Commit 9acdc21b authored by Ofir Bitton's avatar Ofir Bitton Committed by Oded Gabbay
Browse files

habanalabs: add current PI value to cpu packets 



In order to increase cpucp messaging reliability we will add
the current PI value to the descriptor sent to F/W.
F/W will wait for the PI value as an indication of a valid packet.

Signed-off-by: default avatarOfir Bitton <obitton@habana.ai>
Reviewed-by: default avatarOded Gabbay <ogabbay@kernel.org>
Signed-off-by: default avatarOded Gabbay <ogabbay@kernel.org>
parent 7363805b

drivers/misc/habanalabs/common/firmware_if.c

+12 −2
Original line number Diff line number Diff line
@@ -212,7 +212,8 @@ int hl_fw_send_cpu_message(struct hl_device *hdev, u32 hw_queue_id, u32 *msg, 
	struct asic_fixed_properties *prop = &hdev->asic_prop;

	struct cpucp_packet *pkt;

	dma_addr_t pkt_dma_addr;

	u32 tmp, expected_ack_val;

	struct hl_bd *sent_bd;

	u32 tmp, expected_ack_val, pi;

	int rc = 0;



	pkt = hdev->asic_funcs->cpu_accessible_dma_pool_alloc(hdev, len,
@@ -237,6 +238,7 @@ int hl_fw_send_cpu_message(struct hl_device *hdev, u32 hw_queue_id, u32 *msg, 


	/* set fence to a non valid value */

	pkt->fence = cpu_to_le32(UINT_MAX);

	pi = queue->pi;



	/*

	 * The CPU queue is a synchronous queue with an effective depth of
@@ -246,7 +248,7 @@ int hl_fw_send_cpu_message(struct hl_device *hdev, u32 hw_queue_id, u32 *msg, 
	 * Which means that we don't need to lock the access to the entire H/W

	 * queues module when submitting a JOB to the CPU queue.

	 */

	hl_hw_queue_submit_bd(hdev, queue, 0, len, pkt_dma_addr);

	hl_hw_queue_submit_bd(hdev, queue, hl_queue_inc_ptr(queue->pi), len, pkt_dma_addr);



	if (prop->fw_app_cpu_boot_dev_sts0 & CPU_BOOT_DEV_STS0_PKT_PI_ACK_EN)

		expected_ack_val = queue->pi;
@@ -278,6 +280,14 @@ int hl_fw_send_cpu_message(struct hl_device *hdev, u32 hw_queue_id, u32 *msg, 
		*result = le64_to_cpu(pkt->result);

	}



	/* Scrub previous buffer descriptor 'ctl' field which contains the

	 * previous PI value written during packet submission.

	 * We must do this or else F/W can read an old value upon queue wraparound.

	 */

	sent_bd = queue->kernel_address;

	sent_bd += hl_pi_2_offset(pi);

	sent_bd->ctl = cpu_to_le32(UINT_MAX);



out:

	mutex_unlock(&hdev->send_cpu_message_lock);