From a05502e5cfa9abe17a16592be82c2f5692c91f35 Mon Sep 17 00:00:00 2001 From: Horace Chen Date: Fri, 29 Sep 2017 14:41:57 +0800 Subject: drm/amdgpu: Reserve shared memory on VRAM for SR-IOV MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit SR-IOV need to reserve a piece of shared VRAM at the exact place to exchange data betweem PF and VF. The start address and size of the shared mem are passed to guest through VBIOS structure VRAM_UsageByFirmware. VRAM_UsageByFirmware is a general feature in VBIOS, it indicates that VBIOS need to reserve a piece of memory on the VRAM. Because the mem address is specified. Reserve it early in amdgpu_ttm_init to make sure that it can monoplize the space. Signed-off-by: Horace Chen Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a23b8af95319..5c4bed7778d9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1378,6 +1378,18 @@ struct amdgpu_atcs { struct amdgpu_atcs_functions functions; }; +/* + * Firmware VRAM reservation + */ +struct amdgpu_fw_vram_usage { + u64 start_offset; + u64 size; + struct amdgpu_bo *reserved_bo; + void *va; +}; + +int amdgpu_fw_reserve_vram_init(struct amdgpu_device *adev); + /* * CGS */ @@ -1582,6 +1594,8 @@ struct amdgpu_device { struct delayed_work late_init_work; struct amdgpu_virt virt; + /* firmware VRAM reservation */ + struct amdgpu_fw_vram_usage fw_vram_usage; /* link all shadow bo */ struct list_head shadow_list; -- cgit v1.2.3 From b8866c26ec072f1924f5cf601ebea33ca9823326 Mon Sep 17 00:00:00 2001 From: Andres Rodriguez Date: Fri, 28 Apr 2017 20:05:51 -0400 Subject: drm/amdgpu: implement ring set_priority for gfx_v8 compute v9 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Programming CP_HQD_QUEUE_PRIORITY enables a queue to take priority over other queues on the same pipe. Multiple queues on a pipe are timesliced so this gives us full precedence over other queues. Programming CP_HQD_PIPE_PRIORITY changes the SPI_ARB_PRIORITY of the wave as follows: 0x2: CS_H 0x1: CS_M 0x0: CS_L The SPI block will then dispatch work according to the policy set by SPI_ARB_PRIORITY. In the current policy CS_H is higher priority than gfx. In order to prevent getting stuck in loops of resources bouncing between GFX and high priority compute and introducing further latency, we statically reserve a portion of the pipe. v2: fix srbm_select to ring->queue and use ring->funcs->type v3: use AMD_SCHED_PRIORITY_* instead of AMDGPU_CTX_PRIORITY_* v4: switch int to enum amd_sched_priority v5: corresponding changes for srbm_lock v6: change CU reservation to PIPE_PERCENT allocation v7: use kiq instead of MMIO v8: back to MMIO, and make the implementation sleep safe. v9: corresponding changes for splitting HIGH into _HW/_SW Acked-by: Christian König Signed-off-by: Andres Rodriguez Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 + drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 99 ++++++++++++++++++++++++++++++ 3 files changed, 105 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 5c4bed7778d9..715ce4863bc8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1035,6 +1035,10 @@ struct amdgpu_gfx { bool in_suspend; /* NGG */ struct amdgpu_ngg ngg; + + /* pipe reservation */ + struct mutex pipe_reserve_mutex; + DECLARE_BITMAP (pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); }; int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 7b3e3b5461c3..412ad99d8871 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -2094,6 +2094,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, adev->vm_manager.vm_pte_num_rings = 0; adev->gart.gart_funcs = NULL; adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS); + bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES); adev->smc_rreg = &amdgpu_invalid_rreg; adev->smc_wreg = &amdgpu_invalid_wreg; @@ -2122,6 +2123,7 @@ int amdgpu_device_init(struct amdgpu_device *adev, mutex_init(&adev->pm.mutex); mutex_init(&adev->gfx.gpu_clock_mutex); mutex_init(&adev->srbm_mutex); + mutex_init(&adev->gfx.pipe_reserve_mutex); mutex_init(&adev->grbm_idx_mutex); mutex_init(&adev->mn_lock); mutex_init(&adev->virt.vf_errors.lock); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 147e92b3a959..51896b7353b6 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -6394,6 +6394,104 @@ static void gfx_v8_0_ring_set_wptr_compute(struct amdgpu_ring *ring) WDOORBELL32(ring->doorbell_index, lower_32_bits(ring->wptr)); } +static void gfx_v8_0_ring_set_pipe_percent(struct amdgpu_ring *ring, + bool acquire) +{ + struct amdgpu_device *adev = ring->adev; + int pipe_num, tmp, reg; + int pipe_percent = acquire ? SPI_WCL_PIPE_PERCENT_GFX__VALUE_MASK : 0x1; + + pipe_num = ring->me * adev->gfx.mec.num_pipe_per_mec + ring->pipe; + + /* first me only has 2 entries, GFX and HP3D */ + if (ring->me > 0) + pipe_num -= 2; + + reg = mmSPI_WCL_PIPE_PERCENT_GFX + pipe_num; + tmp = RREG32(reg); + tmp = REG_SET_FIELD(tmp, SPI_WCL_PIPE_PERCENT_GFX, VALUE, pipe_percent); + WREG32(reg, tmp); +} + +static void gfx_v8_0_pipe_reserve_resources(struct amdgpu_device *adev, + struct amdgpu_ring *ring, + bool acquire) +{ + int i, pipe; + bool reserve; + struct amdgpu_ring *iring; + + mutex_lock(&adev->gfx.pipe_reserve_mutex); + pipe = amdgpu_gfx_queue_to_bit(adev, ring->me, ring->pipe, 0); + if (acquire) + set_bit(pipe, adev->gfx.pipe_reserve_bitmap); + else + clear_bit(pipe, adev->gfx.pipe_reserve_bitmap); + + if (!bitmap_weight(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES)) { + /* Clear all reservations - everyone reacquires all resources */ + for (i = 0; i < adev->gfx.num_gfx_rings; ++i) + gfx_v8_0_ring_set_pipe_percent(&adev->gfx.gfx_ring[i], + true); + + for (i = 0; i < adev->gfx.num_compute_rings; ++i) + gfx_v8_0_ring_set_pipe_percent(&adev->gfx.compute_ring[i], + true); + } else { + /* Lower all pipes without a current reservation */ + for (i = 0; i < adev->gfx.num_gfx_rings; ++i) { + iring = &adev->gfx.gfx_ring[i]; + pipe = amdgpu_gfx_queue_to_bit(adev, + iring->me, + iring->pipe, + 0); + reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap); + gfx_v8_0_ring_set_pipe_percent(iring, reserve); + } + + for (i = 0; i < adev->gfx.num_compute_rings; ++i) { + iring = &adev->gfx.compute_ring[i]; + pipe = amdgpu_gfx_queue_to_bit(adev, + iring->me, + iring->pipe, + 0); + reserve = test_bit(pipe, adev->gfx.pipe_reserve_bitmap); + gfx_v8_0_ring_set_pipe_percent(iring, reserve); + } + } + + mutex_unlock(&adev->gfx.pipe_reserve_mutex); +} + +static void gfx_v8_0_hqd_set_priority(struct amdgpu_device *adev, + struct amdgpu_ring *ring, + bool acquire) +{ + uint32_t pipe_priority = acquire ? 0x2 : 0x0; + uint32_t queue_priority = acquire ? 0xf : 0x0; + + mutex_lock(&adev->srbm_mutex); + vi_srbm_select(adev, ring->me, ring->pipe, ring->queue, 0); + + WREG32(mmCP_HQD_PIPE_PRIORITY, pipe_priority); + WREG32(mmCP_HQD_QUEUE_PRIORITY, queue_priority); + + vi_srbm_select(adev, 0, 0, 0, 0); + mutex_unlock(&adev->srbm_mutex); +} +static void gfx_v8_0_ring_set_priority_compute(struct amdgpu_ring *ring, + enum amd_sched_priority priority) +{ + struct amdgpu_device *adev = ring->adev; + bool acquire = priority == AMD_SCHED_PRIORITY_HIGH_HW; + + if (ring->funcs->type != AMDGPU_RING_TYPE_COMPUTE) + return; + + gfx_v8_0_hqd_set_priority(adev, ring, acquire); + gfx_v8_0_pipe_reserve_resources(adev, ring, acquire); +} + static void gfx_v8_0_ring_emit_fence_compute(struct amdgpu_ring *ring, u64 addr, u64 seq, unsigned