drm/amdgpu: use IB for copy buffer of eviction

This aids handling buffers moves with the scheduler.

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index c916043..aa2dcf5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -247,7 +247,7 @@
 	unsigned	copy_num_dw;
 
 	/* used for buffer migration */
-	void (*emit_copy_buffer)(struct amdgpu_ring *ring,
+	void (*emit_copy_buffer)(struct amdgpu_ib *ib,
 				 /* src addr in bytes */
 				 uint64_t src_offset,
 				 /* dst addr in bytes */
@@ -518,7 +518,7 @@
 		       uint64_t dst_offset,
 		       uint32_t byte_count,
 		       struct reservation_object *resv,
-		       struct amdgpu_fence **fence);
+		       struct fence **fence);
 int amdgpu_mmap(struct file *filp, struct vm_area_struct *vma);
 
 struct amdgpu_bo_list_entry {
@@ -2247,7 +2247,7 @@
 #define amdgpu_display_add_connector(adev, ci, sd, ct, ib, coi, h, r) (adev)->mode_info.funcs->add_connector((adev), (ci), (sd), (ct), (ib), (coi), (h), (r))
 #define amdgpu_display_stop_mc_access(adev, s) (adev)->mode_info.funcs->stop_mc_access((adev), (s))
 #define amdgpu_display_resume_mc_access(adev, s) (adev)->mode_info.funcs->resume_mc_access((adev), (s))
-#define amdgpu_emit_copy_buffer(adev, r, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((r), (s), (d), (b))
+#define amdgpu_emit_copy_buffer(adev, ib, s, d, b) (adev)->mman.buffer_funcs->emit_copy_buffer((ib),  (s), (d), (b))
 #define amdgpu_emit_fill_buffer(adev, r, s, d, b) (adev)->mman.buffer_funcs->emit_fill_buffer((r), (s), (d), (b))
 #define amdgpu_dpm_get_temperature(adev) (adev)->pm.funcs->get_temperature((adev))
 #define amdgpu_dpm_pre_set_power_state(adev) (adev)->pm.funcs->pre_set_power_state((adev))
@@ -2379,7 +2379,7 @@
 		       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
  */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
index 759482e..98d59ee 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c
@@ -33,7 +33,7 @@
 {
 	unsigned long start_jiffies;
 	unsigned long end_jiffies;
-	struct amdgpu_fence *fence = NULL;
+	struct fence *fence = NULL;
 	int i, r;
 
 	start_jiffies = jiffies;
@@ -42,17 +42,17 @@
 		r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence);
 		if (r)
 			goto exit_do_move;
-		r = fence_wait(&fence->base, false);
+		r = fence_wait(fence, false);
 		if (r)
 			goto exit_do_move;
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 	}
 	end_jiffies = jiffies;
 	r = jiffies_to_msecs(end_jiffies - start_jiffies);
 
 exit_do_move:
 	if (fence)
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
index 962dd55..f80b1a4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_test.c
@@ -77,7 +77,7 @@
 		void *gtt_map, *vram_map;
 		void **gtt_start, **gtt_end;
 		void **vram_start, **vram_end;
-		struct amdgpu_fence *fence = NULL;
+		struct fence *fence = NULL;
 
 		r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
 				     AMDGPU_GEM_DOMAIN_GTT, 0, NULL, gtt_obj + i);
@@ -116,13 +116,13 @@
 			goto out_lclean_unpin;
 		}
 
-		r = fence_wait(&fence->base, false);
+		r = fence_wait(fence, false);
 		if (r) {
 			DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
 			goto out_lclean_unpin;
 		}
 
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 
 		r = amdgpu_bo_kmap(vram_obj, &vram_map);
 		if (r) {
@@ -161,13 +161,13 @@
 			goto out_lclean_unpin;
 		}
 
-		r = fence_wait(&fence->base, false);
+		r = fence_wait(fence, false);
 		if (r) {
 			DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
 			goto out_lclean_unpin;
 		}
 
-		amdgpu_fence_unref(&fence);
+		fence_put(fence);
 
 		r = amdgpu_bo_kmap(gtt_obj[i], &gtt_map);
 		if (r) {
@@ -214,7 +214,7 @@
 			amdgpu_bo_unref(&gtt_obj[i]);
 		}
 		if (fence)
-			amdgpu_fence_unref(&fence);
+			fence_put(fence);
 		break;
 	}
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 4cb8132..3991435 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -228,7 +228,7 @@
 	struct amdgpu_device *adev;
 	struct amdgpu_ring *ring;
 	uint64_t old_start, new_start;
-	struct amdgpu_fence *fence;
+	struct fence *fence;
 	int r;
 
 	adev = amdgpu_get_adev(bo->bdev);
@@ -269,9 +269,9 @@
 			       new_mem->num_pages * PAGE_SIZE, /* bytes */
 			       bo->resv, &fence);
 	/* FIXME: handle copy error */
-	r = ttm_bo_move_accel_cleanup(bo, &fence->base,
+	r = ttm_bo_move_accel_cleanup(bo, fence,
 				      evict, no_wait_gpu, new_mem);
-	amdgpu_fence_unref(&fence);
+	fence_put(fence);
 	return r;
 }
 
@@ -987,52 +987,48 @@
 		       uint64_t dst_offset,
 		       uint32_t byte_count,
 		       struct reservation_object *resv,
-		       struct amdgpu_fence **fence)
+		       struct fence **fence)
 {
 	struct amdgpu_device *adev = ring->adev;
-	struct amdgpu_sync sync;
 	uint32_t max_bytes;
 	unsigned num_loops, num_dw;
+	struct amdgpu_ib *ib;
 	unsigned i;
 	int r;
 
-	/* sync other rings */
-	amdgpu_sync_create(&sync);
-	if (resv) {
-		r = amdgpu_sync_resv(adev, &sync, resv, false);
-		if (r) {
-			DRM_ERROR("sync failed (%d).\n", r);
-			amdgpu_sync_free(adev, &sync, NULL);
-			return r;
-		}
-	}
-
 	max_bytes = adev->mman.buffer_funcs->copy_max_bytes;
 	num_loops = DIV_ROUND_UP(byte_count, max_bytes);
 	num_dw = num_loops * adev->mman.buffer_funcs->copy_num_dw;
 
-	/* for fence and sync */
-	num_dw += 64 + AMDGPU_NUM_SYNCS * 8;
+	/* for IB padding */
+	while (num_dw & 0x7)
+		num_dw++;
 
-	r = amdgpu_sync_wait(&sync);
+	ib = kzalloc(sizeof(struct amdgpu_ib), GFP_KERNEL);
+	if (!ib)
+		return -ENOMEM;
+
+	r = amdgpu_ib_get(ring, NULL, num_dw * 4, ib);
 	if (r) {
-		DRM_ERROR("sync wait failed (%d).\n", r);
-		amdgpu_sync_free(adev, &sync, NULL);
+		kfree(ib);
 		return r;
 	}
 
-	r = amdgpu_ring_lock(ring, num_dw);
-	if (r) {
-		DRM_ERROR("ring lock failed (%d).\n", r);
-		amdgpu_sync_free(adev, &sync, NULL);
-		return r;
+	ib->length_dw = 0;
+
+	if (resv) {
+		r = amdgpu_sync_resv(adev, &ib->sync, resv,
+				     AMDGPU_FENCE_OWNER_UNDEFINED);
+		if (r) {
+			DRM_ERROR("sync failed (%d).\n", r);
+			goto error_free;
+		}
 	}
-	amdgpu_sync_rings(&sync, ring);
 
 	for (i = 0; i < num_loops; i++) {
 		uint32_t cur_size_in_bytes = min(byte_count, max_bytes);
 
-		amdgpu_emit_copy_buffer(adev, ring, src_offset, dst_offset,
+		amdgpu_emit_copy_buffer(adev, ib, src_offset, dst_offset,
 					cur_size_in_bytes);
 
 		src_offset += cur_size_in_bytes;
@@ -1040,17 +1036,24 @@
 		byte_count -= cur_size_in_bytes;
 	}
 
-	r = amdgpu_fence_emit(ring, AMDGPU_FENCE_OWNER_MOVE, fence);
-	if (r) {
-		amdgpu_ring_unlock_undo(ring);
-		amdgpu_sync_free(adev, &sync, NULL);
-		return r;
+	amdgpu_vm_pad_ib(adev, ib);
+	WARN_ON(ib->length_dw > num_dw);
+	r = amdgpu_sched_ib_submit_kernel_helper(adev, ring, ib, 1,
+						 &amdgpu_vm_free_job,
+						 AMDGPU_FENCE_OWNER_MOVE,
+						 fence);
+	if (r)
+		goto error_free;
+
+	if (!amdgpu_enable_scheduler) {
+		amdgpu_ib_free(adev, ib);
+		kfree(ib);
 	}
-
-	amdgpu_ring_unlock_commit(ring);
-	amdgpu_sync_free(adev, &sync, &(*fence)->base);
-
 	return 0;
+error_free:
+	amdgpu_ib_free(adev, ib);
+	kfree(ib);
+	return r;
 }
 
 #if defined(CONFIG_DEBUG_FS)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 677266b..83b7ce6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -316,8 +316,7 @@
 	}
 }
 
-static int amdgpu_vm_free_job(
-	struct amdgpu_job *sched_job)
+int amdgpu_vm_free_job(struct amdgpu_job *sched_job)
 {
 	int i;
 	for (i = 0; i < sched_job->num_ibs; i++)
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index af52655..3920c1e 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -1339,18 +1339,18 @@
  * Used by the amdgpu ttm implementation to move pages if
  * registered as the asic copy callback.
  */
-static void cik_sdma_emit_copy_buffer(struct amdgpu_ring *ring,
+static void cik_sdma_emit_copy_buffer(struct amdgpu_ib *ib,
 				      uint64_t src_offset,
 				      uint64_t dst_offset,
 				      uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0));
-	amdgpu_ring_write(ring, byte_count);
-	amdgpu_ring_write(ring, 0); /* src/dst endian swap */
-	amdgpu_ring_write(ring, lower_32_bits(src_offset));
-	amdgpu_ring_write(ring, upper_32_bits(src_offset));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
+	ib->ptr[ib->length_dw++] = SDMA_PACKET(SDMA_OPCODE_COPY, SDMA_COPY_SUB_OPCODE_LINEAR, 0);
+	ib->ptr[ib->length_dw++] = byte_count;
+	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
+	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 2b0e89e..715e02d 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -1350,19 +1350,19 @@
  * Used by the amdgpu ttm implementation to move pages if
  * registered as the asic copy callback.
  */
-static void sdma_v2_4_emit_copy_buffer(struct amdgpu_ring *ring,
+static void sdma_v2_4_emit_copy_buffer(struct amdgpu_ib *ib,
 				       uint64_t src_offset,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
-			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR));
-	amdgpu_ring_write(ring, byte_count);
-	amdgpu_ring_write(ring, 0); /* src/dst endian swap */
-	amdgpu_ring_write(ring, lower_32_bits(src_offset));
-	amdgpu_ring_write(ring, upper_32_bits(src_offset));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
+		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
+	ib->ptr[ib->length_dw++] = byte_count;
+	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
+	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
 }
 
 /**
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 6f1df03..67128c8 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -1474,19 +1474,19 @@
  * Used by the amdgpu ttm implementation to move pages if
  * registered as the asic copy callback.
  */
-static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ring *ring,
+static void sdma_v3_0_emit_copy_buffer(struct amdgpu_ib *ib,
 				       uint64_t src_offset,
 				       uint64_t dst_offset,
 				       uint32_t byte_count)
 {
-	amdgpu_ring_write(ring, SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
-			  SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR));
-	amdgpu_ring_write(ring, byte_count);
-	amdgpu_ring_write(ring, 0); /* src/dst endian swap */
-	amdgpu_ring_write(ring, lower_32_bits(src_offset));
-	amdgpu_ring_write(ring, upper_32_bits(src_offset));
-	amdgpu_ring_write(ring, lower_32_bits(dst_offset));
-	amdgpu_ring_write(ring, upper_32_bits(dst_offset));
+	ib->ptr[ib->length_dw++] = SDMA_PKT_HEADER_OP(SDMA_OP_COPY) |
+		SDMA_PKT_HEADER_SUB_OP(SDMA_SUBOP_COPY_LINEAR);
+	ib->ptr[ib->length_dw++] = byte_count;
+	ib->ptr[ib->length_dw++] = 0; /* src/dst endian swap */
+	ib->ptr[ib->length_dw++] = lower_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(src_offset);
+	ib->ptr[ib->length_dw++] = lower_32_bits(dst_offset);
+	ib->ptr[ib->length_dw++] = upper_32_bits(dst_offset);
 }
 
 /**