pds_core PF

Quoted 56 lines · On Tue, 4 Apr 2023 12:01:37 -0700 Brett Creeley <[email protected]> wrote:

The pds_core driver will supply adminq services, so find the PF and register with the DSC services.

Use the following commands to enable a VF:
echo 1 > /sys/bus/pci/drivers/pds_core/$PF_BDF/sriov_numvfs

---
 drivers/vfio/pci/pds/Makefile   |  1 +
 drivers/vfio/pci/pds/cmds.c     | 69 +++++++++++++++++++++++++++++++++
 drivers/vfio/pci/pds/cmds.h     | 10 +++++
 drivers/vfio/pci/pds/pci_drv.c  | 16 +++++++-
 drivers/vfio/pci/pds/pci_drv.h  |  9 +++++
 drivers/vfio/pci/pds/vfio_dev.c |  5 +++
 drivers/vfio/pci/pds/vfio_dev.h |  2 +
 include/linux/pds/pds_lm.h      | 12 ++++++
 8 files changed, 123 insertions(+), 1 deletion(-)
 create mode 100644 drivers/vfio/pci/pds/cmds.c
 create mode 100644 drivers/vfio/pci/pds/cmds.h
 create mode 100644 drivers/vfio/pci/pds/pci_drv.h
 create mode 100644 include/linux/pds/pds_lm.h
drivers/vfio/pci/pds/Makefile +1 -0
index e1a55ae0f079..87581111fa17 100644
@@ -4,5 +4,6 @@
 obj-$(CONFIG_PDS_VFIO_PCI) += pds_vfio.o
 
 pds_vfio-y := \
+	cmds.o		\
 	pci_drv.o	\
 	vfio_dev.o
drivers/vfio/pci/pds/cmds.c new file +16 -0
new file mode 100644
index 000000000000..7807dbb2c72c
@@ -0,0 +1,69 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#include <linux/io.h>
+#include <linux/types.h>
+
+#include <linux/pds/pds_common.h>
+#include <linux/pds/pds_core_if.h>
+#include <linux/pds/pds_adminq.h>
+#include <linux/pds/pds_lm.h>
+
+#include "vfio_dev.h"
+#include "cmds.h"
+
+int
+pds_vfio_register_client_cmd(struct pds_vfio_pci_device *pds_vfio)

drivers/vfio/pci/pds/cmds.c line 26

Quoted 10 lines
drivers/vfio/pci/pds/cmds.c new file +10 -0
+{
+	union pds_core_adminq_comp comp = { 0 };
+	union pds_core_adminq_cmd cmd = { 0 };
+	struct device *dev;
+	int err;
+	u32 id;
+	u16 ci;
+
+	id = PCI_DEVID(pds_vfio->pdev->bus->number,
+		       pci_iov_virtfn_devfn(pds_vfio->pdev, pds_vfio->vf_id));

Does this actually work?

pds_vfio_pci_probe() is presumably called with a VF pdev because it calls pdsc_get_pf_struct() -> pci_iov_get_pf_drvdata(), which returns an error if !dev->is_virtfn. pds_vfio_init_device() also stores the vf_id from pci_iov_vf_id(), which requires that it's called on a VF, not PF. This same pdev gets stored at pds_vfio->pdev.

OTOH, pci_iov_virtfn_devfn() used here errors if !dev->is_physfn. So I expect we're calling PCI_DEVID with the second arg as -EINVAL. The first arg would also be suspicious if pds_vfio->pdev were the PF since then we'd be making a PCI_DEVID combining the PF bus number and the VF devfn.

We've also already stored the VF PCI_DEVID at pds_vfio->pci_id, so creating it again seems entirely unnecessary.

Also, since we never check the return of pdsc_get_pf_struct() I'd guess we take a segfault referencing off of pds_vfio->pdsc should the driver bind to something other than expected. Validating the PF drvdata before anything else seems like a good first stop in pds_vfio_pci_probe().

It's also a little curious why we're storing the pdev at all in the pds_vfio_pci_device when it's readily available from the embedded vfio_pci_core_device. Thanks,

Alex

Quoted 167 lines
drivers/vfio/pci/pds/cmds.c new file +43 -0
+
+	dev = &pds_vfio->pdev->dev;
+	cmd.client_reg.opcode = PDS_AQ_CMD_CLIENT_REG;
+	snprintf(cmd.client_reg.devname, sizeof(cmd.client_reg.devname),
+		 "%s.%d-%u", PDS_LM_DEV_NAME,
+		 pci_domain_nr(pds_vfio->pdev->bus), id);
+
+	err = pdsc_adminq_post(pds_vfio->pdsc, &cmd, &comp, false);
+	if (err) {
+		dev_info(dev, "register with DSC failed, status %d: %pe\n",
+			 comp.status, ERR_PTR(err));
+		return err;
+	}
+
+	ci = le16_to_cpu(comp.client_reg.client_id);
+	if (!ci) {
+		dev_err(dev, "%s: device returned null client_id\n", __func__);
+		return -EIO;
+	}
+	pds_vfio->client_id = ci;
+
+	return 0;
+}
+
+void
+pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio)
+{
+	union pds_core_adminq_comp comp = { 0 };
+	union pds_core_adminq_cmd cmd = { 0 };
+	struct device *dev;
+	int err;
+
+	dev = &pds_vfio->pdev->dev;
+	cmd.client_unreg.opcode = PDS_AQ_CMD_CLIENT_UNREG;
+	cmd.client_unreg.client_id = cpu_to_le16(pds_vfio->client_id);
+
+	err = pdsc_adminq_post(pds_vfio->pdsc, &cmd, &comp, false);
+	if (err)
+		dev_info(dev, "unregister from DSC failed, status %d: %pe\n",
+			 comp.status, ERR_PTR(err));
+
+	pds_vfio->client_id = 0;
+}
drivers/vfio/pci/pds/cmds.h new file +10 -0
new file mode 100644
index 000000000000..4c592afccf89
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _CMDS_H_
+#define _CMDS_H_
+
+int pds_vfio_register_client_cmd(struct pds_vfio_pci_device *pds_vfio);
+void pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio);
+
+#endif /* _CMDS_H_ */
drivers/vfio/pci/pds/pci_drv.c +15 -1
index 5e554420792e..46537afdee2d 100644
@@ -8,9 +8,13 @@
 #include <linux/types.h>
 #include <linux/vfio.h>
 
+#include <linux/pds/pds_common.h>
 #include <linux/pds/pds_core_if.h>
+#include <linux/pds/pds_adminq.h>
 
 #include "vfio_dev.h"
+#include "pci_drv.h"
+#include "cmds.h"
 
 #define PDS_VFIO_DRV_NAME		"pds_vfio"
 #define PDS_VFIO_DRV_DESCRIPTION	"AMD/Pensando VFIO Device Driver"
@@ -30,13 +34,23 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
 	dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
 	pds_vfio->pdev = pdev;
+	pds_vfio->pdsc = pdsc_get_pf_struct(pdev);
+
+	err = pds_vfio_register_client_cmd(pds_vfio);
+	if (err) {
+		dev_err(&pdev->dev, "failed to register as client: %pe\n",
+			ERR_PTR(err));
+		goto out_put_vdev;
+	}
 
 	err = vfio_pci_core_register_device(&pds_vfio->vfio_coredev);
 	if (err)
-		goto out_put_vdev;
+		goto out_unreg_client;
 
 	return 0;
 
+out_unreg_client:
+	pds_vfio_unregister_client_cmd(pds_vfio);
 out_put_vdev:
 	vfio_put_device(&pds_vfio->vfio_coredev.vdev);
 	return err;
drivers/vfio/pci/pds/pci_drv.h new file +9 -0
new file mode 100644
index 000000000000..e79bed12ed14
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _PCI_DRV_H
+#define _PCI_DRV_H
+
+#include <linux/pci.h>
+
+#endif /* _PCI_DRV_H */
drivers/vfio/pci/pds/vfio_dev.c +5 -0
index 0f70efec01e1..056715dea512 100644
@@ -31,6 +31,11 @@ pds_vfio_init_device(struct vfio_device *vdev)
 	pds_vfio->vf_id = pci_iov_vf_id(pdev);
 	pds_vfio->pci_id = PCI_DEVID(pdev->bus->number, pdev->devfn);
 
+	dev_dbg(&pdev->dev, "%s: PF %#04x VF %#04x (%d) vf_id %d domain %d pds_vfio %p\n",
+		__func__, pci_dev_id(pdev->physfn),
+		pds_vfio->pci_id, pds_vfio->pci_id, pds_vfio->vf_id,
+		pci_domain_nr(pdev->bus), pds_vfio);
+
 	return 0;
 }
 
drivers/vfio/pci/pds/vfio_dev.h +2 -0
index a66f8069b88c..10557e8dc829 100644
@@ -10,9 +10,11 @@
 struct pds_vfio_pci_device {
 	struct vfio_pci_core_device vfio_coredev;
 	struct pci_dev *pdev;
+	void *pdsc;
 
 	int vf_id;
 	int pci_id;
+	u16 client_id;
 };
 
 const struct vfio_device_ops *pds_vfio_ops_info(void);
include/linux/pds/pds_lm.h new file +12 -0
new file mode 100644
index 000000000000..2bc2bf79426e
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2022 Pensando Systems, Inc */
+
+#ifndef _PDS_LM_H_
+#define _PDS_LM_H_
+
+#include "pds_common.h"
+
+#define PDS_DEV_TYPE_LM_STR	"LM"
+#define PDS_LM_DEV_NAME		PDS_CORE_DRV_NAME "." PDS_DEV_TYPE_LM_STR
+
+#endif /* _PDS_LM_H_ */
#6 lore raw

Re: [PATCH v8 vfio 3/7] vfio/pds: register with the pds_core PF

Quoted 96 lines · Alex Williamson

Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.

Quoted 56 lines · On Tue, 4 Apr 2023 12:01:37 -0700 Brett Creeley <[email protected]> wrote:

The pds_core driver will supply adminq services, so find the PF and register with the DSC services.

Use the following commands to enable a VF:
echo 1 > /sys/bus/pci/drivers/pds_core/$PF_BDF/sriov_numvfs

---
  drivers/vfio/pci/pds/Makefile   |  1 +
  drivers/vfio/pci/pds/cmds.c     | 69 +++++++++++++++++++++++++++++++++
  drivers/vfio/pci/pds/cmds.h     | 10 +++++
  drivers/vfio/pci/pds/pci_drv.c  | 16 +++++++-
  drivers/vfio/pci/pds/pci_drv.h  |  9 +++++
  drivers/vfio/pci/pds/vfio_dev.c |  5 +++
  drivers/vfio/pci/pds/vfio_dev.h |  2 +
  include/linux/pds/pds_lm.h      | 12 ++++++
  8 files changed, 123 insertions(+), 1 deletion(-)

create mode 100644 drivers/vfio/pci/pds/cmds.c create mode 100644 drivers/vfio/pci/pds/cmds.h create mode 100644 drivers/vfio/pci/pds/pci_drv.h create mode 100644 include/linux/pds/pds_lm.h

drivers/vfio/pci/pds/Makefile +1 -0
index e1a55ae0f079..87581111fa17 100644
@@ -4,5 +4,6 @@
  obj-$(CONFIG_PDS_VFIO_PCI) += pds_vfio.o
 
  pds_vfio-y := \
+     cmds.o          \
       pci_drv.o       \
       vfio_dev.o
drivers/vfio/pci/pds/cmds.c new file +16 -0
new file mode 100644
index 000000000000..7807dbb2c72c
@@ -0,0 +1,69 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#include <linux/io.h>
+#include <linux/types.h>
+
+#include <linux/pds/pds_common.h>
+#include <linux/pds/pds_core_if.h>
+#include <linux/pds/pds_adminq.h>
+#include <linux/pds/pds_lm.h>
+
+#include "vfio_dev.h"
+#include "cmds.h"
+
+int
+pds_vfio_register_client_cmd(struct pds_vfio_pci_device *pds_vfio)

drivers/vfio/pci/pds/cmds.c line 26

Quoted 10 lines
drivers/vfio/pci/pds/cmds.c new file +10 -0
+{
+     union pds_core_adminq_comp comp = { 0 };
+     union pds_core_adminq_cmd cmd = { 0 };
+     struct device *dev;
+     int err;
+     u32 id;
+     u16 ci;
+
+     id = PCI_DEVID(pds_vfio->pdev->bus->number,
+                    pci_iov_virtfn_devfn(pds_vfio->pdev, pds_vfio->vf_id));

Does this actually work?

pds_vfio_pci_probe() is presumably called with a VF pdev because it calls pdsc_get_pf_struct() -> pci_iov_get_pf_drvdata(), which returns an error if !dev->is_virtfn. pds_vfio_init_device() also stores the vf_id from pci_iov_vf_id(), which requires that it's called on a VF, not PF. This same pdev gets stored at pds_vfio->pdev.

