[PATCH 1/2] drm/amdgpu:update kernel vcn ring test

[PATCH 1/2] drm/amdgpu:update kernel vcn ring test

[Saleemkhan Jamadar]

add session context buffer to decoder ring test.

v5 - clear the session ct buffer (Christian)
v4 - data type, explain change of ib size change (Christian)
v3 - indent and  v2 changes correction. (Christian)
v2 - put the buffer at the end of the IB (Christian)

Signed-off-by: Saleemkhan Jamadar <saleemkhan.jamadar@amd.com>
Acked-by: Leo Liu <leo.liu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 13 +++++++++++--
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h |  5 ++++-
 2 files changed, 15 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
index 4e1256af80b6..8966ffb2f4a0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
@@ -591,13 +591,15 @@ static int amdgpu_vcn_dec_get_create_msg(struct amdgpu_ring *ring, uint32_t hand
 	int r, i;
 
 	memset(ib, 0, sizeof(*ib));
-	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
+	/* 34 pages : 128KiB  session context buffer size and 8KiB ib msg */
+	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 34,
 			AMDGPU_IB_POOL_DIRECT,
 			ib);
 	if (r)
 		return r;
 
 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
+	memset(msg, 0, (AMDGPU_GPU_PAGE_SIZE * 34));
 	msg[0] = cpu_to_le32(0x00000028);
 	msg[1] = cpu_to_le32(0x00000038);
 	msg[2] = cpu_to_le32(0x00000001);
@@ -626,13 +628,15 @@ static int amdgpu_vcn_dec_get_destroy_msg(struct amdgpu_ring *ring, uint32_t han
 	int r, i;
 
 	memset(ib, 0, sizeof(*ib));
-	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 2,
+	/* 34 pages : 128KiB  session context buffer size and 8KiB ib msg */
+	r = amdgpu_ib_get(adev, NULL, AMDGPU_GPU_PAGE_SIZE * 34,
 			AMDGPU_IB_POOL_DIRECT,
 			ib);
 	if (r)
 		return r;
 
 	msg = (uint32_t *)AMDGPU_GPU_PAGE_ALIGN((unsigned long)ib->ptr);
+	memset(msg, 0, (AMDGPU_GPU_PAGE_SIZE * 34));
 	msg[0] = cpu_to_le32(0x00000028);
 	msg[1] = cpu_to_le32(0x00000018);
 	msg[2] = cpu_to_le32(0x00000000);
@@ -718,6 +722,7 @@ static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
 	struct amdgpu_job *job;
 	struct amdgpu_ib *ib;
 	uint64_t addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
+	uint64_t session_ctx_buf_gaddr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr + 8192);
 	bool sq = amdgpu_vcn_using_unified_queue(ring);
 	uint32_t *ib_checksum;
 	uint32_t ib_pack_in_dw;
@@ -748,6 +753,10 @@ static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
 
+	decode_buffer->valid_buf_flag |=
+				cpu_to_le32(AMDGPU_VCN_CMD_FLAG_SESSION_CONTEXT_BUFFER);
+	decode_buffer->session_context_buffer_address_hi = upper_32_bits(session_ctx_buf_gaddr);
+	decode_buffer->session_context_buffer_address_lo = lower_32_bits(session_ctx_buf_gaddr);
 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
index 1f1d7dc94f90..ba5fefd74197 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
@@ -171,6 +171,7 @@
 
 #define AMDGPU_VCN_IB_FLAG_DECODE_BUFFER	0x00000001
 #define AMDGPU_VCN_CMD_FLAG_MSG_BUFFER		0x00000001
+#define AMDGPU_VCN_CMD_FLAG_SESSION_CONTEXT_BUFFER	0x00100000
 
 #define VCN_CODEC_DISABLE_MASK_AV1  (1 << 0)
 #define VCN_CODEC_DISABLE_MASK_VP9  (1 << 1)
@@ -366,7 +367,9 @@ struct amdgpu_vcn_decode_buffer {
 	uint32_t valid_buf_flag;
 	uint32_t msg_buffer_address_hi;
 	uint32_t msg_buffer_address_lo;
-	uint32_t pad[30];
+	uint32_t session_context_buffer_address_hi;
+	uint32_t session_context_buffer_address_lo;
+	uint32_t pad[28];
 };
 
