Skip to contentSkip to conversation

share semaphore across process

PATCH In review Chunming Zhou 2 participants 9 messages Aug 18, 2016 – Aug 22, 2016 5 messages not in the archive

lore
Share & tools
lore
#0 lore raw

cover share semaphore across process

after ctx id is valid in global side, we share semaphore across process based on BO sharing mechanism. That means we map semaphore object to a bo, then sharing bo with other process, the other process can get the semaphore object from the sharing bo.

Chunming Zhou (4):
amdgpu: use drm_amdgpu_fence instead of amdgpu_cs_fence in semaphore
structure
amdgpu: add export/import semaphore apis
amdgpu: add mutex for across process reason
tests/amdgpu: add semaphore across process test

 amdgpu/amdgpu.h            |  40 ++++++++++++++
 amdgpu/amdgpu_cs.c         | 113 ++++++++++++++++++++++++++++++++------
 amdgpu/amdgpu_internal.h   |   6 ++-
 tests/amdgpu/basic_tests.c | 131 +++++++++++++++++++++++++++++++++++++++++++++
 4 files changed, 273 insertions(+), 17 deletions(-)
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1471506959-905-1-git-send-email-David1.Zhou-5C7GfCeVMHo@public.gmane.org

#2 lore raw

PATCH 1/4 amdgpu: use drm_amdgpu_fence instead of amdgpu_cs_fence in semaphore structure

semaphore just need to store ctx id not context itself, which will bring convenience for sharing semaphore accross process.

Change-Id I46cf54c61ee6143a77d18510c3591bcc97fa8b24
---
 amdgpu/amdgpu_cs.c       | 20 ++++++++++----------
 amdgpu/amdgpu_internal.h |  2 +-
 2 files changed, 11 insertions(+), 11 deletions(-)
amdgpu/amdgpu_cs.c +10 -10
index bb0d79b..c76a675 100644
@@ -284,13 +284,13 @@ static int amdgpu_cs_submit_one(amdgpu_context_handle context,
 		}
 		sem_count = 0;
 		LIST_FOR_EACH_ENTRY_SAFE(sem, tmp, sem_list, list) {
-			struct amdgpu_cs_fence *info = &sem->signal_fence;
+			struct drm_amdgpu_fence *info = &sem->signal_fence;
 			struct drm_amdgpu_cs_chunk_dep *dep = &sem_dependencies[sem_count++];
 			dep->ip_type = info->ip_type;
 			dep->ip_instance = info->ip_instance;
 			dep->ring = info->ring;
-			dep->ctx_id = info->context->id;
-			dep->handle = info->fence;
+			dep->ctx_id = info->ctx_id;
+			dep->handle = info->seq_no;
 
 			list_del(&sem->list);
 			amdgpu_cs_reset_sem(sem);
@@ -550,14 +550,14 @@ int amdgpu_cs_signal_semaphore(amdgpu_context_handle ctx,
 	if (NULL == sem)
 		return -EINVAL;
 	/* sem has been signaled */
-	if (sem->signal_fence.context)
+	if (sem->signal_fence.ctx_id)
 		return -EINVAL;
 	pthread_mutex_lock(&ctx->sequence_mutex);
-	sem->signal_fence.context = ctx;
+	sem->signal_fence.ctx_id = ctx->id;
 	sem->signal_fence.ip_type = ip_type;
 	sem->signal_fence.ip_instance = ip_instance;
 	sem->signal_fence.ring = ring;
-	sem->signal_fence.fence = ctx->last_seq[ip_type][ip_instance][ring];
+	sem->signal_fence.seq_no = ctx->last_seq[ip_type][ip_instance][ring];
 	update_references(NULL, &sem->refcount);
 	pthread_mutex_unlock(&ctx->sequence_mutex);
 	return 0;
@@ -578,7 +578,7 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
 	if (NULL == sem)
 		return -EINVAL;
 	/* must signal first */
-	if (NULL == sem->signal_fence.context)
+	if (0 == sem->signal_fence.ctx_id)
 		return -EINVAL;
 
 	pthread_mutex_lock(&ctx->sequence_mutex);
@@ -591,14 +591,14 @@ static int amdgpu_cs_reset_sem(amdgpu_semaphore_handle sem)
 {
 	if (NULL == sem)
 		return -EINVAL;
-	if (NULL == sem->signal_fence.context)
+	if (0 == sem->signal_fence.ctx_id)
 		return -EINVAL;
 
-	sem->signal_fence.context = NULL;;
+	sem->signal_fence.ctx_id = 0;
 	sem->signal_fence.ip_type = 0;
 	sem->signal_fence.ip_instance = 0;
 	sem->signal_fence.ring = 0;
-	sem->signal_fence.fence = 0;
+	sem->signal_fence.seq_no = 0;
 
 	return 0;
 }
amdgpu/amdgpu_internal.h +1 -1
index 1160a12..ccc85d7 100644
@@ -133,7 +133,7 @@ struct amdgpu_context {
 struct amdgpu_semaphore {
 	atomic_t refcount;
 	struct list_head list;
-	struct amdgpu_cs_fence signal_fence;
+	struct drm_amdgpu_fence signal_fence;
 };
 
 /**
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#3 lore raw

PATCH 2/4 amdgpu: add export/import semaphore apis

In reply to a message not in the archive

They are used for sharing semaphore across process.

