Add support for handling errors that require a complete
device power cycle (cold reset) to recover.
Certain error conditions leave the device in a persistent hardware
error state that cannot be cleared through existing recovery mechanisms
such as driver reload or PCIe reset. In these cases, functionality can
only be restored by performing a cold reset.
To support this, the series introduces a new DRM wedging recovery
method, DRM_WEDGE_RECOVERY_COLD_RESET (BIT(4)). When a device is wedged
with this method, the DRM core notifies userspace via a uevent that a cold
reset is required. This allows userspace to take appropriate action to
power-cycle the device.
Example uevent received:
SUBSYSTEM=drm
WEDGED=cold-reset
DEVPATH=/devices/.../drm/card0
v2:
- Add use case: Handling errors from power management unit,
which requires a complete power cycle to
recover. (Christian)
- Add several instead of number to avoid update. (Jani)
v3:
- Update any scenario that requires cold-reset. (Riana)
- Update document with generic scenario. (Riana)
- Consistent with terminology. (Raag)
- Remove already covered information.
- Use PUNIT instead of PMU. (Riana)
- Use consistent wordingi.
- Remove log. (Raag)
v4:
- Rename cold reset to power cyclce. (Raag)
- Update doc. (Raag/Riana)
- Change commit message. (Raag)
- Make function static. (Raag)
v5:
- Make it consistent with consumer expectations. (Raag)
- Update commit message.
- Remove unbind.
- Simplify cold-reset script.
- Remove kdoc for static function.
- Remove xe_ prefix for static function.
v6:
- Drop "last resort" wording. (Riana)
- Look up the hotplug slot in DEVPATH instead of scanning
every PCI slot on the system. (Raag)
- Drop arbitrary sleep values from the example script.
- Expand commit message to explain why SUR_DN is masked. (Raag/Riana)
- Check Slot Implemented bit before reading Slot Capabilities, per
PCIe spec. (Riana)
- Add debug log.
v7:
- Update recovery script. (Raag)
- Handle surprise link down event properly. (Aravind/Riana)
- Update commit message. (Riana)
- Correct log message.
v8:
- Add rescan instead of reset. (Raag)
- Use find_usp_dev() in punit_error_handler() function.
v9:
- Remove unwanted header. (Sashiko)
- Removed #ifdef CONFIG_PCIEAER. (Riana)
- Used pci_find_ext_capability() instead of usp->aer_cap.
- Clear the PCI_ERR_UNC_SURPDN status bit (W1C) after
reset complete. (Lukas Wunner)
- Use pci_clear_and_set_config_dword() helper.
v10:
- Rebase.
- Fix column width. (Sashiko)
v11:
- Make udev rules in single line. (Sashiko)
v12:
- Trigger punit handler using fault-inject.
v13:
- Rebase.
- Rename inject_punit_error to wedge_cold_reset. (Riana)
- Sashiko corner case issue addressed with
commit 20bc4883c7c0 ("drm/xe/ras: Fix boot-time ras error processing").
Cc: André Almeida <[email protected]>
Cc: Christian König <[email protected]>
Cc: David Airlie <[email protected]>
Cc: Simona Vetter <[email protected]>
Cc: Maxime Ripard <[email protected]>
Cc: Maarten Lankhorst <[email protected]>
Cc: Thomas Zimmermann <[email protected]>
Mallesh Koujalagi (4):
drm: Add DRM_WEDGE_RECOVERY_COLD_RESET recovery method
drm/doc: Document DRM_WEDGE_RECOVERY_COLD_RESET recovery method
drm/xe: Handle PUNIT errors by requesting cold-reset recovery
drm/xe/ras: Use fault-inject to trigger cold-reset wedge
Documentation/gpu/drm-uapi.rst | 93 +++++++++++++++++++++++++++++++--
drivers/gpu/drm/drm_drv.c | 2 +
drivers/gpu/drm/xe/xe_debugfs.c | 4 ++
drivers/gpu/drm/xe/xe_debugfs.h | 2 +
drivers/gpu/drm/xe/xe_ras.c | 15 +++++-
include/drm/drm_device.h | 1 +
6 files changed, 111 insertions(+), 6 deletions(-)
--
2.48.1
