Re: [PATCH 1/2] drm/xe/guc: Add support for w/a KLVs

[John Harrison <john.c.harrison@intel.com>]

On 3/15/2024 04:21, Badal Nilawar wrote:
> To prevent running out of bits, new w/a enable flags are being added
> via a KLV system instead of a 32 bit flags word.
>
> Cc: John Harrison <John.C.Harrison@intel.com>
> Signed-off-by: Badal Nilawar <badal.nilawar@intel.com>
> ---
>   drivers/gpu/drm/xe/xe_guc_ads.c       | 68 ++++++++++++++++++++++++++-
>   drivers/gpu/drm/xe/xe_guc_ads_types.h |  2 +
>   drivers/gpu/drm/xe/xe_guc_fwif.h      |  5 +-
>   3 files changed, 72 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c
> index 6ad4c1a90a78..ee48cf01fe22 100644
> --- a/drivers/gpu/drm/xe/xe_guc_ads.c
> +++ b/drivers/gpu/drm/xe/xe_guc_ads.c
> @@ -80,6 +80,10 @@ ads_to_map(struct xe_guc_ads *ads)
>    *      +---------------------------------------+
>    *      | padding                               |
>    *      +---------------------------------------+ <== 4K aligned
> + *      | w/a KLVs                              |
> + *      +---------------------------------------+
> + *      | padding                               |
> + *      +---------------------------------------+ <== 4K aligned
>    *      | capture lists                         |
>    *      +---------------------------------------+
>    *      | padding                               |
> @@ -131,6 +135,11 @@ static size_t guc_ads_golden_lrc_size(struct xe_guc_ads *ads)
>   	return PAGE_ALIGN(ads->golden_lrc_size);
>   }
>   
> +static u32 guc_ads_waklv_size(struct xe_guc_ads *ads)
> +{
> +	return PAGE_ALIGN(ads->ads_waklv_size);
> +}
> +
>   static size_t guc_ads_capture_size(struct xe_guc_ads *ads)
>   {
>   	/* FIXME: Allocate a proper capture list */
> @@ -167,12 +176,22 @@ static size_t guc_ads_golden_lrc_offset(struct xe_guc_ads *ads)
>   	return PAGE_ALIGN(offset);
>   }
>   
> +static size_t guc_ads_waklv_offset(struct xe_guc_ads *ads)
> +{
> +	u32 offset;
> +
> +	offset = guc_ads_golden_lrc_offset(ads) +
> +		 guc_ads_golden_lrc_size(ads);
> +
> +	return PAGE_ALIGN(offset);
> +}
> +
>   static size_t guc_ads_capture_offset(struct xe_guc_ads *ads)
>   {
>   	size_t offset;
>   
> -	offset = guc_ads_golden_lrc_offset(ads) +
> -		guc_ads_golden_lrc_size(ads);
> +	offset = guc_ads_waklv_offset(ads) +
> +		 guc_ads_waklv_size(ads);
>   
>   	return PAGE_ALIGN(offset);
>   }
> @@ -260,6 +279,48 @@ static size_t calculate_golden_lrc_size(struct xe_guc_ads *ads)
>   	return total_size;
>   }
>   
> +#define GUC_VER(maj, min, pat)	(((maj) << 16) | ((min) << 8) | (pat))
This should be in a common header file, not duplicated in every source 
file that wants to use it. But see below...

> +
> +static void guc_waklv_init(struct xe_guc_ads *ads)
> +{
> +	u32 addr_ggtt, offset, remain, size;
> +	struct xe_uc_fw *uc_fw = &ads_to_guc(ads)->fw;
> +	struct xe_uc_fw_version *version = &uc_fw->versions.found[XE_UC_FW_VER_RELEASE];
> +
> +	if (GUC_VER(version->major, version->minor, version->patch) >= GUC_VER(70, 10, 0))
Note that Xe will not be supporting anything below 70.19 once this patch 
lands:
https://patchwork.freedesktop.org/series/130677/

Which means that any test for <70.19 will need to be removed again.

John.