Change-Id I262adf10913d365bb93368b492e69140af522c64
---
 amdgpu/amdgpu.h          | 40 ++++++++++++++++++++++++++++++
 amdgpu/amdgpu_cs.c       | 63 ++++++++++++++++++++++++++++++++++++++++++++++--
 amdgpu/amdgpu_internal.h |  2 ++
 3 files changed, 103 insertions(+), 2 deletions(-)
amdgpu/amdgpu.h +40 -0
index 693d841..e716855 100644
@@ -1379,6 +1379,19 @@ int amdgpu_svm_commit(amdgpu_va_handle va_range_handle,
 int amdgpu_svm_uncommit(amdgpu_va_handle va_range_handle);
 
 /**
+ *  create shared semaphore
+ *
+ * \param amdgpu_device_handle
+ * \param   sem	   - \c [out] semaphore handle
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem);
+
+/**
  *  create semaphore
  *
  * \param   sem	   - \c [out] semaphore handle
@@ -1439,6 +1452,33 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
 int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem);
 
 /**
+ *  export semaphore
+ *
+ * \param   sem	    - \c [in] semaphore handle
+ * \param   shared_handle    - \c [out] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle);
+/**
+ *  import semaphore
+ *
+ * \param   sem	    - \c [out] semaphore handle
+ * \param   dev	    - \c [in] device handle
+ * \param   shared_handle    - \c [in] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev,
+			       uint32_t shared_handle);
+
+/**
  *  Get the ASIC marketing name
  *
  * \param   dev         - \c [in] Device handle. See #amdgpu_device_initialize()
amdgpu/amdgpu_cs.c +61 -2

85 lines not on this page · raw message

amdgpu/amdgpu_internal.h +2 -0
index ccc85d7..7c422da 100644
@@ -134,6 +134,8 @@ struct amdgpu_semaphore {
 	atomic_t refcount;
 	struct list_head list;
 	struct drm_amdgpu_fence signal_fence;
+	amdgpu_bo_handle buf_handle;
+	uint32_t version;
 };
 
 /**
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 1471506959-905-3-git-send-email-David1.Zhou-5C7GfCeVMHo@public.gmane.org

#5 lore raw

Re: [PATCH 2/4] amdgpu: add export/import semaphore apis

Quoted 74 lines · Chunming Zhou

They are used for sharing semaphore across process.

Change-Id I262adf10913d365bb93368b492e69140af522c64
---
 amdgpu/amdgpu.h          | 40 ++++++++++++++++++++++++++++++
 amdgpu/amdgpu_cs.c       | 63 ++++++++++++++++++++++++++++++++++++++++++++++--
 amdgpu/amdgpu_internal.h |  2 ++
 3 files changed, 103 insertions(+), 2 deletions(-)
amdgpu/amdgpu.h +40 -0
index 693d841..e716855 100644
@@ -1379,6 +1379,19 @@ int amdgpu_svm_commit(amdgpu_va_handle va_range_handle,
 int amdgpu_svm_uncommit(amdgpu_va_handle va_range_handle);
 
 /**
+ *  create shared semaphore
+ *
+ * \param amdgpu_device_handle
+ * \param   sem	   - \c [out] semaphore handle
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem);
+
+/**
  *  create semaphore
  *
  * \param   sem	   - \c [out] semaphore handle
@@ -1439,6 +1452,33 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
 int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem);
 
 /**
+ *  export semaphore
+ *
+ * \param   sem	    - \c [in] semaphore handle
+ * \param   shared_handle    - \c [out] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle);
+/**
+ *  import semaphore
+ *
+ * \param   sem	    - \c [out] semaphore handle
+ * \param   dev	    - \c [in] device handle
+ * \param   shared_handle    - \c [in] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev,
+			       uint32_t shared_handle);
+
+/**
  *  Get the ASIC marketing name
  *
  * \param   dev         - \c [in] Device handle. See #amdgpu_device_initialize()
amdgpu/amdgpu_cs.c +0 -0
index c76a675..a3ff34e 100644
@@ -518,6 +518,34 @@ int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences,
 	return r;

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 528

Quoted 10 lines
amdgpu/amdgpu_cs.c +8 -0
 }
 
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem)
+{
+	struct amdgpu_bo_alloc_request req = {0};
+	amdgpu_bo_handle buf_handle;
+	int r;
+
+	if (NULL == sem)

Since sem is ** then should we not check that *both* sem && *sem are non-NULL too? Further you can use the canonical form of (!sem)

Quoted 5 lines
amdgpu/amdgpu_cs.c +5 -0
+		return -EINVAL;
+
+	req.alloc_size = sizeof(struct amdgpu_semaphore);
+	req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
+

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 543

Quoted 10 lines
amdgpu/amdgpu_cs.c +10 -0
+	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
+	if (r)
+		return r;
+	r = amdgpu_bo_cpu_map(buf_handle, sem);
+	if (r) {
+		amdgpu_bo_free(buf_handle);
+		return r;
+	}
+	(*sem)->buf_handle = buf_handle;
+	atomic_set(&(*sem)->refcount, 1);

Hi Chunming,

When/where was 'amdgpu_semaphore_handle' introduced? I am not sure I like pointers being hidden behind typedef's as opaque types this can lead to really really bad things.. I only noticed sem was a ** because of the weird looking deference then address operator application, then deference again here, &(*sem)->..

Cheers,
Edward.

