From e2fb6e0a7a2194f5ef1fada737217aad71e29e4d Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 9 Jan 2019 15:36:29 +0100 Subject: drm/amdgpu: cleanup amdgpu_ih_process a bit more MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Remove the callback and call the dispatcher directly. Signed-off-by: Christian König Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c | 6 ++--- drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h | 4 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c | 48 ++++++++++++--------------------- drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h | 2 +- 4 files changed, 21 insertions(+), 39 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c index d0a5db777b6d..1c50be3ab8a9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.c @@ -140,9 +140,7 @@ void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) * Interrupt hander (VI), walk the IH ring. * Returns irq process return code. */ -int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, - void (*callback)(struct amdgpu_device *adev, - struct amdgpu_ih_ring *ih)) +int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih) { u32 wptr; @@ -162,7 +160,7 @@ restart_ih: rmb(); while (ih->rptr != wptr) { - callback(adev, ih); + amdgpu_irq_dispatch(adev, ih); ih->rptr &= ih->ptr_mask; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h index 1ccb1831382a..113a1ba13d4a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h @@ -69,8 +69,6 @@ struct amdgpu_ih_funcs { int amdgpu_ih_ring_init(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, unsigned ring_size, bool use_bus_addr); void amdgpu_ih_ring_fini(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); -int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih, - void (*callback)(struct amdgpu_device *adev, - struct amdgpu_ih_ring *ih)); +int amdgpu_ih_process(struct amdgpu_device *adev, struct amdgpu_ih_ring *ih); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c index 8bfb3dab46f7..af4c3b1af322 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c @@ -130,29 +130,6 @@ void amdgpu_irq_disable_all(struct amdgpu_device *adev) spin_unlock_irqrestore(&adev->irq.lock, irqflags); } -/** - * amdgpu_irq_callback - callback from the IH ring - * - * @adev: amdgpu device pointer - * @ih: amdgpu ih ring - * - * Callback from IH ring processing to handle the entry at the current position - * and advance the read pointer. - */ -static void amdgpu_irq_callback(struct amdgpu_device *adev, - struct amdgpu_ih_ring *ih) -{ - u32 ring_index = ih->rptr >> 2; - struct amdgpu_iv_entry entry; - - entry.iv_entry = (const uint32_t *)&ih->ring[ring_index]; - amdgpu_ih_decode_iv(adev, &entry); - - trace_amdgpu_iv(ih - &adev->irq.ih, &entry); - - amdgpu_irq_dispatch(adev, &entry); -} - /** * amdgpu_irq_handler - IRQ handler * @@ -170,7 +147,7 @@ irqreturn_t amdgpu_irq_handler(int irq, void *arg) struct amdgpu_device *adev = dev->dev_private; irqreturn_t ret; - ret = amdgpu_ih_process(adev, &adev->irq.ih, amdgpu_irq_callback); + ret = amdgpu_ih_process(adev, &adev->irq.ih); if (ret == IRQ_HANDLED) pm_runtime_mark_last_busy(dev->dev); return ret; @@ -188,7 +165,7 @@ static void amdgpu_irq_handle_ih1(struct work_struct *work) struct amdgpu_device *adev = container_of(work, struct amdgpu_device, irq.ih1_work); - amdgpu_ih_process(adev, &adev->irq.ih1, amdgpu_irq_callback); + amdgpu_ih_process(adev, &adev->irq.ih1); } /** @@ -203,7 +180,7 @@ static void amdgpu_irq_handle_ih2(struct work_struct *work) struct amdgpu_device *adev = container_of(work, struct amdgpu_device, irq.ih2_work); - amdgpu_ih_process(adev, &adev->irq.ih2, amdgpu_irq_callback); + amdgpu_ih_process(adev, &adev->irq.ih2); } /** @@ -394,14 +371,23 @@ int amdgpu_irq_add_id(struct amdgpu_device *adev, * Dispatches IRQ to IP blocks. */ void amdgpu_irq_dispatch(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry) + struct amdgpu_ih_ring *ih) { - unsigned client_id = entry->client_id; - unsigned src_id = entry->src_id; + u32 ring_index = ih->rptr >> 2; + struct amdgpu_iv_entry entry; + unsigned client_id, src_id; struct amdgpu_irq_src *src; bool handled = false; int r; + entry.iv_entry = (const uint32_t *)&ih->ring[ring_index]; + amdgpu_ih_decode_iv(adev, &entry); + + trace_amdgpu_iv(ih - &adev->irq.ih, &entry); + + client_id = entry.client_id; + src_id = entry.src_id; + if (client_id >= AMDGPU_IRQ_CLIENTID_MAX) { DRM_DEBUG("Invalid client_id in IV: %d\n", client_id); @@ -416,7 +402,7 @@ void amdgpu_irq_dispatch(struct amdgpu_device *adev, client_id, src_id); } else if ((src = adev->irq.client[client_id].sources[src_id])) { - r = src->funcs->process(adev, src, entry); + r = src->funcs->process(adev, src, &entry); if (r < 0) DRM_ERROR("error processing interrupt (%d)\n", r); else if (r) @@ -428,7 +414,7 @@ void amdgpu_irq_dispatch(struct amdgpu_device *adev, /* Send it to amdkfd as well if it isn't already handled */ if (!handled) - amdgpu_amdkfd_interrupt(adev, entry->iv_entry); + amdgpu_amdkfd_interrupt(adev, entry.iv_entry); } /** diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h index c27decfda494..c718e94a55c9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h @@ -108,7 +108,7 @@ int amdgpu_irq_add_id(struct amdgpu_device *adev, unsigned client_id, unsigned src_id, struct amdgpu_irq_src *source); void amdgpu_irq_dispatch(struct amdgpu_device *adev, - struct amdgpu_iv_entry *entry); + struct amdgpu_ih_ring *ih); int amdgpu_irq_update(struct amdgpu_device *adev, struct amdgpu_irq_src *src, unsigned type); int amdgpu_irq_get(struct amdgpu_device *adev, struct amdgpu_irq_src *src, -- cgit v1.2.3 From 2c11ee6ae5533821cf2c64972d6cb5aa4ab4a46e Mon Sep 17 00:00:00 2001 From: wentalou Date: Wed, 30 Jan 2019 11:13:01 +0800 Subject: drm/amdgpu: tighten gpu_recover in mailbox_flr to avoid duplicate recover in sriov sriov's gpu_recover inside xgpu_ai_mailbox_flr_work would cause duplicate recover in TDR. TDR's gpu_recover would be triggered by amdgpu_job_timedout, that could avoid vk-cts failure by unexpected recover. Signed-off-by: Wentao Lou Acked-by: Andrey Grodzovsky Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c index b11a1c17a7f2..73851ebb3833 100644 --- a/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +++ b/drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c @@ -266,7 +266,8 @@ flr_done: } /* Trigger recovery for world switch failure if no TDR */ - if (amdgpu_device_should_recover_gpu(adev)) + if (amdgpu_device_should_recover_gpu(adev) + && amdgpu_lockup_timeout == MAX_SCHEDULE_TIMEOUT) amdgpu_device_gpu_recover(adev, NULL); } -- cgit v1.2.3 From e8e32426b1271d3387208c3a6eaf7c166bea9975 Mon Sep 17 00:00:00 2001 From: Felix Kuehling Date: Mon, 4 Feb 2019 17:46:35 -0500 Subject: drm/amdgpu: Add helper to wait for BO fences using a sync object MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Creates a temporary sync object to wait for the BO reservation. This generalizes amdgpu_vm_wait_pd. Signed-off-by: Felix Kuehling Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 24 ++++++++++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 30 +++--------------------------- 3 files changed, 28 insertions(+), 27 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index fd9c4beeaaa4..ec9e45004bff 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -1284,6 +1284,30 @@ void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence, reservation_object_add_excl_fence(resv, fence); } +/** + * amdgpu_sync_wait_resv - Wait for BO reservation fences + * + * @bo: buffer object + * @owner: fence owner + * @intr: Whether the wait is interruptible + * + * Returns: + * 0 on success, errno otherwise. + */ +int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr) +{ + struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); + struct amdgpu_sync sync; + int r; + + amdgpu_sync_create(&sync); + amdgpu_sync_resv(adev, &sync, bo->tbo.resv, owner, false); + r = amdgpu_sync_wait(&sync, intr); + amdgpu_sync_free(&sync); + + return r; +} + /** * amdgpu_bo_gpu_offset - return GPU offset of bo * @bo: amdgpu object for which we query the offset diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h index 9291c2f837e9..220a6a7b1bc1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h @@ -266,6 +266,7 @@ void amdgpu_bo_move_notify(struct ttm_buffer_object *bo, int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo); void amdgpu_bo_fence(struct amdgpu_bo *bo, struct dma_fence *fence, bool shared); +int amdgpu_bo_sync_wait(struct amdgpu_bo *bo, void *owner, bool