On 10/11/2025 9:38 PM, Michał Winiarski wrote:
> The descriptor reuses the KLV format used by GuC and contains metadata
> that can be used to quickly fail migration when source is incompatible
> with destination.
>
> Signed-off-by: Michał Winiarski <[email protected]>
> ---
> drivers/gpu/drm/xe/xe_sriov_pf_migration.c | 6 +-
> .../gpu/drm/xe/xe_sriov_pf_migration_data.c | 82 ++++++++++++++++++-
> .../gpu/drm/xe/xe_sriov_pf_migration_data.h | 2 +
> 3 files changed, 87 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> b/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> index 9cc178126cbdc..a0cfac456ba0b 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_migration.c
> @@ -186,10 +186,14 @@ xe_sriov_pf_migration_consume(struct xe_device *xe,
> unsigned int vfid)
> static int pf_handle_descriptor(struct xe_device *xe, unsigned int vfid,
> struct xe_sriov_pf_migration_data *data)
> {
> + int ret;
> +
> if (data->tile != 0 || data->gt != 0)
> return -EINVAL;
>
> - xe_sriov_pf_migration_data_free(data);
> + ret = xe_sriov_pf_migration_data_process_desc(xe, vfid, data);
> + if (ret)
> + return ret;
>
> return 0;
> }
> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> index 9a2777dcf9a6b..307b16b027a5e 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.c
> @@ -5,6 +5,7 @@
>
> #include "xe_bo.h"
> #include "xe_device.h"
> +#include "xe_guc_klv_helpers.h"
> #include "xe_sriov_pf_helpers.h"
> #include "xe_sriov_pf_migration.h"
> #include "xe_sriov_pf_migration_data.h"
> @@ -404,11 +405,17 @@ ssize_t xe_sriov_pf_migration_data_write(struct
> xe_device *xe, unsigned int vfid
> return produced;
> }
>
> -#define MIGRATION_DESC_SIZE 4
> +#define MIGRATION_KLV_DEVICE_DEVID_KEY 0xf001u
> +#define MIGRATION_KLV_DEVICE_DEVID_LEN 1u
> +#define MIGRATION_KLV_DEVICE_REVID_KEY 0xf002u
> +#define MIGRATION_KLV_DEVICE_REVID_LEN 1u
as we aim to have unique KLVs across GuC ABI, maybe we should ask GuC team to
reserve some KLVs range (0xf000-0xffff) for our (driver) use ?
> +
> +#define MIGRATION_DESC_DWORDS 4
maybe:
(GUC_KLV_LEN_MIN + MIGRATION_KLV_DEVICE_DEVID_LEN +
GUC_KLV_LEN_MIN + MIGRATION_KLV_DEVICE_REVID_LEN)
> static size_t pf_desc_init(struct xe_device *xe, unsigned int vfid)
> {
> struct xe_sriov_pf_migration_data **desc = pf_pick_descriptor(xe, vfid);
> struct xe_sriov_pf_migration_data *data;
> + u32 *klvs;
> int ret;
>
> data = xe_sriov_pf_migration_data_alloc(xe);
> @@ -416,17 +423,88 @@ static size_t pf_desc_init(struct xe_device *xe,
> unsigned int vfid)
> return -ENOMEM;
>
> ret = xe_sriov_pf_migration_data_init(data, 0, 0,
> XE_SRIOV_MIG_DATA_DESCRIPTOR,
> - 0, MIGRATION_DESC_SIZE);
> + 0, MIGRATION_DESC_DWORDS *
> sizeof(u32));
> if (ret) {
> xe_sriov_pf_migration_data_free(data);
> return ret;
> }
>
> + klvs = data->vaddr;
> + *klvs++ = PREP_GUC_KLV_CONST(MIGRATION_KLV_DEVICE_DEVID_KEY,
> + MIGRATION_KLV_DEVICE_DEVID_LEN);
> + *klvs++ = xe->info.devid;
> + *klvs++ = PREP_GUC_KLV_CONST(MIGRATION_KLV_DEVICE_REVID_KEY,
> + MIGRATION_KLV_DEVICE_REVID_LEN);
> + *klvs++ = xe->info.revid;
> +
> *desc = data;
>
> return 0;
> }
>
> +/**
> + * xe_sriov_pf_migration_data_process_desc() - Process migration data
> descriptor.
> + * @gt: the &struct xe_device
@xe
> + * @vfid: the VF identifier
> + * @data: the &struct xe_sriov_pf_migration_data containing the descriptor
> + *
> + * The descriptor uses the same KLV format as GuC, and contains metadata
> used for
> + * checking migration data compatibility.
> + *
> + * Return: 0 on success, -errno on failure
> + */
> +int xe_sriov_pf_migration_data_process_desc(struct xe_device *xe, unsigned
> int vfid,
> + struct xe_sriov_pf_migration_data
> *data)
> +{
> + u32 num_dwords = data->size / sizeof(u32);
> + u32 *klvs = data->vaddr;
> +
> + xe_assert(xe, data->type == XE_SRIOV_MIG_DATA_DESCRIPTOR);
> + if (data->size % sizeof(u32) != 0)
> + return -EINVAL;
> + if (data->size != num_dwords * sizeof(u32))
> + return -EINVAL;
isn't that redundant ?
> +
> + while (num_dwords >= GUC_KLV_LEN_MIN) {
> + u32 key = FIELD_GET(GUC_KLV_0_KEY, klvs[0]);
> + u32 len = FIELD_GET(GUC_KLV_0_LEN, klvs[0]);
> +
> + klvs += GUC_KLV_LEN_MIN;
> + num_dwords -= GUC_KLV_LEN_MIN;
> +
> + switch (key) {
> + case MIGRATION_KLV_DEVICE_DEVID_KEY:
> + if (*klvs != xe->info.devid) {
> + xe_sriov_info(xe,
maybe it should be more that info() ?
> + "Aborting migration, devid
> mismatch %#04x!=%#04x\n",
> + *klvs, xe->info.devid);
> + return -ENODEV;
> + }
> + break;
> + case MIGRATION_KLV_DEVICE_REVID_KEY:
> + if (*klvs != xe->info.revid) {
> + xe_sriov_info(xe,
> + "Aborting migration, revid
> mismatch %#04x!=%#04x\n",
> + *klvs, xe->info.revid);
> + return -ENODEV;
> + }
> + break;
> + default:
> + xe_sriov_dbg(xe,
> + "Unknown migration descriptor key %#06x -
> skipping\n", key);
> + break;
> + }
> +
> + if (len > num_dwords)
> + return -EINVAL;
> +
> + klvs += len;
> + num_dwords -= len;
> + }
> +
> + return 0;
> +}
> +
> static void pf_pending_init(struct xe_device *xe, unsigned int vfid)
> {
> struct xe_sriov_pf_migration_data **data = pf_pick_pending(xe, vfid);
> diff --git a/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> index 5b96c7f224002..7cfd61005c00f 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> +++ b/drivers/gpu/drm/xe/xe_sriov_pf_migration_data.h
> @@ -32,6 +32,8 @@ ssize_t xe_sriov_pf_migration_data_read(struct xe_device
> *xe, unsigned int vfid,
> char __user *buf, size_t len);
> ssize_t xe_sriov_pf_migration_data_write(struct xe_device *xe, unsigned int
> vfid,
> const char __user *buf, size_t len);
> +int xe_sriov_pf_migration_data_process_desc(struct xe_device *xe, unsigned
> int vfid,
> + struct xe_sriov_pf_migration_data
> *data);
> int xe_sriov_pf_migration_data_save_init(struct xe_device *xe, unsigned int
> vfid);
>
> #endif