 #define VCN_BLOCK_ENCODE_DISABLE_MASK 0x80
-- 
2.25.1

[PATCH 2/2] drm/amdgpu: update kernel vcn ring test

[Saleemkhan Jamadar]

add session context buffer to decoder ring test for vcn v1 to v3.

v3 - correct the cmd for sesssion ctx buf
v2 - add the buffer into IB (Leo liu)

Signed-off-by: Saleemkhan Jamadar <saleemkhan.jamadar@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
index 8966ffb2f4a0..5c2aa925dd79 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
@@ -539,6 +539,7 @@ static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
 				   struct dma_fence **fence)
 {
 	u64 addr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr);
+	uint64_t session_ctx_buf_gaddr = AMDGPU_GPU_PAGE_ALIGN(ib_msg->gpu_addr + 8192);
 	struct amdgpu_device *adev = ring->adev;
 	struct dma_fence *f = NULL;
 	struct amdgpu_job *job;
@@ -552,13 +553,23 @@ static int amdgpu_vcn_dec_send_msg(struct amdgpu_ring *ring,
 		goto err;
 
 	ib = &job->ibs[0];
-	ib->ptr[0] = PACKET0(adev->vcn.internal.data0, 0);
-	ib->ptr[1] = addr;
-	ib->ptr[2] = PACKET0(adev->vcn.internal.data1, 0);
-	ib->ptr[3] = addr >> 32;
-	ib->ptr[4] = PACKET0(adev->vcn.internal.cmd, 0);
-	ib->ptr[5] = 0;
-	for (i = 6; i < 16; i += 2) {
+	ib->length_dw = 0;
+	ib->ptr[ib->length_dw++] = PACKET0(adev->vcn.internal.data0, 0);
+	ib->ptr[ib->length_dw++] = lower_32_bits(session_ctx_buf_gaddr);
+	ib->ptr[ib->length_dw++] = PACKET0(adev->vcn.internal.data1, 0);
+	ib->ptr[ib->length_dw++] = upper_32_bits(session_ctx_buf_gaddr);
+	/* session ctx buffer cmd */
+	ib->ptr[ib->length_dw++] = PACKET0(adev->vcn.internal.cmd, 0);
+	ib->ptr[ib->length_dw++] = 0xa;
+
+	ib->ptr[ib->length_dw++] = PACKET0(adev->vcn.internal.data0, 0);
+	ib->ptr[ib->length_dw++] = lower_32_bits(addr);
+	ib->ptr[ib->length_dw++] = PACKET0(adev->vcn.internal.data1, 0);
+	ib->ptr[ib->length_dw++] = upper_32_bits(addr);
+	ib->ptr[ib->length_dw++] = PACKET0(adev->vcn.internal.cmd, 0);
+	ib->ptr[ib->length_dw++] = 0;
+
+	for (i = ib->length_dw; i < 16; i += 2) {
 		ib->ptr[i] = PACKET0(adev->vcn.internal.nop, 0);
 		ib->ptr[i+1] = 0;
 	}
-- 
2.25.1

Re: [PATCH 1/2] drm/amdgpu:update kernel vcn ring test

[Christian König]

Am 26.06.23 um 16:50 schrieb Saleemkhan Jamadar:
> add session context buffer to decoder ring test.
>
> v2 - put the buffer at the end of the IB (Christian)
>
> Signed-off-by: Saleemkhan Jamadar <saleemkhan.jamadar@amd.com>
> Acked-by: Leo Liu <leo.liu@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 30 ++++++++++++++++++++++---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h |  3 +++
>   2 files changed, 30 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> index 2d94f1b63bd6..04daaaf6ab34 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
> @@ -691,7 +691,8 @@ static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
>   
>   static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
>   				      struct amdgpu_ib *ib_msg,
> -				      struct dma_fence **fence)
> +				      struct dma_fence **fence,
> +					  uint64_t session_ctx_buf_gaddr)

That looks like it isn't correctly indented.

