Merge branch 'drm-next-4.20' of git://people.freedesktop.org/~agd5f/l…

…inux into drm-next

More new features and fixes for 4.20:
- Add dynamic powergating support for VCN on picasso
- Scheduler cleanup
- Vega20 support for KFD
- DC cleanups and bug fixes

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180927184348.2696-1-alexander.deucher@amd.com

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -146,6 +146,7 @@ extern int amdgpu_cik_support;
 #define AMDGPU_DEFAULT_GTT_SIZE_MB		3072ULL /* 3GB by default */
 #define AMDGPU_WAIT_IDLE_TIMEOUT_IN_MS	        3000
 #define AMDGPU_MAX_USEC_TIMEOUT			100000	/* 100 ms */
+#define AMDGPU_FENCE_JIFFIES_TIMEOUT		(HZ / 2)
 /* AMDGPU_IB_POOL_SIZE must be a power of 2 */
 #define AMDGPU_IB_POOL_SIZE			16
 #define AMDGPU_DEBUGFS_MAX_COMPONENTS		32
@@ -408,16 +409,25 @@ typedef enum _AMDGPU_DOORBELL64_ASSIGNMENT
 	AMDGPU_DOORBELL64_GFX_RING0               = 0x8b,
 
 	/*
-	 * Other graphics doorbells can be allocated here: from 0x8c to 0xef
+	 * Other graphics doorbells can be allocated here: from 0x8c to 0xdf
 	 * Graphics voltage island aperture 1
-	 * default non-graphics QWORD index is 0xF0 - 0xFF inclusive
+	 * default non-graphics QWORD index is 0xe0 - 0xFF inclusive
 	 */
 
-	/* sDMA engines */
-	AMDGPU_DOORBELL64_sDMA_ENGINE0            = 0xF0,
-	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0     = 0xF1,
-	AMDGPU_DOORBELL64_sDMA_ENGINE1            = 0xF2,
-	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1     = 0xF3,
+	/* sDMA engines  reserved from 0xe0 -oxef  */
+	AMDGPU_DOORBELL64_sDMA_ENGINE0            = 0xE0,
+	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE0     = 0xE1,
+	AMDGPU_DOORBELL64_sDMA_ENGINE1            = 0xE8,
+	AMDGPU_DOORBELL64_sDMA_HI_PRI_ENGINE1     = 0xE9,
+
+	/* For vega10 sriov, the sdma doorbell must be fixed as follow
+	 * to keep the same setting with host driver, or it will
+	 * happen conflicts
+	 */
+	AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0            = 0xF0,
+	AMDGPU_VEGA10_DOORBELL64_sDMA_HI_PRI_ENGINE0     = 0xF1,
+	AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1            = 0xF2,
+	AMDGPU_VEGA10_DOORBELL64_sDMA_HI_PRI_ENGINE1     = 0xF3,
 
 	/* Interrupt handler */
 	AMDGPU_DOORBELL64_IH                      = 0xF4,  /* For legacy interrupt ring buffer */

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c

@@ -76,6 +76,7 @@ void amdgpu_amdkfd_device_probe(struct amdgpu_device *adev)
 		kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
 		break;
 	case CHIP_VEGA10:
+	case CHIP_VEGA20:
 	case CHIP_RAVEN:
 		kfd2kgd = amdgpu_amdkfd_gfx_9_0_get_functions();
 		break;
@@ -123,7 +124,7 @@ static void amdgpu_doorbell_get_kfd_info(struct amdgpu_device *adev,
 
