[Intel-gfx] [PATCH 41/51] drm/i915/guc: Add golden context to GuC ADS
Matthew Brost
matthew.brost at intel.com
Mon Jul 19 17:24:30 UTC 2021
On Fri, Jul 16, 2021 at 01:17:14PM -0700, Matthew Brost wrote:
> From: John Harrison <John.C.Harrison at Intel.com>
>
> The media watchdog mechanism involves GuC doing a silent reset and
> continue of the hung context. This requires the i915 driver provide a
> golden context to GuC in the ADS.
>
> Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
> Signed-off-by: Matthew Brost <matthew.brost at intel.com>
> ---
> drivers/gpu/drm/i915/gt/intel_gt.c | 2 +
> drivers/gpu/drm/i915/gt/uc/intel_guc.c | 5 +
> drivers/gpu/drm/i915/gt/uc/intel_guc.h | 2 +
> drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c | 213 ++++++++++++++++++---
> drivers/gpu/drm/i915/gt/uc/intel_guc_ads.h | 1 +
> drivers/gpu/drm/i915/gt/uc/intel_uc.c | 5 +
> drivers/gpu/drm/i915/gt/uc/intel_uc.h | 1 +
> 7 files changed, 199 insertions(+), 30 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_gt.c b/drivers/gpu/drm/i915/gt/intel_gt.c
> index acfdd53b2678..ceeb517ba259 100644
> --- a/drivers/gpu/drm/i915/gt/intel_gt.c
> +++ b/drivers/gpu/drm/i915/gt/intel_gt.c
> @@ -654,6 +654,8 @@ int intel_gt_init(struct intel_gt *gt)
> if (err)
> goto err_gt;
>
> + intel_uc_init_late(>->uc);
> +
> err = i915_inject_probe_error(gt->i915, -EIO);
> if (err)
> goto err_gt;
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.c b/drivers/gpu/drm/i915/gt/uc/intel_guc.c
> index 68266cbffd1f..979128e28372 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc.c
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.c
> @@ -180,6 +180,11 @@ void intel_guc_init_early(struct intel_guc *guc)
> }
> }
>
> +void intel_guc_init_late(struct intel_guc *guc)
> +{
> + intel_guc_ads_init_late(guc);
> +}
> +
> static u32 guc_ctl_debug_flags(struct intel_guc *guc)
> {
> u32 level = intel_guc_log_get_level(&guc->log);
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc.h b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
> index bc71635c70b9..dc18ac510ac8 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc.h
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc.h
> @@ -60,6 +60,7 @@ struct intel_guc {
> struct i915_vma *ads_vma;
> struct __guc_ads_blob *ads_blob;
> u32 ads_regset_size;
> + u32 ads_golden_ctxt_size;
>
> struct i915_vma *lrc_desc_pool;
> void *lrc_desc_pool_vaddr;
> @@ -176,6 +177,7 @@ static inline u32 intel_guc_ggtt_offset(struct intel_guc *guc,
> }
>
> void intel_guc_init_early(struct intel_guc *guc);
> +void intel_guc_init_late(struct intel_guc *guc);
> void intel_guc_init_send_regs(struct intel_guc *guc);
> void intel_guc_write_params(struct intel_guc *guc);
> int intel_guc_init(struct intel_guc *guc);
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c
> index 93b0ac35a508..241b3089b658 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.c
> @@ -7,6 +7,7 @@
>
> #include "gt/intel_gt.h"
> #include "gt/intel_lrc.h"
> +#include "gt/shmem_utils.h"
> #include "intel_guc_ads.h"
> #include "intel_guc_fwif.h"
> #include "intel_uc.h"
> @@ -33,6 +34,10 @@
> * +---------------------------------------+ <== dynamic
> * | padding |
> * +---------------------------------------+ <== 4K aligned
> + * | golden contexts |
> + * +---------------------------------------+
> + * | padding |
> + * +---------------------------------------+ <== 4K aligned
> * | private data |
> * +---------------------------------------+
> * | padding |
> @@ -52,6 +57,11 @@ static u32 guc_ads_regset_size(struct intel_guc *guc)
> return guc->ads_regset_size;
> }
>
> +static u32 guc_ads_golden_ctxt_size(struct intel_guc *guc)
> +{
> + return PAGE_ALIGN(guc->ads_golden_ctxt_size);
> +}
> +
> static u32 guc_ads_private_data_size(struct intel_guc *guc)
> {
> return PAGE_ALIGN(guc->fw.private_data_size);
> @@ -62,12 +72,23 @@ static u32 guc_ads_regset_offset(struct intel_guc *guc)
> return offsetof(struct __guc_ads_blob, regset);
> }
>
> -static u32 guc_ads_private_data_offset(struct intel_guc *guc)
> +static u32 guc_ads_golden_ctxt_offset(struct intel_guc *guc)
> {
> u32 offset;
>
> offset = guc_ads_regset_offset(guc) +
> guc_ads_regset_size(guc);
> +
> + return PAGE_ALIGN(offset);
> +}
> +
> +static u32 guc_ads_private_data_offset(struct intel_guc *guc)
> +{
> + u32 offset;
> +
> + offset = guc_ads_golden_ctxt_offset(guc) +
> + guc_ads_golden_ctxt_size(guc);
> +
> return PAGE_ALIGN(offset);
> }
>
> @@ -319,53 +340,163 @@ static void guc_mmio_reg_state_init(struct intel_guc *guc,
> GEM_BUG_ON(temp_set.size);
> }
>
> -/*
> - * The first 80 dwords of the register state context, containing the
> - * execlists and ppgtt registers.