OTOH, pci_iov_virtfn_devfn() used here errors if !dev->is_physfn. So I expect we're calling PCI_DEVID with the second arg as -EINVAL. The first arg would also be suspicious if pds_vfio->pdev were the PF since then we'd be making a PCI_DEVID combining the PF bus number and the VF devfn.

We've also already stored the VF PCI_DEVID at pds_vfio->pci_id, so creating it again seems entirely unnecessary.

Also, since we never check the return of pdsc_get_pf_struct() I'd guess we take a segfault referencing off of pds_vfio->pdsc should the driver bind to something other than expected. Validating the PF drvdata before anything else seems like a good first stop in pds_vfio_pci_probe().

Quoted 6 lines

It's also a little curious why we're storing the pdev at all in the pds_vfio_pci_device when it's readily available from the embedded vfio_pci_core_device. Thanks,

Alex

Yeah, these are all good catches and I'm not sure how I didn't notice them before. I will rework these bits based on your feedback.

Thanks,

Brett

Quoted 168 lines
Quoted 167 lines
drivers/vfio/pci/pds/cmds.c new file +43 -0
+
+     dev = &pds_vfio->pdev->dev;
+     cmd.client_reg.opcode = PDS_AQ_CMD_CLIENT_REG;
+     snprintf(cmd.client_reg.devname, sizeof(cmd.client_reg.devname),
+              "%s.%d-%u", PDS_LM_DEV_NAME,
+              pci_domain_nr(pds_vfio->pdev->bus), id);
+
+     err = pdsc_adminq_post(pds_vfio->pdsc, &cmd, &comp, false);
+     if (err) {
+             dev_info(dev, "register with DSC failed, status %d: %pe\n",
+                      comp.status, ERR_PTR(err));
+             return err;
+     }
+
+     ci = le16_to_cpu(comp.client_reg.client_id);
+     if (!ci) {
+             dev_err(dev, "%s: device returned null client_id\n", __func__);
+             return -EIO;
+     }
+     pds_vfio->client_id = ci;
+
+     return 0;
+}
+
+void
+pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio)
+{
+     union pds_core_adminq_comp comp = { 0 };
+     union pds_core_adminq_cmd cmd = { 0 };
+     struct device *dev;
+     int err;
+
+     dev = &pds_vfio->pdev->dev;
+     cmd.client_unreg.opcode = PDS_AQ_CMD_CLIENT_UNREG;
+     cmd.client_unreg.client_id = cpu_to_le16(pds_vfio->client_id);
+
+     err = pdsc_adminq_post(pds_vfio->pdsc, &cmd, &comp, false);
+     if (err)
+             dev_info(dev, "unregister from DSC failed, status %d: %pe\n",
+                      comp.status, ERR_PTR(err));
+
+     pds_vfio->client_id = 0;
+}
drivers/vfio/pci/pds/cmds.h new file +10 -0
new file mode 100644
index 000000000000..4c592afccf89
@@ -0,0 +1,10 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _CMDS_H_
+#define _CMDS_H_
+
+int pds_vfio_register_client_cmd(struct pds_vfio_pci_device *pds_vfio);
+void pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio);
+
+#endif /* _CMDS_H_ */
drivers/vfio/pci/pds/pci_drv.c +15 -1
index 5e554420792e..46537afdee2d 100644
@@ -8,9 +8,13 @@
  #include <linux/types.h>
  #include <linux/vfio.h>
 
+#include <linux/pds/pds_common.h>
  #include <linux/pds/pds_core_if.h>
+#include <linux/pds/pds_adminq.h>
 
  #include "vfio_dev.h"
+#include "pci_drv.h"
+#include "cmds.h"
 
  #define PDS_VFIO_DRV_NAME            "pds_vfio"
  #define PDS_VFIO_DRV_DESCRIPTION     "AMD/Pensando VFIO Device Driver"
@@ -30,13 +34,23 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
       dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
       pds_vfio->pdev = pdev;
+     pds_vfio->pdsc = pdsc_get_pf_struct(pdev);
+
+     err = pds_vfio_register_client_cmd(pds_vfio);
+     if (err) {
+             dev_err(&pdev->dev, "failed to register as client: %pe\n",
+                     ERR_PTR(err));
+             goto out_put_vdev;
+     }
 
       err = vfio_pci_core_register_device(&pds_vfio->vfio_coredev);
       if (err)
-             goto out_put_vdev;
+             goto out_unreg_client;
 
       return 0;
 
+out_unreg_client:
+     pds_vfio_unregister_client_cmd(pds_vfio);
  out_put_vdev:
       vfio_put_device(&pds_vfio->vfio_coredev.vdev);
       return err;
drivers/vfio/pci/pds/pci_drv.h new file +9 -0
new file mode 100644
index 000000000000..e79bed12ed14
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#ifndef _PCI_DRV_H
+#define _PCI_DRV_H
+
+#include <linux/pci.h>
+
+#endif /* _PCI_DRV_H */
drivers/vfio/pci/pds/vfio_dev.c +5 -0
index 0f70efec01e1..056715dea512 100644
@@ -31,6 +31,11 @@ pds_vfio_init_device(struct vfio_device *vdev)
       pds_vfio->vf_id = pci_iov_vf_id(pdev);
       pds_vfio->pci_id = PCI_DEVID(pdev->bus->number, pdev->devfn);
 
+     dev_dbg(&pdev->dev, "%s: PF %#04x VF %#04x (%d) vf_id %d domain %d pds_vfio %p\n",
+             __func__, pci_dev_id(pdev->physfn),
+             pds_vfio->pci_id, pds_vfio->pci_id, pds_vfio->vf_id,
+             pci_domain_nr(pdev->bus), pds_vfio);
+
       return 0;
  }
 
drivers/vfio/pci/pds/vfio_dev.h +2 -0
index a66f8069b88c..10557e8dc829 100644
@@ -10,9 +10,11 @@
  struct pds_vfio_pci_device {
       struct vfio_pci_core_device vfio_coredev;
       struct pci_dev *pdev;
+     void *pdsc;
 
       int vf_id;
       int pci_id;
+     u16 client_id;
  };
 
  const struct vfio_device_ops *pds_vfio_ops_info(void);
include/linux/pds/pds_lm.h new file +12 -0
new file mode 100644
index 000000000000..2bc2bf79426e
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/* Copyright(c) 2022 Pensando Systems, Inc */
+
+#ifndef _PDS_LM_H_
+#define _PDS_LM_H_
+
+#include "pds_common.h"
+
+#define PDS_DEV_TYPE_LM_STR  "LM"
+#define PDS_LM_DEV_NAME              PDS_CORE_DRV_NAME "." PDS_DEV_TYPE_LM_STR
+
+#endif /* _PDS_LM_H_ */
#7 lore raw

Re: [PATCH v8 vfio 3/7] vfio/pds: register with the pds_core PF

In reply to Brett Creeley

drivers/vfio/pci/pds/pci_drv.c pds_vfio_pci_probe(struct pci_dev *pdev, · line 37

Quoted 5 lines · Brett Creeley
drivers/vfio/pci/pds/pci_drv.c +1 -0
@@ -30,13 +34,23 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
 	dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
 	pds_vfio->pdev = pdev;
+	pds_vfio->pdsc = pdsc_get_pf_struct(pdev);

This should not be a void *, it has a type, looks like it is 'struct pdsc *' - comment applies to all the places in both series that dropped the type here.

Jason

#8 lore raw

Re: [PATCH v8 vfio 3/7] vfio/pds: register with the pds_core PF

Quoted 11 lines · Jason Gunthorpe

drivers/vfio/pci/pds/pci_drv.c pds_vfio_pci_probe(struct pci_dev *pdev, · line 37

Quoted 5 lines · Brett Creeley
drivers/vfio/pci/pds/pci_drv.c +1 -0
@@ -30,13 +34,23 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
       dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
       pds_vfio->pdev = pdev;
+     pds_vfio->pdsc = pdsc_get_pf_struct(pdev);

This should not be a void *, it has a type, looks like it is 'struct pdsc *' - comment applies to all the places in both series that dropped the type here.

Thanks, Jason. I'll fix this up with a proper type in the pds_core patchset so that the pds_vfio can follow.

sln

#9 lore raw

Re: [PATCH v8 vfio 3/7] vfio/pds: register with the pds_core PF

In reply to Jason Gunthorpe
Quoted 6 lines · Jason Gunthorpe

Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.

Quoted 2 lines · Brett Creeley
(quoted diff) +0 -0
@@ -30,13 +34,23 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
Quoted 7 lines

drivers/vfio/pci/pds/pci_drv.c pds_vfio_pci_probe(struct pci_dev *pdev, · line 37

Quoted 3 lines
drivers/vfio/pci/pds/pci_drv.c +1 -0
       dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
       pds_vfio->pdev = pdev;
+     pds_vfio->pdsc = pdsc_get_pf_struct(pdev);

This should not be a void *, it has a type, looks like it is 'struct pdsc *' - comment applies to all the places in both series that dropped the type here.

Will fix.

Quoted 2 lines

Jason

Hey Jason,

Thanks for the responses/feedback.

For some reason Shannon and I didn't get any of your recent responses in our inboxes except this one. We're not really sure why... Due to this, I'm replying to all of your responses in this thread.

Quoted 3 lines

Review pds_vfio_pci_probe(struct pci_dev *pdev, · line 38

Quoted 1 line
(quoted diff) +1 -0
+	union pds_core_adminq_cmd cmd = { 0 };

These should all be = {}, adding the 0 is a subtly different thing in

Will fix.

Quoted 13 lines
Quoted 13 lines
(quoted diff) +13 -0
+int
+pds_vfio_suspend_device_cmd(struct pds_vfio_pci_device *pds_vfio)
+{
+	struct pds_lm_suspend_cmd cmd = {
+		.opcode = PDS_LM_CMD_SUSPEND,
+		.vf_id = cpu_to_le16(pds_vfio->vf_id),
+	};
+	struct pds_lm_suspend_comp comp = {0};
+	struct pci_dev *pdev = pds_vfio->pdev;
+	int err;
+
+	dev_dbg(&pdev->dev, "vf%u: Suspend device\n", pds_vfio->vf_id);
+
Quoted 8 lines

Review pds_vfio_pci_probe(struct pci_dev *pdev, · line 56

Quoted 5 lines
(quoted diff) +5 -0
+	err = pds_client_adminq_cmd(pds_vfio,
+				    (union pds_core_adminq_cmd *)&cmd,
+				    sizeof(cmd),
+				    (union pds_core_adminq_comp *)&comp,
+				    PDS_AQ_FLAG_FASTPOLL);

These casts to a union are really weird, why isn't the union the type on the stack?

