drm/amdgpu: add mes gang contex alloc/free helper

Implement the MES gang contex alloc and free heplers.

Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Michael Chen <michael.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c

@@ -1080,6 +1080,47 @@ void amdgpu_mes_free_proc_ctx_index(struct amdgpu_mes *mes,
 	amdgpu_mes_unlock(mes);
 }
 
+/**
+ * amdgpu_mes_alloc_gang_ctx_index - allocate a gang context slot
+ */
+int amdgpu_mes_alloc_gang_ctx_index(struct amdgpu_mes *mes,
+				    struct amdgpu_usermode_queue *queue)
+{
+	unsigned long bit;
+
+	if (!mes->use_rs64mem || !mes->gang_ctx_bitmap)
+		return -EOPNOTSUPP;
+
+	amdgpu_mes_lock(mes);
+	bit = find_first_zero_bit(mes->gang_ctx_bitmap,
+				  mes->gang_ctx_array_size);
+	if (bit >= mes->gang_ctx_array_size) {
+		amdgpu_mes_unlock(mes);
+		return -ENOSPC;
+	}
+	set_bit(bit, mes->gang_ctx_bitmap);
+	queue->gang_ctx_array_index = bit;
+	amdgpu_mes_unlock(mes);
+
+	return 0;
+}
+
+/**
+ * amdgpu_mes_free_gang_ctx_index - free a gang context slot
+ */
+void amdgpu_mes_free_gang_ctx_index(struct amdgpu_mes *mes,
+				    struct amdgpu_usermode_queue *queue)
+{
+	if (!mes->use_rs64mem || !mes->gang_ctx_bitmap)
+		return;
+	if (queue->gang_ctx_array_index >= mes->gang_ctx_array_size)
+		return;
+
+	amdgpu_mes_lock(mes);
+	clear_bit(queue->gang_ctx_array_index, mes->gang_ctx_bitmap);
+	amdgpu_mes_unlock(mes);
+}
+
 #if defined(CONFIG_DEBUG_FS)
 
 static int amdgpu_debugfs_mes_event_log_show(struct seq_file *m, void *unused)

drivers/gpu/drm/amd/amdgpu/amdgpu_mes.h

@@ -181,6 +181,7 @@ struct amdgpu_mes {
 	uint32_t			proc_ctx_array_size;
 	unsigned long			*proc_ctx_bitmap;
 	uint32_t			gang_ctx_array_size;
+	uint32_t			gang_ctx_array_index;
 	unsigned long			*gang_ctx_bitmap;
 };
 
@@ -280,6 +281,7 @@ struct mes_add_queue_input {
 	uint32_t	sh_mem_config_data;
 	uint32_t	vm_cntx_cntl;
 	uint32_t	process_context_array_index;
+	uint32_t	gang_context_array_index;
 };
 
 struct mes_remove_queue_input {
@@ -288,6 +290,7 @@ struct mes_remove_queue_input {
 	uint64_t	gang_context_addr;
 	uint32_t	queue_type;
 	bool		remove_queue_after_reset;
+	uint32_t	gang_context_array_index;
 };
 
 struct mes_map_legacy_queue_input {
@@ -636,4 +639,8 @@ int amdgpu_mes_alloc_proc_ctx_index(struct amdgpu_mes *mes,
 				    struct amdgpu_usermode_queue *queue);
 void amdgpu_mes_free_proc_ctx_index(struct amdgpu_mes *mes,
 				    struct amdgpu_usermode_queue *queue);
+int amdgpu_mes_alloc_gang_ctx_index(struct amdgpu_mes *mes,
+				    struct amdgpu_usermode_queue *queue);
+void amdgpu_mes_free_gang_ctx_index(struct amdgpu_mes *mes,
+				    struct amdgpu_usermode_queue *queue);
 #endif /* __AMDGPU_MES_H__ */

drivers/gpu/drm/amd/amdgpu/amdgpu_userq.h

@@ -102,6 +102,7 @@ struct amdgpu_usermode_queue {
 	} userq_vas;
 
 	uint32_t	proc_ctx_array_index;
+	uint32_t	gang_ctx_array_index;
 };
 
 struct amdgpu_userq_funcs {

drivers/gpu/drm/amd/amdgpu/mes_userqueue.c

@@ -150,6 +150,8 @@ static int mes_userq_map(struct amdgpu_usermode_queue *queue)
 	if (mes->use_rs64mem) {
 		amdgpu_mes_alloc_proc_ctx_index(mes, queue);
 		queue_input.process_context_array_index = queue->proc_ctx_array_index;
+		amdgpu_mes_alloc_gang_ctx_index(mes, queue);
+		queue_input.gang_context_array_index = queue->gang_ctx_array_index;
 	}
 	amdgpu_mes_lock(&adev->mes);
 	r = adev->mes.funcs->add_hw_queue(&adev->mes, &queue_input);
@@ -173,6 +175,7 @@ static int mes_userq_unmap(struct amdgpu_usermode_queue *queue)
 	int r;
 
 	memset(&queue_input, 0x0, sizeof(struct mes_remove_queue_input));
+	queue_input.gang_context_array_index = queue->gang_ctx_array_index;
 	queue_input.doorbell_offset = queue->doorbell_index;
 	queue_input.gang_context_addr = ctx->gpu_addr;
 	queue_input.queue_type = queue->queue_type;
@@ -180,8 +183,10 @@ static int mes_userq_unmap(struct amdgpu_usermode_queue *queue)
 	amdgpu_mes_lock(&adev->mes);
 	r = adev->mes.funcs->remove_hw_queue(&adev->mes, &queue_input);
 	amdgpu_mes_unlock(&adev->mes);
-	if (mes->use_rs64mem)
+	if (mes->use_rs64mem) {
 		amdgpu_mes_free_proc_ctx_index(mes, queue);
+		amdgpu_mes_free_gang_ctx_index(mes, queue);
+	}
 	if (r)
 		DRM_ERROR("Failed to unmap queue in HW, err (%d)\n", r);
 	return r;

drivers/gpu/drm/amd/amdgpu/mes_v11_0.c

@@ -338,6 +338,7 @@ static int mes_v11_0_add_hw_queue(struct amdgpu_mes *mes,
 	mes_add_queue_pkt.process_context_array_index = input->process_context_array_index;
 	mes_add_queue_pkt.gang_quantum = input->gang_quantum;
 	mes_add_queue_pkt.gang_context_addr = input->gang_context_addr;
+	mes_add_queue_pkt.gang_context_array_index = input->gang_context_array_index;
 	mes_add_queue_pkt.inprocess_gang_priority =
 		convert_to_mes_priority_level(input->inprocess_gang_priority);
 	mes_add_queue_pkt.gang_global_priority_level =
@@ -388,6 +389,7 @@ static int mes_v11_0_remove_hw_queue(struct amdgpu_mes *mes,
 
 	mes_remove_queue_pkt.doorbell_offset = input->doorbell_offset;
 	mes_remove_queue_pkt.gang_context_addr = input->gang_context_addr;
+	mes_remove_queue_pkt.gang_context_array_index = input->gang_context_array_index;
 	mes_remove_queue_pkt.queue_type =
 		convert_to_mes_queue_type(input->queue_type);
 

drivers/gpu/drm/amd/include/mes_v11_api_def.h

@@ -360,6 +360,7 @@ union MESAPI__REMOVE_QUEUE {
 
 		enum MES_QUEUE_TYPE         queue_type;
 		uint64_t                    timestamp;
+		uint32_t                    gang_context_array_index;
 	};
 
 	uint32_t	max_dwords_in_api[API_FRAME_SIZE_IN_DWORDS];