Skip to contentSkip to conversation

drm/amdgpu: pass all the sdma scheds to amdgpu_mman

PATCH Superseded Pierre-Eric Pelloux-Prayer 4 participants 20 messages Apr 3 – Apr 8

Subsystems (4)

v5 of 6 Superseded Latest is v6 →

lore
Share & tools
lore
#0 lore raw

PATCH v5 1/7 drm/amdgpu: pass all the sdma scheds to amdgpu_mman

This will allow the use of all of them for clear/fill buffer operations.
Since drm_sched_entity_init requires a scheduler array, we store schedulers rather than rings. For the few places that need access to a ring, we can get it from the sched using container_of.

Since the code is the same for all sdma versions, add a new helper amdgpu_sdma_set_buffer_funcs_scheds to set buffer_funcs_scheds based on the number of sdma instances.

Note: the new sched array is identical to the amdgpu_vm_manager one. These 2 could be merged.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c    |  4 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 32 ++++++++++++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  3 +-
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c      |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  8 ++----
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c   |  6 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c     |  5 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c     |  5 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/si_dma.c        |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c   |  3 +-
 17 files changed, 48 insertions(+), 43 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 49e7881750fa..e3a8701f0b27 100644
@@ -1467,6 +1467,8 @@ ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring);
 ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset);
 void amdgpu_sdma_set_vm_pte_scheds(struct amdgpu_device *adev,
 				   const struct amdgpu_vm_pte_funcs *vm_pte_funcs);
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+					 const struct amdgpu_buffer_funcs *buffer_funcs);
 
 /* atpx handler */
 #if defined(CONFIG_VGA_SWITCHEROO)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -1
index 0c0489395edf..4da8de34be3d 100644
@@ -3700,7 +3700,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	adev->num_rings = 0;
 	RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub());
 	adev->mman.buffer_funcs = NULL;
-	adev->mman.buffer_funcs_ring = NULL;
+	adev->mman.num_buffer_funcs_scheds = 0;
 	adev->vm_manager.vm_pte_funcs = NULL;
 	adev->vm_manager.vm_pte_num_scheds = 0;
 	adev->gmc.gmc_funcs = NULL;
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +3 -1
index e9e53c7c37d0..1a253e4257ff 100644
@@ -708,12 +708,14 @@ int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev)
 void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 			      uint32_t vmhub, uint32_t flush_type)
 {
-	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+	struct amdgpu_ring *ring;
 	struct amdgpu_vmhub *hub = &adev->vmhub[vmhub];
 	struct dma_fence *fence;
 	struct amdgpu_job *job;
 	int r;
 
+	ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
+
 	if (!hub->sdma_invalidation_workaround || vmid ||
 	    !adev->mman.buffer_funcs_enabled || !adev->ib_pool_ready ||
 	    !ring->sched.ready) {
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +26 -6
index 3c63f87832e4..4ba7321b75e3 100644
@@ -168,7 +168,7 @@ amdgpu_ttm_job_submit(struct amdgpu_device *adev, struct amdgpu_ttm_buffer_entit
 {
 	struct amdgpu_ring *ring;
 
-	ring = adev->mman.buffer_funcs_ring;
+	ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
 	amdgpu_ring_pad_ib(ring, &job->ibs[0]);
 	WARN_ON(job->ibs[0].length_dw > num_dw);
 
@@ -2349,18 +2349,17 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 		return;
 
 	if (enable) {
-		struct amdgpu_ring *ring;
 		struct drm_gpu_scheduler *sched;
 
-		if (!adev->mman.buffer_funcs_ring || !adev->mman.buffer_funcs_ring->sched.ready) {
+		if (!adev->mman.num_buffer_funcs_scheds ||
+		    !adev->mman.buffer_funcs_scheds[0]->ready) {
 			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
 			return;
 		}
 
 		num_clear_entities = 1;
 		num_move_entities = 1;
-		ring = adev->mman.buffer_funcs_ring;
-		sched = &ring->sched;
+		sched = adev->mman.buffer_funcs_scheds[0];
 		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
 						  &adev->mman.default_entity,
 						  DRM_SCHED_PRIORITY_KERNEL,
@@ -2497,7 +2496,7 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
 	unsigned int i;
 	int r;
 
-	ring = adev->mman.buffer_funcs_ring;
+	ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
 
 	if (!ring->sched.ready) {
 		dev_err(adev->dev,
@@ -2730,6 +2729,27 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
 	return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
 }
 
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+					 const struct amdgpu_buffer_funcs *buffer_funcs)
+{
+	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];
+	struct drm_gpu_scheduler *sched;
+	int i;
+
+	adev->mman.buffer_funcs = buffer_funcs;
+
+	for (i = 0; i < adev->sdma.num_instances; i++) {
+		if (adev->sdma.has_page_queue)
+			sched = &adev->sdma.instance[i].page.sched;
+		else
+			sched = &adev->sdma.instance[i].ring.sched;
+		adev->mman.buffer_funcs_scheds[i] = sched;
+	}
+
+	adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ?
+		1 : adev->sdma.num_instances;
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -1
index 3b1973611446..a6249252948b 100644
@@ -67,7 +67,8 @@ struct amdgpu_mman {
 
 	/* buffer handling */
 	const struct amdgpu_buffer_funcs	*buffer_funcs;
-	struct amdgpu_ring			*buffer_funcs_ring;
+	struct drm_gpu_scheduler		*buffer_funcs_scheds[AMDGPU_MAX_RINGS];
+	u32					num_buffer_funcs_scheds;
 	bool					buffer_funcs_enabled;
 
 	/* @default_entity: for workarounds, has no gart windows */
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +1 -2
index 22780c09177d..26276dcfd458 100644
@@ -1340,8 +1340,7 @@ static const struct amdgpu_buffer_funcs cik_sdma_buffer_funcs = {
 
 static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &cik_sdma_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &cik_sdma_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version cik_sdma_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +1 -2
index 0090ace49024..c6a059ca59e5 100644
@@ -1235,8 +1235,7 @@ static const struct amdgpu_buffer_funcs sdma_v2_4_buffer_funcs = {
 
 static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v2_4_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v2_4_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +1 -2
index 2526d393162a..cb516a25210d 100644
@@ -1677,8 +1677,7 @@ static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = {
 
 static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v3_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v3_0_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -6
index 44f0f23e1148..d56be26f216b 100644
@@ -2626,13 +2626,9 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_buffer_funcs = {
 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
 	if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >= IP_VERSION(4, 4, 0))
-		adev->mman.buffer_funcs = &sdma_v4_4_buffer_funcs;
+		amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_buffer_funcs);
 	else
-		adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
-	if (adev->sdma.has_page_queue)
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-	else
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+		amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_0_buffer_funcs);
 }
 
 static void sdma_v4_0_get_ras_error_count(uint32_t value,
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +1 -5
index 78bdfed0a7fd..67e9697301b4 100644
@@ -2316,11 +2316,7 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_2_buffer_funcs = {
 
 static void sdma_v4_4_2_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &sdma_v4_4_2_buffer_funcs;
-	if (adev->sdma.has_page_queue)
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-	else
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_2_buffer_funcs);
 }
 
 /**
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -4
index 52f4e9e099cb..86f5eb784d57 100644
@@ -2052,10 +2052,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_0_buffer_funcs = {
 
 static void sdma_v5_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	if (adev->mman.buffer_funcs == NULL) {
-		adev->mman.buffer_funcs = &sdma_v5_0_buffer_funcs;
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-	}
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v5_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +1 -4
index b4fb90cc8f7d..3fec838374b2 100644
@@ -2056,10 +2056,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_2_buffer_funcs = {
 
 static void sdma_v5_2_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	if (adev->mman.buffer_funcs == NULL) {
-		adev->mman.buffer_funcs = &sdma_v5_2_buffer_funcs;
-		adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-	}
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_2_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v5_2_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c +1 -2
index b005672f2f96..064508cecd11 100644
@@ -1895,8 +1895,7 @@ static const struct amdgpu_buffer_funcs sdma_v6_0_buffer_funcs = {
 
 static void sdma_v6_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &sdma_v6_0_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v6_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v6_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +1 -2
index 5679a94d0815..60447729271e 100644
@@ -1845,8 +1845,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_0_buffer_funcs = {
 
 static void sdma_v7_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &sdma_v7_0_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v7_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c +1 -2
index f20e0fc3fc74..c3428d2731dc 100644
@@ -1764,8 +1764,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_1_buffer_funcs = {
 
 static void sdma_v7_1_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &sdma_v7_1_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_1_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v7_1_ip_block = {
drivers/gpu/drm/amd/amdgpu/si_dma.c +1 -2
index 3e58feb2d5e4..155067c20a0e 100644
@@ -833,8 +833,7 @@ static const struct amdgpu_buffer_funcs si_dma_buffer_funcs = {
 
 static void si_dma_set_buffer_funcs(struct amdgpu_device *adev)
 {
-	adev->mman.buffer_funcs = &si_dma_buffer_funcs;
-	adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+	amdgpu_sdma_set_buffer_funcs_scheds(adev, &si_dma_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version si_dma_ip_block =
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +2 -1
index 964efa325908..28dc6886c1ff 100644
@@ -129,13 +129,14 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys,
 			     struct dma_fence **mfence)
 {
 	const u64 GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE;
-	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+	struct amdgpu_ring *ring;
 	struct amdgpu_ttm_buffer_entity *entity;
 	u64 gart_s, gart_d;
 	struct dma_fence *next;
 	u64 size;
 	int r;
 
+	ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
 	entity = &adev->mman.move_entities[0];
 
 	mutex_lock(&entity->lock);
Signature · 1 line
2.43.0
#1 lore raw

PATCH v5 2/7 drm/amdgpu: only use working sdma schedulers for ttm

It's possible that some sdma instances aren't working so we shouldn't try to use them from TTM. To achieve this, delay the call to amdgpu_sdma_set_buffer_funcs_scheds after the rings have been tested, and then use the 'ready' property to decide if a sched should be used or not.

Note that currently it's not doing much, because if the ring helper fails for any ring, the whole sdma block init fails.

---
v5: check buffer_funcs_enabled from amdgpu_ttm_access_memory_sdma
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  | 23 ++++++++++++++++-------
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c    | 10 ++++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c   |  5 +++--
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c   |  5 +++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c   |  9 +++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c |  6 ++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c   |  6 ++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c   |  9 +++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c   |  9 +++++++--
 drivers/gpu/drm/amd/amdgpu/si_dma.c      |  9 +++++++--
 12 files changed, 68 insertions(+), 27 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +16 -7
index 4ba7321b75e3..bd32113292ec 100644
@@ -1560,7 +1560,7 @@ static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo,
 	if (!adev->mman.sdma_access_ptr)
 		return -EACCES;
 
-	if (!drm_dev_enter(adev_to_drm(adev), &idx))
+	if (!adev->mman.buffer_funcs_enabled || !drm_dev_enter(adev_to_drm(adev), &idx))
 		return -ENODEV;
 
 	if (write)
@@ -2351,8 +2351,7 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 	if (enable) {
 		struct drm_gpu_scheduler *sched;
 
-		if (!adev->mman.num_buffer_funcs_scheds ||
-		    !adev->mman.buffer_funcs_scheds[0]->ready) {
+		if (!adev->mman.num_buffer_funcs_scheds) {
 			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
 			return;
 		}
@@ -2734,20 +2733,30 @@ void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
 {
 	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];
 	struct drm_gpu_scheduler *sched;
-	int i;
+	int i, n;
 
 	adev->mman.buffer_funcs = buffer_funcs;
 
-	for (i = 0; i < adev->sdma.num_instances; i++) {
+	for (i = 0, n = 0; i < adev->sdma.num_instances; i++) {
 		if (adev->sdma.has_page_queue)
 			sched = &adev->sdma.instance[i].page.sched;
 		else
 			sched = &adev->sdma.instance[i].ring.sched;
-		adev->mman.buffer_funcs_scheds[i] = sched;
+
+		if (!sched->ready)
+			continue;
+
+		adev->mman.buffer_funcs_scheds[n++] = sched;
+	}
+
+	if (n == 0) {
+		adev->mman.num_buffer_funcs_scheds = 0;
+		drm_warn(&adev->ddev, "No working sdma ring available\n");
+		return;
 	}
 
 	adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ?
-		1 : adev->sdma.num_instances;
+		1 : n;
 }
 
 #if defined(CONFIG_DEBUG_FS)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +8 -2
index 26276dcfd458..120da838ac28 100644
@@ -939,7 +939,6 @@ static int cik_sdma_early_init(struct amdgpu_ip_block *ip_block)
 
 	cik_sdma_set_ring_funcs(adev);
 	cik_sdma_set_irq_funcs(adev);
-	cik_sdma_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &cik_sdma_vm_pte_funcs);
 
 	return 0;
@@ -1000,8 +999,15 @@ static int cik_sdma_sw_fini(struct amdgpu_ip_block *ip_block)
 static int cik_sdma_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-	return cik_sdma_start(adev);
+	r = cik_sdma_start(adev);
+	if (r)
+		return r;
+
+	cik_sdma_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int cik_sdma_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +3 -2
index c6a059ca59e5..93ec52c1f367 100644
@@ -828,7 +828,6 @@ static int sdma_v2_4_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v2_4_set_ring_funcs(adev);
-	sdma_v2_4_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v2_4_vm_pte_funcs);
 	sdma_v2_4_set_irq_funcs(adev);
 
@@ -898,7 +897,9 @@ static int sdma_v2_4_hw_init(struct amdgpu_ip_block *ip_block)
 	if (r)
 		return r;
 
-	return r;
+	sdma_v2_4_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v2_4_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +3 -2
index cb516a25210d..3fde9be74690 100644
@@ -1108,7 +1108,6 @@ static int sdma_v3_0_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v3_0_set_ring_funcs(adev);
-	sdma_v3_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v3_0_vm_pte_funcs);
 	sdma_v3_0_set_irq_funcs(adev);
 
@@ -1184,7 +1183,9 @@ static int sdma_v3_0_hw_init(struct amdgpu_ip_block *ip_block)
 	if (r)
 		return r;
 
-	return r;
+	sdma_v3_0_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v3_0_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +7 -2
index d56be26f216b..8a2a4e61867e 100644
@@ -1775,7 +1775,6 @@ static int sdma_v4_0_early_init(struct amdgpu_ip_block *ip_block)
 		adev->sdma.has_page_queue = true;
 
 	sdma_v4_0_set_ring_funcs(adev);
-	sdma_v4_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v4_0_vm_pte_funcs);
 	sdma_v4_0_set_irq_funcs(adev);
 	sdma_v4_0_set_ras_funcs(adev);
@@ -1961,6 +1960,7 @@ static int sdma_v4_0_sw_fini(struct amdgpu_ip_block *ip_block)
 static int sdma_v4_0_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
 	if (adev->flags & AMD_IS_APU)
 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false, 0);
@@ -1968,7 +1968,12 @@ static int sdma_v4_0_hw_init(struct amdgpu_ip_block *ip_block)
 	if (!amdgpu_sriov_vf(adev))
 		sdma_v4_0_init_golden_registers(adev);
 