Yeah, this is an artifact of initial development that allowed us to completely de-couple pds_lm.h from pds_adminq.h, but there don't seem to be any conflicts including inluding pds_lm.h in pds_adminq.h. So, it will be fixed in the next revision.

Quoted 7 lines

Review pds_vfio_pci_probe(struct pci_dev *pdev,

Quoted 5 lines
(quoted diff) +5 -0
+
+	/* alloc sgl */
+	sgl = dma_alloc_coherent(dev, lm_file->num_sge *
+				 sizeof(struct pds_lm_sg_elem),
+				 &lm_file->sgl_addr, GFP_KERNEL);

Do you really need a coherent allocation for this?

I don't think it is needed. I will look into this and fix it if dma_alloc_coherent() isn't needed.

Quoted 4 lines
Quoted 1 line

+#define PDS_VFIO_LM_FILENAME "pds_vfio_lm"

This doesn't need a define, it is typical to write the pseudo filename in the only anon_inode_getfile()

Yeah, we technically only use it in one spot, so it makes sense to just have the string inlined to the anon_inode_getfile() call. This also allows us to get rid of the const char *name argument in pds_vfio_get_lm_file(). Will fix in the next revision.

Quoted 44 lines
Quoted 44 lines
(quoted diff) +44 -0
+static struct pds_vfio_lm_file *
+pds_vfio_get_lm_file(const char *name, const struct file_operations *fops,
+		     int flags, u64 size)
+{
+	struct pds_vfio_lm_file *lm_file = NULL;
+	unsigned long long npages;
+	struct page **pages;
+	int err = 0;
+
+	if (!size)
+		return NULL;
+
+	/* Alloc file structure */
+	lm_file = kzalloc(sizeof(*lm_file), GFP_KERNEL);
+	if (!lm_file)
+		return NULL;
+
+	/* Create file */
+	lm_file->filep = anon_inode_getfile(name, fops, lm_file, flags);
+	if (!lm_file->filep)
+		goto err_get_file;
+
+	stream_open(lm_file->filep->f_inode, lm_file->filep);
+	mutex_init(&lm_file->lock);
+
+	lm_file->size = size;
+
+	/* Allocate memory for file pages */
+	npages = DIV_ROUND_UP_ULL(lm_file->size, PAGE_SIZE);
+
+	pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
+	if (!pages)
+		goto err_alloc_pages;
+
+	for (unsigned long long i = 0; i < npages; i++) {
+		pages[i] = alloc_page(GFP_KERNEL);
+		if (!pages[i])
+			goto err_alloc_page;
+	}
+
+	lm_file->pages = pages;
+	lm_file->npages = npages;
+	lm_file->alloc_size = npages * PAGE_SIZE;
+
Quoted 8 lines

Review

Quoted 5 lines
(quoted diff) +5 -0
+	/* Create scatterlist of file pages to use for DMA mapping later */
+	err = sg_alloc_table_from_pages(&lm_file->sg_table, pages, npages,
+					0, size, GFP_KERNEL);
+	if (err)
+		goto err_alloc_sg_table;

This is the same basic thing the mlx5 driver does, you should move the mlx5 code into some common place and just re-use it here.

I looked at the mlx5 code and even though the two drivers are doing the same basic thing, IMHO it doesn't seem like a straight forward task as the mlx5 code seems to have some device/driver specifics mixed in. I'd prefer not trying to refactor/commonize this bit of code at this point in time.

However, it does seem like a good future improvement once things quiet down after getting this initial series merged.

Quoted 8 lines
Quoted 8 lines
drivers/vfio/pci/pds/vfio_dev.h +0 -0
index 10557e8dc829..3f55861ffc7c 100644
@@ -7,10 +7,20 @@
 #include <linux/pci.h>
 #include <linux/vfio_pci_core.h>
 
Quoted 10 lines

drivers/vfio/pci/pds/vfio_dev.h line 16

Quoted 7 lines
drivers/vfio/pci/pds/vfio_dev.h +3 -0
+#include "lm.h"
+
 struct pds_vfio_pci_device {
 	struct vfio_pci_core_device vfio_coredev;
 	struct pci_dev *pdev;
 	void *pdsc;
+	struct device *coredev;

Why? If this is just &pdev->dev it it doesn't need to be in the struct And pdev is just vfio_coredev->pdev, don't need to duplicate it either

This was actually the pds_core's device structure. I have removed this in my local tree and instead use the pci_physfn() to get pds_core's struct device. Will be fixed in the next revision.

Quoted 33 lines
Quoted 33 lines
drivers/vfio/pci/pds/vfio_dev.h +33 -0
+static void
+pds_vfio_recovery_work(struct work_struct *work)
+{
+	struct pds_vfio_pci_device *pds_vfio =
+		container_of(work, struct pds_vfio_pci_device, work);
+	bool deferred_reset_needed = false;
+
+	/* Documentation states that the kernel migration driver must not
+	 * generate asynchronous device state transitions outside of
+	 * manipulation by the user or the VFIO_DEVICE_RESET ioctl.
+	 *
+	 * Since recovery is an asynchronous event received from the device,
+	 * initiate a deferred reset. Only issue the deferred reset if a
+	 * migration is in progress, which will cause the next step of the
+	 * migration to fail. Also, if the device is in a state that will
+	 * be set to VFIO_DEVICE_STATE_RUNNING on the next action (i.e. VM is
+	 * shutdown and device is in VFIO_DEVICE_STATE_STOP) as that will clear
+	 * the VFIO_DEVICE_STATE_ERROR when the VM starts back up.
+	 */
+	mutex_lock(&pds_vfio->state_mutex);
+	if ((pds_vfio->state != VFIO_DEVICE_STATE_RUNNING &&
+	     pds_vfio->state != VFIO_DEVICE_STATE_ERROR) ||
+	    (pds_vfio->state == VFIO_DEVICE_STATE_RUNNING &&
+	     pds_vfio_dirty_is_enabled(pds_vfio)))
+		deferred_reset_needed = true;
+	mutex_unlock(&pds_vfio->state_mutex);
+
+	/* On the next user initiated state transition, the device will
+	 * transition to the VFIO_DEVICE_STATE_ERROR. At this point it's the user's
+	 * responsibility to reset the device.
+	 *
+	 * If a VFIO_DEVICE_RESET is requested post recovery and before the next
+	 * state transition, then the deferred reset state will be set to
Quoted 8 lines

drivers/vfio/pci/pds/vfio_dev.h

Quoted 5 lines
drivers/vfio/pci/pds/vfio_dev.h +5 -0
+	 * VFIO_DEVICE_STATE_RUNNING.
+	 */
+	if (deferred_reset_needed)
+		pds_vfio_deferred_reset(pds_vfio, VFIO_DEVICE_STATE_ERROR);
+}

Why is this a work? it is threaded on a blocking_notifier_chain so it can call the mutex?

I think the work item can be dropped and the contents of the work function can be moved in the notifier callback. I will fix this in the next revision.

Quoted 2 lines

Why is the locking like this, can't you just call
pds_vfio_deferred_reset() under the mutex?

It was done to avoid any lock ordering issues with pds_vfio_state_mutex_unlock() or pds_vfio_reset().

Quoted 8 lines
Quoted 6 lines

Add Kconfig entries and pds_vfio.rst. Also, add an entry in the MAINTAINERS file for this new driver.

It's not clear where documentation for vendor specific VFIO drivers should live, so just re-use the current amd ethernet location.

It would be nice to make a kdoc section for vfio.

It seems like there are already vfio docs in Documentation/driver-api/, but the kdoc added in this patch is slightly different since it's vendor specific. Which of the following locations make the most sense?

[1] Documentation/vfio/<vendor>/<vendor_kdoc>
- Documentation/vfio/amd/pds_vfio.rst

[2] Documentation/vfio/vendor-drivers/<vendor_kdoc>
- Documentation/vfio/vendor-drivers/pds_vfio.rst

Thanks again for the time and feedback,

Brett

author maintainer
#10 lore raw

PATCH v8 4/7 vfio/pds: Add VFIO live migration support

In reply to Brett Creeley

Add live migration support via the VFIO subsystem. The migration implementation aligns with the definition from uapi/vfio.h and uses the pds_core PF's adminq for device configuration.

The ability to suspend, resume, and transfer VF device state data is included along with the required admin queue command structures and implementations.

PDS_LM_CMD_SUSPEND and PDS_LM_CMD_SUSPEND_STATUS are added to support the VF device suspend operation.

PDS_LM_CMD_RESUME is added to support the VF device resume operation.

PDS_LM_CMD_STATUS is added to determine the exact size of the VF device state data.

PDS_LM_CMD_SAVE is added to get the VF device state data.

PDS_LM_CMD_RESTORE is added to restore the VF device with the previously saved data from PDS_LM_CMD_SAVE.

PDS_LM_CMD_HOST_VF_STATUS is added to notify the device when a migration is in/not-in progress from the host's perspective.

---
 drivers/vfio/pci/pds/Makefile   |   1 +
 drivers/vfio/pci/pds/cmds.c     | 322 +++++++++++++++++++++
 drivers/vfio/pci/pds/cmds.h     |   8 +-
 drivers/vfio/pci/pds/lm.c       | 479 ++++++++++++++++++++++++++++++++
 drivers/vfio/pci/pds/lm.h       |  52 ++++
 drivers/vfio/pci/pds/pci_drv.c  |  16 ++
 drivers/vfio/pci/pds/vfio_dev.c | 119 +++++++-
 drivers/vfio/pci/pds/vfio_dev.h |  11 +
 include/linux/pds/pds_lm.h      | 210 +++++++++++++-
 9 files changed, 1213 insertions(+), 5 deletions(-)
 create mode 100644 drivers/vfio/pci/pds/lm.c
 create mode 100644 drivers/vfio/pci/pds/lm.h
drivers/vfio/pci/pds/Makefile +1 -0
index 87581111fa17..dbaf613d3794 100644
@@ -5,5 +5,6 @@ obj-$(CONFIG_PDS_VFIO_PCI) += pds_vfio.o
 
 pds_vfio-y := \
 	cmds.o		\
+	lm.o		\
 	pci_drv.o	\
 	vfio_dev.o
drivers/vfio/pci/pds/cmds.c +322 -0

337 lines not on this page · raw message

drivers/vfio/pci/pds/cmds.h +7 -1
index 4c592afccf89..3d8a5508c733 100644
@@ -6,5 +6,11 @@
 
 int pds_vfio_register_client_cmd(struct pds_vfio_pci_device *pds_vfio);
 void pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio);
-
+int pds_vfio_suspend_device_cmd(struct pds_vfio_pci_device *pds_vfio);
+int pds_vfio_resume_device_cmd(struct pds_vfio_pci_device *pds_vfio);
+int pds_vfio_get_lm_status_cmd(struct pds_vfio_pci_device *pds_vfio, u64 *size);
+int pds_vfio_get_lm_state_cmd(struct pds_vfio_pci_device *pds_vfio);
+int pds_vfio_set_lm_state_cmd(struct pds_vfio_pci_device *pds_vfio);
+void pds_vfio_send_host_vf_lm_status_cmd(struct pds_vfio_pci_device *pds_vfio,
+					 enum pds_lm_host_vf_status vf_status);
 #endif /* _CMDS_H_ */
drivers/vfio/pci/pds/lm.c new file +479 -0

