mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready()
@ 2026-11-20 19:00 Ian Rogers
  2026-11-20 19:00 ` [PATCH v2 1/3] scsi: mpi3mr: Return error on all failure exits " Ian Rogers
                   ` (3 more replies)
  0 siblings, 4 replies; 5+ messages in thread
From: Ian Rogers @ 2026-11-20 19:00 UTC (permalink / raw)
  To: Sathya Prakash Veerichetty, Kashyap Desai, Sumit Saxena,
	Sreekanth Reddy, James E.J. Bottomley, Martin K. Petersen
  Cc: Laurence Oberman, mpi3mr-linuxdrv.pdl, linux-scsi, linux-kernel,
	llvm, Elkin Cruz, Ian Rogers

Split the mpi3mr_bring_ioc_ready() error-handling and retry fixes into
three independent patches with individual Fixes: tags as requested by
Laurence Oberman:

1. Return -1 instead of 0 on failure exits in mpi3mr_bring_ioc_ready()
   when retval is 0 (such as unrecoverable controller faults, hot-unplug
   during reset/diag-save waits, or ready-state faults and timeouts
   after mpi3mr_setup_admin_qpair() sets retval = 0).
2. Initialize start_time = jiffies on entry and guard the post-enable
   timestamp reset with a boolean flag so pre-enable failures and
   subsequent retries never read an uninitialized start_time.
3. Rewrite 'elapsed_time_sec < (mrioc->ready_timeout - 60)' as
   'elapsed_time_sec + 60 < mrioc->ready_timeout' to avoid 32-bit
   unsigned underflow if a controller reports ready_timeout < 60.

v2:
  - Split v1 into three patches, each with its own Fixes: tag (Laurence
    Oberman).
  - Move 'if (!retval) retval = -1;' to 'out_device_not_present:' in
    patch 1 so a single hunk covers both out_failed fallthroughs and
    early out_device_not_present jumps cleanly for stable backports.
  - Add Laurence's Reviewed-by tag on patches 1 and 2, and note the
    510s (570s - 60s) firmware timeout math from commit 9634bb07083c in
    patch 3.

v1: https://lore.kernel.org/r/20261119170900.1530138-1-irogers@google.com

Ian Rogers (3):
  scsi: mpi3mr: Return error on all failure exits in
    mpi3mr_bring_ioc_ready()
  scsi: mpi3mr: Initialize start_time on all paths in
    mpi3mr_bring_ioc_ready()
  scsi: mpi3mr: Avoid ready_timeout underflow in
    mpi3mr_bring_ioc_ready()

 drivers/scsi/mpi3mr/mpi3mr_fw.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)

-- 
2.56.1.750.gf70dc682cf-goog


^ permalink raw reply	[flat|nested] 5+ messages in thread

* [PATCH v2 1/3] scsi: mpi3mr: Return error on all failure exits in mpi3mr_bring_ioc_ready()
  2026-11-20 19:00 [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready() Ian Rogers
@ 2026-11-20 19:00 ` Ian Rogers
  2026-11-20 19:00 ` [PATCH v2 2/3] scsi: mpi3mr: Initialize start_time on all paths " Ian Rogers
                   ` (2 subsequent siblings)
  3 siblings, 0 replies; 5+ messages in thread
From: Ian Rogers @ 2026-11-20 19:00 UTC (permalink / raw)
  To: Sathya Prakash Veerichetty, Kashyap Desai, Sumit Saxena,
	Sreekanth Reddy, James E.J. Bottomley, Martin K. Petersen
  Cc: Laurence Oberman, mpi3mr-linuxdrv.pdl, linux-scsi, linux-kernel,
	llvm, Elkin Cruz, Ian Rogers, stable

In mpi3mr_bring_ioc_ready(), all success paths return 0 directly, while
all failure paths either fall through out_failed or jump to
out_device_not_present and return retval.

However, retval is initialized to 0 at function entry and overwritten
with 0 once mpi3mr_setup_admin_qpair() succeeds. As a result, failure
paths that do not set retval themselves - including unrecoverable fault
checks, PCI device removal during reset or diagnostic-save waits,
failure to reach MRIOC_STATE_RESET, and ready-loop fault/reset-history
exits or timeouts - return 0 (success) to mpi3mr_init_ioc() and
mpi3mr_reinit_ioc(), causing initialization to proceed against a failed
or absent controller.

Set retval = -1 at out_device_not_present whenever retval is 0 so every
failure exit returns a non-zero error while preserving specific error
codes from helper calls.

Found during an x86_64 LLVM=1 allyesconfig warning audit with AI coding
assistant help. Verified with a standalone C logic reproducer and by
building drivers/scsi/mpi3mr/mpi3mr_fw.o with LLVM=1 W=1 (-Werror); not
runtime-tested on MPI3 hardware.