-	return sdma_v4_0_start(adev);
+	r = sdma_v4_0_start(adev);
+	if (r)
+		return r;
+	sdma_v4_0_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v4_0_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +4 -2
index 67e9697301b4..88428b88e00f 100644
@@ -1368,7 +1368,6 @@ static int sdma_v4_4_2_early_init(struct amdgpu_ip_block *ip_block)
 		adev->sdma.has_page_queue = true;
 
 	sdma_v4_4_2_set_ring_funcs(adev);
-	sdma_v4_4_2_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v4_4_2_vm_pte_funcs);
 	sdma_v4_4_2_set_irq_funcs(adev);
 	sdma_v4_4_2_set_ras_funcs(adev);
@@ -1568,8 +1567,11 @@ static int sdma_v4_4_2_hw_init(struct amdgpu_ip_block *ip_block)
 		sdma_v4_4_2_inst_init_golden_registers(adev, inst_mask);
 
 	r = sdma_v4_4_2_inst_start(adev, inst_mask, false);
+	if (r)
+		return r;
+	sdma_v4_4_2_set_buffer_funcs(adev);
 
-	return r;
+	return 0;
 }
 
 static int sdma_v4_4_2_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +4 -2
index 86f5eb784d57..fa02907217e0 100644
@@ -1373,7 +1373,6 @@ static int sdma_v5_0_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v5_0_set_ring_funcs(adev);
-	sdma_v5_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v5_0_vm_pte_funcs);
 	sdma_v5_0_set_irq_funcs(adev);
 	sdma_v5_0_set_mqd_funcs(adev);
@@ -1472,8 +1471,11 @@ static int sdma_v5_0_hw_init(struct amdgpu_ip_block *ip_block)
 	sdma_v5_0_init_golden_registers(adev);
 
 	r = sdma_v5_0_start(adev);
+	if (r)
+		return r;
+	sdma_v5_0_set_buffer_funcs(adev);
 
-	return r;
+	return 0;
 }
 
 static int sdma_v5_0_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +7 -2
index 3fec838374b2..f6ecbc524c9b 100644
@@ -1264,7 +1264,6 @@ static int sdma_v5_2_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v5_2_set_ring_funcs(adev);
-	sdma_v5_2_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v5_2_vm_pte_funcs);
 	sdma_v5_2_set_irq_funcs(adev);
 	sdma_v5_2_set_mqd_funcs(adev);
@@ -1385,8 +1384,14 @@ static int sdma_v5_2_sw_fini(struct amdgpu_ip_block *ip_block)
 static int sdma_v5_2_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-	return sdma_v5_2_start(adev);
+	r = sdma_v5_2_start(adev);
+	if (r)
+		return r;
+	sdma_v5_2_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v5_2_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c +1 -1
index 064508cecd11..ee83d3e5978e 100644
@@ -1313,7 +1313,6 @@ static int sdma_v6_0_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v6_0_set_ring_funcs(adev);
-	sdma_v6_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v6_0_vm_pte_funcs);
 	sdma_v6_0_set_irq_funcs(adev);
 	sdma_v6_0_set_mqd_funcs(adev);
@@ -1477,6 +1476,7 @@ static int sdma_v6_0_hw_init(struct amdgpu_ip_block *ip_block)
 	r = sdma_v6_0_start(adev);
 	if (r)
 		return r;
+	sdma_v6_0_set_buffer_funcs(adev);
 
 	return sdma_v6_0_set_userq_trap_interrupts(adev, true);
 }
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +1 -1
index 60447729271e..c8343981c596 100644
@@ -1299,7 +1299,6 @@ static int sdma_v7_0_early_init(struct amdgpu_ip_block *ip_block)
 	}
 
 	sdma_v7_0_set_ring_funcs(adev);
-	sdma_v7_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v7_0_vm_pte_funcs);
 	sdma_v7_0_set_irq_funcs(adev);
 	sdma_v7_0_set_mqd_funcs(adev);
@@ -1432,6 +1431,7 @@ static int sdma_v7_0_hw_init(struct amdgpu_ip_block *ip_block)
 	r = sdma_v7_0_start(adev);
 	if (r)
 		return r;
+	sdma_v7_0_set_buffer_funcs(adev);
 
 	return sdma_v7_0_set_userq_trap_interrupts(adev, true);
 }
drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c +7 -2
index c3428d2731dc..b1ca95c2623b 100644
@@ -1275,7 +1275,6 @@ static int sdma_v7_1_early_init(struct amdgpu_ip_block *ip_block)
 	}
 
 	sdma_v7_1_set_ring_funcs(adev);
-	sdma_v7_1_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v7_1_vm_pte_funcs);
 	sdma_v7_1_set_irq_funcs(adev);
 	sdma_v7_1_set_mqd_funcs(adev);
@@ -1374,10 +1373,16 @@ static int sdma_v7_1_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
 	uint32_t inst_mask;
+	int r;
 
 	inst_mask = GENMASK(adev->sdma.num_instances - 1, 0);
 
-	return sdma_v7_1_inst_start(adev, inst_mask);
+	r = sdma_v7_1_inst_start(adev, inst_mask);
+	if (r)
+		return r;
+	sdma_v7_1_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v7_1_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/si_dma.c +7 -2
index 155067c20a0e..549708075eb4 100644
@@ -487,7 +487,6 @@ static int si_dma_early_init(struct amdgpu_ip_block *ip_block)
 	adev->sdma.num_instances = SDMA_MAX_INSTANCE;
 
 	si_dma_set_ring_funcs(adev);
-	si_dma_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &si_dma_vm_pte_funcs);
 	si_dma_set_irq_funcs(adev);
 
@@ -543,8 +542,14 @@ static int si_dma_sw_fini(struct amdgpu_ip_block *ip_block)
 static int si_dma_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-	return si_dma_start(adev);
+	r = si_dma_start(adev);
+	if (r)
+		return r;
+	si_dma_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int si_dma_hw_fini(struct amdgpu_ip_block *ip_block)
Signature · 1 line
2.43.0
maintainer Reviewed-by on v5 2/7
#2 lore raw

Re: [PATCH v5 2/7] drm/amdgpu: only use working sdma schedulers for ttm

Quoted 5 lines · Pierre-Eric Pelloux-Prayer

It's possible that some sdma instances aren't working so we shouldn't try to use them from TTM. To achieve this, delay the call to amdgpu_sdma_set_buffer_funcs_scheds after the rings have been tested, and then use the 'ready' property to decide if a sched should be used or not.

Quoted 8 lines

Note that currently it's not doing much, because if the ring helper fails for any ring, the whole sdma block init fails.

---
v5: check buffer_funcs_enabled from amdgpu_ttm_access_memory_sdma
---

Quoted 358 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  | 23 ++++++++++++++++-------
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c    | 10 ++++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c   |  5 +++--
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c   |  5 +++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c   |  9 +++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c |  6 ++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c   |  6 ++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c   |  9 +++++++--
 drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c   |  2 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c   |  9 +++++++--
 drivers/gpu/drm/amd/amdgpu/si_dma.c      |  9 +++++++--
 12 files changed, 68 insertions(+), 27 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +16 -7
index 4ba7321b75e3..bd32113292ec 100644
@@ -1560,7 +1560,7 @@ static int amdgpu_ttm_access_memory_sdma(struct ttm_buffer_object *bo,
 	if (!adev->mman.sdma_access_ptr)
 		return -EACCES;
 
-	if (!drm_dev_enter(adev_to_drm(adev), &idx))
+	if (!adev->mman.buffer_funcs_enabled || !drm_dev_enter(adev_to_drm(adev), &idx))
 		return -ENODEV;
 
 	if (write)
@@ -2351,8 +2351,7 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 	if (enable) {
 		struct drm_gpu_scheduler *sched;
 
-		if (!adev->mman.num_buffer_funcs_scheds ||
-		    !adev->mman.buffer_funcs_scheds[0]->ready) {
+		if (!adev->mman.num_buffer_funcs_scheds) {
 			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
 			return;
 		}
@@ -2734,20 +2733,30 @@ void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
 {
 	struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];
 	struct drm_gpu_scheduler *sched;
-	int i;
+	int i, n;
 
 	adev->mman.buffer_funcs = buffer_funcs;
 
-	for (i = 0; i < adev->sdma.num_instances; i++) {
+	for (i = 0, n = 0; i < adev->sdma.num_instances; i++) {
 		if (adev->sdma.has_page_queue)
 			sched = &adev->sdma.instance[i].page.sched;
 		else
 			sched = &adev->sdma.instance[i].ring.sched;
-		adev->mman.buffer_funcs_scheds[i] = sched;
+
+		if (!sched->ready)
+			continue;
+
+		adev->mman.buffer_funcs_scheds[n++] = sched;
+	}
+
+	if (n == 0) {
+		adev->mman.num_buffer_funcs_scheds = 0;
+		drm_warn(&adev->ddev, "No working sdma ring available\n");
+		return;
 	}
 
 	adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ?
-		1 : adev->sdma.num_instances;
+		1 : n;
 }
 
 #if defined(CONFIG_DEBUG_FS)
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +8 -2
index 26276dcfd458..120da838ac28 100644
@@ -939,7 +939,6 @@ static int cik_sdma_early_init(struct amdgpu_ip_block *ip_block)
 
 	cik_sdma_set_ring_funcs(adev);
 	cik_sdma_set_irq_funcs(adev);
-	cik_sdma_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &cik_sdma_vm_pte_funcs);
 
 	return 0;
@@ -1000,8 +999,15 @@ static int cik_sdma_sw_fini(struct amdgpu_ip_block *ip_block)
 static int cik_sdma_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-	return cik_sdma_start(adev);
+	r = cik_sdma_start(adev);
+	if (r)
+		return r;
+
+	cik_sdma_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int cik_sdma_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +3 -2
index c6a059ca59e5..93ec52c1f367 100644
@@ -828,7 +828,6 @@ static int sdma_v2_4_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v2_4_set_ring_funcs(adev);
-	sdma_v2_4_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v2_4_vm_pte_funcs);
 	sdma_v2_4_set_irq_funcs(adev);
 
@@ -898,7 +897,9 @@ static int sdma_v2_4_hw_init(struct amdgpu_ip_block *ip_block)
 	if (r)
 		return r;
 
-	return r;
+	sdma_v2_4_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v2_4_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +3 -2
index cb516a25210d..3fde9be74690 100644
@@ -1108,7 +1108,6 @@ static int sdma_v3_0_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v3_0_set_ring_funcs(adev);
-	sdma_v3_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v3_0_vm_pte_funcs);
 	sdma_v3_0_set_irq_funcs(adev);
 
@@ -1184,7 +1183,9 @@ static int sdma_v3_0_hw_init(struct amdgpu_ip_block *ip_block)
 	if (r)
 		return r;
 
-	return r;
+	sdma_v3_0_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v3_0_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +7 -2
index d56be26f216b..8a2a4e61867e 100644
@@ -1775,7 +1775,6 @@ static int sdma_v4_0_early_init(struct amdgpu_ip_block *ip_block)
 		adev->sdma.has_page_queue = true;
 
 	sdma_v4_0_set_ring_funcs(adev);
-	sdma_v4_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v4_0_vm_pte_funcs);
 	sdma_v4_0_set_irq_funcs(adev);
 	sdma_v4_0_set_ras_funcs(adev);
@@ -1961,6 +1960,7 @@ static int sdma_v4_0_sw_fini(struct amdgpu_ip_block *ip_block)
 static int sdma_v4_0_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
 	if (adev->flags & AMD_IS_APU)
 		amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_SDMA, false, 0);
@@ -1968,7 +1968,12 @@ static int sdma_v4_0_hw_init(struct amdgpu_ip_block *ip_block)
 	if (!amdgpu_sriov_vf(adev))
 		sdma_v4_0_init_golden_registers(adev);
 
-	return sdma_v4_0_start(adev);
+	r = sdma_v4_0_start(adev);
+	if (r)
+		return r;
+	sdma_v4_0_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v4_0_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +4 -2
index 67e9697301b4..88428b88e00f 100644
@@ -1368,7 +1368,6 @@ static int sdma_v4_4_2_early_init(struct amdgpu_ip_block *ip_block)
 		adev->sdma.has_page_queue = true;
 
 	sdma_v4_4_2_set_ring_funcs(adev);
-	sdma_v4_4_2_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v4_4_2_vm_pte_funcs);
 	sdma_v4_4_2_set_irq_funcs(adev);
 	sdma_v4_4_2_set_ras_funcs(adev);
@@ -1568,8 +1567,11 @@ static int sdma_v4_4_2_hw_init(struct amdgpu_ip_block *ip_block)
 		sdma_v4_4_2_inst_init_golden_registers(adev, inst_mask);
 
 	r = sdma_v4_4_2_inst_start(adev, inst_mask, false);
+	if (r)
+		return r;
+	sdma_v4_4_2_set_buffer_funcs(adev);
 
-	return r;
+	return 0;
 }
 
 static int sdma_v4_4_2_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +4 -2
index 86f5eb784d57..fa02907217e0 100644
@@ -1373,7 +1373,6 @@ static int sdma_v5_0_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v5_0_set_ring_funcs(adev);
-	sdma_v5_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v5_0_vm_pte_funcs);
 	sdma_v5_0_set_irq_funcs(adev);
 	sdma_v5_0_set_mqd_funcs(adev);
@@ -1472,8 +1471,11 @@ static int sdma_v5_0_hw_init(struct amdgpu_ip_block *ip_block)
 	sdma_v5_0_init_golden_registers(adev);
 
 	r = sdma_v5_0_start(adev);
+	if (r)
+		return r;
+	sdma_v5_0_set_buffer_funcs(adev);
 