 void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
 {
-	int i;
+	int i, n;
 	int last_valid_bit;
 	if (adev->kfd) {
 		struct kgd2kfd_shared_resources gpu_resources = {
@@ -162,28 +163,47 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
 				&gpu_resources.doorbell_physical_address,
 				&gpu_resources.doorbell_aperture_size,
 				&gpu_resources.doorbell_start_offset);
-		if (adev->asic_type >= CHIP_VEGA10) {
+
+		if (adev->asic_type < CHIP_VEGA10) {
+			kgd2kfd->device_init(adev->kfd, &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][0] =
-				AMDGPU_DOORBELL64_sDMA_ENGINE0;
-			gpu_resources.sdma_doorbell[0][1] =
-				AMDGPU_DOORBELL64_sDMA_ENGINE0 + 0x200;
-			gpu_resources.sdma_doorbell[1][0] =
-				AMDGPU_DOORBELL64_sDMA_ENGINE1;
-			gpu_resources.sdma_doorbell[1][1] =
-				AMDGPU_DOORBELL64_sDMA_ENGINE1 + 0x200;
-			/* Doorbells 0x0f0-0ff and 0x2f0-2ff are reserved for
-			 * SDMA, IH and VCN. So don't use them for the CP.
-			 */
-			gpu_resources.reserved_doorbell_mask = 0x1f0;
-			gpu_resources.reserved_doorbell_val  = 0x0f0;
+			if (adev->asic_type == CHIP_VEGA10) {
+				gpu_resources.sdma_doorbell[0][i] =
+					AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 + (i >> 1);
+				gpu_resources.sdma_doorbell[0][i+1] =
+					AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE0 + 0x200 + (i >> 1);
+				gpu_resources.sdma_doorbell[1][i] =
+					AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 + (i >> 1);
+				gpu_resources.sdma_doorbell[1][i+1] =
+					AMDGPU_VEGA10_DOORBELL64_sDMA_ENGINE1 + 0x200 + (i >> 1);
+			} else {
+				gpu_resources.sdma_doorbell[0][i] =
+					AMDGPU_DOORBELL64_sDMA_ENGINE0 + (i >> 1);
+				gpu_resources.sdma_doorbell[0][i+1] =
+					AMDGPU_DOORBELL64_sDMA_ENGINE0 + 0x200 + (i >> 1);
+				gpu_resources.sdma_doorbell[1][i] =
+					AMDGPU_DOORBELL64_sDMA_ENGINE1 + (i >> 1);
+				gpu_resources.sdma_doorbell[1][i+1] =
+					AMDGPU_DOORBELL64_sDMA_ENGINE1 + 0x200 + (i >> 1);
+			}
 		}
+		/* Doorbells 0x0e0-0ff and 0x2e0-2ff are reserved for
+		 * SDMA, IH and VCN. So don't use them for the CP.
+		 */
+		gpu_resources.reserved_doorbell_mask = 0x1e0;
+		gpu_resources.reserved_doorbell_val  = 0x0e0;
 
 		kgd2kfd->device_init(adev->kfd, &gpu_resources);
 	}

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h

@@ -174,7 +174,7 @@ void amdgpu_amdkfd_gpuvm_destroy_cb(struct amdgpu_device *adev,
 				struct amdgpu_vm *vm);
 void amdgpu_amdkfd_gpuvm_destroy_process_vm(struct kgd_dev *kgd, void *vm);
 void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *vm);
-uint32_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm);
+uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm);
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 		struct kgd_dev *kgd, uint64_t va, uint64_t size,
 		void *vm, struct kgd_mem **mem,

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c

@@ -142,7 +142,7 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
 static void set_scratch_backing_va(struct kgd_dev *kgd,
 					uint64_t va, uint32_t vmid);
 static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t page_table_base);
+		uint64_t page_table_base);
 static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid);
 static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid);
 static uint32_t read_vmid_from_vmfault_reg(struct kgd_dev *kgd);
@@ -874,15 +874,16 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
 }
 
 static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-			uint32_t page_table_base)
+			uint64_t page_table_base)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 
 	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
 		pr_err("trying to set page table base for wrong VMID\n");
 		return;
 	}
-	WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8, page_table_base);
+	WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8,
+		lower_32_bits(page_table_base));
 }
 
 static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c

@@ -45,8 +45,6 @@ enum hqd_dequeue_request_type {
 	RESET_WAVES
 };
 
-struct vi_sdma_mqd;
-
 /*
  * Register access functions
  */
@@ -100,7 +98,7 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
 static void set_scratch_backing_va(struct kgd_dev *kgd,
 					uint64_t va, uint32_t vmid);
 static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t page_table_base);
+		uint64_t page_table_base);
 static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid);
 static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid);
 
@@ -282,7 +280,8 @@ static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
 
 	lock_srbm(kgd, mec, pipe, 0, 0);
 
-	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK);
+	WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK |
+			CP_INT_CNTL_RING0__OPCODE_ERROR_INT_ENABLE_MASK);
 
 	unlock_srbm(kgd);
 
@@ -834,15 +833,16 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
 }
 
 static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t page_table_base)
+		uint64_t page_table_base)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 
 	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
 		pr_err("trying to set page table base for wrong VMID\n");
 		return;
 	}
-	WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8, page_table_base);
+	WREG32(mmVM_CONTEXT8_PAGE_TABLE_BASE_ADDR + vmid - 8,
+			lower_32_bits(page_table_base));
 }
 
 static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c

@@ -138,7 +138,7 @@ static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
 static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
 		uint8_t vmid);
 static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t page_table_base);
+		uint64_t page_table_base);
 static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
 static void set_scratch_backing_va(struct kgd_dev *kgd,
 					uint64_t va, uint32_t vmid);
@@ -1013,11 +1013,10 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
 }
 
 static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
-		uint32_t page_table_base)
+		uint64_t page_table_base)
 {
 	struct amdgpu_device *adev = get_amdgpu_device(kgd);
-	uint64_t base = (uint64_t)page_table_base << PAGE_SHIFT |
-		AMDGPU_PTE_VALID;
+	uint64_t base = page_table_base | AMDGPU_PTE_VALID;
 
 	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
 		pr_err("trying to set page table base for wrong VMID %u\n",

drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c

@@ -1131,11 +1131,15 @@ void amdgpu_amdkfd_gpuvm_release_process_vm(struct kgd_dev *kgd, void *vm)
 	amdgpu_vm_release_compute(adev, avm);
 }
 
-uint32_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm)
+uint64_t amdgpu_amdkfd_gpuvm_get_process_page_dir(void *vm)
 {
 	struct amdgpu_vm *avm = (struct amdgpu_vm *)vm;
+	struct amdgpu_bo *pd = avm->root.base.bo;
+	struct amdgpu_device *adev = amdgpu_ttm_adev(pd->tbo.bdev);
 
-	return avm->pd_phys_addr >> AMDGPU_GPU_PAGE_SHIFT;
+	if (adev->asic_type < CHIP_VEGA10)
+		return avm->pd_phys_addr >> AMDGPU_GPU_PAGE_SHIFT;
+	return avm->pd_phys_addr;
 }
 
 int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c

@@ -195,6 +195,19 @@ int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s)
 	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_process - check for fence activity
  *
@@ -203,8 +216,10 @@ int amdgpu_fence_emit_polling(struct amdgpu_ring *ring, uint32_t *s)
  * Checks the current fence value and calculates the last
  * signalled fence value. Wakes the fence queue if the
  * sequence number has increased.
+ *
+ * Returns true if fence was processed
  */
-void amdgpu_fence_process(struct amdgpu_ring *ring)
+bool amdgpu_fence_process(struct amdgpu_ring *ring)
 {
 	struct amdgpu_fence_driver *drv = &ring->fence_drv;
 	uint32_t seq, last_seq;
@@ -216,8 +231,12 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
 
 	} while (atomic_cmpxchg(&drv->last_seq, last_seq, seq) != last_seq);
 
+	if (del_timer(&ring->fence_drv.fallback_timer) &&
+	    seq != ring->fence_drv.sync_seq)
+		amdgpu_fence_schedule_fallback(ring);
+
 	if (unlikely(seq == last_seq))
-		return;
+		return false;
 
 	last_seq &= drv->num_fences_mask;
 	seq &= drv->num_fences_mask;
@@ -244,6 +263,24 @@ void amdgpu_fence_process(struct amdgpu_ring *ring)
 
 		dma_fence_put(fence);
 	} while (last_seq != seq);
+
+	return true;
+}
+
+/**
+ * amdgpu_fence_fallback - fallback for hardware interrupts
+ *
+ * @work: delayed work item
+ *
+ * Checks for fence activity.
+ */
+static void amdgpu_fence_fallback(struct timer_list *t)
+{
+	struct amdgpu_ring *ring = from_timer(ring, t,
+					      fence_drv.fallback_timer);
+
+	if (amdgpu_fence_process(ring))
+		DRM_WARN("Fence fallback timer expired on ring %s\n", ring->name);
 }
 
 /**
@@ -393,6 +430,8 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 	atomic_set(&ring->fence_drv.last_seq, 0);
 	ring->fence_drv.initialized = false;
 
+	timer_setup(&ring->fence_drv.fallback_timer, amdgpu_fence_fallback, 0);
+
 	ring->fence_drv.num_fences_mask = num_hw_submission * 2 - 1;
 	spin_lock_init(&ring->fence_drv.lock);
 	ring->fence_drv.fences = kcalloc(num_hw_submission * 2, sizeof(void *),
@@ -468,6 +507,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
 		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
 			       ring->fence_drv.irq_type);
 		drm_sched_fini(&ring->sched);
+		del_timer_sync(&ring->fence_drv.fallback_timer);
 		for (j = 0; j <= ring->fence_drv.num_fences_mask; ++j)
 			dma_fence_put(ring->fence_drv.fences[j]);
 		kfree(ring->fence_drv.fences);
@@ -560,6 +600,27 @@ static const char *amdgpu_fence_get_timeline_name(struct dma_fence *f)
 	return (const char *)fence->ring->name;
 }
 
+/**
+ * amdgpu_fence_enable_signaling - enable signalling on fence
+ * @fence: fence
+ *
+ * This function is called with fence_queue lock held, and adds a callback
+ * to fence_queue that checks if this fence is signaled, and if so it
+ * signals the fence and removes itself.
+ */
+static bool amdgpu_fence_enable_signaling(struct dma_fence *f)
+{
+	struct amdgpu_fence *fence = to_amdgpu_fence(f);
+	struct amdgpu_ring *ring = fence->ring;
+
+	if (!timer_pending(&ring->fence_drv.fallback_timer))
+		amdgpu_fence_schedule_fallback(ring);
+
+	DMA_FENCE_TRACE(&fence->base, "armed on ring %i!\n", ring->idx);
+
+	return true;
+}
+
 /**
  * amdgpu_fence_free - free up the fence memory
  *
@@ -590,6 +651,7 @@ static void amdgpu_fence_release(struct dma_fence *f)
 static const struct dma_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,
 	.release = amdgpu_fence_release,
 };