[PATCH 1/7] drm/amdgpu: Avoid reclaim fs while eviction lock

[PATCH 1/7] drm/amdgpu: Avoid reclaim fs while eviction lock

[Alex Sierra]

[Why]
Avoid reclaim filesystem while eviction lock is held called from
MMU notifier.

[How]
Setting PF_MEMALLOC_NOFS flags while eviction mutex is locked.
Using memalloc_nofs_save / memalloc_nofs_restore API.

Change-Id: I5531c9337836e7d4a430df3f16dcc82888e8018c
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 40 +++++++++++++++++++++-----
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |  6 +++-
 2 files changed, 38 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index b999b67ff57a..d6aba4f9df74 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -82,6 +82,32 @@ struct amdgpu_prt_cb {
 	struct dma_fence_cb cb;
 };
 
+/**
+ * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
+ * happens while holding this lock anywhere to prevent deadlocks when
+ * an MMU notifier runs in reclaim-FS context.
+ */
+static inline void amdgpu_vm_eviction_lock(struct amdgpu_vm *vm)
+{
+	mutex_lock(&vm->eviction_lock);
+	vm->saved_flags = memalloc_nofs_save();
+}
+
+static inline int amdgpu_vm_eviction_trylock(struct amdgpu_vm *vm)
+{
+	if (mutex_trylock(&vm->eviction_lock)) {
+		vm->saved_flags = memalloc_nofs_save();
+		return 1;
+	}
+	return 0;
+}
+
+static inline void amdgpu_vm_eviction_unlock(struct amdgpu_vm *vm)
+{
+	memalloc_nofs_restore(vm->saved_flags);
+	mutex_unlock(&vm->eviction_lock);
+}
+
 /**
  * amdgpu_vm_level_shift - return the addr shift for each level
  *
@@ -678,9 +704,9 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		}
 	}
 
-	mutex_lock(&vm->eviction_lock);
+	amdgpu_vm_eviction_lock(vm);
 	vm->evicting = false;
-	mutex_unlock(&vm->eviction_lock);
+	amdgpu_vm_eviction_unlock(vm);
 
 	return 0;
 }
@@ -1559,7 +1585,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	if (!(flags & AMDGPU_PTE_VALID))
 		owner = AMDGPU_FENCE_OWNER_KFD;
 
-	mutex_lock(&vm->eviction_lock);
+	amdgpu_vm_eviction_lock(vm);
 	if (vm->evicting) {
 		r = -EBUSY;
 		goto error_unlock;
@@ -1576,7 +1602,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 	r = vm->update_funcs->commit(&params, fence);
 
 error_unlock:
-	mutex_unlock(&vm->eviction_lock);
+	amdgpu_vm_eviction_unlock(vm);
 	return r;
 }
 
@@ -2537,18 +2563,18 @@ bool amdgpu_vm_evictable(struct amdgpu_bo *bo)
 		return false;
 
 	/* Try to block ongoing updates */
-	if (!mutex_trylock(&bo_base->vm->eviction_lock))
+	if (!amdgpu_vm_eviction_trylock(bo_base->vm))
 		return false;
 
 	/* Don't evict VM page tables while they are updated */
 	if (!dma_fence_is_signaled(bo_base->vm->last_direct) ||
 	    !dma_fence_is_signaled(bo_base->vm->last_delayed)) {
-		mutex_unlock(&bo_base->vm->eviction_lock);
+		amdgpu_vm_eviction_unlock(bo_base->vm);
 		return false;
 	}
 
 	bo_base->vm->evicting = true;
-	mutex_unlock(&bo_base->vm->eviction_lock);
+	amdgpu_vm_eviction_unlock(bo_base->vm);
 	return true;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 100547f094ff..c21a36bebc0c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -30,6 +30,7 @@
 #include <drm/gpu_scheduler.h>
 #include <drm/drm_file.h>
 #include <drm/ttm/ttm_bo_driver.h>
+#include <linux/sched/mm.h>
 
 #include "amdgpu_sync.h"
 #include "amdgpu_ring.h"
@@ -242,9 +243,12 @@ struct amdgpu_vm {
 	/* tree of virtual addresses mapped */
 	struct rb_root_cached	va;
 
-	/* Lock to prevent eviction while we are updating page tables */
+	/* Lock to prevent eviction while we are updating page tables
+	 * use vm_eviction_lock/unlock(vm)
+	 */
 	struct mutex		eviction_lock;
 	bool			evicting;
+	unsigned int		saved_flags;
 
 	/* BOs who needs a validation */
 	struct list_head	evicted;
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 2/7] drm/amdgpu: kiq pm4 function implementation for gfx_v9

[Alex Sierra]

Functions implemented from kiq_pm4_funcs struct members
for gfx_v9 version.

Change-Id: I8fd3e160c4bd58f19d35d29e39517db967063afe
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 ++++++++++++++++++++++++++
 1 file changed, 115 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index e3d466bd5c4e..ad0179ea2cc5 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -739,6 +739,120 @@ static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
 static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
 				     void *inject_if);
 
+static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
+				uint64_t queue_mask)
+{
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
+	amdgpu_ring_write(kiq_ring,
+		PACKET3_SET_RESOURCES_VMID_MASK(0) |
+		/* vmid_mask:0* queue_type:0 (KIQ) */
+		PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
+	amdgpu_ring_write(kiq_ring,
+			lower_32_bits(queue_mask));	/* queue mask lo */
+	amdgpu_ring_write(kiq_ring,
+			upper_32_bits(queue_mask));	/* queue mask hi */
+	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
+	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
+	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
+	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
+}
+
+static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
+				 struct amdgpu_ring *ring)
+{
+	struct amdgpu_device *adev = kiq_ring->adev;
+	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
+	uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
+	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
+	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
+	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
+			 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
+			 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
+			 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
+			 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
+			 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
+			 /*queue_type: normal compute queue */
+			 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
+			 /* alloc format: all_on_one_pipe */
+			 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
+			 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
+			 /* num_queues: must be 1 */
+			 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
+	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
+	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
+	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
+	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
+}
+
+static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
+				   struct amdgpu_ring *ring,
+				   enum amdgpu_unmap_queues_action action,
+				   u64 gpu_addr, u64 seq)
+{
+	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
+	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
+			  PACKET3_UNMAP_QUEUES_ACTION(action) |
+			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
+			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
+			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
+
+	if (action == PREEMPT_QUEUES_NO_UNMAP) {
+		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
+		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
+		amdgpu_ring_write(kiq_ring, seq);
+	} else {
+		amdgpu_ring_write(kiq_ring, 0);
+		amdgpu_ring_write(kiq_ring, 0);
+		amdgpu_ring_write(kiq_ring, 0);
+	}
+}
+
+static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
+				   struct amdgpu_ring *ring,
+				   u64 addr,
+				   u64 seq)
+{
+	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
+
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
+	amdgpu_ring_write(kiq_ring,
+			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
+			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
+			  PACKET3_QUERY_STATUS_COMMAND(2));
+	/* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
+			PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
+	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
+	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
+	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
+	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
+}
+
+static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
+	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
+	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
+	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
+	.kiq_query_status = gfx_v9_0_kiq_query_status,
+	.set_resources_size = 8,
+	.map_queues_size = 7,
+	.unmap_queues_size = 6,
+	.query_status_size = 7,
+};
+
+static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
+{
+	adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs;
+}
+
 static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
 {
 	switch (adev->asic_type) {
@@ -4260,6 +4374,7 @@ static int gfx_v9_0_early_init(void *handle)
 	else
 		adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
 	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
+	gfx_v9_0_set_kiq_pm4_funcs(adev);
 	gfx_v9_0_set_ring_funcs(adev);
 	gfx_v9_0_set_irq_funcs(adev);
 	gfx_v9_0_set_gds_init(adev);
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 2/7] drm/amdgpu: kiq pm4 function implementation for gfx_v9

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> Functions implemented from kiq_pm4_funcs struct members
> for gfx_v9 version.
>
> Change-Id: I8fd3e160c4bd58f19d35d29e39517db967063afe
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Acked-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 ++++++++++++++++++++++++++
>   1 file changed, 115 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index e3d466bd5c4e..ad0179ea2cc5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -739,6 +739,120 @@ static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
>   static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
>   				     void *inject_if);
>   
> +static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
> +				uint64_t queue_mask)
> +{
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
> +	amdgpu_ring_write(kiq_ring,
> +		PACKET3_SET_RESOURCES_VMID_MASK(0) |
> +		/* vmid_mask:0* queue_type:0 (KIQ) */
> +		PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
> +	amdgpu_ring_write(kiq_ring,
> +			lower_32_bits(queue_mask));	/* queue mask lo */
> +	amdgpu_ring_write(kiq_ring,
> +			upper_32_bits(queue_mask));	/* queue mask hi */
> +	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
> +	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
> +	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
> +	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
> +}
> +
> +static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
> +				 struct amdgpu_ring *ring)
> +{
> +	struct amdgpu_device *adev = kiq_ring->adev;
> +	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
> +	uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
> +	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
> +
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
> +	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
> +	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> +			 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
> +			 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
> +			 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
> +			 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
> +			 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
> +			 /*queue_type: normal compute queue */
> +			 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
> +			 /* alloc format: all_on_one_pipe */
> +			 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
> +			 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
> +			 /* num_queues: must be 1 */
> +			 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
> +}
> +
> +static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
> +				   struct amdgpu_ring *ring,
> +				   enum amdgpu_unmap_queues_action action,
> +				   u64 gpu_addr, u64 seq)
> +{
> +	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
> +
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
> +	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> +			  PACKET3_UNMAP_QUEUES_ACTION(action) |
> +			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
> +			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
> +			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
> +
> +	if (action == PREEMPT_QUEUES_NO_UNMAP) {
> +		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
> +		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
> +		amdgpu_ring_write(kiq_ring, seq);
> +	} else {
> +		amdgpu_ring_write(kiq_ring, 0);
> +		amdgpu_ring_write(kiq_ring, 0);
> +		amdgpu_ring_write(kiq_ring, 0);
> +	}
> +}
> +
> +static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
> +				   struct amdgpu_ring *ring,
> +				   u64 addr,
> +				   u64 seq)
> +{
> +	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
> +
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
> +	amdgpu_ring_write(kiq_ring,
> +			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
> +			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
> +			  PACKET3_QUERY_STATUS_COMMAND(2));
> +	/* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
> +			PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
> +}
> +
> +static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
> +	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
> +	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
> +	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
> +	.kiq_query_status = gfx_v9_0_kiq_query_status,
> +	.set_resources_size = 8,
> +	.map_queues_size = 7,
> +	.unmap_queues_size = 6,
> +	.query_status_size = 7,
> +};
> +
> +static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
> +{
> +	adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs;
> +}
> +
>   static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
>   {
>   	switch (adev->asic_type) {
> @@ -4260,6 +4374,7 @@ static int gfx_v9_0_early_init(void *handle)
>   	else
>   		adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
>   	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
> +	gfx_v9_0_set_kiq_pm4_funcs(adev);
>   	gfx_v9_0_set_ring_funcs(adev);
>   	gfx_v9_0_set_irq_funcs(adev);
>   	gfx_v9_0_set_gds_init(adev);

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 2/7] drm/amdgpu: kiq pm4 function implementation for gfx_v9

[Felix Kuehling]

On 2020-01-11 1:39 p.m., Alex Sierra wrote:
> Functions implemented from kiq_pm4_funcs struct members
> for gfx_v9 version.
>
> Change-Id: I8fd3e160c4bd58f19d35d29e39517db967063afe
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>


