From 96ea47c0ec8c012509116bee8c57414281428fc4 Mon Sep 17 00:00:00 2001 From: Michel Dänzer Date: Thu, 5 Nov 2015 17:25:26 +0900 Subject: drm/radeon: Disable uncacheable CPU mappings of GTT with RV6xx MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit They reportedly cause random GPU hangs. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91268 Reviewed-by: Christian König Cc: stable@vger.kernel.org Signed-off-by: Michel Dänzer Signed-off-by: Alex Deucher --- drivers/gpu/drm/radeon/radeon_object.c | 6 ++++++ 1 file changed, 6 insertions(+) (limited to 'drivers') diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c index d3024883b844..a35f5af77c6a 100644 --- a/drivers/gpu/drm/radeon/radeon_object.c +++ b/drivers/gpu/drm/radeon/radeon_object.c @@ -221,6 +221,12 @@ int radeon_bo_create(struct radeon_device *rdev, if (!(rdev->flags & RADEON_IS_PCIE)) bo->flags &= ~(RADEON_GEM_GTT_WC | RADEON_GEM_GTT_UC); + /* Write-combined CPU mappings of GTT cause GPU hangs with RV6xx + * See https://bugs.freedesktop.org/show_bug.cgi?id=91268 + */ + if (rdev->family >= CHIP_RV610 && rdev->family <= CHIP_RV635) + bo->flags &= ~(RADEON_GEM_GTT_WC | RADEON_GEM_GTT_UC); + #ifdef CONFIG_X86_32 /* XXX: Write-combined CPU mappings of GTT seem broken on 32-bit * See https://bugs.freedesktop.org/show_bug.cgi?id=84627 -- cgit v1.2.3 From a28bbd5824d4a2af98de45b300ab8d8fb39739fc Mon Sep 17 00:00:00 2001 From: Michel Dänzer Date: Thu, 5 Nov 2015 17:25:27 +0900 Subject: drm/radeon: Always disable RADEON_GEM_GTT_UC along with RADEON_GEM_GTT_WC MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Write-combining is a CPU feature. From the GPU POV, these both simply mean no GPU<->CPU cache coherency. Reviewed-by: Christian König Signed-off-by: Michel Dänzer Signed-off-by: Alex Deucher --- drivers/gpu/drm/radeon/radeon_object.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c index a35f5af77c6a..dda2ecfbf056 100644 --- a/drivers/gpu/drm/radeon/radeon_object.c +++ b/drivers/gpu/drm/radeon/radeon_object.c @@ -231,7 +231,7 @@ int radeon_bo_create(struct radeon_device *rdev, /* XXX: Write-combined CPU mappings of GTT seem broken on 32-bit * See https://bugs.freedesktop.org/show_bug.cgi?id=84627 */ - bo->flags &= ~RADEON_GEM_GTT_WC; + bo->flags &= ~(RADEON_GEM_GTT_WC | RADEON_GEM_GTT_UC); #elif defined(CONFIG_X86) && !defined(CONFIG_X86_PAT) /* Don't try to enable write-combining when it can't work, or things * may be slow @@ -243,7 +243,7 @@ int radeon_bo_create(struct radeon_device *rdev, DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for " "better performance thanks to write-combining\n"); - bo->flags &= ~RADEON_GEM_GTT_WC; + bo->flags &= ~(RADEON_GEM_GTT_WC | RADEON_GEM_GTT_UC); #endif radeon_ttm_placement_from_domain(bo, domain); -- cgit v1.2.3 From 938204985c1c46c28a88cb558d9e38610044b67a Mon Sep 17 00:00:00 2001 From: Michel Dänzer Date: Thu, 5 Nov 2015 17:25:28 +0900 Subject: drm/radeon: Only prompt for enabling PAT when we'd allow write-combining MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit No use bothering users about this for whom we disable write-combining for other reasons anyway. Reviewed-by: Christian König Signed-off-by: Michel Dänzer Signed-off-by: Alex Deucher --- drivers/gpu/drm/radeon/radeon_object.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c index dda2ecfbf056..84d45633d28c 100644 --- a/drivers/gpu/drm/radeon/radeon_object.c +++ b/drivers/gpu/drm/radeon/radeon_object.c @@ -241,8 +241,9 @@ int radeon_bo_create(struct radeon_device *rdev, #warning Please enable CONFIG_MTRR and CONFIG_X86_PAT for better performance \ thanks to write-combining - DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for " - "better performance thanks to write-combining\n"); + if (bo->flags & RADEON_GEM_GTT_WC) + DRM_INFO_ONCE("Please enable CONFIG_MTRR and CONFIG_X86_PAT for " + "better performance thanks to write-combining\n"); bo->flags &= ~(RADEON_GEM_GTT_WC | RADEON_GEM_GTT_UC); #endif -- cgit v1.2.3 From fa6760482bed1ed752568fe09135c078e9768595 Mon Sep 17 00:00:00 2001 From: Flora Cui Date: Fri, 30 Oct 2015 18:29:46 +0800 Subject: drm/amdgpu: update Fiji's mmPA_SC_RASTER_CONFIG value Change-Id: I6d138306a878450e5bf8a77a2f1aacc380a39fe5 Signed-off-by: Flora Cui Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 6776cf756d40..4410e5735154 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -2971,10 +2971,13 @@ static int gfx_v8_0_cp_gfx_start(struct amdgpu_device *adev) amdgpu_ring_write(ring, mmPA_SC_RASTER_CONFIG - PACKET3_SET_CONTEXT_REG_START); switch (adev->asic_type) { case CHIP_TONGA: - case CHIP_FIJI: amdgpu_ring_write(ring, 0x16000012); amdgpu_ring_write(ring, 0x0000002A); break; + case CHIP_FIJI: + amdgpu_ring_write(ring, 0x3a00161a); + amdgpu_ring_write(ring, 0x0000002e); + break; case CHIP_TOPAZ: case CHIP_CARRIZO: amdgpu_ring_write(ring, 0x00000002); -- cgit v1.2.3 From 935c186aaecc25258495351adaba34f7c507d298 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 3 Nov 2015 21:23:08 +0100 Subject: drm/amdgpu: remove fence trace points MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Mostly unused and replaced by the common trace points. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Acked-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 43 ------------------------------- 2 files changed, 44 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 257d72205bb5..b92c1937543a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -132,7 +132,6 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner, amdgpu_ring_emit_fence(ring, ring->fence_drv.gpu_addr, (*fence)->seq, AMDGPU_FENCE_FLAG_INT); - trace_amdgpu_fence_emit(ring->adev->ddev, ring->idx, (*fence)->seq); return 0; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index 76ecbaf72a2e..26e2d50a6f64 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -196,49 +196,6 @@ TRACE_EVENT(amdgpu_bo_list_set, TP_printk("list=%p, bo=%p", __entry->list, __entry->bo) ); -DECLARE_EVENT_CLASS(amdgpu_fence_request, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno), - - TP_STRUCT__entry( - __field(u32, dev) - __field(int, ring) - __field(u32, seqno) - ), - - TP_fast_assign( - __entry->dev = dev->primary->index; - __entry->ring = ring; - __entry->seqno = seqno; - ), - - TP_printk("dev=%u, ring=%d, seqno=%u", - __entry->dev, __entry->ring, __entry->seqno) -); - -DEFINE_EVENT(amdgpu_fence_request, amdgpu_fence_emit, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno) -); - -DEFINE_EVENT(amdgpu_fence_request, amdgpu_fence_wait_begin, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno) -); - -DEFINE_EVENT(amdgpu_fence_request, amdgpu_fence_wait_end, - - TP_PROTO(struct drm_device *dev, int ring, u32 seqno), - - TP_ARGS(dev, ring, seqno) -); - DECLARE_EVENT_CLASS(amdgpu_semaphore_request, TP_PROTO(int ring, struct amdgpu_semaphore *sem), -- cgit v1.2.3 From c2776afe740db5598c4c457dcacb94d4427b13f9 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 3 Nov 2015 13:27:39 +0100 Subject: drm/amdgpu: use a timer for fence fallback MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Less overhead than a work item and also adds proper cleanup handling. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Acked-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 3 +- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 78 ++++++++++++++----------------- 2 files changed, 35 insertions(+), 46 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 615ce6d464fb..3630cb545d0d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -389,7 +389,6 @@ struct amdgpu_clock { * Fences. */ struct amdgpu_fence_driver { - struct amdgpu_ring *ring; uint64_t gpu_addr; volatile uint32_t *cpu_addr; /* sync_seq is protected by ring emission lock */ @@ -398,7 +397,7 @@ struct amdgpu_fence_driver { bool initialized; struct amdgpu_irq_src *irq_src; unsigned irq_type; - struct delayed_work lockup_work; + struct timer_list fallback_timer; wait_queue_head_t fence_queue; }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index b92c1937543a..257fce356319 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -84,24 +84,6 @@ static u32 amdgpu_fence_read(struct amdgpu_ring *ring) return seq; } -/** - * amdgpu_fence_schedule_check - schedule lockup check - * - * @ring: pointer to struct amdgpu_ring - * - * Queues a delayed work item to check for lockups. - */ -static void amdgpu_fence_schedule_check(struct amdgpu_ring *ring) -{ - /* - * Do not reset the timer here with mod_delayed_work, - * this can livelock in an interaction with TTM delayed destroy. - */ - queue_delayed_work(system_power_efficient_wq, - &ring->fence_drv.lockup_work, - AMDGPU_FENCE_JIFFIES_TIMEOUT); -} - /** * amdgpu_fence_emit - emit a fence on the requested ring * @@ -135,6 +117,19 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner, return 0; } +/** + * amdgpu_fence_schedule_fallback - schedule fallback check + * + * @ring: pointer to struct amdgpu_ring + * + * Start a timer as fallback to our interrupts. + */ +static void amdgpu_fence_schedule_fallback(struct amdgpu_ring *ring) +{ + mod_timer(&ring->fence_drv.fallback_timer, + jiffies + AMDGPU_FENCE_JIFFIES_TIMEOUT); +} + /** * amdgpu_fence_activity - check for fence activity * @@ -201,45 +196,38 @@ static bool amdgpu_fence_activity(struct amdgpu_ring *ring) } while (atomic64_xchg(&ring->fence_drv.last_seq, seq) > seq); if (seq < last_emitted) - amdgpu_fence_schedule_check(ring); + amdgpu_fence_schedule_fallback(ring); return wake; } /** - * amdgpu_fence_check_lockup - check for hardware lockup + * amdgpu_fence_process - process a fence * - * @work: delayed work item + * @adev: amdgpu_device pointer + * @ring: ring index the fence is associated with * - * Checks for fence activity and if there is none probe - * the hardware if a lockup occured. + * Checks the current fence value and wakes the fence queue + * if the sequence number has increased (all asics). */ -static void amdgpu_fence_check_lockup(struct work_struct *work) +void amdgpu_fence_process(struct amdgpu_ring *ring) { - struct amdgpu_fence_driver *fence_drv; - struct amdgpu_ring *ring; - - fence_drv = container_of(work, struct amdgpu_fence_driver, - lockup_work.work); - ring = fence_drv->ring; - if (amdgpu_fence_activity(ring)) wake_up_all(&ring->fence_drv.fence_queue); } /** - * amdgpu_fence_process - process a fence + * amdgpu_fence_fallback - fallback for hardware interrupts * - * @adev: amdgpu_device pointer - * @ring: ring index the fence is associated with + * @work: delayed work item * - * Checks the current fence value and wakes the fence queue - * if the sequence number has increased (all asics). + * Checks for fence activity. */ -void amdgpu_fence_process(struct amdgpu_ring *ring) +static void amdgpu_fence_fallback(unsigned long arg) { - if (amdgpu_fence_activity(ring)) - wake_up_all(&ring->fence_drv.fence_queue); + struct amdgpu_ring *ring = (void *)arg; + + amdgpu_fence_process(ring); } /** @@ -289,7 +277,7 @@ static int amdgpu_fence_ring_wait_seq(struct amdgpu_ring *ring, uint64_t seq) if (atomic64_read(&ring->fence_drv.last_seq) >= seq) return 0; - amdgpu_fence_schedule_check(ring); + amdgpu_fence_schedule_fallback(ring); wait_event(ring->fence_drv.fence_queue, ( (signaled = amdgpu_fence_seq_signaled(ring, seq)))); @@ -490,9 +478,8 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring) atomic64_set(&ring->fence_drv.last_seq, 0); ring->fence_drv.initialized = false; - INIT_DELAYED_WORK(&ring->fence_drv.lockup_work, - amdgpu_fence_check_lockup); - ring->fence_drv.ring = ring; + setup_timer(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, + (unsigned long)ring); init_waitqueue_head(&ring->fence_drv.fence_queue); @@ -556,6 +543,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) mutex_lock(&adev->ring_lock); for (i = 0; i < AMDGPU_MAX_RINGS; i++) { struct amdgpu_ring *ring = adev->rings[i]; + if (!ring || !ring->fence_drv.initialized) continue; r = amdgpu_fence_wait_empty(ring); @@ -567,6 +555,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) amdgpu_irq_put(adev, ring->fence_drv.irq_src, ring->fence_drv.irq_type); amd_sched_fini(&ring->sched); + del_timer_sync(&ring->fence_drv.fallback_timer); ring->fence_drv.initialized = false; } mutex_unlock(&adev->ring_lock); @@ -750,7 +739,8 @@ static bool amdgpu_fence_enable_signaling(struct fence *f) fence->fence_wake.func = amdgpu_fence_check_signaled; __add_wait_queue(&ring->fence_drv.fence_queue, &fence->fence_wake); fence_get(f); - amdgpu_fence_schedule_check(ring); + if (!timer_pending(&ring->fence_drv.fallback_timer)) + amdgpu_fence_schedule_fallback(ring); FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx); return true; } -- cgit v1.2.3 From b56c22853eabde22a93d77bbd5250c1655405d74 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 29 Oct 2015 17:01:19 +0100 Subject: drm/amdgpu: use fence_is_later() for vm_flush as well v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit v2: remove superfluous check Signed-off-by: Christian König Reviewed-by: Alex Deucher (v1) Reviewed-by: Chunming Zhou --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 21 +++++++++------------ 1 file changed, 9 insertions(+), 12 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 633a32a48560..8c80381b1ea9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -207,24 +207,21 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring, uint64_t pd_addr = amdgpu_bo_gpu_offset(vm->page_directory); struct amdgpu_vm_id *vm_id = &vm->ids[ring->idx]; struct fence *flushed_updates = vm_id->flushed_updates; - bool is_earlier = false; + bool is_later; - if (flushed_updates && updates) { - BUG_ON(flushed_updates->context != updates->context); - is_earlier = (updates->seqno - flushed_updates->seqno <= - INT_MAX) ? true : false; - } - - if (pd_addr != vm_id->pd_gpu_addr || !flushed_updates || - is_earlier) { + if (!flushed_updates) + is_later = true; + else if (!updates) + is_later = false; + else + is_later = fence_is_later(updates, flushed_updates); + if (pd_addr != vm_id->pd_gpu_addr || is_later) { trace_amdgpu_vm_flush(pd_addr, ring->idx, vm_id->id); - if (is_earlier) { + if (is_later) { vm_id->flushed_updates = fence_get(updates); fence_put(flushed_updates); } - if (!flushed_updates) - vm_id->flushed_updates = fence_get(updates); vm_id->pd_gpu_addr = pd_addr; amdgpu_ring_emit_vm_flush(ring, vm_id->id, vm_id->pd_gpu_addr); } -- cgit v1.2.3 From 16ae42feb0ce6cc80473d35dd82a1cd07e2e7178 Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 3 Nov 2015 14:53:28 +0100 Subject: drm/amdgpu: use common fence for amdgpu_vm_fence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Just cleanup the function parameters. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 12 ++++++------ 3 files changed, 8 insertions(+), 8 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 3630cb545d0d..b801b6702812 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -2332,7 +2332,7 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring, struct fence *updates); void amdgpu_vm_fence(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_fence *fence); + struct fence *fence); uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr); int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, struct amdgpu_vm *vm); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index e65987743871..3902e7a9f8cc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -216,7 +216,7 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs, } if (ib->vm) - amdgpu_vm_fence(adev, ib->vm, ib->fence); + amdgpu_vm_fence(adev, ib->vm, &ib->fence->base); amdgpu_ring_unlock_commit(ring); return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 8c80381b1ea9..a9fcc995d27e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -241,16 +241,16 @@ void amdgpu_vm_flush(struct amdgpu_ring *ring, */ void amdgpu_vm_fence(struct amdgpu_device *adev, struct amdgpu_vm *vm, - struct amdgpu_fence *fence) + struct fence *fence) { - unsigned ridx = fence->ring->idx; - unsigned vm_id = vm->ids[ridx].id; + struct amdgpu_ring *ring = amdgpu_ring_from_fence(fence); + unsigned vm_id = vm->ids[ring->idx].id; fence_put(adev->vm_manager.active[vm_id]); - adev->vm_manager.active[vm_id] = fence_get(&fence->base); + adev->vm_manager.active[vm_id] = fence_get(fence); - fence_put(vm->ids[ridx].last_id_use); - vm->ids[ridx].last_id_use = fence_get(&fence->base); + fence_put(vm->ids[ring->idx].last_id_use); + vm->ids[ring->idx].last_id_use = fence_get(fence); } /** -- cgit v1.2.3 From 451f698bcac84c49dd4e25dd8e89dbd66796b4cd Mon Sep 17 00:00:00 2001 From: Flora Cui Date: Thu, 5 Nov 2015 15:24:46 +0800 Subject: drm/amdgpu: update fiji_mgcg_cgcg_init table Change-Id: If44b8057741c78208f1976f60f31b535c944d0bd Signed-off-by: Flora Cui Reviewed-by: Jammy Zhou --- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 5 ----- 1 file changed, 5 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 4410e5735154..3de6a8894e2d 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -268,7 +268,6 @@ static const u32 fiji_mgcg_cgcg_init[] = mmCGTT_CP_CLK_CTRL, 0xffffffff, 0x00000100, mmCGTT_CPC_CLK_CTRL, 0xffffffff, 0x00000100, mmCGTT_CPF_CLK_CTRL, 0xffffffff, 0x40000100, - mmCGTT_DRM_CLK_CTRL0, 0xffffffff, 0x00600100, mmCGTT_GDS_CLK_CTRL, 0xffffffff, 0x00000100, mmCGTT_IA_CLK_CTRL, 0xffffffff, 0x06000100, mmCGTT_PA_CLK_CTRL, 0xffffffff, 0x00000100, @@ -296,10 +295,6 @@ static const u32 fiji_mgcg_cgcg_init[] = mmCGTS_SM_CTRL_REG, 0xffffffff, 0x96e00200, mmCP_RB_WPTR_POLL_CNTL, 0xffffffff, 0x00900100, mmRLC_CGCG_CGLS_CTRL, 0xffffffff, 0x0020003c, - mmPCIE_INDEX, 0xffffffff, 0x0140001c, - mmPCIE_DATA, 0x000f0000, 0x00000000, - mmCGTT_DRM_CLK_CTRL0, 0xff000fff, 0x00000100, - mmHDP_XDP_CGTT_BLK_CTRL, 0xc0000fff, 0x00000104, mmCP_MEM_SLP_CNTL, 0x00000001, 0x00000001, }; -- cgit v1.2.3 From b49c84a5765cf68b8e2fdb1dc4eded9c208e6d58 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Thu, 5 Nov 2015 11:28:28 +0800 Subject: drm/amdgpu: add kmem cache for amdgpu fence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: I5ad8dd156ccf27a6f18004aa0a215a0925b6e67b Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 22 ++++++++++++++++++++-- 1 file changed, 20 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c index 257fce356319..3671f9f220bd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c @@ -47,6 +47,9 @@ * that the the relevant GPU caches have been flushed. */ +static struct kmem_cache *amdgpu_fence_slab; +static atomic_t amdgpu_fence_slab_ref = ATOMIC_INIT(0); + /** * amdgpu_fence_write - write a fence value * @@ -100,7 +103,7 @@ int amdgpu_fence_emit(struct amdgpu_ring *ring, void *owner, struct amdgpu_device *adev = ring->adev; /* we are protected by the ring emission mutex */ - *fence = kmalloc(sizeof(struct amdgpu_fence), GFP_KERNEL); + *fence = kmem_cache_alloc(amdgpu_fence_slab, GFP_KERNEL); if ((*fence) == NULL) { return -ENOMEM; } @@ -522,6 +525,13 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring) */ int amdgpu_fence_driver_init(struct amdgpu_device *adev) { + if (atomic_inc_return(&amdgpu_fence_slab_ref) == 1) { + amdgpu_fence_slab = kmem_cache_create( + "amdgpu_fence", sizeof(struct amdgpu_fence), 0, + SLAB_HWCACHE_ALIGN, NULL); + if (!amdgpu_fence_slab) + return -ENOMEM; + } if (amdgpu_debugfs_fence_init(adev)) dev_err(adev->dev, "fence debugfs file creation failed\n"); @@ -540,6 +550,8 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev) { int i, r; + if (atomic_dec_and_test(&amdgpu_fence_slab_ref)) + kmem_cache_destroy(amdgpu_fence_slab); mutex_lock(&adev->ring_lock); for (i = 0; i < AMDGPU_MAX_RINGS; i++) { struct amdgpu_ring *ring = adev->rings[i]; @@ -745,13 +757,19 @@ static bool amdgpu_fence_enable_signaling(struct fence *f) return true; } +static void amdgpu_fence_release(struct fence *f) +{ + struct amdgpu_fence *fence = to_amdgpu_fence(f); + kmem_cache_free(amdgpu_fence_slab, fence); +} + const struct fence_ops amdgpu_fence_ops = { .get_driver_name = amdgpu_fence_get_driver_name, .get_timeline_name = amdgpu_fence_get_timeline_name, .enable_signaling = amdgpu_fence_enable_signaling, .signaled = amdgpu_fence_is_signaled, .wait = fence_default_wait, - .release = NULL, + .release = amdgpu_fence_release, }; /* -- cgit v1.2.3 From f5617f9dde5ae2466560f7cb008c741e2b88adab Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Thu, 5 Nov 2015 11:41:50 +0800 Subject: drm/amd: add kmem cache for sched fence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: I45bb8ff10ef05dc3b15e31a77fbcf31117705f11 Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 12 ++++++++++++ drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 3 +++ drivers/gpu/drm/amd/scheduler/sched_fence.c | 10 ++++++++-- 3 files changed, 23 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 89619a5a4289..fe5b3c415f18 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -34,6 +34,9 @@ static struct amd_sched_job * amd_sched_entity_pop_job(struct amd_sched_entity *entity); static void amd_sched_wakeup(struct amd_gpu_scheduler *sched); +struct kmem_cache *sched_fence_slab; +atomic_t sched_fence_slab_ref = ATOMIC_INIT(0); + /* Initialize a given run queue struct */ static void amd_sched_rq_init(struct amd_sched_rq *rq) { @@ -450,6 +453,13 @@ int amd_sched_init(struct amd_gpu_scheduler *sched, init_waitqueue_head(&sched->wake_up_worker); init_waitqueue_head(&sched->job_scheduled); atomic_set(&sched->hw_rq_count, 0); + if (atomic_inc_return(&sched_fence_slab_ref) == 1) { + sched_fence_slab = kmem_cache_create( + "amd_sched_fence", sizeof(struct amd_sched_fence), 0, + SLAB_HWCACHE_ALIGN, NULL); + if (!sched_fence_slab) + return -ENOMEM; + } /* Each scheduler will run on a seperate kernel thread */ sched->thread = kthread_run(amd_sched_main, sched, sched->name); @@ -470,4 +480,6 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched) { if (sched->thread) kthread_stop(sched->thread); + if (atomic_dec_and_test(&sched_fence_slab_ref)) + kmem_cache_destroy(sched_fence_slab); } diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 929e9aced041..4d05ca6fb057 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -30,6 +30,9 @@ struct amd_gpu_scheduler; struct amd_sched_rq; +extern struct kmem_cache *sched_fence_slab; +extern atomic_t sched_fence_slab_ref; + /** * A scheduler entity is a wrapper around a job queue or a group * of other entities. Entities take turns emitting jobs from their diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c index d802638094f4..8d2130b9ff05 100644 --- a/drivers/gpu/drm/amd/scheduler/sched_fence.c +++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c @@ -32,7 +32,7 @@ struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *s_entity struct amd_sched_fence *fence = NULL; unsigned seq; - fence = kzalloc(sizeof(struct amd_sched_fence), GFP_KERNEL); + fence = kmem_cache_zalloc(sched_fence_slab, GFP_KERNEL); if (fence == NULL) return NULL; fence->owner = owner; @@ -71,11 +71,17 @@ static bool amd_sched_fence_enable_signaling(struct fence *f) return true; } +static void amd_sched_fence_release(struct fence *f) +{ + struct amd_sched_fence *fence = to_amd_sched_fence(f); + kmem_cache_free(sched_fence_slab, fence); +} + const struct fence_ops amd_sched_fence_ops = { .get_driver_name = amd_sched_fence_get_driver_name, .get_timeline_name = amd_sched_fence_get_timeline_name, .enable_signaling = amd_sched_fence_enable_signaling, .signaled = NULL, .wait = fence_default_wait, - .release = NULL, + .release = amd_sched_fence_release, }; -- cgit v1.2.3 From e4a58a28b50f30e72292b6659d94410cbf7355ad Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 5 Nov 2015 17:00:25 +0100 Subject: drm/amdgpu: fix leaking the IBs on error MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fixing a memory leak when the scheduler is enabled. Signed-off-by: Christian König Reviewed-by: Junwei Zhang Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 25 ++++++++++++------------- 1 file changed, 12 insertions(+), 13 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index dfc4d02c7a38..ecc82dfe83f8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -499,16 +499,12 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) for (i = 0; i < parser->nchunks; i++) drm_free_large(parser->chunks[i].kdata); kfree(parser->chunks); - if (!amdgpu_enable_scheduler) - { - if (parser->ibs) - for (i = 0; i < parser->num_ibs; i++) - amdgpu_ib_free(parser->adev, &parser->ibs[i]); - kfree(parser->ibs); - if (parser->uf.bo) - drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); - } - + if (parser->ibs) + for (i = 0; i < parser->num_ibs; i++) + amdgpu_ib_free(parser->adev, &parser->ibs[i]); + kfree(parser->ibs); + if (parser->uf.bo) + drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); kfree(parser); } @@ -888,11 +884,14 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) job->base.owner = parser->filp; mutex_init(&job->job_lock); if (job->ibs[job->num_ibs - 1].user) { - memcpy(&job->uf, &parser->uf, - sizeof(struct amdgpu_user_fence)); + job->uf = parser->uf; job->ibs[job->num_ibs - 1].user = &job->uf; + parser->uf.bo = NULL; } + parser->ibs = NULL; + parser->num_ibs = 0; + job->free_job = amdgpu_cs_free_job; mutex_lock(&job->job_lock); r = amd_sched_entity_push_job(&job->base); @@ -905,7 +904,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) cs->out.handle = amdgpu_ctx_add_fence(parser->ctx, ring, &job->base.s_fence->base); - parser->ibs[parser->num_ibs - 1].sequence = cs->out.handle; + job->ibs[job->num_ibs - 1].sequence = cs->out.handle; list_sort(NULL, &parser->validated, cmp_size_smaller_first); ttm_eu_fence_buffer_objects(&parser->ticket, -- cgit v1.2.3 From 7e52a81c2f0326a85d3ebc005829bcd604731c6d Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 4 Nov 2015 15:44:39 +0100 Subject: drm/amdgpu: cleanup amdgpu_cs_parser handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit No need any more to allocate that structure dynamically, just put it on the stack. This is a start to cleanup some of the scheduler fallouts. Signed-off-by: Christian König Reviewed-by: Junwei Zhang Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 -- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 88 +++++++++++++--------------------- 2 files changed, 33 insertions(+), 60 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index b801b6702812..7b02e3455172 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -2256,11 +2256,6 @@ void amdgpu_pci_config_reset(struct amdgpu_device *adev); bool amdgpu_card_posted(struct amdgpu_device *adev); void amdgpu_update_display_priority(struct amdgpu_device *adev); bool amdgpu_boot_test_post_card(struct amdgpu_device *adev); -struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev, - struct drm_file *filp, - struct amdgpu_ctx *ctx, - struct amdgpu_ib *ibs, - uint32_t num_ibs); int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data); int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index ecc82dfe83f8..bf32096b8eb7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -127,30 +127,6 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type, return 0; } -struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev, - struct drm_file *filp, - struct amdgpu_ctx *ctx, - struct amdgpu_ib *ibs, - uint32_t num_ibs) -{ - struct amdgpu_cs_parser *parser; - int i; - - parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL); - if (!parser) - return NULL; - - parser->adev = adev; - parser->filp = filp; - parser->ctx = ctx; - parser->ibs = ibs; - parser->num_ibs = num_ibs; - for (i = 0; i < num_ibs; i++) - ibs[i].ctx = ctx; - - return parser; -} - int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) { union drm_amdgpu_cs *cs = data; @@ -490,6 +466,7 @@ static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int err static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) { unsigned i; + if (parser->ctx) amdgpu_ctx_put(parser->ctx); if (parser->bo_list) @@ -505,7 +482,6 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) kfree(parser->ibs); if (parser->uf.bo) drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); - kfree(parser); } /** @@ -824,36 +800,36 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) union drm_amdgpu_cs *cs = data; struct amdgpu_fpriv *fpriv = filp->driver_priv; struct amdgpu_vm *vm = &fpriv->vm; - struct amdgpu_cs_parser *parser; + struct amdgpu_cs_parser parser = {}; bool reserved_buffers = false; int i, r; if (!adev->accel_working) return -EBUSY; - parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0); - if (!parser) - return -ENOMEM; - r = amdgpu_cs_parser_init(parser, data); + parser.adev = adev; + parser.filp = filp; + + r = amdgpu_cs_parser_init(&parser, data); if (r) { DRM_ERROR("Failed to initialize parser !\n"); - amdgpu_cs_parser_fini(parser, r, false); + amdgpu_cs_parser_fini(&parser, r, false); r = amdgpu_cs_handle_lockup(adev, r); return r; } mutex_lock(&vm->mutex); - r = amdgpu_cs_parser_relocs(parser); + r = amdgpu_cs_parser_relocs(&parser); if (r == -ENOMEM) DRM_ERROR("Not enough memory for command submission!\n"); else if (r && r != -ERESTARTSYS) DRM_ERROR("Failed to process the buffer list %d!\n", r); else if (!r) { reserved_buffers = true; - r = amdgpu_cs_ib_fill(adev, parser); + r = amdgpu_cs_ib_fill(adev, &parser); } if (!r) { - r = amdgpu_cs_dependencies(adev, parser); + r = amdgpu_cs_dependencies(adev, &parser); if (r) DRM_ERROR("Failed in the dependencies handling %d!\n", r); } @@ -861,36 +837,38 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) if (r) goto out; - for (i = 0; i < parser->num_ibs; i++) - trace_amdgpu_cs(parser, i); + for (i = 0; i < parser.num_ibs; i++) + trace_amdgpu_cs(&parser, i); - r = amdgpu_cs_ib_vm_chunk(adev, parser); + r = amdgpu_cs_ib_vm_chunk(adev, &parser); if (r) goto out; - if (amdgpu_enable_scheduler && parser->num_ibs) { + if (amdgpu_enable_scheduler && parser.num_ibs) { struct amdgpu_job *job; - struct amdgpu_ring * ring = parser->ibs->ring; + struct amdgpu_ring * ring = parser.ibs->ring; + job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL); if (!job) { r = -ENOMEM; goto out; } + job->base.sched = &ring->sched; - job->base.s_entity = &parser->ctx->rings[ring->idx].entity; - job->adev = parser->adev; - job->ibs = parser->ibs; - job->num_ibs = parser->num_ibs; - job->base.owner = parser->filp; + job->base.s_entity = &parser.ctx->rings[ring->idx].entity; + job->adev = parser.adev; + job->ibs = parser.ibs; + job->num_ibs = parser.num_ibs; + job->base.owner = parser.filp; mutex_init(&job->job_lock); if (job->ibs[job->num_ibs - 1].user) { - job->uf = parser->uf; + job->uf = parser.uf; job->ibs[job->num_ibs - 1].user = &job->uf; - parser->uf.bo = NULL; + parser.uf.bo = NULL; } - parser->ibs = NULL; - parser->num_ibs = 0; + parser.ibs = NULL; + parser.num_ibs = 0; job->free_job = amdgpu_cs_free_job; mutex_lock(&job->job_lock); @@ -902,24 +880,24 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) goto out; } cs->out.handle = - amdgpu_ctx_add_fence(parser->ctx, ring, + amdgpu_ctx_add_fence(parser.ctx, ring, &job->base.s_fence->base); job->ibs[job->num_ibs - 1].sequence = cs->out.handle; - list_sort(NULL, &parser->validated, cmp_size_smaller_first); - ttm_eu_fence_buffer_objects(&parser->ticket, - &parser->validated, + list_sort(NULL, &parser.validated, cmp_size_smaller_first); + ttm_eu_fence_buffer_objects(&parser.ticket, + &parser.validated, &job->base.s_fence->base); mutex_unlock(&job->job_lock); - amdgpu_cs_parser_fini_late(parser); + amdgpu_cs_parser_fini_late(&parser); mutex_unlock(&vm->mutex); return 0; } - cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence; + cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence; out: - amdgpu_cs_parser_fini(parser, r, reserved_buffers); + amdgpu_cs_parser_fini(&parser, r, reserved_buffers); mutex_unlock(&vm->mutex); r = amdgpu_cs_handle_lockup(adev, r); return r; -- cgit v1.2.3 From 24dd2f64c5a877392925202321c7c2c46c2b0ddf Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Tue, 10 Nov 2015 13:01:35 -0500 Subject: drm/radeon: unconditionally set sysfs_initialized Avoids spew on resume for systems where sysfs may fail even on init. bug: https://bugzilla.kernel.org/show_bug.cgi?id=106851 Signed-off-by: Alex Deucher Cc: stable@vger.kernel.org --- drivers/gpu/drm/radeon/radeon_pm.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c index 6d80dde23400..f4f03dcc1530 100644 --- a/drivers/gpu/drm/radeon/radeon_pm.c +++ b/drivers/gpu/drm/radeon/radeon_pm.c @@ -1542,8 +1542,7 @@ int radeon_pm_late_init(struct radeon_device *rdev) ret = device_create_file(rdev->dev, &dev_attr_power_method); if (ret) DRM_ERROR("failed to create device file for power method\n"); - if (!ret) - rdev->pm.sysfs_initialized = true; + rdev->pm.sysfs_initialized = true; } mutex_lock(&rdev->pm.mutex); -- cgit v1.2.3 From a80b30476d240482d360a25a1b2e8c13036ed750 Mon Sep 17 00:00:00 2001 From: Jay Cornwall Date: Thu, 5 Nov 2015 13:06:15 -0600 Subject: drm/amdgpu: Fix default page access routing MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The VM default page (used when a VM translation fails) is allocated in system memory. The VM is misconfigured to interpret the physical address as referencing a VRAM physical page. Route default page accesses to system memory. Reviewed-by: Christian König Signed-off-by: Jay Cornwall Cc: # v4.2+ Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 1 + drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 1 + 2 files changed, 2 insertions(+) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index 85bbcdc73fff..b93b649d9e36 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -501,6 +501,7 @@ static int gmc_v7_0_gart_enable(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1); + tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1); WREG32(mmVM_L2_CNTL, tmp); tmp = REG_SET_FIELD(0, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_L2_CACHE, 1); diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index 1bcc4e74e3b4..c9209b45d901 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -629,6 +629,7 @@ static int gmc_v8_0_gart_enable(struct amdgpu_device *adev) tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_L2_PDE0_CACHE_LRU_UPDATE_BY_WRITE, 1); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, EFFECTIVE_L2_QUEUE_SIZE, 7); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, CONTEXT1_IDENTITY_ACCESS_MODE, 1); + tmp = REG_SET_FIELD(tmp, VM_L2_CNTL, ENABLE_DEFAULT_PAGE_OUT_TO_SYSTEM_MEMORY, 1); WREG32(mmVM_L2_CNTL, tmp); tmp = RREG32(mmVM_L2_CNTL2); tmp = REG_SET_FIELD(tmp, VM_L2_CNTL2, INVALIDATE_ALL_L1_TLBS, 1); -- cgit v1.2.3 From 515c752dabee9945c1e8686c87f7cdeb3935eea4 Mon Sep 17 00:00:00 2001 From: Maxim Sheviakov Date: Tue, 10 Nov 2015 13:09:13 -0500 Subject: drm/radeon: fix quirk for MSI R7 370 Armor 2X There was a typo in the original. bug: https://bugs.freedesktop.org/show_bug.cgi?id=92865 Signed-off-by: Maxim Sheviakov Signed-off-by: Alex Deucher --- drivers/gpu/drm/radeon/si_dpm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/radeon/si_dpm.c b/drivers/gpu/drm/radeon/si_dpm.c index e72bf46042e0..a82b891ae1fe 100644 --- a/drivers/gpu/drm/radeon/si_dpm.c +++ b/drivers/gpu/drm/radeon/si_dpm.c @@ -2927,7 +2927,7 @@ static struct si_dpm_quirk si_dpm_quirk_list[] = { { PCI_VENDOR_ID_ATI, 0x6810, 0x1462, 0x3036, 0, 120000 }, { PCI_VENDOR_ID_ATI, 0x6811, 0x174b, 0xe271, 0, 120000 }, { PCI_VENDOR_ID_ATI, 0x6810, 0x174b, 0xe271, 85000, 90000 }, - { PCI_VENDOR_ID_ATI, 0x6811, 0x1762, 0x2015, 0, 120000 }, + { PCI_VENDOR_ID_ATI, 0x6811, 0x1462, 0x2015, 0, 120000 }, { PCI_VENDOR_ID_ATI, 0x6811, 0x1043, 0x2015, 0, 120000 }, { 0, 0, 0, 0 }, }; -- cgit v1.2.3 From e1b35f6103b37e0d81184b32906b7010170dda02 Mon Sep 17 00:00:00 2001 From: Arnd Bergmann Date: Tue, 10 Nov 2015 13:17:55 +0100 Subject: drm/amdgpu: fix seq_printf format string The amdgpu driver has a debugfs interface that shows the amount of VRAM in use, but the newly added code causes a build error on all 32-bit architectures: drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c:1076:17: warning: format '%lu' expects argument of type 'long unsigned int', but argument 4 has type 'long long int' [-Wformat=] This fixes the format string to use "%llu" for printing 64-bit numbers, which works everywhere, as long as we also cast to 'u64'. Unlike atomic64_t, u64 is defined as 'unsigned long long' on all architectures. Signed-off-by: Arnd Bergmann Fixes: a2ef8a974931 ("drm/amdgpu: add vram usage into debugfs") Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 81bb8e9fc26d..d4bac5f49939 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1073,10 +1073,10 @@ static int amdgpu_mm_dump_table(struct seq_file *m, void *data) ret = drm_mm_dump_table(m, mm); spin_unlock(&glob->lru_lock); if (ttm_pl == TTM_PL_VRAM) - seq_printf(m, "man size:%llu pages, ram usage:%luMB, vis usage:%luMB\n", + seq_printf(m, "man size:%llu pages, ram usage:%lluMB, vis usage:%lluMB\n", adev->mman.bdev.man[ttm_pl].size, - atomic64_read(&adev->vram_usage) >> 20, - atomic64_read(&adev->vram_vis_usage) >> 20); + (u64)atomic64_read(&adev->vram_usage) >> 20, + (u64)atomic64_read(&adev->vram_vis_usage) >> 20); return ret; } -- cgit v1.2.3 From c305fd5fffae5b826b9f8f7ef32fe72ee83354fc Mon Sep 17 00:00:00 2001 From: Rex Zhu Date: Tue, 13 Oct 2015 13:57:52 +0800 Subject: drm/amdgpu: fix bug that can't enter thermal interrupt for bonaire. Set reversed bit to enable/disable thermal interrupt. Signed-off-by: Rex Zhu Reviewed-by: Jammy Zhou --- drivers/gpu/drm/amd/amdgpu/ci_dpm.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c index a1a35a5df8e7..57a2e347f04d 100644 --- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c @@ -6569,12 +6569,12 @@ static int ci_dpm_set_interrupt_state(struct amdgpu_device *adev, switch (state) { case AMDGPU_IRQ_STATE_DISABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; + cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; case AMDGPU_IRQ_STATE_ENABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; + cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTH_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; default: @@ -6586,12 +6586,12 @@ static int ci_dpm_set_interrupt_state(struct amdgpu_device *adev, switch (state) { case AMDGPU_IRQ_STATE_DISABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; + cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; case AMDGPU_IRQ_STATE_ENABLE: cg_thermal_int = RREG32_SMC(ixCG_THERMAL_INT); - cg_thermal_int |= CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; + cg_thermal_int &= ~CG_THERMAL_INT_CTRL__THERM_INTL_MASK_MASK; WREG32_SMC(ixCG_THERMAL_INT, cg_thermal_int); break; default: -- cgit v1.2.3 From 5f2e816b2957fdde3c7d51d55db6751a980f4bea Mon Sep 17 00:00:00 2001 From: Flora Cui Date: Thu, 5 Nov 2015 09:50:21 +0800 Subject: drm/amdgpu: update Fiji's tiling mode table Change-Id: I925c15015390113f7e27746ec5751eaa6a92c2a7 Signed-off-by: Flora Cui Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 292 +++++++++++++++++++++++++++++++++- 1 file changed, 291 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 3de6a8894e2d..e1dcab98e249 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -995,7 +995,7 @@ static void gfx_v8_0_gpu_early_init(struct amdgpu_device *adev) adev->gfx.config.max_cu_per_sh = 16; adev->gfx.config.max_sh_per_se = 1; adev->gfx.config.max_backends_per_se = 4; - adev->gfx.config.max_texture_channel_caches = 8; + adev->gfx.config.max_texture_channel_caches = 16; adev->gfx.config.max_gprs = 256; adev->gfx.config.max_gs_threads = 32; adev->gfx.config.max_hw_contexts = 8; @@ -1608,6 +1608,296 @@ static void gfx_v8_0_tiling_mode_table_init(struct amdgpu_device *adev) WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); } case CHIP_FIJI: + for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { + switch (reg_offset) { + case 0: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_64B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 1: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_128B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 2: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_256B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 3: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_512B) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 4: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 5: + gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 6: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 7: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + TILE_SPLIT(ADDR_SURF_TILE_SPLIT_2KB) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DEPTH_MICRO_TILING)); + break; + case 8: + gb_tile_moden = (ARRAY_MODE(ARRAY_LINEAR_ALIGNED) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16)); + break; + case 9: + gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 10: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 11: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + break; + case 12: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_DISPLAY_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + break; + case 13: + gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 14: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 15: + gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 16: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + break; + case 17: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + break; + case 18: + gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 19: + gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 20: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 21: + gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 22: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 23: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 24: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THIN_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 25: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 26: + gb_tile_moden = (ARRAY_MODE(ARRAY_3D_TILED_XTHICK) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_THICK_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_1)); + break; + case 27: + gb_tile_moden = (ARRAY_MODE(ARRAY_1D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 28: + gb_tile_moden = (ARRAY_MODE(ARRAY_2D_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_2)); + break; + case 29: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P16_32x32_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + break; + case 30: + gb_tile_moden = (ARRAY_MODE(ARRAY_PRT_TILED_THIN1) | + PIPE_CONFIG(ADDR_SURF_P4_16x16) | + MICRO_TILE_MODE_NEW(ADDR_SURF_ROTATED_MICRO_TILING) | + SAMPLE_SPLIT(ADDR_SURF_SAMPLE_SPLIT_8)); + break; + default: + gb_tile_moden = 0; + break; + } + adev->gfx.config.tile_mode_array[reg_offset] = gb_tile_moden; + WREG32(mmGB_TILE_MODE0 + reg_offset, gb_tile_moden); + } + for (reg_offset = 0; reg_offset < num_secondary_tile_mode_states; reg_offset++) { + switch (reg_offset) { + case 0: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 1: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 2: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 3: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 4: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 5: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 6: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 8: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_8) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 9: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_4) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 10: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 11: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 12: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_2) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 13: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_2) | + NUM_BANKS(ADDR_SURF_8_BANK)); + break; + case 14: + gb_tile_moden = (BANK_WIDTH(ADDR_SURF_BANK_WIDTH_1) | + BANK_HEIGHT(ADDR_SURF_BANK_HEIGHT_1) | + MACRO_TILE_ASPECT(ADDR_SURF_MACRO_ASPECT_1) | + NUM_BANKS(ADDR_SURF_4_BANK)); + break; + case 7: + /* unused idx */ + continue; + default: + gb_tile_moden = 0; + break; + } + adev->gfx.config.macrotile_mode_array[reg_offset] = gb_tile_moden; + WREG32(mmGB_MACROTILE_MODE0 + reg_offset, gb_tile_moden); + } + break; case CHIP_TONGA: for (reg_offset = 0; reg_offset < num_tile_mode_states; reg_offset++) { switch (reg_offset) { -- cgit v1.2.3 From 7034decf6a5b1ff778d83ff9d7ce1f0b404804e4 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Wed, 11 Nov 2015 14:56:00 +0800 Subject: drm/amdgpu: add command submission workflow tracepoint OGL needs these tracepoints to investigate performance issue. Change-Id: I5e58187d061253f7d665dfce8e4e163ba91d3e2b Signed-off-by: Chunming Zhou --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 51 +++++++++++++++++++++++++ drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h | 24 ++++++++++-- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 + 5 files changed, 76 insertions(+), 4 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index bf32096b8eb7..2ae73d5232dd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -888,7 +888,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated, &job->base.s_fence->base); - + trace_amdgpu_cs_ioctl(job); mutex_unlock(&job->job_lock); amdgpu_cs_parser_fini_late(&parser); mutex_unlock(&vm->mutex); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index dcf4a8aca680..67f778f6eedb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -26,6 +26,7 @@ #include #include #include "amdgpu.h" +#include "amdgpu_trace.h" static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job) { @@ -45,6 +46,7 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job) } job = to_amdgpu_job(sched_job); mutex_lock(&job->job_lock); + trace_amdgpu_sched_run_job(job); r = amdgpu_ib_schedule(job->adev, job->num_ibs, job->ibs, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index 26e2d50a6f64..8f9834ab1bd5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -48,6 +48,57 @@ TRACE_EVENT(amdgpu_cs, __entry->fences) ); +TRACE_EVENT(amdgpu_cs_ioctl, + TP_PROTO(struct amdgpu_job *job), + TP_ARGS(job), + TP_STRUCT__entry( + __field(struct amdgpu_device *, adev) + __field(struct amd_sched_job *, sched_job) + __field(struct amdgpu_ib *, ib) + __field(struct fence *, fence) + __field(char *, ring_name) + __field(u32, num_ibs) + ), + + TP_fast_assign( + __entry->adev = job->adev; + __entry->sched_job = &job->base; + __entry->ib = job->ibs; + __entry->fence = &job->base.s_fence->base; + __entry->ring_name = job->ibs[0].ring->name; + __entry->num_ibs = job->num_ibs; + ), + TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u", + __entry->adev, __entry->sched_job, __entry->ib, + __entry->fence, __entry->ring_name, __entry->num_ibs) +); + +TRACE_EVENT(amdgpu_sched_run_job, + TP_PROTO(struct amdgpu_job *job), + TP_ARGS(job), + TP_STRUCT__entry( + __field(struct amdgpu_device *, adev) + __field(struct amd_sched_job *, sched_job) + __field(struct amdgpu_ib *, ib) + __field(struct fence *, fence) + __field(char *, ring_name) + __field(u32, num_ibs) + ), + + TP_fast_assign( + __entry->adev = job->adev; + __entry->sched_job = &job->base; + __entry->ib = job->ibs; + __entry->fence = &job->base.s_fence->base; + __entry->ring_name = job->ibs[0].ring->name; + __entry->num_ibs = job->num_ibs; + ), + TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u", + __entry->adev, __entry->sched_job, __entry->ib, + __entry->fence, __entry->ring_name, __entry->num_ibs) +); + + TRACE_EVENT(amdgpu_vm_grab_id, TP_PROTO(unsigned vmid, int ring), TP_ARGS(vmid, ring), diff --git a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h index 144f50acc971..c89dc777768f 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h @@ -16,6 +16,8 @@ TRACE_EVENT(amd_sched_job, TP_ARGS(sched_job), TP_STRUCT__entry( __field(struct amd_sched_entity *, entity) + __field(struct amd_sched_job *, sched_job) + __field(struct fence *, fence) __field(const char *, name) __field(u32, job_count) __field(int, hw_job_count) @@ -23,16 +25,32 @@ TRACE_EVENT(amd_sched_job, TP_fast_assign( __entry->entity = sched_job->s_entity; + __entry->sched_job = sched_job; + __entry->fence = &sched_job->s_fence->base; __entry->name = sched_job->sched->name; __entry->job_count = kfifo_len( &sched_job->s_entity->job_queue) / sizeof(sched_job); __entry->hw_job_count = atomic_read( &sched_job->sched->hw_rq_count); ), - TP_printk("entity=%p, ring=%s, job count:%u, hw job count:%d", - __entry->entity, __entry->name, __entry->job_count, - __entry->hw_job_count) + TP_printk("entity=%p, sched job=%p, fence=%p, ring=%s, job count:%u, hw job count:%d", + __entry->entity, __entry->sched_job, __entry->fence, __entry->name, + __entry->job_count, __entry->hw_job_count) ); + +TRACE_EVENT(amd_sched_process_job, + TP_PROTO(struct amd_sched_fence *fence), + TP_ARGS(fence), + TP_STRUCT__entry( + __field(struct fence *, fence) + ), + + TP_fast_assign( + __entry->fence = &fence->base; + ), + TP_printk("fence=%p signaled", __entry->fence) +); + #endif /* This part must be outside protection */ diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index fe5b3c415f18..b8925fea577c 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -346,6 +346,7 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) list_del_init(&s_fence->list); spin_unlock_irqrestore(&sched->fence_list_lock, flags); } + trace_amd_sched_process_job(s_fence); fence_put(&s_fence->base); wake_up_interruptible(&sched->wake_up_worker); } -- cgit v1.2.3 From 4a562283376197722b295d27633134401bbc80f5 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 6 Nov 2015 14:09:21 +0100 Subject: drm/amdgpu: cleanup scheduler fence get/put dance MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The code was correct, but getting two references when the ownership is linearly moved on is a bit awkward and just overhead. Signed: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 1 - drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 - 2 files changed, 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 67f778f6eedb..8ef9e4415fcc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -64,7 +64,6 @@ err: job->free_job(job); mutex_unlock(&job->job_lock); - fence_put(&job->base.s_fence->base); kfree(job); return fence ? &fence->base : NULL; } diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index b8925fea577c..ccb7c1554f5e 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -285,7 +285,6 @@ int amd_sched_entity_push_job(struct amd_sched_job *sched_job) if (!fence) return -ENOMEM; - fence_get(&fence->base); sched_job->s_fence = fence; wait_event(entity->sched->job_scheduled, -- cgit v1.2.3 From e284022163716ecf11c37fd1057c35d689ef2c11 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 5 Nov 2015 19:49:48 +0100 Subject: drm/amdgpu: fix incorrect mutex usage v3 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Before this patch the scheduler fence was created when we push the job into the queue, so we could only get the fence after pushing it. The mutex now was necessary to prevent the thread pushing the jobs to the hardware from running faster than the thread pushing the jobs into the queue. Otherwise the thread pushing jobs into the queue would have accessed possible freed up memory when it tries to get a reference to the fence. So what you get in the end is thread A: mutex_lock(&job->lock); ... Kick of thread B. ... mutex_unlock(&job->lock); And thread B: mutex_lock(&job->lock); .... mutex_unlock(&job->lock); kfree(job); I'm actually not sure if I'm still up to date on this, but this usage pattern used to be not allowed with mutexes. See here as well https://lwn.net/Articles/575460/. v2: remove unrelated changes, fix missing owner v3: rebased, add more commit message Signed-off-by: Christian König Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 43 +++++++++++++++------------ drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 27 +++++++---------- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 10 +------ drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 3 +- 5 files changed, 37 insertions(+), 48 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 7b02e3455172..0f187027c753 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1225,7 +1225,7 @@ struct amdgpu_job { struct amdgpu_device *adev; struct amdgpu_ib *ibs; uint32_t num_ibs; - struct mutex job_lock; + void *owner; struct amdgpu_user_fence uf; int (*free_job)(struct amdgpu_job *job); }; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 2ae73d5232dd..44cf977ae4f6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -845,8 +845,9 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) goto out; if (amdgpu_enable_scheduler && parser.num_ibs) { - struct amdgpu_job *job; struct amdgpu_ring * ring = parser.ibs->ring; + struct amd_sched_fence *fence; + struct amdgpu_job *job; job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL); if (!job) { @@ -859,37 +860,41 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) job->adev = parser.adev; job->ibs = parser.ibs; job->num_ibs = parser.num_ibs; - job->base.owner = parser.filp; - mutex_init(&job->job_lock); + job->owner = parser.filp; + job->free_job = amdgpu_cs_free_job; + if (job->ibs[job->num_ibs - 1].user) { job->uf = parser.uf; job->ibs[job->num_ibs - 1].user = &job->uf; parser.uf.bo = NULL; } - parser.ibs = NULL; - parser.num_ibs = 0; - - job->free_job = amdgpu_cs_free_job; - mutex_lock(&job->job_lock); - r = amd_sched_entity_push_job(&job->base); - if (r) { - mutex_unlock(&job->job_lock); + fence = amd_sched_fence_create(job->base.s_entity, + parser.filp); + if (!fence) { + r = -ENOMEM; amdgpu_cs_free_job(job); kfree(job); goto out; } - cs->out.handle = - amdgpu_ctx_add_fence(parser.ctx, ring, - &job->base.s_fence->base); + job->base.s_fence = fence; + fence_get(&fence->base); + + cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring, + &fence->base); job->ibs[job->num_ibs - 1].sequence = cs->out.handle; - list_sort(NULL, &parser.validated, cmp_size_smaller_first); - ttm_eu_fence_buffer_objects(&parser.ticket, - &parser.validated, - &job->base.s_fence->base); + parser.ibs = NULL; + parser.num_ibs = 0; + trace_amdgpu_cs_ioctl(job); - mutex_unlock(&job->job_lock); + amd_sched_entity_push_job(&job->base); + + list_sort(NULL, &parser.validated, cmp_size_smaller_first); + ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated, + &fence->base); + fence_put(&fence->base); + amdgpu_cs_parser_fini_late(&parser); mutex_unlock(&vm->mutex); return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 8ef9e4415fcc..438c05254695 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -45,12 +45,8 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job) return NULL; } job = to_amdgpu_job(sched_job); - mutex_lock(&job->job_lock); trace_amdgpu_sched_run_job(job); - r = amdgpu_ib_schedule(job->adev, - job->num_ibs, - job->ibs, - job->base.owner); + r = amdgpu_ib_schedule(job->adev, job->num_ibs, job->ibs, job->owner); if (r) { DRM_ERROR("Error scheduling IBs (%d)\n", r); goto err; @@ -63,7 +59,6 @@ err: if (job->free_job) job->free_job(job); - mutex_unlock(&job->job_lock); kfree(job); return fence ? &fence->base : NULL; } @@ -89,21 +84,19 @@ int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev, return -ENOMEM; job->base.sched = &ring->sched; job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity; + job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner); + if (!job->base.s_fence) { + kfree(job); + return -ENOMEM; + } + *f = fence_get(&job->base.s_fence->base); + job->adev = adev; job->ibs = ibs; job->num_ibs = num_ibs; - job->base.owner = owner; - mutex_init(&job->job_lock); + job->owner = owner; job->free_job = free_job; - mutex_lock(&job->job_lock); - r = amd_sched_entity_push_job(&job->base); - if (r) { - mutex_unlock(&job->job_lock); - kfree(job); - return r; - } - *f = fence_get(&job->base.s_fence->base); - mutex_unlock(&job->job_lock); + amd_sched_entity_push_job(&job->base); } else { r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner); if (r) diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index ccb7c1554f5e..ea30d6ad4c13 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -276,21 +276,13 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job) * * Returns 0 for success, negative error code otherwise. */ -int amd_sched_entity_push_job(struct amd_sched_job *sched_job) +void amd_sched_entity_push_job(struct amd_sched_job *sched_job) { struct amd_sched_entity *entity = sched_job->s_entity; - struct amd_sched_fence *fence = amd_sched_fence_create( - entity, sched_job->owner); - - if (!fence) - return -ENOMEM; - - sched_job->s_fence = fence; wait_event(entity->sched->job_scheduled, amd_sched_entity_in(sched_job)); trace_amd_sched_job(sched_job); - return 0; } /** diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 4d05ca6fb057..939692b14f4b 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -79,7 +79,6 @@ struct amd_sched_job { struct