Quoted 68 lines
amdgpu/amdgpu_cs.c +38 -2
+	(*sem)->version = 2;
+
+	return 0;
+}
+
 int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 {
 	struct amdgpu_semaphore *gpu_semaphore;
@@ -529,6 +557,7 @@ int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 	if (NULL == gpu_semaphore)
 		return -ENOMEM;
 
+	gpu_semaphore->version = 1;
 	atomic_set(&gpu_semaphore->refcount, 1);
 	*sem = gpu_semaphore;
 
@@ -608,8 +637,15 @@ static int amdgpu_cs_unreference_sem(amdgpu_semaphore_handle sem)
 	if (NULL == sem)
 		return -EINVAL;
 
-	if (update_references(&sem->refcount, NULL))
-		free(sem);
+	if (update_references(&sem->refcount, NULL)) {
+		if (sem->version == 1)
+			free(sem);
+		else if (sem->version == 2) {
+			amdgpu_bo_handle buf_handle = sem->buf_handle;
+			amdgpu_bo_cpu_unmap(buf_handle);
+			amdgpu_bo_free(buf_handle);
+		}
+	}
 	return 0;
 }
 
@@ -618,4 +654,27 @@ int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem)
 	return amdgpu_cs_unreference_sem(sem);
 }
 
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle)
+{
+	return amdgpu_bo_export(sem->buf_handle,
+				amdgpu_bo_handle_type_dma_buf_fd,
+				shared_handle);
+
+}
+
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev, uint32_t shared_handle)
+{
+	struct amdgpu_bo_import_result output;
+	int r;
+
+	r = amdgpu_bo_import(dev,
+			     amdgpu_bo_handle_type_dma_buf_fd,
+			     shared_handle,
+			     &output);
+	if (r)
+		return r;
 
+	return amdgpu_bo_cpu_map(output.buf_handle, sem);
+}
amdgpu/amdgpu_internal.h +0 -0
index ccc85d7..7c422da 100644

amdgpu/amdgpu_internal.h struct amdgpu_semaphore { · line 141

Quoted 8 lines
amdgpu/amdgpu_internal.h +2 -0
@@ -134,6 +134,8 @@ struct amdgpu_semaphore {
 	atomic_t refcount;
 	struct list_head list;
 	struct drm_amdgpu_fence signal_fence;
+	amdgpu_bo_handle buf_handle;
+	uint32_t version;
 };
 
 /**

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): cf458d97-d40a-d363-b5c4-2ae45dde9179-dczkZgxz+BNUPWh3PAxdjQ@public.gmane.org

author
#7 lore raw

Re: [PATCH 2/4] amdgpu: add export/import semaphore apis

Quoted 106 lines · Edward O'Callaghan
Quoted 74 lines · Chunming Zhou

They are used for sharing semaphore across process.

Change-Id I262adf10913d365bb93368b492e69140af522c64
---
  amdgpu/amdgpu.h          | 40 ++++++++++++++++++++++++++++++
  amdgpu/amdgpu_cs.c       | 63 ++++++++++++++++++++++++++++++++++++++++++++++--
  amdgpu/amdgpu_internal.h |  2 ++
  3 files changed, 103 insertions(+), 2 deletions(-)
amdgpu/amdgpu.h +40 -0
index 693d841..e716855 100644
@@ -1379,6 +1379,19 @@ int amdgpu_svm_commit(amdgpu_va_handle va_range_handle,
  int amdgpu_svm_uncommit(amdgpu_va_handle va_range_handle);
 
  /**
+ *  create shared semaphore
+ *
+ * \param amdgpu_device_handle
+ * \param   sem	   - \c [out] semaphore handle
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem);
+
+/**
   *  create semaphore
   *
   * \param   sem	   - \c [out] semaphore handle
@@ -1439,6 +1452,33 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
  int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem);
 
  /**
+ *  export semaphore
+ *
+ * \param   sem	    - \c [in] semaphore handle
+ * \param   shared_handle    - \c [out] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle);
+/**
+ *  import semaphore
+ *
+ * \param   sem	    - \c [out] semaphore handle
+ * \param   dev	    - \c [in] device handle
+ * \param   shared_handle    - \c [in] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev,
+			       uint32_t shared_handle);
+
+/**
   *  Get the ASIC marketing name
   *
   * \param   dev         - \c [in] Device handle. See #amdgpu_device_initialize()
amdgpu/amdgpu_cs.c +0 -0
index c76a675..a3ff34e 100644
@@ -518,6 +518,34 @@ int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences,
  	return r;

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 528

Quoted 10 lines
amdgpu/amdgpu_cs.c +8 -0
  }
 
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem)
+{
+	struct amdgpu_bo_alloc_request req = {0};
+	amdgpu_bo_handle buf_handle;
+	int r;
+
+	if (NULL == sem)

Since sem is ** then should we not check that *both* sem && *sem are non-NULL too? Further you can use the canonical form of (!sem)

Quoted 5 lines
amdgpu/amdgpu_cs.c +5 -0
+		return -EINVAL;
+
+	req.alloc_size = sizeof(struct amdgpu_semaphore);
+	req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
+

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 543

Quoted 10 lines
amdgpu/amdgpu_cs.c +10 -0
+	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
+	if (r)
+		return r;
+	r = amdgpu_bo_cpu_map(buf_handle, sem);
+	if (r) {
+		amdgpu_bo_free(buf_handle);
+		return r;
+	}
+	(*sem)->buf_handle = buf_handle;
+	atomic_set(&(*sem)->refcount, 1);

Hi Chunming,

Quoted 5 lines

When/where was 'amdgpu_semaphore_handle' introduced? I am not sure I like pointers being hidden behind typedef's as opaque types this can lead to really really bad things.. I only noticed sem was a ** because of the weird looking deference then address operator application, then deference again here, &(*sem)->..

Hi Edward,
semaphore was introduced last year, which is the wrapper of existing dependency. Yeah, this whole sharing semaphore approach has been NAKed by Christian. So we can skip this series now, we are going to use sync_file instead.

Thanks,
David Zhou

Quoted 78 lines

Cheers,
Edward.