Fixes: 59bd9cfe3fa0 ("scsi: mpi3mr: Code refactor of IOC init - part2")
Cc: stable@vger.kernel.org
Reviewed-by: Laurence Oberman <loberman@redhat.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 drivers/scsi/mpi3mr/mpi3mr_fw.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/scsi/mpi3mr/mpi3mr_fw.c b/drivers/scsi/mpi3mr/mpi3mr_fw.c
index f0d3cd398dd0..416f27c37535 100644
--- a/drivers/scsi/mpi3mr/mpi3mr_fw.c
+++ b/drivers/scsi/mpi3mr/mpi3mr_fw.c
@@ -1744,6 +1744,8 @@ out_failed:
 	    "failed to bring to ready state,  current state: %s\n",
 	    mpi3mr_iocstate_name(ioc_state));
 out_device_not_present:
+	if (!retval)
+		retval = -1;
 	return retval;
 }
 
-- 
2.56.1.750.gf70dc682cf-goog


^ permalink raw reply	[flat|nested] 5+ messages in thread

* [PATCH v2 2/3] scsi: mpi3mr: Initialize start_time on all paths in mpi3mr_bring_ioc_ready()
  2026-11-20 19:00 [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready() Ian Rogers
  2026-11-20 19:00 ` [PATCH v2 1/3] scsi: mpi3mr: Return error on all failure exits " Ian Rogers
@ 2026-11-20 19:00 ` Ian Rogers
  2026-11-20 19:00 ` [PATCH v2 3/3] scsi: mpi3mr: Avoid ready_timeout underflow " Ian Rogers
  2026-11-20 20:30 ` [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check " Laurence Oberman
  3 siblings, 0 replies; 5+ messages in thread
From: Ian Rogers @ 2026-11-20 19:00 UTC (permalink / raw)
  To: Sathya Prakash Veerichetty, Kashyap Desai, Sumit Saxena,
	Sreekanth Reddy, James E.J. Bottomley, Martin K. Petersen
  Cc: Laurence Oberman, mpi3mr-linuxdrv.pdl, linux-scsi, linux-kernel,
	llvm, Elkin Cruz, Ian Rogers, stable

In mpi3mr_bring_ioc_ready(), start_time is only assigned after writing
MPI3_SYSIF_IOC_CONFIG_ENABLE_IOC when retry == 0, which sits after three
earlier 'goto out_failed' error paths (soft reset failure, failure to
reach MRIOC_STATE_RESET, and admin queue setup failure).

If any of those pre-enable steps fails on the first attempt, out_failed
computes elapsed_time_sec from an uninitialized start_time. Furthermore,
if out_failed decides to retry (retry == 1) and the second attempt
reaches the ready-polling loop, 'if (retry == 0)' skips assigning
start_time, leaving the ready-polling loop comparing against
uninitialized stack memory as well.

Initialize start_time = jiffies at function entry to cover pre-enable
failure exits, and replace 'if (retry == 0)' with a boolean
enable_started check so start_time is reset the first time IOC enable is
written without pre-enable waits draining ready_timeout.

Found via Clang -Wconditional-uninitialized on an x86_64 allyesconfig
build with AI coding assistant help. Verified with a standalone C
reproducer under ASan/UBSan and by building
drivers/scsi/mpi3mr/mpi3mr_fw.o with LLVM=1 W=1 (-Werror) and
-Wconditional-uninitialized; not runtime-tested on MPI3 hardware.

Fixes: 9634bb07083c ("scsi: mpi3mr: Enhance the Enable Controller retry logic")
Cc: stable@vger.kernel.org
Reviewed-by: Laurence Oberman <loberman@redhat.com>
Signed-off-by: Ian Rogers <irogers@google.com>
---
 drivers/scsi/mpi3mr/mpi3mr_fw.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/scsi/mpi3mr/mpi3mr_fw.c b/drivers/scsi/mpi3mr/mpi3mr_fw.c
index 416f27c37535..1aeae8a55b1b 100644
--- a/drivers/scsi/mpi3mr/mpi3mr_fw.c
+++ b/drivers/scsi/mpi3mr/mpi3mr_fw.c
@@ -1582,8 +1582,9 @@ static int mpi3mr_bring_ioc_ready(struct mpi3mr_ioc *mrioc)
 	int retval = 0;
 	enum mpi3mr_iocstate ioc_state;
 	u64 base_info;
+	bool enable_started = false;
 	u8 retry = 0;
-	u64 start_time, elapsed_time_sec;
+	u64 start_time = jiffies, elapsed_time_sec;
 
 retry_bring_ioc_ready:
 