amd_gpu_scheduler *sched; struct amd_sched_entity *s_entity; struct amd_sched_fence *s_fence; - void *owner; }; extern const struct fence_ops amd_sched_fence_ops; @@ -131,7 +130,7 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched, uint32_t jobs); void amd_sched_entity_fini(struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity); -int amd_sched_entity_push_job(struct amd_sched_job *sched_job); +void amd_sched_entity_push_job(struct amd_sched_job *sched_job); struct amd_sched_fence *amd_sched_fence_create( struct amd_sched_entity *s_entity, void *owner); -- cgit v1.2.3 From 5d82730af746abca2aa74e00de6370d338df7e95 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 13 Nov 2015 13:04:50 +0100 Subject: drm/amdgpu: fix handling order in scheduler CS MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We need to clear parser.ibs and num_ibs before amd_sched_fence_create, otherwise the IB could be freed twice if fence creates fails. Signed-off-by: Christian König Reviewed-by: Chunming Zhou --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 +++++----- 1 file changed, 5 insertions(+), 5 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 44cf977ae4f6..6096effd6a56 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -858,11 +858,14 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) job->base.sched = &ring->sched; job->base.s_entity = &parser.ctx->rings[ring->idx].entity; job->adev = parser.adev; - job->ibs = parser.ibs; - job->num_ibs = parser.num_ibs; job->owner = parser.filp; job->free_job = amdgpu_cs_free_job; + job->ibs = parser.ibs; + job->num_ibs = parser.num_ibs; + parser.ibs = NULL; + parser.num_ibs = 0; + if (job->ibs[job->num_ibs - 1].user) { job->uf = parser.uf; job->ibs[job->num_ibs - 1].user = &job->uf; @@ -884,9 +887,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) &fence->base); job->ibs[job->num_ibs - 1].sequence = cs->out.handle; - parser.ibs = NULL; - parser.num_ibs = 0; - trace_amdgpu_cs_ioctl(job); amd_sched_entity_push_job(&job->base); -- cgit v1.2.3 From 43c27fb5c21eb238ce5ffc7766bc59ab773bfd7c Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Thu, 12 Nov 2015 15:33:09 +0800 Subject: drm/amdgpu: update pd while updating vm as well MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: I93a861cd6707f7d91672b9e19757cc50008cd7a2 Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 087332858853..16dca46314bd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -483,6 +483,9 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev, if (domain == AMDGPU_GEM_DOMAIN_CPU) goto error_unreserve; } + r = amdgpu_vm_update_page_directory(adev, bo_va->vm); + if (r) + goto error_unreserve; r = amdgpu_vm_clear_freed(adev, bo_va->vm); if (r) -- cgit v1.2.3 From 680513cc0accf14cb447663b65e02ae307ac3811 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 10 Sep 2015 15:03:50 +0200 Subject: drm/amdgpu: wait interruptible when semaphores are disabled v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Otherwise debugging locked up processes isn't possible. v2: rebased Signed-off-by: Christian König Reviewed-by: Alex Deucher (v1) --- drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index a6697fd05217..dd005c336c97 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -302,8 +302,14 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync, return -EINVAL; } - if (amdgpu_enable_scheduler || !amdgpu_enable_semaphores || - (count >= AMDGPU_NUM_SYNCS)) { + if (amdgpu_enable_scheduler || !amdgpu_enable_semaphores) { + r = fence_wait(&fence->base, true); + if (r) + return r; + continue; + } + + if (count >= AMDGPU_NUM_SYNCS) { /* not enough room, wait manually */ r = fence_wait(&fence->base, false); if (r) -- cgit v1.2.3 From bbf0b34578a71f42d19bd6c966f09221044921fe Mon Sep 17 00:00:00 2001 From: Junwei Zhang Date: Sun, 6 Sep 2015 14:00:46 +0800 Subject: drm/amdgpu: remove the unnecessary parameter adev for amdgpu_sa_bo_new() MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Junwei Zhang Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 7 +++---- drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c | 3 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c | 2 +- 4 files changed, 6 insertions(+), 8 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c index 3902e7a9f8cc..9e25edafa721 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c @@ -62,7 +62,7 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm, int r; if (size) { - r = amdgpu_sa_bo_new(adev, &adev->ring_tmp_bo, + r = amdgpu_sa_bo_new(&adev->ring_tmp_bo, &ib->sa_bo, size, 256); if (r) { dev_err(adev->dev, "failed to get a new IB (%d)\n", r); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h index 3c2ff4567798..ea756e77b023 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h @@ -189,10 +189,9 @@ int amdgpu_sa_bo_manager_start(struct amdgpu_device *adev, struct amdgpu_sa_manager *sa_manager); int amdgpu_sa_bo_manager_suspend(struct amdgpu_device *adev, struct amdgpu_sa_manager *sa_manager); -int amdgpu_sa_bo_new(struct amdgpu_device *adev, - struct amdgpu_sa_manager *sa_manager, - struct amdgpu_sa_bo **sa_bo, - unsigned size, unsigned align); +int amdgpu_sa_bo_new(struct amdgpu_sa_manager *sa_manager, + struct amdgpu_sa_bo **sa_bo, + unsigned size, unsigned align); void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo, struct fence *fence); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c index 0212b31dc194..8b88edb0434b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sa.c @@ -311,8 +311,7 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager, return false; } -int amdgpu_sa_bo_new(struct amdgpu_device *adev, - struct amdgpu_sa_manager *sa_manager, +int amdgpu_sa_bo_new(struct amdgpu_sa_manager *sa_manager, struct amdgpu_sa_bo **sa_bo, unsigned size, unsigned align) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c index ff3ca52ec6fe..1caaf201b708 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_semaphore.c @@ -40,7 +40,7 @@ int amdgpu_semaphore_create(struct amdgpu_device *adev, if (*semaphore == NULL) { return -ENOMEM; } - r = amdgpu_sa_bo_new(adev, &adev->ring_tmp_bo, + r = amdgpu_sa_bo_new(&adev->ring_tmp_bo, &(*semaphore)->sa_bo, 8, 8); if (r) { kfree(*semaphore); -- cgit v1.2.3 From 2269a395790dd1ed3ae6e7708d76731e92fd46ff Mon Sep 17 00:00:00 2001 From: Christian König Date: Tue, 17 Nov 2015 11:02:47 +0100 Subject: drm/amdgpu: fix typo in firmware name MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Christian König Reviewed-by: Chunming Zhou --- drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index b93b649d9e36..98ee39001c45 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -40,7 +40,7 @@ static void gmc_v7_0_set_gart_funcs(struct amdgpu_device *adev); static void gmc_v7_0_set_irq_funcs(struct amdgpu_device *adev); -MODULE_FIRMWARE("radeon/boniare_mc.bin"); +MODULE_FIRMWARE("radeon/bonaire_mc.bin"); MODULE_FIRMWARE("radeon/hawaii_mc.bin"); /** -- cgit v1.2.3 From 984810fc45389c545719fbb4219e8a12b27032a4 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sat, 14 Nov 2015 21:05:35 +0100 Subject: drm/amdgpu: cleanup scheduler command submission MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Unify the two code path again, cause they do pretty much the same thing. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 63 +++++++++++++--------------------- 2 files changed, 25 insertions(+), 39 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 0f187027c753..0ce5c1e45ec8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -1210,6 +1210,7 @@ struct amdgpu_cs_parser { /* relocations */ struct amdgpu_bo_list_entry *vm_bos; struct list_head validated; + struct fence *fence; struct amdgpu_ib *ibs; uint32_t num_ibs; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 6096effd6a56..3afcf0237c25 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -439,8 +439,18 @@ static int cmp_size_smaller_first(void *priv, struct list_head *a, return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages; } -static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff) +/** + * cs_parser_fini() - clean parser states + * @parser: parser structure holding parsing context. + * @error: error number + * + * If error is set than unvalidate buffer, otherwise just free memory + * used by parsing context. + **/ +static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff) { + unsigned i; + if (!error) { /* Sort the buffer list from the smallest to largest buffer, * which affects the order of buffers in the LRU list. @@ -455,17 +465,13 @@ static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int err list_sort(NULL, &parser->validated, cmp_size_smaller_first); ttm_eu_fence_buffer_objects(&parser->ticket, - &parser->validated, - &parser->ibs[parser->num_ibs-1].fence->base); + &parser->validated, + parser->fence); } else if (backoff) { ttm_eu_backoff_reservation(&parser->ticket, &parser->validated); } -} - -static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) -{ - unsigned i; + fence_put(parser->fence); if (parser->ctx) amdgpu_ctx_put(parser->ctx); @@ -484,20 +490,6 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser) drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base); } -/** - * cs_parser_fini() - clean parser states - * @parser: parser structure holding parsing context. - * @error: error number - * - * If error is set than unvalidate buffer, otherwise just free memory - * used by parsing context. - **/ -static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff) -{ - amdgpu_cs_parser_fini_early(parser, error, backoff); - amdgpu_cs_parser_fini_late(parser); -} - static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p, struct amdgpu_vm *vm) { @@ -582,15 +574,9 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev, } r = amdgpu_bo_vm_update_pte(parser, vm); - if (r) { - goto out; - } - amdgpu_cs_sync_rings(parser); - if (!amdgpu_enable_scheduler) - r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs, - parser->filp); + if (!r) + amdgpu_cs_sync_rings(parser); -out: return r; } @@ -881,7 +867,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) goto out; } job->base.s_fence = fence; - fence_get(&fence->base); + parser.fence = fence_get(&fence->base); cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring, &fence->base); @@ -890,17 +876,16 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) trace_amdgpu_cs_ioctl(job); amd_sched_entity_push_job(&job->base); - list_sort(NULL, &parser.validated, cmp_size_smaller_first); - ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated, - &fence->base); - fence_put(&fence->base); + } else { + struct amdgpu_fence *fence; - amdgpu_cs_parser_fini_late(&parser); - mutex_unlock(&vm->mutex); - return 0; + r = amdgpu_ib_schedule(adev, parser.num_ibs, parser.ibs, + parser.filp); + fence = parser.ibs[parser.num_ibs - 1].fence; + parser.fence = fence_get(&fence->base); + cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence; } - cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence; out: amdgpu_cs_parser_fini(&parser, r, reserved_buffers); mutex_unlock(&vm->mutex); -- cgit v1.2.3 From eeed25ab83bbd68b2c9dba3cc2ce5913537594f6 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sun, 15 Nov 2015 15:50:07 +0100 Subject: drm/amdgpu: remove unused VM manager field MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 -- 1 file changed, 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 0ce5c1e45ec8..819d1473de30 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -967,8 +967,6 @@ struct amdgpu_vm_manager { u64 vram_base_offset; /* is vm enabled? */ bool enabled; - /* for hw to save the PD addr on suspend/resume */ - uint32_t saved_table_addr[AMDGPU_NUM_VM]; /* vm pte handling */ const struct amdgpu_vm_pte_funcs *vm_pte_funcs; struct amdgpu_ring *vm_pte_funcs_ring; -- cgit v1.2.3 From 8b4fb00b5db271bb1e117a078d5d7a161d8d8af1 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sun, 15 Nov 2015 16:04:16 +0100 Subject: drm/amdgpu: cleanup VM coding style MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Fix the indentation and move the VM functions to the structures. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Junwei Zhang Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 98 ++++++++++++++++++------------------- 1 file changed, 48 insertions(+), 50 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 819d1473de30..d1e16726e225 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -916,8 +916,8 @@ struct amdgpu_ring { #define AMDGPU_VM_FAULT_STOP_ALWAYS 2 struct amdgpu_vm_pt { - struct amdgpu_bo *bo; - uint64_t addr; + struct amdgpu_bo *bo; + uint64_t addr; }; struct amdgpu_vm_id { @@ -959,19 +959,60 @@ struct amdgpu_vm { }; struct amdgpu_vm_manager { - struct fence *active[AMDGPU_NUM_VM]; - uint32_t max_pfn; + struct fence *active[AMDGPU_NUM_VM]; + uint32_t max_pfn; /* number of VMIDs */ - unsigned nvm; + unsigned nvm; /* vram base address for page table entry */ - u64 vram_base_offset; + u64 vram_base_offset; /* is vm enabled? */ - bool enabled; + bool enabled; /* vm pte handling */ const struct amdgpu_vm_pte_funcs *vm_pte_funcs; struct amdgpu_ring *vm_pte_funcs_ring; }; +int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm); +void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); +struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, + struct amdgpu_vm *vm, + struct list_head *head); +int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, + struct amdgpu_sync *sync); +void amdgpu_vm_flush(struct amdgpu_ring *ring, + struct amdgpu_vm *vm, + struct fence *updates); +void amdgpu_vm_fence(struct amdgpu_device *adev, + struct amdgpu_vm *vm, + struct fence *fence); +uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr); +int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, + struct amdgpu_vm *vm); +int amdgpu_vm_clear_freed(struct amdgpu_device *adev, + struct amdgpu_vm *vm); +int amdgpu_vm_clear_invalids(struct amdgpu_device *adev, struct amdgpu_vm *vm, + struct amdgpu_sync *sync); +int amdgpu_vm_bo_update(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va, + struct ttm_mem_reg *mem); +void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, + struct amdgpu_bo *bo); +struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm, + struct amdgpu_bo *bo); +struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, + struct amdgpu_vm *vm, + struct amdgpu_bo *bo); +int amdgpu_vm_bo_map(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va, + uint64_t addr, uint64_t offset, + uint64_t size, uint32_t flags); +int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va, + uint64_t addr); +void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, + struct amdgpu_bo_va *bo_va); +int amdgpu_vm_free_job(struct amdgpu_job *job); + /* * context related structures */ @@ -2311,49 +2352,6 @@ int amdgpu_get_vblank_timestamp_kms(struct drm_device *dev, unsigned int pipe, long amdgpu_kms_compat_ioctl(struct file *filp, unsigned int cmd, unsigned long arg); -/* - * vm - */ -int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm); -void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); -struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct list_head *head); -int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, - struct amdgpu_sync *sync); -void amdgpu_vm_flush(struct amdgpu_ring *ring, - struct amdgpu_vm *vm, - struct fence *updates); -void amdgpu_vm_fence(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct fence *fence); -uint64_t amdgpu_vm_map_gart(struct amdgpu_device *adev, uint64_t addr); -int amdgpu_vm_update_page_directory(struct amdgpu_device *adev, - struct amdgpu_vm *vm); -int amdgpu_vm_clear_freed(struct amdgpu_device *adev, - struct amdgpu_vm *vm); -int amdgpu_vm_clear_invalids(struct amdgpu_device *adev, - struct amdgpu_vm *vm, struct amdgpu_sync *sync); -int amdgpu_vm_bo_update(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va, - struct ttm_mem_reg *mem); -void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev, - struct amdgpu_bo *bo); -struct amdgpu_bo_va *amdgpu_vm_bo_find(struct amdgpu_vm *vm, - struct amdgpu_bo *bo); -struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, - struct amdgpu_vm *vm, - struct amdgpu_bo *bo); -int amdgpu_vm_bo_map(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va, - uint64_t addr, uint64_t offset, - uint64_t size, uint32_t flags); -int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va, - uint64_t addr); -void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, - struct amdgpu_bo_va *bo_va); -int amdgpu_vm_free_job(struct amdgpu_job *job); /* * functions used by amdgpu_encoder.c */ -- cgit v1.2.3 From ea89f8c9e8ba8a7b75446eef36917da50d2186d9 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sun, 15 Nov 2015 20:52:06 +0100 Subject: drm/amdgpu: move VM manager clean into the VM code again MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit It's not a good idea to duplicate that code. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Reviewed-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++++ drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 8 ++------ drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 8 ++------ 4 files changed, 20 insertions(+), 12 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index d1e16726e225..beb74854a8a3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -972,6 +972,7 @@ struct amdgpu_vm_manager { struct amdgpu_ring *vm_pte_funcs_ring; }; +void amdgpu_vm_manager_fini(struct amdgpu_device *adev); int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm); void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm); struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index a9fcc995d27e..6bb209bc0d36 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1318,3 +1318,18 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) mutex_destroy(&vm->mutex); } + +/** + * amdgpu_vm_manager_fini - cleanup VM manager + * + * @adev: amdgpu_device pointer + * + * Cleanup the VM manager and free resources. + */ +void amdgpu_vm_manager_fini(struct amdgpu_device *adev) +{ + unsigned i; + + for (i = 0; i < AMDGPU_NUM_VM; ++i) + fence_put(adev->vm_manager.active[i]); +} diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c index 98ee39001c45..7427d8cd4c43 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c @@ -961,12 +961,10 @@ static int gmc_v7_0_sw_init(void *handle) static int gmc_v7_0_sw_fini(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1011,12 +1009,10 @@ static int gmc_v7_0_hw_fini(void *handle) static int gmc_v7_0_suspend(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v7_0_vm_fini(adev); adev->vm_manager.enabled = false; } diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c index c9209b45d901..cb0e50ebb528 100644 --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c @@ -980,12 +980,10 @@ static int gmc_v8_0_sw_init(void *handle) static int gmc_v8_0_sw_fini(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } @@ -1032,12 +1030,10 @@ static int gmc_v8_0_hw_fini(void *handle) static int gmc_v8_0_suspend(void *handle) { - int i; struct amdgpu_device *adev = (struct amdgpu_device *)handle; if (adev->vm_manager.enabled) { - for (i = 0; i < AMDGPU_NUM_VM; ++i) - fence_put(adev->vm_manager.active[i]); + amdgpu_vm_manager_fini(adev); gmc_v8_0_vm_fini(adev); adev->vm_manager.enabled = false; } -- cgit v1.2.3 From 1c16c0a7b26c6c905dc79c4194135ca2f360f0f5 Mon Sep 17 00:00:00 2001 From: Christian König Date: Sat, 14 Nov 2015 21:31:40 +0100 Subject: drm/amdgpu: keep the owner for VMIDs MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We don't need the last VM use any more, keep the owner directly. Signed-off-by: Christian König Reviewed-by: Chunming Zhou --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 8 +++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 33 +++++++++++++++++++-------------- 2 files changed, 24 insertions(+), 17 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index beb74854a8a3..a5692624070a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -925,8 +925,6 @@ struct amdgpu_vm_id { uint64_t pd_gpu_addr; /* last flushed PD/PT update */ struct fence *flushed_updates; - /* last use of vmid */ - struct fence *last_id_use; }; struct amdgpu_vm { @@ -959,7 +957,11 @@ struct amdgpu_vm { }; struct amdgpu_vm_manager { - struct fence *active[AMDGPU_NUM_VM]; + struct { + struct fence *active; + atomic_long_t owner; + } ids[AMDGPU_NUM_VM]; + uint32_t max_pfn; /* number of VMIDs */ unsigned nvm; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 6bb209bc0d36..0bdbb2480f9b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -143,10 +143,15 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, unsigned i; /* check if the id is still valid */ - if (vm_id->id && vm_id->last_id_use && - vm_id->last_id_use == adev->vm_manager.active[vm_id->id]) { - trace_amdgpu_vm_grab_id(vm_id->id, ring->idx); - return 0; + if (vm_id->id) { + unsigned id = vm_id->id; + long owner; + + owner = atomic_long_read(&adev->vm_manager.ids[id].owner); + if (owner == (long)vm) { + trace_amdgpu_vm_grab_id(vm_id->id, ring->idx); + return 0; + } } /* we definately need to flush */ @@ -154,7 +159,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, /* skip over VMID 0, since it is the system VM */ for (i = 1; i < adev->vm_manager.nvm; ++i) { - struct fence *fence = adev->vm_manager.active[i]; + struct fence *fence = adev->vm_manager.ids[i].active; struct amdgpu_ring *fring; if (fence == NULL) { @@ -176,7 +181,7 @@ int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring, if (choices[i]) { struct fence *fence; - fence = adev->vm_manager.active[choices[i]]; + fence = adev->vm_manager.ids[choices[i]].active; vm_id->id = choices[i]; trace_amdgpu_vm_grab_id(choices[i], ring->idx); @@ -246,11 +251,9 @@ void amdgpu_vm_fence(struct amdgpu_device *adev, struct amdgpu_ring *ring = amdgpu_ring_from_fence(fence); unsigned vm_id = vm->ids[ring->idx].id; - fence_put(adev->vm_manager.active[vm_id]); - adev->vm_manager.active[vm_id] = fence_get(fence); - - fence_put(vm->ids[ring->idx].last_id_use); - vm->ids[ring->idx].last_id_use = fence_get(fence); + fence_put(adev->vm_manager.ids[vm_id].active); + adev->vm_manager.ids[vm_id].active = fence_get(fence); + atomic_long_set(&adev->vm_manager.ids[vm_id].owner, (long)vm); } /** @@ -1238,7 +1241,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { vm->ids[i].id = 0; vm->ids[i].flushed_updates = NULL; - vm->ids[i].last_id_use = NULL; } mutex_init(&vm->mutex); vm->va = RB_ROOT; @@ -1312,8 +1314,11 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) fence_put(vm->page_directory_fence); for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + unsigned id = vm->ids[i].id; + + atomic_long_cmpxchg(&adev->vm_manager.ids[id].owner, + (long)vm, 