intr); u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo); int amdgpu_bo_validate(struct amdgpu_bo *bo); int amdgpu_bo_restore_shadow(struct amdgpu_bo *shadow, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 75481cf3348f..3f33286c3cd5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -1332,31 +1332,6 @@ static void amdgpu_vm_cpu_set_ptes(struct amdgpu_pte_update_params *params, } } - -/** - * amdgpu_vm_wait_pd - Wait for PT BOs to be free. - * - * @adev: amdgpu_device pointer - * @vm: related vm - * @owner: fence owner - * - * Returns: - * 0 on success, errno otherwise. - */ -static int amdgpu_vm_wait_pd(struct amdgpu_device *adev, struct amdgpu_vm *vm, - void *owner) -{ - struct amdgpu_sync sync; - int r; - - amdgpu_sync_create(&sync); - amdgpu_sync_resv(adev, &sync, vm->root.base.bo->tbo.resv, owner, false); - r = amdgpu_sync_wait(&sync, true); - amdgpu_sync_free(&sync); - - return r; -} - /** * amdgpu_vm_update_func - helper to call update function * @@ -1451,7 +1426,8 @@ restart: params.adev = adev; if (vm->use_cpu_for_update) { - r = amdgpu_vm_wait_pd(adev, vm, AMDGPU_FENCE_OWNER_VM); + r = amdgpu_bo_sync_wait(vm->root.base.bo, + AMDGPU_FENCE_OWNER_VM, true); if (unlikely(r)) return r; @@ -1784,7 +1760,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, /* Wait for PT BOs to be idle. PTs share the same resv. object * as the root PD BO */ - r = amdgpu_vm_wait_pd(adev, vm, owner); + r = amdgpu_bo_sync_wait(vm->root.base.bo, owner, true); if (unlikely(r)) return r; -- cgit v1.2.3 From c60cd590cb7da0d7dbb423727bb67350182a371c Mon Sep 17 00:00:00 2001 From: Felix Kuehling Date: Mon, 4 Feb 2019 17:53:05 -0500 Subject: drm/amdgpu: Replace ttm_bo_wait with amdgpu_bo_sync_wait MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The fence_owner logic in amdgpu_sync_wait will allow waiting without having to temporarily remove eviction fences. Signed-off-by: Felix Kuehling Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index d7b10d79f1de..44a1581f6b8f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -355,7 +355,7 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain, if (ret) goto validate_fail; - ttm_bo_wait(&bo->tbo, false, false); + amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false); amdgpu_amdkfd_add_eviction_fence(bo, ef_list, ef_count); } @@ -1002,7 +1002,7 @@ static int init_kfd_vm(struct amdgpu_vm *vm, void **process_info, pr_err("validate_pt_pd_bos() failed\n"); goto validate_pd_fail; } - ret = ttm_bo_wait(&vm->root.base.bo->tbo, false, false); + amdgpu_bo_sync_wait(vm->root.base.bo, AMDGPU_FENCE_OWNER_KFD, false); if (ret) goto wait_pd_fail; amdgpu_bo_fence(vm->root.base.bo, -- cgit v1.2.3 From 8db588d5d35e6ab0f8fa404c0d3eade6bbd272bc Mon Sep 17 00:00:00 2001 From: Felix Kuehling Date: Thu, 7 Feb 2019 12:08:14 -0500 Subject: drm/amdgpu: Avoid setting off KFD eviction fences in amdgpu_vm MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Use FENCE_OWNER_KFD to synchronize PT/PD initialization and clearing of page table entries. This avoids triggering KFD eviction fences on the PD reservation objects of compute VMs. Signed-off-by: Felix Kuehling Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 3f33286c3cd5..6dda415bb3ee 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -828,7 +828,7 @@ static int amdgpu_vm_clear_bo(struct amdgpu_device *adev, WARN_ON(job->ibs[0].length_dw > 64); r = amdgpu_sync_resv(adev, &job->sync, bo->tbo.resv, - AMDGPU_FENCE_OWNER_UNDEFINED, false); + AMDGPU_FENCE_OWNER_KFD, false); if (r) goto error_free; @@ -1748,9 +1748,9 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev, params.adev = adev; params.vm = vm; - /* sync to everything on unmapping */ + /* sync to everything except eviction fences on unmapping */ if (!(flags & AMDGPU_PTE_VALID)) - owner = AMDGPU_FENCE_OWNER_UNDEFINED; + owner = AMDGPU_FENCE_OWNER_KFD; if (vm->use_cpu_for_update) { /* params.src is used as flag to indicate system Memory */ -- cgit v1.2.3 From 2d086fded1021d6b482dfc705bab8f085d10a496 Mon Sep 17 00:00:00 2001 From: Felix Kuehling Date: Mon, 4 Feb 2019 18:17:26 -0500 Subject: drm/amdgpu: Simplify eviction fence handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Temporarily removing eviction fences to avoid triggering them by accident is no longer necessary due to the fence_owner logic in amdgpu_sync_resv. As a result the ef_list usage of amdgpu_amdkfd_remove_eviction_fence and amdgpu_amdkfd_add_eviction_fence are no longer needed. Signed-off-by: Felix Kuehling Acked-by: Christian König Reviewed-by: Harish Kasiviswanathan Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 134 ++--------------------- 1 file changed, 11 insertions(+), 123 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c index 44a1581f6b8f..1921dec3df7a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c @@ -204,38 +204,25 @@ void amdgpu_amdkfd_unreserve_memory_limit(struct amdgpu_bo *bo) } -/* amdgpu_amdkfd_remove_eviction_fence - Removes eviction fence(s) from BO's +/* amdgpu_amdkfd_remove_eviction_fence - Removes eviction fence from BO's * reservation object. * * @bo: [IN] Remove eviction fence(s) from this BO - * @ef: [IN] If ef is specified, then this eviction fence is removed if it + * @ef: [IN] This eviction fence is removed if it * is present in the shared list. - * @ef_list: [OUT] Returns list of eviction fences. These fences are removed - * from BO's reservation object shared list. - * @ef_count: [OUT] Number of fences in ef_list. * - * NOTE: If called with ef_list, then amdgpu_amdkfd_add_eviction_fence must be - * called to restore the eviction fences and to avoid memory leak. This is - * useful for shared BOs. * NOTE: Must be called with BO reserved i.e. bo->tbo.resv->lock held. */ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, - struct amdgpu_amdkfd_fence *ef, - struct amdgpu_amdkfd_fence ***ef_list, - unsigned int *ef_count) + struct amdgpu_amdkfd_fence *ef) { struct reservation_object *resv = bo->tbo.resv; struct reservation_object_list *old, *new; unsigned int i, j, k; - if (!ef && !ef_list) + if (!ef) return -EINVAL; - if (ef_list) { - *ef_list = NULL; - *ef_count = 0; - } - old = reservation_object_get_list(resv); if (!old) return 0; @@ -254,8 +241,7 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, f = rcu_dereference_protected(old->shared[i], reservation_object_held(resv)); - if ((ef && f->context == ef->base.context) || - (!ef && to_amdgpu_amdkfd_fence(f))) + if (f->context == ef->base.context) RCU_INIT_POINTER(new->shared[--j], f); else RCU_INIT_POINTER(new->shared[k++], f); @@ -263,21 +249,6 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, new->shared_max = old->shared_max; new->shared_count = k; - if (!ef) { - unsigned int count = old->shared_count - j; - - /* Alloc memory for count number of eviction fence pointers. - * Fill the ef_list array and ef_count - */ - *ef_list = kcalloc(count, sizeof(**ef_list), GFP_KERNEL); - *ef_count = count; - - if (!*ef_list) { - kfree(new); - return -ENOMEM; - } - } - /* Install the new fence list, seqcount provides the barriers */ preempt_disable(); write_seqcount_begin(&resv->seq); @@ -291,46 +262,13 @@ static int amdgpu_amdkfd_remove_eviction_fence(struct amdgpu_bo *bo, f = rcu_dereference_protected(new->shared[i], reservation_object_held(resv)); - if (!ef) - (*ef_list)[k++] = to_amdgpu_amdkfd_fence(f); - else - dma_fence_put(f); + dma_fence_put(f); } kfree_rcu(old, rcu); return 0; } -/* amdgpu_amdkfd_add_eviction_fence - Adds eviction fence(s) back into BO's - * reservation object. - * - * @bo: [IN] Add eviction fences to this BO - * @ef_list: [IN] List of eviction fences to be added - * @ef_count: [IN] Number of fences in ef_list. - * - * NOTE: Must call amdgpu_amdkfd_remove_eviction_fence before calling this - * function. - */ -static void amdgpu_amdkfd_add_eviction_fence(struct amdgpu_bo *bo, - struct amdgpu_amdkfd_fence **ef_list, - unsigned int ef_count) -{ - int i; - - if (!ef_list || !ef_count) - return; - - for (i = 0; i < ef_count; i++) { - amdgpu_bo_fence(bo, &ef_list[i]->base, true); - /* Re-adding the fence takes an additional reference. Drop that - * reference. - */ - dma_fence_put(&ef_list[i]->base); - } - - kfree(ef_list); -} - static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain, bool wait) { @@ -346,18 +284,8 @@ static int amdgpu_amdkfd_bo_validate(struct amdgpu_bo *bo, uint32_t domain, ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx); if (ret) goto validate_fail; - if (wait) { - struct amdgpu_amdkfd_fence **ef_list; - unsigned int ef_count; - - ret = amdgpu_amdkfd_remove_eviction_fence(bo, NULL, &ef_list, - &ef_count); - if (ret) - goto validate_fail; - + if (wait) amdgpu_bo_sync_wait(bo, AMDGPU_FENCE_OWNER_KFD, false); - amdgpu_amdkfd_add_eviction_fence(bo, ef_list, ef_count); - } validate_fail: return ret; @@ -444,7 +372,6 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, { int ret; struct kfd_bo_va_list *bo_va_entry; - struct amdgpu_bo *pd = vm->root.base.bo; struct amdgpu_bo *bo = mem->bo; uint64_t va = mem->va; struct list_head *list_bo_va = &mem->bo_va_list; @@ -484,14 +411,8 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, *p_bo_va_entry = bo_va_entry; /* Allocate new page tables if needed and validate - * them. Clearing of new page tables and validate need to wait - * on move fences. We don't want that to trigger the eviction - * fence, so remove it temporarily. + * them. */ - amdgpu_amdkfd_remove_eviction_fence(pd, - vm->process_info->eviction_fence, - NULL, NULL); - ret = amdgpu_vm_alloc_pts(adev, vm, va, amdgpu_bo_size(bo)); if (ret) { pr_err("Failed to allocate pts, err=%d\n", ret); @@ -504,13 +425,9 @@ static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem, goto err_alloc_pts; } - /* Add the eviction fence back */ - amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true); - return 0; err_alloc_pts: - amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true); amdgpu_vm_bo_rmv(adev, bo_va_entry->bo_va); list_del(&bo_va_entry->bo_list); err_vmadd: @@ -809,24 +726,11 @@ static int unmap_bo_from_gpuvm(struct amdgpu_device *adev, { struct amdgpu_bo_va *bo_va = entry->bo_va; struct amdgpu_vm *vm = bo_va->base.vm; - struct amdgpu_bo *pd = vm->root.base.bo; - /* Remove eviction fence from PD (and thereby from PTs too as - * they share the resv. object). Otherwise during PT update - * job (see amdgpu_vm_bo_update_mapping), eviction fence would - * get added to job->sync object and job execution would - * trigger the eviction fence. - */ - amdgpu_amdkfd_remove_eviction_fence(pd, - vm->process_info->eviction_fence, - NULL, NULL); amdgpu_vm_bo_unmap(adev, bo_va, entry->va); amdgpu_vm_clear_freed(adev, vm, &bo_va->last_pt_update); - /* Add the eviction fence back */ - amdgpu_bo_fence(pd, &vm->process_info->eviction_fence->base, true); - amdgpu_sync_fence(NULL, sync, bo_va->last_pt_update, false); return 0; @@ -1389,8 +1293,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu( * attached */ amdgpu_amdkfd_remove_eviction_fence(mem->bo, - process_info->eviction_fence, - NULL, NULL); + process_info->eviction_fence); pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va, mem->va + bo_size * (1 + mem->aql_queue)); @@ -1617,8 +1520,7 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu( if (mem->mapped_to_gpu_memory == 0 && !amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm) && !mem->bo->pin_count) amdgpu_amdkfd_remove_eviction_fence(mem->bo, - process_info->eviction_fence, - NULL, NULL); + process_info->eviction_fence); unreserve_out: unreserve_bo_and_vms(&ctx, false, false); @@ -1679,7 +1581,7 @@ int amdgpu_amdkfd_gpuvm_map_gtt_bo_to_kernel(struct kgd_dev *kgd, } amdgpu_amdkfd_remove_eviction_fence( - bo, mem->process_info->eviction_fence, NULL, NULL); + bo, mem->process_info->eviction_fence); list_del_init(&mem->validate_list.head); if (size) @@ -1945,16 +1847,6 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) amdgpu_sync_create(&sync); - /* Avoid triggering eviction fences when unmapping invalid - * userptr BOs (waits for all fences, doesn't use - * FENCE_OWNER_VM) - */ - list_for_each_entry(peer_vm, &process_info->vm_list_head, - vm_list_node) - amdgpu_amdkfd_remove_eviction_fence(peer_vm->root.base.bo, - process_info->eviction_fence, - NULL, NULL); - ret = process_validate_vms(process_info); if (ret) goto unreserve_out; @@ -2015,10 +1907,6 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info) ret = process_update_pds(process_info, &sync); unreserve_out: - list_for_each_entry(peer_vm, &process_info->vm_list_head, - vm_list_node) - amdgpu_bo_fence(peer_vm->root.base.bo, - &process_info->eviction_fence->base, true); ttm_eu_backoff_reservation(&ticket, &resv_list); amdgpu_sync_wait(&sync, false); amdgpu_sync_free(&sync); -- cgit v1.2.3 From 74b9b3ea0cb355d668eb0309d1ab57f403d52c69 Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Tue, 5 Feb 2019 15:17:40 -0500 Subject: drm/amdgpu: Fix bugs in setting CP RB/MEC DOORBELL_RANGE registers CP_RB_DOORBELL_RANGE_LOWER/UPPER and CP_MEC_DOORBELL_RANGE_LOWER/UPPER are used for waking up an idle scheduler and for power gating support. Usually the first few doorbells in pci doorbell bar are used for RB and all leftover for MEC. This patch fixes the incorrect settings. Theoretically, gfx ring doorbells should come before all MEC doorbells to be consistent with the design. However, since the doorbell allocations are agreed by all and we are not free to change them, also considering the kernel MEC ring doorbells which are before gfx ring doorbells are not used often, we compromise by leaving the doorbell allocations unchanged. Signed-off-by: Yong Zhao Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 19 +++++++++++++++---- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 19 ++++++++++++++----- 2 files changed, 29 insertions(+), 9 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index b8e50a34bdb3..41cf9d0224d9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -4223,8 +4223,8 @@ static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu adev->doorbell_index.gfx_ring0); WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp); - WREG32(mmCP_RB_DOORBELL_RANGE_UPPER, - CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); + /* There is only one GFX queue */ + WREG32(mmCP_RB_DOORBELL_RANGE_UPPER, tmp); } static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev) @@ -4646,8 +4646,19 @@ static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring) static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev) { if (adev->asic_type > CHIP_TONGA) { - WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2); - WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2); + /* The first few doorbells in pci doorbell bar are for GFX RB + * rings and all the leftover for MEC. + * So CP_MEC_DOORBELL_RANGE_LOWER should be set one index after + * CP_RB_DOORBELL_RANGE_UPPER, as we assume there is only one + * GFX RB rings. + */ + u32 tmp = REG_SET_FIELD(0, CP_MEC_DOORBELL_RANGE_LOWER, + DOORBELL_RANGE_LOWER, + adev->gfx.gfx_ring[0].doorbell_index + 1); + + WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, tmp); + WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, + CP_MEC_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); } /* enable doorbells */ WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 5533f6e4f4a4..36f417fd6ba9 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -2631,8 +2631,8 @@ static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev) DOORBELL_RANGE_LOWER, ring->doorbell_index); WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp); - WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER, - CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); + /* There is only one GFX queue */ + WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER, tmp); /* start the ring */ @@ -2995,10 +2995,19 @@ static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring) /* enable the doorbell if requested */ if (ring->use_doorbell) { - WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, - (adev->doorbell_index.kiq * 2) << 2); + /* The first few doorbells in pci doorbell bar are for GFX RB + * rings and all the leftover for MEC. + * So CP_MEC_DOORBELL_RANGE_LOWER should be set one index after + * CP_RB_DOORBELL_RANGE_UPPER, as we assume there is only one + * GFX RB rings. + */ + u32 tmp = REG_SET_FIELD(0, CP_MEC_DOORBELL_RANGE_LOWER, + DOORBELL_RANGE_LOWER, + adev->gfx.gfx_ring[0].doorbell_index + 2); + + WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, tmp); WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER, - (adev->doorbell_index.userqueue_end * 2) << 2); + CP_MEC_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); } WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, -- cgit v1.2.3 From c0d9271ecbd891cdeb0fad1edcdd99ee717a655f Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Fri, 1 Feb 2019 18:36:21 -0500 Subject: drm/amdgpu: Delete user queue doorbell variables They are no longer used, so delete them to avoid confusion. Signed-off-by: Yong Zhao Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h | 8 -------- drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c | 2 -- drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c | 2 -- 3 files changed, 12 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h index 1cfec06f81d4..43546500ec26 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h @@ -48,8 +48,6 @@ struct amdgpu_doorbell_index { uint32_t mec_ring5; uint32_t mec_ring6; uint32_t mec_ring7; - uint32_t userqueue_start; - uint32_t userqueue_end; uint32_t gfx_ring0; uint32_t sdma_engine[8]; uint32_t ih; @@ -111,8 +109,6 @@ typedef enum _AMDGPU_VEGA20_DOORBELL_ASSIGNMENT AMDGPU_VEGA20_DOORBELL_MEC_RING5 = 0x008, AMDGPU_VEGA20_DOORBELL_MEC_RING6 = 0x009, AMDGPU_VEGA20_DOORBELL_MEC_RING7 = 0x00A, - AMDGPU_VEGA20_DOORBELL_USERQUEUE_START = 0x00B, - AMDGPU_VEGA20_DOORBELL_USERQUEUE_END = 0x08A, AMDGPU_VEGA20_DOORBELL_GFX_RING0 = 0x08B, /* SDMA:256~335*/ AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0 = 0x100, @@ -176,10 +172,6 @@ typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT AMDGPU_DOORBELL64_MEC_RING6 = 0x09, AMDGPU_DOORBELL64_MEC_RING7 = 0x0a, - /* User queue doorbell range (128 doorbells) */ - AMDGPU_DOORBELL64_USERQUEUE_START = 0x0b, - AMDGPU_DOORBELL64_USERQUEUE_END = 0x8a, - /* Graphics engine */ AMDGPU_DOORBELL64_GFX_RING0 = 0x8b, diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c index 4b5d60ea3e78..62f49c895314 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c @@ -67,8 +67,6 @@ void vega10_doorbell_index_init(struct amdgpu_device *adev) adev->doorbell_index.mec_ring5 = AMDGPU_DOORBELL64_MEC_RING5; adev->doorbell_index.mec_ring6 = AMDGPU_DOORBELL64_MEC_RING6; adev->doorbell_index.mec_ring7 = AMDGPU_DOORBELL64_MEC_RING7; - adev->doorbell_index.userqueue_start = AMDGPU_DOORBELL64_USERQUEUE_START; - adev->doorbell_index.userqueue_end = AMDGPU_DOORBELL64_USERQUEUE_END; adev->doorbell_index.gfx_ring0 = AMDGPU_DOORBELL64_GFX_RING0; adev->doorbell_index.sdma_engine[0] = AMDGPU_DOORBELL64_sDMA_ENGINE0; adev->doorbell_index.sdma_engine[1] = AMDGPU_DOORBELL64_sDMA_ENGINE1; diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c index 53716c593b2b..1271e1702ad4 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c @@ -65,8 +65,6 @@ void vega20_doorbell_index_init(struct amdgpu_device *adev) adev->doorbell_index.mec_ring5 = AMDGPU_VEGA20_DOORBELL_MEC_RING5; adev->doorbell_index.mec_ring6 = AMDGPU_VEGA20_DOORBELL_MEC_RING6; adev->doorbell_index.mec_ring7 = AMDGPU_VEGA20_DOORBELL_MEC_RING7; - adev->doorbell_index.userqueue_start = AMDGPU_VEGA20_DOORBELL_USERQUEUE_START; - adev->doorbell_index.userqueue_end = AMDGPU_VEGA20_DOORBELL_USERQUEUE_END; adev->doorbell_index.gfx_ring0 = AMDGPU_VEGA20_DOORBELL_GFX_RING0; adev->doorbell_index.sdma_engine[0] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0; adev->doorbell_index.sdma_engine[1] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1; -- cgit v1.2.3 From 9b49c19766a854feefa7fe5288c1915658a939f5 Mon Sep 17 00:00:00 2001 From: Colin Ian King Date: Tue, 12 Feb 2019 14:05:08 +0000 Subject: drm/amdgpu: fix several indentation issues MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit There are several statements that are incorrectly indented. Fix these. Reviewed-by: Christian König Signed-off-by: Colin Ian King Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 2 +- drivers/gpu/drm/amd/amdgpu/dce_v6_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c | 2 +- drivers/gpu/drm/amd/amdgpu/si.c | 2 +- drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c | 2 +- drivers/gpu/drm/amd/powerplay/hwmgr/vega20_hwmgr.c | 2 +- 6 files changed, 6 insertions(+), 6 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c index bc62bf41b7e9..b65e18101108 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c @@ -207,7 +207,7 @@ int amdgpu_driver_load_kms(struct drm_device *dev, unsigned long flags) if (!r) { acpi_status = amdgpu_acpi_init(adev); if (acpi_status) - dev_dbg(&dev->pdev->dev, + dev_dbg(&dev->pdev->dev, "Error during ACPI methods call\n"); } diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c index db443ec53d3a..bea32f076b91 100644 --- a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c @@ -2980,7 +2980,7 @@ static int dce_v6_0_pageflip_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) { - unsigned long flags; + unsigned long flags; unsigned crtc_id; struct amdgpu_crtc *amdgpu_crtc; struct amdgpu_flip_work *works; diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c b/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c index e347b407bd03..727370ccd854 100644 --- a/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c +++ b/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c @@ -32,7 +32,7 @@ static u32 nbio_v7_4_get_rev_id(struct amdgpu_device *adev) { - u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0); + u32 tmp = RREG32_SOC15(NBIO, 0, mmRCC_DEV0_EPF0_STRAP0); tmp &= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0_MASK; tmp >>= RCC_DEV0_EPF0_STRAP0__STRAP_ATI_REV_ID_DEV0_F0__SHIFT; diff --git a/drivers/gpu/drm/amd/amdgpu/si.c b/drivers/gpu/drm/amd/amdgpu/si.c index 79c1a9bbcc21..9d8df68893b9 100644 --- a/drivers/gpu/drm/amd/amdgpu/si.c +++ b/drivers/gpu/drm/amd/amdgpu/si.c @@ -1436,7 +1436,7 @@ static int si_common_early_init(void *handle) AMD_CG_SUPPORT_UVD_MGCG | AMD_CG_SUPPORT_HDP_LS | AMD_CG_SUPPORT_HDP_MGCG; - adev->pg_flags = 0; + adev->pg_flags = 0; adev->external_rev_id = (adev->rev_id == 0) ? 1 : (adev->rev_id == 1) ? 5 : 6; break; diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c b/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c index d138ddae563d..58f5589aaf12 100644 --- a/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c +++ b/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c @@ -1211,7 +1211,7 @@ int smu7_power_control_set_level(struct pp_hwmgr *hwmgr) hwmgr->platform_descriptor.TDPAdjustment : (-1 * hwmgr->platform_descriptor.TDPAdjustment); - if (hwmgr->chip_id > CHIP_TONGA) + if (hwmgr->chip_id > CHIP_TONGA) target_tdp = ((100 + adjust_percent) * (int)(cac_table->usTDP * 256)) / 100; else target_tdp = ((100 + adjust_percent) * (int)(cac_table->usConfigurableTDP * 256)) / 100; diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/vega20_hwmgr.c b/drivers/gpu/drm/amd/powerplay/hwmgr/vega20_hwmgr.c index 0769b1ec562b..aad79affb081 100644 --- a/drivers/gpu/drm/amd/powerplay/hwmgr/vega20_hwmgr.c +++ b/drivers/gpu/drm/amd/powerplay/hwmgr/vega20_hwmgr.c @@ -3456,7 +3456,7 @@ static int vega20_apply_clocks_adjust_rules(struct pp_hwmgr *hwmgr) disable_mclk_switching = ((1 < hwmgr->display_config->num_display) && !hwmgr->display_config->multi_monitor_in_sync) || vblank_too_short; - latency = hwmgr->display_config->dce_tolerable_mclk_in_active_latency; + latency = hwmgr->display_config->dce_tolerable_mclk_in_active_latency; /* gfxclk */ dpm_table = &(data->dpm_table.gfx_table); -- cgit v1.2.3 From 7a5e0d9ab6d9564f5f82d7a16355118aadd25ef8 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Mon, 11 Feb 2019 16:49:47 -0500 Subject: drm/amdgpu: don't clamp debugfs register access to the BAR size MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This prevents us from accessing extended registers in tools like umr. The register access functions already check if the offset is beyond the BAR size and use the indirect accessors with locking so this is safe. Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 3 --- 1 file changed, 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c index dd9a4fb9ce39..4ae3ff9a1d4c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c @@ -158,9 +158,6 @@ static int amdgpu_debugfs_process_reg_op(bool read, struct file *f, while (size) { uint32_t value; - if (*pos > adev->rmmio_size) - goto end; - if (read) { value = RREG32(*pos >> 2); r = put_user(value, (uint32_t *)buf); -- cgit v1.2.3 From 2a84e48e9712ea8591a10dd59d59ccab3d54efd6 Mon Sep 17 00:00:00 2001 From: Bas Nieuwenhuizen Date: Wed, 30 Jan 2019 02:53:20 +0100 Subject: drm/amdgpu: Only add rqs for initialized rings. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit I don't see another way to figure out if a ring is initialized if the hardware block might not be initialized. Entities have been fixed up to handle num_rqs = 0. Signed-off-by: Bas Nieuwenhuizen Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index d85184b5b35c..7b526593eb77 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -124,6 +124,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ring *rings[AMDGPU_MAX_RINGS]; struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS]; unsigned num_rings; + unsigned num_rqs = 0; switch (i) { case AMDGPU_HW_IP_GFX: @@ -166,12 +167,16 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, break; } - for (j = 0; j < num_rings; ++j) - rqs[j] = &rings[j]->sched.sched_rq[priority]; + for (j = 0; j < num_rings; ++j) { + if (!rings[j]->adev) + continue; + + rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority]; + } for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) r = drm_sched_entity_init(&ctx->entities[i][j].entity, - rqs, num_rings, &ctx->guilty); + rqs, num_rqs, &ctx->guilty); if (r) goto error_cleanup_entities; } -- cgit v1.2.3 From 021830d24ba55a578f602979274965344c8e6284 Mon Sep 17 00:00:00 2001 From: Bas Nieuwenhuizen Date: Wed, 30 Jan 2019 02:53:21 +0100 Subject: drm/amdgpu: Check if fd really is an amdgpu fd. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Otherwise we interpret the file private data as drm & amdgpu data while it might not be, possibly allowing one to get memory corruption. Signed-off-by: Bas Nieuwenhuizen Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 16 ++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 10 +++++++--- 3 files changed, 25 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index 9efa681d0878..8d0d7f3dd5fb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -411,6 +411,8 @@ struct amdgpu_fpriv { struct amdgpu_ctx_mgr ctx_mgr; }; +int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv); + int amdgpu_ib_get(struct amdgpu_device *adev, struct amdgpu_vm *vm, unsigned size, struct amdgpu_ib *ib); void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 7f3aa7b7e1d8..637f927e3675 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -1178,6 +1178,22 @@ static const struct file_operations amdgpu_driver_kms_fops = { #endif }; +int amdgpu_file_to_fpriv(struct file *filp, struct amdgpu_fpriv **fpriv) +{ + struct drm_file *file; + + if (!filp) + return -EINVAL; + + if (filp->f_op != &amdgpu_driver_kms_fops) { + return -EINVAL; + } + + file = filp->private_data; + *fpriv = file->driver_priv; + return 0; +} + static bool amdgpu_get_crtc_scanout_position(struct drm_device *dev, unsigned int pipe, bool in_vblank_irq, int *vpos, int *hpos, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 1cafe8d83a4d..0b70410488b6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -54,16 +54,20 @@ static int amdgpu_sched_process_priority_override(struct amdgpu_device *adev, enum drm_sched_priority priority) { struct file *filp = fget(fd); - struct drm_file *file; struct amdgpu_fpriv *fpriv; struct amdgpu_ctx *ctx; uint32_t id; + int r; if (!filp) return -EINVAL; - file = filp->private_data; - fpriv = file->driver_priv; + r = amdgpu_file_to_fpriv(filp, &fpriv); + if (r) { + fput(filp); + return r; + } + idr_for_each_entry(&fpriv->ctx_mgr.ctx_handles, ctx, id) amdgpu_ctx_priority_override(ctx, priority); -- cgit v1.2.3 From b5bb37eddb63b16b7ab959598d108b1c444be77d Mon Sep 17 00:00:00 2001 From: Bas Nieuwenhuizen Date: Wed, 30 Jan 2019 02:53:22 +0100 Subject: drm/amdgpu: Add command to override the context priority. MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Given a master fd we can then override the priority of the context in another fd. Using these overrides was recommended by Christian instead of trying to submit from a master fd, and I am adding a way to override a single context instead of the entire process so we can only upgrade a single Vulkan queue and not effectively the entire process. Reused the flags field as it was checked to be 0 anyways, so nothing used it. This is source-incompatible (due to the name change), but ABI compatible. Signed-off-by: Bas Nieuwenhuizen Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 41 ++++++++++++++++++++++++++++++- include/uapi/drm/amdgpu_drm.h | 3 ++- 2 files changed, 42 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c index 0b70410488b6..0767a93e4d91 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c @@ -76,6 +76,39 @@ static int amdgpu_sched_process_priority_override(struct amdgpu_device *adev, return 0; } +static int amdgpu_sched_context_priority_override(struct amdgpu_device *adev, + int fd, + unsigned ctx_id, + enum drm_sched_priority priority) +{ + struct file *filp = fget(fd); + struct amdgpu_fpriv *fpriv; + struct amdgpu_ctx *ctx; + int r; + + if (!filp) + return -EINVAL; + + r = amdgpu_file_to_fpriv(filp, &fpriv); + if (r) { + fput(filp); + return r; + } + + ctx = amdgpu_ctx_get(fpriv, ctx_id); + + if (!ctx) { + fput(filp); + return -EINVAL; + } + + amdgpu_ctx_priority_override(ctx, priority); + amdgpu_ctx_put(ctx); + fput(filp); + + return 0; +} + int amdgpu_sched_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { @@ -85,7 +118,7 @@ int amdgpu_sched_ioctl(struct drm_device *dev, void *data, int r; priority = amdgpu_to_sched_priority(args->in.priority); - if (args->in.flags || priority == DRM_SCHED_PRIORITY_INVALID) + if (priority == DRM_SCHED_PRIORITY_INVALID) return -EINVAL; switch (args->in.op) { @@ -94,6 +127,12 @@ int amdgpu_sched_ioctl(struct drm_device *dev, void *data, args->in.fd, priority); break; + case AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE: + r = amdgpu_sched_context_priority_override(adev, + args->in.fd, + args->in.ctx_id, + priority); + break; default: DRM_ERROR("Invalid sched op specified: %d\n", args->in.op); r = -EINVAL; diff --git a/include/uapi/drm/amdgpu_drm.h b/include/uapi/drm/amdgpu_drm.h index 2acbc8bc465b..4a53f6cfa034 100644 --- a/include/uapi/drm/amdgpu_drm.h +++ b/include/uapi/drm/amdgpu_drm.h @@ -272,13 +272,14 @@ union drm_amdgpu_vm { /* sched ioctl */ #define AMDGPU_SCHED_OP_PROCESS_PRIORITY_OVERRIDE 1 +#define AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE 2 struct drm_amdgpu_sched_in { /* AMDGPU_SCHED_OP_* */ __u32 op; __u32 fd; __s32 priority; - __u32 flags; + __u32 ctx_id; }; union drm_amdgpu_sched { -- cgit v1.2.3 From 478168e1cbd0b3df52ca223e1d7af3c257588166 Mon Sep 17 00:00:00 2001 From: Alex Deucher Date: Thu, 14 Feb 2019 15:54:57 -0500 Subject: drm/amdgpu: remove some old unused dpm helpers MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Carried over from radeon, but no longer used. Reviewed-by: Michel Dänzer Acked-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c | 88 --------------------------------- drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h | 9 ---- 2 files changed, 97 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c index 1c4595562f8f..344967df3137 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.c @@ -184,61 +184,6 @@ u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev) return vrefresh; } -void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b, - u32 *p, u32 *u) -{ - u32 b_c = 0; - u32 i_c; - u32 tmp; - - i_c = (i * r_c) / 100; - tmp = i_c >> p_b; - - while (tmp) { - b_c++; - tmp >>= 1; - } - - *u = (b_c + 1) / 2; - *p = i_c / (1 << (2 * (*u))); -} - -int amdgpu_calculate_at(u32 t, u32 h, u32 fh, u32 fl, u32 *tl, u32 *th) -{ - u32 k, a, ah, al; - u32 t1; - - if ((fl == 0) || (fh == 0) || (fl > fh)) - return -EINVAL; - - k = (100 * fh) / fl; - t1 = (t * (k - 100)); - a = (1000 * (100 * h + t1)) / (10000 + (t1 / 100)); - a = (a + 5) / 10; - ah = ((a * t) + 5000) / 10000; - al = a - ah; - - *th = t - ah; - *tl = t + al; - - return 0; -} - -bool amdgpu_is_uvd_state(u32 class, u32 class2) -{ - if (class & ATOM_PPLIB_CLASSIFICATION_UVDSTATE) - return true; - if (class & ATOM_PPLIB_CLASSIFICATION_HD2STATE) - return true; - if (class & ATOM_PPLIB_CLASSIFICATION_HDSTATE) - return true; - if (class & ATOM_PPLIB_CLASSIFICATION_SDSTATE) - return true; - if (class2 & ATOM_PPLIB_CLASSIFICATION2_MVC) - return true; - return false; -} - bool amdgpu_is_internal_thermal_sensor(enum amdgpu_int_thermal_type sensor) { switch (sensor) { @@ -949,39 +894,6 @@ enum amdgpu_pcie_gen amdgpu_get_pcie_gen_support(struct amdgpu_device *adev, return AMDGPU_PCIE_GEN1; } -u16 amdgpu_get_pcie_lane_support(struct amdgpu_device *adev, - u16 asic_lanes, - u16 default_lanes) -{ - switch (asic_lanes) { - case 0: - default: - return default_lanes; - case 1: - return 1; - case 2: - return 2; - case 4: - return 4; - case 8: - return 8; - case 12: - return 12; - case 16: - return 16; - } -} - -u8 amdgpu_encode_pci_lane_width(u32 lanes) -{ - u8 encoded_lanes[] = { 0, 1, 2, 0, 3, 0, 0, 0, 4, 0, 0, 0, 5, 0, 0, 0, 6 }; - - if (lanes > 16) - return 0; - - return encoded_lanes[lanes]; -} - struct amd_vce_state* amdgpu_get_vce_clock_state(void *handle, u32 idx) { diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h