Always allocate space for control stack at beginning of save area

Hardware block testing is done with the workgroup state offset
initialized to the control stack size on all ASICs. MEC microcode
assumes this space is available when the workgroup state offset is
reset after a context restore event.

Fixes context save area overrun when the full save area is used.

Change-Id: I8eeb62f97140c6fe409fe78b4497d833584feea8
Signed-off-by: Jay Cornwall <Jay.Cornwall@amd.com>

src/queues.c

@@ -122,7 +122,6 @@ struct queue {
 	void *ctx_save_restore;
 	uint32_t ctx_save_restore_size;
 	uint32_t ctl_stack_size;
-	void *ctl_stack_copy;
 	const struct device_info *dev_info;
 	/* This queue structure is allocated from GPU with page aligned size
 	 * but only small bytes are used. We use the extra space in the end for
@@ -337,11 +336,7 @@ static bool update_ctx_save_restore_size(uint32_t nodeid, struct queue *q)
 		ctl_stack_size = cu_num * WAVES_PER_CU_VI * 8 + 8;
 		wg_data_size = cu_num * WG_CONTEXT_DATA_SIZE_PER_CU_VI;
 		q->ctl_stack_size = PAGE_ALIGN_UP(ctl_stack_size);
-		q->ctx_save_restore_size = PAGE_ALIGN_UP(wg_data_size);
-
-		if (q->dev_info->asic_family < CHIP_VEGA10)
-			/* GFX8 chips store ctl-stack with WG data */
-			q->ctx_save_restore_size += q->ctl_stack_size;
+		q->ctx_save_restore_size = q->ctl_stack_size + PAGE_ALIGN_UP(wg_data_size);
 
 		return true;
 	}
@@ -423,8 +418,6 @@ static void free_queue(struct queue *q)
 		free_exec_aligned_memory(q->ctx_save_restore,
 					 q->ctx_save_restore_size,
 					 PAGE_SIZE, q->dev_info->asic_family);
-	if (q->ctl_stack_copy)
-		free(q->ctl_stack_copy);
 
 	free_exec_aligned_memory((void *)q, sizeof(*q), PAGE_SIZE, q->dev_info->asic_family);
 }
@@ -461,11 +454,6 @@ static int handle_concrete_asic(struct queue *q,
 				return HSAKMT_STATUS_NO_MEMORY;
 
 			args->ctx_save_restore_address = (uintptr_t)q->ctx_save_restore;
-			if (IS_SOC15(q->dev_info->asic_family)) {
-				q->ctl_stack_copy = malloc(q->ctl_stack_size);
-				if (q->ctl_stack_copy == NULL)
-					return HSAKMT_STATUS_NO_MEMORY;
-			}
 		}
 	}
 
@@ -706,7 +694,6 @@ hsaKmtGetQueueInfo(
 {
 	struct queue *q = PORT_UINT64_TO_VPTR(QueueId);
 	struct kfd_ioctl_get_queue_wave_state_args args;
-	uintptr_t ctl_stack_base_addr = 0;
 
 	CHECK_KFD_OPEN();
 
@@ -718,30 +705,15 @@ hsaKmtGetQueueInfo(
 
 	memset(&args, 0, sizeof(args));
 	args.queue_id = q->queue_id;
-
-	if (IS_SOC15(q->dev_info->asic_family)) {
-		/* From SOC15 onwards the control stack is kept in the MQD.
-		 * The save area is at the beginning of the allocated space.
-		 * Request a copy of the control stack from the KFD.
-		 */
-		ctl_stack_base_addr = (uintptr_t)q->ctl_stack_copy;
-		QueueInfo->UserContextSaveArea = q->ctx_save_restore;
-		args.ctl_stack_address = ctl_stack_base_addr;
-	} else {
-		/* Before SOC15 the control stack is already in userspace.
-		 * The save area immediately follows the control stack.
-		 */
-		ctl_stack_base_addr = (uintptr_t)q->ctx_save_restore;
-		QueueInfo->UserContextSaveArea = (void *)
-			((uintptr_t)q->ctx_save_restore + q->ctl_stack_size);
-		args.ctl_stack_address = 0;
-	}
+	args.ctl_stack_address = (uintptr_t)q->ctx_save_restore;
 
 	if (kmtIoctl(kfd_fd, AMDKFD_IOC_GET_QUEUE_WAVE_STATE, &args) < 0)
 		return HSAKMT_STATUS_ERROR;
 
-	QueueInfo->ControlStackTop = (void *)(ctl_stack_base_addr +
+	QueueInfo->ControlStackTop = (void *)(args.ctl_stack_address +
 				q->ctl_stack_size - args.ctl_stack_used_size);
+	QueueInfo->UserContextSaveArea = (void *)
+				 (args.ctl_stack_address + q->ctl_stack_size);
 	QueueInfo->SaveAreaSizeInBytes = args.save_area_used_size;
 	QueueInfo->ControlStackUsedInBytes = args.ctl_stack_used_size;
 	QueueInfo->NumCUAssigned = q->cu_mask_count;