>   {
>   	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
>   	unsigned int ib_size_dw = 64;
> @@ -730,6 +731,14 @@ static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
>   	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
>   	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
>   
> +	if (session_ctx_buf_gaddr) {
> +		decode_buffer->valid_buf_flag |=
> +					cpu_to_le32(AMDGPU_VCN_CMD_FLAG_SESSION_CONTEXT_BUFFER);
> +		decode_buffer->session_context_buffer_address_hi =
> +							cpu_to_le32(session_ctx_buf_gaddr >> 32);
> +		decode_buffer->session_context_buffer_address_lo =
> +							cpu_to_le32(session_ctx_buf_gaddr);
> +	}
>   	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
>   	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
>   	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
> @@ -763,20 +772,34 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
>   {
>   	struct dma_fence *fence = NULL;
>   	struct amdgpu_ib ib;
> +	struct amdgpu_bo *session_ctx_buf = NULL;
> +	void *cpu_addr = NULL;
> +	uint64_t gpu_addr = 0;
>   	long r;
>   
> +	r = amdgpu_bo_create_kernel(ring->adev, 128*1024, PAGE_SIZE,
> +				    AMDGPU_GEM_DOMAIN_VRAM |
> +				    AMDGPU_GEM_DOMAIN_GTT,
> +				    &session_ctx_buf,
> +				    &gpu_addr,
> +				    &cpu_addr);

That still creates a buffer for the session ctx instead of putting it 
into the IB.

> +	if (r) {
> +		dev_err(ring->adev->dev, "VCN ib test:%ld failed to allocate session ctx bo\n", r);
> +		return r;
> +	}
> +
>   	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
>   	if (r)
>   		goto error;
>   
> -	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
> +	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL, gpu_addr);
>   	if (r)
>   		goto error;
>   	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
>   	if (r)
>   		goto error;
>   
> -	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
> +	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence, gpu_addr);
>   	if (r)
>   		goto error;
>   
> @@ -788,6 +811,7 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
>   
>   	dma_fence_put(fence);
>   error:
> +	amdgpu_bo_free_kernel(&session_ctx_buf, &gpu_addr, &cpu_addr);ebd59851c796c

Which is freed here.

>   	return r;
>   }
>   
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
> index f1397ef66fd7..06f9ee91a1e1 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
> @@ -166,6 +166,7 @@
>   
>   #define AMDGPU_VCN_IB_FLAG_DECODE_BUFFER	0x00000001
>   #define AMDGPU_VCN_CMD_FLAG_MSG_BUFFER		0x00000001
> +#define AMDGPU_VCN_CMD_FLAG_SESSION_CONTEXT_BUFFER	0x00100000
>   
>   #define VCN_CODEC_DISABLE_MASK_AV1  (1 << 0)
>   #define VCN_CODEC_DISABLE_MASK_VP9  (1 << 1)
> @@ -357,6 +358,8 @@ struct amdgpu_vcn_decode_buffer {
>   	uint32_t valid_buf_flag;
>   	uint32_t msg_buffer_address_hi;
>   	uint32_t msg_buffer_address_lo;
> +	unsigned int session_context_buffer_address_hi;
> +	unsigned int session_context_buffer_address_lo;
>   	uint32_t pad[30];

that here looks incorrect as well.

Christian.

>   };
>   

[PATCH 1/2] drm/amdgpu:update kernel vcn ring test

[Saleemkhan Jamadar]

add session context buffer to decoder ring test.

v2 - put the buffer at the end of the IB (Christian)