flags) @@ -6839,6 +6937,7 @@ static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_compute = { .test_ib = gfx_v8_0_ring_test_ib, .insert_nop = amdgpu_ring_insert_nop, .pad_ib = amdgpu_ring_generic_pad_ib, + .set_priority = gfx_v8_0_ring_set_priority_compute, }; static const struct amdgpu_ring_funcs gfx_v8_0_ring_funcs_kiq = { -- cgit v1.2.3 From c23be4ae1d50e8d93f805de7ae3e8ea49a4ad781 Mon Sep 17 00:00:00 2001 From: Andres Rodriguez Date: Tue, 6 Jun 2017 20:20:38 -0400 Subject: drm/amdgpu: add plumbing for ctx priority changes v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Introduce amdgpu_ctx_priority_override(). A mechanism to override a context's priority. An override can be terminated by setting the override to AMD_SCHED_PRIORITY_UNSET. v2: change refcounted interface for a direct set Signed-off-by: Andres Rodriguez Acked-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 6 +++++- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 29 +++++++++++++++++++++++++++++ 2 files changed, 34 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 715ce4863bc8..951c8db01412 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -735,7 +735,9 @@ struct amdgpu_ctx { spinlock_t ring_lock; struct dma_fence **fences; struct amdgpu_ctx_ring rings[AMDGPU_MAX_RINGS]; - bool preamble_presented; + bool preamble_presented; + enum amd_sched_priority init_priority; + enum amd_sched_priority override_priority; }; struct amdgpu_ctx_mgr { @@ -752,6 +754,8 @@ int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, struct dma_fence *fence, uint64_t *seq); struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, uint64_t seq); +void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, + enum amd_sched_priority priority); int amdgpu_ctx_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 29eebdc30a4c..d2ef24f4b56d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -72,6 +72,8 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, } ctx->reset_counter = atomic_read(&adev->gpu_reset_counter); + ctx->init_priority = priority; + ctx->override_priority = AMD_SCHED_PRIORITY_UNSET; /* create context entity for each ring */ for (i = 0; i < adev->num_rings; i++) { @@ -362,6 +364,33 @@ struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, return fence; } +void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, + enum amd_sched_priority priority) +{ + int i; + struct amdgpu_device *adev = ctx->adev; + struct amd_sched_rq *rq; + struct amd_sched_entity *entity; + struct amdgpu_ring *ring; + enum amd_sched_priority ctx_prio; + + ctx->override_priority = priority; + + ctx_prio = (ctx->override_priority == AMD_SCHED_PRIORITY_UNSET) ? + ctx->init_priority : ctx->override_priority; + + for (i = 0; i < adev->num_rings; i++) { + ring = adev->rings[i]; + entity = &ctx->rings[i].entity; + rq = &ring->sched.sched_rq[ctx_prio]; + + if (ring->funcs->type == AMDGPU_RING_TYPE_KIQ) + continue; + + amd_sched_entity_set_rq(entity, rq); + } +} + void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr) { mutex_init(&mgr->lock); -- cgit v1.2.3 From 0ae94444c08a0adf2fab4aab26be0646ee445a19 Mon Sep 17 00:00:00 2001 From: Andrey Grodzovsky Date: Tue, 10 Oct 2017 16:50:17 -0400 Subject: drm/amdgpu: Move old fence waiting before reservation lock is aquired v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Helps avoiding deadlock during GPU reset. Added mutex to amdgpu_ctx to preserve order of fences on a ring. v2: Put waiting logic in a function in a seperate function in amdgpu_ctx.c Signed-off-by: Andrey Grodzovsky Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 4 ++++ drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 ++++++---- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 