Quoted 74 lines
amdgpu/amdgpu_cs.c +38 -2
+	(*sem)->version = 2;
+
+	return 0;
+}
+
  int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
  {
  	struct amdgpu_semaphore *gpu_semaphore;
@@ -529,6 +557,7 @@ int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
  	if (NULL == gpu_semaphore)
  		return -ENOMEM;
 
+	gpu_semaphore->version = 1;
  	atomic_set(&gpu_semaphore->refcount, 1);
  	*sem = gpu_semaphore;
 
@@ -608,8 +637,15 @@ static int amdgpu_cs_unreference_sem(amdgpu_semaphore_handle sem)
  	if (NULL == sem)
  		return -EINVAL;
 
-	if (update_references(&sem->refcount, NULL))
-		free(sem);
+	if (update_references(&sem->refcount, NULL)) {
+		if (sem->version == 1)
+			free(sem);
+		else if (sem->version == 2) {
+			amdgpu_bo_handle buf_handle = sem->buf_handle;
+			amdgpu_bo_cpu_unmap(buf_handle);
+			amdgpu_bo_free(buf_handle);
+		}
+	}
  	return 0;
  }
 
@@ -618,4 +654,27 @@ int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem)
  	return amdgpu_cs_unreference_sem(sem);
  }
 
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle)
+{
+	return amdgpu_bo_export(sem->buf_handle,
+				amdgpu_bo_handle_type_dma_buf_fd,
+				shared_handle);
+
+}
+
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev, uint32_t shared_handle)
+{
+	struct amdgpu_bo_import_result output;
+	int r;
+
+	r = amdgpu_bo_import(dev,
+			     amdgpu_bo_handle_type_dma_buf_fd,
+			     shared_handle,
+			     &output);
+	if (r)
+		return r;
 
+	return amdgpu_bo_cpu_map(output.buf_handle, sem);
+}
amdgpu/amdgpu_internal.h +2 -0
index ccc85d7..7c422da 100644
@@ -134,6 +134,8 @@ struct amdgpu_semaphore {
  	atomic_t refcount;
  	struct list_head list;
  	struct drm_amdgpu_fence signal_fence;
+	amdgpu_bo_handle buf_handle;
+	uint32_t version;
  };
 
Quoted 8 lines

amdgpu/amdgpu_internal.h struct amdgpu_semaphore { · line 141

Quoted 2 lines
amdgpu/amdgpu_internal.h +0 -0
  /**

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): [email protected]

#9 lore raw

Re: [PATCH 2/4] amdgpu: add export/import semaphore apis

Quoted 109 lines · On 08/22/2016 12:42 PM, zhoucm1 wrote:
Quoted 106 lines · Edward O'Callaghan
Quoted 74 lines · Chunming Zhou

They are used for sharing semaphore across process.

Change-Id I262adf10913d365bb93368b492e69140af522c64
---
 amdgpu/amdgpu.h          | 40 ++++++++++++++++++++++++++++++
 amdgpu/amdgpu_cs.c       | 63 ++++++++++++++++++++++++++++++++++++++++++++++--
 amdgpu/amdgpu_internal.h |  2 ++
 3 files changed, 103 insertions(+), 2 deletions(-)
amdgpu/amdgpu.h +40 -0
index 693d841..e716855 100644
@@ -1379,6 +1379,19 @@ int amdgpu_svm_commit(amdgpu_va_handle va_range_handle,
 int amdgpu_svm_uncommit(amdgpu_va_handle va_range_handle);
 
 /**
+ *  create shared semaphore
+ *
+ * \param amdgpu_device_handle
+ * \param   sem	   - \c [out] semaphore handle
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem);
+
+/**
  *  create semaphore
  *
  * \param   sem	   - \c [out] semaphore handle
@@ -1439,6 +1452,33 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
 int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem);
 
 /**
+ *  export semaphore
+ *
+ * \param   sem	    - \c [in] semaphore handle
+ * \param   shared_handle    - \c [out] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle);
+/**
+ *  import semaphore
+ *
+ * \param   sem	    - \c [out] semaphore handle
+ * \param   dev	    - \c [in] device handle
+ * \param   shared_handle    - \c [in] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev,
+			       uint32_t shared_handle);
+
+/**
  *  Get the ASIC marketing name
  *
  * \param   dev         - \c [in] Device handle. See #amdgpu_device_initialize()
amdgpu/amdgpu_cs.c +0 -0
index c76a675..a3ff34e 100644
@@ -518,6 +518,34 @@ int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences,
 	return r;

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 528

Quoted 10 lines
amdgpu/amdgpu_cs.c +8 -0
 }
 
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem)
+{
+	struct amdgpu_bo_alloc_request req = {0};
+	amdgpu_bo_handle buf_handle;
+	int r;
+
+	if (NULL == sem)

Since sem is ** then should we not check that *both* sem && *sem are non-NULL too? Further you can use the canonical form of (!sem)

Quoted 5 lines
amdgpu/amdgpu_cs.c +5 -0
+		return -EINVAL;
+
+	req.alloc_size = sizeof(struct amdgpu_semaphore);
+	req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
+

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 543

Quoted 10 lines
amdgpu/amdgpu_cs.c +10 -0
+	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
+	if (r)
+		return r;
+	r = amdgpu_bo_cpu_map(buf_handle, sem);
+	if (r) {
+		amdgpu_bo_free(buf_handle);
+		return r;
+	}
+	(*sem)->buf_handle = buf_handle;
+	atomic_set(&(*sem)->refcount, 1);

Hi Chunming,

Quoted 8 lines
Quoted 5 lines

When/where was 'amdgpu_semaphore_handle' introduced? I am not sure I like pointers being hidden behind typedef's as opaque types this can lead to really really bad things.. I only noticed sem was a ** because of the weird looking deference then address operator application, then deference again here, &(*sem)->..

Hi Edward,
semaphore was introduced last year, which is the wrapper of existing dependency.

Well no, I mean 'amdgpu_semaphore_handle' in particular as I didn't see that in mainline. Where was it introduced?

Quoted 2 lines

Yeah, this whole sharing semaphore approach has been NAKed by Christian. So we can skip this series now, we are going to use sync_file instead.

No worries.

Kind Regards,
Edward.

Quoted 93 lines

Thanks,
David Zhou

Quoted 78 lines

Cheers,
Edward.

Quoted 74 lines
amdgpu/amdgpu_cs.c +38 -2
+	(*sem)->version = 2;
+
+	return 0;
+}
+
 int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 {
 	struct amdgpu_semaphore *gpu_semaphore;
@@ -529,6 +557,7 @@ int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 	if (NULL == gpu_semaphore)
 		return -ENOMEM;
 
+	gpu_semaphore->version = 1;
 	atomic_set(&gpu_semaphore->refcount, 1);
 	*sem = gpu_semaphore;
 
@@ -608,8 +637,15 @@ static int amdgpu_cs_unreference_sem(amdgpu_semaphore_handle sem)
 	if (NULL == sem)
 		return -EINVAL;
 
-	if (update_references(&sem->refcount, NULL))
-		free(sem);
+	if (update_references(&sem->refcount, NULL)) {
+		if (sem->version == 1)
+			free(sem);
+		else if (sem->version == 2) {
+			amdgpu_bo_handle buf_handle = sem->buf_handle;
+			amdgpu_bo_cpu_unmap(buf_handle);
+			amdgpu_bo_free(buf_handle);
+		}
+	}
 	return 0;
 }
 
@@ -618,4 +654,27 @@ int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem)
 	return amdgpu_cs_unreference_sem(sem);
 }
 
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle)
+{
+	return amdgpu_bo_export(sem->buf_handle,
+				amdgpu_bo_handle_type_dma_buf_fd,
+				shared_handle);
+
+}
+
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev, uint32_t shared_handle)
+{
+	struct amdgpu_bo_import_result output;
+	int r;
+
+	r = amdgpu_bo_import(dev,
+			     amdgpu_bo_handle_type_dma_buf_fd,
+			     shared_handle,
+			     &output);
+	if (r)
+		return r;
 
+	return amdgpu_bo_cpu_map(output.buf_handle, sem);
+}
amdgpu/amdgpu_internal.h +2 -0
index ccc85d7..7c422da 100644
@@ -134,6 +134,8 @@ struct amdgpu_semaphore {
 	atomic_t refcount;
 	struct list_head list;
 	struct drm_amdgpu_fence signal_fence;
+	amdgpu_bo_handle buf_handle;
+	uint32_t version;
 };
 
Quoted 8 lines

amdgpu/amdgpu_internal.h struct amdgpu_semaphore { · line 141

Quoted 2 lines
amdgpu/amdgpu_internal.h +0 -0
 /**

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Quoted 5 lines

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 414bf811-0e72-6f79-350f-8995a2b5167c-dczkZgxz+BNUPWh3PAxdjQ@public.gmane.org

#11 lore raw

Re: [PATCH 2/4] amdgpu: add export/import semaphore apis

Quoted 117 lines · Edward O'Callaghan
Quoted 114 lines · On 08/22/2016 12:42 PM, zhoucm1 wrote:
Quoted 111 lines · Edward O'Callaghan
Quoted 74 lines · Chunming Zhou

They are used for sharing semaphore across process.

Change-Id I262adf10913d365bb93368b492e69140af522c64
---
 amdgpu/amdgpu.h          | 40 ++++++++++++++++++++++++++++++
 amdgpu/amdgpu_cs.c       | 63 ++++++++++++++++++++++++++++++++++++++++++++++--
 amdgpu/amdgpu_internal.h |  2 ++
 3 files changed, 103 insertions(+), 2 deletions(-)
amdgpu/amdgpu.h +40 -0
index 693d841..e716855 100644
@@ -1379,6 +1379,19 @@ int amdgpu_svm_commit(amdgpu_va_handle va_range_handle,
 int amdgpu_svm_uncommit(amdgpu_va_handle va_range_handle);
 
 /**
+ *  create shared semaphore
+ *
+ * \param amdgpu_device_handle
+ * \param   sem	   - \c [out] semaphore handle
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem);
+
+/**
  *  create semaphore
  *
  * \param   sem	   - \c [out] semaphore handle
@@ -1439,6 +1452,33 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
 int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem);
 
 /**
+ *  export semaphore
+ *
+ * \param   sem	    - \c [in] semaphore handle
+ * \param   shared_handle    - \c [out] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle);
+/**
+ *  import semaphore
+ *
+ * \param   sem	    - \c [out] semaphore handle
+ * \param   dev	    - \c [in] device handle
+ * \param   shared_handle    - \c [in] handle across process
+ *
+ * \return   0 on success\n
+ *          <0 - Negative POSIX Error code
+ *
+*/
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev,
+			       uint32_t shared_handle);
+
+/**
  *  Get the ASIC marketing name
  *
  * \param   dev         - \c [in] Device handle. See #amdgpu_device_initialize()
amdgpu/amdgpu_cs.c +0 -0
index c76a675..a3ff34e 100644
@@ -518,6 +518,34 @@ int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences,
 	return r;

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 528

Quoted 10 lines
amdgpu/amdgpu_cs.c +8 -0
 }
 
+int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
+				      amdgpu_semaphore_handle *sem)
+{
+	struct amdgpu_bo_alloc_request req = {0};
+	amdgpu_bo_handle buf_handle;
+	int r;
+
+	if (NULL == sem)

Since sem is ** then should we not check that *both* sem && *sem are non-NULL too? Further you can use the canonical form of (!sem)

Quoted 5 lines
amdgpu/amdgpu_cs.c +5 -0
+		return -EINVAL;
+
+	req.alloc_size = sizeof(struct amdgpu_semaphore);
+	req.preferred_heap = AMDGPU_GEM_DOMAIN_GTT;
+

amdgpu/amdgpu_cs.c int amdgpu_cs_wait_fences(struct amdgpu_cs_fence *fences, · line 543

Quoted 10 lines
amdgpu/amdgpu_cs.c +10 -0
+	r = amdgpu_bo_alloc(device_handle, &req, &buf_handle);
+	if (r)
+		return r;
+	r = amdgpu_bo_cpu_map(buf_handle, sem);
+	if (r) {
+		amdgpu_bo_free(buf_handle);
+		return r;
+	}
+	(*sem)->buf_handle = buf_handle;
+	atomic_set(&(*sem)->refcount, 1);

Hi Chunming,

When/where was 'amdgpu_semaphore_handle' introduced? I am not sure I like pointers being hidden behind typedef's as opaque types this can lead to really really bad things.. I only noticed sem was a ** because of the weird looking deference then address operator application, then deference again here, &(*sem)->..