> - */
> -#define LR_HW_CONTEXT_SIZE (80 * sizeof(u32))
> +static void fill_engine_enable_masks(struct intel_gt *gt,
> + struct guc_gt_system_info *info)
> +{
> + info->engine_enabled_masks[GUC_RENDER_CLASS] = 1;
> + info->engine_enabled_masks[GUC_BLITTER_CLASS] = 1;
> + info->engine_enabled_masks[GUC_VIDEO_CLASS] = VDBOX_MASK(gt);
> + info->engine_enabled_masks[GUC_VIDEOENHANCE_CLASS] = VEBOX_MASK(gt);
> +}
>
> -static void __guc_ads_init(struct intel_guc *guc)
> +/* Skip execlist and PPGTT registers */
> +#define LR_HW_CONTEXT_SIZE (80 * sizeof(u32))
> +#define SKIP_SIZE (LRC_PPHWSP_SZ * PAGE_SIZE + LR_HW_CONTEXT_SIZE)
> +
> +static int guc_prep_golden_context(struct intel_guc *guc,
> + struct __guc_ads_blob *blob)
> {
> struct intel_gt *gt = guc_to_gt(guc);
> - struct drm_i915_private *i915 = gt->i915;
> + u32 addr_ggtt, offset;
> + u32 total_size = 0, alloc_size, real_size;
> + u8 engine_class, guc_class;
> + struct guc_gt_system_info *info, local_info;
> +
> + /*
> + * Reserve the memory for the golden contexts and point GuC at it but
> + * leave it empty for now. The context data will be filled in later
> + * once there is something available to put there.
> + *
> + * Note that the HWSP and ring context are not included.
> + *
> + * Note also that the storage must be pinned in the GGTT, so that the
> + * address won't change after GuC has been told where to find it. The
> + * GuC will also validate that the LRC base + size fall within the
> + * allowed GGTT range.