30 ++++++++++++++++++++++++------ 3 files changed, 34 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 951c8db01412..76033e2cdba8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -738,6 +738,7 @@ struct amdgpu_ctx { bool preamble_presented; enum amd_sched_priority init_priority; enum amd_sched_priority override_priority; + struct mutex lock; }; struct amdgpu_ctx_mgr { @@ -760,9 +761,12 @@ void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, int amdgpu_ctx_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); +int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx, unsigned ring_id); + void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr); void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr); + /* * file private structure */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 9166d5e1e557..5de092eab0fa 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -90,6 +90,8 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) goto free_chunk; } + mutex_lock(&p->ctx->lock); + /* get chunks */ chunk_array_user = u64_to_user_ptr(cs->in.chunks); if (copy_from_user(chunk_array, chunk_array_user, @@ -737,8 +739,10 @@ static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, dma_fence_put(parser->fence); - if (parser->ctx) + if (parser->ctx) { + mutex_unlock(&parser->ctx->lock); amdgpu_ctx_put(parser->ctx); + } if (parser->bo_list) amdgpu_bo_list_put(parser->bo_list); @@ -895,9 +899,7 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev, r = amdgpu_ring_parse_cs(ring, p, j); if (r) return r; - } - j++; } @@ -985,7 +987,7 @@ static int amdgpu_cs_ib_fill(struct amdgpu_device *adev, parser->job->ring->funcs->type == AMDGPU_RING_TYPE_VCE)) return -EINVAL; - return 0; + return amdgpu_ctx_wait_prev_fence(parser->ctx, parser->job->ring->idx); } static int amdgpu_cs_process_fence_dep(struct amdgpu_cs_parser *p, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index a78b03f65c69..4309820658c4 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -67,6 +67,8 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, if (!ctx->fences) return -ENOMEM; + mutex_init(&ctx->lock); + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { ctx->rings[i].sequence = 1; ctx->rings[i].fences = &ctx->fences[amdgpu_sched_jobs * i]; @@ -126,6 +128,8 @@ static void amdgpu_ctx_fini(struct amdgpu_ctx *ctx) &ctx->rings[i].entity); amdgpu_queue_mgr_fini(adev, &ctx->queue_mgr); + + mutex_destroy(&ctx->lock); } static int amdgpu_ctx_alloc(struct amdgpu_device *adev, @@ -296,12 +300,8 @@ int amdgpu_ctx_add_fence(struct amdgpu_ctx *ctx, struct amdgpu_ring *ring, idx = seq & (amdgpu_sched_jobs - 1); other = cring->fences[idx]; - if (other) { - signed long r; - r = dma_fence_wait_timeout(other, true, MAX_SCHEDULE_TIMEOUT); - if (r < 0) - return r; - } + if (other) + BUG_ON(!dma_fence_is_signaled(other)); dma_fence_get(fence); @@ -372,6 +372,24 @@ void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, } } +int amdgpu_ctx_wait_prev_fence(struct amdgpu_ctx *ctx, unsigned ring_id) +{ + struct amdgpu_ctx_ring *cring = &ctx->rings[ring_id]; + unsigned idx = cring->sequence & (amdgpu_sched_jobs - 1); + struct dma_fence *other = cring->fences[idx]; + + if (other) { + signed long r; + r = dma_fence_wait_timeout(other, false, MAX_SCHEDULE_TIMEOUT); + if (r < 0) { + DRM_ERROR("Error (%ld) waiting for fence!\n", r); + return r; + } + } + + return 0; +} + void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr) { mutex_init(&mgr->lock); -- cgit v1.2.3 From 14e47f93c5cc4a1237dbacc137e174706093b69c Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 9 Oct 2017 15:04:41 +0200 Subject: drm/amdgpu: keep copy of VRAM lost counter in job MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Instead of reading the current counter from fpriv. Signed-off-by: Christian König Reviewed-by: Nicolai Hähnle Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 13 +++++++------ 3 files changed, 10 insertions(+), 6 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 76033e2cdba8..aa70f8c045b1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1125,6 +1125,7 @@ struct amdgpu_job { uint32_t gds_base, gds_size; uint32_t gws_base, gws_size; uint32_t oa_base, oa_size; + uint32_t vram_lost_counter; /* user fence handling */ uint64_t uf_addr; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 0c07df72743c..9daa7cac0ffb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -172,6 +172,8 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) if (ret) goto free_all_kdata; + p->job->vram_lost_counter = fpriv->vram_lost_counter; + if (p->uf_entry.robj) p->job->uf_addr = uf_offset; kfree(chunk_array); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 83d13431cbdd..4f2b5acc8743 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -61,6 +61,7 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs, (*job)->vm = vm; (*job)->ibs = (void *)&(*job)[1]; (*job)->num_ibs = num_ibs; + (*job)->vram_lost_counter = atomic_read(&adev->vram_lost_counter); amdgpu_sync_create(&(*job)->sync); amdgpu_sync_create(&(*job)->dep_sync); @@ -180,8 +181,8 @@ static struct dma_fence *amdgpu_job_dependency(struct amd_sched_job *sched_job) static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job) { struct dma_fence *fence = NULL; + struct amdgpu_device *adev; struct amdgpu_job *job; - struct amdgpu_fpriv *fpriv = NULL; int r; if (!sched_job) { @@ -189,17 +190,17 @@ static struct dma_fence *amdgpu_job_run(struct amd_sched_job *sched_job) return NULL; } job = to_amdgpu_job(sched_job); + adev = job->adev; BUG_ON(amdgpu_sync_peek_fence(&job->sync, NULL)); trace_amdgpu_sched_run_job(job); - if (job->vm) - fpriv = container_of(job->vm, struct amdgpu_fpriv, vm); /* skip ib schedule when vram is lost */ - if (fpriv && amdgpu_kms_vram_lost(job->adev, fpriv)) + if (job->vram_lost_counter != atomic_read(&adev->vram_lost_counter)) { DRM_ERROR("Skip scheduling IBs!\n"); - else { - r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, &fence); + } else { + r = amdgpu_ib_schedule(job->ring, job->num_ibs, job->ibs, job, + &fence); if (r) DRM_ERROR("Error scheduling IBs (%d)\n", r); } -- cgit v1.2.3 From e55f2b646df3318e24f12b8388ab6e5cccb3e92d Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 9 Oct 2017 15:18:43 +0200 Subject: drm/amdgpu: move the VRAM lost counter per context MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Instead of per device track the VRAM lost per context and return ECANCELED instead of ENODEV. Signed-off-by: Christian König Reviewed-by: Nicolai Hähnle Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 6 ++---- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 9 +++++---- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 16 ---------------- 4 files changed, 8 insertions(+), 24 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index aa70f8c045b1..67b864436be1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -732,10 +732,11 @@ struct amdgpu_ctx { struct amdgpu_device *adev; struct amdgpu_queue_mgr queue_mgr; unsigned reset_counter; + uint32_t vram_lost_counter; spinlock_t ring_lock; struct dma_fence **fences; struct amdgpu_ctx_ring rings[AMDGPU_MAX_RINGS]; - bool preamble_presented; + bool preamble_presented; enum amd_sched_priority init_priority; enum amd_sched_priority override_priority; struct mutex lock; @@ -778,7 +779,6 @@ struct amdgpu_fpriv { struct mutex bo_list_lock; struct idr bo_list_handles; struct amdgpu_ctx_mgr ctx_mgr; - u32 vram_lost_counter; }; /* @@ -1860,8 +1860,6 @@ static inline bool amdgpu_has_atpx(void) { return false; } extern const struct drm_ioctl_desc amdgpu_ioctls_kms[]; extern const int amdgpu_max_kms_ioctl; -bool amdgpu_kms_vram_lost(struct amdgpu_device *adev, - struct amdgpu_fpriv *fpriv); int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags); void amdgpu_driver_unload_kms(struct drm_device *dev); void amdgpu_driver_lastclose_kms(struct drm_device *dev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 9daa7cac0ffb..b355189533d2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -172,7 +172,11 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) if (ret) goto free_all_kdata; - p->job->vram_lost_counter = fpriv->vram_lost_counter; + p->job->vram_lost_counter = atomic_read(&p->adev->vram_lost_counter); + if (p->ctx->vram_lost_counter != p->job->vram_lost_counter) { + ret = -ECANCELED; + goto free_all_kdata; + } if (p->uf_entry.robj) p->job->uf_addr = uf_offset; @@ -1205,7 +1209,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { struct amdgpu_device *adev = dev->dev_private; - struct amdgpu_fpriv *fpriv = filp->driver_priv; union drm_amdgpu_cs *cs = data; struct amdgpu_cs_parser parser = {}; bool reserved_buffers = false; @@ -1213,8 +1216,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) if (!adev->accel_working) return -EBUSY; - if (amdgpu_kms_vram_lost(adev, fpriv)) - return -ENODEV; parser.adev = adev; parser.filp = filp; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 4309820658c4..c184468e2b2b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -75,6 +75,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, } ctx->reset_counter = atomic_read(&adev->gpu_reset_counter); + ctx->vram_lost_counter = atomic_read(&adev->vram_lost_counter); ctx->init_priority = priority; ctx->override_priority = AMD_SCHED_PRIORITY_UNSET; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index f759836d10ef..ff1a416a66c9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -789,21 +789,6 @@ void amdgpu_driver_lastclose_kms(struct drm_device *dev) vga_switcheroo_process_delayed_switch(); } -/** - * amdgpu_kms_vram_lost - check if VRAM was lost for this client - * - * @adev: amdgpu device - * @fpriv: client private - * - * Check if all CS is blocked for the client because of lost VRAM - */ -bool amdgpu_kms_vram_lost(struct amdgpu_device *adev, - struct amdgpu_fpriv *fpriv) -{ - return fpriv->vram_lost_counter != - atomic_read(&adev->vram_lost_counter); -} - /** * amdgpu_driver_open_kms - drm callback for open * @@ -860,7 +845,6 @@ int amdgpu_driver_open_kms(struct drm_device *dev, struct drm_file *file_priv) amdgpu_ctx_mgr_init(&fpriv->ctx_mgr); - fpriv->vram_lost_counter = atomic_read(&adev->vram_lost_counter); file_priv->driver_priv = fpriv; out_suspend: -- cgit v1.2.3 From 896a664c969307a4954cd8c7754fd5d4e5358396 Mon Sep 17 00:00:00 2001 From: Monk Liu Date: Tue, 17 Oct 2017 19:23:42 +0800 Subject: drm/amdgpu:reduce wb to 512 slot MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit with current WB usage we only use 57 slots, so 512 is extreamly sufficient, and reduce to 512 can make WB fit into one page. Signed-off-by: Monk Liu Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 67b864436be1..b8ba1f5ae5e7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1151,7 +1151,7 @@ static inline void amdgpu_set_ib_value(struct amdgpu_cs_parser *p, /* * Writeback */ -#define AMDGPU_MAX_WB 1024 /* Reserve at most 1024 WB slots for amdgpu-owned rings. */ +#define AMDGPU_MAX_WB 512 /* Reserve at most 512 WB slots for amdgpu-owned rings. */ struct amdgpu_wb { struct amdgpu_bo *wb_obj; -- cgit v1.2.3 From 43ca8efa46d9b1c4defa1b27c4dd1ef3866aaad9 Mon Sep 17 00:00:00 2001 From: pding Date: Fri, 13 Oct 2017 15:38:35 +0800 Subject: drm/amdgpu: busywait KIQ register accessing (v4) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Register accessing is performed when IRQ is disabled. Never sleep in this function. Known issue: dead sleep in many use cases of index/data registers. v2: - wrap polling fence functions. - don't trigger IRQ for polling in case of wrongly fence signal. v3: - handle wrap round gracefully. - add comments for polling function v4: - don't return negative timeout confused with error code Signed-off-by: pding Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 8 ++--- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 50 ++++++++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h | 4 +++ drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c | 30 ++++++++---------- 6 files changed, 71 insertions(+), 25 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu.h') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index b8ba1f5ae5e7..cbcb6a153aba 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -879,7 +879,7 @@ struct amdgpu_mec { struct amdgpu_kiq { u64 eop_gpu_addr; struct amdgpu_bo *eop_obj; - struct mutex ring_mutex; + spinlock_t ring_lock; struct amdgpu_ring ring; struct amdgpu_irq_src irq; }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index 3a6ce6386ad0..0b9332e65a4c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -109,10 +109,8 @@ uint32_t amdgpu_mm_rreg(struct amdgpu_device *adev, uint32_t reg, { uint32_t ret; - if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) { - BUG_ON(in_interrupt()); + if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) return amdgpu_virt_kiq_rreg(adev, reg); - } if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX)) ret = readl(((void __iomem *)adev->rmmio) + (reg * 4)); @@ -137,10 +135,8 @@ void amdgpu_mm_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v, adev->last_mm_index = v; } - if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) { - BUG_ON(in_interrupt()); + if (!(acc_flags & AMDGPU_REGS_NO_KIQ) && amdgpu_sriov_runtime(adev)) return amdgpu_virt_kiq_wreg(adev, reg, v); - } if ((reg * 4) < adev->rmmio_size && !(acc_flags & AMDGPU_REGS_IDX)) writel(v, ((void __iomem *)adev->rmmio) + (reg * 4)); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 7bdedd788f5a..fb9f88ef6059 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -168,6 +168,32 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **f) return 0; } +/** + * amdgpu_fence_emit_polling - emit a fence on the requeste ring + * + * @ring: ring the fence is associated with + * @s: resulting sequence number + * + * Emits a fence command on the requested ring (all asics). + * Used For polling fence. + * Returns 0 on success, -ENOMEM on failure. + */ +int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s) +{ + uint32_t seq; + + if (!s) + return -EINVAL; + + seq = ++ring->fence_drv.sync_seq; + amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, + seq, AMDGPU_FENCE_FLAG_INT); + + *s = seq; + + return 0; +} + /** * amdgpu_fence_schedule_fallback - schedule fallback check * @@ -281,6 +307,30 @@ int amdgpu_fence_wait_empty(struct amdgpu_ring *ring) return r; } +/** + * amdgpu_fence_wait_polling - busy wait for givn sequence number + * + * @ring: ring index the fence is associated with + * @wait_seq: sequence number to wait + * @timeout: the timeout for waiting in usecs + * + * Wait for all fences on the requested ring to signal (all asics). + * Returns left time if no timeout, 0 or minus if timeout. + */ +signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring, + uint32_t wait_seq, + signed long timeout) +{ + uint32_t seq; + + do { + seq = amdgpu_fence_read(ring); + udelay(5); + timeout -= 5; + } while ((int32_t)(wait_seq - seq) > 0 && timeout > 0); + + return timeout > 0 ? timeout : 0; +} /** * amdgpu_fence_count_emitted - get the count of emitted fences * diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c index 83435ccbad44..ef043361009f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c @@ -201,7 +201,7 @@ int amdgpu_gfx_kiq_init_ring(struct amdgpu_device *adev, struct amdgpu_kiq *kiq = &adev->gfx.kiq; int r = 0; - mutex_init(&kiq->ring_mutex); + spin_lock_init(&kiq->ring_lock); r = amdgpu_wb_get(adev, &adev->virt.reg_val_offs); if (r) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h index 0d9ce141404c..b18c2b96691f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h @@ -90,8 +90,12 @@ int amdgpu_fence_driver_start_ring(struct amdgpu_ring *ring, void amdgpu_fence_driver_suspend(struct amdgpu_device *adev); void amdgpu_fence_driver_resume(struct amdgpu_device *adev); int amdgpu_fence_emit(struct amdgpu_ring *ring, struct dma_fence **fence); +int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s); void amdgpu_fence_process(struct amdgpu_ring *ring); int amdgpu_fence_wait_empty(struct amdgpu_ring *ring); +signed long amdgpu_fence_wait_polling(struct amdgpu_ring *ring, + uint32_t wait_seq, + signed long timeout); unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring); /* diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c index ed7be2eb24b0..e97f80f86005 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c @@ -22,7 +22,7 @@ */ #include "amdgpu.h" -#define MAX_KIQ_REG_WAIT 100000 +#define MAX_KIQ_REG_WAIT 100000000 /* in usecs */ int amdgpu_allocate_static_csa(struct amdgpu_device *adev) { @@ -114,27 +114,24 @@ void amdgpu_virt_init_setting(struct amdgpu_device *adev) uint32_t amdgpu_virt_kiq_rreg(struct amdgpu_device *adev, uint32_t reg) { signed long r; - uint32_t val; - struct dma_fence *f; + uint32_t val, seq; struct amdgpu_kiq *kiq = &adev->gfx.kiq; struct amdgpu_ring *ring = &kiq->ring; BUG_ON(!ring->funcs->emit_rreg); - mutex_lock(&kiq->ring_mutex); + spin_lock(&kiq->ring_lock); amdgpu_ring_alloc(ring, 32); amdgpu_ring_emit_rreg(ring, reg); - amdgpu_fence_emit(ring, &f); + amdgpu_fence_emit_polling(ring, &seq); amdgpu_ring_commit(ring); - mutex_unlock(&kiq->ring_mutex); + spin_unlock(&kiq->ring_lock); - r = dma_fence_wait_timeout(f, false, msecs_to_jiffies(MAX_KIQ_REG_WAIT)); - dma_fence_put(f); + r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); if (r < 1) { - DRM_ERROR("wait for kiq fence error: %ld.\n", r); + DRM_ERROR("wait for kiq fence error: %ld\n", r); return ~0; } - val = adev->wb.wb[adev->virt.reg_val_offs]; return val; @@ -143,23 +140,22 @@ uint32_t amdgpu_virt_kiq_rreg(struct amdgpu_device *adev, uint32_t reg) void amdgpu_virt_kiq_wreg(struct amdgpu_device *adev, uint32_t reg, uint32_t v) { signed long r; - struct dma_fence *f; + uint32_t seq; struct amdgpu_kiq *kiq = &adev->gfx.kiq; struct amdgpu_ring *ring = &kiq->ring; BUG_ON(!ring->funcs->emit_wreg); - mutex_lock(&kiq->ring_mutex); + spin_lock(&kiq->ring_lock); amdgpu_ring_alloc(ring, 32); amdgpu_ring_emit_wreg(ring, reg, v); - amdgpu_fence_emit(ring, &f); + amdgpu_fence_emit_polling(ring, &seq); amdgpu_ring_commit(ring); - mutex_unlock(&kiq->ring_mutex); + spin_unlock(&kiq->ring_lock); - r = dma_fence_wait_timeout(f, false, msecs_to_jiffies(MAX_KIQ_REG_WAIT)); + r = amdgpu_fence_wait_polling(ring, seq, MAX_KIQ_REG_WAIT); if (r < 1) - DRM_ERROR("wait for kiq fence error: %ld.\n", r); - dma_fence_put(f); + DRM_ERROR("wait for kiq fence error: %ld\n", r); } /** -- cgit v1.2.3