From: Chris Wilson Date: Fri, 9 Sep 2016 13:11:57 +0000 (+0100) Subject: drm/i915/guc: Prepare for nonblocking execbuf submission X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=dadd481bfe559f09bc3c42e9d552149c1cb84e6a;p=GitHub%2Fmoto-9609%2Fandroid_kernel_motorola_exynos9610.git drm/i915/guc: Prepare for nonblocking execbuf submission Currently the presumption is that the request construction and its submission to the GuC are all under the same holding of struct_mutex. We wish to relax this to separate the request construction and the later submission to the GuC. This requires us to reserve some space in the GuC command queue for the future submission. For flexibility to handle out-of-order request submission we do not preallocate the next slot in the GuC command queue during request construction, just ensuring that there is enough space later. Signed-off-by: Chris Wilson Link: http://patchwork.freedesktop.org/patch/msgid/20160909131201.16673-17-chris@chris-wilson.co.uk --- diff --git a/drivers/gpu/drm/i915/i915_guc_submission.c b/drivers/gpu/drm/i915/i915_guc_submission.c index 0eb6b71935cf..279a4d060288 100644 --- a/drivers/gpu/drm/i915/i915_guc_submission.c +++ b/drivers/gpu/drm/i915/i915_guc_submission.c @@ -432,20 +432,23 @@ int i915_guc_wq_check_space(struct drm_i915_gem_request *request) { const size_t wqi_size = sizeof(struct guc_wq_item); struct i915_guc_client *gc = request->i915->guc.execbuf_client; - struct guc_process_desc *desc; + struct guc_process_desc *desc = gc->client_base + gc->proc_desc_offset; u32 freespace; + int ret; - GEM_BUG_ON(gc == NULL); - - desc = gc->client_base + gc->proc_desc_offset; - + spin_lock(&gc->wq_lock); freespace = CIRC_SPACE(gc->wq_tail, desc->head, gc->wq_size); - if (likely(freespace >= wqi_size)) - return 0; - - gc->no_wq_space += 1; + freespace -= gc->wq_rsvd; + if (likely(freespace >= wqi_size)) { + gc->wq_rsvd += wqi_size; + ret = 0; + } else { + gc->no_wq_space++; + ret = -EAGAIN; + } + spin_unlock(&gc->wq_lock); - return -EAGAIN; + return ret; } static void guc_add_workqueue_item(struct i915_guc_client *gc, @@ -480,12 +483,14 @@ static void guc_add_workqueue_item(struct i915_guc_client *gc, * workqueue buffer dw by dw. */ BUILD_BUG_ON(wqi_size != 16); + GEM_BUG_ON(gc->wq_rsvd < wqi_size); /* postincrement WQ tail for next time */ wq_off = gc->wq_tail; + GEM_BUG_ON(wq_off & (wqi_size - 1)); gc->wq_tail += wqi_size; gc->wq_tail &= gc->wq_size - 1; - GEM_BUG_ON(wq_off & (wqi_size - 1)); + gc->wq_rsvd -= wqi_size; /* WQ starts from the page after doorbell / process_desc */ wq_page = (wq_off + GUC_DB_SIZE) >> PAGE_SHIFT; @@ -589,6 +594,7 @@ static void i915_guc_submit(struct drm_i915_gem_request *rq) struct i915_guc_client *client = guc->execbuf_client; int b_ret; + spin_lock(&client->wq_lock); guc_add_workqueue_item(client, rq); b_ret = guc_ring_doorbell(client); @@ -599,6 +605,7 @@ static void i915_guc_submit(struct drm_i915_gem_request *rq) guc->submissions[engine_id] += 1; guc->last_seqno[engine_id] = rq->fence.seqno; + spin_unlock(&client->wq_lock); } /* @@ -789,6 +796,8 @@ guc_client_alloc(struct drm_i915_private *dev_priv, /* We'll keep just the first (doorbell/proc) page permanently kmap'd. */ client->vma = vma; client->client_base = kmap(i915_vma_first_page(vma)); + + spin_lock_init(&client->wq_lock); client->wq_offset = GUC_DB_SIZE; client->wq_size = GUC_WQ_SIZE; @@ -1015,9 +1024,11 @@ int i915_guc_submission_enable(struct drm_i915_private *dev_priv) engine->submit_request = i915_guc_submit; /* Replay the current set of previously submitted requests */ - list_for_each_entry(request, &engine->request_list, link) + list_for_each_entry(request, &engine->request_list, link) { + client->wq_rsvd += sizeof(struct guc_wq_item); if (i915_sw_fence_done(&request->submit)) i915_guc_submit(request); + } } return 0; diff --git a/drivers/gpu/drm/i915/intel_guc.h b/drivers/gpu/drm/i915/intel_guc.h index c97326269588..467845967e0b 100644 --- a/drivers/gpu/drm/i915/intel_guc.h +++ b/drivers/gpu/drm/i915/intel_guc.h @@ -78,9 +78,11 @@ struct i915_guc_client { uint16_t doorbell_id; uint16_t padding[3]; /* Maintain alignment */ + spinlock_t wq_lock; uint32_t wq_offset; uint32_t wq_size; uint32_t wq_tail; + uint32_t wq_rsvd; uint32_t no_wq_space; uint32_t b_fail; int retcode;