479 lines not on this page · raw message

drivers/vfio/pci/pds/lm.h new file +52 -0

52 lines not on this page · raw message

drivers/vfio/pci/pds/pci_drv.c +16 -0
index 46537afdee2d..b0781d9f4246 100644
@@ -34,6 +34,7 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 
 	dev_set_drvdata(&pdev->dev, &pds_vfio->vfio_coredev);
 	pds_vfio->pdev = pdev;
+	pds_vfio->coredev = &pdev->dev;
 	pds_vfio->pdsc = pdsc_get_pf_struct(pdev);
 
 	err = pds_vfio_register_client_cmd(pds_vfio);
@@ -72,12 +73,27 @@ pds_vfio_pci_table[] = {
 };
 MODULE_DEVICE_TABLE(pci, pds_vfio_pci_table);
 
+static void
+pds_vfio_pci_aer_reset_done(struct pci_dev *pdev)
+{
+	struct pds_vfio_pci_device *pds_vfio = pds_vfio_pci_drvdata(pdev);
+
+	pds_vfio_reset(pds_vfio);
+}
+
+static const struct
+pci_error_handlers pds_vfio_pci_err_handlers = {
+	.reset_done = pds_vfio_pci_aer_reset_done,
+	.error_detected = vfio_pci_core_aer_err_detected,
+};
+
 static struct pci_driver
 pds_vfio_pci_driver = {
 	.name = PDS_VFIO_DRV_NAME,
 	.id_table = pds_vfio_pci_table,
 	.probe = pds_vfio_pci_probe,
 	.remove = pds_vfio_pci_remove,
+	.err_handler = &pds_vfio_pci_err_handlers,
 	.driver_managed_dma = true,
 };
 
drivers/vfio/pci/pds/vfio_dev.c +118 -1

154 lines not on this page · raw message

drivers/vfio/pci/pds/vfio_dev.h +11 -0
index 10557e8dc829..3f55861ffc7c 100644
@@ -7,10 +7,20 @@
 #include <linux/pci.h>
 #include <linux/vfio_pci_core.h>
 
+#include "lm.h"
+
 struct pds_vfio_pci_device {
 	struct vfio_pci_core_device vfio_coredev;
 	struct pci_dev *pdev;
 	void *pdsc;
+	struct device *coredev;
+
+	struct pds_vfio_lm_file *save_file;
+	struct pds_vfio_lm_file *restore_file;
+	struct mutex state_mutex; /* protect migration state */
+	enum vfio_device_mig_state state;
+	spinlock_t reset_lock; /* protect reset_done flow */
+	u8 deferred_reset;
 
 	int vf_id;
 	int pci_id;
@@ -19,5 +29,6 @@ struct pds_vfio_pci_device {
 
 const struct vfio_device_ops *pds_vfio_ops_info(void);
 struct pds_vfio_pci_device *pds_vfio_pci_drvdata(struct pci_dev *pdev);
+void pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio);
 
 #endif /* _VFIO_DEV_H_ */
include/linux/pds/pds_lm.h +207 -3

219 lines not on this page · raw message

Signature · 1 line
2.17.1
#11 lore raw

Re: [PATCH v8 vfio 4/7] vfio/pds: Add VFIO live migration support

Quoted 32 lines · On Tue, 4 Apr 2023 12:01:38 -0700 Brett Creeley <[email protected]> wrote:
drivers/vfio/pci/pds/lm.c new file +26 -0
new file mode 100644
index 000000000000..855f5da9b99a
@@ -0,0 +1,479 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#include <linux/anon_inodes.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/highmem.h>
+#include <linux/vfio.h>
+#include <linux/vfio_pci_core.h>
+
+#include "vfio_dev.h"
+#include "cmds.h"
+
+#define PDS_VFIO_LM_FILENAME	"pds_vfio_lm"
+
+const char *
+pds_vfio_lm_state(enum vfio_device_mig_state state)
+{
+	switch (state) {
+	case VFIO_DEVICE_STATE_ERROR:
+		return "VFIO_DEVICE_STATE_ERROR";
+	case VFIO_DEVICE_STATE_STOP:
+		return "VFIO_DEVICE_STATE_STOP";
+	case VFIO_DEVICE_STATE_RUNNING:
+		return "VFIO_DEVICE_STATE_RUNNING";
+	case VFIO_DEVICE_STATE_STOP_COPY:

drivers/vfio/pci/pds/lm.c line 36

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+		return "VFIO_DEVICE_STATE_STOP_COPY";
+	case VFIO_DEVICE_STATE_RESUMING:
+		return "VFIO_DEVICE_STATE_RESUMING";
+	case VFIO_DEVICE_STATE_RUNNING_P2P:
+		return "VFIO_DEVICE_STATE_RUNNING_P2P";
+	default:
+		return "VFIO_DEVICE_STATE_INVALID";
+	}
+
+	return "VFIO_DEVICE_STATE_INVALID";

Seems a tad redundant.

Quoted 2 lines
drivers/vfio/pci/pds/lm.c new file +2 -0
+}
+

[snip]

Quoted 5 lines
drivers/vfio/pci/pds/lm.c new file +5 -0
+struct file *
+pds_vfio_step_device_state_locked(struct pds_vfio_pci_device *pds_vfio,
+				  enum vfio_device_mig_state next)
+{
+	enum vfio_device_mig_state cur = pds_vfio->state;

drivers/vfio/pci/pds/lm.c line 402

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+	struct device *dev = &pds_vfio->pdev->dev;
+	int err = 0;
+
+	dev_dbg(dev, "%s => %s\n",
+		pds_vfio_lm_state(cur), pds_vfio_lm_state(next));
+
+	if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_STOP_COPY) {
+		/* Device is already stopped
+		 * create save device data file & get device state from firmware
+		 */

Standard multi-line comment style please, we're not under net/ here.

Quoted 45 lines
drivers/vfio/pci/pds/lm.c new file +45 -0
+		err = pds_vfio_get_save_file(pds_vfio);
+		if (err)
+			return ERR_PTR(err);
+
+		/* Get device state */
+		err = pds_vfio_get_lm_state_cmd(pds_vfio);
+		if (err) {
+			pds_vfio_put_save_file(pds_vfio);
+			return ERR_PTR(err);
+		}
+
+		return pds_vfio->save_file->filep;
+	}
+
+	if (cur == VFIO_DEVICE_STATE_STOP_COPY && next == VFIO_DEVICE_STATE_STOP) {
+		/* Device is already stopped
+		 * delete the save device state file
+		 */
+		pds_vfio_put_save_file(pds_vfio);
+		pds_vfio_send_host_vf_lm_status_cmd(pds_vfio,
+						    PDS_LM_STA_NONE);
+		return NULL;
+	}
+
+	if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RESUMING) {
+		/* create resume device data file */
+		err = pds_vfio_get_restore_file(pds_vfio);
+		if (err)
+			return ERR_PTR(err);
+
+		return pds_vfio->restore_file->filep;
+	}
+
+	if (cur == VFIO_DEVICE_STATE_RESUMING && next == VFIO_DEVICE_STATE_STOP) {
+		/* Set device state */
+		err = pds_vfio_set_lm_state_cmd(pds_vfio);
+		if (err)
+			return ERR_PTR(err);
+
+		/* delete resume device data file */
+		pds_vfio_put_restore_file(pds_vfio);
+		return NULL;
+	}
+
+	if (cur == VFIO_DEVICE_STATE_RUNNING && next == VFIO_DEVICE_STATE_RUNNING_P2P) {

drivers/vfio/pci/pds/lm.c line 457

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+		pds_vfio_send_host_vf_lm_status_cmd(pds_vfio, PDS_LM_STA_IN_PROGRESS);
+		/* Device should be stopped
+		 * no interrupts, dma or change in internal state
+		 */
+		err = pds_vfio_suspend_device_cmd(pds_vfio);
+		if (err)
+			return ERR_PTR(err);
+
+		return NULL;
+	}

The comment here, as well as the no-op transitions from STOP->P2P and P2P->STOP has me concerned whether a device in this suspend state really meets the definition of our P2P state. The RUNNING_P2P state is a quiescent state where the device must accept access, including peer-to-peer DMA, but it cannot initiate DMA or interrupts. Is that consistent with this suspend state? Thanks,

Alex

Quoted 22 lines
drivers/vfio/pci/pds/lm.c new file +22 -0
+
+	if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_RUNNING) {
+		/* Device should be functional
+		 * interrupts, dma, mmio or changes to internal state is allowed
+		 */
+		err = pds_vfio_resume_device_cmd(pds_vfio);
+		if (err)
+			return ERR_PTR(err);
+
+		pds_vfio_send_host_vf_lm_status_cmd(pds_vfio,
+						    PDS_LM_STA_NONE);
+		return NULL;
+	}
+
+	if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RUNNING_P2P)
+		return NULL;
+
+	if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_STOP)
+		return NULL;
+
+	return ERR_PTR(-EINVAL);
+}
#12 lore raw

Re: [PATCH v8 vfio 4/7] vfio/pds: Add VFIO live migration support

Quoted 41 lines · Alex Williamson

Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.

Quoted 36 lines · On Tue, 4 Apr 2023 12:01:38 -0700 Brett Creeley <[email protected]> wrote:
drivers/vfio/pci/pds/lm.c new file +30 -0
new file mode 100644
index 000000000000..855f5da9b99a
@@ -0,0 +1,479 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#include <linux/anon_inodes.h>
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/highmem.h>
+#include <linux/vfio.h>
+#include <linux/vfio_pci_core.h>
+
+#include "vfio_dev.h"
+#include "cmds.h"
+
+#define PDS_VFIO_LM_FILENAME "pds_vfio_lm"
+
+const char *
+pds_vfio_lm_state(enum vfio_device_mig_state state)
+{
+     switch (state) {
+     case VFIO_DEVICE_STATE_ERROR:
+             return "VFIO_DEVICE_STATE_ERROR";
+     case VFIO_DEVICE_STATE_STOP:
+             return "VFIO_DEVICE_STATE_STOP";
+     case VFIO_DEVICE_STATE_RUNNING:
+             return "VFIO_DEVICE_STATE_RUNNING";
+     case VFIO_DEVICE_STATE_STOP_COPY:
+             return "VFIO_DEVICE_STATE_STOP_COPY";
+     case VFIO_DEVICE_STATE_RESUMING:
+             return "VFIO_DEVICE_STATE_RESUMING";
+     case VFIO_DEVICE_STATE_RUNNING_P2P:
Quoted 8 lines

drivers/vfio/pci/pds/lm.c line 36

Quoted 6 lines
drivers/vfio/pci/pds/lm.c new file +6 -0
+             return "VFIO_DEVICE_STATE_RUNNING_P2P";
+     default:
+             return "VFIO_DEVICE_STATE_INVALID";
+     }
+
+     return "VFIO_DEVICE_STATE_INVALID";

Seems a tad redundant.

I can remove this on the next revision. However, it is nice to be able to see the state transitions via enabling the dynamic debug statement in pds_vfio_step_device_state_locked(), but maybe this would make more sense in the layer above the device specific VFIO drivers?

Quoted 13 lines
Quoted 3 lines
drivers/vfio/pci/pds/lm.c new file +2 -0
+}
+

[snip]

Quoted 9 lines
drivers/vfio/pci/pds/lm.c new file +9 -0
+struct file *
+pds_vfio_step_device_state_locked(struct pds_vfio_pci_device *pds_vfio,
+                               enum vfio_device_mig_state next)
+{
+     enum vfio_device_mig_state cur = pds_vfio->state;
+     struct device *dev = &pds_vfio->pdev->dev;
+     int err = 0;
+
+     dev_dbg(dev, "%s => %s\n",
Quoted 8 lines

drivers/vfio/pci/pds/lm.c line 402

Quoted 6 lines
drivers/vfio/pci/pds/lm.c new file +6 -0
+             pds_vfio_lm_state(cur), pds_vfio_lm_state(next));
+
+     if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_STOP_COPY) {
+             /* Device is already stopped
+              * create save device data file & get device state from firmware
+              */

Standard multi-line comment style please, we're not under net/ here.

Will fix. Thanks.

Quoted 64 lines
Quoted 46 lines
drivers/vfio/pci/pds/lm.c new file +45 -0
+             err = pds_vfio_get_save_file(pds_vfio);
+             if (err)
+                     return ERR_PTR(err);
+
+             /* Get device state */
+             err = pds_vfio_get_lm_state_cmd(pds_vfio);
+             if (err) {
+                     pds_vfio_put_save_file(pds_vfio);
+                     return ERR_PTR(err);
+             }
+
+             return pds_vfio->save_file->filep;
+     }
+
+     if (cur == VFIO_DEVICE_STATE_STOP_COPY && next == VFIO_DEVICE_STATE_STOP) {
+             /* Device is already stopped
+              * delete the save device state file
+              */
+             pds_vfio_put_save_file(pds_vfio);
+             pds_vfio_send_host_vf_lm_status_cmd(pds_vfio,
+                                                 PDS_LM_STA_NONE);
+             return NULL;
+     }
+
+     if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RESUMING) {
+             /* create resume device data file */
+             err = pds_vfio_get_restore_file(pds_vfio);
+             if (err)
+                     return ERR_PTR(err);
+
+             return pds_vfio->restore_file->filep;
+     }
+
+     if (cur == VFIO_DEVICE_STATE_RESUMING && next == VFIO_DEVICE_STATE_STOP) {
+             /* Set device state */
+             err = pds_vfio_set_lm_state_cmd(pds_vfio);
+             if (err)
+                     return ERR_PTR(err);
+
+             /* delete resume device data file */
+             pds_vfio_put_restore_file(pds_vfio);
+             return NULL;
+     }
+
+     if (cur == VFIO_DEVICE_STATE_RUNNING && next == VFIO_DEVICE_STATE_RUNNING_P2P) {

drivers/vfio/pci/pds/lm.c line 457

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+             pds_vfio_send_host_vf_lm_status_cmd(pds_vfio, PDS_LM_STA_IN_PROGRESS);
+             /* Device should be stopped
+              * no interrupts, dma or change in internal state
+              */
+             err = pds_vfio_suspend_device_cmd(pds_vfio);
+             if (err)
+                     return ERR_PTR(err);
+
+             return NULL;
+     }

The comment here, as well as the no-op transitions from STOP->P2P and P2P->STOP has me concerned whether a device in this suspend state really meets the definition of our P2P state. The RUNNING_P2P state is a quiescent state where the device must accept access, including peer-to-peer DMA, but it cannot initiate DMA or interrupts. Is that consistent with this suspend state? Thanks,

Quoted 1 line

Alex

Yeah, these comments are misleading. Our device in the suspend state does meet the definition of the P2P and STOP states based on how you described them here and how they are defined in the VFIO header file.

I will remove the comments here as they are redundant since the states align with the documentation.

Thanks,

Brett

Quoted 28 lines
Quoted 13 lines
drivers/vfio/pci/pds/lm.c new file +12 -0
+
+     if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_RUNNING) {
+             /* Device should be functional
+              * interrupts, dma, mmio or changes to internal state is allowed
+              */
+             err = pds_vfio_resume_device_cmd(pds_vfio);
+             if (err)
+                     return ERR_PTR(err);
+
+             pds_vfio_send_host_vf_lm_status_cmd(pds_vfio,
+                                                 PDS_LM_STA_NONE);
+             return NULL;

drivers/vfio/pci/pds/lm.c line 479

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+     }
+
+     if (cur == VFIO_DEVICE_STATE_STOP && next == VFIO_DEVICE_STATE_RUNNING_P2P)
+             return NULL;
+
+     if (cur == VFIO_DEVICE_STATE_RUNNING_P2P && next == VFIO_DEVICE_STATE_STOP)
+             return NULL;
+
+     return ERR_PTR(-EINVAL);
+}

--
You received this message because you are subscribed to the Google Groups "Pensando Drivers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
To view this discussion on the web visit https://groups.google.com/a/pensando.io/d/msgid/drivers/20230410160538.35c1a5a6.alex.williamson%40redhat.com.

#13 lore raw

Re: [PATCH v8 vfio 4/7] vfio/pds: Add VFIO live migration support

In reply to Brett Creeley

drivers/vfio/pci/pds/cmds.c line 26

Quoted 1 line · Brett Creeley
drivers/vfio/pci/pds/cmds.c +1 -0
+	union pds_core_adminq_cmd cmd = { 0 };

These should all be = {}, adding the 0 is a subtly different thing in

Quoted 8 lines
drivers/vfio/pci/pds/cmds.c +8 -0
+int
+pds_vfio_suspend_device_cmd(struct pds_vfio_pci_device *pds_vfio)
+{
+	struct pds_lm_suspend_cmd cmd = {
+		.opcode = PDS_LM_CMD_SUSPEND,
+		.vf_id = cpu_to_le16(pds_vfio->vf_id),
+	};
+	struct pds_lm_suspend_comp comp = {0};

drivers/vfio/pci/pds/cmds.c pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio) · line 161

Quoted 10 lines
drivers/vfio/pci/pds/cmds.c +10 -0
+	struct pci_dev *pdev = pds_vfio->pdev;
+	int err;
+
+	dev_dbg(&pdev->dev, "vf%u: Suspend device\n", pds_vfio->vf_id);
+
+	err = pds_client_adminq_cmd(pds_vfio,
+				    (union pds_core_adminq_cmd *)&cmd,
+				    sizeof(cmd),
+				    (union pds_core_adminq_comp *)&comp,
+				    PDS_AQ_FLAG_FASTPOLL);

These casts to a union are really weird, why isn't the union the type on the stack?

Jason

Quoted 13 lines
drivers/vfio/pci/pds/cmds.c +13 -0
+static int
+pds_vfio_dma_map_lm_file(struct device *dev, enum dma_data_direction dir,
+			 struct pds_vfio_lm_file *lm_file)
+{
+	struct pds_lm_sg_elem *sgl, *sge;
+	struct scatterlist *sg;
+	int err;
+	int i;
+
+	if (!lm_file)
+		return -EINVAL;
+
+	/* dma map file pages */

drivers/vfio/pci/pds/cmds.c pds_vfio_unregister_client_cmd(struct pds_vfio_pci_device *pds_vfio) · line 236

Quoted 10 lines
drivers/vfio/pci/pds/cmds.c +10 -0
+	err = dma_map_sgtable(dev, &lm_file->sg_table, dir, 0);
+	if (err)
+		return err;
+
+	lm_file->num_sge = lm_file->sg_table.nents;
+
+	/* alloc sgl */
+	sgl = dma_alloc_coherent(dev, lm_file->num_sge *
+				 sizeof(struct pds_lm_sg_elem),
+				 &lm_file->sgl_addr, GFP_KERNEL);

Do you really need a coherent allocation for this?

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +4 -0
new file mode 100644
index 000000000000..855f5da9b99a
@@ -0,0 +1,479 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright(c) 2023 Advanced Micro Devices, Inc. */
+
+#include <linux/anon_inodes.h>

drivers/vfio/pci/pds/lm.c line 14

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+#include <linux/file.h>
+#include <linux/fs.h>
+#include <linux/highmem.h>
+#include <linux/vfio.h>
+#include <linux/vfio_pci_core.h>
+
+#include "vfio_dev.h"
+#include "cmds.h"
+
+#define PDS_VFIO_LM_FILENAME	"pds_vfio_lm"

This doesn't need a define, it is typical to write the pseudo filename in the only anon_inode_getfile()

Quoted 39 lines
drivers/vfio/pci/pds/lm.c new file +39 -0
+static struct pds_vfio_lm_file *
+pds_vfio_get_lm_file(const char *name, const struct file_operations *fops,
+		     int flags, u64 size)
+{
+	struct pds_vfio_lm_file *lm_file = NULL;
+	unsigned long long npages;
+	struct page **pages;
+	int err = 0;
+
+	if (!size)
+		return NULL;
+
+	/* Alloc file structure */
+	lm_file = kzalloc(sizeof(*lm_file), GFP_KERNEL);
+	if (!lm_file)
+		return NULL;
+
+	/* Create file */
+	lm_file->filep = anon_inode_getfile(name, fops, lm_file, flags);
+	if (!lm_file->filep)
+		goto err_get_file;
+
+	stream_open(lm_file->filep->f_inode, lm_file->filep);
+	mutex_init(&lm_file->lock);
+
+	lm_file->size = size;
+
+	/* Allocate memory for file pages */
+	npages = DIV_ROUND_UP_ULL(lm_file->size, PAGE_SIZE);
+
+	pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
+	if (!pages)
+		goto err_alloc_pages;
+
+	for (unsigned long long i = 0; i < npages; i++) {
+		pages[i] = alloc_page(GFP_KERNEL);
+		if (!pages[i])
+			goto err_alloc_page;
+	}

drivers/vfio/pci/pds/lm.c line 87

Quoted 10 lines
drivers/vfio/pci/pds/lm.c new file +10 -0
+
+	lm_file->pages = pages;
+	lm_file->npages = npages;
+	lm_file->alloc_size = npages * PAGE_SIZE;
+
+	/* Create scatterlist of file pages to use for DMA mapping later */
+	err = sg_alloc_table_from_pages(&lm_file->sg_table, pages, npages,
+					0, size, GFP_KERNEL);
+	if (err)
+		goto err_alloc_sg_table;

This is the same basic thing the mlx5 driver does, you should move the mlx5 code into some common place and just re-use it here.

Quoted 5 lines
drivers/vfio/pci/pds/vfio_dev.h +0 -0
index 10557e8dc829..3f55861ffc7c 100644

drivers/vfio/pci/pds/vfio_dev.h line 16

Quoted 10 lines
drivers/vfio/pci/pds/vfio_dev.h +3 -0
@@ -7,10 +7,20 @@
 #include <linux/pci.h>
 #include <linux/vfio_pci_core.h>
 
+#include "lm.h"
+
 struct pds_vfio_pci_device {
 	struct vfio_pci_core_device vfio_coredev;
 	struct pci_dev *pdev;
 	void *pdsc;
+	struct device *coredev;

Why? If this is just &pdev->dev it it doesn't need to be in the struct

And pdev is just vfio_coredev->pdev, don't need to duplicate it either

Jason

author maintainer
#14 lore raw

PATCH v8 5/7 vfio/pds: Add support for dirty page tracking

In reply to Brett Creeley

In order to support dirty page tracking, the driver has to implement the VFIO subsystem's vfio_log_ops. This includes log_start, log_stop, and log_read_and_clear.

All of the tracker resources are allocated and dirty tracking on the device is started during log_start. The resources are cleaned up and dirty tracking on the device is stopped during log_stop. The dirty pages are determined and reported during log_read_and_clear.

In order to support these callbacks admin queue commands are used. All of the adminq queue command structures and implementations are included as part of this patch.

PDS_LM_CMD_DIRTY_STATUS is added to query the current status of dirty tracking on the device. This includes if it's enabled (i.e. number of regions being tracked from the device's perspective) and the maximum number of regions supported from the device's perspective.

PDS_LM_CMD_DIRTY_ENABLE is added to enable dirty tracking on the specified number of regions and their iova ranges.

PDS_LM_CMD_DIRTY_DISABLE is added to disable dirty tracking for all regions on the device.

PDS_LM_CMD_READ_SEQ and PDS_LM_CMD_DIRTY_WRITE_ACK are added to support reading and acknowledging the currently dirtied pages.