> ---
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 115 ++++++++++++++++++++++++++
>   1 file changed, 115 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index e3d466bd5c4e..ad0179ea2cc5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -739,6 +739,120 @@ static int gfx_v9_0_query_ras_error_count(struct amdgpu_device *adev,
>   static int gfx_v9_0_ras_error_inject(struct amdgpu_device *adev,
>   				     void *inject_if);
>   
> +static void gfx_v9_0_kiq_set_resources(struct amdgpu_ring *kiq_ring,
> +				uint64_t queue_mask)
> +{
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
> +	amdgpu_ring_write(kiq_ring,
> +		PACKET3_SET_RESOURCES_VMID_MASK(0) |
> +		/* vmid_mask:0* queue_type:0 (KIQ) */
> +		PACKET3_SET_RESOURCES_QUEUE_TYPE(0));
> +	amdgpu_ring_write(kiq_ring,
> +			lower_32_bits(queue_mask));	/* queue mask lo */
> +	amdgpu_ring_write(kiq_ring,
> +			upper_32_bits(queue_mask));	/* queue mask hi */
> +	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
> +	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
> +	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
> +	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
> +}
> +
> +static void gfx_v9_0_kiq_map_queues(struct amdgpu_ring *kiq_ring,
> +				 struct amdgpu_ring *ring)
> +{
> +	struct amdgpu_device *adev = kiq_ring->adev;
> +	uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
> +	uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
> +	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
> +
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
> +	/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
> +	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> +			 PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
> +			 PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
> +			 PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
> +			 PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
> +			 PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
> +			 /*queue_type: normal compute queue */
> +			 PACKET3_MAP_QUEUES_QUEUE_TYPE(0) |
> +			 /* alloc format: all_on_one_pipe */
> +			 PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) |
> +			 PACKET3_MAP_QUEUES_ENGINE_SEL(eng_sel) |
> +			 /* num_queues: must be 1 */
> +			 PACKET3_MAP_QUEUES_NUM_QUEUES(1));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
> +}
> +
> +static void gfx_v9_0_kiq_unmap_queues(struct amdgpu_ring *kiq_ring,
> +				   struct amdgpu_ring *ring,
> +				   enum amdgpu_unmap_queues_action action,
> +				   u64 gpu_addr, u64 seq)
> +{
> +	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
> +
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_UNMAP_QUEUES, 4));
> +	amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> +			  PACKET3_UNMAP_QUEUES_ACTION(action) |
> +			  PACKET3_UNMAP_QUEUES_QUEUE_SEL(0) |
> +			  PACKET3_UNMAP_QUEUES_ENGINE_SEL(eng_sel) |
> +			  PACKET3_UNMAP_QUEUES_NUM_QUEUES(1));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_UNMAP_QUEUES_DOORBELL_OFFSET0(ring->doorbell_index));
> +
> +	if (action == PREEMPT_QUEUES_NO_UNMAP) {
> +		amdgpu_ring_write(kiq_ring, lower_32_bits(gpu_addr));
> +		amdgpu_ring_write(kiq_ring, upper_32_bits(gpu_addr));
> +		amdgpu_ring_write(kiq_ring, seq);
> +	} else {
> +		amdgpu_ring_write(kiq_ring, 0);
> +		amdgpu_ring_write(kiq_ring, 0);
> +		amdgpu_ring_write(kiq_ring, 0);
> +	}
> +}
> +
> +static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
> +				   struct amdgpu_ring *ring,
> +				   u64 addr,
> +				   u64 seq)
> +{
> +	uint32_t eng_sel = ring->funcs->type == AMDGPU_RING_TYPE_GFX ? 4 : 0;
> +
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_QUERY_STATUS, 5));
> +	amdgpu_ring_write(kiq_ring,
> +			  PACKET3_QUERY_STATUS_CONTEXT_ID(0) |
> +			  PACKET3_QUERY_STATUS_INTERRUPT_SEL(0) |
> +			  PACKET3_QUERY_STATUS_COMMAND(2));
> +	/* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_QUERY_STATUS_DOORBELL_OFFSET(ring->doorbell_index) |
> +			PACKET3_QUERY_STATUS_ENG_SEL(eng_sel));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(addr));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(addr));
> +	amdgpu_ring_write(kiq_ring, lower_32_bits(seq));
> +	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
> +}
> +
> +static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
> +	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
> +	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
> +	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
> +	.kiq_query_status = gfx_v9_0_kiq_query_status,
> +	.set_resources_size = 8,
> +	.map_queues_size = 7,
> +	.unmap_queues_size = 6,
> +	.query_status_size = 7,
> +};
> +
> +static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
> +{
> +	adev->gfx.kiq.pmf = &gfx_v9_0_kiq_pm4_funcs;
> +}
> +
>   static void gfx_v9_0_init_golden_registers(struct amdgpu_device *adev)
>   {
>   	switch (adev->asic_type) {
> @@ -4260,6 +4374,7 @@ static int gfx_v9_0_early_init(void *handle)
>   	else
>   		adev->gfx.num_gfx_rings = GFX9_NUM_GFX_RINGS;
>   	adev->gfx.num_compute_rings = AMDGPU_MAX_COMPUTE_RINGS;
> +	gfx_v9_0_set_kiq_pm4_funcs(adev);
>   	gfx_v9_0_set_ring_funcs(adev);
>   	gfx_v9_0_set_irq_funcs(adev);
>   	gfx_v9_0_set_gds_init(adev);
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/7] drm/amdgpu: implement tlbs invalidate on gfx9 gfx10

[Alex Sierra]

tlbs invalidate pointer function added to kiq_pm4_funcs struct.
This way, tlb flush can be done through kiq member.
TLBs invalidatation implemented for gfx9 and gfx10.

Change-Id: I1b77b364f3ae0038ff3e70e869be5f2ef6e6d293
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  4 +++
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 33 +++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 32 ++++++++++++++++++++++++
 3 files changed, 69 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index 8e88e0411662..2927837bd401 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -76,11 +76,15 @@ struct kiq_pm4_funcs {
 					struct amdgpu_ring *ring,
 					u64 addr,
 					u64 seq);
+	int (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub);
 	/* Packet sizes */
 	int set_resources_size;
 	int map_queues_size;
 	int unmap_queues_size;
 	int query_status_size;
+	int invalidate_tlbs_size;
 };
 
 struct amdgpu_kiq {
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
index 379e46c1b7f6..2e82213f57eb 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
@@ -40,6 +40,7 @@
 #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
 
 #include "soc15.h"
+#include "soc15d.h"
 #include "soc15_common.h"
 #include "clearstate_gfx10.h"
 #include "v10_structs.h"
@@ -346,15 +347,47 @@ static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
 }
 
+static int gfx10_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub)
+{
+	signed long r;
+	uint32_t seq;
+	struct amdgpu_device *adev = kiq_ring->adev;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
+
+	spin_lock(&adev->gfx.kiq.ring_lock);
+	amdgpu_ring_alloc(kiq_ring, kiq->pmf->invalidate_tlbs_size);
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
+			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
+			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
+			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
+	amdgpu_fence_emit_polling(kiq_ring, &seq);
+	amdgpu_ring_commit(kiq_ring);
+	spin_unlock(&adev->gfx.kiq.ring_lock);
+
+	r = amdgpu_fence_wait_polling(kiq_ring, seq, adev->usec_timeout);
+	if (r < 1) {
+		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
+		return -ETIME;
+	}
+
+	return 0;
+}
+
 static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
 	.kiq_set_resources = gfx10_kiq_set_resources,
 	.kiq_map_queues = gfx10_kiq_map_queues,
 	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
 	.kiq_query_status = gfx10_kiq_query_status,
+	.kiq_invalidate_tlbs = gfx10_kiq_invalidate_tlbs,
 	.set_resources_size = 8,
 	.map_queues_size = 7,
 	.unmap_queues_size = 6,
 	.query_status_size = 7,
+	.invalidate_tlbs_size = 12,
 };
 
 static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index ad0179ea2cc5..5be6fab55b73 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -837,15 +837,47 @@ static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
 }
 
+static int gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub)
+{
+	signed long r;
+	uint32_t seq;
+	struct amdgpu_device *adev = kiq_ring->adev;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
+
+	spin_lock(&adev->gfx.kiq.ring_lock);
+	amdgpu_ring_alloc(kiq_ring, kiq->pmf->invalidate_tlbs_size);
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
+			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
+			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
+			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
+	amdgpu_fence_emit_polling(kiq_ring, &seq);
+	amdgpu_ring_commit(kiq_ring);
+	spin_unlock(&adev->gfx.kiq.ring_lock);
+
+	r = amdgpu_fence_wait_polling(kiq_ring, seq, adev->usec_timeout);
+	if (r < 1) {
+		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
+		return -ETIME;
+	}
+
+	return 0;
+}
+
 static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
 	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
 	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
 	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
 	.kiq_query_status = gfx_v9_0_kiq_query_status,
+	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
 	.set_resources_size = 8,
 	.map_queues_size = 7,
 	.unmap_queues_size = 6,
 	.query_status_size = 7,
+	.invalidate_tlbs_size = 12,
 };
 
 static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 3/7] drm/amdgpu: implement tlbs invalidate on gfx9 gfx10

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> tlbs invalidate pointer function added to kiq_pm4_funcs struct.
> This way, tlb flush can be done through kiq member.
> TLBs invalidatation implemented for gfx9 and gfx10.
>
> Change-Id: I1b77b364f3ae0038ff3e70e869be5f2ef6e6d293
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Please note that I can't judge the correctness of the PM4 packets, but 
the interface looks really nice and clean now.

Reviewed-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  4 +++
>   drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 33 +++++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 32 ++++++++++++++++++++++++
>   3 files changed, 69 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> index 8e88e0411662..2927837bd401 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> @@ -76,11 +76,15 @@ struct kiq_pm4_funcs {
>   					struct amdgpu_ring *ring,
>   					u64 addr,
>   					u64 seq);
> +	int (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub);
>   	/* Packet sizes */
>   	int set_resources_size;
>   	int map_queues_size;
>   	int unmap_queues_size;
>   	int query_status_size;
> +	int invalidate_tlbs_size;
>   };
>   
>   struct amdgpu_kiq {
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> index 379e46c1b7f6..2e82213f57eb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> @@ -40,6 +40,7 @@
>   #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   #include "clearstate_gfx10.h"
>   #include "v10_structs.h"
> @@ -346,15 +347,47 @@ static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static int gfx10_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	signed long r;
> +	uint32_t seq;
> +	struct amdgpu_device *adev = kiq_ring->adev;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	spin_lock(&adev->gfx.kiq.ring_lock);
> +	amdgpu_ring_alloc(kiq_ring, kiq->pmf->invalidate_tlbs_size);
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +	amdgpu_fence_emit_polling(kiq_ring, &seq);
> +	amdgpu_ring_commit(kiq_ring);
> +	spin_unlock(&adev->gfx.kiq.ring_lock);
> +
> +	r = amdgpu_fence_wait_polling(kiq_ring, seq, adev->usec_timeout);
> +	if (r < 1) {
> +		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
> +		return -ETIME;
> +	}
> +
> +	return 0;
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx10_kiq_set_resources,
>   	.kiq_map_queues = gfx10_kiq_map_queues,
>   	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
>   	.kiq_query_status = gfx10_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx10_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index ad0179ea2cc5..5be6fab55b73 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -837,15 +837,47 @@ static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static int gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	signed long r;
> +	uint32_t seq;
> +	struct amdgpu_device *adev = kiq_ring->adev;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	spin_lock(&adev->gfx.kiq.ring_lock);
> +	amdgpu_ring_alloc(kiq_ring, kiq->pmf->invalidate_tlbs_size);
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +	amdgpu_fence_emit_polling(kiq_ring, &seq);
> +	amdgpu_ring_commit(kiq_ring);
> +	spin_unlock(&adev->gfx.kiq.ring_lock);
> +
> +	r = amdgpu_fence_wait_polling(kiq_ring, seq, adev->usec_timeout);
> +	if (r < 1) {
> +		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
> +		return -ETIME;
> +	}
> +
> +	return 0;
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
>   	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
>   	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
>   	.kiq_query_status = gfx_v9_0_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 3/7] drm/amdgpu: implement tlbs invalidate on gfx9 gfx10

[Felix Kuehling]

On 2020-01-11 1:39 p.m., Alex Sierra wrote:
> tlbs invalidate pointer function added to kiq_pm4_funcs struct.
> This way, tlb flush can be done through kiq member.
> TLBs invalidatation implemented for gfx9 and gfx10.
>
> Change-Id: I1b77b364f3ae0038ff3e70e869be5f2ef6e6d293
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  4 +++
>   drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 33 +++++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 32 ++++++++++++++++++++++++
>   3 files changed, 69 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> index 8e88e0411662..2927837bd401 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> @@ -76,11 +76,15 @@ struct kiq_pm4_funcs {
>   					struct amdgpu_ring *ring,
>   					u64 addr,
>   					u64 seq);
> +	int (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub);
>   	/* Packet sizes */
>   	int set_resources_size;
>   	int map_queues_size;
>   	int unmap_queues_size;
>   	int query_status_size;
> +	int invalidate_tlbs_size;
>   };
>   
>   struct amdgpu_kiq {
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> index 379e46c1b7f6..2e82213f57eb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> @@ -40,6 +40,7 @@
>   #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   #include "clearstate_gfx10.h"
>   #include "v10_structs.h"
> @@ -346,15 +347,47 @@ static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static int gfx10_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	signed long r;
> +	uint32_t seq;
> +	struct amdgpu_device *adev = kiq_ring->adev;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	spin_lock(&adev->gfx.kiq.ring_lock);
> +	amdgpu_ring_alloc(kiq_ring, kiq->pmf->invalidate_tlbs_size);
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +	amdgpu_fence_emit_polling(kiq_ring, &seq);
> +	amdgpu_ring_commit(kiq_ring);
> +	spin_unlock(&adev->gfx.kiq.ring_lock);

The other KIQ functions don't include the emit_polling, commit and 
locking. I think the way the KIQ-funcs interface is meant to be used, 
all that should be outside the IP-version-specific functions. For 
consistency all you should do here is the amdgpu_ring_write calls with 
IP-version-specific packets.

Regards,
   Felix


> +
> +	r = amdgpu_fence_wait_polling(kiq_ring, seq, adev->usec_timeout);
> +	if (r < 1) {
> +		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
> +		return -ETIME;
> +	}
> +
> +	return 0;
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx10_kiq_set_resources,
>   	.kiq_map_queues = gfx10_kiq_map_queues,
>   	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
>   	.kiq_query_status = gfx10_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx10_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index ad0179ea2cc5..5be6fab55b73 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -837,15 +837,47 @@ static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static int gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	signed long r;
> +	uint32_t seq;
> +	struct amdgpu_device *adev = kiq_ring->adev;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	spin_lock(&adev->gfx.kiq.ring_lock);
> +	amdgpu_ring_alloc(kiq_ring, kiq->pmf->invalidate_tlbs_size);
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +	amdgpu_fence_emit_polling(kiq_ring, &seq);
> +	amdgpu_ring_commit(kiq_ring);
> +	spin_unlock(&adev->gfx.kiq.ring_lock);
> +
> +	r = amdgpu_fence_wait_polling(kiq_ring, seq, adev->usec_timeout);
> +	if (r < 1) {
> +		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
> +		return -ETIME;
> +	}
> +
> +	return 0;
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
>   	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
>   	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
>   	.kiq_query_status = gfx_v9_0_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 4/7] drm/amdgpu: replace kcq enable function on gfx_v9

[Alex Sierra]

[Why]
There's a HW-indpendent function that enables kcq. This function uses
the kiq_pm4_funcs implementation.

