[PATCH v3] s390/pci: Avoid deadlock between PCI error recovery and mlx5 crdump

[Gerd Bayer <gbayer@linux.ibm.com>]

Do not block PCI config accesses through pci_cfg_access_lock() when
executing the s390 variant of PCI error recovery: Acquire just
device_lock() instead of pci_dev_lock() as powerpc's EEH and
generig PCI AER processing do.

During error recovery testing a pair of tasks was reported to be hung:

mlx5_core 0000:00:00.1: mlx5_health_try_recover:338:(pid 5553): health recovery flow aborted, PCI reads still not working
INFO: task kmcheck:72 blocked for more than 122 seconds.
      Not tainted 5.14.0-570.12.1.bringup7.el9.s390x #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:kmcheck         state:D stack:0     pid:72    tgid:72    ppid:2      flags:0x00000000
Call Trace:
 [<000000065256f030>] __schedule+0x2a0/0x590
 [<000000065256f356>] schedule+0x36/0xe0
 [<000000065256f572>] schedule_preempt_disabled+0x22/0x30
 [<0000000652570a94>] __mutex_lock.constprop.0+0x484/0x8a8
 [<000003ff800673a4>] mlx5_unload_one+0x34/0x58 [mlx5_core]
 [<000003ff8006745c>] mlx5_pci_err_detected+0x94/0x140 [mlx5_core]
 [<0000000652556c5a>] zpci_event_attempt_error_recovery+0xf2/0x398
 [<0000000651b9184a>] __zpci_event_error+0x23a/0x2c0
INFO: task kworker/u1664:6:1514 blocked for more than 122 seconds.
      Not tainted 5.14.0-570.12.1.bringup7.el9.s390x #1
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:kworker/u1664:6 state:D stack:0     pid:1514  tgid:1514  ppid:2      flags:0x00000000
Workqueue: mlx5_health0000:00:00.0 mlx5_fw_fatal_reporter_err_work [mlx5_core]
Call Trace:
 [<000000065256f030>] __schedule+0x2a0/0x590
 [<000000065256f356>] schedule+0x36/0xe0
 [<0000000652172e28>] pci_wait_cfg+0x80/0xe8
 [<0000000652172f94>] pci_cfg_access_lock+0x74/0x88
 [<000003ff800916b6>] mlx5_vsc_gw_lock+0x36/0x178 [mlx5_core]
 [<000003ff80098824>] mlx5_crdump_collect+0x34/0x1c8 [mlx5_core]
 [<000003ff80074b62>] mlx5_fw_fatal_reporter_dump+0x6a/0xe8 [mlx5_core]
 [<0000000652512242>] devlink_health_do_dump.part.0+0x82/0x168
 [<0000000652513212>] devlink_health_report+0x19a/0x230
 [<000003ff80075a12>] mlx5_fw_fatal_reporter_err_work+0xba/0x1b0 [mlx5_core]

No kernel log of the exact same error with an upstream kernel is
available - but the very same deadlock situation can be constructed there,
too:

- task: kmcheck
  mlx5_unload_one() tries to acquire devlink lock while the PCI error
  recovery code has set pdev->block_cfg_access by way of
  pci_cfg_access_lock()
- task: kworker
  mlx5_crdump_collect() tries to set block_cfg_access through
  pci_cfg_access_lock() while devlink_health_report() had acquired
  the devlink lock.

A similar deadlock situation can be reproduced by requesting a
crdump with
  > devlink health dump show pci/<BDF> reporter fw_fatal

while PCI error recovery is executed on the same <BDF> physical function
by mlx5_core's pci_error_handlers. On s390 this can be injected with
  > zpcictl --reset-fw <BDF>

Tests with this patch failed to reproduce that second deadlock situation,
the devlink command is rejected with "kernel answers: Permission denied" -
and we get a kernel log message of:

mlx5_core 1ed0:00:00.1: mlx5_crdump_collect:50:(pid 254382): crdump: failed to lock vsc gw err -5

because the config read of VSC_SEMAPHORE is rejected by the underlying
hardware.

Two prior attempts to address this issue have been discussed and
ultimately rejected [see link], with the primary argument that s390's
implementation of PCI error recovery is imposing restrictions that
neither powerpc's EEH nor PCI AER handling need. Tests show that PCI
error recovery on s390 is running to completion even without blocking
access to PCI config space.

Link: https://lore.kernel.org/all/20251007144826.2825134-1-gbayer@linux.ibm.com/
Cc: stable@vger.kernel.org
Fixes: 4cdf2f4e24ff ("s390/pci: implement minimal PCI error recovery")
Reviewed-by: Niklas Schnelle <schnelle@linux.ibm.com>
Signed-off-by: Gerd Bayer <gbayer@linux.ibm.com>
---
Hi Niklas, Shay, Jason,

by now I believe fixing this in s390/pci is the right way to go, since
the other PCI error recovery implementations apparently don't require
this strict blocking of accesses to the PCI config space.
    
Hi Alexander, Vasily, Heiko,
    
while I sent this to netdev since prior versions were discussed there,
I assume this patch will go through the s390 tree, right?
    
Thanks,
Gerd

    
---
Changes in v3:
- Incorporate changes to commit message as suggested by Niklas.
- Link to v2: https://lore.kernel.org/r/20251015-fix_pcirecov_master-v2-1-e07962fe9558@linux.ibm.com

Changes in v2:
- Rebase to upstream master
---
 arch/s390/pci/pci_event.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/s390/pci/pci_event.c b/arch/s390/pci/pci_event.c
index b95376041501f479eee20705d45fb8c68553da71..27db1e72c623f8a289cae457e87f0a9896ed241d 100644
--- a/arch/s390/pci/pci_event.c
+++ b/arch/s390/pci/pci_event.c
@@ -188,7 +188,7 @@ static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev)
 	 * is unbound or probed and that userspace can't access its
 	 * configuration space while we perform recovery.
 	 */
-	pci_dev_lock(pdev);
+	device_lock(&pdev->dev);
 	if (pdev->error_state == pci_channel_io_perm_failure) {
 		ers_res = PCI_ERS_RESULT_DISCONNECT;
 		goto out_unlock;
@@ -257,7 +257,7 @@ static pci_ers_result_t zpci_event_attempt_error_recovery(struct pci_dev *pdev)
 		driver->err_handler->resume(pdev);
 	pci_uevent_ers(pdev, PCI_ERS_RESULT_RECOVERED);
 out_unlock:
-	pci_dev_unlock(pdev);
+	device_unlock(&pdev->dev);
 	zpci_report_status(zdev, "recovery", status_str);
 
 	return ers_res;

---
base-commit: 9b332cece987ee1790b2ed4c989e28162fa47860
change-id: 20251015-fix_pcirecov_master-55fe3705c6c6

Best regards,
-- 
Gerd Bayer <gbayer@linux.ibm.com>