---
 drivers/vfio/pci/pds/Makefile   |   1 +
 drivers/vfio/pci/pds/cmds.c     | 139 +++++++++
 drivers/vfio/pci/pds/cmds.h     |   9 +
 drivers/vfio/pci/pds/dirty.c    | 536 ++++++++++++++++++++++++++++++++
 drivers/vfio/pci/pds/dirty.h    |  45 +++
 drivers/vfio/pci/pds/lm.c       |   3 +-
 drivers/vfio/pci/pds/vfio_dev.c |  10 +
 drivers/vfio/pci/pds/vfio_dev.h |   2 +
 include/linux/pds/pds_lm.h      | 173 +++++++++++
 9 files changed, 916 insertions(+), 2 deletions(-)
 create mode 100644 drivers/vfio/pci/pds/dirty.c
 create mode 100644 drivers/vfio/pci/pds/dirty.h
drivers/vfio/pci/pds/Makefile +1 -0
index dbaf613d3794..805176f7be9f 100644
@@ -5,6 +5,7 @@ obj-$(CONFIG_PDS_VFIO_PCI) += pds_vfio.o
 
 pds_vfio-y := \
 	cmds.o		\
+	dirty.o		\
 	lm.o		\
 	pci_drv.o	\
 	vfio_dev.o
drivers/vfio/pci/pds/cmds.c +139 -0

142 lines not on this page · raw message

drivers/vfio/pci/pds/cmds.h +9 -0
index 3d8a5508c733..fc1f4ae611eb 100644
@@ -13,4 +13,13 @@ int pds_vfio_get_lm_state_cmd(struct pds_vfio_pci_device *pds_vfio);
 int pds_vfio_set_lm_state_cmd(struct pds_vfio_pci_device *pds_vfio);
 void pds_vfio_send_host_vf_lm_status_cmd(struct pds_vfio_pci_device *pds_vfio,
 					 enum pds_lm_host_vf_status vf_status);
+int pds_vfio_dirty_status_cmd(struct pds_vfio_pci_device *pds_vfio,
+			      u64 regions_dma, u8 *max_regions,
+			      u8 *num_regions);
+int pds_vfio_dirty_enable_cmd(struct pds_vfio_pci_device *pds_vfio,
+			      u64 regions_dma, u8 num_regions);
+int pds_vfio_dirty_disable_cmd(struct pds_vfio_pci_device *pds_vfio);
+int pds_vfio_dirty_seq_ack_cmd(struct pds_vfio_pci_device *pds_vfio,
+			       u64 sgl_dma, u16 num_sge, u32 offset,
+			       u32 total_len, bool read_seq);
 #endif /* _CMDS_H_ */
drivers/vfio/pci/pds/dirty.c new file +536 -0

536 lines not on this page · raw message

drivers/vfio/pci/pds/dirty.h new file +45 -0

45 lines not on this page · raw message

drivers/vfio/pci/pds/lm.c +1 -2
index 855f5da9b99a..6f38b9731390 100644
@@ -419,8 +419,7 @@ pds_vfio_step_device_state_locked(struct pds_vfio_pci_device *pds_vfio,
 		 * delete the save device state file
 		 */
 		pds_vfio_put_save_file(pds_vfio);
-		pds_vfio_send_host_vf_lm_status_cmd(pds_vfio,
-						    PDS_LM_STA_NONE);
+		pds_vfio_dirty_disable(pds_vfio);
 		return NULL;
 	}
 
drivers/vfio/pci/pds/vfio_dev.c +10 -0
index e6d431347c26..86f5cbc7eb49 100644
@@ -5,6 +5,7 @@
 #include <linux/vfio_pci_core.h>
 
 #include "lm.h"
+#include "dirty.h"
 #include "vfio_dev.h"
 
 struct pds_vfio_pci_device *
@@ -113,6 +114,13 @@ pds_vfio_lm_ops = {
 	.migration_get_data_size = pds_vfio_get_device_state_size
 };
 
+static const struct vfio_log_ops
+pds_vfio_log_ops = {
+	.log_start = pds_vfio_dma_logging_start,
+	.log_stop = pds_vfio_dma_logging_stop,
+	.log_read_and_clear = pds_vfio_dma_logging_report,
+};
+
 static int
 pds_vfio_init_device(struct vfio_device *vdev)
 {
@@ -131,6 +139,7 @@ pds_vfio_init_device(struct vfio_device *vdev)
 
 	vdev->migration_flags = VFIO_MIGRATION_STOP_COPY | VFIO_MIGRATION_P2P;
 	vdev->mig_ops = &pds_vfio_lm_ops;
+	vdev->log_ops = &pds_vfio_log_ops;
 
 	dev_dbg(&pdev->dev, "%s: PF %#04x VF %#04x (%d) vf_id %d domain %d pds_vfio %p\n",
 		__func__, pci_dev_id(pdev->physfn),
@@ -169,6 +178,7 @@ pds_vfio_close_device(struct vfio_device *vdev)
 		container_of(vdev, struct pds_vfio_pci_device,
 			     vfio_coredev.vdev);
 
+	pds_vfio_dirty_disable(pds_vfio);
 	mutex_destroy(&pds_vfio->state_mutex);
 	vfio_pci_core_close_device(vdev);
 }
drivers/vfio/pci/pds/vfio_dev.h +2 -0
index 3f55861ffc7c..7c968a1214d6 100644
@@ -7,6 +7,7 @@
 #include <linux/pci.h>
 #include <linux/vfio_pci_core.h>
 
+#include "dirty.h"
 #include "lm.h"
 
 struct pds_vfio_pci_device {
@@ -17,6 +18,7 @@ struct pds_vfio_pci_device {
 
 	struct pds_vfio_lm_file *save_file;
 	struct pds_vfio_lm_file *restore_file;
+	struct pds_vfio_dirty dirty;
 	struct mutex state_mutex; /* protect migration state */
 	enum vfio_device_mig_state state;
 	spinlock_t reset_lock; /* protect reset_done flow */
include/linux/pds/pds_lm.h +173 -0

183 lines not on this page · raw message

Signature · 1 line
2.17.1
#15 lore raw

Re: [PATCH v8 vfio 5/7] vfio/pds: Add support for dirty page tracking

Quoted 40 lines · On Tue, 4 Apr 2023 12:01:39 -0700 Brett Creeley <[email protected]> wrote:
drivers/vfio/pci/pds/dirty.c new file +40 -0
+static int
+pds_vfio_dirty_enable(struct pds_vfio_pci_device *pds_vfio,
+		      struct rb_root_cached *ranges, u32 nnodes,
+		      u64 *page_size)
+{
+	struct pds_vfio_dirty *dirty = &pds_vfio->dirty;
+	u64 region_start, region_size, region_page_size;
+	struct pds_lm_dirty_region_info *region_info;
+	struct interval_tree_node *node = NULL;
+	struct pci_dev *pdev = pds_vfio->pdev;
+	u8 max_regions = 0, num_regions;
+	dma_addr_t regions_dma = 0;
+	u32 num_ranges = nnodes;
+	u32 page_count;
+	u16 len;
+	int err;
+
+	dev_dbg(&pdev->dev, "vf%u: Start dirty page tracking\n", pds_vfio->vf_id);
+
+	if (pds_vfio_dirty_is_enabled(pds_vfio))
+		return -EINVAL;
+
+	pds_vfio_dirty_set_enabled(pds_vfio);
+
+	/* find if dirty tracking is disabled, i.e. num_regions == 0 */
+	err = pds_vfio_dirty_status_cmd(pds_vfio, 0, &max_regions, &num_regions);
+	if (num_regions) {
+		dev_err(&pdev->dev, "Dirty tracking already enabled for %d regions\n",
+			num_regions);
+		err = -EEXIST;
+		goto err_out;
+	} else if (!max_regions) {
+		dev_err(&pdev->dev, "Device doesn't support dirty tracking, max_regions %d\n",
+			max_regions);
+		err = -EOPNOTSUPP;
+		goto err_out;
+	} else if (err) {
+		dev_err(&pdev->dev, "Failed to get dirty status, err %pe\n",
+			ERR_PTR(err));
+		goto err_out;

drivers/vfio/pci/pds/dirty.c line 229

Quoted 10 lines
drivers/vfio/pci/pds/dirty.c new file +10 -0
+	}
+
+	/* Only support 1 region for now. If there are any large gaps in the
+	 * VM's address regions, then this would be a waste of memory as we are
+	 * generating 2 bitmaps (ack/seq) from the min address to the max
+	 * address of the VM's address regions. In the future, if we support
+	 * more than one region in the device/driver we can split the bitmaps
+	 * on the largest address region gaps. We can do this split up to the
+	 * max_regions times returned from the dirty_status command.
+	 */

Large gaps in a VM are very possible, particularly after QEMU has relocated RAM above 4GB to above the reserved hypertransport range on AMD systems. Thanks,

Alex

Quoted 79 lines
drivers/vfio/pci/pds/dirty.c new file +79 -0
+	max_regions = 1;
+	if (num_ranges > max_regions) {
+		vfio_combine_iova_ranges(ranges, nnodes, max_regions);
+		num_ranges = max_regions;
+	}
+
+	node = interval_tree_iter_first(ranges, 0, ULONG_MAX);
+	if (!node) {
+		err = -EINVAL;
+		goto err_out;
+	}
+
+	region_size = node->last - node->start + 1;
+	region_start = node->start;
+	region_page_size = *page_size;
+
+	len = sizeof(*region_info);
+	region_info = kzalloc(len, GFP_KERNEL);
+	if (!region_info) {
+		err = -ENOMEM;
+		goto err_out;
+	}
+
+	page_count = DIV_ROUND_UP(region_size, region_page_size);
+
+	region_info->dma_base = cpu_to_le64(region_start);
+	region_info->page_count = cpu_to_le32(page_count);
+	region_info->page_size_log2 = ilog2(region_page_size);
+
+	regions_dma = dma_map_single(pds_vfio->coredev, (void *)region_info, len,
+				     DMA_BIDIRECTIONAL);
+	if (dma_mapping_error(pds_vfio->coredev, regions_dma)) {
+		err = -ENOMEM;
+		kfree(region_info);
+		goto err_out;
+	}
+
+	err = pds_vfio_dirty_enable_cmd(pds_vfio, regions_dma, max_regions);
+	dma_unmap_single(pds_vfio->coredev, regions_dma, len, DMA_BIDIRECTIONAL);
+	/* page_count might be adjusted by the device,
+	 * update it before freeing region_info DMA
+	 */
+	page_count = le32_to_cpu(region_info->page_count);
+
+	dev_dbg(&pdev->dev, "region_info: regions_dma 0x%llx dma_base 0x%llx page_count %u page_size_log2 %u\n",
+		regions_dma, region_start, page_count, (u8)ilog2(region_page_size));
+
+	kfree(region_info);
+	if (err)
+		goto err_out;
+
+	err = pds_vfio_dirty_alloc_bitmaps(dirty, page_count);
+	if (err) {
+		dev_err(&pdev->dev, "Failed to alloc dirty bitmaps: %pe\n",
+			ERR_PTR(err));
+		goto err_out;
+	}
+
+	err = pds_vfio_dirty_alloc_sgl(pds_vfio, page_count);
+	if (err) {
+		dev_err(&pdev->dev, "Failed to alloc dirty sg lists: %pe\n",
+			ERR_PTR(err));
+		goto err_free_bitmaps;
+	}
+
+	dirty->region_start = region_start;
+	dirty->region_size = region_size;
+	dirty->region_page_size = region_page_size;
+
+	pds_vfio_print_guest_region_info(pds_vfio, max_regions);
+
+	return 0;
+
+err_free_bitmaps:
+	pds_vfio_dirty_free_bitmaps(dirty);
+err_out:
+	pds_vfio_dirty_set_disabled(pds_vfio);
+	return err;
+}
author maintainer
#16 lore raw

PATCH v8 6/7 vfio/pds: Add support for firmware recovery

In reply to Brett Creeley

It's possible that the device firmware crashes and is able to recover due to some configuration and/or other issue. If a live migration is in progress while the firmware crashes, the live migration will fail. However, the VF PCI device should still be functional post crash recovery and subsequent migrations should go through as expected.

When the pds_core device notices that firmware crashes it sends an event to all its client drivers. When the pds_vfio driver receives this event while migration is in progress it will request a deferred reset on the next migration state transition. This state transition will report failure as well as any subsequent state transition requests from the VMM/VFIO. Based on uapi/vfio.h the only way out of VFIO_DEVICE_STATE_ERROR is by issuing VFIO_DEVICE_RESET. Once this reset is done, the migration state will be reset to VFIO_DEVICE_STATE_RUNNING and migration can be performed.

If the event is received while no migration is in progress (i.e. the VM is in normal operating mode), then no actions are taken and the migration state remains VFIO_DEVICE_STATE_RUNNING.

---
 drivers/vfio/pci/pds/pci_drv.c  | 110 +++++++++++++++++++++++++++++++-
 drivers/vfio/pci/pds/vfio_dev.c |  34 +++++++++-
 drivers/vfio/pci/pds/vfio_dev.h |   6 +-
 3 files changed, 146 insertions(+), 4 deletions(-)
drivers/vfio/pci/pds/pci_drv.c +109 -1
index b0781d9f4246..b155ac9b98ae 100644
@@ -20,6 +20,104 @@
 #define PDS_VFIO_DRV_DESCRIPTION	"AMD/Pensando VFIO Device Driver"
 #define PCI_VENDOR_ID_PENSANDO		0x1dd8
 
+static void
+pds_vfio_recovery_work(struct work_struct *work)
+{
+	struct pds_vfio_pci_device *pds_vfio =
+		container_of(work, struct pds_vfio_pci_device, work);
+	bool deferred_reset_needed = false;
+
+	/* Documentation states that the kernel migration driver must not
+	 * generate asynchronous device state transitions outside of
+	 * manipulation by the user or the VFIO_DEVICE_RESET ioctl.
+	 *
+	 * Since recovery is an asynchronous event received from the device,
+	 * initiate a deferred reset. Only issue the deferred reset if a
+	 * migration is in progress, which will cause the next step of the
+	 * migration to fail. Also, if the device is in a state that will
+	 * be set to VFIO_DEVICE_STATE_RUNNING on the next action (i.e. VM is
+	 * shutdown and device is in VFIO_DEVICE_STATE_STOP) as that will clear
+	 * the VFIO_DEVICE_STATE_ERROR when the VM starts back up.
+	 */
+	mutex_lock(&pds_vfio->state_mutex);
+	if ((pds_vfio->state != VFIO_DEVICE_STATE_RUNNING &&
+	     pds_vfio->state != VFIO_DEVICE_STATE_ERROR) ||
+	    (pds_vfio->state == VFIO_DEVICE_STATE_RUNNING &&
+	     pds_vfio_dirty_is_enabled(pds_vfio)))
+		deferred_reset_needed = true;
+	mutex_unlock(&pds_vfio->state_mutex);
+
+	/* On the next user initiated state transition, the device will
+	 * transition to the VFIO_DEVICE_STATE_ERROR. At this point it's the user's
+	 * responsibility to reset the device.
+	 *
+	 * If a VFIO_DEVICE_RESET is requested post recovery and before the next
+	 * state transition, then the deferred reset state will be set to
+	 * VFIO_DEVICE_STATE_RUNNING.
+	 */
+	if (deferred_reset_needed)
+		pds_vfio_deferred_reset(pds_vfio, VFIO_DEVICE_STATE_ERROR);
+}
+
+static int
+pds_vfio_pci_notify_handler(struct notifier_block *nb,
+			    unsigned long ecode,
+			    void *data)
+{
+	struct pds_vfio_pci_device *pds_vfio = container_of(nb,
+							    struct pds_vfio_pci_device,
+							    nb);
+	union pds_core_notifyq_comp *event = data;
+
+	dev_info(pds_vfio->coredev, "%s: event code %lu\n", __func__, ecode);
+
+	/* We don't need to do anything for RESET state==0 as there is no notify
+	 * or feedback mechanism available, and it is possible that we won't
+	 * even see a state==0 event.
+	 *
+	 * Any requests from VFIO while state==0 will fail, which will return
+	 * error and may cause migration to fail.
+	 */
+	if (ecode == PDS_EVENT_RESET) {
+		dev_info(pds_vfio->coredev, "%s: PDS_EVENT_RESET event received, state==%d\n",
+			 __func__, event->reset.state);
+		if (event->reset.state == 1)
+			schedule_work(&pds_vfio->work);
+	}
+
+	return 0;
+}
+
+static int
+pds_vfio_pci_register_event_handler(struct pds_vfio_pci_device *pds_vfio)
+{
+	struct notifier_block *nb = &pds_vfio->nb;
+	int err;
+
+	if (!nb->notifier_call) {
+		nb->notifier_call = pds_vfio_pci_notify_handler;
+		err = pdsc_register_notify(nb);
+		if (err) {
+			nb->notifier_call = NULL;
+			dev_err(pds_vfio->coredev, "failed to register pds event handler: %pe\n",
+				ERR_PTR(err));
+			return -EINVAL;
+		}
+		dev_dbg(pds_vfio->coredev, "pds event handler registered\n");
+	}
+
+	return 0;
+}
+
+static void
+pds_vfio_pci_unregister_event_handler(struct pds_vfio_pci_device *pds_vfio)
+{
+	if (pds_vfio->nb.notifier_call) {
+		pdsc_unregister_notify(&pds_vfio->nb);
+		pds_vfio->nb.notifier_call = NULL;
+	}
+}
+
 static int
 pds_vfio_pci_probe(struct pci_dev *pdev,
 		   const struct pci_device_id *id)
@@ -44,14 +142,22 @@ pds_vfio_pci_probe(struct pci_dev *pdev,
 		goto out_put_vdev;
 	}
 
-	err = vfio_pci_core_register_device(&pds_vfio->vfio_coredev);
+	INIT_WORK(&pds_vfio->work, pds_vfio_recovery_work);
+	err = pds_vfio_pci_register_event_handler(pds_vfio);
 	if (err)
 		goto out_unreg_client;
 
+	err = vfio_pci_core_register_device(&pds_vfio->vfio_coredev);
+	if (err)
+		goto out_unreg_notify;
+
 	return 0;
 
+out_unreg_notify:
+	pds_vfio_pci_unregister_event_handler(pds_vfio);
 out_unreg_client:
 	pds_vfio_unregister_client_cmd(pds_vfio);
+	cancel_work_sync(&pds_vfio->work);
 out_put_vdev:
 	vfio_put_device(&pds_vfio->vfio_coredev.vdev);
 	return err;
@@ -62,6 +168,8 @@ pds_vfio_pci_remove(struct pci_dev *pdev)
 {
 	struct pds_vfio_pci_device *pds_vfio = pds_vfio_pci_drvdata(pdev);
 
+	pds_vfio_pci_unregister_event_handler(pds_vfio);
+	cancel_work_sync(&pds_vfio->work);
 	vfio_pci_core_unregister_device(&pds_vfio->vfio_coredev);
 	vfio_put_device(&pds_vfio->vfio_coredev.vdev);
 }
drivers/vfio/pci/pds/vfio_dev.c +32 -2
index 86f5cbc7eb49..f4ed0082a6c0 100644
@@ -25,10 +25,17 @@ pds_vfio_state_mutex_unlock(struct pds_vfio_pci_device *pds_vfio)
 	if (pds_vfio->deferred_reset) {
 		pds_vfio->deferred_reset = false;
 		if (pds_vfio->state == VFIO_DEVICE_STATE_ERROR) {
-			pds_vfio->state = VFIO_DEVICE_STATE_RUNNING;
+			dev_dbg(&pds_vfio->pdev->dev, "Transitioning from VFIO_DEVICE_STATE_ERROR to %s\n",
+				pds_vfio_lm_state(pds_vfio->deferred_reset_state));
+			pds_vfio->state = pds_vfio->deferred_reset_state;
 			pds_vfio_put_restore_file(pds_vfio);
 			pds_vfio_put_save_file(pds_vfio);
+		} else if (pds_vfio->deferred_reset_state == VFIO_DEVICE_STATE_ERROR) {
+			dev_dbg(&pds_vfio->pdev->dev, "Transitioning from %s to VFIO_DEVICE_STATE_ERROR based on deferred_reset request\n",
+				pds_vfio_lm_state(pds_vfio->state));
+			pds_vfio->state = VFIO_DEVICE_STATE_ERROR;
 		}
+		pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_RUNNING;
 		spin_unlock(&pds_vfio->reset_lock);
 		goto again;
 	}
@@ -41,6 +48,7 @@ pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio)
 {
 	spin_lock(&pds_vfio->reset_lock);
 	pds_vfio->deferred_reset = true;
+	pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_RUNNING;
 	if (!mutex_trylock(&pds_vfio->state_mutex)) {
 		spin_unlock(&pds_vfio->reset_lock);
 		return;
@@ -49,6 +57,18 @@ pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio)
 	pds_vfio_state_mutex_unlock(pds_vfio);
 }
 
+void
+pds_vfio_deferred_reset(struct pds_vfio_pci_device *pds_vfio,
+			enum vfio_device_mig_state reset_state)
+{
+	dev_info(&pds_vfio->pdev->dev, "Requesting deferred_reset to state %s\n",
+		 pds_vfio_lm_state(reset_state));
+	spin_lock(&pds_vfio->reset_lock);
+	pds_vfio->deferred_reset = true;
+	pds_vfio->deferred_reset_state = reset_state;
+	spin_unlock(&pds_vfio->reset_lock);
+}
+
 static struct file *
 pds_vfio_set_device_state(struct vfio_device *vdev,
 			  enum vfio_device_mig_state new_state)
@@ -59,7 +79,13 @@ pds_vfio_set_device_state(struct vfio_device *vdev,
 	struct file *res = NULL;
 
 	mutex_lock(&pds_vfio->state_mutex);
-	while (new_state != pds_vfio->state) {
+	/* only way to transition out of VFIO_DEVICE_STATE_ERROR is via
+	 * VFIO_DEVICE_RESET, so prevent the state machine from running since
+	 * vfio_mig_get_next_state() will throw a WARN_ON() when transitioning
+	 * from VFIO_DEVICE_STATE_ERROR to any other state
+	 */
+	while (pds_vfio->state != VFIO_DEVICE_STATE_ERROR &&
+	       new_state != pds_vfio->state) {
 		enum vfio_device_mig_state next_state;
 
 		int err = vfio_mig_get_next_state(vdev, pds_vfio->state,
@@ -81,6 +107,9 @@ pds_vfio_set_device_state(struct vfio_device *vdev,
 		}
 	}
 	pds_vfio_state_mutex_unlock(pds_vfio);
+	/* still waiting on a deferred_reset */
+	if (pds_vfio->state == VFIO_DEVICE_STATE_ERROR)
+		res = ERR_PTR(-EIO);
 
 	return res;
 }
@@ -165,6 +194,7 @@ pds_vfio_open_device(struct vfio_device *vdev)
 	dev_dbg(&pds_vfio->pdev->dev, "%s: %s => VFIO_DEVICE_STATE_RUNNING\n",
 		__func__, pds_vfio_lm_state(pds_vfio->state));
 	pds_vfio->state = VFIO_DEVICE_STATE_RUNNING;
+	pds_vfio->deferred_reset_state = VFIO_DEVICE_STATE_RUNNING;
 
 	vfio_pci_core_finish_enable(&pds_vfio->vfio_coredev);
 
drivers/vfio/pci/pds/vfio_dev.h +5 -1
index 7c968a1214d6..46a239c16e1b 100644
@@ -23,6 +23,9 @@ struct pds_vfio_pci_device {
 	enum vfio_device_mig_state state;
 	spinlock_t reset_lock; /* protect reset_done flow */
 	u8 deferred_reset;
+	enum vfio_device_mig_state deferred_reset_state;
+	struct work_struct work;
+	struct notifier_block nb;
 
 	int vf_id;
 	int pci_id;
@@ -32,5 +35,6 @@ struct pds_vfio_pci_device {
 const struct vfio_device_ops *pds_vfio_ops_info(void);
 struct pds_vfio_pci_device *pds_vfio_pci_drvdata(struct pci_dev *pdev);
 void pds_vfio_reset(struct pds_vfio_pci_device *pds_vfio);
-
+void pds_vfio_deferred_reset(struct pds_vfio_pci_device *pds_vfio,
+			     enum vfio_device_mig_state reset_state);
 #endif /* _VFIO_DEV_H_ */
Signature · 1 line
2.17.1
#17 lore raw

Re: [PATCH v8 vfio 6/7] vfio/pds: Add support for firmware recovery

Quoted 65 lines · Brett Creeley

It's possible that the device firmware crashes and is able to recover due to some configuration and/or other issue. If a live migration is in progress while the firmware crashes, the live migration will fail. However, the VF PCI device should still be functional post crash recovery and subsequent migrations should go through as expected.

When the pds_core device notices that firmware crashes it sends an event to all its client drivers. When the pds_vfio driver receives this event while migration is in progress it will request a deferred reset on the next migration state transition. This state transition will report failure as well as any subsequent state transition requests from the VMM/VFIO. Based on uapi/vfio.h the only way out of VFIO_DEVICE_STATE_ERROR is by issuing VFIO_DEVICE_RESET. Once this reset is done, the migration state will be reset to VFIO_DEVICE_STATE_RUNNING and migration can be performed.

If the event is received while no migration is in progress (i.e. the VM is in normal operating mode), then no actions are taken and the migration state remains VFIO_DEVICE_STATE_RUNNING.

---
 drivers/vfio/pci/pds/pci_drv.c  | 110 +++++++++++++++++++++++++++++++-
 drivers/vfio/pci/pds/vfio_dev.c |  34 +++++++++-
 drivers/vfio/pci/pds/vfio_dev.h |   6 +-
 3 files changed, 146 insertions(+), 4 deletions(-)
drivers/vfio/pci/pds/pci_drv.c +28 -0
index b0781d9f4246..b155ac9b98ae 100644
@@ -20,6 +20,104 @@
 #define PDS_VFIO_DRV_DESCRIPTION	"AMD/Pensando VFIO Device Driver"
 #define PCI_VENDOR_ID_PENSANDO		0x1dd8
 
+static void
+pds_vfio_recovery_work(struct work_struct *work)
+{
+	struct pds_vfio_pci_device *pds_vfio =
+		container_of(work, struct pds_vfio_pci_device, work);
+	bool deferred_reset_needed = false;
+
+	/* Documentation states that the kernel migration driver must not
+	 * generate asynchronous device state transitions outside of
+	 * manipulation by the user or the VFIO_DEVICE_RESET ioctl.
+	 *
+	 * Since recovery is an asynchronous event received from the device,
+	 * initiate a deferred reset. Only issue the deferred reset if a
+	 * migration is in progress, which will cause the next step of the
+	 * migration to fail. Also, if the device is in a state that will
+	 * be set to VFIO_DEVICE_STATE_RUNNING on the next action (i.e. VM is
+	 * shutdown and device is in VFIO_DEVICE_STATE_STOP) as that will clear
+	 * the VFIO_DEVICE_STATE_ERROR when the VM starts back up.
+	 */
+	mutex_lock(&pds_vfio->state_mutex);
+	if ((pds_vfio->state != VFIO_DEVICE_STATE_RUNNING &&
+	     pds_vfio->state != VFIO_DEVICE_STATE_ERROR) ||
+	    (pds_vfio->state == VFIO_DEVICE_STATE_RUNNING &&
+	     pds_vfio_dirty_is_enabled(pds_vfio)))
+		deferred_reset_needed = true;
+	mutex_unlock(&pds_vfio->state_mutex);
+
+	/* On the next user initiated state transition, the device will

drivers/vfio/pci/pds/pci_drv.c line 60

Quoted 10 lines
drivers/vfio/pci/pds/pci_drv.c +10 -0
+	 * transition to the VFIO_DEVICE_STATE_ERROR. At this point it's the user's
+	 * responsibility to reset the device.
+	 *
+	 * If a VFIO_DEVICE_RESET is requested post recovery and before the next
+	 * state transition, then the deferred reset state will be set to
+	 * VFIO_DEVICE_STATE_RUNNING.
+	 */
+	if (deferred_reset_needed)
+		pds_vfio_deferred_reset(pds_vfio, VFIO_DEVICE_STATE_ERROR);
+}

Why is this a work? it is threaded on a blocking_notifier_chain so it can call the mutex?

Why is the locking like this, can't you just call
pds_vfio_deferred_reset() under the mutex?

Jason

author maintainer
#18 lore raw

PATCH v8 7/7 vfio/pds: Add Kconfig and documentation

In reply to Brett Creeley

Add Kconfig entries and pds_vfio.rst. Also, add an entry in the MAINTAINERS file for this new driver.

It's not clear where documentation for vendor specific VFIO drivers should live, so just re-use the current amd ethernet location.

---
 .../device_drivers/ethernet/amd/pds_vfio.rst  | 79 +++++++++++++++++++
 .../device_drivers/ethernet/index.rst         |  1 +
 MAINTAINERS                                   |  7 ++
 drivers/vfio/pci/Kconfig                      |  2 +
 drivers/vfio/pci/pds/Kconfig                  | 20 +++++
 5 files changed, 109 insertions(+)
 create mode 100644 Documentation/networking/device_drivers/ethernet/amd/pds_vfio.rst
 create mode 100644 drivers/vfio/pci/pds/Kconfig
Documentation/networking/device_drivers/ethernet/amd/pds_vfio.rst new file +79 -0
new file mode 100644
index 000000000000..7bddde0c7c9d
@@ -0,0 +1,79 @@
+.. SPDX-License-Identifier: GPL-2.0+
+.. note: can be edited and viewed with /usr/bin/formiko-vim
+
+==========================================================
+PCI VFIO driver for the AMD/Pensando(R) DSC adapter family
+==========================================================
+
+AMD/Pensando Linux VFIO PCI Device Driver
+Copyright(c) 2023 Advanced Micro Devices, Inc.
+
+Overview
+========
+
+The ``pds_vfio`` module is a PCI driver that supports Live Migration
+capable Virtual Function (VF) devices in the DSC hardware.
+
+Using the device
+================
+
+The pds_vfio device is enabled via multiple configuration steps and
+depends on the ``pds_core`` driver to create and enable SR-IOV Virtual
+Function devices.
+
+Shown below are the steps to bind the driver to a VF and also to the
+associated auxiliary device created by the ``pds_core`` driver. This
+example assumes the pds_core and pds_vfio modules are already
+loaded.
+
+.. code-block:: bash
+  :name: example-setup-script
+
+  #!/bin/bash
+
+  PF_BUS="0000:60"
+  PF_BDF="0000:60:00.0"
+  VF_BDF="0000:60:00.1"
+
+  # Prevent non-vfio VF driver from probing the VF device
+  echo 0 > /sys/class/pci_bus/$PF_BUS/device/$PF_BDF/sriov_drivers_autoprobe
+
+  # Create single VF for Live Migration via VFIO
+  echo 1 > /sys/bus/pci/drivers/pds_core/$PF_BDF/sriov_numvfs
+
+  # Allow the VF to be bound to the pds_vfio driver
+  echo "pds_vfio" > /sys/class/pci_bus/$PF_BUS/device/$VF_BDF/driver_override
+
+  # Bind the VF to the pds_vfio driver
+  echo "$VF_BDF" > /sys/bus/pci/drivers/pds_vfio/bind
+
+After performing the steps above, a file in /dev/vfio/<iommu_group>
+should have been created.
+
+
+Enabling the driver
+===================
+
+The driver is enabled via the standard kernel configuration system,
+using the make command::
+
+  make oldconfig/menuconfig/etc.
+
+The driver is located in the menu structure at:
+
+  -> Device Drivers
+    -> VFIO Non-Privileged userspace driver framework
+      -> VFIO support for PDS PCI devices
+
+Support
+=======
+
+For general Linux networking support, please use the netdev mailing
+list, which is monitored by Pensando personnel::
+
+  [email protected]
+
+For more specific support needs, please use the Pensando driver support
+email::
+
+  [email protected]
Documentation/networking/device_drivers/ethernet/index.rst +1 -0
index eaaf284e69e6..3b456406ff85 100644
@@ -14,6 +14,7 @@ Contents:
    3com/vortex
    amazon/ena
    amd/pds_core
+   amd/pds_vfio
    altera/altera_tse
    aquantia/atlantic
    chelsio/cxgb
MAINTAINERS +7 -0
index 97e2919baf4b..73b2b8225acc 100644
@@ -21959,6 +21959,13 @@ S:	Maintained
 P:	Documentation/driver-api/vfio-pci-device-specific-driver-acceptance.rst
 F:	drivers/vfio/pci/*/
 
+VFIO PDS PCI DRIVER
+M:	Brett Creeley <[email protected]>
+L:	[email protected]
+S:	Maintained
+F:	Documentation/networking/device_drivers/ethernet/amd/pds_vfio.rst
+F:	drivers/vfio/pci/pds/
+
 VFIO PLATFORM DRIVER
 M:	Eric Auger <[email protected]>
 L:	[email protected]
drivers/vfio/pci/Kconfig +2 -0
index f9d0c908e738..2c3831dd60ef 100644
@@ -59,4 +59,6 @@ source "drivers/vfio/pci/mlx5/Kconfig"
 
 source "drivers/vfio/pci/hisilicon/Kconfig"
 
+source "drivers/vfio/pci/pds/Kconfig"
+
 endif
drivers/vfio/pci/pds/Kconfig new file +20 -0
new file mode 100644
index 000000000000..149d4986bf43
@@ -0,0 +1,20 @@
+# SPDX-License-Identifier: GPL-2.0
+# Copyright (c) 2023 Advanced Micro Devices, Inc.
+
+config PDS_VFIO_PCI
+	tristate "VFIO support for PDS PCI devices"
+	depends on PDS_CORE
+	depends on VFIO_PCI_CORE
+	help
+	  This provides generic PCI support for PDS devices using the VFIO
+	  framework.
+
+	  More specific information on this driver can be
+	  found in
+	  <file:Documentation/networking/device_drivers/ethernet/amd/pds_vfio.rst>.
+
+	  To compile this driver as a module, choose M here. The module
+	  will be called pds_vfio.
+
+	  If you don't know what to do here, say N.
+
Signature · 1 line
2.17.1
#19 lore raw

Re: [PATCH v8 vfio 7/7] vfio/pds: Add Kconfig and documentation

Quoted 6 lines · Brett Creeley

Add Kconfig entries and pds_vfio.rst. Also, add an entry in the MAINTAINERS file for this new driver.

It's not clear where documentation for vendor specific VFIO drivers should live, so just re-use the current amd ethernet location.

It would be nice to make a kdoc section for vfio.

Jason

#20 lore raw
In reply to Brett Creeley
Quoted 6 lines · Brett Creeley

This is a patchset for a new vendor specific VFIO driver (pds_vfio) for use with the AMD/Pensando Distributed Services Card (DSC). This driver makes use of the newly introduced pds_core driver, which the latest version can be referenced at:

https://lore.kernel.org/netdev/[email protected]/

You don't need to post one of these every little change - the base netdev stuff and the use of aux dev still hasn't got a review from the relevant people, let alone this part..

Jason