This is a note to let you know that I've just added the patch titled
RDMA/bnxt_re: Avoid the command wait if firmware is inactive
to the 6.4-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
rdma-bnxt_re-avoid-the-command-wait-if-firmware-is-i.patch
and it can be found in the queue-6.4 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit 15a3a71e5ca04ee2b1ab2e62f847209400a429e2
Author: Kashyap Desai <kashyap.desai@xxxxxxxxxxxx>
Date: Fri Jun 9 04:01:43 2023 -0700
RDMA/bnxt_re: Avoid the command wait if firmware is inactive
[ Upstream commit 3022cc15119733cebaef05feddb5d87b9e401c0e ]
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@xxxxxxxxxxxx>
Signed-off-by: Selvin Xavier <selvin.xavier@xxxxxxxxxxxx>
Link: https://lore.kernel.org/r/1686308514-11996-7-git-send-email-selvin.xavier@xxxxxxxxxxxx
Signed-off-by: Leon Romanovsky <leon@xxxxxxxxxx>
Stable-dep-of: 29900bf351e1 ("RDMA/bnxt_re: Fix hang during driver unload")
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
index 918e588588885..bfa0f29c7abf4 100644
--- a/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
+++ b/drivers/infiniband/hw/bnxt_re/qplib_rcfw.c
@@ -53,10 +53,47 @@
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",
