From d1675198ed1f21aec6e036336e4340c40b726497 Mon Sep 17 00:00:00 2001 From: Alex Dai Date: Wed, 12 Aug 2015 15:43:43 +0100 Subject: [PATCH] drm/i915: Integrate GuC-based command submission GuC-based submission is mostly the same as execlist mode, up to intel_logical_ring_advance_and_submit(), where the context being dispatched would be added to the execlist queue; at this point we submit the context to the GuC backend instead. There are, however, a few other changes also required, notably: 1. Contexts must be pinned at GGTT addresses accessible by the GuC i.e. NOT in the range [0..WOPCM_SIZE), so we have to add the PIN_OFFSET_BIAS flag to the relevant GGTT-pinning calls. 2. The GuC's TLB must be invalidated after a context is pinned at a new GGTT address. 3. GuC firmware uses the one page before Ring Context as shared data. Therefore, whenever driver wants to get base address of LRC, we will offset one page for it. LRC_PPHWSP_PN is defined as the page number of LRCA. 4. In the work queue used to pass requests to the GuC, the GuC firmware requires the ring-tail-offset to be represented as an 11-bit value, expressed in QWords. Therefore, the ringbuffer size must be reduced to the representable range (4 pages). v2: Defer adding #defines until needed [Chris Wilson] Rationalise type declarations [Chris Wilson] v4: Squashed kerneldoc patch into here [Daniel Vetter] v5: Update request->tail in code common to both GuC and execlist modes. Add a private version of lr_context_update(), as sharing the execlist version leads to race conditions when the CPU and the GuC both update TAIL in the context image. Conversion of error-captured HWS page to string must account for offset from start of object to actual HWS (LRC_PPHWSP_PN). Issue: VIZ-4884 Signed-off-by: Alex Dai Signed-off-by: Dave Gordon Reviewed-by: Tom O'Rourke Signed-off-by: Daniel Vetter --- Documentation/DocBook/drm.tmpl | 14 ++++ drivers/gpu/drm/i915/i915_debugfs.c | 2 +- drivers/gpu/drm/i915/i915_gpu_error.c | 14 ++-- drivers/gpu/drm/i915/i915_guc_submission.c | 79 ++++++++++++++++++++-- drivers/gpu/drm/i915/intel_guc.h | 1 + drivers/gpu/drm/i915/intel_lrc.c | 55 ++++++++++----- drivers/gpu/drm/i915/intel_lrc.h | 6 ++ 7 files changed, 142 insertions(+), 29 deletions(-) diff --git a/Documentation/DocBook/drm.tmpl b/Documentation/DocBook/drm.tmpl index 9ddf8c6cb887..069958c00f1c 100644 --- a/Documentation/DocBook/drm.tmpl +++ b/Documentation/DocBook/drm.tmpl @@ -4237,6 +4237,20 @@ int num_ioctls; !Idrivers/gpu/drm/i915/i915_gem_shrinker.c + + GuC-based Command Submission + + GuC +!Pdrivers/gpu/drm/i915/intel_guc_loader.c GuC-specific firmware loader +!Idrivers/gpu/drm/i915/intel_guc_loader.c + + + GuC Client +!Pdrivers/gpu/drm/i915/intel_guc_submission.c GuC-based command submissison +!Idrivers/gpu/drm/i915/intel_guc_submission.c + + + Tracing diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c index 1123b8e3b10a..a815fcc407c6 100644 --- a/drivers/gpu/drm/i915/i915_debugfs.c +++ b/drivers/gpu/drm/i915/i915_debugfs.c @@ -1995,7 +1995,7 @@ static void i915_dump_lrc_obj(struct seq_file *m, return; } - page = i915_gem_object_get_page(ctx_obj, 1); + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); if (!WARN_ON(page == NULL)) { reg_state = kmap_atomic(page); diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c b/drivers/gpu/drm/i915/i915_gpu_error.c index 0d0a7b1a6b4b..493e9b294a69 100644 --- a/drivers/gpu/drm/i915/i915_gpu_error.c +++ b/drivers/gpu/drm/i915/i915_gpu_error.c @@ -462,17 +462,17 @@ int i915_error_state_to_str(struct drm_i915_error_state_buf *m, } if ((obj = error->ring[i].hws_page)) { - err_printf(m, "%s --- HW Status = 0x%08x\n", - dev_priv->ring[i].name, - lower_32_bits(obj->gtt_offset)); + err_printf(m, "%s --- HW Status = 0x%08llx\n", + dev_priv->ring[i].name, + obj->gtt_offset + LRC_PPHWSP_PN * PAGE_SIZE); offset = 0; for (elt = 0; elt < PAGE_SIZE/16; elt += 4) { err_printf(m, "[%04x] %08x %08x %08x %08x\n", offset, - obj->pages[0][elt], - obj->pages[0][elt+1], - obj->pages[0][elt+2], - obj->pages[0][elt+3]); + obj->pages[LRC_PPHWSP_PN][elt], + obj->pages[LRC_PPHWSP_PN][elt+1], + obj->pages[LRC_PPHWSP_PN][elt+2], + obj->pages[LRC_PPHWSP_PN][elt+3]); offset += 16; } } diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c index 3352b8521ddd..ec703937f363 100644 --- a/drivers/gpu/drm/i915/i915_guc_submission.c +++ b/drivers/gpu/drm/i915/i915_guc_submission.c @@ -364,18 +364,58 @@ static void guc_init_proc_desc(struct intel_guc *guc, static void guc_init_ctx_desc(struct intel_guc *guc, struct i915_guc_client *client) { + struct intel_context *ctx = client->owner; struct guc_context_desc desc; struct sg_table *sg; + int i; memset(&desc, 0, sizeof(desc)); desc.attribute = GUC_CTX_DESC_ATTR_ACTIVE | GUC_CTX_DESC_ATTR_KERNEL; desc.context_id = client->ctx_index; desc.priority = client->priority; - desc.engines_used = (1 << RCS) | (1 << VCS) | (1 << BCS) | - (1 << VECS) | (1 << VCS2); /* all engines */ desc.db_id = client->doorbell_id; + for (i = 0; i < I915_NUM_RINGS; i++) { + struct guc_execlist_context *lrc = &desc.lrc[i]; + struct intel_ringbuffer *ringbuf = ctx->engine[i].ringbuf; + struct intel_engine_cs *ring; + struct drm_i915_gem_object *obj; + uint64_t ctx_desc; + + /* TODO: We have a design issue to be solved here. Only when we + * receive the first batch, we know which engine is used by the + * user. But here GuC expects the lrc and ring to be pinned. It + * is not an issue for default context, which is the only one + * for now who owns a GuC client. But for future owner of GuC + * client, need to make sure lrc is pinned prior to enter here. + */ + obj = ctx->engine[i].state; + if (!obj) + break; /* XXX: continue? */ + + ring = ringbuf->ring; + ctx_desc = intel_lr_context_descriptor(ctx, ring); + lrc->context_desc = (u32)ctx_desc; + + /* The state page is after PPHWSP */ + lrc->ring_lcra = i915_gem_obj_ggtt_offset(obj) + + LRC_STATE_PN * PAGE_SIZE; + lrc->context_id = (client->ctx_index << GUC_ELC_CTXID_OFFSET) | + (ring->id << GUC_ELC_ENGINE_OFFSET); + + obj = ringbuf->obj; + + lrc->ring_begin = i915_gem_obj_ggtt_offset(obj); + lrc->ring_end = lrc->ring_begin + obj->base.size - 1; + lrc->ring_next_free_location = lrc->ring_begin; + lrc->ring_current_tail_pointer_value = 0; + + desc.engines_used |= (1 << ring->id); + } + + WARN_ON(desc.engines_used == 0); + /* * The CPU address is only needed at certain points, so kmap_atomic on * demand instead of storing it in the ctx descriptor. @@ -501,6 +541,29 @@ static int guc_add_workqueue_item(struct i915_guc_client *gc, return 0; } +#define CTX_RING_BUFFER_START 0x08 + +/* Update the ringbuffer pointer in a saved context image */ +static void lr_context_update(struct drm_i915_gem_request *rq) +{ + enum intel_ring_id ring_id = rq->ring->id; + struct drm_i915_gem_object *ctx_obj = rq->ctx->engine[ring_id].state; + struct drm_i915_gem_object *rb_obj = rq->ringbuf->obj; + struct page *page; + uint32_t *reg_state; + + BUG_ON(!ctx_obj); + WARN_ON(!i915_gem_obj_is_pinned(ctx_obj)); + WARN_ON(!i915_gem_obj_is_pinned(rb_obj)); + + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); + reg_state = kmap_atomic(page); + + reg_state[CTX_RING_BUFFER_START+1] = i915_gem_obj_ggtt_offset(rb_obj); + + kunmap_atomic(reg_state); +} + /** * i915_guc_submit() - Submit commands through GuC * @client: the guc client where commands will go through @@ -517,6 +580,10 @@ int i915_guc_submit(struct i915_guc_client *client, unsigned long flags; int q_ret, b_ret; + /* Need this because of the deferred pin ctx and ring */ + /* Shall we move this right after ring is pinned? */ + lr_context_update(rq); + spin_lock_irqsave(&client->wq_lock, flags); q_ret = guc_add_workqueue_item(client, rq); @@ -643,11 +710,13 @@ static void guc_client_free(struct drm_device *dev, * The kernel client to replace ExecList submission is created with * NORMAL priority. Priority of a client for scheduler can be HIGH, * while a preemption context can use CRITICAL. + * @ctx the context to own the client (we use the default render context) * * Return: An i915_guc_client object if success. */ static struct i915_guc_client *guc_client_alloc(struct drm_device *dev, - uint32_t priority) + uint32_t priority, + struct intel_context *ctx) { struct i915_guc_client *client; struct drm_i915_private *dev_priv = dev->dev_private; @@ -660,6 +729,7 @@ static struct i915_guc_client *guc_client_alloc(struct drm_device *dev, client->doorbell_id = GUC_INVALID_DOORBELL_ID; client->priority = priority; + client->owner = ctx; client->guc = guc; client->ctx_index = (uint32_t)ida_simple_get(&guc->ctx_ids, 0, @@ -793,10 +863,11 @@ int i915_guc_submission_enable(struct drm_device *dev) { struct drm_i915_private *dev_priv = dev->dev_private; struct intel_guc *guc = &dev_priv->guc; + struct intel_context *ctx = dev_priv->ring[RCS].default_context; struct i915_guc_client *client; /* client for execbuf submission */ - client = guc_client_alloc(dev, GUC_CTX_PRIORITY_KMD_NORMAL); + client = guc_client_alloc(dev, GUC_CTX_PRIORITY_KMD_NORMAL, ctx); if (!client) { DRM_ERROR("Failed to create execbuf guc_client\n"); return -ENOMEM; diff --git a/drivers/gpu/drm/i915/intel_guc.h b/drivers/gpu/drm/i915/intel_guc.h index 92658ce45a61..4ec2d27a557e 100644 --- a/drivers/gpu/drm/i915/intel_guc.h +++ b/drivers/gpu/drm/i915/intel_guc.h @@ -29,6 +29,7 @@ struct i915_guc_client { struct drm_i915_gem_object *client_obj; + struct intel_context *owner; struct intel_guc *guc; uint32_t priority; uint32_t ctx_index; diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c index f3411f835a5a..e77b6b0d44d9 100644 --- a/drivers/gpu/drm/i915/intel_lrc.c +++ b/drivers/gpu/drm/i915/intel_lrc.c @@ -263,7 +263,8 @@ int intel_sanitize_enable_execlists(struct drm_device *dev, int enable_execlists */ u32 intel_execlists_ctx_id(struct drm_i915_gem_object *ctx_obj) { - u32 lrca = i915_gem_obj_ggtt_offset(ctx_obj); + u32 lrca = i915_gem_obj_ggtt_offset(ctx_obj) + + LRC_PPHWSP_PN * PAGE_SIZE; /* LRCA is required to be 4K aligned so the more significant 20 bits * are globally unique */ @@ -276,7 +277,8 @@ uint64_t intel_lr_context_descriptor(struct intel_context *ctx, struct drm_device *dev = ring->dev; struct drm_i915_gem_object *ctx_obj = ctx->engine[ring->id].state; uint64_t desc; - uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj); + uint64_t lrca = i915_gem_obj_ggtt_offset(ctx_obj) + + LRC_PPHWSP_PN * PAGE_SIZE; WARN_ON(lrca & 0xFFFFFFFF00000FFFULL); @@ -350,7 +352,7 @@ static int execlists_update_context(struct drm_i915_gem_request *rq) WARN_ON(!i915_gem_obj_is_pinned(ctx_obj)); WARN_ON(!i915_gem_obj_is_pinned(rb_obj)); - page = i915_gem_object_get_page(ctx_obj, 1); + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); reg_state = kmap_atomic(page); reg_state[CTX_RING_TAIL+1] = rq->tail; @@ -548,8 +550,6 @@ static int execlists_context_queue(struct drm_i915_gem_request *request) i915_gem_request_reference(request); - request->tail = request->ringbuf->tail; - spin_lock_irq(&ring->execlist_lock); list_for_each_entry(cursor, &ring->execlist_queue, execlist_link) @@ -702,13 +702,19 @@ static void intel_logical_ring_advance_and_submit(struct drm_i915_gem_request *request) { struct intel_engine_cs *ring = request->ring; + struct drm_i915_private *dev_priv = request->i915; intel_logical_ring_advance(request->ringbuf); + request->tail = request->ringbuf->tail; + if (intel_ring_stopped(ring)) return; - execlists_context_queue(request); + if (dev_priv->guc.execbuf_client) + i915_guc_submit(dev_priv->guc.execbuf_client, request); + else + execlists_context_queue(request); } static void __wrap_ring_buffer(struct intel_ringbuffer *ringbuf) @@ -998,6 +1004,7 @@ int logical_ring_flush_all_caches(struct drm_i915_gem_request *req) static int intel_lr_context_pin(struct drm_i915_gem_request *rq) { + struct drm_i915_private *dev_priv = rq->i915; struct intel_engine_cs *ring = rq->ring; struct drm_i915_gem_object *ctx_obj = rq->ctx->engine[ring->id].state; struct intel_ringbuffer *ringbuf = rq->ringbuf; @@ -1005,14 +1012,18 @@ static int intel_lr_context_pin(struct drm_i915_gem_request *rq) WARN_ON(!mutex_is_locked(&ring->dev->struct_mutex)); if (rq->ctx->engine[ring->id].pin_count++ == 0) { - ret = i915_gem_obj_ggtt_pin(ctx_obj, - GEN8_LR_CONTEXT_ALIGN, 0); + ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, + PIN_OFFSET_BIAS | GUC_WOPCM_TOP); if (ret) goto reset_pin_count; ret = intel_pin_and_map_ringbuffer_obj(ring->dev, ringbuf); if (ret) goto unpin_ctx_obj; + + /* Invalidate GuC TLB. */ + if (i915.enable_guc_submission) + I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); } return ret; @@ -2137,7 +2148,7 @@ populate_lr_context(struct intel_context *ctx, struct drm_i915_gem_object *ctx_o /* The second page of the context object contains some fields which must * be set up prior to the first execution. */ - page = i915_gem_object_get_page(ctx_obj, 1); + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); reg_state = kmap_atomic(page); /* A context is actually a big batch buffer with several MI_LOAD_REGISTER_IMM @@ -2307,12 +2318,13 @@ static void lrc_setup_hardware_status_page(struct intel_engine_cs *ring, struct drm_i915_gem_object *default_ctx_obj) { struct drm_i915_private *dev_priv = ring->dev->dev_private; + struct page *page; - /* The status page is offset 0 from the default context object - * in LRC mode. */ - ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj); - ring->status_page.page_addr = - kmap(sg_page(default_ctx_obj->pages->sgl)); + /* The HWSP is part of the default context object in LRC mode. */ + ring->status_page.gfx_addr = i915_gem_obj_ggtt_offset(default_ctx_obj) + + LRC_PPHWSP_PN * PAGE_SIZE; + page = i915_gem_object_get_page(default_ctx_obj, LRC_PPHWSP_PN); + ring->status_page.page_addr = kmap(page); ring->status_page.obj = default_ctx_obj; I915_WRITE(RING_HWS_PGA(ring->mmio_base), @@ -2338,6 +2350,7 @@ int intel_lr_context_deferred_create(struct intel_context *ctx, { const bool is_global_default_ctx = (ctx == ring->default_context); struct drm_device *dev = ring->dev; + struct drm_i915_private *dev_priv = dev->dev_private; struct drm_i915_gem_object *ctx_obj; uint32_t context_size; struct intel_ringbuffer *ringbuf; @@ -2348,6 +2361,9 @@ int intel_lr_context_deferred_create(struct intel_context *ctx, context_size = round_up(get_lr_context_size(ring), 4096); + /* One extra page as the sharing data between driver and GuC */ + context_size += PAGE_SIZE * LRC_PPHWSP_PN; + ctx_obj = i915_gem_alloc_object(dev, context_size); if (!ctx_obj) { DRM_DEBUG_DRIVER("Alloc LRC backing obj failed.\n"); @@ -2355,13 +2371,18 @@ int intel_lr_context_deferred_create(struct intel_context *ctx, } if (is_global_default_ctx) { - ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, 0); + ret = i915_gem_obj_ggtt_pin(ctx_obj, GEN8_LR_CONTEXT_ALIGN, + PIN_OFFSET_BIAS | GUC_WOPCM_TOP); if (ret) { DRM_DEBUG_DRIVER("Pin LRC backing obj failed: %d\n", ret); drm_gem_object_unreference(&ctx_obj->base); return ret; } + + /* Invalidate GuC TLB. */ + if (i915.enable_guc_submission) + I915_WRITE(GEN8_GTCR, GEN8_GTCR_INVALIDATE); } ringbuf = kzalloc(sizeof(*ringbuf), GFP_KERNEL); @@ -2374,7 +2395,7 @@ int intel_lr_context_deferred_create(struct intel_context *ctx, ringbuf->ring = ring; - ringbuf->size = 32 * PAGE_SIZE; + ringbuf->size = 4 * PAGE_SIZE; ringbuf->effective_size = ringbuf->size; ringbuf->head = 0; ringbuf->tail = 0; @@ -2474,7 +2495,7 @@ void intel_lr_context_reset(struct drm_device *dev, WARN(1, "Failed get_pages for context obj\n"); continue; } - page = i915_gem_object_get_page(ctx_obj, 1); + page = i915_gem_object_get_page(ctx_obj, LRC_STATE_PN); reg_state = kmap_atomic(page); reg_state[CTX_RING_HEAD+1] = 0; diff --git a/drivers/gpu/drm/i915/intel_lrc.h b/drivers/gpu/drm/i915/intel_lrc.h index 5e5788c83eeb..4cc54b371afd 100644 --- a/drivers/gpu/drm/i915/intel_lrc.h +++ b/drivers/gpu/drm/i915/intel_lrc.h @@ -68,6 +68,12 @@ static inline void intel_logical_ring_emit(struct intel_ringbuffer *ringbuf, } /* Logical Ring Contexts */ + +/* One extra page is added before LRC for GuC as shared data */ +#define LRC_GUCSHR_PN (0) +#define LRC_PPHWSP_PN (LRC_GUCSHR_PN + 1) +#define LRC_STATE_PN (LRC_PPHWSP_PN + 1) + void intel_lr_context_free(struct intel_context *ctx); int intel_lr_context_deferred_create(struct intel_context *ctx, struct intel_engine_cs *ring); -- 2.39.5