[PATCH 0/2][SRU][impish/OEM-5.14] NVMe devices fail to probe due to ACPI power state change
Heitor Alves de Siqueira
halves at canonical.com
Wed Apr 13 16:50:14 UTC 2022
BugLink: https://bugs.launchpad.net/bugs/1942624
Hi kernel-team,
Users have reported specific NVMe disks failing to probe due to a power
management regression in ACPI. Our affected kernels include 5.13 (impish) and
5.14 (focal-oem). For the impish kernel, there's an additional dependency patch
that has reworked how unused resources are turned off.
I've provided test builds of the affected kernels to the Launchpad bug and to
some affected users, and they've confirmed the patches fix the issues without
noticeable regressions.
[Impact]
* Specific NVMe devices fail to probe and become unusable after boot
* Caused by an ACPI regression that doesn't correctly handle power states
* Upstream regression commit:
7e4fdeafa61f ACPI: power: Turn off unused power resources unconditionally
* Regression window for Ubuntu includes 5.13- and 5.14-based kernels
[Test Plan]
* Boot affected kernel and validate whether NVMe device is usable
* Check kernel logs for failed probe message:
"can't change power state from D3Cold to D0 (config space inaccessible)"
[Fix]
* Fixed by not turning off power resources in unknown state
* Fix was introduced by commit:
bc2836859643 ACPI: PM: Do not turn off power resources in unknown state
* Kernels starting with 5.15 (e.g. Jammy) not affected, as they already contain
the fix above
[Regression Potential]
* NVMe devices continue failing to probe
* Other devices become unusable after power state changes
* Further regressions would affect power state of devices, possibly after boot
Rafael J. Wysocki (2):
ACPI: power: Rework turning off unused power resources
ACPI: PM: Do not turn off power resources in unknown state
drivers/acpi/internal.h | 2 +-
drivers/acpi/power.c | 47 ++++++++++++-----------------------------
drivers/acpi/scan.c | 2 +-
drivers/acpi/sleep.c | 2 +-
4 files changed, 17 insertions(+), 36 deletions(-)
--
2.35.1
More information about the kernel-team
mailing list