[How]
Local kcq enable function removed and replace it by the generic kcq
enable under amdgpu_gfx

Change-Id: I7709bdba93742c234941a5936c82eb67e346077c
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 70 +--------------------------
 1 file changed, 1 insertion(+), 69 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 5be6fab55b73..7219eacad9ce 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -3252,74 +3252,6 @@ static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
 	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
 }
 
-static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev)
-{
-	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
-	uint64_t queue_mask = 0;
-	int r, i;
-
-	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
-		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
-			continue;
-
-		/* This situation may be hit in the future if a new HW
-		 * generation exposes more than 64 queues. If so, the
-		 * definition of queue_mask needs updating */
-		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
-			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
-			break;
-		}
-
-		queue_mask |= (1ull << i);
-	}
-
-	r = amdgpu_ring_alloc(kiq_ring, (7 * adev->gfx.num_compute_rings) + 8);
-	if (r) {
-		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
-		return r;
-	}
-
-	/* set resources */
-	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
-	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
-			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
-	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
-	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
-	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
-	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
-	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
-	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
-	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
-		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
-		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
-		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
-
-		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
-		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
-		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
-				  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
-				  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
-				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
-				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
-				  PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
-				  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
-				  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
-				  PACKET3_MAP_QUEUES_ENGINE_SEL(0) | /* engine_sel: compute */
-				  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
-		amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
-		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
-		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
-		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
-		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
-	}
-
-	r = amdgpu_ring_test_helper(kiq_ring);
-	if (r)
-		DRM_ERROR("KCQ enable failed\n");
-
-	return r;
-}
-
 static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
 {
 	struct amdgpu_device *adev = ring->adev;
@@ -3726,7 +3658,7 @@ static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
 			goto done;
 	}
 
-	r = gfx_v9_0_kiq_kcq_enable(adev);
+	r = amdgpu_gfx_enable_kcq(adev);
 done:
 	return r;
 }
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 4/7] drm/amdgpu: replace kcq enable function on gfx_v9

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> [Why]
> There's a HW-indpendent function that enables kcq. This function uses
> the kiq_pm4_funcs implementation.
>
> [How]
> Local kcq enable function removed and replace it by the generic kcq
> enable under amdgpu_gfx
>
> Change-Id: I7709bdba93742c234941a5936c82eb67e346077c
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Acked-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 70 +--------------------------
>   1 file changed, 1 insertion(+), 69 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index 5be6fab55b73..7219eacad9ce 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -3252,74 +3252,6 @@ static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
>   	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
>   }
>   
> -static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev)
> -{
> -	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
> -	uint64_t queue_mask = 0;
> -	int r, i;
> -
> -	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
> -		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
> -			continue;
> -
> -		/* This situation may be hit in the future if a new HW
> -		 * generation exposes more than 64 queues. If so, the
> -		 * definition of queue_mask needs updating */
> -		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
> -			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
> -			break;
> -		}
> -
> -		queue_mask |= (1ull << i);
> -	}
> -
> -	r = amdgpu_ring_alloc(kiq_ring, (7 * adev->gfx.num_compute_rings) + 8);
> -	if (r) {
> -		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
> -		return r;
> -	}
> -
> -	/* set resources */
> -	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
> -	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
> -			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
> -	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
> -	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
> -	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
> -	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
> -	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
> -	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
> -	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
> -		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
> -		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
> -		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
> -
> -		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
> -		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
> -		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> -				  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
> -				  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
> -				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
> -				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
> -				  PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
> -				  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
> -				  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
> -				  PACKET3_MAP_QUEUES_ENGINE_SEL(0) | /* engine_sel: compute */
> -				  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
> -		amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
> -		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
> -		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
> -		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
> -		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
> -	}
> -
> -	r = amdgpu_ring_test_helper(kiq_ring);
> -	if (r)
> -		DRM_ERROR("KCQ enable failed\n");
> -
> -	return r;
> -}
> -
>   static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
>   {
>   	struct amdgpu_device *adev = ring->adev;
> @@ -3726,7 +3658,7 @@ static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
>   			goto done;
>   	}
>   
> -	r = gfx_v9_0_kiq_kcq_enable(adev);
> +	r = amdgpu_gfx_enable_kcq(adev);
>   done:
>   	return r;
>   }

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 4/7] drm/amdgpu: replace kcq enable function on gfx_v9

[Felix Kuehling]

Can you include kcq_disable in the patch as well?

Thanks,
   Felix

On 2020-01-11 1:39 p.m., Alex Sierra wrote:
> [Why]
> There's a HW-indpendent function that enables kcq. This function uses
> the kiq_pm4_funcs implementation.
>
> [How]
> Local kcq enable function removed and replace it by the generic kcq
> enable under amdgpu_gfx
>
> Change-Id: I7709bdba93742c234941a5936c82eb67e346077c
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 70 +--------------------------
>   1 file changed, 1 insertion(+), 69 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index 5be6fab55b73..7219eacad9ce 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -3252,74 +3252,6 @@ static void gfx_v9_0_kiq_setting(struct amdgpu_ring *ring)
>   	WREG32_SOC15_RLC(GC, 0, mmRLC_CP_SCHEDULERS, tmp);
>   }
>   
> -static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev)
> -{
> -	struct amdgpu_ring *kiq_ring = &adev->gfx.kiq.ring;
> -	uint64_t queue_mask = 0;
> -	int r, i;
> -
> -	for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
> -		if (!test_bit(i, adev->gfx.mec.queue_bitmap))
> -			continue;
> -
> -		/* This situation may be hit in the future if a new HW
> -		 * generation exposes more than 64 queues. If so, the
> -		 * definition of queue_mask needs updating */
> -		if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
> -			DRM_ERROR("Invalid KCQ enabled: %d\n", i);
> -			break;
> -		}
> -
> -		queue_mask |= (1ull << i);
> -	}
> -
> -	r = amdgpu_ring_alloc(kiq_ring, (7 * adev->gfx.num_compute_rings) + 8);
> -	if (r) {
> -		DRM_ERROR("Failed to lock KIQ (%d).\n", r);
> -		return r;
> -	}
> -
> -	/* set resources */
> -	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_SET_RESOURCES, 6));
> -	amdgpu_ring_write(kiq_ring, PACKET3_SET_RESOURCES_VMID_MASK(0) |
> -			  PACKET3_SET_RESOURCES_QUEUE_TYPE(0));	/* vmid_mask:0 queue_type:0 (KIQ) */
> -	amdgpu_ring_write(kiq_ring, lower_32_bits(queue_mask));	/* queue mask lo */
> -	amdgpu_ring_write(kiq_ring, upper_32_bits(queue_mask));	/* queue mask hi */
> -	amdgpu_ring_write(kiq_ring, 0);	/* gws mask lo */
> -	amdgpu_ring_write(kiq_ring, 0);	/* gws mask hi */
> -	amdgpu_ring_write(kiq_ring, 0);	/* oac mask */
> -	amdgpu_ring_write(kiq_ring, 0);	/* gds heap base:0, gds heap size:0 */
> -	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
> -		struct amdgpu_ring *ring = &adev->gfx.compute_ring[i];
> -		uint64_t mqd_addr = amdgpu_bo_gpu_offset(ring->mqd_obj);
> -		uint64_t wptr_addr = adev->wb.gpu_addr + (ring->wptr_offs * 4);
> -
> -		amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_MAP_QUEUES, 5));
> -		/* Q_sel:0, vmid:0, vidmem: 1, engine:0, num_Q:1*/
> -		amdgpu_ring_write(kiq_ring, /* Q_sel: 0, vmid: 0, engine: 0, num_Q: 1 */
> -				  PACKET3_MAP_QUEUES_QUEUE_SEL(0) | /* Queue_Sel */
> -				  PACKET3_MAP_QUEUES_VMID(0) | /* VMID */
> -				  PACKET3_MAP_QUEUES_QUEUE(ring->queue) |
> -				  PACKET3_MAP_QUEUES_PIPE(ring->pipe) |
> -				  PACKET3_MAP_QUEUES_ME((ring->me == 1 ? 0 : 1)) |
> -				  PACKET3_MAP_QUEUES_QUEUE_TYPE(0) | /*queue_type: normal compute queue */
> -				  PACKET3_MAP_QUEUES_ALLOC_FORMAT(0) | /* alloc format: all_on_one_pipe */
> -				  PACKET3_MAP_QUEUES_ENGINE_SEL(0) | /* engine_sel: compute */
> -				  PACKET3_MAP_QUEUES_NUM_QUEUES(1)); /* num_queues: must be 1 */
> -		amdgpu_ring_write(kiq_ring, PACKET3_MAP_QUEUES_DOORBELL_OFFSET(ring->doorbell_index));
> -		amdgpu_ring_write(kiq_ring, lower_32_bits(mqd_addr));
> -		amdgpu_ring_write(kiq_ring, upper_32_bits(mqd_addr));
> -		amdgpu_ring_write(kiq_ring, lower_32_bits(wptr_addr));
> -		amdgpu_ring_write(kiq_ring, upper_32_bits(wptr_addr));
> -	}
> -
> -	r = amdgpu_ring_test_helper(kiq_ring);
> -	if (r)
> -		DRM_ERROR("KCQ enable failed\n");
> -
> -	return r;
> -}
> -
>   static int gfx_v9_0_mqd_init(struct amdgpu_ring *ring)
>   {
>   	struct amdgpu_device *adev = ring->adev;
> @@ -3726,7 +3658,7 @@ static int gfx_v9_0_kcq_resume(struct amdgpu_device *adev)
>   			goto done;
>   	}
>   
> -	r = gfx_v9_0_kiq_kcq_enable(adev);
> +	r = amdgpu_gfx_enable_kcq(adev);
>   done:
>   	return r;
>   }
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 5/7] drm/amdgpu: export function to flush TLB via pasid

[Alex Sierra]

This can be used directly from amdgpu and amdkfd to invalidate
TLB through pasid.
It supports gmc v7, v8, v9 and v10.

Change-Id: I6563a8eba2e42d1a67fa2547156c20da41d1e490
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h |  6 +++
 drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c  | 59 ++++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c   | 33 +++++++++++++
 drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c   | 34 ++++++++++++++
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c   | 61 +++++++++++++++++++++++++
 5 files changed, 193 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
index c91dd602d5f1..d3c27a3c43f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
@@ -92,6 +92,9 @@ struct amdgpu_gmc_funcs {
 	/* flush the vm tlb via mmio */
 	void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid,
 				uint32_t vmhub, uint32_t flush_type);
+	/* flush the vm tlb via pasid */
+	int (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
+					uint32_t flush_type, bool all_hub);
 	/* flush the vm tlb via ring */
 	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
 				       uint64_t pd_addr);
@@ -216,6 +219,9 @@ struct amdgpu_gmc {
 };
 
 #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, vmhub, type) ((adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (vmhub), (type)))
+#define amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, type, allhub) \
+	((adev)->gmc.gmc_funcs->flush_gpu_tlb_pasid \
+	((adev), (pasid), (type), (allhub)))
 #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
 #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
 #define amdgpu_gmc_map_mtype(adev, flags) (adev)->gmc.gmc_funcs->map_mtype((adev),(flags))
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
index 5ad89bb6f3ba..09408b8b390f 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
@@ -30,6 +30,8 @@
 #include "hdp/hdp_5_0_0_sh_mask.h"
 #include "gc/gc_10_1_0_sh_mask.h"
 #include "mmhub/mmhub_2_0_0_sh_mask.h"
+#include "athub/athub_2_0_0_sh_mask.h"
+#include "athub/athub_2_0_0_offset.h"
 #include "dcn/dcn_2_0_0_offset.h"
 #include "dcn/dcn_2_0_0_sh_mask.h"
 #include "oss/osssys_5_0_0_offset.h"
@@ -37,6 +39,7 @@
 #include "navi10_enum.h"
 
 #include "soc15.h"
+#include "soc15d.h"
 #include "soc15_common.h"
 
 #include "nbio_v2_3.h"
@@ -234,6 +237,19 @@ static bool gmc_v10_0_use_invalidate_semaphore(struct amdgpu_device *adev,
 		(!amdgpu_sriov_vf(adev)));
 }
 
+static bool gmc_v10_0_get_atc_vmid_pasid_mapping_info(
+					struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
+{
+	uint32_t value;
+
+	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
+}
+
 /*
  * GART
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -380,6 +396,48 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	DRM_ERROR("Error flushing GPU TLB using the SDMA (%d)!\n", r);
 }
 
+/**
+ * gmc_v10_0_flush_gpu_tlb_pasid - tlb flush via pasid
+ *
+ * @adev: amdgpu_device pointer
+ * @pasid: pasid to be flush
+ *
+ * Flush the TLB for the requested pasid.
+ */
+static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
+					uint16_t pasid, uint32_t flush_type,
+					bool all_hub)
+{
+	int vmid, i;
+	uint16_t queried_pasid;
+	bool ret;
+	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
+
+	if (amdgpu_emu_mode == 0 && ring->sched.ready)
+		return kiq->pmf->kiq_invalidate_tlbs(ring,
+						pasid, flush_type, all_hub);
+
+	for (vmid = 1; vmid < 16; vmid++) {
+
+		ret = gmc_v10_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
+				&queried_pasid);
+		if (ret	&& queried_pasid == pasid) {
+			if (all_hub) {
+				for (i = 0; i < adev->num_vmhubs; i++)
+					gmc_v10_0_flush_gpu_tlb(adev, vmid,
+							i, 0);
+			} else {
+				gmc_v10_0_flush_gpu_tlb(adev, vmid,
+						AMDGPU_GFXHUB_0, 0);
+			}
+			break;
+		}
+	}
+
+	return 0;
+}
+
 static uint64_t gmc_v10_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 					     unsigned vmid, uint64_t pd_addr)
 {
@@ -531,6 +589,7 @@ static void gmc_v10_0_get_vm_pte(struct amdgpu_device *adev,
 
 static const struct amdgpu_gmc_funcs gmc_v10_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v10_0_flush_gpu_tlb,
+	.flush_gpu_tlb_pasid = gmc_v10_0_flush_gpu_tlb_pasid,
 	.emit_flush_gpu_tlb = gmc_v10_0_emit_flush_gpu_tlb,
 	.emit_pasid_mapping = gmc_v10_0_emit_pasid_mapping,
 	.map_mtype = gmc_v10_0_map_mtype,
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
index f08e5330642d..19d5b133e1d7 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
@@ -418,6 +418,38 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev)
 	return 0;
 }
 
+/**
+ * gmc_v7_0_flush_gpu_tlb_pasid - tlb flush via pasid
+ *
+ * @adev: amdgpu_device pointer
+ * @pasid: pasid to be flush
+ *
+ * Flush the TLB for the requested pasid.
+ */
+static int gmc_v7_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
+					uint16_t pasid, uint32_t flush_type,
+					bool all_hub)
+{
+	int vmid;
+	unsigned int tmp;
+
+	if (adev->in_gpu_reset)
+		return -EIO;
+
+	for (vmid = 1; vmid < 16; vmid++) {
+
+		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
+			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
+			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+			RREG32(mmVM_INVALIDATE_RESPONSE);
+			break;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * GART
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -1333,6 +1365,7 @@ static const struct amd_ip_funcs gmc_v7_0_ip_funcs = {
 
 static const struct amdgpu_gmc_funcs gmc_v7_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v7_0_flush_gpu_tlb,
+	.flush_gpu_tlb_pasid = gmc_v7_0_flush_gpu_tlb_pasid,
 	.emit_flush_gpu_tlb = gmc_v7_0_emit_flush_gpu_tlb,
 	.emit_pasid_mapping = gmc_v7_0_emit_pasid_mapping,
 	.set_prt = gmc_v7_0_set_prt,
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
index 6d96d40fbcb8..27d83204fa2b 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
@@ -620,6 +620,39 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
 	return 0;
 }
 
+/**
+ * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
+ *
+ * @adev: amdgpu_device pointer
+ * @pasid: pasid to be flush
+ *
+ * Flush the TLB for the requested pasid.
+ */
+static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
+					uint16_t pasid, uint32_t flush_type,
+					bool all_hub)
+{
+	int vmid;
+	unsigned int tmp;
+
+	if (adev->in_gpu_reset)
+		return -EIO;
+
+	for (vmid = 1; vmid < 16; vmid++) {
+
+		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
+			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
+			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+			RREG32(mmVM_INVALIDATE_RESPONSE);
+			break;
+		}
+	}
+
+	return 0;
+
+}
+
 /*
  * GART
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -1700,6 +1733,7 @@ static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
 
 static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
+	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
 	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
 	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
 	.set_prt = gmc_v8_0_set_prt,
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index b83c8d745f42..95cce54999b7 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -38,10 +38,12 @@
 #include "dce/dce_12_0_sh_mask.h"
 #include "vega10_enum.h"
 #include "mmhub/mmhub_1_0_offset.h"
+#include "athub/athub_1_0_sh_mask.h"
 #include "athub/athub_1_0_offset.h"
 #include "oss/osssys_4_0_offset.h"
 
 #include "soc15.h"
+#include "soc15d.h"
 #include "soc15_common.h"
 #include "umc/umc_6_0_sh_mask.h"
 
@@ -441,6 +443,18 @@ static bool gmc_v9_0_use_invalidate_semaphore(struct amdgpu_device *adev,
 		   adev->pdev->device == 0x15d8)));
 }
 
+static bool gmc_v9_0_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
+					uint8_t vmid, uint16_t *p_pasid)
+{
+	uint32_t value;
+
+	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
+		     + vmid);
+	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
+
+	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
+}
+
 /*
  * GART
  * VMID 0 is the physical GPU addresses as used by the kernel.
@@ -539,6 +553,52 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
 	DRM_ERROR("Timeout waiting for VM flush ACK!\n");
 }
 
+/**
+ * gmc_v9_0_flush_gpu_tlb_pasid - tlb flush via pasid
+ *
+ * @adev: amdgpu_device pointer
+ * @pasid: pasid to be flush
+ *
+ * Flush the TLB for the requested pasid.
+ */
+static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
+					uint16_t pasid, uint32_t flush_type,
+					bool all_hub)
+{
+	int vmid, i;
+	uint16_t queried_pasid;
+	bool ret;
+	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
+	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
+
+	if (adev->in_gpu_reset)
+		return -EIO;
+
+	if (ring->sched.ready)
+		return kiq->pmf->kiq_invalidate_tlbs(ring,
+						pasid, flush_type, all_hub);
+
+	for (vmid = 1; vmid < 16; vmid++) {
+
+		ret = gmc_v9_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
+				&queried_pasid);
+		if (ret && queried_pasid == pasid) {
+			if (all_hub) {
+				for (i = 0; i < adev->num_vmhubs; i++)
+					gmc_v9_0_flush_gpu_tlb(adev, vmid,
+							i, 0);
+			} else {
+				gmc_v9_0_flush_gpu_tlb(adev, vmid,
+						AMDGPU_GFXHUB_0, 0);
+			}
+			break;
+		}
+	}
+
+	return 0;
+
+}
+
 static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 					    unsigned vmid, uint64_t pd_addr)
 {
@@ -700,6 +760,7 @@ static void gmc_v9_0_get_vm_pte(struct amdgpu_device *adev,
 
 static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
 	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
+	.flush_gpu_tlb_pasid = gmc_v9_0_flush_gpu_tlb_pasid,
 	.emit_flush_gpu_tlb = gmc_v9_0_emit_flush_gpu_tlb,
 	.emit_pasid_mapping = gmc_v9_0_emit_pasid_mapping,
 	.map_mtype = gmc_v9_0_map_mtype,
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 5/7] drm/amdgpu: export function to flush TLB via pasid

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> This can be used directly from amdgpu and amdkfd to invalidate
> TLB through pasid.
> It supports gmc v7, v8, v9 and v10.
>
> Change-Id: I6563a8eba2e42d1a67fa2547156c20da41d1e490
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Flushing by pasid is in principle racy, but I don't see a way to avoid that.

The worst thing that could happen is that we flush a VMID while we 
wouldn't have to.

Reviewed-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h |  6 +++
>   drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c  | 59 ++++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c   | 33 +++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c   | 34 ++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c   | 61 +++++++++++++++++++++++++
>   5 files changed, 193 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> index c91dd602d5f1..d3c27a3c43f6 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> @@ -92,6 +92,9 @@ struct amdgpu_gmc_funcs {
>   	/* flush the vm tlb via mmio */
>   	void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid,
>   				uint32_t vmhub, uint32_t flush_type);
> +	/* flush the vm tlb via pasid */
> +	int (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
> +					uint32_t flush_type, bool all_hub);
>   	/* flush the vm tlb via ring */
>   	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
>   				       uint64_t pd_addr);
> @@ -216,6 +219,9 @@ struct amdgpu_gmc {
>   };
>   
>   #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, vmhub, type) ((adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (vmhub), (type)))
> +#define amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, type, allhub) \
> +	((adev)->gmc.gmc_funcs->flush_gpu_tlb_pasid \
> +	((adev), (pasid), (type), (allhub)))
>   #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
>   #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
>   #define amdgpu_gmc_map_mtype(adev, flags) (adev)->gmc.gmc_funcs->map_mtype((adev),(flags))
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
> index 5ad89bb6f3ba..09408b8b390f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
> @@ -30,6 +30,8 @@
>   #include "hdp/hdp_5_0_0_sh_mask.h"
>   #include "gc/gc_10_1_0_sh_mask.h"
>   #include "mmhub/mmhub_2_0_0_sh_mask.h"
> +#include "athub/athub_2_0_0_sh_mask.h"
> +#include "athub/athub_2_0_0_offset.h"
>   #include "dcn/dcn_2_0_0_offset.h"
>   #include "dcn/dcn_2_0_0_sh_mask.h"
>   #include "oss/osssys_5_0_0_offset.h"
> @@ -37,6 +39,7 @@
>   #include "navi10_enum.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   
>   #include "nbio_v2_3.h"
> @@ -234,6 +237,19 @@ static bool gmc_v10_0_use_invalidate_semaphore(struct amdgpu_device *adev,
>   		(!amdgpu_sriov_vf(adev)));
>   }
>   
> +static bool gmc_v10_0_get_atc_vmid_pasid_mapping_info(
> +					struct amdgpu_device *adev,
> +					uint8_t vmid, uint16_t *p_pasid)
> +{
> +	uint32_t value;
> +
> +	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
> +		     + vmid);
> +	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
> +
> +	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -380,6 +396,48 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
>   	DRM_ERROR("Error flushing GPU TLB using the SDMA (%d)!\n", r);
>   }
>   
> +/**
> + * gmc_v10_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid, i;
> +	uint16_t queried_pasid;
> +	bool ret;
> +	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	if (amdgpu_emu_mode == 0 && ring->sched.ready)
> +		return kiq->pmf->kiq_invalidate_tlbs(ring,
> +						pasid, flush_type, all_hub);
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		ret = gmc_v10_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
> +				&queried_pasid);
> +		if (ret	&& queried_pasid == pasid) {
> +			if (all_hub) {
> +				for (i = 0; i < adev->num_vmhubs; i++)
> +					gmc_v10_0_flush_gpu_tlb(adev, vmid,
> +							i, 0);
> +			} else {
> +				gmc_v10_0_flush_gpu_tlb(adev, vmid,
> +						AMDGPU_GFXHUB_0, 0);
> +			}
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>   static uint64_t gmc_v10_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
>   					     unsigned vmid, uint64_t pd_addr)
>   {
> @@ -531,6 +589,7 @@ static void gmc_v10_0_get_vm_pte(struct amdgpu_device *adev,
>   
>   static const struct amdgpu_gmc_funcs gmc_v10_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v10_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v10_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v10_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v10_0_emit_pasid_mapping,
>   	.map_mtype = gmc_v10_0_map_mtype,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
> index f08e5330642d..19d5b133e1d7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
> @@ -418,6 +418,38 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev)
>   	return 0;
>   }
>   
> +/**
> + * gmc_v7_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v7_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid;
> +	unsigned int tmp;
> +
> +	if (adev->in_gpu_reset)
> +		return -EIO;
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
> +		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
> +			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
> +			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> +			RREG32(mmVM_INVALIDATE_RESPONSE);
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -1333,6 +1365,7 @@ static const struct amd_ip_funcs gmc_v7_0_ip_funcs = {
>   
>   static const struct amdgpu_gmc_funcs gmc_v7_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v7_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v7_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v7_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v7_0_emit_pasid_mapping,
>   	.set_prt = gmc_v7_0_set_prt,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
> index 6d96d40fbcb8..27d83204fa2b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
> @@ -620,6 +620,39 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
>   	return 0;
>   }
>   
> +/**
> + * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid;
> +	unsigned int tmp;
> +
> +	if (adev->in_gpu_reset)
> +		return -EIO;
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
> +		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
> +			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
> +			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> +			RREG32(mmVM_INVALIDATE_RESPONSE);
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -1700,6 +1733,7 @@ static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
>   
>   static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
>   	.set_prt = gmc_v8_0_set_prt,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> index b83c8d745f42..95cce54999b7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> @@ -38,10 +38,12 @@
>   #include "dce/dce_12_0_sh_mask.h"
>   #include "vega10_enum.h"
>   #include "mmhub/mmhub_1_0_offset.h"
> +#include "athub/athub_1_0_sh_mask.h"
>   #include "athub/athub_1_0_offset.h"
>   #include "oss/osssys_4_0_offset.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   #include "umc/umc_6_0_sh_mask.h"
>   
> @@ -441,6 +443,18 @@ static bool gmc_v9_0_use_invalidate_semaphore(struct amdgpu_device *adev,
>   		   adev->pdev->device == 0x15d8)));
>   }
>   
> +static bool gmc_v9_0_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
> +					uint8_t vmid, uint16_t *p_pasid)
> +{
> +	uint32_t value;
> +
> +	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
> +		     + vmid);
> +	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
> +
> +	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -539,6 +553,52 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
>   	DRM_ERROR("Timeout waiting for VM flush ACK!\n");
>   }
>   
> +/**
> + * gmc_v9_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid, i;
> +	uint16_t queried_pasid;
> +	bool ret;
> +	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	if (adev->in_gpu_reset)
> +		return -EIO;
> +
> +	if (ring->sched.ready)
> +		return kiq->pmf->kiq_invalidate_tlbs(ring,
> +						pasid, flush_type, all_hub);
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		ret = gmc_v9_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
> +				&queried_pasid);
> +		if (ret && queried_pasid == pasid) {
> +			if (all_hub) {
> +				for (i = 0; i < adev->num_vmhubs; i++)
> +					gmc_v9_0_flush_gpu_tlb(adev, vmid,
> +							i, 0);
> +			} else {
> +				gmc_v9_0_flush_gpu_tlb(adev, vmid,
> +						AMDGPU_GFXHUB_0, 0);
> +			}
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +
> +}
> +
>   static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
>   					    unsigned vmid, uint64_t pd_addr)
>   {
> @@ -700,6 +760,7 @@ static void gmc_v9_0_get_vm_pte(struct amdgpu_device *adev,
>   
>   static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v9_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v9_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v9_0_emit_pasid_mapping,
>   	.map_mtype = gmc_v9_0_map_mtype,

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 5/7] drm/amdgpu: export function to flush TLB via pasid

[Felix Kuehling]

On 2020-01-11 1:39 p.m., Alex Sierra wrote:
> This can be used directly from amdgpu and amdkfd to invalidate
> TLB through pasid.
> It supports gmc v7, v8, v9 and v10.
>
> Change-Id: I6563a8eba2e42d1a67fa2547156c20da41d1e490
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h |  6 +++
>   drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c  | 59 ++++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c   | 33 +++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c   | 34 ++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c   | 61 +++++++++++++++++++++++++
>   5 files changed, 193 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> index c91dd602d5f1..d3c27a3c43f6 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gmc.h
> @@ -92,6 +92,9 @@ struct amdgpu_gmc_funcs {
>   	/* flush the vm tlb via mmio */
>   	void (*flush_gpu_tlb)(struct amdgpu_device *adev, uint32_t vmid,
>   				uint32_t vmhub, uint32_t flush_type);
> +	/* flush the vm tlb via pasid */
> +	int (*flush_gpu_tlb_pasid)(struct amdgpu_device *adev, uint16_t pasid,
> +					uint32_t flush_type, bool all_hub);
>   	/* flush the vm tlb via ring */
>   	uint64_t (*emit_flush_gpu_tlb)(struct amdgpu_ring *ring, unsigned vmid,
>   				       uint64_t pd_addr);
> @@ -216,6 +219,9 @@ struct amdgpu_gmc {
>   };
>   
>   #define amdgpu_gmc_flush_gpu_tlb(adev, vmid, vmhub, type) ((adev)->gmc.gmc_funcs->flush_gpu_tlb((adev), (vmid), (vmhub), (type)))
> +#define amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, type, allhub) \
> +	((adev)->gmc.gmc_funcs->flush_gpu_tlb_pasid \
> +	((adev), (pasid), (type), (allhub)))
>   #define amdgpu_gmc_emit_flush_gpu_tlb(r, vmid, addr) (r)->adev->gmc.gmc_funcs->emit_flush_gpu_tlb((r), (vmid), (addr))
>   #define amdgpu_gmc_emit_pasid_mapping(r, vmid, pasid) (r)->adev->gmc.gmc_funcs->emit_pasid_mapping((r), (vmid), (pasid))
>   #define amdgpu_gmc_map_mtype(adev, flags) (adev)->gmc.gmc_funcs->map_mtype((adev),(flags))
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
> index 5ad89bb6f3ba..09408b8b390f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c
> @@ -30,6 +30,8 @@
>   #include "hdp/hdp_5_0_0_sh_mask.h"
>   #include "gc/gc_10_1_0_sh_mask.h"
>   #include "mmhub/mmhub_2_0_0_sh_mask.h"
> +#include "athub/athub_2_0_0_sh_mask.h"
> +#include "athub/athub_2_0_0_offset.h"
>   #include "dcn/dcn_2_0_0_offset.h"
>   #include "dcn/dcn_2_0_0_sh_mask.h"
>   #include "oss/osssys_5_0_0_offset.h"
> @@ -37,6 +39,7 @@
>   #include "navi10_enum.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   
>   #include "nbio_v2_3.h"
> @@ -234,6 +237,19 @@ static bool gmc_v10_0_use_invalidate_semaphore(struct amdgpu_device *adev,
>   		(!amdgpu_sriov_vf(adev)));
>   }
>   
> +static bool gmc_v10_0_get_atc_vmid_pasid_mapping_info(
> +					struct amdgpu_device *adev,
> +					uint8_t vmid, uint16_t *p_pasid)
> +{
> +	uint32_t value;
> +
> +	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
> +		     + vmid);
> +	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
> +
> +	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -380,6 +396,48 @@ static void gmc_v10_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
>   	DRM_ERROR("Error flushing GPU TLB using the SDMA (%d)!\n", r);
>   }
>   
> +/**
> + * gmc_v10_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v10_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid, i;
> +	uint16_t queried_pasid;
> +	bool ret;
> +	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	if (amdgpu_emu_mode == 0 && ring->sched.ready)
> +		return kiq->pmf->kiq_invalidate_tlbs(ring,
> +						pasid, flush_type, all_hub);

This is where you should do the locking, ring alloc, and commit.

Regards,
   Felix


> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		ret = gmc_v10_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
> +				&queried_pasid);
> +		if (ret	&& queried_pasid == pasid) {
> +			if (all_hub) {
> +				for (i = 0; i < adev->num_vmhubs; i++)
> +					gmc_v10_0_flush_gpu_tlb(adev, vmid,
> +							i, 0);
> +			} else {
> +				gmc_v10_0_flush_gpu_tlb(adev, vmid,
> +						AMDGPU_GFXHUB_0, 0);
> +			}
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>   static uint64_t gmc_v10_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
>   					     unsigned vmid, uint64_t pd_addr)
>   {
> @@ -531,6 +589,7 @@ static void gmc_v10_0_get_vm_pte(struct amdgpu_device *adev,
>   
>   static const struct amdgpu_gmc_funcs gmc_v10_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v10_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v10_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v10_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v10_0_emit_pasid_mapping,
>   	.map_mtype = gmc_v10_0_map_mtype,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
> index f08e5330642d..19d5b133e1d7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c
> @@ -418,6 +418,38 @@ static int gmc_v7_0_mc_init(struct amdgpu_device *adev)
>   	return 0;
>   }
>   
> +/**
> + * gmc_v7_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v7_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid;
> +	unsigned int tmp;
> +
> +	if (adev->in_gpu_reset)
> +		return -EIO;
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
> +		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
> +			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
> +			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> +			RREG32(mmVM_INVALIDATE_RESPONSE);
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -1333,6 +1365,7 @@ static const struct amd_ip_funcs gmc_v7_0_ip_funcs = {
>   
>   static const struct amdgpu_gmc_funcs gmc_v7_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v7_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v7_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v7_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v7_0_emit_pasid_mapping,
>   	.set_prt = gmc_v7_0_set_prt,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
> index 6d96d40fbcb8..27d83204fa2b 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c
> @@ -620,6 +620,39 @@ static int gmc_v8_0_mc_init(struct amdgpu_device *adev)
>   	return 0;
>   }
>   
> +/**
> + * gmc_v8_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v8_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid;
> +	unsigned int tmp;
> +
> +	if (adev->in_gpu_reset)
> +		return -EIO;
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
> +		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
> +			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
> +			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> +			RREG32(mmVM_INVALIDATE_RESPONSE);
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -1700,6 +1733,7 @@ static const struct amd_ip_funcs gmc_v8_0_ip_funcs = {
>   
>   static const struct amdgpu_gmc_funcs gmc_v8_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v8_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v8_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v8_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v8_0_emit_pasid_mapping,
>   	.set_prt = gmc_v8_0_set_prt,
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> index b83c8d745f42..95cce54999b7 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> @@ -38,10 +38,12 @@
>   #include "dce/dce_12_0_sh_mask.h"
>   #include "vega10_enum.h"
>   #include "mmhub/mmhub_1_0_offset.h"
> +#include "athub/athub_1_0_sh_mask.h"
>   #include "athub/athub_1_0_offset.h"
>   #include "oss/osssys_4_0_offset.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   #include "umc/umc_6_0_sh_mask.h"
>   
> @@ -441,6 +443,18 @@ static bool gmc_v9_0_use_invalidate_semaphore(struct amdgpu_device *adev,
>   		   adev->pdev->device == 0x15d8)));
>   }
>   
> +static bool gmc_v9_0_get_atc_vmid_pasid_mapping_info(struct amdgpu_device *adev,
> +					uint8_t vmid, uint16_t *p_pasid)
> +{
> +	uint32_t value;
> +
> +	value = RREG32(SOC15_REG_OFFSET(ATHUB, 0, mmATC_VMID0_PASID_MAPPING)
> +		     + vmid);
> +	*p_pasid = value & ATC_VMID0_PASID_MAPPING__PASID_MASK;
> +
> +	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
> +}
> +
>   /*
>    * GART
>    * VMID 0 is the physical GPU addresses as used by the kernel.
> @@ -539,6 +553,52 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev, uint32_t vmid,
>   	DRM_ERROR("Timeout waiting for VM flush ACK!\n");
>   }
>   
> +/**
> + * gmc_v9_0_flush_gpu_tlb_pasid - tlb flush via pasid
> + *
> + * @adev: amdgpu_device pointer
> + * @pasid: pasid to be flush
> + *
> + * Flush the TLB for the requested pasid.
> + */
> +static int gmc_v9_0_flush_gpu_tlb_pasid(struct amdgpu_device *adev,
> +					uint16_t pasid, uint32_t flush_type,
> +					bool all_hub)
> +{
> +	int vmid, i;
> +	uint16_t queried_pasid;
> +	bool ret;
> +	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> +	struct amdgpu_kiq *kiq = &adev->gfx.kiq;
> +
> +	if (adev->in_gpu_reset)
> +		return -EIO;
> +
> +	if (ring->sched.ready)
> +		return kiq->pmf->kiq_invalidate_tlbs(ring,
> +						pasid, flush_type, all_hub);
> +
> +	for (vmid = 1; vmid < 16; vmid++) {
> +
> +		ret = gmc_v9_0_get_atc_vmid_pasid_mapping_info(adev, vmid,
> +				&queried_pasid);
> +		if (ret && queried_pasid == pasid) {
> +			if (all_hub) {
> +				for (i = 0; i < adev->num_vmhubs; i++)
> +					gmc_v9_0_flush_gpu_tlb(adev, vmid,
> +							i, 0);
> +			} else {
> +				gmc_v9_0_flush_gpu_tlb(adev, vmid,
> +						AMDGPU_GFXHUB_0, 0);
> +			}
> +			break;
> +		}
> +	}
> +
> +	return 0;
> +
> +}
> +
>   static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
>   					    unsigned vmid, uint64_t pd_addr)
>   {
> @@ -700,6 +760,7 @@ static void gmc_v9_0_get_vm_pte(struct amdgpu_device *adev,
>   
>   static const struct amdgpu_gmc_funcs gmc_v9_0_gmc_funcs = {
>   	.flush_gpu_tlb = gmc_v9_0_flush_gpu_tlb,
> +	.flush_gpu_tlb_pasid = gmc_v9_0_flush_gpu_tlb_pasid,
>   	.emit_flush_gpu_tlb = gmc_v9_0_emit_flush_gpu_tlb,
>   	.emit_pasid_mapping = gmc_v9_0_emit_pasid_mapping,
>   	.map_mtype = gmc_v9_0_map_mtype,
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 6/7] drm/amdgpu: GPU TLB flush API moved to amdgpu_amdkfd

[Alex Sierra]

[Why]
TLB flush method has been deprecated using kfd2kgd interface.
This implementation is now on the amdgpu_amdkfd API.

[How]
TLB flush functions now implemented in amdgpu_amdkfd.

Change-Id: Ic51cccdfe6e71288d78da772b6e1b6ced72f8ef7
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 32 ++++++++++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h |  2 ++
 drivers/gpu/drm/amd/amdkfd/kfd_process.c   |  8 ++++--
 3 files changed, 39 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 88e10b956413..8609287620ea 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -628,6 +628,38 @@ bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
 	return false;
 }
 
+int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+
+	if (adev->family == AMDGPU_FAMILY_AI) {
+		int i;
+
+		for (i = 0; i < adev->num_vmhubs; i++)
+			amdgpu_gmc_flush_gpu_tlb(adev, vmid, i, 0);
+	} else {
+		amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB_0, 0);
+	}
+
+	return 0;
+}
+
+int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid)
+{
+	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
+	uint32_t flush_type = 0;
+	bool all_hub = false;
+
+	if (adev->gmc.xgmi.num_physical_nodes &&
+		adev->asic_type == CHIP_VEGA20)
+		flush_type = 2;
+
+	if (adev->family == AMDGPU_FAMILY_AI)
+		all_hub = true;
+
+	return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub);
+}
+
 bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 069d5d230810..47b0f2957d1f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -136,6 +136,8 @@ int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
 				uint32_t *ib_cmd, uint32_t ib_len);
 void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle);
 bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd);
+int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid);
+int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid);
 
 bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
 
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 536a153ac9a4..25b90f70aecd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -32,6 +32,7 @@
 #include <linux/mman.h>
 #include <linux/file.h>
 #include "amdgpu_amdkfd.h"
+#include "amdgpu.h"
 
 struct mm_struct;
 
@@ -1152,16 +1153,17 @@ int kfd_reserved_mem_mmap(struct kfd_dev *dev, struct kfd_process *process,
 void kfd_flush_tlb(struct kfd_process_device *pdd)
 {
 	struct kfd_dev *dev = pdd->dev;
-	const struct kfd2kgd_calls *f2g = dev->kfd2kgd;
 
 	if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
 		/* Nothing to flush until a VMID is assigned, which
 		 * only happens when the first queue is created.
 		 */
 		if (pdd->qpd.vmid)
-			f2g->invalidate_tlbs_vmid(dev->kgd, pdd->qpd.vmid);
+			amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->kgd,
+							pdd->qpd.vmid);
 	} else {
-		f2g->invalidate_tlbs(dev->kgd, pdd->process->pasid);
+		amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->kgd,
+						pdd->process->pasid);
 	}
 }
 
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 6/7] drm/amdgpu: GPU TLB flush API moved to amdgpu_amdkfd

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> [Why]
> TLB flush method has been deprecated using kfd2kgd interface.
> This implementation is now on the amdgpu_amdkfd API.
>
> [How]
> TLB flush functions now implemented in amdgpu_amdkfd.
>
> Change-Id: Ic51cccdfe6e71288d78da772b6e1b6ced72f8ef7
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Acked-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 32 ++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h |  2 ++
>   drivers/gpu/drm/amd/amdkfd/kfd_process.c   |  8 ++++--
>   3 files changed, 39 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> index 88e10b956413..8609287620ea 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> @@ -628,6 +628,38 @@ bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
>   	return false;
>   }
>   
> +int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid)
> +{
> +	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
> +
> +	if (adev->family == AMDGPU_FAMILY_AI) {
> +		int i;
> +
> +		for (i = 0; i < adev->num_vmhubs; i++)
> +			amdgpu_gmc_flush_gpu_tlb(adev, vmid, i, 0);
> +	} else {
> +		amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB_0, 0);
> +	}
> +
> +	return 0;
> +}
> +
> +int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid)
> +{
> +	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
> +	uint32_t flush_type = 0;
> +	bool all_hub = false;
> +
> +	if (adev->gmc.xgmi.num_physical_nodes &&
> +		adev->asic_type == CHIP_VEGA20)
> +		flush_type = 2;
> +
> +	if (adev->family == AMDGPU_FAMILY_AI)
> +		all_hub = true;
> +
> +	return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub);
> +}
> +
>   bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd)
>   {
>   	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> index 069d5d230810..47b0f2957d1f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> @@ -136,6 +136,8 @@ int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
>   				uint32_t *ib_cmd, uint32_t ib_len);
>   void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle);
>   bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd);
> +int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid);
> +int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid);
>   
>   bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
>   
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> index 536a153ac9a4..25b90f70aecd 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> @@ -32,6 +32,7 @@
>   #include <linux/mman.h>
>   #include <linux/file.h>
>   #include "amdgpu_amdkfd.h"
> +#include "amdgpu.h"
>   
>   struct mm_struct;
>   
> @@ -1152,16 +1153,17 @@ int kfd_reserved_mem_mmap(struct kfd_dev *dev, struct kfd_process *process,
>   void kfd_flush_tlb(struct kfd_process_device *pdd)
>   {
>   	struct kfd_dev *dev = pdd->dev;
> -	const struct kfd2kgd_calls *f2g = dev->kfd2kgd;
>   
>   	if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
>   		/* Nothing to flush until a VMID is assigned, which
>   		 * only happens when the first queue is created.
>   		 */
>   		if (pdd->qpd.vmid)
> -			f2g->invalidate_tlbs_vmid(dev->kgd, pdd->qpd.vmid);
> +			amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->kgd,
> +							pdd->qpd.vmid);
>   	} else {
> -		f2g->invalidate_tlbs(dev->kgd, pdd->process->pasid);
> +		amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->kgd,
> +						pdd->process->pasid);
>   	}
>   }
>   

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 6/7] drm/amdgpu: GPU TLB flush API moved to amdgpu_amdkfd

[Felix Kuehling]

On 2020-01-11 1:39 p.m., Alex Sierra wrote:
> [Why]
> TLB flush method has been deprecated using kfd2kgd interface.
> This implementation is now on the amdgpu_amdkfd API.
>
> [How]
> TLB flush functions now implemented in amdgpu_amdkfd.
>
> Change-Id: Ic51cccdfe6e71288d78da772b6e1b6ced72f8ef7
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>

Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>


> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 32 ++++++++++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h |  2 ++
>   drivers/gpu/drm/amd/amdkfd/kfd_process.c   |  8 ++++--
>   3 files changed, 39 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> index 88e10b956413..8609287620ea 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> @@ -628,6 +628,38 @@ bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
>   	return false;
>   }
>   
> +int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid)
> +{
> +	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
> +
> +	if (adev->family == AMDGPU_FAMILY_AI) {
> +		int i;
> +
> +		for (i = 0; i < adev->num_vmhubs; i++)
> +			amdgpu_gmc_flush_gpu_tlb(adev, vmid, i, 0);
> +	} else {
> +		amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB_0, 0);
> +	}
> +
> +	return 0;
> +}
> +
> +int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid)
> +{
> +	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
> +	uint32_t flush_type = 0;
> +	bool all_hub = false;
> +
> +	if (adev->gmc.xgmi.num_physical_nodes &&
> +		adev->asic_type == CHIP_VEGA20)
> +		flush_type = 2;
> +
> +	if (adev->family == AMDGPU_FAMILY_AI)
> +		all_hub = true;
> +
> +	return amdgpu_gmc_flush_gpu_tlb_pasid(adev, pasid, flush_type, all_hub);
> +}
> +
>   bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd)
>   {
>   	struct amdgpu_device *adev = (struct amdgpu_device *)kgd;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> index 069d5d230810..47b0f2957d1f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
> @@ -136,6 +136,8 @@ int amdgpu_amdkfd_submit_ib(struct kgd_dev *kgd, enum kgd_engine_type engine,
>   				uint32_t *ib_cmd, uint32_t ib_len);
>   void amdgpu_amdkfd_set_compute_idle(struct kgd_dev *kgd, bool idle);
>   bool amdgpu_amdkfd_have_atomics_support(struct kgd_dev *kgd);
> +int amdgpu_amdkfd_flush_gpu_tlb_vmid(struct kgd_dev *kgd, uint16_t vmid);
> +int amdgpu_amdkfd_flush_gpu_tlb_pasid(struct kgd_dev *kgd, uint16_t pasid);
>   
>   bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid);
>   
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> index 536a153ac9a4..25b90f70aecd 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
> @@ -32,6 +32,7 @@
>   #include <linux/mman.h>
>   #include <linux/file.h>
>   #include "amdgpu_amdkfd.h"
> +#include "amdgpu.h"
>   
>   struct mm_struct;
>   
> @@ -1152,16 +1153,17 @@ int kfd_reserved_mem_mmap(struct kfd_dev *dev, struct kfd_process *process,
>   void kfd_flush_tlb(struct kfd_process_device *pdd)
>   {
>   	struct kfd_dev *dev = pdd->dev;
> -	const struct kfd2kgd_calls *f2g = dev->kfd2kgd;
>   
>   	if (dev->dqm->sched_policy == KFD_SCHED_POLICY_NO_HWS) {
>   		/* Nothing to flush until a VMID is assigned, which
>   		 * only happens when the first queue is created.
>   		 */
>   		if (pdd->qpd.vmid)
> -			f2g->invalidate_tlbs_vmid(dev->kgd, pdd->qpd.vmid);
> +			amdgpu_amdkfd_flush_gpu_tlb_vmid(dev->kgd,
> +							pdd->qpd.vmid);
>   	} else {
> -		f2g->invalidate_tlbs(dev->kgd, pdd->process->pasid);
> +		amdgpu_amdkfd_flush_gpu_tlb_pasid(dev->kgd,
> +						pdd->process->pasid);
>   	}
>   }
>   
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 7/7] drm/amdgpu: flush TLB functions removal from kfd2kgd interface

[Alex Sierra]

[Why]
kfd2kgd interface will be deprecated. This removal only covers TLB
invalidation for now. They have been replaced in amdgpu_amdkfd API.

[How]
TLB invalidate functions removed from the different amdkfd_gfx_v*
versions.

Change-Id: Ic2c7d4a0d19fe1e884dee1ff10a520d31252afee
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
---
 .../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c   |  2 -
 .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c    | 67 -------------
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 41 --------
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 41 --------
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 96 -------------------
 .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h |  2 -
 .../gpu/drm/amd/include/kgd_kfd_interface.h   |  2 -
 7 files changed, 251 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
index f9011a07cb90..562e7a7f51a8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
@@ -317,7 +317,5 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
 			kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
 	.get_tile_config = kgd_gfx_v9_get_tile_config,
 	.set_vm_context_page_table_base = kgd_set_vm_context_page_table_base,
-	.invalidate_tlbs = kgd_gfx_v9_invalidate_tlbs,
-	.invalidate_tlbs_vmid = kgd_gfx_v9_invalidate_tlbs_vmid,
 	.get_hive_id = amdgpu_amdkfd_get_hive_id,
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
index 61cd707158e4..6132b4874498 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
@@ -686,71 +686,6 @@ static bool get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
 	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
 }
 
-static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid)
-{
-	signed long r;
-	uint32_t seq;
-	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
-
-	spin_lock(&adev->gfx.kiq.ring_lock);
-	amdgpu_ring_alloc(ring, 12); /* fence + invalidate_tlbs package*/
-	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
-	amdgpu_ring_write(ring,
-			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
-			PACKET3_INVALIDATE_TLBS_PASID(pasid));
-	amdgpu_fence_emit_polling(ring, &seq);
-	amdgpu_ring_commit(ring);
-	spin_unlock(&adev->gfx.kiq.ring_lock);
-
-	r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
-	if (r < 1) {
-		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
-		return -ETIME;
-	}
-
-	return 0;
-}
-
-static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	int vmid;
-	uint16_t queried_pasid;
-	bool ret;
-	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
-
-	if (amdgpu_emu_mode == 0 && ring->sched.ready)
-		return invalidate_tlbs_with_kiq(adev, pasid);
-
-	for (vmid = 0; vmid < 16; vmid++) {
-		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
-			continue;
-
-		ret = get_atc_vmid_pasid_mapping_info(kgd, vmid,
-				&queried_pasid);
-		if (ret	&& queried_pasid == pasid) {
-			amdgpu_gmc_flush_gpu_tlb(adev, vmid,
-					AMDGPU_GFXHUB_0, 0);
-			break;
-		}
-	}
-
-	return 0;
-}
-
-static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
-	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
-		pr_err("non kfd vmid %d\n", vmid);
-		return 0;
-	}
-
-	amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB_0, 0);
-	return 0;
-}
-
 static int kgd_address_watch_disable(struct kgd_dev *kgd)
 {
 	return 0;
@@ -832,7 +767,5 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
 			get_atc_vmid_pasid_mapping_info,
 	.get_tile_config = amdgpu_amdkfd_get_tile_config,
 	.set_vm_context_page_table_base = set_vm_context_page_table_base,
-	.invalidate_tlbs = invalidate_tlbs,
-	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
 	.get_hive_id = amdgpu_amdkfd_get_hive_id,
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
index 6e6f0a99ec06..8f052e98a3c6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
@@ -696,45 +696,6 @@ static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
 		lower_32_bits(page_table_base));
 }
 
-static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	int vmid;
-	unsigned int tmp;
-
-	if (adev->in_gpu_reset)
-		return -EIO;
-
-	for (vmid = 0; vmid < 16; vmid++) {
-		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
-			continue;
-
-		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
-		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
-			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
-			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
-			RREG32(mmVM_INVALIDATE_RESPONSE);
-			break;
-		}
-	}
-
-	return 0;
-}
-
-static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
-	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
-		pr_err("non kfd vmid\n");
-		return 0;
-	}
-
-	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
-	RREG32(mmVM_INVALIDATE_RESPONSE);
-	return 0;
-}
-
  /**
   * read_vmid_from_vmfault_reg - read vmid from register
   *
@@ -771,7 +732,5 @@ const struct kfd2kgd_calls gfx_v7_kfd2kgd = {
 	.set_scratch_backing_va = set_scratch_backing_va,
 	.get_tile_config = get_tile_config,
 	.set_vm_context_page_table_base = set_vm_context_page_table_base,
-	.invalidate_tlbs = invalidate_tlbs,
-	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
 	.read_vmid_from_vmfault_reg = read_vmid_from_vmfault_reg,
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index bfbddedb2380..19a10db93d68 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -657,45 +657,6 @@ static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
 			lower_32_bits(page_table_base));
 }
 
-static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	int vmid;
-	unsigned int tmp;
-
-	if (adev->in_gpu_reset)
-		return -EIO;
-
-	for (vmid = 0; vmid < 16; vmid++) {
-		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
-			continue;
-
-		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
-		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
-			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
-			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
-			RREG32(mmVM_INVALIDATE_RESPONSE);
-			break;
-		}
-	}
-
-	return 0;
-}
-
-static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-
-	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
-		pr_err("non kfd vmid %d\n", vmid);
-		return -EINVAL;
-	}
-
-	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
-	RREG32(mmVM_INVALIDATE_RESPONSE);
-	return 0;
-}
-
 const struct kfd2kgd_calls gfx_v8_kfd2kgd = {
 	.program_sh_mem_settings = kgd_program_sh_mem_settings,
 	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
@@ -717,6 +678,4 @@ const struct kfd2kgd_calls gfx_v8_kfd2kgd = {
 	.set_scratch_backing_va = set_scratch_backing_va,
 	.get_tile_config = get_tile_config,
 	.set_vm_context_page_table_base = set_vm_context_page_table_base,
-	.invalidate_tlbs = invalidate_tlbs,
-	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
index e7861f0ef415..932ae85d97e2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -617,100 +617,6 @@ bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
 	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
 }
 
-static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid,
-			uint32_t flush_type)
-{
-	signed long r;
-	uint32_t seq;
-	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
-
-	spin_lock(&adev->gfx.kiq.ring_lock);
-	amdgpu_ring_alloc(ring, 12); /* fence + invalidate_tlbs package*/
-	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
-	amdgpu_ring_write(ring,
-			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
-			PACKET3_INVALIDATE_TLBS_ALL_HUB(1) |
-			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
-			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
-	amdgpu_fence_emit_polling(ring, &seq);
-	amdgpu_ring_commit(ring);
-	spin_unlock(&adev->gfx.kiq.ring_lock);
-
-	r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
-	if (r < 1) {
-		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
-		return -ETIME;
-	}
-
-	return 0;
-}
-
-int kgd_gfx_v9_invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	int vmid, i;
-	uint16_t queried_pasid;
-	bool ret;
-	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
-	uint32_t flush_type = 0;
-
-	if (adev->in_gpu_reset)
-		return -EIO;
-	if (adev->gmc.xgmi.num_physical_nodes &&
-		adev->asic_type == CHIP_VEGA20)
-		flush_type = 2;
-
-	if (ring->sched.ready)
-		return invalidate_tlbs_with_kiq(adev, pasid, flush_type);
-
-	for (vmid = 0; vmid < 16; vmid++) {
-		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
-			continue;
-
-		ret = kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(kgd, vmid,
-				&queried_pasid);
-		if (ret && queried_pasid == pasid) {
-			for (i = 0; i < adev->num_vmhubs; i++)
-				amdgpu_gmc_flush_gpu_tlb(adev, vmid,
-							i, flush_type);
-			break;
-		}
-	}
-
-	return 0;
-}
-
-int kgd_gfx_v9_invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
-{
-	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	int i;
-
-	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
-		pr_err("non kfd vmid %d\n", vmid);
-		return 0;
-	}
-
-	/* Use legacy mode tlb invalidation.
-	 *
-	 * Currently on Raven the code below is broken for anything but
-	 * legacy mode due to a MMHUB power gating problem. A workaround
-	 * is for MMHUB to wait until the condition PER_VMID_INVALIDATE_REQ
-	 * == PER_VMID_INVALIDATE_ACK instead of simply waiting for the ack
-	 * bit.
-	 *
-	 * TODO 1: agree on the right set of invalidation registers for
-	 * KFD use. Use the last one for now. Invalidate both GC and
-	 * MMHUB.
-	 *
-	 * TODO 2: support range-based invalidation, requires kfg2kgd
-	 * interface change
-	 */
-	for (i = 0; i < adev->num_vmhubs; i++)
-		amdgpu_gmc_flush_gpu_tlb(adev, vmid, i, 0);
-
-	return 0;
-}
-
 int kgd_gfx_v9_address_watch_disable(struct kgd_dev *kgd)
 {
 	return 0;
@@ -793,7 +699,5 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
 			kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
 	.get_tile_config = kgd_gfx_v9_get_tile_config,
 	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
-	.invalidate_tlbs = kgd_gfx_v9_invalidate_tlbs,
-	.invalidate_tlbs_vmid = kgd_gfx_v9_invalidate_tlbs_vmid,
 	.get_hive_id = amdgpu_amdkfd_get_hive_id,
 };
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
index 02b1426d17d1..dfafa28b7559 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
@@ -57,7 +57,5 @@ uint32_t kgd_gfx_v9_address_watch_get_offset(struct kgd_dev *kgd,
 
 bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
 					uint8_t vmid, uint16_t *p_pasid);
-int kgd_gfx_v9_invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid);
-int kgd_gfx_v9_invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid);
 int kgd_gfx_v9_get_tile_config(struct kgd_dev *kgd,
 		struct tile_config *config);
diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
index 2cd217e60125..a01ef836ad58 100644
--- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
+++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
@@ -307,8 +307,6 @@ struct kfd2kgd_calls {
 
 	void (*set_vm_context_page_table_base)(struct kgd_dev *kgd,
 			uint32_t vmid, uint64_t page_table_base);
-	int (*invalidate_tlbs)(struct kgd_dev *kgd, uint16_t pasid);
-	int (*invalidate_tlbs_vmid)(struct kgd_dev *kgd, uint16_t vmid);
 	uint32_t (*read_vmid_from_vmfault_reg)(struct kgd_dev *kgd);
 	uint64_t (*get_hive_id)(struct kgd_dev *kgd);
 
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 7/7] drm/amdgpu: flush TLB functions removal from kfd2kgd interface

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> [Why]
> kfd2kgd interface will be deprecated. This removal only covers TLB
> invalidation for now. They have been replaced in amdgpu_amdkfd API.
>
> [How]
> TLB invalidate functions removed from the different amdkfd_gfx_v*
> versions.
>
> Change-Id: Ic2c7d4a0d19fe1e884dee1ff10a520d31252afee
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>

Nice work for the patch set. Acked-by: Christian König 
<christian.koenig@amd.com>

> ---
>   .../drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c   |  2 -
>   .../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c    | 67 -------------
>   .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 41 --------
>   .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 41 --------
>   .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 96 -------------------
>   .../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h |  2 -
>   .../gpu/drm/amd/include/kgd_kfd_interface.h   |  2 -
>   7 files changed, 251 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> index f9011a07cb90..562e7a7f51a8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_arcturus.c
> @@ -317,7 +317,5 @@ const struct kfd2kgd_calls arcturus_kfd2kgd = {
>   			kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
>   	.get_tile_config = kgd_gfx_v9_get_tile_config,
>   	.set_vm_context_page_table_base = kgd_set_vm_context_page_table_base,
> -	.invalidate_tlbs = kgd_gfx_v9_invalidate_tlbs,
> -	.invalidate_tlbs_vmid = kgd_gfx_v9_invalidate_tlbs_vmid,
>   	.get_hive_id = amdgpu_amdkfd_get_hive_id,
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
> index 61cd707158e4..6132b4874498 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v10.c
> @@ -686,71 +686,6 @@ static bool get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
>   	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
>   }
>   
> -static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid)
> -{
> -	signed long r;
> -	uint32_t seq;
> -	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> -
> -	spin_lock(&adev->gfx.kiq.ring_lock);
> -	amdgpu_ring_alloc(ring, 12); /* fence + invalidate_tlbs package*/
> -	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> -	amdgpu_ring_write(ring,
> -			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> -			PACKET3_INVALIDATE_TLBS_PASID(pasid));
> -	amdgpu_fence_emit_polling(ring, &seq);
> -	amdgpu_ring_commit(ring);
> -	spin_unlock(&adev->gfx.kiq.ring_lock);
> -
> -	r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
> -	if (r < 1) {
> -		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
> -		return -ETIME;
> -	}
> -
> -	return 0;
> -}
> -
> -static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -	int vmid;
> -	uint16_t queried_pasid;
> -	bool ret;
> -	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> -
> -	if (amdgpu_emu_mode == 0 && ring->sched.ready)
> -		return invalidate_tlbs_with_kiq(adev, pasid);
> -
> -	for (vmid = 0; vmid < 16; vmid++) {
> -		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
> -			continue;
> -
> -		ret = get_atc_vmid_pasid_mapping_info(kgd, vmid,
> -				&queried_pasid);
> -		if (ret	&& queried_pasid == pasid) {
> -			amdgpu_gmc_flush_gpu_tlb(adev, vmid,
> -					AMDGPU_GFXHUB_0, 0);
> -			break;
> -		}
> -	}
> -
> -	return 0;
> -}
> -
> -static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -
> -	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
> -		pr_err("non kfd vmid %d\n", vmid);
> -		return 0;
> -	}
> -
> -	amdgpu_gmc_flush_gpu_tlb(adev, vmid, AMDGPU_GFXHUB_0, 0);
> -	return 0;
> -}
> -
>   static int kgd_address_watch_disable(struct kgd_dev *kgd)
>   {
>   	return 0;
> @@ -832,7 +767,5 @@ const struct kfd2kgd_calls gfx_v10_kfd2kgd = {
>   			get_atc_vmid_pasid_mapping_info,
>   	.get_tile_config = amdgpu_amdkfd_get_tile_config,
>   	.set_vm_context_page_table_base = set_vm_context_page_table_base,
> -	.invalidate_tlbs = invalidate_tlbs,
> -	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
>   	.get_hive_id = amdgpu_amdkfd_get_hive_id,
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
> index 6e6f0a99ec06..8f052e98a3c6 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
> @@ -696,45 +696,6 @@ static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
>   		lower_32_bits(page_table_base));
>   }
>   
> -static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -	int vmid;
> -	unsigned int tmp;
> -
> -	if (adev->in_gpu_reset)
> -		return -EIO;
> -
> -	for (vmid = 0; vmid < 16; vmid++) {
> -		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
> -			continue;
> -
> -		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
> -		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
> -			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
> -			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> -			RREG32(mmVM_INVALIDATE_RESPONSE);
> -			break;
> -		}
> -	}
> -
> -	return 0;
> -}
> -
> -static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -
> -	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
> -		pr_err("non kfd vmid\n");
> -		return 0;
> -	}
> -
> -	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> -	RREG32(mmVM_INVALIDATE_RESPONSE);
> -	return 0;
> -}
> -
>    /**
>     * read_vmid_from_vmfault_reg - read vmid from register
>     *
> @@ -771,7 +732,5 @@ const struct kfd2kgd_calls gfx_v7_kfd2kgd = {
>   	.set_scratch_backing_va = set_scratch_backing_va,
>   	.get_tile_config = get_tile_config,
>   	.set_vm_context_page_table_base = set_vm_context_page_table_base,
> -	.invalidate_tlbs = invalidate_tlbs,
> -	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
>   	.read_vmid_from_vmfault_reg = read_vmid_from_vmfault_reg,
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
> index bfbddedb2380..19a10db93d68 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
> @@ -657,45 +657,6 @@ static void set_vm_context_page_table_base(struct kgd_dev *kgd, uint32_t vmid,
>   			lower_32_bits(page_table_base));
>   }
>   
> -static int invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -	int vmid;
> -	unsigned int tmp;
> -
> -	if (adev->in_gpu_reset)
> -		return -EIO;
> -
> -	for (vmid = 0; vmid < 16; vmid++) {
> -		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
> -			continue;
> -
> -		tmp = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
> -		if ((tmp & ATC_VMID0_PASID_MAPPING__VALID_MASK) &&
> -			(tmp & ATC_VMID0_PASID_MAPPING__PASID_MASK) == pasid) {
> -			WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> -			RREG32(mmVM_INVALIDATE_RESPONSE);
> -			break;
> -		}
> -	}
> -
> -	return 0;
> -}
> -
> -static int invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -
> -	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
> -		pr_err("non kfd vmid %d\n", vmid);
> -		return -EINVAL;
> -	}
> -
> -	WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
> -	RREG32(mmVM_INVALIDATE_RESPONSE);
> -	return 0;
> -}
> -
>   const struct kfd2kgd_calls gfx_v8_kfd2kgd = {
>   	.program_sh_mem_settings = kgd_program_sh_mem_settings,
>   	.set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
> @@ -717,6 +678,4 @@ const struct kfd2kgd_calls gfx_v8_kfd2kgd = {
>   	.set_scratch_backing_va = set_scratch_backing_va,
>   	.get_tile_config = get_tile_config,
>   	.set_vm_context_page_table_base = set_vm_context_page_table_base,
> -	.invalidate_tlbs = invalidate_tlbs,
> -	.invalidate_tlbs_vmid = invalidate_tlbs_vmid,
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> index e7861f0ef415..932ae85d97e2 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> @@ -617,100 +617,6 @@ bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
>   	return !!(value & ATC_VMID0_PASID_MAPPING__VALID_MASK);
>   }
>   
> -static int invalidate_tlbs_with_kiq(struct amdgpu_device *adev, uint16_t pasid,
> -			uint32_t flush_type)
> -{
> -	signed long r;
> -	uint32_t seq;
> -	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> -
> -	spin_lock(&adev->gfx.kiq.ring_lock);
> -	amdgpu_ring_alloc(ring, 12); /* fence + invalidate_tlbs package*/
> -	amdgpu_ring_write(ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> -	amdgpu_ring_write(ring,
> -			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> -			PACKET3_INVALIDATE_TLBS_ALL_HUB(1) |
> -			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> -			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> -	amdgpu_fence_emit_polling(ring, &seq);
> -	amdgpu_ring_commit(ring);
> -	spin_unlock(&adev->gfx.kiq.ring_lock);
> -
> -	r = amdgpu_fence_wait_polling(ring, seq, adev->usec_timeout);
> -	if (r < 1) {
> -		DRM_ERROR("wait for kiq fence error: %ld.\n", r);
> -		return -ETIME;
> -	}
> -
> -	return 0;
> -}
> -
> -int kgd_gfx_v9_invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -	int vmid, i;
> -	uint16_t queried_pasid;
> -	bool ret;
> -	struct amdgpu_ring *ring = &adev->gfx.kiq.ring;
> -	uint32_t flush_type = 0;
> -
> -	if (adev->in_gpu_reset)
> -		return -EIO;
> -	if (adev->gmc.xgmi.num_physical_nodes &&
> -		adev->asic_type == CHIP_VEGA20)
> -		flush_type = 2;
> -
> -	if (ring->sched.ready)
> -		return invalidate_tlbs_with_kiq(adev, pasid, flush_type);
> -
> -	for (vmid = 0; vmid < 16; vmid++) {
> -		if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid))
> -			continue;
> -
> -		ret = kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(kgd, vmid,
> -				&queried_pasid);
> -		if (ret && queried_pasid == pasid) {
> -			for (i = 0; i < adev->num_vmhubs; i++)
> -				amdgpu_gmc_flush_gpu_tlb(adev, vmid,
> -							i, flush_type);
> -			break;
> -		}
> -	}
> -
> -	return 0;
> -}
> -
> -int kgd_gfx_v9_invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid)
> -{
> -	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
> -	int i;
> -
> -	if (!amdgpu_amdkfd_is_kfd_vmid(adev, vmid)) {
> -		pr_err("non kfd vmid %d\n", vmid);
> -		return 0;
> -	}
> -
> -	/* Use legacy mode tlb invalidation.
> -	 *
> -	 * Currently on Raven the code below is broken for anything but
> -	 * legacy mode due to a MMHUB power gating problem. A workaround
> -	 * is for MMHUB to wait until the condition PER_VMID_INVALIDATE_REQ
> -	 * == PER_VMID_INVALIDATE_ACK instead of simply waiting for the ack
> -	 * bit.
> -	 *
> -	 * TODO 1: agree on the right set of invalidation registers for
> -	 * KFD use. Use the last one for now. Invalidate both GC and
> -	 * MMHUB.
> -	 *
> -	 * TODO 2: support range-based invalidation, requires kfg2kgd
> -	 * interface change
> -	 */
> -	for (i = 0; i < adev->num_vmhubs; i++)
> -		amdgpu_gmc_flush_gpu_tlb(adev, vmid, i, 0);
> -
> -	return 0;
> -}
> -
>   int kgd_gfx_v9_address_watch_disable(struct kgd_dev *kgd)
>   {
>   	return 0;
> @@ -793,7 +699,5 @@ const struct kfd2kgd_calls gfx_v9_kfd2kgd = {
>   			kgd_gfx_v9_get_atc_vmid_pasid_mapping_info,
>   	.get_tile_config = kgd_gfx_v9_get_tile_config,
>   	.set_vm_context_page_table_base = kgd_gfx_v9_set_vm_context_page_table_base,
> -	.invalidate_tlbs = kgd_gfx_v9_invalidate_tlbs,
> -	.invalidate_tlbs_vmid = kgd_gfx_v9_invalidate_tlbs_vmid,
>   	.get_hive_id = amdgpu_amdkfd_get_hive_id,
>   };
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
> index 02b1426d17d1..dfafa28b7559 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.h
> @@ -57,7 +57,5 @@ uint32_t kgd_gfx_v9_address_watch_get_offset(struct kgd_dev *kgd,
>   
>   bool kgd_gfx_v9_get_atc_vmid_pasid_mapping_info(struct kgd_dev *kgd,
>   					uint8_t vmid, uint16_t *p_pasid);
> -int kgd_gfx_v9_invalidate_tlbs(struct kgd_dev *kgd, uint16_t pasid);
> -int kgd_gfx_v9_invalidate_tlbs_vmid(struct kgd_dev *kgd, uint16_t vmid);
>   int kgd_gfx_v9_get_tile_config(struct kgd_dev *kgd,
>   		struct tile_config *config);
> diff --git a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
> index 2cd217e60125..a01ef836ad58 100644
> --- a/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
> +++ b/drivers/gpu/drm/amd/include/kgd_kfd_interface.h
> @@ -307,8 +307,6 @@ struct kfd2kgd_calls {
>   
>   	void (*set_vm_context_page_table_base)(struct kgd_dev *kgd,
>   			uint32_t vmid, uint64_t page_table_base);
> -	int (*invalidate_tlbs)(struct kgd_dev *kgd, uint16_t pasid);
> -	int (*invalidate_tlbs_vmid)(struct kgd_dev *kgd, uint16_t vmid);
>   	uint32_t (*read_vmid_from_vmfault_reg)(struct kgd_dev *kgd);
>   	uint64_t (*get_hive_id)(struct kgd_dev *kgd);
>   

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 1/7] drm/amdgpu: Avoid reclaim fs while eviction lock

[Christian König]

Am 11.01.20 um 19:39 schrieb Alex Sierra:
> [Why]
> Avoid reclaim filesystem while eviction lock is held called from
> MMU notifier.
>
> [How]
> Setting PF_MEMALLOC_NOFS flags while eviction mutex is locked.
> Using memalloc_nofs_save / memalloc_nofs_restore API.
>
> Change-Id: I5531c9337836e7d4a430df3f16dcc82888e8018c
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>

Reviewed-by: Christian König <christian.koenig@amd.com>

> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 40 +++++++++++++++++++++-----
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |  6 +++-
>   2 files changed, 38 insertions(+), 8 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> index b999b67ff57a..d6aba4f9df74 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
> @@ -82,6 +82,32 @@ struct amdgpu_prt_cb {
>   	struct dma_fence_cb cb;
>   };
>   
> +/**
> + * vm eviction_lock can be taken in MMU notifiers. Make sure no reclaim-FS
> + * happens while holding this lock anywhere to prevent deadlocks when
> + * an MMU notifier runs in reclaim-FS context.
> + */
> +static inline void amdgpu_vm_eviction_lock(struct amdgpu_vm *vm)
> +{
> +	mutex_lock(&vm->eviction_lock);
> +	vm->saved_flags = memalloc_nofs_save();
> +}
> +
> +static inline int amdgpu_vm_eviction_trylock(struct amdgpu_vm *vm)
> +{
> +	if (mutex_trylock(&vm->eviction_lock)) {
> +		vm->saved_flags = memalloc_nofs_save();
> +		return 1;
> +	}
> +	return 0;
> +}
> +
> +static inline void amdgpu_vm_eviction_unlock(struct amdgpu_vm *vm)
> +{
> +	memalloc_nofs_restore(vm->saved_flags);
> +	mutex_unlock(&vm->eviction_lock);
> +}
> +
>   /**
>    * amdgpu_vm_level_shift - return the addr shift for each level
>    *
> @@ -678,9 +704,9 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
>   		}
>   	}
>   
> -	mutex_lock(&vm->eviction_lock);
> +	amdgpu_vm_eviction_lock(vm);
>   	vm->evicting = false;
> -	mutex_unlock(&vm->eviction_lock);
> +	amdgpu_vm_eviction_unlock(vm);
>   
>   	return 0;
>   }
> @@ -1559,7 +1585,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
>   	if (!(flags & AMDGPU_PTE_VALID))
>   		owner = AMDGPU_FENCE_OWNER_KFD;
>   
> -	mutex_lock(&vm->eviction_lock);
> +	amdgpu_vm_eviction_lock(vm);
>   	if (vm->evicting) {
>   		r = -EBUSY;
>   		goto error_unlock;
> @@ -1576,7 +1602,7 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
>   	r = vm->update_funcs->commit(&params, fence);
>   
>   error_unlock:
> -	mutex_unlock(&vm->eviction_lock);
> +	amdgpu_vm_eviction_unlock(vm);
>   	return r;
>   }
>   
> @@ -2537,18 +2563,18 @@ bool amdgpu_vm_evictable(struct amdgpu_bo *bo)
>   		return false;
>   
>   	/* Try to block ongoing updates */
> -	if (!mutex_trylock(&bo_base->vm->eviction_lock))
> +	if (!amdgpu_vm_eviction_trylock(bo_base->vm))
>   		return false;
>   
>   	/* Don't evict VM page tables while they are updated */
>   	if (!dma_fence_is_signaled(bo_base->vm->last_direct) ||
>   	    !dma_fence_is_signaled(bo_base->vm->last_delayed)) {
> -		mutex_unlock(&bo_base->vm->eviction_lock);
> +		amdgpu_vm_eviction_unlock(bo_base->vm);
>   		return false;
>   	}
>   
>   	bo_base->vm->evicting = true;
> -	mutex_unlock(&bo_base->vm->eviction_lock);
> +	amdgpu_vm_eviction_unlock(bo_base->vm);
>   	return true;
>   }
>   
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> index 100547f094ff..c21a36bebc0c 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
> @@ -30,6 +30,7 @@
>   #include <drm/gpu_scheduler.h>
>   #include <drm/drm_file.h>
>   #include <drm/ttm/ttm_bo_driver.h>
> +#include <linux/sched/mm.h>
>   
>   #include "amdgpu_sync.h"
>   #include "amdgpu_ring.h"
> @@ -242,9 +243,12 @@ struct amdgpu_vm {
>   	/* tree of virtual addresses mapped */
>   	struct rb_root_cached	va;
>   
> -	/* Lock to prevent eviction while we are updating page tables */
> +	/* Lock to prevent eviction while we are updating page tables
> +	 * use vm_eviction_lock/unlock(vm)
> +	 */
>   	struct mutex		eviction_lock;
>   	bool			evicting;
> +	unsigned int		saved_flags;
>   
>   	/* BOs who needs a validation */
>   	struct list_head	evicted;

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

[PATCH 3/7] drm/amdgpu: implement tlbs invalidate on gfx9 gfx10

[Alex Sierra]

tlbs invalidate pointer function added to kiq_pm4_funcs struct.
This way, tlb flush can be done through kiq member.
TLBs invalidatation implemented for gfx9 and gfx10.

Change-Id: I1b77b364f3ae0038ff3e70e869be5f2ef6e6d293
Signed-off-by: Alex Sierra <alex.sierra@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  4 ++++
 drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 15 +++++++++++++++
 drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 14 ++++++++++++++
 3 files changed, 33 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index 8e88e0411662..af4bd279f42f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -76,11 +76,15 @@ struct kiq_pm4_funcs {
 					struct amdgpu_ring *ring,
 					u64 addr,
 					u64 seq);
+	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub);
 	/* Packet sizes */
 	int set_resources_size;
 	int map_queues_size;
 	int unmap_queues_size;
 	int query_status_size;
+	int invalidate_tlbs_size;
 };
 
 struct amdgpu_kiq {
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
index 379e46c1b7f6..d72b60f997c8 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
@@ -40,6 +40,7 @@
 #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
 
 #include "soc15.h"
+#include "soc15d.h"
 #include "soc15_common.h"
 #include "clearstate_gfx10.h"
 #include "v10_structs.h"
@@ -346,15 +347,29 @@ static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
 }
 
+static void gfx10_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub)
+{
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
+			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
+			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
+			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
+}
+
 static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
 	.kiq_set_resources = gfx10_kiq_set_resources,
 	.kiq_map_queues = gfx10_kiq_map_queues,
 	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
 	.kiq_query_status = gfx10_kiq_query_status,
+	.kiq_invalidate_tlbs = gfx10_kiq_invalidate_tlbs,
 	.set_resources_size = 8,
 	.map_queues_size = 7,
 	.unmap_queues_size = 6,
 	.query_status_size = 7,
+	.invalidate_tlbs_size = 12,
 };
 
 static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index ad0179ea2cc5..b8759386dcbb 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -837,15 +837,29 @@ static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
 	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
 }
 
+static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
+				uint16_t pasid, uint32_t flush_type,
+				bool all_hub)
+{
+	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
+	amdgpu_ring_write(kiq_ring,
+			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
+			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
+			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
+			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
+}
+
 static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
 	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
 	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
 	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
 	.kiq_query_status = gfx_v9_0_kiq_query_status,
+	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
 	.set_resources_size = 8,
 	.map_queues_size = 7,
 	.unmap_queues_size = 6,
 	.query_status_size = 7,
+	.invalidate_tlbs_size = 12,
 };
 
 static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
-- 
2.17.1

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 3/7] drm/amdgpu: implement tlbs invalidate on gfx9 gfx10

[Felix Kuehling]

The series is

Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>


On 2020-01-13 3:26 p.m., Alex Sierra wrote:
> tlbs invalidate pointer function added to kiq_pm4_funcs struct.
> This way, tlb flush can be done through kiq member.
> TLBs invalidatation implemented for gfx9 and gfx10.
>
> Change-Id: I1b77b364f3ae0038ff3e70e869be5f2ef6e6d293
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  4 ++++
>   drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 15 +++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 14 ++++++++++++++
>   3 files changed, 33 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> index 8e88e0411662..af4bd279f42f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> @@ -76,11 +76,15 @@ struct kiq_pm4_funcs {
>   					struct amdgpu_ring *ring,
>   					u64 addr,
>   					u64 seq);
> +	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub);
>   	/* Packet sizes */
>   	int set_resources_size;
>   	int map_queues_size;
>   	int unmap_queues_size;
>   	int query_status_size;
> +	int invalidate_tlbs_size;
>   };
>   
>   struct amdgpu_kiq {
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> index 379e46c1b7f6..d72b60f997c8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> @@ -40,6 +40,7 @@
>   #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   #include "clearstate_gfx10.h"
>   #include "v10_structs.h"
> @@ -346,15 +347,29 @@ static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static void gfx10_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx10_kiq_set_resources,
>   	.kiq_map_queues = gfx10_kiq_map_queues,
>   	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
>   	.kiq_query_status = gfx10_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx10_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index ad0179ea2cc5..b8759386dcbb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -837,15 +837,29 @@ static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
>   	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
>   	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
>   	.kiq_query_status = gfx_v9_0_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 3/7] drm/amdgpu: implement tlbs invalidate on gfx9 gfx10

[Felix Kuehling]

On 2020-01-13 3:26 p.m., Alex Sierra wrote:
> tlbs invalidate pointer function added to kiq_pm4_funcs struct.
> This way, tlb flush can be done through kiq member.
> TLBs invalidatation implemented for gfx9 and gfx10.
>
> Change-Id: I1b77b364f3ae0038ff3e70e869be5f2ef6e6d293
> Signed-off-by: Alex Sierra <alex.sierra@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h |  4 ++++
>   drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c  | 15 +++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c   | 14 ++++++++++++++
>   3 files changed, 33 insertions(+)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> index 8e88e0411662..af4bd279f42f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> @@ -76,11 +76,15 @@ struct kiq_pm4_funcs {
>   					struct amdgpu_ring *ring,
>   					u64 addr,
>   					u64 seq);
> +	void (*kiq_invalidate_tlbs)(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub);
>   	/* Packet sizes */
>   	int set_resources_size;
>   	int map_queues_size;
>   	int unmap_queues_size;
>   	int query_status_size;
> +	int invalidate_tlbs_size;
>   };
>   
>   struct amdgpu_kiq {
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> index 379e46c1b7f6..d72b60f997c8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
> @@ -40,6 +40,7 @@
>   #include "ivsrcid/gfx/irqsrcs_gfx_10_1.h"
>   
>   #include "soc15.h"
> +#include "soc15d.h"
>   #include "soc15_common.h"
>   #include "clearstate_gfx10.h"
>   #include "v10_structs.h"
> @@ -346,15 +347,29 @@ static void gfx10_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static void gfx10_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v10_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx10_kiq_set_resources,
>   	.kiq_map_queues = gfx10_kiq_map_queues,
>   	.kiq_unmap_queues = gfx10_kiq_unmap_queues,
>   	.kiq_query_status = gfx10_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx10_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,

This looks like it was copied from the function that emits both flush 
and fence. Now that the function only emits the flush, this number 
should be smaller. Only 2 dwords. And it seems like 12 was 
over-estimated, because the fence is only 8 dwords.

Regards,
   Felix


>   };
>   
>   static void gfx_v10_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index ad0179ea2cc5..b8759386dcbb 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -837,15 +837,29 @@ static void gfx_v9_0_kiq_query_status(struct amdgpu_ring *kiq_ring,
>   	amdgpu_ring_write(kiq_ring, upper_32_bits(seq));
>   }
>   
> +static void gfx_v9_0_kiq_invalidate_tlbs(struct amdgpu_ring *kiq_ring,
> +				uint16_t pasid, uint32_t flush_type,
> +				bool all_hub)
> +{
> +	amdgpu_ring_write(kiq_ring, PACKET3(PACKET3_INVALIDATE_TLBS, 0));
> +	amdgpu_ring_write(kiq_ring,
> +			PACKET3_INVALIDATE_TLBS_DST_SEL(1) |
> +			PACKET3_INVALIDATE_TLBS_ALL_HUB(all_hub) |
> +			PACKET3_INVALIDATE_TLBS_PASID(pasid) |
> +			PACKET3_INVALIDATE_TLBS_FLUSH_TYPE(flush_type));
> +}
> +
>   static const struct kiq_pm4_funcs gfx_v9_0_kiq_pm4_funcs = {
>   	.kiq_set_resources = gfx_v9_0_kiq_set_resources,
>   	.kiq_map_queues = gfx_v9_0_kiq_map_queues,
>   	.kiq_unmap_queues = gfx_v9_0_kiq_unmap_queues,
>   	.kiq_query_status = gfx_v9_0_kiq_query_status,
> +	.kiq_invalidate_tlbs = gfx_v9_0_kiq_invalidate_tlbs,
>   	.set_resources_size = 8,
>   	.map_queues_size = 7,
>   	.unmap_queues_size = 6,
>   	.query_status_size = 7,
> +	.invalidate_tlbs_size = 12,
>   };
>   
>   static void gfx_v9_0_set_kiq_pm4_funcs(struct amdgpu_device *adev)
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx