Re: [PATCH v4 3/6] drm/amdgpu: Introduce SRIOV critical regions v2 during VF init

["Lazar, Lijo" <lijo.lazar@amd.com>]

On 10/15/2025 1:43 AM, Ellen Pan wrote:
>      1. Introduced amdgpu_virt_init_critical_region during VF init.
>       - VFs use init_data_header_offset and init_data_header_size_kb
>              transmitted via PF2VF mailbox to fetch the offset of
>              critical regions' offsets/sizes in VRAM and save to
>              adev->virt.crit_region_offsets and adev->virt.crit_region_sizes_kb.
> 
> Signed-off-by: Ellen Pan <yunru.pan@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c  |   4 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c    | 106 ++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h    |   9 ++
>   drivers/gpu/drm/amd/amdgpu/amdgv_sriovmsg.h |  31 ++++++
>   4 files changed, 150 insertions(+)
> 
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 929936c8d87c..351cfe03a1aa 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -2754,6 +2754,10 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
>   		r = amdgpu_virt_request_full_gpu(adev, true);
>   		if (r)
>   			return r;
> +
> +		r = amdgpu_virt_init_critical_region(adev);
> +		if (r)
> +			return r;
>   	}
>   
>   	switch (adev->asic_type) {
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
> index 39ab7d00379b..27235f3f3b81 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c
> @@ -44,6 +44,8 @@
>   		vf2pf_info->ucode_info[ucode].version = ver; \
>   	} while (0)
>   
> +#define mmRCC_CONFIG_MEMSIZE    0xde3
> +
>   bool amdgpu_virt_mmio_blocked(struct amdgpu_device *adev)
>   {
>   	/* By now all MMIO pages except mailbox are blocked */
> @@ -843,6 +845,110 @@ static void amdgpu_virt_init_ras(struct amdgpu_device *adev)
>   	adev->virt.ras.cper_rptr = 0;
>   }
>   
> +static uint8_t amdgpu_virt_crit_region_calc_checksum(uint8_t *buf_start, uint8_t *buf_end)
> +{
> +	uint32_t sum = 0;
> +
> +	if (buf_start >= buf_end)
> +		return 0;
> +
> +	for (; buf_start < buf_end; buf_start++)
> +		sum += buf_start[0];
> +
> +	return 0xffffffff - sum;
> +}
> +
> +int amdgpu_virt_init_critical_region(struct amdgpu_device *adev)
> +{
> +	struct amd_sriov_msg_init_data_header *init_data_hdr = NULL;
> +	uint32_t init_hdr_offset = adev->virt.init_data_header.offset;
> +	uint32_t init_hdr_size = adev->virt.init_data_header.size_kb << 10;
> +	uint64_t vram_size;
> +	int r = 0;
> +	uint8_t checksum = 0;
> +
> +	/* Skip below init if critical region version != v2 */
> +	if (adev->virt.req_init_data_ver != GPU_CRIT_REGION_V2)
> +		return 0;
> +
> +	if (init_hdr_offset < 0) {
> +		dev_err(adev->dev, "Invalid init header offset\n");
> +		return -EINVAL;
> +	}
> +
> +	vram_size = RREG32(mmRCC_CONFIG_MEMSIZE);
> +	if (!vram_size || vram_size == U32_MAX)
> +		return -EINVAL;
> +	vram_size <<= 20;
> +
> +	if ((init_hdr_offset + init_hdr_size) > vram_size) {
> +		dev_err(adev->dev, "init_data_header exceeds VRAM size, exiting\n");
> +		return -EINVAL;
> +	}
> +
> +	/* Allocate for init_data_hdr */
> +	init_data_hdr = kzalloc(sizeof(struct amd_sriov_msg_init_data_header), GFP_KERNEL);
> +	if (!init_data_hdr)
> +		return -ENOMEM;
> +
> +	amdgpu_device_vram_access(adev, (uint64_t)init_hdr_offset, (uint32_t *)init_data_hdr,
> +					sizeof(struct amd_sriov_msg_init_data_header), false);
> +
> +	if (strncmp(init_data_hdr->signature,
> +				AMDGPU_SRIOV_CRIT_DATA_SIGNATURE,
> +				AMDGPU_SRIOV_CRIT_DATA_SIG_LEN) != 0) {
> +		dev_err(adev->dev, "Invalid init data signature: %.4s\n",
> +			init_data_hdr->signature);
> +		r = -EINVAL;
> +		goto out;
> +	}
> +
> +	checksum = amdgpu_virt_crit_region_calc_checksum(
> +			(uint8_t *)&init_data_hdr->initdata_offset,
> +			(uint8_t *)init_data_hdr +
> +			sizeof(struct amd_sriov_msg_init_data_header));
> +	if (checksum != init_data_hdr->checksum) {
> +		dev_err(adev->dev, "Found unmatching checksum from calculation 0x%x and init_data 0x%x\n",
> +				checksum, init_data_hdr->checksum);
> +		r = -EINVAL;
> +		goto out;
> +	}
> +
> +	/* Initialize critical region offsets */
> +	adev->virt.crit_regn.offset = init_data_hdr->initdata_offset;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_IPD_TABLE_ID].offset =
> +		init_data_hdr->ip_discovery_offset;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_VBIOS_IMG_TABLE_ID].offset =