> + */
> + if (blob) {
> + offset = guc_ads_golden_ctxt_offset(guc);
> + addr_ggtt = intel_guc_ggtt_offset(guc, guc->ads_vma) + offset;
> + info = &blob->system_info;
> + } else {
> + memset(&local_info, 0, sizeof(local_info));
> + info = &local_info;
> + fill_engine_enable_masks(gt, info);
> + }
> +
> + for (engine_class = 0; engine_class <= MAX_ENGINE_CLASS; ++engine_class) {
> + if (engine_class == OTHER_CLASS)
> + continue;
> +
> + guc_class = engine_class_to_guc_class(engine_class);
> +
> + if (!info->engine_enabled_masks[guc_class])
> + continue;
> +
> + real_size = intel_engine_context_size(gt, engine_class);
> + alloc_size = PAGE_ALIGN(real_size);
> + total_size += alloc_size;
> +
> + if (!blob)
> + continue;
> +
> + blob->ads.eng_state_size[guc_class] = real_size;
> + blob->ads.golden_context_lrca[guc_class] = addr_ggtt;
> + addr_ggtt += alloc_size;
> + }
> +
> + if (!blob)
> + return total_size;
> +
> + GEM_BUG_ON(guc->ads_golden_ctxt_size != total_size);
> + return total_size;
> +}
> +
> +static struct intel_engine_cs *find_engine_state(struct intel_gt *gt, u8 engine_class)
> +{
> + struct intel_engine_cs *engine;
> + enum intel_engine_id id;
> +
> + for_each_engine(engine, gt, id) {
> + if (engine->class != engine_class)
> + continue;
> +
> + if (!engine->default_state)
> + continue;
> +
> + return engine;
> + }
> +
> + return NULL;
> +}
> +
> +static void guc_init_golden_context(struct intel_guc *guc)
> +{
> struct __guc_ads_blob *blob = guc->ads_blob;
> - const u32 skipped_size = LRC_PPHWSP_SZ * PAGE_SIZE + LR_HW_CONTEXT_SIZE;
> - u32 base;
> + struct intel_engine_cs *engine;
> + struct intel_gt *gt = guc_to_gt(guc);
> + u32 addr_ggtt, offset;
> + u32 total_size = 0, alloc_size, real_size;
> u8 engine_class, guc_class;
> + u8 *ptr;
>
> - /* GuC scheduling policies */
> - guc_policies_init(guc, &blob->policies);
> + if (!intel_uc_uses_guc_submission(>->uc))
> + return;
> +
> + GEM_BUG_ON(!blob);
>
> /*
> - * GuC expects a per-engine-class context image and size
> - * (minus hwsp and ring context). The context image will be
> - * used to reinitialize engines after a reset. It must exist
> - * and be pinned in the GGTT, so that the address won't change after
> - * we have told GuC where to find it. The context size will be used
> - * to validate that the LRC base + size fall within allowed GGTT.
> + * Go back and fill in the golden context data now that it is
> + * available.
> */
> + offset = guc_ads_golden_ctxt_offset(guc);
> + addr_ggtt = intel_guc_ggtt_offset(guc, guc->ads_vma) + offset;
> + ptr = ((u8 *) blob) + offset;
> +
> for (engine_class = 0; engine_class <= MAX_ENGINE_CLASS; ++engine_class) {
> if (engine_class == OTHER_CLASS)
> continue;
>
> guc_class = engine_class_to_guc_class(engine_class);
>
> - /*
> - * TODO: Set context pointer to default state to allow
> - * GuC to re-init guilty contexts after internal reset.
> - */
> - blob->ads.golden_context_lrca[guc_class] = 0;
> - blob->ads.eng_state_size[guc_class] =
> - intel_engine_context_size(gt, engine_class) -
> - skipped_size;
> + if (!blob->system_info.engine_enabled_masks[guc_class])
> + continue;
> +
> + real_size = intel_engine_context_size(gt, engine_class);
> + alloc_size = PAGE_ALIGN(real_size);
> + total_size += alloc_size;
> +
> + engine = find_engine_state(gt, engine_class);
> + if (!engine) {
> + drm_err(>->i915->drm, "No engine state recorded for class %d!\n", engine_class);
> + blob->ads.eng_state_size[guc_class] = 0;
> + blob->ads.golden_context_lrca[guc_class] = 0;
> + continue;
> + }
> +
> + GEM_BUG_ON(blob->ads.eng_state_size[guc_class] != real_size);
> + GEM_BUG_ON(blob->ads.golden_context_lrca[guc_class] != addr_ggtt);
> + addr_ggtt += alloc_size;
> +
> + shmem_read(engine->default_state, SKIP_SIZE, ptr + SKIP_SIZE, real_size);
There is a confirmed memory corruption here, should be:
shmem_read(engine->default_state, SKIP_SIZE, ptr + SKIP_SIZE, real_size - SKIP_SIZE);
Also a bit of nit but I think s/SKIP_SIZE/GOLDEN_CONTEXT_SKIP_SIZE/g
As long you agree John, I'll fix both of these in the next rev.
