On 10/11/2025 9:38 PM, Michał Winiarski wrote:
> Now that it's possible to free the packets - connect the restore
> handling logic with the ring.
> The helpers will also be used in upcoming changes that will start producing
> migration data packets.
>
> Signed-off-by: Michał Winiarski <[email protected]>
> ---
> drivers/gpu/drm/xe/Makefile | 1 +
> drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c | 48 ++++++-
> drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c | 10 +-
> drivers/gpu/drm/xe/xe_sriov_pf_migration.c | 1 +
> .../gpu/drm/xe/xe_sriov_pf_migration_data.c | 135 ++++++++++++++++++
> .../gpu/drm/xe/xe_sriov_pf_migration_data.h | 32 +++++
while this is used by the PF only, maybe those files don't have to include _pf_
tag (like xe_pci_sriov.c or xe_sriov_vfio.c ?)
.../gpu/drm/xe/xe_sriov_migration_data.c | 135 ++++++++++++++++++
.../gpu/drm/xe/xe_sriov_migration_data.h | 32 +++++
or
.../gpu/drm/xe/xe_sriov_vfio_data.c | 135 ++++++++++++++++++
.../gpu/drm/xe/xe_sriov_vfio_data.h | 32 +++++
> 6 files changed, 224 insertions(+), 3 deletions(-)
> create mode 100644 drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> create mode 100644 drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
>
> diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
> index 71f685a315dca..e253d65366de4 100644
> --- a/drivers/gpu/drm/xe/Makefile
> +++ b/drivers/gpu/drm/xe/Makefile
> @@ -177,6 +177,7 @@ xe-$(CONFIG_PCI_IOV) += \
> xe_sriov_pf_control.o \
> xe_sriov_pf_debugfs.o \
> xe_sriov_pf_migration.o \
> + xe_sriov_pf_migration_data.o \
> xe_sriov_pf_service.o \
> xe_tile_sriov_pf_debugfs.o
>
> diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
> b/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
> index 16a88e7599f6d..04a4e92133c2e 100644
> --- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
> +++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
> @@ -20,6 +20,7 @@
> #include "xe_sriov.h"
> #include "xe_sriov_pf_control.h"
> #include "xe_sriov_pf_migration.h"
> +#include "xe_sriov_pf_migration_data.h"
> #include "xe_sriov_pf_service.h"
> #include "xe_tile.h"
>
> @@ -949,14 +950,57 @@ static void pf_exit_vf_restored(struct xe_gt *gt,
> unsigned int vfid)
> pf_exit_vf_state(gt, vfid, XE_GT_SRIOV_STATE_RESTORED);
> }
>
> +static void pf_enter_vf_restore_failed(struct xe_gt *gt, unsigned int vfid)
> +{
> + pf_enter_vf_state(gt, vfid, XE_GT_SRIOV_STATE_RESTORE_FAILED);
> + pf_exit_vf_wip(gt, vfid);
> +}
> +
> +static int pf_handle_vf_restore_data(struct xe_gt *gt, unsigned int vfid,
> + struct xe_sriov_pf_migration_data *data)
> +{
> + switch (data->type) {
> + default:
> + xe_gt_sriov_notice(gt, "Skipping VF%u invalid data type: %d\n",
> vfid, data->type);
> + pf_enter_vf_restore_failed(gt, vfid);
shouldn't this be done in pf_handle_vf_restore_wip() where all other state
transitions are done?
> + }
> +
> + return -EINVAL;
> +}
> +
> static bool pf_handle_vf_restore_wip(struct xe_gt *gt, unsigned int vfid)
> {
> + struct xe_sriov_pf_migration_data *data;
> + int ret;
> +
> if (!pf_check_vf_state(gt, vfid, XE_GT_SRIOV_STATE_RESTORE_WIP))
in other places in VF control state machine we use slightly different pattern:
// can we exit state AAA?