-	return r;
+	return 0;
 }
 
 static int sdma_v5_0_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +7 -2
index 3fec838374b2..f6ecbc524c9b 100644
@@ -1264,7 +1264,6 @@ static int sdma_v5_2_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v5_2_set_ring_funcs(adev);
-	sdma_v5_2_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v5_2_vm_pte_funcs);
 	sdma_v5_2_set_irq_funcs(adev);
 	sdma_v5_2_set_mqd_funcs(adev);
@@ -1385,8 +1384,14 @@ static int sdma_v5_2_sw_fini(struct amdgpu_ip_block *ip_block)
 static int sdma_v5_2_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-	return sdma_v5_2_start(adev);
+	r = sdma_v5_2_start(adev);
+	if (r)
+		return r;
+	sdma_v5_2_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v5_2_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c +1 -1
index 064508cecd11..ee83d3e5978e 100644
@@ -1313,7 +1313,6 @@ static int sdma_v6_0_early_init(struct amdgpu_ip_block *ip_block)
 		return r;
 
 	sdma_v6_0_set_ring_funcs(adev);
-	sdma_v6_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v6_0_vm_pte_funcs);
 	sdma_v6_0_set_irq_funcs(adev);
 	sdma_v6_0_set_mqd_funcs(adev);
@@ -1477,6 +1476,7 @@ static int sdma_v6_0_hw_init(struct amdgpu_ip_block *ip_block)
 	r = sdma_v6_0_start(adev);
 	if (r)
 		return r;
+	sdma_v6_0_set_buffer_funcs(adev);
 
 	return sdma_v6_0_set_userq_trap_interrupts(adev, true);
 }
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +1 -1
index 60447729271e..c8343981c596 100644
@@ -1299,7 +1299,6 @@ static int sdma_v7_0_early_init(struct amdgpu_ip_block *ip_block)
 	}
 
 	sdma_v7_0_set_ring_funcs(adev);
-	sdma_v7_0_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v7_0_vm_pte_funcs);
 	sdma_v7_0_set_irq_funcs(adev);
 	sdma_v7_0_set_mqd_funcs(adev);
@@ -1432,6 +1431,7 @@ static int sdma_v7_0_hw_init(struct amdgpu_ip_block *ip_block)
 	r = sdma_v7_0_start(adev);
 	if (r)
 		return r;
+	sdma_v7_0_set_buffer_funcs(adev);
 
 	return sdma_v7_0_set_userq_trap_interrupts(adev, true);
 }
drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c +7 -2
index c3428d2731dc..b1ca95c2623b 100644
@@ -1275,7 +1275,6 @@ static int sdma_v7_1_early_init(struct amdgpu_ip_block *ip_block)
 	}
 
 	sdma_v7_1_set_ring_funcs(adev);
-	sdma_v7_1_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &sdma_v7_1_vm_pte_funcs);
 	sdma_v7_1_set_irq_funcs(adev);
 	sdma_v7_1_set_mqd_funcs(adev);
@@ -1374,10 +1373,16 @@ static int sdma_v7_1_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
 	uint32_t inst_mask;
+	int r;
 
 	inst_mask = GENMASK(adev->sdma.num_instances - 1, 0);
 
-	return sdma_v7_1_inst_start(adev, inst_mask);
+	r = sdma_v7_1_inst_start(adev, inst_mask);
+	if (r)
+		return r;
+	sdma_v7_1_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int sdma_v7_1_hw_fini(struct amdgpu_ip_block *ip_block)
drivers/gpu/drm/amd/amdgpu/si_dma.c +7 -2
index 155067c20a0e..549708075eb4 100644
@@ -487,7 +487,6 @@ static int si_dma_early_init(struct amdgpu_ip_block *ip_block)
 	adev->sdma.num_instances = SDMA_MAX_INSTANCE;
 
 	si_dma_set_ring_funcs(adev);
-	si_dma_set_buffer_funcs(adev);
 	amdgpu_sdma_set_vm_pte_scheds(adev, &si_dma_vm_pte_funcs);
 	si_dma_set_irq_funcs(adev);
 
@@ -543,8 +542,14 @@ static int si_dma_sw_fini(struct amdgpu_ip_block *ip_block)
 static int si_dma_hw_init(struct amdgpu_ip_block *ip_block)
 {
 	struct amdgpu_device *adev = ip_block->adev;
+	int r;
 
-	return si_dma_start(adev);
+	r = si_dma_start(adev);
+	if (r)
+		return r;
+	si_dma_set_buffer_funcs(adev);
+
+	return 0;
 }
 
 static int si_dma_hw_fini(struct amdgpu_ip_block *ip_block)
#3 lore raw

PATCH v5 3/7 drm/amdgpu: create multiple clear/move ttm entities

In reply to Pierre-Eric Pelloux-Prayer

This enables parallelism of operations.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +2 -2
index bd32113292ec..83f6d00dc3a0 100644
@@ -2356,8 +2356,8 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 			return;
 		}
 
-		num_clear_entities = 1;
-		num_move_entities = 1;
+		num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
+		num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
 		sched = adev->mman.buffer_funcs_scheds[0];
 		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
 						  &adev->mman.default_entity,
Signature · 1 line
2.43.0
maintainer Reviewed-by on v5 3/7
#4 lore raw

Re: [PATCH v5 3/7] drm/amdgpu: create multiple clear/move ttm entities

Quoted 3 lines · Pierre-Eric Pelloux-Prayer

This enables parallelism of operations.

Quoted 19 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +2 -2
index bd32113292ec..83f6d00dc3a0 100644
@@ -2356,8 +2356,8 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 			return;
 		}
 
-		num_clear_entities = 1;
-		num_move_entities = 1;
+		num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
+		num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
 		sched = adev->mman.buffer_funcs_scheds[0];
 		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
 						  &adev->mman.default_entity,
#5 lore raw

PATCH v5 4/7 drm/amdgpu: give ttm entities access to all the sdma scheds

In reply to Pierre-Eric Pelloux-Prayer

With this change we now have as many clear and move entities as we have sdma engines (limited to TTM_NUM_MOVE_FENCES).

To enable load-balancing this patch gives access to all sdma schedulers to all entities.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +9 -7
index 83f6d00dc3a0..648ad344e89c 100644
@@ -2349,8 +2349,6 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 		return;
 
 	if (enable) {
-		struct drm_gpu_scheduler *sched;
-
 		if (!adev->mman.num_buffer_funcs_scheds) {
 			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
 			return;
@@ -2358,11 +2356,10 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 		num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
 		num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
-		sched = adev->mman.buffer_funcs_scheds[0];
 		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
 						  &adev->mman.default_entity,
 						  DRM_SCHED_PRIORITY_KERNEL,
-						  &sched, 1, 0);
+						  adev->mman.buffer_funcs_scheds, 1, 0);
 		if (r < 0) {
 			dev_err(adev->dev,
 				"Failed setting up TTM entity (%d)\n", r);
@@ -2380,8 +2377,11 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 		for (i = 0; i < num_clear_entities; i++) {
 			r = amdgpu_ttm_buffer_entity_init(
-				&adev->mman.gtt_mgr, &adev->mman.clear_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 1);
+				&adev->mman.gtt_mgr,
+				&adev->mman.clear_entities[i],
+				DRM_SCHED_PRIORITY_NORMAL,
+				adev->mman.buffer_funcs_scheds,
+				adev->mman.num_buffer_funcs_scheds, 1);
 
 			if (r < 0) {
 				for (j = 0; j < i; j++)
@@ -2400,7 +2400,9 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 			r = amdgpu_ttm_buffer_entity_init(
 				&adev->mman.gtt_mgr,
 				&adev->mman.move_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 2);
+				DRM_SCHED_PRIORITY_NORMAL,
+				adev->mman.buffer_funcs_scheds,
+				adev->mman.num_buffer_funcs_scheds, 2);
 
 			if (r < 0) {
 				for (j = 0; j < i; j++)
Signature · 1 line
2.43.0
maintainer
#6 lore raw

Re: [PATCH v5 4/7] drm/amdgpu: give ttm entities access to all the sdma scheds

Quoted 26 lines · Pierre-Eric Pelloux-Prayer

With this change we now have as many clear and move entities as we have sdma engines (limited to TTM_NUM_MOVE_FENCES).

To enable load-balancing this patch gives access to all sdma schedulers to all entities.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 +++++++++-------
 1 file changed, 9 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -2
index 83f6d00dc3a0..648ad344e89c 100644
@@ -2349,8 +2349,6 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 		return;
 
 	if (enable) {
-		struct drm_gpu_scheduler *sched;
-
 		if (!adev->mman.num_buffer_funcs_scheds) {
 			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
 			return;

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) · line 2362

Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +1 -2
@@ -2358,11 +2356,10 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 		num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
 		num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
-		sched = adev->mman.buffer_funcs_scheds[0];
 		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
 						  &adev->mman.default_entity,
 						  DRM_SCHED_PRIORITY_KERNEL,
-						  &sched, 1, 0);
+						  adev->mman.buffer_funcs_scheds, 1, 0);

Why still giving num_schedulers as 1 here???

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) · line 2382

Quoted 12 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +3 -2
 		if (r < 0) {
 			dev_err(adev->dev,
 				"Failed setting up TTM entity (%d)\n", r);
@@ -2380,8 +2377,11 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 		for (i = 0; i < num_clear_entities; i++) {
 			r = amdgpu_ttm_buffer_entity_init(
-				&adev->mman.gtt_mgr, &adev->mman.clear_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 1);
+				&adev->mman.gtt_mgr,
+				&adev->mman.clear_entities[i],
+				DRM_SCHED_PRIORITY_NORMAL,

That should be DRM_SCHED_PRIORITY_KERNEL, same below.

Regards,
Christian.

Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5 -1
+				adev->mman.buffer_funcs_scheds,
+				adev->mman.num_buffer_funcs_scheds, 1);
 
 			if (r < 0) {
 				for (j = 0; j < i; j++)
@@ -2400,7 +2400,9 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 			r = amdgpu_ttm_buffer_entity_init(
 				&adev->mman.gtt_mgr,
 				&adev->mman.move_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 2);
+				DRM_SCHED_PRIORITY_NORMAL,
+				adev->mman.buffer_funcs_scheds,
+				adev->mman.num_buffer_funcs_scheds, 2);
 
 			if (r < 0) {
 				for (j = 0; j < i; j++)
#7 lore raw

Re: [PATCH v5 4/7] drm/amdgpu: give ttm entities access to all the sdma scheds

Le 07/04/2026 à 12:05, Christian König a écrit :

Quoted 28 lines · Pierre-Eric Pelloux-Prayer
Quoted 27 lines

With this change we now have as many clear and move entities as we have sdma engines (limited to TTM_NUM_MOVE_FENCES).

To enable load-balancing this patch gives access to all sdma schedulers to all entities.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 +++++++++-------
  1 file changed, 9 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -2
index 83f6d00dc3a0..648ad344e89c 100644
@@ -2349,8 +2349,6 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
  		return;
 
  	if (enable) {
-		struct drm_gpu_scheduler *sched;
-
  		if (!adev->mman.num_buffer_funcs_scheds) {
  			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
  			return;
@@ -2358,11 +2356,10 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
Quoted 11 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) · line 2362

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +1 -2
  		num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
  		num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
-		sched = adev->mman.buffer_funcs_scheds[0];
  		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
  						  &adev->mman.default_entity,
  						  DRM_SCHED_PRIORITY_KERNEL,
-						  &sched, 1, 0);
+						  adev->mman.buffer_funcs_scheds, 1, 0);

Why still giving num_schedulers as 1 here???

Because I think multiple schedulers aren't useful for this entity. But if you prefer I can pass all available schedulers to all ttm entities (in which case I'd remove the parameters from amdgpu_ttm_buffer_entity_init).

Quoted 5 lines
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -0
  		if (r < 0) {
  			dev_err(adev->dev,
  				"Failed setting up TTM entity (%d)\n", r);
@@ -2380,8 +2377,11 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
Quoted 9 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) · line 2382

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +3 -2
  		for (i = 0; i < num_clear_entities; i++) {
  			r = amdgpu_ttm_buffer_entity_init(
-				&adev->mman.gtt_mgr, &adev->mman.clear_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 1);
+				&adev->mman.gtt_mgr,
+				&adev->mman.clear_entities[i],
+				DRM_SCHED_PRIORITY_NORMAL,

That should be DRM_SCHED_PRIORITY_KERNEL, same below.

OK, will update in v6.

Thanks,
Pierre-Eric

Quoted 20 lines

Regards,
Christian.

Quoted 16 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5 -1
+				adev->mman.buffer_funcs_scheds,
+				adev->mman.num_buffer_funcs_scheds, 1);
 
  			if (r < 0) {
  				for (j = 0; j < i; j++)
@@ -2400,7 +2400,9 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
  			r = amdgpu_ttm_buffer_entity_init(
  				&adev->mman.gtt_mgr,
  				&adev->mman.move_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 2);
+				DRM_SCHED_PRIORITY_NORMAL,
+				adev->mman.buffer_funcs_scheds,
+				adev->mman.num_buffer_funcs_scheds, 2);
 
  			if (r < 0) {
  				for (j = 0; j < i; j++)
maintainer
#8 lore raw

Re: [PATCH v5 4/7] drm/amdgpu: give ttm entities access to all the sdma scheds

Quoted 41 lines · Pierre-Eric Pelloux-Prayer

Le 07/04/2026 à 12:05, Christian König a écrit :

Quoted 28 lines · Pierre-Eric Pelloux-Prayer
Quoted 27 lines

With this change we now have as many clear and move entities as we have sdma engines (limited to TTM_NUM_MOVE_FENCES).

To enable load-balancing this patch gives access to all sdma schedulers to all entities.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 +++++++++-------

  1 file changed, 9 insertions(+), 7 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -2
index 83f6d00dc3a0..648ad344e89c 100644
@@ -2349,8 +2349,6 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
           return;
         if (enable) {
-        struct drm_gpu_scheduler *sched;
-
           if (!adev->mman.num_buffer_funcs_scheds) {
               dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
               return;
@@ -2358,11 +2356,10 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
             num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
           num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
Quoted 9 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) · line 2362

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +1 -2
-        sched = adev->mman.buffer_funcs_scheds[0];
           r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
                             &adev->mman.default_entity,
                             DRM_SCHED_PRIORITY_KERNEL,
-                          &sched, 1, 0);
+                          adev->mman.buffer_funcs_scheds, 1, 0);

Why still giving num_schedulers as 1 here???

Quoted 1 line

Because I think multiple schedulers aren't useful for this entity. But if you prefer I can pass all available schedulers to all ttm entities (in which case I'd remove the parameters from amdgpu_ttm_buffer_entity_init).

Ah! Yeah that makes sense, but please add a comment why we do that.

I completely missed that this is for the default_entity.

Regards,
Christian.

Quoted 39 lines
Quoted 14 lines
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -0
           if (r < 0) {
               dev_err(adev->dev,
                   "Failed setting up TTM entity (%d)\n", r);
@@ -2380,8 +2377,11 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
             for (i = 0; i < num_clear_entities; i++) {
Quoted 6 lines

              r = amdgpu_ttm_buffer_entity_init(
-                &adev->mman.gtt_mgr, &adev->mman.clear_entities[i], -                DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 1); +                &adev->mman.gtt_mgr,
+                &adev->mman.clear_entities[i],
+                DRM_SCHED_PRIORITY_NORMAL,

That should be DRM_SCHED_PRIORITY_KERNEL, same below.

OK, will update in v6.

Thanks,
Pierre-Eric

Quoted 18 lines

Regards,
Christian.

Quoted 14 lines

+                adev->mman.buffer_funcs_scheds, +                adev->mman.num_buffer_funcs_scheds, 1); if (r < 0) {
                  for (j = 0; j < i; j++)

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -0
@@ -2400,7 +2400,9 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)

              r = amdgpu_ttm_buffer_entity_init( &adev->mman.gtt_mgr, &adev->mman.move_entities[i],
-                DRM_SCHED_PRIORITY_NORMAL, &sched, 1, 2); +                DRM_SCHED_PRIORITY_NORMAL,
+                adev->mman.buffer_funcs_scheds, +                adev->mman.num_buffer_funcs_scheds, 2); if (r < 0) {
                  for (j = 0; j < i; j++)

#9 lore raw

PATCH v5 5/7 drm/amdgpu: get rid of amdgpu_ttm_clear_buffer

In reply to Pierre-Eric Pelloux-Prayer

It's doing the same thing as amdgpu_fill_buffer(src_data=0), so drop it.