@@ -1690,8 +1691,10 @@ retry_bring_ioc_ready:
 	ioc_config |= MPI3_SYSIF_IOC_CONFIG_ENABLE_IOC;
 	writel(ioc_config, &mrioc->sysif_regs->ioc_configuration);
 
-	if (retry == 0)
+	if (!enable_started) {
 		start_time = jiffies;
+		enable_started = true;
+	}
 
 	timeout = mrioc->ready_timeout * 10;
 	do {
-- 
2.56.1.750.gf70dc682cf-goog


^ permalink raw reply	[flat|nested] 5+ messages in thread

* [PATCH v2 3/3] scsi: mpi3mr: Avoid ready_timeout underflow in mpi3mr_bring_ioc_ready()
  2026-11-20 19:00 [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready() Ian Rogers
  2026-11-20 19:00 ` [PATCH v2 1/3] scsi: mpi3mr: Return error on all failure exits " Ian Rogers
  2026-11-20 19:00 ` [PATCH v2 2/3] scsi: mpi3mr: Initialize start_time on all paths " Ian Rogers
@ 2026-11-20 19:00 ` Ian Rogers
  2026-11-20 20:30 ` [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check " Laurence Oberman
  3 siblings, 0 replies; 5+ messages in thread
From: Ian Rogers @ 2026-11-20 19:00 UTC (permalink / raw)
  To: Sathya Prakash Veerichetty, Kashyap Desai, Sumit Saxena,
	Sreekanth Reddy, James E.J. Bottomley, Martin K. Petersen
  Cc: Laurence Oberman, mpi3mr-linuxdrv.pdl, linux-scsi, linux-kernel,
	llvm, Elkin Cruz, Ian Rogers, stable

Commit 9634bb07083c ("scsi: mpi3mr: Enhance the Enable Controller retry
logic") caps cumulative enable retries at 'mrioc->ready_timeout - 60'
(510 seconds on firmware advertising a 570-second ready timeout) so a
retry is not started with less than 60 seconds remaining.

Because mrioc->ready_timeout is a u32 read from the controller's
ioc_information register, any value below 60 (for example 0 from a
broken or disconnected controller) wraps around in 32-bit unsigned
arithmetic (e.g. 0xffffffc4) before zero-extending to u64 for the
comparison with elapsed_time_sec, spuriously retrying instead of
aborting.

Rewrite the check as 'elapsed_time_sec + 60 < mrioc->ready_timeout' to
avoid unsigned underflow while preserving the 60-second margin when
ready_timeout >= 60. (Note that if any firmware legitimately advertises
ready_timeout < 60, no retries will be attempted.)

Found during code inspection alongside a Clang warning audit with AI
coding assistant help. Verified with a standalone C reproducer under
UBSan and by building drivers/scsi/mpi3mr/mpi3mr_fw.o with LLVM=1 W=1
(-Werror); not runtime-tested on MPI3 hardware.

Fixes: 9634bb07083c ("scsi: mpi3mr: Enhance the Enable Controller retry logic")
Cc: stable@vger.kernel.org
Signed-off-by: Ian Rogers <irogers@google.com>
---
 drivers/scsi/mpi3mr/mpi3mr_fw.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/scsi/mpi3mr/mpi3mr_fw.c b/drivers/scsi/mpi3mr/mpi3mr_fw.c
index 1aeae8a55b1b..45544ad21a6e 100644
--- a/drivers/scsi/mpi3mr/mpi3mr_fw.c
+++ b/drivers/scsi/mpi3mr/mpi3mr_fw.c
@@ -1734,7 +1734,7 @@ retry_bring_ioc_ready:
 
 out_failed:
 	elapsed_time_sec = jiffies_to_msecs(jiffies - start_time)/1000;
-	if ((retry < 2) && (elapsed_time_sec < (mrioc->ready_timeout - 60))) {
+	if (retry < 2 && elapsed_time_sec + 60 < mrioc->ready_timeout) {
 		retry++;
 
 		ioc_warn(mrioc, "retrying to bring IOC ready, retry_count:%d\n"
-- 
2.56.1.750.gf70dc682cf-goog


^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready()
  2026-11-20 19:00 [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready() Ian Rogers
                   ` (2 preceding siblings ...)
  2026-11-20 19:00 ` [PATCH v2 3/3] scsi: mpi3mr: Avoid ready_timeout underflow " Ian Rogers
@ 2026-11-20 20:30 ` Laurence Oberman
  3 siblings, 0 replies; 5+ messages in thread
From: Laurence Oberman @ 2026-11-20 20:30 UTC (permalink / raw)
  To: Ian Rogers, Sathya Prakash Veerichetty, Kashyap Desai,
	Sumit Saxena, Sreekanth Reddy, James E.J. Bottomley,
	Martin K. Petersen
  Cc: mpi3mr-linuxdrv.pdl, linux-scsi, linux-kernel, llvm, Elkin Cruz