index 2f61e9edb1c1..e871e022c129 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_dpm.h @@ -486,10 +486,6 @@ void amdgpu_dpm_print_ps_status(struct amdgpu_device *adev, u32 amdgpu_dpm_get_vblank_time(struct amdgpu_device *adev); u32 amdgpu_dpm_get_vrefresh(struct amdgpu_device *adev); void amdgpu_dpm_get_active_displays(struct amdgpu_device *adev); -bool amdgpu_is_uvd_state(u32 class, u32 class2); -void amdgpu_calculate_u_and_p(u32 i, u32 r_c, u32 p_b, - u32 *p, u32 *u); -int amdgpu_calculate_at(u32 t, u32 h, u32 fh, u32 fl, u32 *tl, u32 *th); bool amdgpu_is_internal_thermal_sensor(enum amdgpu_int_thermal_type sensor); @@ -505,11 +501,6 @@ enum amdgpu_pcie_gen amdgpu_get_pcie_gen_support(struct amdgpu_device *adev, enum amdgpu_pcie_gen asic_gen, enum amdgpu_pcie_gen default_gen); -u16 amdgpu_get_pcie_lane_support(struct amdgpu_device *adev, - u16 asic_lanes, - u16 default_lanes); -u8 amdgpu_encode_pci_lane_width(u32 lanes); - struct amd_vce_state* amdgpu_get_vce_clock_state(void *handle, u32 idx); -- cgit v1.2.3 From 828845b7c86c5338f6ca02aaaaf4b525718f31b2 Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Wed, 13 Feb 2019 13:13:50 -0500 Subject: drm/amdgpu: Add first_non_cp and last_non_cp in amdgpu_doorbell_index They will be used to inform KFD the doorbell range not usable for CP. Signed-off-by: Yong Zhao Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h | 9 +++++++++ drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c | 4 ++++ drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c | 4 ++++ 3 files changed, 17 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h index 43546500ec26..5587fac671bb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h @@ -69,6 +69,8 @@ struct amdgpu_doorbell_index { uint32_t vce_ring6_7; } uvd_vce; }; + uint32_t first_non_cp; + uint32_t last_non_cp; uint32_t max_assignment; /* Per engine SDMA doorbell size in dword */ uint32_t sdma_doorbell_range; @@ -139,6 +141,10 @@ typedef enum _AMDGPU_VEGA20_DOORBELL_ASSIGNMENT AMDGPU_VEGA20_DOORBELL64_VCE_RING2_3 = 0x18D, AMDGPU_VEGA20_DOORBELL64_VCE_RING4_5 = 0x18E, AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7 = 0x18F, + + AMDGPU_VEGA20_DOORBELL64_FIRST_NON_CP = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0, + AMDGPU_VEGA20_DOORBELL64_LAST_NON_CP = AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7, + AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT = 0x18F, AMDGPU_VEGA20_DOORBELL_INVALID = 0xFFFF } AMDGPU_VEGA20_DOORBELL_ASSIGNMENT; @@ -214,6 +220,9 @@ typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT AMDGPU_DOORBELL64_VCE_RING4_5 = 0xFE, AMDGPU_DOORBELL64_VCE_RING6_7 = 0xFF, + AMDGPU_DOORBELL64_FIRST_NON_CP = AMDGPU_DOORBELL64_sDMA_ENGINE0, + AMDGPU_DOORBELL64_LAST_NON_CP = AMDGPU_DOORBELL64_VCE_RING6_7, + AMDGPU_DOORBELL64_MAX_ASSIGNMENT = 0xFF, AMDGPU_DOORBELL64_INVALID = 0xFFFF } AMDGPU_DOORBELL64_ASSIGNMENT; diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c index 62f49c895314..5e9e53143a8e 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c @@ -79,6 +79,10 @@ void vega10_doorbell_index_init(struct amdgpu_device *adev) adev->doorbell_index.uvd_vce.vce_ring2_3 = AMDGPU_DOORBELL64_VCE_RING2_3; adev->doorbell_index.uvd_vce.vce_ring4_5 = AMDGPU_DOORBELL64_VCE_RING4_5; adev->doorbell_index.uvd_vce.vce_ring6_7 = AMDGPU_DOORBELL64_VCE_RING6_7; + + adev->doorbell_index.first_non_cp = AMDGPU_DOORBELL64_FIRST_NON_CP; + adev->doorbell_index.last_non_cp = AMDGPU_DOORBELL64_LAST_NON_CP; + /* In unit of dword doorbell */ adev->doorbell_index.max_assignment = AMDGPU_DOORBELL64_MAX_ASSIGNMENT << 1; adev->doorbell_index.sdma_doorbell_range = 4; diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c index 1271e1702ad4..fb6398e38be9 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c @@ -83,6 +83,10 @@ void vega20_doorbell_index_init(struct amdgpu_device *adev) adev->doorbell_index.uvd_vce.vce_ring2_3 = AMDGPU_VEGA20_DOORBELL64_VCE_RING2_3; adev->doorbell_index.uvd_vce.vce_ring4_5 = AMDGPU_VEGA20_DOORBELL64_VCE_RING4_5; adev->doorbell_index.uvd_vce.vce_ring6_7 = AMDGPU_VEGA20_DOORBELL64_VCE_RING6_7; + + adev->doorbell_index.first_non_cp = AMDGPU_VEGA20_DOORBELL64_FIRST_NON_CP; + adev->doorbell_index.last_non_cp = AMDGPU_VEGA20_DOORBELL64_LAST_NON_CP; + adev->doorbell_index.max_assignment = AMDGPU_VEGA20_DOORBELL_MAX_ASSIGNMENT << 1; adev->doorbell_index.sdma_doorbell_range = 20; } -- cgit v1.2.3 From 1f86805adc3432e92e7d87e1ff5da9826ef56eab Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Wed, 13 Feb 2019 13:15:05 -0500 Subject: drm/amdkfd: Fix bugs regarding CP queue doorbell mask on SOC15 Reserved doorbells for SDMA IH and VCN were not properly masked out when allocating doorbells for CP user queues. This patch fixed that. Signed-off-by: Yong Zhao Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 16 ++++++++++++---- drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 11 +++++++++++ drivers/gpu/drm/amd/amdkfd/kfd_process.c | 14 +++++++++----- drivers/gpu/drm/amd/include/kgd_kfd_interface.h | 15 ++++++--------- 4 files changed, 38 insertions(+), 18 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index e957e42c539a..30e2b371578e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -196,11 +196,19 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) gpu_resources.sdma_doorbell[1][i+1] = adev->doorbell_index.sdma_engine[1] + 0x200 + (i >> 1); } - /* Doorbells 0x0e0-0ff and 0x2e0-2ff are reserved for - * SDMA, IH and VCN. So don't use them for the CP. + + /* Since SOC15, BIF starts to statically use the + * lower 12 bits of doorbell addresses for routing + * based on settings in registers like + * SDMA0_DOORBELL_RANGE etc.. + * In order to route a doorbell to CP engine, the lower + * 12 bits of its address has to be outside the range + * set for SDMA, VCN, and IH blocks. */ - gpu_resources.reserved_doorbell_mask = 0x1e0; - gpu_resources.reserved_doorbell_val = 0x0e0; + gpu_resources.non_cp_doorbells_start = + adev->doorbell_index.first_non_cp; + gpu_resources.non_cp_doorbells_end = + adev->doorbell_index.last_non_cp; kgd2kfd_device_init(adev->kfd.dev, &gpu_resources); } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h index e5ebcca7f031..0eeee3c6d6dc 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h @@ -103,6 +103,17 @@ #define KFD_KERNEL_QUEUE_SIZE 2048 +/* + * 512 = 0x200 + * The doorbell index distance between SDMA RLC (2*i) and (2*i+1) in the + * same SDMA engine on SOC15, which has 8-byte doorbells for SDMA. + * 512 8-byte doorbell distance (i.e. one page away) ensures that SDMA RLC + * (2*i+1) doorbells (in terms of the lower 12 bit address) lie exactly in + * the OFFSET and SIZE set in registers like BIF_SDMA0_DOORBELL_RANGE. + */ +#define KFD_QUEUE_DOORBELL_MIRROR_OFFSET 512 + + /* * Kernel module parameter to specify maximum number of supported queues per * device diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c index 80b36e860a0a..4bdae78bab8e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c @@ -607,13 +607,17 @@ static int init_doorbell_bitmap(struct qcm_process_device *qpd, if (!qpd->doorbell_bitmap) return -ENOMEM; - /* Mask out any reserved doorbells */ - for (i = 0; i < KFD_MAX_NUM_OF_QUEUES_PER_PROCESS; i++) - if ((dev->shared_resources.reserved_doorbell_mask & i) == - dev->shared_resources.reserved_doorbell_val) { + /* Mask out doorbells reserved for SDMA, IH, and VCN on SOC15. */ + for (i = 0; i < KFD_MAX_NUM_OF_QUEUES_PER_PROCESS / 2; i++) { + if (i >= dev->shared_resources.non_cp_doorbells_start + && i <= dev->shared_resources.non_cp_doorbells_end) { set_bit(i, qpd->doorbell_bitmap); - pr_debug("reserved doorbell 0x%03x\n", i); + set_bit(i + KFD_QUEUE_DOORBELL_MIRROR_OFFSET, + qpd->doorbell_bitmap); + pr_debug("reserved doorbell 0x%03x and 0x%03x\n", i, + i + KFD_QUEUE_DOORBELL_MIRROR_OFFSET); } + } return 0; } diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h index 83d960110d23..0b6b34f4e5a1 100644 --- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h +++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h @@ -140,17 +140,14 @@ struct kgd2kfd_shared_resources { /* Doorbell assignments (SOC15 and later chips only). Only * specific doorbells are routed to each SDMA engine. Others * are routed to IH and VCN. They are not usable by the CP. - * - * Any doorbell number D that satisfies the following condition - * is reserved: (D & reserved_doorbell_mask) == reserved_doorbell_val - * - * KFD currently uses 1024 (= 0x3ff) doorbells per process. If - * doorbells 0x0e0-0x0ff and 0x2e0-0x2ff are reserved, that means - * mask would be set to 0x1e0 and val set to 0x0e0. */ unsigned int sdma_doorbell[2][8]; - unsigned int reserved_doorbell_mask; - unsigned int reserved_doorbell_val; + + /* From SOC15 onward, the doorbell index range not usable for CP + * queues. + */ + uint32_t non_cp_doorbells_start; + uint32_t non_cp_doorbells_end; /* Base address of doorbell aperture. */ phys_addr_t doorbell_physical_address; -- cgit v1.2.3 From 234441dd49bcd917d0acd23290d1c49f332a816b Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Wed, 9 Jan 2019 23:31:14 -0500 Subject: drm/amdkfd: Optimize out sdma doorbell array in kgd2kfd_shared_resources We can directly calculate sdma doorbell indexes in the process doorbell pages through the doorbell_index structure in amdgpu_device, so no need to cache them in kgd2kfd_shared_resources any more. This alleviates the adaptation needs when new SDMA configurations are introduced. Signed-off-by: Yong Zhao Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 41 ++++++---------------- .