[PATCH 1/3] drm/amdkfd: Remove unnecessary register setting when invalidating tlb in kfd

[PATCH 1/3] drm/amdkfd: Remove unnecessary register setting when invalidating tlb in kfd

[Zhao, Yong]

Those register settings have been done in gfxhub_v1_0_program_invalidation()
and mmhub_v1_0_program_invalidation().

Change-Id: I9b9b44f17ac2a6ff0c9c78f91885665da75543d0
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 17 -----------------
 1 file changed, 17 deletions(-)

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 60b5f56c..3ade5d5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -60,11 +60,6 @@
 #define mmMMHUB_VM_INVALIDATE_ENG16_ACK				0x0705
 #define mmMMHUB_VM_INVALIDATE_ENG16_ACK_BASE_IDX		0
 
-#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32		0x0727
-#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32_BASE_IDX	0
-#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32		0x0728
-#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32_BASE_IDX	0
-
 #define V9_PIPE_PER_MEC		(4)
 #define V9_QUEUES_PER_PIPE_MEC	(8)
 
@@ -772,18 +767,6 @@ static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
 	 * TODO 2: support range-based invalidation, requires kfg2kgd
 	 * interface change
 	 */
-	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_LO32),
-				0xffffffff);
-	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_HI32),
-				0x0000001f);
-
-	WREG32(SOC15_REG_OFFSET(MMHUB, 0,
-				mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32),
-				0xffffffff);
-	WREG32(SOC15_REG_OFFSET(MMHUB, 0,
-				mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32),
-				0x0000001f);
-
 	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_REQ), req);
 
 	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_INVALIDATE_ENG16_REQ),
-- 
2.7.4

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

[PATCH 2/3] drm/amdgpu: Expose gmc_v9_0_flush_gpu_tlb_helper() for kfd to use

[Zhao, Yong]

Change-Id: I3dcd71955297c53b181f82e7078981230c642c01
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 64 ++++++++++++++++++++---------------
 drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h |  3 ++
 2 files changed, 40 insertions(+), 27 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index f35d7a5..6f96545 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -293,14 +293,15 @@ static void gmc_v9_0_set_irq_funcs(struct amdgpu_device *adev)
 	adev->gmc.vm_fault.funcs = &gmc_v9_0_irq_funcs;
 }
 
-static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid)
+static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid,
+					uint32_t flush_type)
 {
 	u32 req = 0;
 
 	/* invalidate using legacy mode on vmid*/
 	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ,
 			    PER_VMID_INVALIDATE_REQ, 1 << vmid);
-	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 0);
+	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, flush_type);
 	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 1);
 	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 1);
 	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 1);
@@ -354,32 +355,15 @@ static signed long  amdgpu_kiq_reg_write_reg_wait(struct amdgpu_device *adev,
 	return r;
 }
 
-/*
- * GART
- * VMID 0 is the physical GPU addresses as used by the kernel.
- * VMIDs 1-15 are used for userspace clients and are handled
- * by the amdgpu vm/hsa code.
- */
-
-/**
- * gmc_v9_0_flush_gpu_tlb - gart tlb flush callback
- *
- * @adev: amdgpu_device pointer
- * @vmid: vm instance to flush
- *
- * Flush the TLB for the requested page table.
- */
-static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
-					uint32_t vmid)
+void gmc_v9_0_flush_gpu_tlb_helper(struct amdgpu_device *adev, uint32_t vmid,
+				uint32_t flush_type, uint32_t eng, bool lock)
 {
-	/* Use register 17 for GART */
-	const unsigned eng = 17;
 	unsigned i, j;
 	int r;
 
 	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
 		struct amdgpu_vmhub *hub = &adev->vmhub[i];
-		u32 tmp = gmc_v9_0_get_invalidate_req(vmid);
+		u32 tmp = gmc_v9_0_get_invalidate_req(vmid, flush_type);
 
 		if (adev->gfx.kiq.ring.ready &&
 		    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) &&
@@ -390,7 +374,8 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
 				continue;
 		}
 
-		spin_lock(&adev->gmc.invalidate_lock);
+		if (lock)
+			spin_lock(&adev->gmc.invalidate_lock);
 
 		WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp);
 
@@ -403,7 +388,8 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
 			cpu_relax();
 		}
 		if (j < 100) {
-			spin_unlock(&adev->gmc.invalidate_lock);
+			if (lock)
+				spin_unlock(&adev->gmc.invalidate_lock);
 			continue;
 		}
 