Quoted 6 lines
Quoted 3 lines

Hi Edward,
semaphore was introduced last year, which is the wrapper of existing dependency.

Well no, I mean 'amdgpu_semaphore_handle' in particular as I didn't see that in mainline. Where was it introduced?

woops, ignore me. 11pm should be in bed :p .... sorry for the noise.

Quoted 110 lines
Quoted 3 lines

Yeah, this whole sharing semaphore approach has been NAKed by Christian. So we can skip this series now, we are going to use sync_file instead.

No worries.

Kind Regards,
Edward.

Quoted 93 lines

Thanks,
David Zhou

Quoted 78 lines

Cheers,
Edward.

Quoted 74 lines
amdgpu/amdgpu_cs.c +38 -2
+	(*sem)->version = 2;
+
+	return 0;
+}
+
 int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 {
 	struct amdgpu_semaphore *gpu_semaphore;
@@ -529,6 +557,7 @@ int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 	if (NULL == gpu_semaphore)
 		return -ENOMEM;
 
+	gpu_semaphore->version = 1;
 	atomic_set(&gpu_semaphore->refcount, 1);
 	*sem = gpu_semaphore;
 
@@ -608,8 +637,15 @@ static int amdgpu_cs_unreference_sem(amdgpu_semaphore_handle sem)
 	if (NULL == sem)
 		return -EINVAL;
 
-	if (update_references(&sem->refcount, NULL))
-		free(sem);
+	if (update_references(&sem->refcount, NULL)) {
+		if (sem->version == 1)
+			free(sem);
+		else if (sem->version == 2) {
+			amdgpu_bo_handle buf_handle = sem->buf_handle;
+			amdgpu_bo_cpu_unmap(buf_handle);
+			amdgpu_bo_free(buf_handle);
+		}
+	}
 	return 0;
 }
 
@@ -618,4 +654,27 @@ int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem)
 	return amdgpu_cs_unreference_sem(sem);
 }
 
+int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
+			       uint32_t *shared_handle)
+{
+	return amdgpu_bo_export(sem->buf_handle,
+				amdgpu_bo_handle_type_dma_buf_fd,
+				shared_handle);
+
+}
+
+int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
+			       amdgpu_device_handle dev, uint32_t shared_handle)
+{
+	struct amdgpu_bo_import_result output;
+	int r;
+
+	r = amdgpu_bo_import(dev,
+			     amdgpu_bo_handle_type_dma_buf_fd,
+			     shared_handle,
+			     &output);
+	if (r)
+		return r;
 
+	return amdgpu_bo_cpu_map(output.buf_handle, sem);
+}
amdgpu/amdgpu_internal.h +2 -0
index ccc85d7..7c422da 100644
@@ -134,6 +134,8 @@ struct amdgpu_semaphore {
 	atomic_t refcount;
 	struct list_head list;
 	struct drm_amdgpu_fence signal_fence;
+	amdgpu_bo_handle buf_handle;
+	uint32_t version;
 };
 
Quoted 8 lines

amdgpu/amdgpu_internal.h struct amdgpu_semaphore { · line 141

Quoted 2 lines
amdgpu/amdgpu_internal.h +0 -0
 /**

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Quoted 5 lines

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

Quoted 5 lines

_______________________________________________
amd-gfx mailing list [email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

#12 lore raw

PATCH 3/4 amdgpu: add mutex for across process reason

In reply to a message not in the archive
Change-Id I69b5c8d86f9e1ed32bb20e899b74ad4146c0e988
---
 amdgpu/amdgpu_cs.c       | 36 +++++++++++++++++++++++++++++-------
 amdgpu/amdgpu_internal.h |  2 ++
 2 files changed, 31 insertions(+), 7 deletions(-)
amdgpu/amdgpu_cs.c +29 -7
index a3ff34e..ecbd4d7 100644
@@ -541,6 +541,12 @@ int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
 	}
 	(*sem)->buf_handle = buf_handle;
 	atomic_set(&(*sem)->refcount, 1);
+	r = sem_init(&(*sem)->mutex, 1, 1);
+	if (r) {
+		amdgpu_bo_cpu_unmap(buf_handle);
+		amdgpu_bo_free(buf_handle);
+		return r;
+	}
 	(*sem)->version = 2;
 
 	return 0;
@@ -549,6 +555,7 @@ int amdgpu_cs_create_semaphore_object(amdgpu_device_handle device_handle,
 int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 {
 	struct amdgpu_semaphore *gpu_semaphore;
+	int r;
 
 	if (NULL == sem)
 		return -EINVAL;
@@ -559,6 +566,11 @@ int amdgpu_cs_create_semaphore(amdgpu_semaphore_handle *sem)
 
 	gpu_semaphore->version = 1;
 	atomic_set(&gpu_semaphore->refcount, 1);
+	r = sem_init(&gpu_semaphore->mutex, 1, 1);
+	if (r) {
+		free(gpu_semaphore);
+		return r;
+	}
 	*sem = gpu_semaphore;
 
 	return 0;
@@ -581,6 +593,7 @@ int amdgpu_cs_signal_semaphore(amdgpu_context_handle ctx,
 	/* sem has been signaled */
 	if (sem->signal_fence.ctx_id)
 		return -EINVAL;
+	sem_wait(&sem->mutex);
 	pthread_mutex_lock(&ctx->sequence_mutex);
 	sem->signal_fence.ctx_id = ctx->id;
 	sem->signal_fence.ip_type = ip_type;
@@ -589,6 +602,7 @@ int amdgpu_cs_signal_semaphore(amdgpu_context_handle ctx,
 	sem->signal_fence.seq_no = ctx->last_seq[ip_type][ip_instance][ring];
 	update_references(NULL, &sem->refcount);
 	pthread_mutex_unlock(&ctx->sequence_mutex);
+	sem_post(&sem->mutex);
 	return 0;
 }
 