0); fence_put(vm->ids[i].flushed_updates); - fence_put(vm->ids[i].last_id_use); } mutex_destroy(&vm->mutex); @@ -1331,5 +1336,5 @@ void amdgpu_vm_manager_fini(struct amdgpu_device *adev) unsigned i; for (i = 0; i < AMDGPU_NUM_VM; ++i) - fence_put(adev->vm_manager.active[i]); + fence_put(adev->vm_manager.ids[i].active); } -- cgit v1.2.3 From c25867dfabf045a8148fd179fa759bb17f670e42 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Fri, 13 Nov 2015 13:32:01 +0800 Subject: drm/amdgpu: add lock for interval tree in vm MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: I62b892a22af37b32e6b4aefca80a25cf45426ed2 Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 15 +++++++++++++-- 2 files changed, 15 insertions(+), 2 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a5692624070a..306f75700bf8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -954,6 +954,8 @@ struct amdgpu_vm { /* for id and flush management per ring */ struct amdgpu_vm_id ids[AMDGPU_MAX_RINGS]; + /* for interval tree */ + spinlock_t it_lock; }; struct amdgpu_vm_manager { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 0bdbb2480f9b..0513f3fed2c2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1028,7 +1028,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, saddr /= AMDGPU_GPU_PAGE_SIZE; eaddr /= AMDGPU_GPU_PAGE_SIZE; + spin_lock(&vm->it_lock); it = interval_tree_iter_first(&vm->va, saddr, eaddr - 1); + spin_unlock(&vm->it_lock); if (it) { struct amdgpu_bo_va_mapping *tmp; tmp = container_of(it, struct amdgpu_bo_va_mapping, it); @@ -1055,7 +1057,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, mapping->flags = flags; list_add(&mapping->list, &bo_va->invalids); + spin_lock(&vm->it_lock); interval_tree_insert(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_map(bo_va, mapping); /* Make sure the page tables are allocated */ @@ -1101,7 +1105,9 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, error_free: list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_unmap(bo_va, mapping); kfree(mapping); @@ -1151,7 +1157,9 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, } list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_unmap(bo_va, mapping); if (valid) @@ -1187,13 +1195,17 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, list_for_each_entry_safe(mapping, next, &bo_va->valids, list) { list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); trace_amdgpu_vm_bo_unmap(bo_va, mapping); list_add(&mapping->list, &vm->freed); } list_for_each_entry_safe(mapping, next, &bo_va->invalids, list) { list_del(&mapping->list); + spin_lock(&vm->it_lock); interval_tree_remove(&mapping->it, &vm->va); + spin_unlock(&vm->it_lock); kfree(mapping); } @@ -1248,7 +1260,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) INIT_LIST_HEAD(&vm->invalidated); INIT_LIST_HEAD(&vm->cleared); INIT_LIST_HEAD(&vm->freed); - + spin_lock_init(&vm->it_lock); pd_size = amdgpu_vm_directory_size(adev); pd_entries = amdgpu_vm_num_pdes(adev); @@ -1312,7 +1324,6 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) amdgpu_bo_unref(&vm->page_directory); fence_put(vm->page_directory_fence); - for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { unsigned id = vm->ids[i].id; -- cgit v1.2.3 From ef9f0a83d68cecca241ee5d61b797d598722772e Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Fri, 13 Nov 2015 13:43:22 +0800 Subject: drm/amdgpu: move bo_reserve out of amdgpu_vm_clear_bo MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: Ifbb0c06680494bfa04d0be5e5941d31ae2e5ef28 Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 33 ++++++++++++++++++--------------- 1 file changed, 18 insertions(+), 15 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 0513f3fed2c2..e6dc19bc2dd8 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -332,6 +332,8 @@ int amdgpu_vm_free_job(struct amdgpu_job *job) * * @adev: amdgpu_device pointer * @bo: bo to clear + * + * need to reserve bo first before calling it. */ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, struct amdgpu_bo *bo) @@ -343,24 +345,20 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, uint64_t addr; int r; - r = amdgpu_bo_reserve(bo, false); - if (r) - return r; - r = reservation_object_reserve_shared(bo->tbo.resv); if (r) return r; r = ttm_bo_validate(&bo->tbo, &bo->placement, true, false); if (r) - goto error_unreserve; + goto error; addr = amdgpu_bo_gpu_offset(bo); entries = amdgpu_bo_size(bo) / 8; ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL); if (!ib) - goto error_unreserve; + goto error; r = amdgpu_ib_get(ring, NULL, entries * 2 + 64, ib); if (r) @@ -378,16 +376,14 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, if (!r) amdgpu_bo_fence(bo, fence, true); fence_put(fence); - if (amdgpu_enable_scheduler) { - amdgpu_bo_unreserve(bo); + if (amdgpu_enable_scheduler) return 0; - } + error_free: amdgpu_ib_free(adev, ib); kfree(ib); -error_unreserve: - amdgpu_bo_unreserve(bo); +error: return r; } @@ -1087,11 +1083,12 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, AMDGPU_GEM_DOMAIN_VRAM, AMDGPU_GEM_CREATE_NO_CPU_ACCESS, NULL, resv, &pt); - ww_mutex_unlock(&resv->lock); - if (r) + if (r) { + ww_mutex_unlock(&resv->lock); goto error_free; - + } r = amdgpu_vm_clear_bo(adev, pt); + ww_mutex_unlock(&resv->lock); if (r) { amdgpu_bo_unref(&pt); goto error_free; @@ -1280,8 +1277,14 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm) NULL, NULL, &vm->page_directory); if (r) return r; - + r = amdgpu_bo_reserve(vm->page_directory, false); + if (r) { + amdgpu_bo_unref(&vm->page_directory); + vm->page_directory = NULL; + return r; + } r = amdgpu_vm_clear_bo(adev, vm->page_directory); + amdgpu_bo_unreserve(vm->page_directory); if (r) { amdgpu_bo_unref(&vm->page_directory); vm->page_directory = NULL; -- cgit v1.2.3 From 49b02b180a541d6fb31031aaffe75496e9238942 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Fri, 13 Nov 2015 14:18:38 +0800 Subject: drm/amdgpu: reserve/unreserve objects out of map/unmap operations MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Change-Id: Id6514f2fb6e002437fdbe99353d5d35f4ac736c7 Signed-off-by: Chunming Zhou Reviewed-by: Christian König --- drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 21 ++++++++++++++++++--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 29 +++++++---------------------- 2 files changed, 25 insertions(+), 25 deletions(-) (limited to 'drivers') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c index 16dca46314bd..00c5b580f56c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c @@ -515,6 +515,9 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, struct amdgpu_fpriv *fpriv = filp->driver_priv; struct amdgpu_bo *rbo; struct amdgpu_bo_va *bo_va; + struct ttm_validate_buffer tv, tv_pd; + struct ww_acquire_ctx ticket; + struct list_head list, duplicates; uint32_t invalid_flags, va_flags = 0; int r = 0; @@ -552,7 +555,18 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, return -ENOENT; mutex_lock(&fpriv->vm.mutex); rbo = gem_to_amdgpu_bo(gobj); - r = amdgpu_bo_reserve(rbo, false); + INIT_LIST_HEAD(&list); + INIT_LIST_HEAD(&duplicates); + tv.bo = &rbo->tbo; + tv.shared = true; + list_add(&tv.head, &list); + + if (args->operation == AMDGPU_VA_OP_MAP) { + tv_pd.bo = &fpriv->vm.page_directory->tbo; + tv_pd.shared = true; + list_add(&tv_pd.head, &list); + } + r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates); if (r) { mutex_unlock(&fpriv->vm.mutex); drm_gem_object_unreference_unlocked(gobj); @@ -561,7 +575,8 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, bo_va = amdgpu_vm_bo_find(&fpriv->vm, rbo); if (!bo_va) { - amdgpu_bo_unreserve(rbo); + ttm_eu_backoff_reservation(&ticket, &list); + drm_gem_object_unreference_unlocked(gobj); mutex_unlock(&fpriv->vm.mutex); return -ENOENT; } @@ -584,7 +599,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data, default: break; } - + ttm_eu_backoff_reservation(&ticket, &list); if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE)) amdgpu_gem_va_update_vm(adev, bo_va, args->operation); mutex_unlock(&fpriv->vm.mutex); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index e6dc19bc2dd8..159ce54bbd8d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -985,7 +985,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev, * Add a mapping of the BO at the specefied addr into the VM. * Returns 0 for success, error for failure. * - * Object has to be reserved and gets unreserved by this function! + * Object has to be reserved and unreserved outside! */ int amdgpu_vm_bo_map(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, @@ -1001,23 +1001,18 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, /* validate the parameters */ if (saddr & AMDGPU_GPU_PAGE_MASK || offset & AMDGPU_GPU_PAGE_MASK || - size == 0 || size & AMDGPU_GPU_PAGE_MASK) { - amdgpu_bo_unreserve(bo_va->bo); + size == 0 || size & AMDGPU_GPU_PAGE_MASK) return -EINVAL; - } /* make sure object fit at this offset */ eaddr = saddr + size; - if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo))) { - amdgpu_bo_unreserve(bo_va->bo); + if ((saddr >= eaddr) || (offset + size > amdgpu_bo_size(bo_va->bo))) return -EINVAL; - } last_pfn = eaddr / AMDGPU_GPU_PAGE_SIZE; if (last_pfn > adev->vm_manager.max_pfn) { dev_err(adev->dev, "va above limit (0x%08X > 0x%08X)\n", last_pfn, adev->vm_manager.max_pfn); - amdgpu_bo_unreserve(bo_va->bo); return -EINVAL; } @@ -1034,14 +1029,12 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, dev_err(adev->dev, "bo %p va 0x%010Lx-0x%010Lx conflict with " "0x%010lx-0x%010lx\n", bo_va->bo, saddr, eaddr, tmp->it.start, tmp->it.last + 1); - amdgpu_bo_unreserve(bo_va->bo); r = -EINVAL; goto error; } mapping = kmalloc(sizeof(*mapping), GFP_KERNEL); if (!mapping) { - amdgpu_bo_unreserve(bo_va->bo); r = -ENOMEM; goto error; } @@ -1067,8 +1060,6 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, if (eaddr > vm->max_pde_used) vm->max_pde_used = eaddr; - amdgpu_bo_unreserve(bo_va->bo); - /* walk over the address space and allocate the page tables */ for (pt_idx = saddr; pt_idx <= eaddr; ++pt_idx) { struct reservation_object *resv = vm->page_directory->tbo.resv; @@ -1077,18 +1068,15 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev, if (vm->page_tables[pt_idx].bo) continue; - ww_mutex_lock(&resv->lock, NULL); r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8, AMDGPU_GPU_PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM, AMDGPU_GEM_CREATE_NO_CPU_ACCESS, NULL, resv, &pt); - if (r) { - ww_mutex_unlock(&resv->lock); + if (r) goto error_free; - } + r = amdgpu_vm_clear_bo(adev, pt); - ww_mutex_unlock(&resv->lock); if (r) { amdgpu_bo_unref(&pt); goto error_free; @@ -1122,7 +1110,7 @@ error: * Remove a mapping of the BO at the specefied addr from the VM. * Returns 0 for success, error for failure. * - * Object has to be reserved and gets unreserved by this function! + * Object has to be reserved and unreserved outside! */ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, struct amdgpu_bo_va *bo_va, @@ -1147,10 +1135,8 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, break; } - if (&mapping->list == &bo_va->invalids) { - amdgpu_bo_unreserve(bo_va->bo); + if (&mapping->list == &bo_va->invalids) return -ENOENT; - } } list_del(&mapping->list); @@ -1163,7 +1149,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev, list_add(&mapping->list, &vm->freed); else kfree(mapping); - amdgpu_bo_unreserve(bo_va->bo); return 0; } -- cgit v1.2.3