@@ -416,20 +402,44 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
 			udelay(1);
 		}
 		if (j < adev->usec_timeout) {
-			spin_unlock(&adev->gmc.invalidate_lock);
+			if (lock)
+				spin_unlock(&adev->gmc.invalidate_lock);
 			continue;
 		}
-		spin_unlock(&adev->gmc.invalidate_lock);
+		if (lock)
+			spin_unlock(&adev->gmc.invalidate_lock);
 		DRM_ERROR("Timeout waiting for VM flush ACK!\n");
 	}
 }
 
+/*
+ * GART
+ * VMID 0 is the physical GPU addresses as used by the kernel.
+ * VMIDs 1-15 are used for userspace clients and are handled
+ * by the amdgpu vm/hsa code.
+ */
+
+/**
+ * gmc_v9_0_flush_gpu_tlb - gart tlb flush callback
+ *
+ * @adev: amdgpu_device pointer
+ * @vmid: vm instance to flush
+ *
+ * Flush the TLB for the requested page table.
+ */
+static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
+					uint32_t vmid)
+{
+	/* Use engine 17 for amdgpu */
+	gmc_v9_0_flush_gpu_tlb_helper(adev, vmid, 0, 17, true);
+}
+
 static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
 					    unsigned vmid, uint64_t pd_addr)
 {
 	struct amdgpu_device *adev = ring->adev;
 	struct amdgpu_vmhub *hub = &adev->vmhub[ring->funcs->vmhub];
-	uint32_t req = gmc_v9_0_get_invalidate_req(vmid);
+	uint32_t req = gmc_v9_0_get_invalidate_req(vmid, 0);
 	unsigned eng = ring->vm_inv_eng;
 
 	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_lo32 + (2 * vmid),
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h
index 1fd178a6..56f504a 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h
@@ -33,4 +33,7 @@ void gfxhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
 void mmhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
 				uint64_t page_table_base);
 
+void gmc_v9_0_flush_gpu_tlb_helper(struct amdgpu_device *adev, uint32_t vmid,
+				uint32_t flush_type, uint32_t eng, bool lock);
+
 #endif
-- 
2.7.4

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

[PATCH 3/3] drm/amdkfd: Use functions from amdgpu to invalidate vmid in kfd

[Zhao, Yong]

Change-Id: I306305e43d4b4032316909b3f4e3f9f5ca4520ae
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 32 +----------------------
 1 file changed, 1 insertion(+), 31 deletions(-)

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 3ade5d5..18f161b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
@@ -48,17 +48,6 @@
 #include "soc15d.h"
 #include "gmc_v9_0.h"
 
-/* HACK: MMHUB and GC both have VM-related register with the same
- * names but different offsets. Define the MMHUB register we need here
- * with a prefix. A proper solution would be to move the functions
- * programming these registers into gfx_v9_0.c and mmhub_v1_0.c
- * respectively.
- */
-#define mmMMHUB_VM_INVALIDATE_ENG16_REQ				0x06f3
-#define mmMMHUB_VM_INVALIDATE_ENG16_REQ_BASE_IDX		0
-
-#define mmMMHUB_VM_INVALIDATE_ENG16_ACK				0x0705
-#define mmMMHUB_VM_INVALIDATE_ENG16_ACK_BASE_IDX		0
 
 #define V9_PIPE_PER_MEC		(4)
 #define V9_QUEUES_PER_PIPE_MEC	(8)
@@ -742,13 +731,6 @@ static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
 static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
 {
 	struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
-	uint32_t req = (1 << vmid) |
-		(0 << VM_INVALIDATE_ENG16_REQ__FLUSH_TYPE__SHIFT) | /* legacy */
-		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PTES_MASK |
-		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE0_MASK |
-		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE1_MASK |
-		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L2_PDE2_MASK |
-		VM_INVALIDATE_ENG16_REQ__INVALIDATE_L1_PTES_MASK;
 
 	mutex_lock(&adev->srbm_mutex);
 
@@ -767,19 +749,7 @@ static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
 	 * TODO 2: support range-based invalidation, requires kfg2kgd
 	 * interface change
 	 */
-	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_REQ), req);
-
-	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_INVALIDATE_ENG16_REQ),
-				req);
-
-	while (!(RREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ACK)) &
-					(1 << vmid)))
-		cpu_relax();
-
-	while (!(RREG32(SOC15_REG_OFFSET(MMHUB, 0,
-					mmMMHUB_VM_INVALIDATE_ENG16_ACK)) &
-					(1 << vmid)))
-		cpu_relax();
+	gmc_v9_0_flush_gpu_tlb_helper(adev, vmid, 0, 16, false);
 
 	mutex_unlock(&adev->srbm_mutex);
 