> +		init_data_hdr->vbios_img_offset;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_RAS_TELEMETRY_TABLE_ID].offset =
> +		init_data_hdr->ras_tele_info_offset;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_DATAEXCHANGE_TABLE_ID].offset =
> +		init_data_hdr->dataexchange_offset;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_BAD_PAGE_INFO_TABLE_ID].offset =
> +		init_data_hdr->bad_page_info_offset;
> +
> +	/* Initialize critical region sizes */
> +	adev->virt.crit_regn.size_kb = init_data_hdr->initdata_size_in_kb;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_IPD_TABLE_ID].size_kb =
> +		init_data_hdr->ip_discovery_size_in_kb;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_VBIOS_IMG_TABLE_ID].size_kb =
> +		init_data_hdr->vbios_img_size_in_kb;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_RAS_TELEMETRY_TABLE_ID].size_kb =
> +		init_data_hdr->ras_tele_info_size_in_kb;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_DATAEXCHANGE_TABLE_ID].size_kb =
> +		init_data_hdr->dataexchange_size_in_kb;
> +	adev->virt.crit_regn_tbl[AMD_SRIOV_MSG_BAD_PAGE_INFO_TABLE_ID].size_kb =
> +		init_data_hdr->bad_page_size_in_kb;
> +
> +	adev->virt.is_dynamic_crit_regn_enabled = true;
> +
> +out:
> +	kfree(init_data_hdr);
> +	init_data_hdr = NULL;
> +
> +	return r;
> +}
> +
>   void amdgpu_virt_init(struct amdgpu_device *adev)
>   {
>   	bool is_sriov = false;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h
> index 36247a160aa6..5c1dce9731e1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.h
> @@ -54,6 +54,10 @@
>   
>   #define AMDGPU_VF2PF_UPDATE_MAX_RETRY_LIMIT 2
>   
> +/* Signature used to validate the SR-IOV dynamic critical region init data header ("INDA") */
> +#define AMDGPU_SRIOV_CRIT_DATA_SIGNATURE "INDA"
> +#define AMDGPU_SRIOV_CRIT_DATA_SIG_LEN   4
> +
>   enum amdgpu_sriov_vf_mode {
>   	SRIOV_VF_MODE_BARE_METAL = 0,
>   	SRIOV_VF_MODE_ONE_VF,
> @@ -296,6 +300,9 @@ struct amdgpu_virt {
>   
>   	/* dynamic(v2) critical regions */
>   	struct amdgpu_virt_region init_data_header;
> +	struct amdgpu_virt_region crit_regn;
> +	struct amdgpu_virt_region crit_regn_tbl[AMD_SRIOV_MSG_MAX_TABLE_ID];
> +	bool is_dynamic_crit_regn_enabled;
>   
>   	/* vf2pf message */
>   	struct delayed_work vf2pf_work;
> @@ -432,6 +439,8 @@ void amdgpu_virt_exchange_data(struct amdgpu_device *adev);
>   void amdgpu_virt_fini_data_exchange(struct amdgpu_device *adev);
>   void amdgpu_virt_init(struct amdgpu_device *adev);
>   
> +int amdgpu_virt_init_critical_region(struct amdgpu_device *adev);
> +
>   bool amdgpu_virt_can_access_debugfs(struct amdgpu_device *adev);
>   int amdgpu_virt_enable_access_debugfs(struct amdgpu_device *adev);
>   void amdgpu_virt_disable_access_debugfs(struct amdgpu_device *adev);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgv_sriovmsg.h b/drivers/gpu/drm/amd/amdgpu/amdgv_sriovmsg.h
> index b53caab5b706..d15c256f9abd 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgv_sriovmsg.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgv_sriovmsg.h
> @@ -70,6 +70,37 @@ enum amd_sriov_crit_region_version {
>   	GPU_CRIT_REGION_V2 = 2,
>   };
>   
> +/* v2 layout offset enum (in order of allocation) */
> +enum amd_sriov_msg_table_id_enum {
> +	AMD_SRIOV_MSG_IPD_TABLE_ID = 0,
> +	AMD_SRIOV_MSG_VBIOS_IMG_TABLE_ID,
> +	AMD_SRIOV_MSG_RAS_TELEMETRY_TABLE_ID,
> +	AMD_SRIOV_MSG_DATAEXCHANGE_TABLE_ID,
> +	AMD_SRIOV_MSG_BAD_PAGE_INFO_TABLE_ID,
> +	AMD_SRIOV_MSG_INITD_H_TABLE_ID,
> +	AMD_SRIOV_MSG_MAX_TABLE_ID,
> +};
> +
> +struct amd_sriov_msg_init_data_header {
> +	char     signature[4];  /* "INDA"  */
> +	uint32_t version;
> +	uint32_t checksum;
> +	uint32_t initdata_offset; /* 0 */
> +	uint32_t initdata_size_in_kb; /* 5MB */
> +	uint32_t valid_tables;

Missed this earlier. Is this a mask of table_ids that are valid? Not 
seeing it used before assigning the offsets.

Thanks,
Lijo

> +	uint32_t vbios_img_offset;
> +	uint32_t vbios_img_size_in_kb;
> +	uint32_t dataexchange_offset;
> +	uint32_t dataexchange_size_in_kb;
> +	uint32_t ras_tele_info_offset;
> +	uint32_t ras_tele_info_size_in_kb;
> +	uint32_t ip_discovery_offset;
> +	uint32_t ip_discovery_size_in_kb;
> +	uint32_t bad_page_info_offset;
> +	uint32_t bad_page_size_in_kb;
> +	uint32_t reserved[8];
> +};
> +
>   /*
>    * PF2VF history log:
>    * v1 defined in amdgim