Signed-off-by: Saleemkhan Jamadar <saleemkhan.jamadar@amd.com>
Acked-by: Leo Liu <leo.liu@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c | 30 ++++++++++++++++++++++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h |  3 +++
 2 files changed, 30 insertions(+), 3 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
index 2d94f1b63bd6..04daaaf6ab34 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.c
@@ -691,7 +691,8 @@ static void amdgpu_vcn_unified_ring_ib_checksum(uint32_t **ib_checksum,
 
 static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
 				      struct amdgpu_ib *ib_msg,
-				      struct dma_fence **fence)
+				      struct dma_fence **fence,
+					  uint64_t session_ctx_buf_gaddr)
 {
 	struct amdgpu_vcn_decode_buffer *decode_buffer = NULL;
 	unsigned int ib_size_dw = 64;
@@ -730,6 +731,14 @@ static int amdgpu_vcn_dec_sw_send_msg(struct amdgpu_ring *ring,
 	ib->length_dw += sizeof(struct amdgpu_vcn_decode_buffer) / 4;
 	memset(decode_buffer, 0, sizeof(struct amdgpu_vcn_decode_buffer));
 
+	if (session_ctx_buf_gaddr) {
+		decode_buffer->valid_buf_flag |=
+					cpu_to_le32(AMDGPU_VCN_CMD_FLAG_SESSION_CONTEXT_BUFFER);
+		decode_buffer->session_context_buffer_address_hi =
+							cpu_to_le32(session_ctx_buf_gaddr >> 32);
+		decode_buffer->session_context_buffer_address_lo =
+							cpu_to_le32(session_ctx_buf_gaddr);
+	}
 	decode_buffer->valid_buf_flag |= cpu_to_le32(AMDGPU_VCN_CMD_FLAG_MSG_BUFFER);
 	decode_buffer->msg_buffer_address_hi = cpu_to_le32(addr >> 32);
 	decode_buffer->msg_buffer_address_lo = cpu_to_le32(addr);
@@ -763,20 +772,34 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 {
 	struct dma_fence *fence = NULL;
 	struct amdgpu_ib ib;
+	struct amdgpu_bo *session_ctx_buf = NULL;
+	void *cpu_addr = NULL;
+	uint64_t gpu_addr = 0;
 	long r;
 
+	r = amdgpu_bo_create_kernel(ring->adev, 128*1024, PAGE_SIZE,
+				    AMDGPU_GEM_DOMAIN_VRAM |
+				    AMDGPU_GEM_DOMAIN_GTT,
+				    &session_ctx_buf,
+				    &gpu_addr,
+				    &cpu_addr);
+	if (r) {
+		dev_err(ring->adev->dev, "VCN ib test:%ld failed to allocate session ctx bo\n", r);
+		return r;
+	}
+
 	r = amdgpu_vcn_dec_get_create_msg(ring, 1, &ib);
 	if (r)
 		goto error;
 
-	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL);
+	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, NULL, gpu_addr);
 	if (r)
 		goto error;
 	r = amdgpu_vcn_dec_get_destroy_msg(ring, 1, &ib);
 	if (r)
 		goto error;
 
-	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence);
+	r = amdgpu_vcn_dec_sw_send_msg(ring, &ib, &fence, gpu_addr);
 	if (r)
 		goto error;
 
@@ -788,6 +811,7 @@ int amdgpu_vcn_dec_sw_ring_test_ib(struct amdgpu_ring *ring, long timeout)
 
 	dma_fence_put(fence);
 error:
+	amdgpu_bo_free_kernel(&session_ctx_buf, &gpu_addr, &cpu_addr);
 	return r;
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
index f1397ef66fd7..06f9ee91a1e1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
@@ -166,6 +166,7 @@
 
 #define AMDGPU_VCN_IB_FLAG_DECODE_BUFFER	0x00000001
 #define AMDGPU_VCN_CMD_FLAG_MSG_BUFFER		0x00000001
+#define AMDGPU_VCN_CMD_FLAG_SESSION_CONTEXT_BUFFER	0x00100000
 
 #define VCN_CODEC_DISABLE_MASK_AV1  (1 << 0)
 #define VCN_CODEC_DISABLE_MASK_VP9  (1 << 1)
@@ -357,6 +358,8 @@ struct amdgpu_vcn_decode_buffer {
 	uint32_t valid_buf_flag;
 	uint32_t msg_buffer_address_hi;
 	uint32_t msg_buffer_address_lo;
+	unsigned int session_context_buffer_address_hi;
+	unsigned int session_context_buffer_address_lo;
 	uint32_t pad[30];
 };
 
-- 
2.25.1