@@ -609,10 +623,11 @@ int amdgpu_cs_wait_semaphore(amdgpu_context_handle ctx,
 	/* must signal first */
 	if (0 == sem->signal_fence.ctx_id)
 		return -EINVAL;
-
+	sem_wait(&sem->mutex);
 	pthread_mutex_lock(&ctx->sequence_mutex);
 	list_add(&sem->list, &ctx->sem_list[ip_type][ip_instance][ring]);
 	pthread_mutex_unlock(&ctx->sequence_mutex);
+	sem_post(&sem->mutex);
 	return 0;
 }
 
@@ -622,13 +637,13 @@ static int amdgpu_cs_reset_sem(amdgpu_semaphore_handle sem)
 		return -EINVAL;
 	if (0 == sem->signal_fence.ctx_id)
 		return -EINVAL;
-
+	sem_wait(&sem->mutex);
 	sem->signal_fence.ctx_id = 0;
 	sem->signal_fence.ip_type = 0;
 	sem->signal_fence.ip_instance = 0;
 	sem->signal_fence.ring = 0;
 	sem->signal_fence.seq_no = 0;
-
+	sem_post(&sem->mutex);
 	return 0;
 }
 
@@ -636,8 +651,9 @@ static int amdgpu_cs_unreference_sem(amdgpu_semaphore_handle sem)
 {
 	if (NULL == sem)
 		return -EINVAL;
-
+	sem_wait(&sem->mutex);
 	if (update_references(&sem->refcount, NULL)) {
+		sem_post(&sem->mutex);
 		if (sem->version == 1)
 			free(sem);
 		else if (sem->version == 2) {
@@ -645,7 +661,9 @@ static int amdgpu_cs_unreference_sem(amdgpu_semaphore_handle sem)
 			amdgpu_bo_cpu_unmap(buf_handle);
 			amdgpu_bo_free(buf_handle);
 		}
+		return 0;
 	}
+	sem_post(&sem->mutex);
 	return 0;
 }
 
@@ -657,10 +675,14 @@ int amdgpu_cs_destroy_semaphore(amdgpu_semaphore_handle sem)
 int amdgpu_cs_export_semaphore(amdgpu_semaphore_handle sem,
 			       uint32_t *shared_handle)
 {
-	return amdgpu_bo_export(sem->buf_handle,
-				amdgpu_bo_handle_type_dma_buf_fd,
-				shared_handle);
+	int r;
 
+	sem_wait(&sem->mutex);
+	r = amdgpu_bo_export(sem->buf_handle,
+			     amdgpu_bo_handle_type_dma_buf_fd,
+			     shared_handle);
+	sem_post(&sem->mutex);
+	return r;
 }
 
 int amdgpu_cs_import_semaphore(amdgpu_semaphore_handle *sem,
amdgpu/amdgpu_internal.h +2 -0
index 7c422da..f035786 100644
@@ -31,6 +31,7 @@
 
 #include <assert.h>
 #include <pthread.h>
+#include <semaphore.h>
 
 #include "libdrm_macros.h"
 #include "xf86atomic.h"
@@ -135,6 +136,7 @@ struct amdgpu_semaphore {
 	struct list_head list;
 	struct drm_amdgpu_fence signal_fence;
 	amdgpu_bo_handle buf_handle;
+	sem_t mutex;
 	uint32_t version;
 };
 
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
#13 lore raw

PATCH 4/4 tests/amdgpu: add semaphore across process test

In reply to a message not in the archive
Change-Id I6e8c8cfa1a05f51f3c03670baea68ed6da94fa11
---
 tests/amdgpu/basic_tests.c | 131 +++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 131 insertions(+)
tests/amdgpu/basic_tests.c +131 -0
index 02e863a..c7da54d 100644
@@ -47,6 +47,7 @@ static void amdgpu_command_submission_sdma(void);
 static void amdgpu_command_submission_multi_fence(void);
 static void amdgpu_userptr_test(void);
 static void amdgpu_semaphore_test(void);
+static void amdgpu_semaphore_across_process_test(void);
 static void amdgpu_svm_test(void);
 static void amdgpu_multi_svm_test(void);
 static void amdgpu_va_range_test(void);
@@ -60,6 +61,7 @@ CU_TestInfo basic_tests[] = {
 	{ "Command submission Test (SDMA)", amdgpu_command_submission_sdma },
 	{ "Command submission Test (Multi-fence)", amdgpu_command_submission_multi_fence },
 	{ "SW semaphore Test",  amdgpu_semaphore_test },
+	{ "SW semaphore across process Test",  amdgpu_semaphore_across_process_test },
 	{ "VA range Test", amdgpu_va_range_test},
 	{ "SVM Test", amdgpu_svm_test },
 	{ "SVM Test (multi-GPUs)", amdgpu_multi_svm_test },
@@ -515,6 +517,135 @@ static void amdgpu_command_submission_gfx(void)
 	amdgpu_command_submission_gfx_shared_ib();
 }
 