On Fri, 2026-11-20 at 11:00 -0800, Ian Rogers wrote:
> Split the mpi3mr_bring_ioc_ready() error-handling and retry fixes
> into
> three independent patches with individual Fixes: tags as requested by
> Laurence Oberman:
> 
> 1. Return -1 instead of 0 on failure exits in
> mpi3mr_bring_ioc_ready()
>    when retval is 0 (such as unrecoverable controller faults, hot-
> unplug
>    during reset/diag-save waits, or ready-state faults and timeouts
>    after mpi3mr_setup_admin_qpair() sets retval = 0).
> 2. Initialize start_time = jiffies on entry and guard the post-enable
>    timestamp reset with a boolean flag so pre-enable failures and
>    subsequent retries never read an uninitialized start_time.
> 3. Rewrite 'elapsed_time_sec < (mrioc->ready_timeout - 60)' as
>    'elapsed_time_sec + 60 < mrioc->ready_timeout' to avoid 32-bit
>    unsigned underflow if a controller reports ready_timeout < 60.
> 
> v2:
>   - Split v1 into three patches, each with its own Fixes: tag
> (Laurence
>     Oberman).
>   - Move 'if (!retval) retval = -1;' to 'out_device_not_present:' in
>     patch 1 so a single hunk covers both out_failed fallthroughs and
>     early out_device_not_present jumps cleanly for stable backports.
>   - Add Laurence's Reviewed-by tag on patches 1 and 2, and note the
>     510s (570s - 60s) firmware timeout math from commit 9634bb07083c
> in
>     patch 3.
> 
> v1:
> https://lore.kernel.org/r/20261119170900.1530138-1-irogers@google.com
> 
> Ian Rogers (3):
>   scsi: mpi3mr: Return error on all failure exits in
>     mpi3mr_bring_ioc_ready()
>   scsi: mpi3mr: Initialize start_time on all paths in
>     mpi3mr_bring_ioc_ready()
>   scsi: mpi3mr: Avoid ready_timeout underflow in
>     mpi3mr_bring_ioc_ready()
> 
>  drivers/scsi/mpi3mr/mpi3mr_fw.c | 11 ++++++++---
>  1 file changed, 8 insertions(+), 3 deletions(-)

Hello Ian

Thanks for the quick v2. Tested the series on 7.4.0-rc1 with a
Broadcom SAS40xx/41xx (FW 8.5.1.0.0.1), two SAS disks attached.

Firmware advertises "ready timeout: 510 seconds" (ioc_info
0xff000000), before and after the series.

Tests, all with mpi3mr_bring_ioc_ready() reaching ready on the first
attempt and no "retrying to bring IOC ready" or "failed to bring"
messages:

 - boot probe
 - modprobe -r / modprobe (mpi3mr_init_ioc path)
 - sg_reset -H under fio 4k randread QD32 (mpi3mr_reinit_ioc path);
   host reset completed in ~45s including diag save and port enable,
   fio completed with no errors
 - PCI remove/rescan of the controller
 - 20 iterations of reload + host reset under fio: 40 successful
   ready transitions, no retries, no failures, no fio errors

The failure paths themselves were not exercised, as they are hard to
trigger on healthy hardware; this confirms no regression on the good
path.

Minor, 3/3 changelog: the IOC info timeout field is 8 bits in 2s
units, so ready_timeout tops out at 510s and the retry window at 450s;
the "510 seconds" in 9634bb07083c most likely refers to the maximum
ready timeout rather than the retry cap, so 570s cannot be advertised.
Doesn't affect the code change. Martin may be able to fix it up on
apply.

The question to Broadcom on firmware advertising ready_timeout < 60
is still open for 3/3; on this controller it makes no difference.

For the series:

Tested-by: Laurence Oberman <loberman@redhat.com>

Regards
Laurence


^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2026-11-20 20:30 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-20 19:00 [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check in mpi3mr_bring_ioc_ready() Ian Rogers
2026-11-20 19:00 ` [PATCH v2 1/3] scsi: mpi3mr: Return error on all failure exits " Ian Rogers
2026-11-20 19:00 ` [PATCH v2 2/3] scsi: mpi3mr: Initialize start_time on all paths " Ian Rogers
2026-11-20 19:00 ` [PATCH v2 3/3] scsi: mpi3mr: Avoid ready_timeout underflow " Ian Rogers
2026-11-20 20:30 ` [PATCH v2 0/3] scsi: mpi3mr: Fix error return, start_time, and timeout check " Laurence Oberman

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®