The only caveat is that amdgpu_res_cleared() return value is only valid right after allocation.

---
v2: introduce new "bool consider_clear_status" arg
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 16 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 88 +++++-----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  6 +-
 3 files changed, 32 insertions(+), 78 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +10 -6
index 66c20dd46d12..d0884bbffa75 100644
@@ -717,13 +717,17 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
 	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
 		struct dma_fence *fence;
 
-		r = amdgpu_ttm_clear_buffer(bo, bo->tbo.base.resv, &fence);
+		r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
+				       bo, 0, NULL, &fence,
+				       true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
 		if (unlikely(r))
 			goto fail_unreserve;
 
-		dma_resv_add_fence(bo->tbo.base.resv, fence,
-				   DMA_RESV_USAGE_KERNEL);
-		dma_fence_put(fence);
+		if (fence) {
+			dma_resv_add_fence(bo->tbo.base.resv, fence,
+					   DMA_RESV_USAGE_KERNEL);
+			dma_fence_put(fence);
+		}
 	}
 	if (!bp->resv)
 		amdgpu_bo_unreserve(bo);
@@ -1326,8 +1330,8 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
 		goto out;
 
 	r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-			       abo, 0, &bo->base._resv,
-			       &fence, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+			       abo, 0, &bo->base._resv, &fence,
+			       false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
 	if (WARN_ON(r))
 		goto out;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +20 -68

118 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -4
index a6249252948b..436a3e09a178 100644
@@ -187,14 +187,12 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct dma_resv *resv,
 		       struct dma_fence **fence,
 		       bool vm_needs_flush, uint32_t copy_flags);
-int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,
-			    struct dma_resv *resv,
-			    struct dma_fence **fence);
 int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 		       struct amdgpu_bo *bo,
 		       uint32_t src_data,
 		       struct dma_resv *resv,
-		       struct dma_fence **f,
+		       struct dma_fence **out_fence,
+		       bool consider_clear_status,
 		       u64 k_job_id);
 struct amdgpu_ttm_buffer_entity *amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev);
 
Signature · 1 line
2.43.0
maintainer
#10 lore raw

Re: [PATCH v5 5/7] drm/amdgpu: get rid of amdgpu_ttm_clear_buffer

Quoted 137 lines · Pierre-Eric Pelloux-Prayer

It's doing the same thing as amdgpu_fill_buffer(src_data=0), so drop it.

The only caveat is that amdgpu_res_cleared() return value is only valid right after allocation.

---
v2: introduce new "bool consider_clear_status" arg
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 16 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 88 +++++-----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  6 +-
 3 files changed, 32 insertions(+), 78 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +10 -6
index 66c20dd46d12..d0884bbffa75 100644
@@ -717,13 +717,17 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
 	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
 		struct dma_fence *fence;
 
-		r = amdgpu_ttm_clear_buffer(bo, bo->tbo.base.resv, &fence);
+		r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
+				       bo, 0, NULL, &fence,
+				       true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
 		if (unlikely(r))
 			goto fail_unreserve;
 
-		dma_resv_add_fence(bo->tbo.base.resv, fence,
-				   DMA_RESV_USAGE_KERNEL);
-		dma_fence_put(fence);
+		if (fence) {
+			dma_resv_add_fence(bo->tbo.base.resv, fence,
+					   DMA_RESV_USAGE_KERNEL);
+			dma_fence_put(fence);
+		}
 	}
 	if (!bp->resv)
 		amdgpu_bo_unreserve(bo);
@@ -1326,8 +1330,8 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
 		goto out;
 
 	r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-			       abo, 0, &bo->base._resv,
-			       &fence, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+			       abo, 0, &bo->base._resv, &fence,
+			       false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
 	if (WARN_ON(r))
 		goto out;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +12 -55
index 648ad344e89c..a3d0998b8cec 100644
@@ -418,7 +418,7 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
 		struct dma_fence *wipe_fence = NULL;
 		r = amdgpu_fill_buffer(entity, abo, 0, NULL, &wipe_fence,
-				       AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
+				       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
 		if (r) {
 			goto error;
 		} else if (wipe_fence) {
@@ -2572,76 +2572,25 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_ttm_clear_buffer - clear memory buffers
- * @bo: amdgpu buffer object
- * @resv: reservation object
- * @fence: dma_fence associated with the operation
+ * amdgpu_fill_buffer - fill a buffer with a given value
+ * @entity: entity to use
+ * @bo: the bo to fill
+ * @src_data: the value to set
+ * @resv: fences contained in this reservation will be used as dependencies.
+ * @out_fence: the fence from the last clear will be stored here. It might be
+ *             NULL if no job was run.
+ * @dependency: optional input dependency fence.
+ * @consider_clear_status: true if region reported as cleared by amdgpu_res_cleared()
+ *                         are skipped.
+ * @k_job_id: trace id
  *
- * Clear the memory buffer resource.
- *
- * Returns:
- * 0 for success or a negative error code on failure.
  */
-int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,
-			    struct dma_resv *resv,
-			    struct dma_fence **fence)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	struct amdgpu_ttm_buffer_entity *entity;
-	struct amdgpu_res_cursor cursor;
-	u64 addr;
-	int r = 0;
-
-	if (!adev->mman.buffer_funcs_enabled)
-		return -EINVAL;
-
-	if (!fence)
-		return -EINVAL;
-	entity = &adev->mman.clear_entities[0];
-	*fence = dma_fence_get_stub();
-
-	amdgpu_res_first(bo->tbo.resource, 0, amdgpu_bo_size(bo), &cursor);
-
-	mutex_lock(&entity->lock);
-	while (cursor.remaining) {
-		struct dma_fence *next = NULL;
-		u64 size;
-
-		if (amdgpu_res_cleared(&cursor)) {
-			amdgpu_res_next(&cursor, cursor.size);
-			continue;
-		}
-
-		/* Never clear more than 256MiB at once to avoid timeouts */
-		size = min(cursor.size, 256ULL << 20);
-
-		r = amdgpu_ttm_map_buffer(entity, &bo->tbo, bo->tbo.resource, &cursor,
-					  0, false, &size, &addr);
-		if (r)
-			goto err;
-
-		r = amdgpu_ttm_fill_mem(adev, entity, 0, addr, size, resv,
-					&next, true,
-					AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
-		if (r)
-			goto err;
-
-		dma_fence_put(*fence);
-		*fence = next;

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev, · line 2588

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -9
-
-		amdgpu_res_next(&cursor, size);
-	}
-err:
-	mutex_unlock(&entity->lock);
-
-	return r;
-}
-
 int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,

We should probably rename the function to amdgpu_ttm_clear_buffer() while doing this.

Apart from that looks good to me.

Regards,
Christian.

Quoted 53 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +8 -4
 		       struct amdgpu_bo *bo,
 		       uint32_t src_data,
 		       struct dma_resv *resv,
-		       struct dma_fence **f,
+		       struct dma_fence **out_fence,
+		       bool consider_clear_status,
 		       u64 k_job_id)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
@@ -2659,6 +2608,11 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 		struct dma_fence *next;
 		uint64_t cur_size, to;
 
+		if (consider_clear_status && amdgpu_res_cleared(&dst)) {
+			amdgpu_res_next(&dst, dst.size);
+			continue;
+		}
+
 		/* Never fill more than 256MiB at once to avoid timeouts */
 		cur_size = min(dst.size, 256ULL << 20);
 
@@ -2680,9 +2634,7 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 	}
 error:
 	mutex_unlock(&entity->lock);
-	if (f)
-		*f = dma_fence_get(fence);
-	dma_fence_put(fence);
+	*out_fence = fence;
 	return r;
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -4
index a6249252948b..436a3e09a178 100644
@@ -187,14 +187,12 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct dma_resv *resv,
 		       struct dma_fence **fence,
 		       bool vm_needs_flush, uint32_t copy_flags);
-int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,
-			    struct dma_resv *resv,
-			    struct dma_fence **fence);
 int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 		       struct amdgpu_bo *bo,
 		       uint32_t src_data,
 		       struct dma_resv *resv,
-		       struct dma_fence **f,
+		       struct dma_fence **out_fence,
+		       bool consider_clear_status,
 		       u64 k_job_id);
 struct amdgpu_ttm_buffer_entity *amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev);
 
maintainer
#11 lore raw

Re: [PATCH v5 5/7] drm/amdgpu: get rid of amdgpu_ttm_clear_buffer

Quoted 144 lines · Christian König
Quoted 141 lines · Pierre-Eric Pelloux-Prayer

It's doing the same thing as amdgpu_fill_buffer(src_data=0), so drop it.

The only caveat is that amdgpu_res_cleared() return value is only valid right after allocation.

---
v2: introduce new "bool consider_clear_status" arg
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 16 ++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 88 +++++-----------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  6 +-
 3 files changed, 32 insertions(+), 78 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +10 -6
index 66c20dd46d12..d0884bbffa75 100644
@@ -717,13 +717,17 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
 	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
 		struct dma_fence *fence;
 
-		r = amdgpu_ttm_clear_buffer(bo, bo->tbo.base.resv, &fence);
+		r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
+				       bo, 0, NULL, &fence,
+				       true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
 		if (unlikely(r))
 			goto fail_unreserve;
 
-		dma_resv_add_fence(bo->tbo.base.resv, fence,
-				   DMA_RESV_USAGE_KERNEL);
-		dma_fence_put(fence);
+		if (fence) {
+			dma_resv_add_fence(bo->tbo.base.resv, fence,
+					   DMA_RESV_USAGE_KERNEL);
+			dma_fence_put(fence);
+		}
 	}
 	if (!bp->resv)
 		amdgpu_bo_unreserve(bo);
@@ -1326,8 +1330,8 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
 		goto out;
 
 	r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-			       abo, 0, &bo->base._resv,
-			       &fence, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+			       abo, 0, &bo->base._resv, &fence,
+			       false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
 	if (WARN_ON(r))
 		goto out;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +12 -59
index 648ad344e89c..a3d0998b8cec 100644
@@ -418,7 +418,7 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
 		struct dma_fence *wipe_fence = NULL;
 		r = amdgpu_fill_buffer(entity, abo, 0, NULL, &wipe_fence,
-				       AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
+				       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
 		if (r) {
 			goto error;
 		} else if (wipe_fence) {
@@ -2572,76 +2572,25 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_ttm_clear_buffer - clear memory buffers
- * @bo: amdgpu buffer object
- * @resv: reservation object
- * @fence: dma_fence associated with the operation
+ * amdgpu_fill_buffer - fill a buffer with a given value
+ * @entity: entity to use
+ * @bo: the bo to fill
+ * @src_data: the value to set
+ * @resv: fences contained in this reservation will be used as dependencies.
+ * @out_fence: the fence from the last clear will be stored here. It might be
+ *             NULL if no job was run.
+ * @dependency: optional input dependency fence.
+ * @consider_clear_status: true if region reported as cleared by amdgpu_res_cleared()
+ *                         are skipped.
+ * @k_job_id: trace id
  *
- * Clear the memory buffer resource.
- *
- * Returns:
- * 0 for success or a negative error code on failure.
  */
-int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,
-			    struct dma_resv *resv,
-			    struct dma_fence **fence)
-{
-	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
-	struct amdgpu_ttm_buffer_entity *entity;
-	struct amdgpu_res_cursor cursor;
-	u64 addr;
-	int r = 0;
-
-	if (!adev->mman.buffer_funcs_enabled)
-		return -EINVAL;
-
-	if (!fence)
-		return -EINVAL;
-	entity = &adev->mman.clear_entities[0];
-	*fence = dma_fence_get_stub();
-
-	amdgpu_res_first(bo->tbo.resource, 0, amdgpu_bo_size(bo), &cursor);
-
-	mutex_lock(&entity->lock);
-	while (cursor.remaining) {
-		struct dma_fence *next = NULL;
-		u64 size;
-
-		if (amdgpu_res_cleared(&cursor)) {
-			amdgpu_res_next(&cursor, cursor.size);
-			continue;
-		}
-
-		/* Never clear more than 256MiB at once to avoid timeouts */
-		size = min(cursor.size, 256ULL << 20);
-
-		r = amdgpu_ttm_map_buffer(entity, &bo->tbo, bo->tbo.resource, &cursor,
-					  0, false, &size, &addr);
-		if (r)
-			goto err;
-
-		r = amdgpu_ttm_fill_mem(adev, entity, 0, addr, size, resv,
-					&next, true,
-					AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
-		if (r)
-			goto err;
-
-		dma_fence_put(*fence);
-		*fence = next;
-
-		amdgpu_res_next(&cursor, size);
-	}
-err:
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev, · line 2588

Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +0 -5
-	mutex_unlock(&entity->lock);
-
-	return r;
-}
-
 int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,

We should probably rename the function to amdgpu_ttm_clear_buffer() while doing this.

Forget that, I've just seen that patch #6 does exactly that.

Feel free to add Reviewed-by: Christian König <[email protected]>.

Regards,
Christian.

Quoted 60 lines

Apart from that looks good to me.

Regards,
Christian.

Quoted 53 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +8 -4
 		       struct amdgpu_bo *bo,
 		       uint32_t src_data,
 		       struct dma_resv *resv,
-		       struct dma_fence **f,
+		       struct dma_fence **out_fence,
+		       bool consider_clear_status,
 		       u64 k_job_id)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
@@ -2659,6 +2608,11 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 		struct dma_fence *next;
 		uint64_t cur_size, to;
 
+		if (consider_clear_status && amdgpu_res_cleared(&dst)) {
+			amdgpu_res_next(&dst, dst.size);
+			continue;
+		}
+
 		/* Never fill more than 256MiB at once to avoid timeouts */
 		cur_size = min(dst.size, 256ULL << 20);
 
@@ -2680,9 +2634,7 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 	}
 error:
 	mutex_unlock(&entity->lock);
-	if (f)
-		*f = dma_fence_get(fence);
-	dma_fence_put(fence);
+	*out_fence = fence;
 	return r;
 }
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -4
index a6249252948b..436a3e09a178 100644
@@ -187,14 +187,12 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct dma_resv *resv,
 		       struct dma_fence **fence,
 		       bool vm_needs_flush, uint32_t copy_flags);
-int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,
-			    struct dma_resv *resv,
-			    struct dma_fence **fence);
 int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 		       struct amdgpu_bo *bo,
 		       uint32_t src_data,
 		       struct dma_resv *resv,
-		       struct dma_fence **f,
+		       struct dma_fence **out_fence,
+		       bool consider_clear_status,
 		       u64 k_job_id);
 struct amdgpu_ttm_buffer_entity *amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev);
 
#12 lore raw

PATCH v5 6/7 drm/amdgpu: rename amdgpu_fill_buffer as amdgpu_ttm_clear_buffer

In reply to Pierre-Eric Pelloux-Prayer

This is the only use case for this function.