+static void amdgpu_semaphore_across_process_test(void)
+{
+	struct amdgpu_context  *context_handle;
+	amdgpu_semaphore_handle sem;
+	amdgpu_bo_handle ib_result_handle[2];
+	void *ib_result_cpu[2];
+	uint64_t ib_result_mc_address[2];
+	struct amdgpu_cs_request ibs_request[2] = {0};
+	struct amdgpu_cs_ib_info ib_info[2] = {0};
+	struct amdgpu_cs_fence fence_status = {0};
+	uint32_t *ptr;
+	uint32_t expired;
+	uint32_t shared_handle;
+	amdgpu_bo_list_handle bo_list[2];
+	amdgpu_va_handle va_handle[2];
+	int r, i, pid;
+
+	r = amdgpu_cs_create_semaphore_object(device_handle, &sem);
+	CU_ASSERT_EQUAL(r, 0);
+	r = amdgpu_cs_ctx_create(device_handle, &context_handle);
+	CU_ASSERT_EQUAL(r, 0);
+
+	r = amdgpu_bo_alloc_and_map(device_handle, 4096, 4096,
+				    AMDGPU_GEM_DOMAIN_GTT, 0,
+				    &ib_result_handle[0], &ib_result_cpu[0],
+				    &ib_result_mc_address[0], &va_handle[0]);
+	CU_ASSERT_EQUAL(r, 0);
+
+	r = amdgpu_get_bo_list(device_handle, ib_result_handle[0],
+			       NULL, &bo_list[0]);
+	CU_ASSERT_EQUAL(r, 0);
+
+	ptr = ib_result_cpu[0];
+	ptr[0] = SDMA_NOP;
+	ib_info[0].ib_mc_address = ib_result_mc_address[0];
+	ib_info[0].size = 1;
+
+	ibs_request[0].ip_type = AMDGPU_HW_IP_DMA;
+	ibs_request[0].number_of_ibs = 1;
+	ibs_request[0].ibs = &ib_info[0];
+	ibs_request[0].resources = bo_list[0];
+	ibs_request[0].fence_info.handle = NULL;
+	r = amdgpu_cs_submit(context_handle, 0,&ibs_request[0], 1);
+	CU_ASSERT_EQUAL(r, 0);
+	r = amdgpu_cs_signal_semaphore(context_handle, AMDGPU_HW_IP_DMA, 0, 0, sem);
+	CU_ASSERT_EQUAL(r, 0);
+	r = amdgpu_cs_export_semaphore(sem, &shared_handle);
+	CU_ASSERT_EQUAL(r, 0);
+
+	pid = fork();
+	/* child process */
+	if (pid == 0) {
+		amdgpu_device_handle child_device_handle;
+		uint32_t  child_major_version;
+		uint32_t  child_minor_version;
+		amdgpu_semaphore_handle child_sem;
+		amdgpu_context_handle context_handle1;
+
+		r = amdgpu_device_initialize(drm_amdgpu[0], &child_major_version,
+					     &child_minor_version, &child_device_handle);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_bo_alloc_and_map(child_device_handle, 4096, 4096,
+					    AMDGPU_GEM_DOMAIN_GTT, 0,
+					    &ib_result_handle[1], &ib_result_cpu[1],
+					    &ib_result_mc_address[1], &va_handle[1]);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_get_bo_list(child_device_handle, ib_result_handle[1],
+				       NULL, &bo_list[1]);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_cs_import_semaphore(&child_sem, child_device_handle, shared_handle);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_cs_ctx_create(child_device_handle, &context_handle1);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_cs_wait_semaphore(context_handle1, AMDGPU_HW_IP_GFX, 0, 0, child_sem);
+		CU_ASSERT_EQUAL(r, 0);
+		ptr = ib_result_cpu[1];
+		ptr[0] = GFX_COMPUTE_NOP;
+		ib_info[1].ib_mc_address = ib_result_mc_address[1];
+		ib_info[1].size = 1;
+
+		ibs_request[1].ip_type = AMDGPU_HW_IP_GFX;
+		ibs_request[1].number_of_ibs = 1;
+		ibs_request[1].ibs = &ib_info[1];
+		ibs_request[1].resources = bo_list[1];
+		ibs_request[1].fence_info.handle = NULL;
+
+		r = amdgpu_cs_submit(context_handle1, 0,&ibs_request[1], 1);
+		CU_ASSERT_EQUAL(r, 0);
+
+		fence_status.context = context_handle1;
+		fence_status.ip_type = AMDGPU_HW_IP_GFX;
+		fence_status.fence = ibs_request[1].seq_no;
+		r = amdgpu_cs_query_fence_status(&fence_status,
+						 500000000, 0, &expired);
+		CU_ASSERT_EQUAL(r, 0);
+		CU_ASSERT_EQUAL(expired, true);
+
+		r = amdgpu_bo_unmap_and_free(ib_result_handle[1], va_handle[1],
+					     ib_result_mc_address[1], 4096);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_bo_list_destroy(bo_list[1]);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_cs_ctx_free(context_handle1);
+		CU_ASSERT_EQUAL(r, 0);
+
+		amdgpu_device_deinitialize(child_device_handle);
+	} else {
+		r = amdgpu_bo_unmap_and_free(ib_result_handle[0], va_handle[0],
+					     ib_result_mc_address[0], 4096);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_bo_list_destroy(bo_list[0]);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_cs_ctx_free(context_handle);
+		CU_ASSERT_EQUAL(r, 0);
+
+		r = amdgpu_cs_destroy_semaphore(sem);
+		CU_ASSERT_EQUAL(r, 0);
+	}
+}
+
 static void amdgpu_semaphore_test(void)
 {
 	amdgpu_context_handle context_handle[2];
Signature · 6 lines
1.9.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx