Commit 975ab7b3 authored by Oded Gabbay's avatar Oded Gabbay
Browse files

habanalabs: count dropped CS because max CS in-flight 



There is a case where the user reaches the maximum number of CS in-flight.
In that case, the driver rejects the new CS of the user with EAGAIN. Count
that event so the user can query the driver later to see if it happened.

Reviewed-by: default avatarTomer Tayar <ttayar@habana.ai>
Signed-off-by: default avatarOded Gabbay <oded.gabbay@gmail.com>
parent 0db57535

drivers/misc/habanalabs/common/command_submission.c

+4 −1
Original line number Diff line number Diff line
@@ -252,6 +252,8 @@ static void cs_counters_aggregate(struct hl_device *hdev, struct hl_ctx *ctx) 
			ctx->cs_counters.parsing_drop_cnt;

	hdev->aggregated_cs_counters.queue_full_drop_cnt +=

			ctx->cs_counters.queue_full_drop_cnt;

	hdev->aggregated_cs_counters.max_cs_in_flight_drop_cnt +=

			ctx->cs_counters.max_cs_in_flight_drop_cnt;

}



static void cs_do_release(struct kref *ref)
@@ -431,8 +433,9 @@ static int allocate_cs(struct hl_device *hdev, struct hl_ctx *ctx, 
				(hdev->asic_prop.max_pending_cs - 1)];



	if (other && !completion_done(&other->completion)) {

		dev_dbg(hdev->dev,

		dev_dbg_ratelimited(hdev->dev,

			"Rejecting CS because of too many in-flights CS\n");

		ctx->cs_counters.max_cs_in_flight_drop_cnt++;

		rc = -EAGAIN;

		goto free_fence;

	}

include/uapi/misc/habanalabs.h

+2 −0
Original line number Diff line number Diff line
@@ -401,12 +401,14 @@ struct hl_info_sync_manager { 
 * @parsing_drop_cnt: dropped due to error in packet parsing

 * @queue_full_drop_cnt: dropped due to queue full

 * @device_in_reset_drop_cnt: dropped due to device in reset

 * @max_cs_in_flight_drop_cnt: dropped due to maximum CS in-flight

 */

struct hl_cs_counters {

	__u64 out_of_mem_drop_cnt;

	__u64 parsing_drop_cnt;

	__u64 queue_full_drop_cnt;

	__u64 device_in_reset_drop_cnt;

	__u64 max_cs_in_flight_drop_cnt;

};



struct hl_info_cs_counters {

CRA Git | Maintained and supported by SUSTech CRA and CCSE