---
v2: amdgpu_ttm_clear_buffer instead of amdgpu_clear_buffer
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 12 +++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 23 ++++++++++------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    | 13 ++++++------
 3 files changed, 22 insertions(+), 26 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +6 -6
index d0884bbffa75..195cb1c814d1 100644
@@ -717,9 +717,9 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
 	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
 		struct dma_fence *fence;
 
-		r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-				       bo, 0, NULL, &fence,
-				       true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
+		r = amdgpu_ttm_clear_buffer(amdgpu_ttm_next_clear_entity(adev),
+					    bo, NULL, &fence,
+					    true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
 		if (unlikely(r))
 			goto fail_unreserve;
 
@@ -1329,9 +1329,9 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
 	if (r)
 		goto out;
 
-	r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-			       abo, 0, &bo->base._resv, &fence,
-			       false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+	r = amdgpu_ttm_clear_buffer(amdgpu_ttm_next_clear_entity(adev),
+				    abo, &bo->base._resv, &fence,
+				    false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
 	if (WARN_ON(r))
 		goto out;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +10 -13
index a3d0998b8cec..e74f9f8a88dc 100644
@@ -417,8 +417,8 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 	if (old_mem->mem_type == TTM_PL_VRAM &&
 	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
 		struct dma_fence *wipe_fence = NULL;
-		r = amdgpu_fill_buffer(entity, abo, 0, NULL, &wipe_fence,
-				       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
+		r = amdgpu_ttm_clear_buffer(entity, abo, NULL, &wipe_fence,
+					    false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
 		if (r) {
 			goto error;
 		} else if (wipe_fence) {
@@ -2572,26 +2572,23 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_fill_buffer - fill a buffer with a given value
+ * amdgpu_ttm_clear_buffer - fill a buffer with 0
  * @entity: entity to use
  * @bo: the bo to fill
- * @src_data: the value to set
  * @resv: fences contained in this reservation will be used as dependencies.
  * @out_fence: the fence from the last clear will be stored here. It might be
  *             NULL if no job was run.
- * @dependency: optional input dependency fence.
  * @consider_clear_status: true if region reported as cleared by amdgpu_res_cleared()
  *                         are skipped.
  * @k_job_id: trace id
  *
  */
-int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
-		       struct amdgpu_bo *bo,
-		       uint32_t src_data,
-		       struct dma_resv *resv,
-		       struct dma_fence **out_fence,
-		       bool consider_clear_status,
-		       u64 k_job_id)
+int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
+			    struct amdgpu_bo *bo,
+			    struct dma_resv *resv,
+			    struct dma_fence **out_fence,
+			    bool consider_clear_status,
+			    u64 k_job_id)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
 	struct dma_fence *fence = NULL;
@@ -2622,7 +2619,7 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 			goto error;
 
 		r = amdgpu_ttm_fill_mem(adev, entity,
-					src_data, to, cur_size, resv,
+					0, to, cur_size, resv,
 					&next, true, k_job_id);
 		if (r)
 			goto error;
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +6 -7
index 436a3e09a178..d7b14d5cac77 100644
@@ -187,13 +187,12 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct dma_resv *resv,
 		       struct dma_fence **fence,
 		       bool vm_needs_flush, uint32_t copy_flags);
-int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
-		       struct amdgpu_bo *bo,
-		       uint32_t src_data,
-		       struct dma_resv *resv,
-		       struct dma_fence **out_fence,
-		       bool consider_clear_status,
-		       u64 k_job_id);
+int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
+			    struct amdgpu_bo *bo,
+			    struct dma_resv *resv,
+			    struct dma_fence **out_fence,
+			    bool consider_clear_status,
+			    u64 k_job_id);
 struct amdgpu_ttm_buffer_entity *amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev);
 
 int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
Signature · 1 line
2.43.0
maintainer
#13 lore raw

Re: [PATCH v5 6/7] drm/amdgpu: rename amdgpu_fill_buffer as amdgpu_ttm_clear_buffer

Quoted 19 lines · Pierre-Eric Pelloux-Prayer

This is the only use case for this function.

---
v2: amdgpu_ttm_clear_buffer instead of amdgpu_clear_buffer
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 12 +++++------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 23 ++++++++++------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    | 13 ++++++------
 3 files changed, 22 insertions(+), 26 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +0 -0
index d0884bbffa75..195cb1c814d1 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c int amdgpu_bo_create(struct amdgpu_device *adev, · line 721

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +2 -3
@@ -717,9 +717,9 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
 	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
 		struct dma_fence *fence;
 
-		r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-				       bo, 0, NULL, &fence,
-				       true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
+		r = amdgpu_ttm_clear_buffer(amdgpu_ttm_next_clear_entity(adev),
+					    bo, NULL, &fence,

Wait a second. Why is the dependency NULL here?

Could it be that we relied on the fact that clears and moves were serialized in the past?

Regards,
Christian.

Quoted 99 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +4 -3
+					    true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
 		if (unlikely(r))
 			goto fail_unreserve;
 
@@ -1329,9 +1329,9 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
 	if (r)
 		goto out;
 
-	r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-			       abo, 0, &bo->base._resv, &fence,
-			       false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+	r = amdgpu_ttm_clear_buffer(amdgpu_ttm_next_clear_entity(adev),
+				    abo, &bo->base._resv, &fence,
+				    false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
 	if (WARN_ON(r))
 		goto out;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +10 -13
index a3d0998b8cec..e74f9f8a88dc 100644
@@ -417,8 +417,8 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
 	if (old_mem->mem_type == TTM_PL_VRAM &&
 	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
 		struct dma_fence *wipe_fence = NULL;
-		r = amdgpu_fill_buffer(entity, abo, 0, NULL, &wipe_fence,
-				       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
+		r = amdgpu_ttm_clear_buffer(entity, abo, NULL, &wipe_fence,
+					    false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
 		if (r) {
 			goto error;
 		} else if (wipe_fence) {
@@ -2572,26 +2572,23 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev,
 }
 
 /**
- * amdgpu_fill_buffer - fill a buffer with a given value
+ * amdgpu_ttm_clear_buffer - fill a buffer with 0
  * @entity: entity to use
  * @bo: the bo to fill
- * @src_data: the value to set
  * @resv: fences contained in this reservation will be used as dependencies.
  * @out_fence: the fence from the last clear will be stored here. It might be
  *             NULL if no job was run.
- * @dependency: optional input dependency fence.
  * @consider_clear_status: true if region reported as cleared by amdgpu_res_cleared()
  *                         are skipped.
  * @k_job_id: trace id
  *
  */
-int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
-		       struct amdgpu_bo *bo,
-		       uint32_t src_data,
-		       struct dma_resv *resv,
-		       struct dma_fence **out_fence,
-		       bool consider_clear_status,
-		       u64 k_job_id)
+int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
+			    struct amdgpu_bo *bo,
+			    struct dma_resv *resv,
+			    struct dma_fence **out_fence,
+			    bool consider_clear_status,
+			    u64 k_job_id)
 {
 	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
 	struct dma_fence *fence = NULL;
@@ -2622,7 +2619,7 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
 			goto error;
 
 		r = amdgpu_ttm_fill_mem(adev, entity,
-					src_data, to, cur_size, resv,
+					0, to, cur_size, resv,
 					&next, true, k_job_id);
 		if (r)
 			goto error;
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +6 -7
index 436a3e09a178..d7b14d5cac77 100644
@@ -187,13 +187,12 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct dma_resv *resv,
 		       struct dma_fence **fence,
 		       bool vm_needs_flush, uint32_t copy_flags);
-int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
-		       struct amdgpu_bo *bo,
-		       uint32_t src_data,
-		       struct dma_resv *resv,
-		       struct dma_fence **out_fence,
-		       bool consider_clear_status,
-		       u64 k_job_id);
+int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
+			    struct amdgpu_bo *bo,
+			    struct dma_resv *resv,
+			    struct dma_fence **out_fence,
+			    bool consider_clear_status,
+			    u64 k_job_id);
 struct amdgpu_ttm_buffer_entity *amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev);
 
 int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
#14 lore raw

Re: [PATCH v5 6/7] drm/amdgpu: rename amdgpu_fill_buffer as amdgpu_ttm_clear_buffer

Le 07/04/2026 à 12:16, Christian König a écrit :

Quoted 25 lines · Pierre-Eric Pelloux-Prayer
Quoted 22 lines

This is the only use case for this function.

---
v2: amdgpu_ttm_clear_buffer instead of amdgpu_clear_buffer
---

---
  drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 12 +++++------
  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 23 ++++++++++------------
  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    | 13 ++++++------
  3 files changed, 22 insertions(+), 26 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +0 -0
index d0884bbffa75..195cb1c814d1 100644
@@ -717,9 +717,9 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
  	    bo->tbo.resource->mem_type == TTM_PL_VRAM) {
  		struct dma_fence *fence;
 
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c int amdgpu_bo_create(struct amdgpu_device *adev, · line 721

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +2 -3
-		r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-				       bo, 0, NULL, &fence,
-				       true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
+		r = amdgpu_ttm_clear_buffer(amdgpu_ttm_next_clear_entity(adev),
+					    bo, NULL, &fence,

Wait a second. Why is the dependency NULL here?

Good catch.

Quoted 2 lines

Could it be that we relied on the fact that clears and moves were serialized in the past?

No it's a mistake in the previous patch where the resv param was dropped. I've fixed it in v6.

Thanks,
PE

Quoted 102 lines

Regards,
Christian.

Quoted 99 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +4 -3
+					    true, AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
  		if (unlikely(r))
  			goto fail_unreserve;
 
@@ -1329,9 +1329,9 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
  	if (r)
  		goto out;
 
-	r = amdgpu_fill_buffer(amdgpu_ttm_next_clear_entity(adev),
-			       abo, 0, &bo->base._resv, &fence,
-			       false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
+	r = amdgpu_ttm_clear_buffer(amdgpu_ttm_next_clear_entity(adev),
+				    abo, &bo->base._resv, &fence,
+				    false, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
  	if (WARN_ON(r))
  		goto out;
 
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +10 -13
index a3d0998b8cec..e74f9f8a88dc 100644
@@ -417,8 +417,8 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
  	if (old_mem->mem_type == TTM_PL_VRAM &&
  	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
  		struct dma_fence *wipe_fence = NULL;
-		r = amdgpu_fill_buffer(entity, abo, 0, NULL, &wipe_fence,
-				       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
+		r = amdgpu_ttm_clear_buffer(entity, abo, NULL, &wipe_fence,
+					    false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
  		if (r) {
  			goto error;
  		} else if (wipe_fence) {
@@ -2572,26 +2572,23 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_device *adev,
  }
 
  /**
- * amdgpu_fill_buffer - fill a buffer with a given value
+ * amdgpu_ttm_clear_buffer - fill a buffer with 0
   * @entity: entity to use
   * @bo: the bo to fill
- * @src_data: the value to set
   * @resv: fences contained in this reservation will be used as dependencies.
   * @out_fence: the fence from the last clear will be stored here. It might be
   *             NULL if no job was run.
- * @dependency: optional input dependency fence.
   * @consider_clear_status: true if region reported as cleared by amdgpu_res_cleared()
   *                         are skipped.
   * @k_job_id: trace id
   *
   */
-int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
-		       struct amdgpu_bo *bo,
-		       uint32_t src_data,
-		       struct dma_resv *resv,
-		       struct dma_fence **out_fence,
-		       bool consider_clear_status,
-		       u64 k_job_id)
+int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
+			    struct amdgpu_bo *bo,
+			    struct dma_resv *resv,
+			    struct dma_fence **out_fence,
+			    bool consider_clear_status,
+			    u64 k_job_id)
  {
  	struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
  	struct dma_fence *fence = NULL;
@@ -2622,7 +2619,7 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
  			goto error;
 
  		r = amdgpu_ttm_fill_mem(adev, entity,
-					src_data, to, cur_size, resv,
+					0, to, cur_size, resv,
  					&next, true, k_job_id);
  		if (r)
  			goto error;
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +6 -7
index 436a3e09a178..d7b14d5cac77 100644
@@ -187,13 +187,12 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
  		       struct dma_resv *resv,
  		       struct dma_fence **fence,
  		       bool vm_needs_flush, uint32_t copy_flags);
-int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity,
-		       struct amdgpu_bo *bo,
-		       uint32_t src_data,
-		       struct dma_resv *resv,
-		       struct dma_fence **out_fence,
-		       bool consider_clear_status,
-		       u64 k_job_id);
+int amdgpu_ttm_clear_buffer(struct amdgpu_ttm_buffer_entity *entity,
+			    struct amdgpu_bo *bo,
+			    struct dma_resv *resv,
+			    struct dma_fence **out_fence,
+			    bool consider_clear_status,
+			    u64 k_job_id);
  struct amdgpu_ttm_buffer_entity *amdgpu_ttm_next_clear_entity(struct amdgpu_device *adev);
 
  int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
#15 lore raw

PATCH v5 7/7 drm/amdgpu: split amdgpu_ttm_set_buffer_funcs_status in 2 funcs

In reply to Pierre-Eric Pelloux-Prayer

Makes a code slightly clearer and reduces indentation.

---
v5: use amdgpu_in_reset in amdgpu_ttm_disable_buffer_funcs
---

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  14 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 186 +++++++++++----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |   4 +-
 3 files changed, 108 insertions(+), 96 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +7 -7
index 4da8de34be3d..a24f52e71850 100644
@@ -2463,7 +2463,7 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 	if (r)
 		goto init_failed;
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_ttm_enable_buffer_funcs(adev);
 
 	/* Don't init kfd if whole hive need to be reset during init */
 	if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
@@ -3147,7 +3147,7 @@ static int amdgpu_device_ip_suspend(struct amdgpu_device *adev)
 		amdgpu_virt_request_full_gpu(adev, false);
 	}
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 
 	r = amdgpu_device_ip_suspend_phase1(adev);
 	if (r)
@@ -3362,7 +3362,7 @@ static int amdgpu_device_ip_resume(struct amdgpu_device *adev)
 
 	r = amdgpu_device_ip_resume_phase2(adev);
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_ttm_enable_buffer_funcs(adev);
 
 	if (r)
 		return r;
@@ -4215,7 +4215,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
 	/* disable ras feature must before hw fini */
 	amdgpu_ras_pre_fini(adev);
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 
 	/*
 	 * device went through surprise hotplug; we need to destroy topology
@@ -4482,7 +4482,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 	if (r)
 		goto unwind_userq;
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 
 	amdgpu_fence_driver_hw_fini(adev);
 
@@ -4496,7 +4496,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 	return 0;
 
 unwind_evict:
-	amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_ttm_enable_buffer_funcs(adev);
 	amdgpu_fence_driver_hw_init(adev);
 
 unwind_userq:
@@ -5230,7 +5230,7 @@ int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context)
 				if (r)
 					goto out;
 
-				amdgpu_ttm_set_buffer_funcs_status(tmp_adev, true);
+				amdgpu_ttm_enable_buffer_funcs(tmp_adev);
 
 				r = amdgpu_device_ip_resume_phase3(tmp_adev);
 				if (r)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +99 -87

227 lines not on this page · raw message

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -2
index d7b14d5cac77..8a5f34aaabac 100644
@@ -178,8 +178,8 @@ bool amdgpu_res_cpu_visible(struct amdgpu_device *adev,
 
 int amdgpu_ttm_init(struct amdgpu_device *adev);
 void amdgpu_ttm_fini(struct amdgpu_device *adev);
-void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev,
-					bool enable);
+void amdgpu_ttm_enable_buffer_funcs(struct amdgpu_device *adev);
+void amdgpu_ttm_disable_buffer_funcs(struct amdgpu_device *adev);
 int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct amdgpu_ttm_buffer_entity *entity,
 		       uint64_t src_offset,
Signature · 1 line
2.43.0
maintainer Reviewed-by on v5 7/7
#16 lore raw

Re: [PATCH v5 7/7] drm/amdgpu: split amdgpu_ttm_set_buffer_funcs_status in 2 funcs

Quoted 8 lines · Pierre-Eric Pelloux-Prayer

Makes a code slightly clearer and reduces indentation.