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 16 ++++++--- drivers/gpu/drm/amd/include/kgd_kfd_interface.h | 4 +-- 3 files changed, 23 insertions(+), 38 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c index 30e2b371578e..fe1d7368c1e6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c @@ -131,7 +131,7 @@ static void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev, void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) { - int i, n; + int i; int last_valid_bit; if (adev->kfd.dev) { @@ -142,7 +142,9 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) .gpuvm_size = min(adev->vm_manager.max_pfn << AMDGPU_GPU_PAGE_SHIFT, AMDGPU_GMC_HOLE_START), - .drm_render_minor = adev->ddev->render->index + .drm_render_minor = adev->ddev->render->index, + .sdma_doorbell_idx = adev->doorbell_index.sdma_engine, + }; /* this is going to have a few of the MSBs set that we need to @@ -172,31 +174,6 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) &gpu_resources.doorbell_aperture_size, &gpu_resources.doorbell_start_offset); - if (adev->asic_type < CHIP_VEGA10) { - kgd2kfd_device_init(adev->kfd.dev, &gpu_resources); - return; - } - - n = (adev->asic_type < CHIP_VEGA20) ? 2 : 8; - - for (i = 0; i < n; i += 2) { - /* On SOC15 the BIF is involved in routing - * doorbells using the low 12 bits of the - * address. Communicate the assignments to - * KFD. KFD uses two doorbell pages per - * process in case of 64-bit doorbells so we - * can use each doorbell assignment twice. - */ - gpu_resources.sdma_doorbell[0][i] = - adev->doorbell_index.sdma_engine[0] + (i >> 1); - gpu_resources.sdma_doorbell[0][i+1] = - adev->doorbell_index.sdma_engine[0] + 0x200 + (i >> 1); - gpu_resources.sdma_doorbell[1][i] = - adev->doorbell_index.sdma_engine[1] + (i >> 1); - gpu_resources.sdma_doorbell[1][i+1] = - adev->doorbell_index.sdma_engine[1] + 0x200 + (i >> 1); - } - /* Since SOC15, BIF starts to statically use the * lower 12 bits of doorbell addresses for routing * based on settings in registers like @@ -205,10 +182,12 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev) * 12 bits of its address has to be outside the range * set for SDMA, VCN, and IH blocks. */ - gpu_resources.non_cp_doorbells_start = - adev->doorbell_index.first_non_cp; - gpu_resources.non_cp_doorbells_end = - adev->doorbell_index.last_non_cp; + if (adev->asic_type >= CHIP_VEGA10) { + gpu_resources.non_cp_doorbells_start = + adev->doorbell_index.first_non_cp; + gpu_resources.non_cp_doorbells_end = + adev->doorbell_index.last_non_cp; + } kgd2kfd_device_init(adev->kfd.dev, &gpu_resources); } diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c index 8372556b52eb..c6c9530e704e 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -134,12 +134,18 @@ static int allocate_doorbell(struct qcm_process_device *qpd, struct queue *q) */ q->doorbell_id = q->properties.queue_id; } else if (q->properties.type == KFD_QUEUE_TYPE_SDMA) { - /* For SDMA queues on SOC15, use static doorbell - * assignments based on the engine and queue. + /* For SDMA queues on SOC15 with 8-byte doorbell, use static + * doorbell assignments based on the engine and queue id. + * The doobell index distance between RLC (2*i) and (2*i+1) + * for a SDMA engine is 512. */ - q->doorbell_id = dev->shared_resources.sdma_doorbell - [q->properties.sdma_engine_id] - [q->properties.sdma_queue_id]; + uint32_t *idx_offset = + dev->shared_resources.sdma_doorbell_idx; + + q->doorbell_id = idx_offset[q->properties.sdma_engine_id] + + (q->properties.sdma_queue_id & 1) + * KFD_QUEUE_DOORBELL_MIRROR_OFFSET + + (q->properties.sdma_queue_id >> 1); } else { /* For CP queues on SOC15 reserve a free doorbell ID */ unsigned int found; diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h index 0b6b34f4e5a1..5f3c10ebff08 100644 --- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h +++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h @@ -137,11 +137,11 @@ struct kgd2kfd_shared_resources { /* Bit n == 1 means Queue n is available for KFD */ DECLARE_BITMAP(queue_bitmap, KGD_MAX_QUEUES); - /* Doorbell assignments (SOC15 and later chips only). Only + /* SDMA doorbell assignments (SOC15 and later chips only). Only * specific doorbells are routed to each SDMA engine. Others * are routed to IH and VCN. They are not usable by the CP. */ - unsigned int sdma_doorbell[2][8]; + uint32_t *sdma_doorbell_idx; /* From SOC15 onward, the doorbell index range not usable for CP * queues. -- cgit v1.2.3 From 2f54a0412f31c1312a105d6c02adba9b2ae41c64 Mon Sep 17 00:00:00 2001 From: "Gustavo A. R. Silva" Date: Fri, 15 Feb 2019 10:46:43 -0600 Subject: drm/amdgpu/si_dpm: Mark expected switch fall-throughs In preparation to enabling -Wimplicit-fallthrough, mark switch cases where we are expecting to fall through. Warning level 3 was used: -Wimplicit-fallthrough=3 This patch is part of the ongoing efforts to enable -Wimplicit-fallthrough. Signed-off-by: Gustavo A. R. Silva Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/si_dpm.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/si_dpm.c b/drivers/gpu/drm/amd/amdgpu/si_dpm.c index da58040fdbdc..41e01a7f57a4 100644 --- a/drivers/gpu/drm/amd/amdgpu/si_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/si_dpm.c @@ -6216,10 +6216,12 @@ static void si_request_link_speed_change_before_state_change(struct amdgpu_devic si_pi->force_pcie_gen = AMDGPU_PCIE_GEN2; if (current_link_speed == AMDGPU_PCIE_GEN2) break; + /* fall through */ case AMDGPU_PCIE_GEN2: if (amdgpu_acpi_pcie_performance_request(adev, PCIE_PERF_REQ_PECI_GEN2, false) == 0) break; #endif + /* fall through */ default: si_pi->force_pcie_gen = si_get_current_pcie_speed(adev); break; -- cgit v1.2.3 From e02c80d60c606ea9810565421e4e70c3996e3df7 Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Tue, 19 Feb 2019 11:21:41 -0500 Subject: Revert "drm/amdgpu: Delete user queue doorbell variables" This reverts commit 9006c6bd9059cb9807fa863bafc1d776222cb61b. Signed-off-by: Yong Zhao Acked-by: Alex Deucher Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h | 8 ++++++++ drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c | 2 ++ drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c | 2 ++ 3 files changed, 12 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h index 5587fac671bb..68959b923f89 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_doorbell.h @@ -48,6 +48,8 @@ struct amdgpu_doorbell_index { uint32_t mec_ring5; uint32_t mec_ring6; uint32_t mec_ring7; + uint32_t userqueue_start; + uint32_t userqueue_end; uint32_t gfx_ring0; uint32_t sdma_engine[8]; uint32_t ih; @@ -111,6 +113,8 @@ typedef enum _AMDGPU_VEGA20_DOORBELL_ASSIGNMENT AMDGPU_VEGA20_DOORBELL_MEC_RING5 = 0x008, AMDGPU_VEGA20_DOORBELL_MEC_RING6 = 0x009, AMDGPU_VEGA20_DOORBELL_MEC_RING7 = 0x00A, + AMDGPU_VEGA20_DOORBELL_USERQUEUE_START = 0x00B, + AMDGPU_VEGA20_DOORBELL_USERQUEUE_END = 0x08A, AMDGPU_VEGA20_DOORBELL_GFX_RING0 = 0x08B, /* SDMA:256~335*/ AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0 = 0x100, @@ -178,6 +182,10 @@ typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT AMDGPU_DOORBELL64_MEC_RING6 = 0x09, AMDGPU_DOORBELL64_MEC_RING7 = 0x0a, + /* User queue doorbell range (128 doorbells) */ + AMDGPU_DOORBELL64_USERQUEUE_START = 0x0b, + AMDGPU_DOORBELL64_USERQUEUE_END = 0x8a, + /* Graphics engine */ AMDGPU_DOORBELL64_GFX_RING0 = 0x8b, diff --git a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c index 5e9e53143a8e..a8e92638a2e8 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega10_reg_init.c @@ -67,6 +67,8 @@ void vega10_doorbell_index_init(struct amdgpu_device *adev) adev->doorbell_index.mec_ring5 = AMDGPU_DOORBELL64_MEC_RING5; adev->doorbell_index.mec_ring6 = AMDGPU_DOORBELL64_MEC_RING6; adev->doorbell_index.mec_ring7 = AMDGPU_DOORBELL64_MEC_RING7; + adev->doorbell_index.userqueue_start = AMDGPU_DOORBELL64_USERQUEUE_START; + adev->doorbell_index.userqueue_end = AMDGPU_DOORBELL64_USERQUEUE_END; adev->doorbell_index.gfx_ring0 = AMDGPU_DOORBELL64_GFX_RING0; adev->doorbell_index.sdma_engine[0] = AMDGPU_DOORBELL64_sDMA_ENGINE0; adev->doorbell_index.sdma_engine[1] = AMDGPU_DOORBELL64_sDMA_ENGINE1; diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c index fb6398e38be9..0db84386252a 100644 --- a/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c +++ b/drivers/gpu/drm/amd/amdgpu/vega20_reg_init.c @@ -65,6 +65,8 @@ void vega20_doorbell_index_init(struct amdgpu_device *adev) adev->doorbell_index.mec_ring5 = AMDGPU_VEGA20_DOORBELL_MEC_RING5; adev->doorbell_index.mec_ring6 = AMDGPU_VEGA20_DOORBELL_MEC_RING6; adev->doorbell_index.mec_ring7 = AMDGPU_VEGA20_DOORBELL_MEC_RING7; + adev->doorbell_index.userqueue_start = AMDGPU_VEGA20_DOORBELL_USERQUEUE_START; + adev->doorbell_index.userqueue_end = AMDGPU_VEGA20_DOORBELL_USERQUEUE_END; adev->doorbell_index.gfx_ring0 = AMDGPU_VEGA20_DOORBELL_GFX_RING0; adev->doorbell_index.sdma_engine[0] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE0; adev->doorbell_index.sdma_engine[1] = AMDGPU_VEGA20_DOORBELL_sDMA_ENGINE1; -- cgit v1.2.3 From a4a3ad3523359923dc93dfbbb13a89f2b8d5589a Mon Sep 17 00:00:00 2001 From: Yong Zhao Date: Tue, 19 Feb 2019 11:21:51 -0500 Subject: Revert "drm/amdgpu: Fix bugs in setting CP RB/MEC DOORBELL_RANGE registers" The original change caused a regression, so revert it until the new fix is ready. BUG: https://bugs.freedesktop.org/show_bug.cgi?id=109650 This reverts commit 764c85fef41722db0f21558c6c2fb38bee172d19. Signed-off-by: Yong Zhao Acked-by: Alex Deucher Reviewed-by: Felix Kuehling Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 19 ++++--------------- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 19 +++++-------------- 2 files changed, 9 insertions(+), 29 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 41cf9d0224d9..b8e50a34bdb3 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -4223,8 +4223,8 @@ static void gfx_v8_0_set_cpg_door_bell(struct amdgpu_device *adev, struct amdgpu adev->doorbell_index.gfx_ring0); WREG32(mmCP_RB_DOORBELL_RANGE_LOWER, tmp); - /* There is only one GFX queue */ - WREG32(mmCP_RB_DOORBELL_RANGE_UPPER, tmp); + WREG32(mmCP_RB_DOORBELL_RANGE_UPPER, + CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); } static int gfx_v8_0_cp_gfx_resume(struct amdgpu_device *adev) @@ -4646,19 +4646,8 @@ static int gfx_v8_0_kcq_init_queue(struct amdgpu_ring *ring) static void gfx_v8_0_set_mec_doorbell_range(struct amdgpu_device *adev) { if (adev->asic_type > CHIP_TONGA) { - /* The first few doorbells in pci doorbell bar are for GFX RB - * rings and all the leftover for MEC. - * So CP_MEC_DOORBELL_RANGE_LOWER should be set one index after - * CP_RB_DOORBELL_RANGE_UPPER, as we assume there is only one - * GFX RB rings. - */ - u32 tmp = REG_SET_FIELD(0, CP_MEC_DOORBELL_RANGE_LOWER, - DOORBELL_RANGE_LOWER, - adev->gfx.gfx_ring[0].doorbell_index + 1); - - WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, tmp); - WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, - CP_MEC_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); + WREG32(mmCP_MEC_DOORBELL_RANGE_LOWER, adev->doorbell_index.kiq << 2); + WREG32(mmCP_MEC_DOORBELL_RANGE_UPPER, adev->doorbell_index.mec_ring7 << 2); } /* enable doorbells */ WREG32_FIELD(CP_PQ_STATUS, DOORBELL_ENABLE, 1); diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index 36f417fd6ba9..5533f6e4f4a4 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -2631,8 +2631,8 @@ static int gfx_v9_0_cp_gfx_resume(struct amdgpu_device *adev) DOORBELL_RANGE_LOWER, ring->doorbell_index); WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_LOWER, tmp); - /* There is only one GFX queue */ - WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER, tmp); + WREG32_SOC15(GC, 0, mmCP_RB_DOORBELL_RANGE_UPPER, + CP_RB_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); /* start the ring */ @@ -2995,19 +2995,10 @@ static int gfx_v9_0_kiq_init_register(struct amdgpu_ring *ring) /* enable the doorbell if requested */ if (ring->use_doorbell) { - /* The first few doorbells in pci doorbell bar are for GFX RB - * rings and all the leftover for MEC. - * So CP_MEC_DOORBELL_RANGE_LOWER should be set one index after - * CP_RB_DOORBELL_RANGE_UPPER, as we assume there is only one - * GFX RB rings. - */ - u32 tmp = REG_SET_FIELD(0, CP_MEC_DOORBELL_RANGE_LOWER, - DOORBELL_RANGE_LOWER, - adev->gfx.gfx_ring[0].doorbell_index + 2); - - WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, tmp); + WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_LOWER, + (adev->doorbell_index.kiq * 2) << 2); WREG32_SOC15(GC, 0, mmCP_MEC_DOORBELL_RANGE_UPPER, - CP_MEC_DOORBELL_RANGE_UPPER__DOORBELL_RANGE_UPPER_MASK); + (adev->doorbell_index.userqueue_end * 2) << 2); } WREG32_SOC15(GC, 0, mmCP_HQD_PQ_DOORBELL_CONTROL, -- cgit v1.2.3 From 8466cc61da89d33441e0d7a98de1ba98697cd465 Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 28 Jan 2019 13:41:58 +0100 Subject: drm/amdgpu: cleanup setting bulk_movable MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We only need to set this to false now when BOs are removed from the LRU. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 6 ------ 1 file changed, 6 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 6dda415bb3ee..d4c733885e87 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -332,7 +332,6 @@ static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base, if (bo->tbo.resv != vm->root.base.bo->tbo.resv) return; - vm->bulk_moveable = false; if (bo->tbo.type == ttm_bo_type_kernel) amdgpu_vm_bo_relocated(base); else @@ -698,8 +697,6 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm, struct amdgpu_vm_bo_base *bo_base, *tmp; int r = 0; - vm->bulk_moveable &= list_empty(&vm->evicted); - list_for_each_entry_safe(bo_base, tmp, &vm->evicted, vm_status) { struct amdgpu_bo *bo = bo_base->bo; @@ -2758,9 +2755,6 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, struct amdgpu_vm_bo_base **base; if (bo) { - if (bo->tbo.resv == vm->root.base.bo->tbo.resv) - vm->bulk_moveable = false; - for (base = &bo_va->base.bo->vm_bo; *base; base = &(*base)->next) { if (*base != &bo_va->base) -- cgit v1.2.3 From 661b96b21c25e05b9a02178b4f19fea83b5dd774 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 30 Jan 2019 13:41:05 +0100 Subject: drm/amdgpu: partial revert cleanup setting bulk_movable v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We still need to set bulk_movable to false when new BOs are added or removed. v2: also set it to false on removal Signed-off-by: Christian König Tested-by: StDenis, Tom Tested-by: Przemek Socha Reviewed-by: Zhou, David(ChunMing) Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index d4c733885e87..b06ffd29ccc6 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -332,6 +332,7 @@ static void amdgpu_vm_bo_base_init(struct amdgpu_vm_bo_base *base, if (bo->tbo.resv != vm->root.base.bo->tbo.resv) return; + vm->bulk_moveable = false; if (bo->tbo.type == ttm_bo_type_kernel) amdgpu_vm_bo_relocated(base); else @@ -2755,6 +2756,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev, struct amdgpu_vm_bo_base **base; if (bo) { + if (bo->tbo.resv == vm->root.base.bo->tbo.resv) + vm->bulk_moveable = false; + for (base = &bo_va->base.bo->vm_bo; *base; base = &(*base)->next) { if (*base != &bo_va->base) -- cgit v1.2.3 From 767e06a9924162ce8ca5890533932174b04471f3 Mon Sep 17 00:00:00 2001 From: Bas Nieuwenhuizen Date: Sun, 3 Feb 2019 01:45:24 +0100 Subject: drm/amdgpu: Bump amdgpu version for context priority override. Signed-off-by: Bas Nieuwenhuizen Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index 637f927e3675..7419ea8a388b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -73,9 +73,10 @@ * - 3.27.0 - Add new chunk to to AMDGPU_CS to enable BO_LIST creation. * - 3.28.0 - Add AMDGPU_CHUNK_ID_SCHEDULED_DEPENDENCIES * - 3.29.0 - Add AMDGPU_IB_FLAG_RESET_GDS_MAX_WAVE_ID + * - 3.30.0 - Add AMDGPU_SCHED_OP_CONTEXT_PRIORITY_OVERRIDE. */ #define KMS_DRIVER_MAJOR 3 -#define KMS_DRIVER_MINOR 29 +#define KMS_DRIVER_MINOR 30 #define KMS_DRIVER_PATCHLEVEL 0 int amdgpu_vram_limit = 0; -- cgit v1.2.3