> +		return;
> +
> +	offset = guc_ads_waklv_offset(ads);
> +	remain = guc_ads_waklv_size(ads);
> +
> +	/*
> +	 * Add workarounds here:
> +	 *
> +	 * if (want_wa_<name>) {
> +	 *      size = guc_waklv_<name>(guc, offset, remain);
> +	 *      offset += size;
> +	 *      remain -= size;
> +	 * }
> +	 */
> +
> +	size = guc_ads_waklv_size(ads) - remain;
> +	if (!size)
> +		return;
> +
> +	offset = guc_ads_waklv_offset(ads);
> +	addr_ggtt = xe_bo_ggtt_addr(ads->bo) + offset;
> +
> +	ads_blob_write(ads, ads.wa_klv_addr_lo, addr_ggtt);
> +	ads_blob_write(ads, ads.wa_klv_addr_hi, 0);
> +	ads_blob_write(ads, ads.wa_klv_size, size);
> +}
> +
> +static int calculate_waklv_size(struct xe_guc_ads *ads)
> +{
> +	/* Fudge something chunky for now: */
> +	return PAGE_SIZE;
> +}
> +
>   #define MAX_GOLDEN_LRC_SIZE	(SZ_4K * 64)
>   
>   int xe_guc_ads_init(struct xe_guc_ads *ads)
> @@ -271,6 +332,7 @@ int xe_guc_ads_init(struct xe_guc_ads *ads)
>   
>   	ads->golden_lrc_size = calculate_golden_lrc_size(ads);
>   	ads->regset_size = calculate_regset_size(gt);
> +	ads->ads_waklv_size = calculate_waklv_size(ads);
>   
>   	bo = xe_managed_bo_create_pin_map(xe, tile, guc_ads_size(ads) + MAX_GOLDEN_LRC_SIZE,
>   					  XE_BO_CREATE_SYSTEM_BIT |
> @@ -597,6 +659,8 @@ void xe_guc_ads_populate(struct xe_guc_ads *ads)
>   	guc_mapping_table_init(gt, &info_map);
>   	guc_capture_list_init(ads);
>   	guc_doorbell_init(ads);
> +	/* Workaround KLV list */
> +	guc_waklv_init(ads);
>   
>   	if (xe->info.has_usm) {
>   		guc_um_init_params(ads);
> diff --git a/drivers/gpu/drm/xe/xe_guc_ads_types.h b/drivers/gpu/drm/xe/xe_guc_ads_types.h
> index 4afe44bece4b..62235b2a6fe3 100644
> --- a/drivers/gpu/drm/xe/xe_guc_ads_types.h
> +++ b/drivers/gpu/drm/xe/xe_guc_ads_types.h
> @@ -20,6 +20,8 @@ struct xe_guc_ads {
>   	size_t golden_lrc_size;
>   	/** @regset_size: size of register set passed to GuC for save/restore */
>   	u32 regset_size;
> +	/** @ads_waklv_size: waklv size */
> +	u32 ads_waklv_size;
>   };
>   
>   #endif
> diff --git a/drivers/gpu/drm/xe/xe_guc_fwif.h b/drivers/gpu/drm/xe/xe_guc_fwif.h
> index c281fdbfd2d6..52503719d2aa 100644
> --- a/drivers/gpu/drm/xe/xe_guc_fwif.h
> +++ b/drivers/gpu/drm/xe/xe_guc_fwif.h
> @@ -207,7 +207,10 @@ struct guc_ads {
>   	u32 capture_instance[GUC_CAPTURE_LIST_INDEX_MAX][GUC_MAX_ENGINE_CLASSES];
>   	u32 capture_class[GUC_CAPTURE_LIST_INDEX_MAX][GUC_MAX_ENGINE_CLASSES];
>   	u32 capture_global[GUC_CAPTURE_LIST_INDEX_MAX];
> -	u32 reserved[14];
> +	u32 wa_klv_addr_lo;
> +	u32 wa_klv_addr_hi;
> +	u32 wa_klv_size;
> +	u32 reserved[11];
>   } __packed;
>   
>   /* Engine usage stats */