-- 
2.7.4

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

Re: [PATCH 1/3] drm/amdkfd: Remove unnecessary register setting when invalidating tlb in kfd

[Kuehling, Felix]

Patch 1 is Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>

Patch 2: I'm not sure we need the "lock" parameter and the invalidation
engine parameter. If we're serious about consolidating TLB invalidation
between amdgpu and KFD, I think we should use the same invalidation
engine and the same lock. Then you also don't need to take the
adev->srbm_mutex any more in write_vmid_invalidate_request, which we
were abusing for this purpose.

Regards,
  Felix


On 2018-10-22 2:41 p.m., Zhao, Yong wrote:
> Those register settings have been done in gfxhub_v1_0_program_invalidation()
> and mmhub_v1_0_program_invalidation().
>
> Change-Id: I9b9b44f17ac2a6ff0c9c78f91885665da75543d0
> Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c | 17 -----------------
>  1 file changed, 17 deletions(-)
>
> 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 60b5f56c..3ade5d5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
> @@ -60,11 +60,6 @@
>  #define mmMMHUB_VM_INVALIDATE_ENG16_ACK				0x0705
>  #define mmMMHUB_VM_INVALIDATE_ENG16_ACK_BASE_IDX		0
>  
> -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32		0x0727
> -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32_BASE_IDX	0
> -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32		0x0728
> -#define mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32_BASE_IDX	0
> -
>  #define V9_PIPE_PER_MEC		(4)
>  #define V9_QUEUES_PER_PIPE_MEC	(8)
>  
> @@ -772,18 +767,6 @@ static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
>  	 * TODO 2: support range-based invalidation, requires kfg2kgd
>  	 * interface change
>  	 */
> -	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_LO32),
> -				0xffffffff);
> -	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_ADDR_RANGE_HI32),
> -				0x0000001f);
> -
> -	WREG32(SOC15_REG_OFFSET(MMHUB, 0,
> -				mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_LO32),
> -				0xffffffff);
> -	WREG32(SOC15_REG_OFFSET(MMHUB, 0,
> -				mmMMHUB_VM_INVALIDATE_ENG16_ADDR_RANGE_HI32),
> -				0x0000001f);
> -
>  	WREG32(SOC15_REG_OFFSET(GC, 0, mmVM_INVALIDATE_ENG16_REQ), req);
>  
>  	WREG32(SOC15_REG_OFFSET(MMHUB, 0, mmMMHUB_VM_INVALIDATE_ENG16_REQ),
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Re: [PATCH 2/3] drm/amdgpu: Expose gmc_v9_0_flush_gpu_tlb_helper() for kfd to use

[Koenig, Christian]

Am 22.10.18 um 20:41 schrieb Zhao, Yong:
> Change-Id: I3dcd71955297c53b181f82e7078981230c642c01
> Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 64 ++++++++++++++++++++---------------
>   drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h |  3 ++
>   2 files changed, 40 insertions(+), 27 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> index f35d7a5..6f96545 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
> @@ -293,14 +293,15 @@ static void gmc_v9_0_set_irq_funcs(struct amdgpu_device *adev)
>   	adev->gmc.vm_fault.funcs = &gmc_v9_0_irq_funcs;
>   }
>   
> -static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid)
> +static uint32_t gmc_v9_0_get_invalidate_req(unsigned int vmid,
> +					uint32_t flush_type)
>   {
>   	u32 req = 0;
>   
>   	/* invalidate using legacy mode on vmid*/
>   	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ,
>   			    PER_VMID_INVALIDATE_REQ, 1 << vmid);
> -	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, 0);
> +	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, FLUSH_TYPE, flush_type);
>   	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PTES, 1);
>   	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE0, 1);
>   	req = REG_SET_FIELD(req, VM_INVALIDATE_ENG0_REQ, INVALIDATE_L2_PDE1, 1);
> @@ -354,32 +355,15 @@ static signed long  amdgpu_kiq_reg_write_reg_wait(struct amdgpu_device *adev,
>   	return r;
>   }
>   
> -/*
> - * GART
> - * VMID 0 is the physical GPU addresses as used by the kernel.
> - * VMIDs 1-15 are used for userspace clients and are handled
> - * by the amdgpu vm/hsa code.
> - */
> -
> -/**
> - * gmc_v9_0_flush_gpu_tlb - gart tlb flush callback
> - *
> - * @adev: amdgpu_device pointer
> - * @vmid: vm instance to flush
> - *
> - * Flush the TLB for the requested page table.
> - */
> -static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
> -					uint32_t vmid)
> +void gmc_v9_0_flush_gpu_tlb_helper(struct amdgpu_device *adev, uint32_t vmid,
> +				uint32_t flush_type, uint32_t eng, bool lock)

This one needs kernel documentation and a better name would be nice.

In general sounds like a nice cleanup, but I exposing ASIC specific 
functions is not necessarily a good idea.

We should probably move the ASIC specific KFD code into this file instead.

Christian.

>   {
> -	/* Use register 17 for GART */
> -	const unsigned eng = 17;
>   	unsigned i, j;
>   	int r;
>   
>   	for (i = 0; i < AMDGPU_MAX_VMHUBS; ++i) {
>   		struct amdgpu_vmhub *hub = &adev->vmhub[i];
> -		u32 tmp = gmc_v9_0_get_invalidate_req(vmid);
> +		u32 tmp = gmc_v9_0_get_invalidate_req(vmid, flush_type);
>   
>   		if (adev->gfx.kiq.ring.ready &&
>   		    (amdgpu_sriov_runtime(adev) || !amdgpu_sriov_vf(adev)) &&
> @@ -390,7 +374,8 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
>   				continue;
>   		}
>   
> -		spin_lock(&adev->gmc.invalidate_lock);
> +		if (lock)
> +			spin_lock(&adev->gmc.invalidate_lock);
>   
>   		WREG32_NO_KIQ(hub->vm_inv_eng0_req + eng, tmp);
>   
> @@ -403,7 +388,8 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
>   			cpu_relax();
>   		}
>   		if (j < 100) {
> -			spin_unlock(&adev->gmc.invalidate_lock);
> +			if (lock)
> +				spin_unlock(&adev->gmc.invalidate_lock);
>   			continue;
>   		}
>   
> @@ -416,20 +402,44 @@ static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
>   			udelay(1);
>   		}
>   		if (j < adev->usec_timeout) {
> -			spin_unlock(&adev->gmc.invalidate_lock);
> +			if (lock)
> +				spin_unlock(&adev->gmc.invalidate_lock);
>   			continue;
>   		}
> -		spin_unlock(&adev->gmc.invalidate_lock);
> +		if (lock)
> +			spin_unlock(&adev->gmc.invalidate_lock);
>   		DRM_ERROR("Timeout waiting for VM flush ACK!\n");
>   	}
>   }
>   
> +/*
> + * GART
> + * VMID 0 is the physical GPU addresses as used by the kernel.
> + * VMIDs 1-15 are used for userspace clients and are handled
> + * by the amdgpu vm/hsa code.
> + */
> +
> +/**
> + * gmc_v9_0_flush_gpu_tlb - gart tlb flush callback
> + *
> + * @adev: amdgpu_device pointer
> + * @vmid: vm instance to flush
> + *
> + * Flush the TLB for the requested page table.
> + */
> +static void gmc_v9_0_flush_gpu_tlb(struct amdgpu_device *adev,
> +					uint32_t vmid)
> +{
> +	/* Use engine 17 for amdgpu */
> +	gmc_v9_0_flush_gpu_tlb_helper(adev, vmid, 0, 17, true);
> +}
> +
>   static uint64_t gmc_v9_0_emit_flush_gpu_tlb(struct amdgpu_ring *ring,
>   					    unsigned vmid, uint64_t pd_addr)
>   {
>   	struct amdgpu_device *adev = ring->adev;
>   	struct amdgpu_vmhub *hub = &adev->vmhub[ring->funcs->vmhub];
> -	uint32_t req = gmc_v9_0_get_invalidate_req(vmid);
> +	uint32_t req = gmc_v9_0_get_invalidate_req(vmid, 0);
>   	unsigned eng = ring->vm_inv_eng;
>   
>   	amdgpu_ring_emit_wreg(ring, hub->ctx0_ptb_addr_lo32 + (2 * vmid),
> diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h
> index 1fd178a6..56f504a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h
> +++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.h
> @@ -33,4 +33,7 @@ void gfxhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
>   void mmhub_v1_0_setup_vm_pt_regs(struct amdgpu_device *adev, uint32_t vmid,
>   				uint64_t page_table_base);
>   
> +void gmc_v9_0_flush_gpu_tlb_helper(struct amdgpu_device *adev, uint32_t vmid,
> +				uint32_t flush_type, uint32_t eng, bool lock);
> +
>   #endif

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