Matt
> + ptr += alloc_size;
> }
>
> + GEM_BUG_ON(guc->ads_golden_ctxt_size != total_size);
> +}
> +
> +static void __guc_ads_init(struct intel_guc *guc)
> +{
> + struct intel_gt *gt = guc_to_gt(guc);
> + struct drm_i915_private *i915 = gt->i915;
> + struct __guc_ads_blob *blob = guc->ads_blob;
> + u32 base;
> +
> + /* GuC scheduling policies */
> + guc_policies_init(guc, &blob->policies);
> +
> /* System info */
> - blob->system_info.engine_enabled_masks[GUC_RENDER_CLASS] = 1;
> - blob->system_info.engine_enabled_masks[GUC_BLITTER_CLASS] = 1;
> - blob->system_info.engine_enabled_masks[GUC_VIDEO_CLASS] = VDBOX_MASK(gt);
> - blob->system_info.engine_enabled_masks[GUC_VIDEOENHANCE_CLASS] = VEBOX_MASK(gt);
> + fill_engine_enable_masks(gt, &blob->system_info);
>
> blob->system_info.generic_gt_sysinfo[GUC_GENERIC_GT_SYSINFO_SLICE_ENABLED] =
> hweight8(gt->info.sseu.slice_mask);
> @@ -380,6 +511,9 @@ static void __guc_ads_init(struct intel_guc *guc)
> GEN12_DOORBELLS_PER_SQIDI) + 1;
> }
>
> + /* Golden contexts for re-initialising after a watchdog reset */
> + guc_prep_golden_context(guc, blob);
> +
> guc_mapping_table_init(guc_to_gt(guc), &blob->system_info);
>
> base = intel_guc_ggtt_offset(guc, guc->ads_vma);
> @@ -417,6 +551,13 @@ int intel_guc_ads_create(struct intel_guc *guc)
> return ret;
> guc->ads_regset_size = ret;
>
> + /* Likewise the golden contexts: */
> + ret = guc_prep_golden_context(guc, NULL);
> + if (ret < 0)
> + return ret;
> + guc->ads_golden_ctxt_size = ret;
> +
> + /* Now the total size can be determined: */
> size = guc_ads_blob_size(guc);
>
> ret = intel_guc_allocate_and_map_vma(guc, size, &guc->ads_vma,
> @@ -429,6 +570,18 @@ int intel_guc_ads_create(struct intel_guc *guc)
> return 0;
> }
>
> +void intel_guc_ads_init_late(struct intel_guc *guc)
> +{
> + /*
> + * The golden context setup requires the saved engine state from
> + * __engines_record_defaults(). However, that requires engines to be
> + * operational which means the ADS must already have been configured.
> + * Fortunately, the golden context state is not needed until a hang
> + * occurs, so it can be filled in during this late init phase.
> + */
> + guc_init_golden_context(guc);
> +}
> +
> void intel_guc_ads_destroy(struct intel_guc *guc)
> {
> i915_vma_unpin_and_release(&guc->ads_vma, I915_VMA_RELEASE_MAP);
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.h b/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.h
> index bdcb339a5321..3d85051d57e4 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.h
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ads.h
> @@ -11,6 +11,7 @@ struct drm_printer;
>
> int intel_guc_ads_create(struct intel_guc *guc);
> void intel_guc_ads_destroy(struct intel_guc *guc);
> +void intel_guc_ads_init_late(struct intel_guc *guc);
> void intel_guc_ads_reset(struct intel_guc *guc);
> void intel_guc_ads_print_policy_info(struct intel_guc *guc,
> struct drm_printer *p);
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_uc.c b/drivers/gpu/drm/i915/gt/uc/intel_uc.c
> index 77c1fe2ed883..7a69c3c027e9 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_uc.c
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_uc.c
> @@ -120,6 +120,11 @@ void intel_uc_init_early(struct intel_uc *uc)
> uc->ops = &uc_ops_off;
> }
>
> +void intel_uc_init_late(struct intel_uc *uc)
> +{
> + intel_guc_init_late(&uc->guc);
> +}
> +
> void intel_uc_driver_late_release(struct intel_uc *uc)
> {
> }
> diff --git a/drivers/gpu/drm/i915/gt/uc/intel_uc.h b/drivers/gpu/drm/i915/gt/uc/intel_uc.h
> index 91315e3f1c58..e2da2b6e76e1 100644
> --- a/drivers/gpu/drm/i915/gt/uc/intel_uc.h
> +++ b/drivers/gpu/drm/i915/gt/uc/intel_uc.h
> @@ -35,6 +35,7 @@ struct intel_uc {
> };
>
> void intel_uc_init_early(struct intel_uc *uc);
> +void intel_uc_init_late(struct intel_uc *uc);
> void intel_uc_driver_late_release(struct intel_uc *uc);
> void intel_uc_driver_remove(struct intel_uc *uc);
> void intel_uc_init_mmio(struct intel_uc *uc);
> --
> 2.28.0
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx
More information about the dri-devel
mailing list