---
v5: use amdgpu_in_reset in amdgpu_ttm_disable_buffer_funcs
---

Quoted 322 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  14 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 186 +++++++++++----------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |   4 +-
 3 files changed, 108 insertions(+), 96 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +7 -7
index 4da8de34be3d..a24f52e71850 100644
@@ -2463,7 +2463,7 @@ static int amdgpu_device_ip_init(struct amdgpu_device *adev)
 	if (r)
 		goto init_failed;
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_ttm_enable_buffer_funcs(adev);
 
 	/* Don't init kfd if whole hive need to be reset during init */
 	if (adev->init_lvl->level != AMDGPU_INIT_LEVEL_MINIMAL_XGMI) {
@@ -3147,7 +3147,7 @@ static int amdgpu_device_ip_suspend(struct amdgpu_device *adev)
 		amdgpu_virt_request_full_gpu(adev, false);
 	}
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 
 	r = amdgpu_device_ip_suspend_phase1(adev);
 	if (r)
@@ -3362,7 +3362,7 @@ static int amdgpu_device_ip_resume(struct amdgpu_device *adev)
 
 	r = amdgpu_device_ip_resume_phase2(adev);
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_ttm_enable_buffer_funcs(adev);
 
 	if (r)
 		return r;
@@ -4215,7 +4215,7 @@ void amdgpu_device_fini_hw(struct amdgpu_device *adev)
 	/* disable ras feature must before hw fini */
 	amdgpu_ras_pre_fini(adev);
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 
 	/*
 	 * device went through surprise hotplug; we need to destroy topology
@@ -4482,7 +4482,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 	if (r)
 		goto unwind_userq;
 
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 
 	amdgpu_fence_driver_hw_fini(adev);
 
@@ -4496,7 +4496,7 @@ int amdgpu_device_suspend(struct drm_device *dev, bool notify_clients)
 	return 0;
 
 unwind_evict:
-	amdgpu_ttm_set_buffer_funcs_status(adev, true);
+	amdgpu_ttm_enable_buffer_funcs(adev);
 	amdgpu_fence_driver_hw_init(adev);
 
 unwind_userq:
@@ -5230,7 +5230,7 @@ int amdgpu_device_reinit_after_reset(struct amdgpu_reset_context *reset_context)
 				if (r)
 					goto out;
 
-				amdgpu_ttm_set_buffer_funcs_status(tmp_adev, true);
+				amdgpu_ttm_enable_buffer_funcs(tmp_adev);
 
 				r = amdgpu_device_ip_resume_phase3(tmp_adev);
 				if (r)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +99 -87
index e74f9f8a88dc..9bbd8149ffd2 100644
@@ -2100,7 +2100,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 	}
 
 	/* Change the size here instead of the init above so only lpfn is affected */
-	amdgpu_ttm_set_buffer_funcs_status(adev, false);
+	amdgpu_ttm_disable_buffer_funcs(adev);
 #ifdef CONFIG_64BIT
 #ifdef CONFIG_X86
 	if (adev->gmc.xgmi.connected_to_cpu)
@@ -2329,115 +2329,91 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_ttm_set_buffer_funcs_status - enable/disable use of buffer functions
+ * amdgpu_ttm_enable_buffer_funcs - enable use of buffer functions
  *
  * @adev: amdgpu_device pointer
- * @enable: true when we can use buffer functions.
  *
- * Enable/disable use of buffer functions during suspend/resume. This should
+ * Enable use of buffer functions during suspend/resume. This should
  * only be called at bootup or when userspace isn't running.
  */
-void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
+void amdgpu_ttm_enable_buffer_funcs(struct amdgpu_device *adev)
 {
 	struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
 	u32 num_clear_entities, num_move_entities;
-	uint64_t size;
 	int r, i, j;
 
 	if (!adev->mman.initialized || amdgpu_in_reset(adev) ||
-	    adev->mman.buffer_funcs_enabled == enable || adev->gmc.is_app_apu)
+	    adev->mman.buffer_funcs_enabled || adev->gmc.is_app_apu)
 		return;
 
-	if (enable) {
-		if (!adev->mman.num_buffer_funcs_scheds) {
-			dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
-			return;
-		}
+	if (!adev->mman.num_buffer_funcs_scheds) {
+		dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
+		return;
+	}
+
+	r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
+						&adev->mman.default_entity,
+						DRM_SCHED_PRIORITY_KERNEL,
+						adev->mman.buffer_funcs_scheds, 1, 0);
+	if (r < 0) {
+		dev_err(adev->dev,
+			"Failed setting up TTM entity (%d)\n", r);
+		return;
+	}
+
+	num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
+	num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
+
+	adev->mman.clear_entities = kcalloc(num_clear_entities,
+						sizeof(struct amdgpu_ttm_buffer_entity),
+						GFP_KERNEL);
+	atomic_set(&adev->mman.next_clear_entity, 0);
+	if (!adev->mman.clear_entities)
+		goto error_free_default_entity;
+
+	adev->mman.num_clear_entities = num_clear_entities;
+
+	for (i = 0; i < num_clear_entities; i++) {
+		r = amdgpu_ttm_buffer_entity_init(
+			&adev->mman.gtt_mgr,
+			&adev->mman.clear_entities[i],
+			DRM_SCHED_PRIORITY_NORMAL,
+			adev->mman.buffer_funcs_scheds,
+			adev->mman.num_buffer_funcs_scheds, 1);
 
-		num_clear_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
-		num_move_entities = MIN(adev->mman.num_buffer_funcs_scheds, TTM_NUM_MOVE_FENCES);
-		r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
-						  &adev->mman.default_entity,
-						  DRM_SCHED_PRIORITY_KERNEL,
-						  adev->mman.buffer_funcs_scheds, 1, 0);
 		if (r < 0) {
-			dev_err(adev->dev,
-				"Failed setting up TTM entity (%d)\n", r);
-			return;
-		}
-
-		adev->mman.clear_entities = kcalloc(num_clear_entities,
-						    sizeof(struct amdgpu_ttm_buffer_entity),
-						    GFP_KERNEL);
-		atomic_set(&adev->mman.next_clear_entity, 0);
-		if (!adev->mman.clear_entities)
+			for (j = 0; j < i; j++)
+				amdgpu_ttm_buffer_entity_fini(
+					&adev->mman.gtt_mgr, &adev->mman.clear_entities[j]);
+			adev->mman.num_clear_entities = 0;
+			kfree(adev->mman.clear_entities);
 			goto error_free_default_entity;
-
-		adev->mman.num_clear_entities = num_clear_entities;
-
-		for (i = 0; i < num_clear_entities; i++) {
-			r = amdgpu_ttm_buffer_entity_init(
-				&adev->mman.gtt_mgr,
-				&adev->mman.clear_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL,
-				adev->mman.buffer_funcs_scheds,
-				adev->mman.num_buffer_funcs_scheds, 1);
-
-			if (r < 0) {
-				for (j = 0; j < i; j++)
-					amdgpu_ttm_buffer_entity_fini(
-						&adev->mman.gtt_mgr, &adev->mman.clear_entities[j]);
-				kfree(adev->mman.clear_entities);
-				adev->mman.num_clear_entities = 0;
-				adev->mman.clear_entities = NULL;
-				goto error_free_default_entity;
-			}
 		}
+	}
 
-		adev->mman.num_move_entities = num_move_entities;
-		atomic_set(&adev->mman.next_move_entity, 0);
-		for (i = 0; i < num_move_entities; i++) {
-			r = amdgpu_ttm_buffer_entity_init(
-				&adev->mman.gtt_mgr,
-				&adev->mman.move_entities[i],
-				DRM_SCHED_PRIORITY_NORMAL,
-				adev->mman.buffer_funcs_scheds,
-				adev->mman.num_buffer_funcs_scheds, 2);
+	adev->mman.num_move_entities = num_move_entities;
+	atomic_set(&adev->mman.next_move_entity, 0);
+	for (i = 0; i < num_move_entities; i++) {
+		r = amdgpu_ttm_buffer_entity_init(
+			&adev->mman.gtt_mgr,
+			&adev->mman.move_entities[i],
+			DRM_SCHED_PRIORITY_NORMAL,
+			adev->mman.buffer_funcs_scheds,
+			adev->mman.num_buffer_funcs_scheds, 2);
 
-			if (r < 0) {
-				for (j = 0; j < i; j++)
-					amdgpu_ttm_buffer_entity_fini(
-						&adev->mman.gtt_mgr, &adev->mman.move_entities[j]);
-				adev->mman.num_move_entities = 0;
-				goto error_free_clear_entities;
-			}
+		if (r < 0) {
+			for (j = 0; j < i; j++)
+				amdgpu_ttm_buffer_entity_fini(
+					&adev->mman.gtt_mgr,
+					&adev->mman.move_entities[j]);
+			adev->mman.num_move_entities = 0;
+			goto error_free_clear_entities;
 		}
-	} else {
-		amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
-					      &adev->mman.default_entity);
-		for (i = 0; i < adev->mman.num_clear_entities; i++)
-			amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
-						      &adev->mman.clear_entities[i]);
-		for (i = 0; i < adev->mman.num_move_entities; i++)
-			amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
-						      &adev->mman.move_entities[i]);
-		/* Drop all the old fences since re-creating the scheduler entities
-		 * will allocate new contexts.
-		 */
-		ttm_resource_manager_cleanup(man);
-		kfree(adev->mman.clear_entities);
-		adev->mman.clear_entities = NULL;
-		adev->mman.num_clear_entities = 0;
-		adev->mman.num_move_entities = 0;
 	}
 
 	/* this just adjusts TTM size idea, which sets lpfn to the correct value */
-	if (enable)
-		size = adev->gmc.real_vram_size;
-	else
-		size = adev->gmc.visible_vram_size;
-	man->size = size;
-	adev->mman.buffer_funcs_enabled = enable;
+	man->size = adev->gmc.real_vram_size;
+	adev->mman.buffer_funcs_enabled = true;
 
 	return;
 
@@ -2453,6 +2429,42 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 				      &adev->mman.default_entity);
 }
 
+/**
+ * amdgpu_ttm_disable_buffer_funcs - disable use of buffer functions
+ *
+ * @adev: amdgpu_device pointer
+ */
+void amdgpu_ttm_disable_buffer_funcs(struct amdgpu_device *adev)
+{
+	struct ttm_resource_manager *man =
+		ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM);
+	int i;
+
+	if (!adev->mman.buffer_funcs_enabled || amdgpu_in_reset(adev))
+		return;
+
+	amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
+				      &adev->mman.default_entity);
+	for (i = 0; i < adev->mman.num_move_entities; i++)
+		amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
+					      &adev->mman.move_entities[i]);
+	for (i = 0; i < adev->mman.num_clear_entities; i++)
+		amdgpu_ttm_buffer_entity_fini(&adev->mman.gtt_mgr,
+					      &adev->mman.clear_entities[i]);
+	/* Drop all the old fences since re-creating the scheduler entities
+	 * will allocate new contexts.
+	 */
+	ttm_resource_manager_cleanup(man);
+
+	kfree(adev->mman.clear_entities);
+	adev->mman.clear_entities = NULL;
+	adev->mman.num_clear_entities = 0;
+	adev->mman.num_move_entities = 0;
+
+	man->size = adev->gmc.visible_vram_size;
+	adev->mman.buffer_funcs_enabled = false;
+}
+
 static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,
 				  struct amdgpu_ttm_buffer_entity *entity,
 				  unsigned int num_dw,
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -2
index d7b14d5cac77..8a5f34aaabac 100644
@@ -178,8 +178,8 @@ bool amdgpu_res_cpu_visible(struct amdgpu_device *adev,
 
 int amdgpu_ttm_init(struct amdgpu_device *adev);
 void amdgpu_ttm_fini(struct amdgpu_device *adev);
-void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev,
-					bool enable);
+void amdgpu_ttm_enable_buffer_funcs(struct amdgpu_device *adev);
+void amdgpu_ttm_disable_buffer_funcs(struct amdgpu_device *adev);
 int amdgpu_copy_buffer(struct amdgpu_device *adev,
 		       struct amdgpu_ttm_buffer_entity *entity,
 		       uint64_t src_offset,
#17 lore raw
In reply to Pierre-Eric Pelloux-Prayer
Quoted 127 lines · Pierre-Eric Pelloux-Prayer

This will allow the use of all of them for clear/fill buffer operations.
Since drm_sched_entity_init requires a scheduler array, we store schedulers rather than rings. For the few places that need access to a ring, we can get it from the sched using container_of.

Since the code is the same for all sdma versions, add a new helper amdgpu_sdma_set_buffer_funcs_scheds to set buffer_funcs_scheds based on the number of sdma instances.