if (!pf_exit_vf_state(AAA))
return false; // no, we are not in this state
// try to process other state
// yes, we _were_ in AAA, process this state
ret = handle_state_aaa();
// now decide where to go next
if (ret == -EAGAIN)
pf_enter_vf_state(AAA); // back
else if (ret < 0)
pf_enter_vf_state(AAA_FAILED) // failed
else
pf_enter_vf_state(AAA_DONE) // next
// state was processed, start next iteration
return true;
> return false;
>
> - pf_exit_vf_restore_wip(gt, vfid);
> - pf_enter_vf_restored(gt, vfid);
> + data = xe_gt_sriov_pf_migration_ring_consume(gt, vfid);
> + if (IS_ERR(data)) {
> + if (PTR_ERR(data) == -ENODATA &&
> + !xe_gt_sriov_pf_control_check_vf_data_wip(gt, vfid)) {
> + pf_exit_vf_restore_wip(gt, vfid);
> + pf_enter_vf_restored(gt, vfid);
> + } else {
> + pf_enter_vf_restore_failed(gt, vfid);
> + }
> + return false;
this should be 'true' as we completed this state processing
> + }
> +
> + xe_gt_assert(gt, gt->info.id == data->gt);
> + xe_gt_assert(gt, gt->tile->id == data->tile);
> +
> + ret = pf_handle_vf_restore_data(gt, vfid, data);
> + if (ret) {
> + xe_gt_sriov_err(gt, "VF%u failed to restore data type: %d
> (%d)\n",
use %pe for error
> + vfid, data->type, ret);
maybe for debug try to dump here more details about failing data packet
> + xe_sriov_pf_migration_data_free(data);
> + pf_enter_vf_restore_failed(gt, vfid);
> + return false;
> + }
>
> + xe_sriov_pf_migration_data_free(data);
> return true;
> }
>
> diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c
> b/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c
> index af5952f42fff1..582aaf062cbd4 100644
> --- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c
> +++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_migration.c
> @@ -15,6 +15,7 @@
> #include "xe_guc_ct.h"
> #include "xe_sriov.h"
> #include "xe_sriov_pf_migration.h"
> +#include "xe_sriov_pf_migration_data.h"
>
> #define XE_GT_SRIOV_PF_MIGRATION_RING_TIMEOUT (HZ * 20)
> #define XE_GT_SRIOV_PF_MIGRATION_RING_SIZE 5
> @@ -523,11 +524,18 @@ xe_gt_sriov_pf_migration_ring_consume_nowait(struct
> xe_gt *gt, unsigned int vfid
> return ERR_PTR(-EAGAIN);
> }
>
> +static void pf_mig_data_destroy(void *ptr)
> +{
> + struct xe_sriov_pf_migration_data *data = ptr;
> +
> + xe_sriov_pf_migration_data_free(data);
> +}
> +
> static void pf_gt_migration_cleanup(struct drm_device *dev, void *arg)
> {
> struct xe_gt_sriov_pf_migration *migration = arg;
>
> - ptr_ring_cleanup(&migration->ring, NULL);
> + ptr_ring_cleanup(&migration->ring, pf_mig_data_destroy);
> }
>
> /**
> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> b/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> index 347682f29a03c..d39cee66589b5 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> @@ -12,6 +12,7 @@
> #include "xe_pm.h"
> #include "xe_sriov_pf_helpers.h"
> #include "xe_sriov_pf_migration.h"
> +#include "xe_sriov_pf_migration_data.h"
> #include "xe_sriov_printk.h"
>
> static struct xe_sriov_pf_migration *pf_pick_migration(struct xe_device *xe,
> unsigned int vfid)
> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> new file mode 100644
> index 0000000000000..cfc6b512c6674
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> @@ -0,0 +1,135 @@
> +// SPDX-License-Identifier: MIT
> +/*
> + * Copyright © 2025 Intel Corporation
> + */
> +
> +#include "xe_bo.h"
> +#include "xe_device.h"
> +#include "xe_sriov_pf_migration_data.h"
> +
> +static bool data_needs_bo(struct xe_sriov_pf_migration_data *data)
> +{
> + unsigned int type = data->type;
> +
> + return type == XE_SRIOV_MIG_DATA_CCS ||
> + type == XE_SRIOV_MIG_DATA_VRAM;
> +}
> +
> +/**
> + * xe_sriov_pf_migration_data_alloc() - Allocate migration data packet
> + * @xe: the &struct xe_device
> + *
> + * Only allocates the "outer" structure, without initializing the migration
> + * data backing storage.
> + *
> + * Return: Pointer to &struct xe_sriov_pf_migration_data on success,
> + * NULL in case of error.
> + */
> +struct xe_sriov_pf_migration_data *
> +xe_sriov_pf_migration_data_alloc(struct xe_device *xe)
> +{
> + struct xe_sriov_pf_migration_data *data;
> +
> + data = kzalloc(sizeof(*data), GFP_KERNEL);
> + if (!data)
> + return NULL;
> +
> + data->xe = xe;
> + data->hdr_remaining = sizeof(data->hdr);
> +
> + return data;
> +}
> +
> +/**
> + * xe_sriov_pf_migration_data_free() - Free migration data packet
> + * @data: the &struct xe_sriov_pf_migration_data packet
> + */
> +void xe_sriov_pf_migration_data_free(struct xe_sriov_pf_migration_data *data)
> +{
> + if (data_needs_bo(data)) {
> + if (data->bo)
not needed, xe_bo_unpin_map_no_vm() checks for NULL
> + xe_bo_unpin_map_no_vm(data->bo);
> + } else {
> + if (data->buff)
not needed, kvfree() also checks for NULL
> + kvfree(data->buff);
> + }
> +
> + kfree(data);
> +}
> +
> +static int mig_data_init(struct xe_sriov_pf_migration_data *data)
> +{
> + struct xe_gt *gt = xe_device_get_gt(data->xe, data->gt);
> +
> + if (!gt || data->tile != gt->tile->id)
> + return -EINVAL;
didn't we check that already in xe_sriov_pf_migration_produce() ?
