diff mbox series

[v2,for-next,06/17] RDMA/bnxt_re: Avoid the command wait if firmware is inactive

Message ID 1686308514-11996-7-git-send-email-selvin.xavier@broadcom.com (mailing list archive)
State Accepted
Headers show
Series RDMA/bnxt_re: Control path updates | expand

Commit Message

Selvin Xavier June 9, 2023, 11:01 a.m. UTC 
From: Kashyap Desai <kashyap.desai@broadcom.com>

Add a check to avoid waiting if driver already detects a
FW timeout. Return success for resource destroy in case
the device is detached. Add helper function to map timeout
error code to success.

Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Selvin Xavier <selvin.xavier@broadcom.com>
---
 drivers/infiniband/hw/bnxt_re/qplib_rcfw.c | 52 +++++++++++++++++++++++++++---
 1 file changed, 48 insertions(+), 4 deletions(-)
diff mbox series

Patch

diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
index 3b242cc..3c4f72a 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
@@ -54,9 +54,46 @@ 
 static void bnxt_qplib_service_creq(struct tasklet_struct *t);
 
 /**
+ * bnxt_qplib_map_rc  -  map return type based on opcode
+ * @opcode    -  roce slow path opcode
+ *
+ * In some cases like firmware halt is detected, the driver is supposed to
+ * remap the error code of the timed out command.
+ *
+ * It is not safe to assume hardware is really inactive so certain opcodes
+ * like destroy qp etc are not safe to be returned success, but this function
+ * will be called when FW already reports a timeout. This would be possible
+ * only when FW crashes and resets. This will clear all the HW resources.
+ *
+ * Returns:
+ * 0 to communicate success to caller.
+ * Non zero error code to communicate failure to caller.
+ */
+static int bnxt_qplib_map_rc(u8 opcode)
+{
+	switch (opcode) {
+	case CMDQ_BASE_OPCODE_DESTROY_QP:
+	case CMDQ_BASE_OPCODE_DESTROY_SRQ:
+	case CMDQ_BASE_OPCODE_DESTROY_CQ:
+	case CMDQ_BASE_OPCODE_DEALLOCATE_KEY:
+	case CMDQ_BASE_OPCODE_DEREGISTER_MR:
+	case CMDQ_BASE_OPCODE_DELETE_GID:
+	case CMDQ_BASE_OPCODE_DESTROY_QP1:
+	case CMDQ_BASE_OPCODE_DESTROY_AH:
+	case CMDQ_BASE_OPCODE_DEINITIALIZE_FW:
+	case CMDQ_BASE_OPCODE_MODIFY_ROCE_CC:
+	case CMDQ_BASE_OPCODE_SET_LINK_AGGR_MODE:
+		return 0;
+	default:
+		return -ETIMEDOUT;
+	}
+}
+
+/**
  * __wait_for_resp   -	Don't hold the cpu context and wait for response
  * @rcfw      -   rcfw channel instance of rdev
  * @cookie    -   cookie to track the command
+ * @opcode    -   rcfw submitted for given opcode
  *
  * Wait for command completion in sleepable context.
  *
@@ -64,7 +101,7 @@  static void bnxt_qplib_service_creq(struct tasklet_struct *t);
  * 0 if command is completed by firmware.
  * Non zero error code for rest of the case.
  */
-static int __wait_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie)
+static int __wait_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie, u8 opcode)
 {
 	struct bnxt_qplib_cmdq_ctx *cmdq;
 	u16 cbit;
@@ -74,6 +111,9 @@  static int __wait_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie)
 	cbit = cookie % rcfw->cmdq_depth;
 
 	do {
+		if (test_bit(ERR_DEVICE_DETACHED, &cmdq->flags))
+			return bnxt_qplib_map_rc(opcode);
+
 		/* Non zero means command completed */
 		ret = wait_event_timeout(cmdq->waitq,
 					 !test_bit(cbit, cmdq->cmdq_bitmap),
@@ -94,6 +134,7 @@  static int __wait_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie)
  * __block_for_resp   -	hold the cpu context and wait for response
  * @rcfw      -   rcfw channel instance of rdev
  * @cookie    -   cookie to track the command
+ * @opcode    -   rcfw submitted for given opcode
  *
  * This function will hold the cpu (non-sleepable context) and
  * wait for command completion. Maximum holding interval is 8 second.
@@ -102,7 +143,7 @@  static int __wait_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie)
  * -ETIMEOUT if command is not completed in specific time interval.
  * 0 if command is completed by firmware.
  */
-static int __block_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie)
+static int __block_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie, u8 opcode)
 {
 	struct bnxt_qplib_cmdq_ctx *cmdq = &rcfw->cmdq;
 	unsigned long issue_time = 0;
@@ -112,6 +153,9 @@  static int __block_for_resp(struct bnxt_qplib_rcfw *rcfw, u16 cookie)
 	issue_time = jiffies;
 
 	do {
+		if (test_bit(ERR_DEVICE_DETACHED, &cmdq->flags))
+			return bnxt_qplib_map_rc(opcode);
+
 		udelay(1);
 
 		bnxt_qplib_service_creq(&rcfw->creq.creq_tasklet);
@@ -267,9 +311,9 @@  int bnxt_qplib_rcfw_send_message(struct bnxt_qplib_rcfw *rcfw,
 	} while (retry_cnt--);
 
 	if (msg->block)
-		rc = __block_for_resp(rcfw, cookie);
+		rc = __block_for_resp(rcfw, cookie, opcode);
 	else
-		rc = __wait_for_resp(rcfw, cookie);
+		rc = __wait_for_resp(rcfw, cookie, opcode);
 	if (rc) {
 		/* timed out */
 		dev_err(&rcfw->pdev->dev, "cmdq[%#x]=%#x timedout (%d)msec\n",