Note: the new sched array is identical to the amdgpu_vm_manager one. These 2 could be merged.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c    |  4 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 32 ++++++++++++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  3 +-
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c      |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  8 ++----
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c   |  6 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c     |  5 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c     |  5 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/si_dma.c        |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c   |  3 +-
 17 files changed, 48 insertions(+), 43 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 49e7881750fa..e3a8701f0b27 100644
@@ -1467,6 +1467,8 @@ ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring);
 ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset);
 void amdgpu_sdma_set_vm_pte_scheds(struct amdgpu_device *adev,
                                   const struct amdgpu_vm_pte_funcs *vm_pte_funcs);
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+                                        const struct amdgpu_buffer_funcs *buffer_funcs);
 
 /* atpx handler */
 #if defined(CONFIG_VGA_SWITCHEROO)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -1
index 0c0489395edf..4da8de34be3d 100644
@@ -3700,7 +3700,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
        adev->num_rings = 0;
        RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub());
        adev->mman.buffer_funcs = NULL;
-       adev->mman.buffer_funcs_ring = NULL;
+       adev->mman.num_buffer_funcs_scheds = 0;
        adev->vm_manager.vm_pte_funcs = NULL;
        adev->vm_manager.vm_pte_num_scheds = 0;
        adev->gmc.gmc_funcs = NULL;
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +3 -1
index e9e53c7c37d0..1a253e4257ff 100644
@@ -708,12 +708,14 @@ int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev)
 void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
                              uint32_t vmhub, uint32_t flush_type)
 {
-       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+       struct amdgpu_ring *ring;
        struct amdgpu_vmhub *hub = &adev->vmhub[vmhub];
        struct dma_fence *fence;
        struct amdgpu_job *job;
        int r;
 
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
+
        if (!hub->sdma_invalidation_workaround || vmid ||
            !adev->mman.buffer_funcs_enabled || !adev->ib_pool_ready ||
            !ring->sched.ready) {
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5 -6
index 3c63f87832e4..4ba7321b75e3 100644
@@ -168,7 +168,7 @@ amdgpu_ttm_job_submit(struct amdgpu_device *adev, struct amdgpu_ttm_buffer_entit
 {
        struct amdgpu_ring *ring;
 
-       ring = adev->mman.buffer_funcs_ring;
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
        amdgpu_ring_pad_ib(ring, &job->ibs[0]);
        WARN_ON(job->ibs[0].length_dw > num_dw);
 
@@ -2349,18 +2349,17 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
                return;
 
        if (enable) {
-               struct amdgpu_ring *ring;
                struct drm_gpu_scheduler *sched;
 
-               if (!adev->mman.buffer_funcs_ring || !adev->mman.buffer_funcs_ring->sched.ready) {
+               if (!adev->mman.num_buffer_funcs_scheds ||
+                   !adev->mman.buffer_funcs_scheds[0]->ready) {
                        dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
                        return;
                }
 
                num_clear_entities = 1;
                num_move_entities = 1;
-               ring = adev->mman.buffer_funcs_ring;
-               sched = &ring->sched;
+               sched = adev->mman.buffer_funcs_scheds[0];
                r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
                                                  &adev->mman.default_entity,
                                                  DRM_SCHED_PRIORITY_KERNEL,
@@ -2497,7 +2496,7 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
        unsigned int i;
        int r;
 
-       ring = adev->mman.buffer_funcs_ring;
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
 
        if (!ring->sched.ready) {
                dev_err(adev->dev,

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type) · line 2735

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +4 -0
@@ -2730,6 +2729,27 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
        return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
 }
 
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+                                        const struct amdgpu_buffer_funcs *buffer_funcs)
+{
+       struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];

Only gfx9 and newer have multiple hubs. Additionally, sdma 4.x is attached to mmhub. sdma 5.x is attached to gfxhub. Might be worth a comment here that this is just for the sdma_invalidation_workaround check so it's only relevant for specific chips that have that set in case someone attempts to do something else here in the future.

Alex

Quoted 224 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +17 -0
+       struct drm_gpu_scheduler *sched;
+       int i;
+
+       adev->mman.buffer_funcs = buffer_funcs;
+
+       for (i = 0; i < adev->sdma.num_instances; i++) {
+               if (adev->sdma.has_page_queue)
+                       sched = &adev->sdma.instance[i].page.sched;
+               else
+                       sched = &adev->sdma.instance[i].ring.sched;
+               adev->mman.buffer_funcs_scheds[i] = sched;
+       }
+
+       adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ?
+               1 : adev->sdma.num_instances;
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -1
index 3b1973611446..a6249252948b 100644
@@ -67,7 +67,8 @@ struct amdgpu_mman {
 
        /* buffer handling */
        const struct amdgpu_buffer_funcs        *buffer_funcs;
-       struct amdgpu_ring                      *buffer_funcs_ring;
+       struct drm_gpu_scheduler                *buffer_funcs_scheds[AMDGPU_MAX_RINGS];
+       u32                                     num_buffer_funcs_scheds;
        bool                                    buffer_funcs_enabled;
 
        /* @default_entity: for workarounds, has no gart windows */
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +1 -2
index 22780c09177d..26276dcfd458 100644
@@ -1340,8 +1340,7 @@ static const struct amdgpu_buffer_funcs cik_sdma_buffer_funcs = {
 
 static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &cik_sdma_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &cik_sdma_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version cik_sdma_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +1 -2
index 0090ace49024..c6a059ca59e5 100644
@@ -1235,8 +1235,7 @@ static const struct amdgpu_buffer_funcs sdma_v2_4_buffer_funcs = {
 
 static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v2_4_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v2_4_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +1 -2
index 2526d393162a..cb516a25210d 100644
@@ -1677,8 +1677,7 @@ static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = {
 
 static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v3_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v3_0_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -6
index 44f0f23e1148..d56be26f216b 100644
@@ -2626,13 +2626,9 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_buffer_funcs = {
 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
        if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >= IP_VERSION(4, 4, 0))
-               adev->mman.buffer_funcs = &sdma_v4_4_buffer_funcs;
+               amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_buffer_funcs);
        else
-               adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
-       if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-       else
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+               amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_0_buffer_funcs);
 }
 
 static void sdma_v4_0_get_ras_error_count(uint32_t value,
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +1 -5
index 78bdfed0a7fd..67e9697301b4 100644
@@ -2316,11 +2316,7 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_2_buffer_funcs = {
 
 static void sdma_v4_4_2_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v4_4_2_buffer_funcs;
-       if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-       else
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_2_buffer_funcs);
 }
 
 /**
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -4
index 52f4e9e099cb..86f5eb784d57 100644
@@ -2052,10 +2052,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_0_buffer_funcs = {
 
 static void sdma_v5_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       if (adev->mman.buffer_funcs == NULL) {
-               adev->mman.buffer_funcs = &sdma_v5_0_buffer_funcs;
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-       }
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v5_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +1 -4
index b4fb90cc8f7d..3fec838374b2 100644
@@ -2056,10 +2056,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_2_buffer_funcs = {
 
 static void sdma_v5_2_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       if (adev->mman.buffer_funcs == NULL) {
-               adev->mman.buffer_funcs = &sdma_v5_2_buffer_funcs;
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-       }
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_2_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v5_2_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c +1 -2
index b005672f2f96..064508cecd11 100644
@@ -1895,8 +1895,7 @@ static const struct amdgpu_buffer_funcs sdma_v6_0_buffer_funcs = {
 
 static void sdma_v6_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v6_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v6_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v6_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +1 -2
index 5679a94d0815..60447729271e 100644
@@ -1845,8 +1845,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_0_buffer_funcs = {
 
 static void sdma_v7_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v7_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v7_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c +1 -2
index f20e0fc3fc74..c3428d2731dc 100644
@@ -1764,8 +1764,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_1_buffer_funcs = {
 
 static void sdma_v7_1_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v7_1_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_1_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v7_1_ip_block = {
drivers/gpu/drm/amd/amdgpu/si_dma.c +1 -2
index 3e58feb2d5e4..155067c20a0e 100644
@@ -833,8 +833,7 @@ static const struct amdgpu_buffer_funcs si_dma_buffer_funcs = {
 
 static void si_dma_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &si_dma_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &si_dma_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version si_dma_ip_block =
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +2 -2
index 964efa325908..28dc6886c1ff 100644
@@ -129,13 +129,14 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys,
                             struct dma_fence **mfence)
 {
        const u64 GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE;
-       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+       struct amdgpu_ring *ring;
        struct amdgpu_ttm_buffer_entity *entity;
        u64 gart_s, gart_d;
        struct dma_fence *next;
        u64 size;
        int r;
 
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
        entity = &adev->mman.move_entities[0];
 
        mutex_lock(&entity->lock);
--

2.43.0

maintainer
#18 lore raw
Quoted 132 lines · Alex Deucher
Quoted 130 lines · Pierre-Eric Pelloux-Prayer

This will allow the use of all of them for clear/fill buffer operations.
Since drm_sched_entity_init requires a scheduler array, we store schedulers rather than rings. For the few places that need access to a ring, we can get it from the sched using container_of.

Since the code is the same for all sdma versions, add a new helper amdgpu_sdma_set_buffer_funcs_scheds to set buffer_funcs_scheds based on the number of sdma instances.

Note: the new sched array is identical to the amdgpu_vm_manager one. These 2 could be merged.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c    |  4 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 32 ++++++++++++++++++----
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  3 +-
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c      |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  8 ++----
 drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c   |  6 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c     |  5 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c     |  5 +---
 drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c     |  3 +-
 drivers/gpu/drm/amd/amdgpu/si_dma.c        |  3 +-
 drivers/gpu/drm/amd/amdkfd/kfd_migrate.c   |  3 +-
 17 files changed, 48 insertions(+), 43 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 49e7881750fa..e3a8701f0b27 100644
@@ -1467,6 +1467,8 @@ ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring);
 ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset);
 void amdgpu_sdma_set_vm_pte_scheds(struct amdgpu_device *adev,
                                   const struct amdgpu_vm_pte_funcs *vm_pte_funcs);
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+                                        const struct amdgpu_buffer_funcs *buffer_funcs);
 
 /* atpx handler */
 #if defined(CONFIG_VGA_SWITCHEROO)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -1
index 0c0489395edf..4da8de34be3d 100644
@@ -3700,7 +3700,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
        adev->num_rings = 0;
        RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub());
        adev->mman.buffer_funcs = NULL;
-       adev->mman.buffer_funcs_ring = NULL;
+       adev->mman.num_buffer_funcs_scheds = 0;
        adev->vm_manager.vm_pte_funcs = NULL;
        adev->vm_manager.vm_pte_num_scheds = 0;
        adev->gmc.gmc_funcs = NULL;
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +3 -1
index e9e53c7c37d0..1a253e4257ff 100644
@@ -708,12 +708,14 @@ int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev)
 void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
                              uint32_t vmhub, uint32_t flush_type)
 {
-       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+       struct amdgpu_ring *ring;
        struct amdgpu_vmhub *hub = &adev->vmhub[vmhub];
        struct dma_fence *fence;
        struct amdgpu_job *job;
        int r;
 
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
+
        if (!hub->sdma_invalidation_workaround || vmid ||
            !adev->mman.buffer_funcs_enabled || !adev->ib_pool_ready ||
            !ring->sched.ready) {
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5 -6
index 3c63f87832e4..4ba7321b75e3 100644
@@ -168,7 +168,7 @@ amdgpu_ttm_job_submit(struct amdgpu_device *adev, struct amdgpu_ttm_buffer_entit
 {
        struct amdgpu_ring *ring;
 
-       ring = adev->mman.buffer_funcs_ring;
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
        amdgpu_ring_pad_ib(ring, &job->ibs[0]);
        WARN_ON(job->ibs[0].length_dw > num_dw);
 
@@ -2349,18 +2349,17 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
                return;
 
        if (enable) {
-               struct amdgpu_ring *ring;
                struct drm_gpu_scheduler *sched;
 
-               if (!adev->mman.buffer_funcs_ring || !adev->mman.buffer_funcs_ring->sched.ready) {
+               if (!adev->mman.num_buffer_funcs_scheds ||
+                   !adev->mman.buffer_funcs_scheds[0]->ready) {
                        dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
                        return;
                }
 
                num_clear_entities = 1;
                num_move_entities = 1;
-               ring = adev->mman.buffer_funcs_ring;
-               sched = &ring->sched;
+               sched = adev->mman.buffer_funcs_scheds[0];
                r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
                                                  &adev->mman.default_entity,
                                                  DRM_SCHED_PRIORITY_KERNEL,
@@ -2497,7 +2496,7 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
        unsigned int i;
        int r;
 
-       ring = adev->mman.buffer_funcs_ring;
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
 
        if (!ring->sched.ready) {
                dev_err(adev->dev,
@@ -2730,6 +2729,27 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
        return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
 }
 
Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type) · line 2735

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +4 -0
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+                                        const struct amdgpu_buffer_funcs *buffer_funcs)
+{
+       struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];

Only gfx9 and newer have multiple hubs. Additionally, sdma 4.x is attached to mmhub. sdma 5.x is attached to gfxhub. Might be worth a comment here that this is just for the sdma_invalidation_workaround check so it's only relevant for specific chips that have that set in case someone attempts to do something else here in the future.

IIRC this workaround only applies to Navi 1x and there the SDMA is connected to the gfxhub.

But I agree that this really needs at least a comment explaining why we do this.

Christian.

Quoted 227 lines

Alex

Quoted 224 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +17 -0
+       struct drm_gpu_scheduler *sched;
+       int i;
+
+       adev->mman.buffer_funcs = buffer_funcs;
+
+       for (i = 0; i < adev->sdma.num_instances; i++) {
+               if (adev->sdma.has_page_queue)
+                       sched = &adev->sdma.instance[i].page.sched;
+               else
+                       sched = &adev->sdma.instance[i].ring.sched;
+               adev->mman.buffer_funcs_scheds[i] = sched;
+       }
+
+       adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ?
+               1 : adev->sdma.num_instances;
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -1
index 3b1973611446..a6249252948b 100644
@@ -67,7 +67,8 @@ struct amdgpu_mman {
 
        /* buffer handling */
        const struct amdgpu_buffer_funcs        *buffer_funcs;
-       struct amdgpu_ring                      *buffer_funcs_ring;
+       struct drm_gpu_scheduler                *buffer_funcs_scheds[AMDGPU_MAX_RINGS];
+       u32                                     num_buffer_funcs_scheds;
        bool                                    buffer_funcs_enabled;
 
        /* @default_entity: for workarounds, has no gart windows */
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +1 -2
index 22780c09177d..26276dcfd458 100644
@@ -1340,8 +1340,7 @@ static const struct amdgpu_buffer_funcs cik_sdma_buffer_funcs = {
 
 static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &cik_sdma_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &cik_sdma_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version cik_sdma_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +1 -2
index 0090ace49024..c6a059ca59e5 100644
@@ -1235,8 +1235,7 @@ static const struct amdgpu_buffer_funcs sdma_v2_4_buffer_funcs = {
 
 static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v2_4_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v2_4_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +1 -2
index 2526d393162a..cb516a25210d 100644
@@ -1677,8 +1677,7 @@ static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = {
 
 static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v3_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v3_0_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -6
index 44f0f23e1148..d56be26f216b 100644
@@ -2626,13 +2626,9 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_buffer_funcs = {
 static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
        if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >= IP_VERSION(4, 4, 0))
-               adev->mman.buffer_funcs = &sdma_v4_4_buffer_funcs;
+               amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_buffer_funcs);
        else
-               adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
-       if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-       else
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+               amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_0_buffer_funcs);
 }
 
 static void sdma_v4_0_get_ras_error_count(uint32_t value,
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +1 -5
index 78bdfed0a7fd..67e9697301b4 100644
@@ -2316,11 +2316,7 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_2_buffer_funcs = {
 
 static void sdma_v4_4_2_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v4_4_2_buffer_funcs;
-       if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-       else
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_2_buffer_funcs);
 }
 
 /**
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -4
index 52f4e9e099cb..86f5eb784d57 100644
@@ -2052,10 +2052,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_0_buffer_funcs = {
 
 static void sdma_v5_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       if (adev->mman.buffer_funcs == NULL) {
-               adev->mman.buffer_funcs = &sdma_v5_0_buffer_funcs;
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-       }
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v5_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +1 -4
index b4fb90cc8f7d..3fec838374b2 100644
@@ -2056,10 +2056,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_2_buffer_funcs = {
 
 static void sdma_v5_2_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       if (adev->mman.buffer_funcs == NULL) {
-               adev->mman.buffer_funcs = &sdma_v5_2_buffer_funcs;
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-       }
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_2_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v5_2_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c +1 -2
index b005672f2f96..064508cecd11 100644
@@ -1895,8 +1895,7 @@ static const struct amdgpu_buffer_funcs sdma_v6_0_buffer_funcs = {
 
 static void sdma_v6_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v6_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v6_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v6_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +1 -2
index 5679a94d0815..60447729271e 100644
@@ -1845,8 +1845,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_0_buffer_funcs = {
 
 static void sdma_v7_0_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v7_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_0_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v7_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c +1 -2
index f20e0fc3fc74..c3428d2731dc 100644
@@ -1764,8 +1764,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_1_buffer_funcs = {
 
 static void sdma_v7_1_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &sdma_v7_1_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_1_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version sdma_v7_1_ip_block = {
drivers/gpu/drm/amd/amdgpu/si_dma.c +1 -2
index 3e58feb2d5e4..155067c20a0e 100644
@@ -833,8 +833,7 @@ static const struct amdgpu_buffer_funcs si_dma_buffer_funcs = {
 
 static void si_dma_set_buffer_funcs(struct amdgpu_device *adev)
 {
-       adev->mman.buffer_funcs = &si_dma_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &si_dma_buffer_funcs);
 }
 
 const struct amdgpu_ip_block_version si_dma_ip_block =
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +2 -2
index 964efa325908..28dc6886c1ff 100644
@@ -129,13 +129,14 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys,
                             struct dma_fence **mfence)
 {
        const u64 GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE;
-       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+       struct amdgpu_ring *ring;
        struct amdgpu_ttm_buffer_entity *entity;
        u64 gart_s, gart_d;
        struct dma_fence *next;
        u64 size;
        int r;
 
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
        entity = &adev->mman.move_entities[0];
 
        mutex_lock(&entity->lock);
--

2.43.0

#19 lore raw

Le 07/04/2026 à 10:07, Christian König a écrit :

Quoted 144 lines · Alex Deucher
Quoted 132 lines
Quoted 130 lines · Pierre-Eric Pelloux-Prayer

This will allow the use of all of them for clear/fill buffer operations.
Since drm_sched_entity_init requires a scheduler array, we store schedulers rather than rings. For the few places that need access to a ring, we can get it from the sched using container_of.

Since the code is the same for all sdma versions, add a new helper amdgpu_sdma_set_buffer_funcs_scheds to set buffer_funcs_scheds based on the number of sdma instances.

Note: the new sched array is identical to the amdgpu_vm_manager one. These 2 could be merged.

---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +-
  drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c    |  4 ++-
  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c    | 32 ++++++++++++++++++----
  drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h    |  3 +-
  drivers/gpu/drm/amd/amdgpu/cik_sdma.c      |  3 +-
  drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c     |  3 +-
  drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c     |  3 +-
  drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  8 ++----
  drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c   |  6 +---
  drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c     |  5 +---
  drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c     |  5 +---
  drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c     |  3 +-
  drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c     |  3 +-
  drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c     |  3 +-
  drivers/gpu/drm/amd/amdgpu/si_dma.c        |  3 +-
  drivers/gpu/drm/amd/amdkfd/kfd_migrate.c   |  3 +-
  17 files changed, 48 insertions(+), 43 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +2 -0
index 49e7881750fa..e3a8701f0b27 100644
@@ -1467,6 +1467,8 @@ ssize_t amdgpu_get_soft_full_reset_mask(struct amdgpu_ring *ring);
  ssize_t amdgpu_show_reset_mask(char *buf, uint32_t supported_reset);
  void amdgpu_sdma_set_vm_pte_scheds(struct amdgpu_device *adev,
                                    const struct amdgpu_vm_pte_funcs *vm_pte_funcs);
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+                                        const struct amdgpu_buffer_funcs *buffer_funcs);
 
  /* atpx handler */
  #if defined(CONFIG_VGA_SWITCHEROO)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +1 -1
index 0c0489395edf..4da8de34be3d 100644
@@ -3700,7 +3700,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
         adev->num_rings = 0;
         RCU_INIT_POINTER(adev->gang_submit, dma_fence_get_stub());
         adev->mman.buffer_funcs = NULL;
-       adev->mman.buffer_funcs_ring = NULL;
+       adev->mman.num_buffer_funcs_scheds = 0;
         adev->vm_manager.vm_pte_funcs = NULL;
         adev->vm_manager.vm_pte_num_scheds = 0;
         adev->gmc.gmc_funcs = NULL;
drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.c +3 -1
index e9e53c7c37d0..1a253e4257ff 100644
@@ -708,12 +708,14 @@ int amdgpu_gmc_allocate_vm_inv_eng(struct amdgpu_device *adev)
  void amdgpu_gmc_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
                               uint32_t vmhub, uint32_t flush_type)
  {
-       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+       struct amdgpu_ring *ring;
         struct amdgpu_vmhub *hub = &adev->vmhub[vmhub];
         struct dma_fence *fence;
         struct amdgpu_job *job;
         int r;
 
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
+
         if (!hub->sdma_invalidation_workaround || vmid ||
             !adev->mman.buffer_funcs_enabled || !adev->ib_pool_ready ||
             !ring->sched.ready) {
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +5 -6
index 3c63f87832e4..4ba7321b75e3 100644
@@ -168,7 +168,7 @@ amdgpu_ttm_job_submit(struct amdgpu_device *adev, struct amdgpu_ttm_buffer_entit
  {
         struct amdgpu_ring *ring;
 
-       ring = adev->mman.buffer_funcs_ring;
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
         amdgpu_ring_pad_ib(ring, &job->ibs[0]);
         WARN_ON(job->ibs[0].length_dw > num_dw);
 
@@ -2349,18 +2349,17 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
                 return;
 
         if (enable) {
-               struct amdgpu_ring *ring;
                 struct drm_gpu_scheduler *sched;
 
-               if (!adev->mman.buffer_funcs_ring || !adev->mman.buffer_funcs_ring->sched.ready) {
+               if (!adev->mman.num_buffer_funcs_scheds ||
+                   !adev->mman.buffer_funcs_scheds[0]->ready) {
                         dev_warn(adev->dev, "Not enabling DMA transfers for in kernel use");
                         return;
                 }
 
                 num_clear_entities = 1;
                 num_move_entities = 1;
-               ring = adev->mman.buffer_funcs_ring;
-               sched = &ring->sched;
+               sched = adev->mman.buffer_funcs_scheds[0];
                 r = amdgpu_ttm_buffer_entity_init(&adev->mman.gtt_mgr,
                                                   &adev->mman.default_entity,
                                                   DRM_SCHED_PRIORITY_KERNEL,
@@ -2497,7 +2496,7 @@ int amdgpu_copy_buffer(struct amdgpu_device *adev,
         unsigned int i;
         int r;
 
-       ring = adev->mman.buffer_funcs_ring;
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
 
         if (!ring->sched.ready) {
                 dev_err(adev->dev,
@@ -2730,6 +2729,27 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
         return ttm_resource_manager_evict_all(&adev->mman.bdev, man);
  }
 
Quoted 10 lines

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type) · line 2735

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +4 -0
+void amdgpu_sdma_set_buffer_funcs_scheds(struct amdgpu_device *adev,
+                                        const struct amdgpu_buffer_funcs *buffer_funcs)
+{
+       struct amdgpu_vmhub *hub = &adev->vmhub[AMDGPU_GFXHUB(0)];

Only gfx9 and newer have multiple hubs. Additionally, sdma 4.x is attached to mmhub. sdma 5.x is attached to gfxhub. Might be worth a comment here that this is just for the sdma_invalidation_workaround check so it's only relevant for specific chips that have that set in case someone attempts to do something else here in the future.

Quoted 3 lines

IIRC this workaround only applies to Navi 1x and there the SDMA is connected to the gfxhub.

But I agree that this really needs at least a comment explaining why we do this.

I added a comment to v6 of this patch, thanks.

Pierre-Eric

Quoted 230 lines

Christian.

Quoted 227 lines

Alex

Quoted 224 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +17 -0
+       struct drm_gpu_scheduler *sched;
+       int i;
+
+       adev->mman.buffer_funcs = buffer_funcs;
+
+       for (i = 0; i < adev->sdma.num_instances; i++) {
+               if (adev->sdma.has_page_queue)
+                       sched = &adev->sdma.instance[i].page.sched;
+               else
+                       sched = &adev->sdma.instance[i].ring.sched;
+               adev->mman.buffer_funcs_scheds[i] = sched;
+       }
+
+       adev->mman.num_buffer_funcs_scheds = hub->sdma_invalidation_workaround ?
+               1 : adev->sdma.num_instances;
+}
+
  #if defined(CONFIG_DEBUG_FS)
 
  static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +2 -1
index 3b1973611446..a6249252948b 100644
@@ -67,7 +67,8 @@ struct amdgpu_mman {
 
         /* buffer handling */
         const struct amdgpu_buffer_funcs        *buffer_funcs;
-       struct amdgpu_ring                      *buffer_funcs_ring;
+       struct drm_gpu_scheduler                *buffer_funcs_scheds[AMDGPU_MAX_RINGS];
+       u32                                     num_buffer_funcs_scheds;
         bool                                    buffer_funcs_enabled;
 
         /* @default_entity: for workarounds, has no gart windows */
drivers/gpu/drm/amd/amdgpu/cik_sdma.c +1 -2
index 22780c09177d..26276dcfd458 100644
@@ -1340,8 +1340,7 @@ static const struct amdgpu_buffer_funcs cik_sdma_buffer_funcs = {
 
  static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &cik_sdma_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &cik_sdma_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version cik_sdma_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c +1 -2
index 0090ace49024..c6a059ca59e5 100644
@@ -1235,8 +1235,7 @@ static const struct amdgpu_buffer_funcs sdma_v2_4_buffer_funcs = {
 
  static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v2_4_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v2_4_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c +1 -2
index 2526d393162a..cb516a25210d 100644
@@ -1677,8 +1677,7 @@ static const struct amdgpu_buffer_funcs sdma_v3_0_buffer_funcs = {
 
  static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v3_0_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v3_0_ip_block =
drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c +2 -6
index 44f0f23e1148..d56be26f216b 100644
@@ -2626,13 +2626,9 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_buffer_funcs = {
  static void sdma_v4_0_set_buffer_funcs(struct amdgpu_device *adev)
  {
         if (amdgpu_ip_version(adev, SDMA0_HWIP, 0) >= IP_VERSION(4, 4, 0))
-               adev->mman.buffer_funcs = &sdma_v4_4_buffer_funcs;
+               amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_buffer_funcs);
         else
-               adev->mman.buffer_funcs = &sdma_v4_0_buffer_funcs;
-       if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-       else
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+               amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_0_buffer_funcs);
  }
 
  static void sdma_v4_0_get_ras_error_count(uint32_t value,
drivers/gpu/drm/amd/amdgpu/sdma_v4_4_2.c +1 -5
index 78bdfed0a7fd..67e9697301b4 100644
@@ -2316,11 +2316,7 @@ static const struct amdgpu_buffer_funcs sdma_v4_4_2_buffer_funcs = {
 
  static void sdma_v4_4_2_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &sdma_v4_4_2_buffer_funcs;
-       if (adev->sdma.has_page_queue)
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].page;
-       else
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v4_4_2_buffer_funcs);
  }
 
  /**
drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c +1 -4
index 52f4e9e099cb..86f5eb784d57 100644
@@ -2052,10 +2052,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_0_buffer_funcs = {
 
  static void sdma_v5_0_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       if (adev->mman.buffer_funcs == NULL) {
-               adev->mman.buffer_funcs = &sdma_v5_0_buffer_funcs;
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-       }
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_0_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v5_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v5_2.c +1 -4
index b4fb90cc8f7d..3fec838374b2 100644
@@ -2056,10 +2056,7 @@ static const struct amdgpu_buffer_funcs sdma_v5_2_buffer_funcs = {
 
  static void sdma_v5_2_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       if (adev->mman.buffer_funcs == NULL) {
-               adev->mman.buffer_funcs = &sdma_v5_2_buffer_funcs;
-               adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
-       }
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v5_2_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v5_2_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v6_0.c +1 -2
index b005672f2f96..064508cecd11 100644
@@ -1895,8 +1895,7 @@ static const struct amdgpu_buffer_funcs sdma_v6_0_buffer_funcs = {
 
  static void sdma_v6_0_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &sdma_v6_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v6_0_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v6_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_0.c +1 -2
index 5679a94d0815..60447729271e 100644
@@ -1845,8 +1845,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_0_buffer_funcs = {
 
  static void sdma_v7_0_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &sdma_v7_0_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_0_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v7_0_ip_block = {
drivers/gpu/drm/amd/amdgpu/sdma_v7_1.c +1 -2
index f20e0fc3fc74..c3428d2731dc 100644
@@ -1764,8 +1764,7 @@ static const struct amdgpu_buffer_funcs sdma_v7_1_buffer_funcs = {
 
  static void sdma_v7_1_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &sdma_v7_1_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &sdma_v7_1_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version sdma_v7_1_ip_block = {
drivers/gpu/drm/amd/amdgpu/si_dma.c +1 -2
index 3e58feb2d5e4..155067c20a0e 100644
@@ -833,8 +833,7 @@ static const struct amdgpu_buffer_funcs si_dma_buffer_funcs = {
 
  static void si_dma_set_buffer_funcs(struct amdgpu_device *adev)
  {
-       adev->mman.buffer_funcs = &si_dma_buffer_funcs;
-       adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
+       amdgpu_sdma_set_buffer_funcs_scheds(adev, &si_dma_buffer_funcs);
  }
 
  const struct amdgpu_ip_block_version si_dma_ip_block =
drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +2 -2
index 964efa325908..28dc6886c1ff 100644
@@ -129,13 +129,14 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys,
                              struct dma_fence **mfence)
  {
         const u64 GTT_MAX_PAGES = AMDGPU_GTT_MAX_TRANSFER_SIZE;
-       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
+       struct amdgpu_ring *ring;
         struct amdgpu_ttm_buffer_entity *entity;
         u64 gart_s, gart_d;
         struct dma_fence *next;
         u64 size;
         int r;
 
+       ring = to_amdgpu_ring(adev->mman.buffer_funcs_scheds[0]);
         entity = &adev->mman.move_entities[0];
 
         mutex_lock(&entity->lock);
--

2.43.0