in other places we call xe_sriov_pf_migration_data_init() using ids from real
tile and gt
> +
> + if (data->size == 0)
> + return 0;
> +
> + if (data_needs_bo(data)) {
> + struct xe_bo *bo = xe_bo_create_pin_map_novm(data->xe, gt->tile,
> +
> PAGE_ALIGN(data->size),
> + ttm_bo_type_kernel,
> + XE_BO_FLAG_SYSTEM
> | XE_BO_FLAG_PINNED,
> + false);
> + if (IS_ERR(bo))
> + return PTR_ERR(bo);
> +
> + data->bo = bo;
> + data->vaddr = bo->vmap.vaddr;
> + } else {
> + void *buff = kvzalloc(data->size, GFP_KERNEL);
> + if (!buff)
> + return -ENOMEM;
> +
> + data->buff = buff;
> + data->vaddr = buff;
> + }
> +
> + return 0;
> +}
> +
> +/**
> + * xe_sriov_pf_migration_data_init() - Initialize the migration data header
> and backing storage
> + * @data: the &struct xe_sriov_pf_migration_data packet
> + * @tile_id: tile identifier
> + * @gt_id: GT identifier
> + * @type: &enum xe_sriov_pf_migration_data_type
here type is enum
> + * @offset: offset of data packet payload (within wider resource)
> + * @size: size of data packet payload
> + *
> + * Return: 0 on success or a negative error code on failure.
> + */
> +int xe_sriov_pf_migration_data_init(struct xe_sriov_pf_migration_data *data,
> u8 tile_id, u8 gt_id,
> + unsigned int type, loff_t offset, size_t
> size)
but here is plain int ?
> +{
> + xe_assert(data->xe, type < XE_SRIOV_MIG_DATA_MAX);
if it's "enum" - no need to check
if it's "int" and type is coming from outside of our code, then assert is not
sufficient anyway
nit: if assert stays, add sep line here
> + data->version = 1;
magic "1" needs its own #define
> + data->type = type;
> + data->tile = tile_id;
> + data->gt = gt_id;
> + data->offset = offset;
> + data->size = size;
> + data->remaining = size;
> +
> + return mig_data_init(data);
> +}
> +
> +/**
> + * xe_sriov_pf_migration_data_init() - Initialize the migration data backing
> storage based on header
> + * @data: the &struct xe_sriov_pf_migration_data packet
> + *
> + * Header data is expected to be filled prior to calling this function
> + *
> + * Return: 0 on success or a negative error code on failure.
> + */
> +int xe_sriov_pf_migration_data_init_from_hdr(struct
> xe_sriov_pf_migration_data *data)
> +{
> + if (WARN_ON(data->hdr_remaining))
better: xe_WARN_ON(xe, ....)
but does it really deserves any WARN here?
we rather know who is the caller
> + return -EINVAL;
> +
> + data->remaining = data->size;
> +
> + return mig_data_init(data);
> +}
> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> new file mode 100644
> index 0000000000000..1dde4cfcdbc47
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> @@ -0,0 +1,32 @@
> +/* SPDX-License-Identifier: MIT */
> +/*
> + * Copyright © 2025 Intel Corporation
> + */
> +
> +#ifndef _XE_SRIOV_PF_MIGRATION_DATA_H_
> +#define _XE_SRIOV_PF_MIGRATION_DATA_H_
> +
> +#include <linux/types.h>
> +
> +struct xe_device;
> +
> +enum xe_sriov_pf_migration_data_type {
maybe add a note that default 0 was skipped on purpose to catch
uninitialized/invalid data
> + XE_SRIOV_MIG_DATA_DESCRIPTOR = 1,
shouldn't we try to match enumerator names with enum name?
XE_SRIOV_PF_MIGRATION_DATA_TYPE_DESCRIPTOR = 1,
XE_SRIOV_PF_MIGRATION_DATA_TYPE_TRAILER,
XE_SRIOV_PF_MIGRATION_DATA_TYPE_...,
or change the enum (and file) name:
xe_sriov_migration_data.c
XE_SRIOV_MIGRATION_DATA_TYPE_DESCRIPTOR = 1,
XE_SRIOV_MIGRATION_DATA_TYPE_TRAILER,
XE_SRIOV_MIGRATION_DATA_TYPE_...,
> + XE_SRIOV_MIG_DATA_TRAILER,
> + XE_SRIOV_MIG_DATA_GGTT,
> + XE_SRIOV_MIG_DATA_MMIO,
> + XE_SRIOV_MIG_DATA_GUC,
> + XE_SRIOV_MIG_DATA_CCS,
> + XE_SRIOV_MIG_DATA_VRAM,
> + XE_SRIOV_MIG_DATA_MAX,
please drop it
> +};
> +
> +struct xe_sriov_pf_migration_data *
> +xe_sriov_pf_migration_data_alloc(struct xe_device *xe);
> +void xe_sriov_pf_migration_data_free(struct xe_sriov_pf_migration_data
> *snapshot);
> +
> +int xe_sriov_pf_migration_data_init(struct xe_sriov_pf_migration_data *data,
> u8 tile_id, u8 gt_id,
> + unsigned int type, loff_t offset, size_t
> size);
> +int xe_sriov_pf_migration_data_init_from_hdr(struct
> xe_sriov_pf_migration_data *snapshot);
> +
> +#endif
