Public bug reported:
BugLink: https://bugs.launchpad.net/bugs/2166862
[ Impact ]
7.0 kernel introduced a bug where offlining enough of CPUs before s0i3
causes system hang. This is platform independent (it can reproduce in
QEMU for example).
To reproduce, on a ZBook Ultra G1a (with 32 cores), with
7.0.0-31-generic kernel, offlining processors:
$ printf '0\n' | sudo tee /sys/devices/system/cpu/cpu{8..31}/online
And let the system supend (e.g. systemctl suspend). Then after a resume,
the system would end up hang or extremely laggy.
[ Fix ]
Cherry-pick the following patch:
- [PATCH v4] lib/group_cpus: Snapshot cluster masks to keep grouping
hotplug invariant
Which is currently under upstream review. The hope would be for it to
land in one of the 7.3-rcX, or in 7.4.
[ Test plan ]
On the said platform, offline some processors, say:
$ printf '0\n' | sudo tee /sys/devices/system/cpu/cpu{8..31}/online
Suspend the system:
$ systemctl suspend
And resume. The system should resume normally.
[ Where problem could occur ]
From the discussion in the mail thread, this would be a general
scheduler code change. Ideally would need test coverage from other
platforms.
[ Additional information ]
The offending commit 89802ca36c96 ("lib/group_cpus: make group CPU
cluster aware") was included in 7.0 kernel, so older kernel shouldn't
get impacted.
[1]
https://lore.kernel.org/all/[email protected]/
** Affects: linux (Ubuntu)
Importance: Undecided
Status: New
** Summary changed:
- [SRU] System hang after S0ix if CPUs are offlined
+ [SRU] System hang after S0ix if enough of CPUs are offlined beforehand
** Description changed:
[ Impact ]
7.0 kernel introduced a bug where offlining enough of CPUs before s0i3
causes system hang. This is platform independent (it can reproduce in
QEMU for example).
To reproduce, on a ZBook Ultra G1a (with 32 cores), with
7.0.0-31-generic kernel, offlining processors:
$ printf '0\n' | sudo tee /sys/devices/system/cpu/cpu{8..31}/online
And let the system supend (e.g. systemctl suspend). Then after a resume,
the system would end up hang or extremely laggy.
[ Fix ]
Cherry-pick the following patch:
- [PATCH v4] lib/group_cpus: Snapshot cluster masks to keep grouping
hotplug invariant
Which is currently under upstream review. The hope would be for it to
land in one of the 7.3-rcX, or in 7.4.
[ Test plan ]
On the said platform, offline some processors, say:
$ printf '0\n' | sudo tee /sys/devices/system/cpu/cpu{8..31}/online
Suspend the system:
$ systemctl suspend
And resume. The system should resume normally.
+ [ Where problem could occur ]
+
+ From the discussion in the mail thread, this would be a general
+ scheduler code change. Ideally would need test coverage from other
+ platforms.
+
+
[ Additional information ]
+
+ The offending commit 89802ca36c96 ("lib/group_cpus: make group CPU
+ cluster aware") was included in 7.0 kernel, so older kernel shouldn't
+ get impacted.
[1]
https://lore.kernel.org/all/[email protected]/
** Description changed:
+ BugLink: https://bugs.launchpad.net/bugs/2166862
+
[ Impact ]
7.0 kernel introduced a bug where offlining enough of CPUs before s0i3
causes system hang. This is platform independent (it can reproduce in
QEMU for example).
To reproduce, on a ZBook Ultra G1a (with 32 cores), with
7.0.0-31-generic kernel, offlining processors:
$ printf '0\n' | sudo tee /sys/devices/system/cpu/cpu{8..31}/online
And let the system supend (e.g. systemctl suspend). Then after a resume,
the system would end up hang or extremely laggy.
-
[ Fix ]
Cherry-pick the following patch:
- [PATCH v4] lib/group_cpus: Snapshot cluster masks to keep grouping
hotplug invariant
Which is currently under upstream review. The hope would be for it to
land in one of the 7.3-rcX, or in 7.4.
-
[ Test plan ]
On the said platform, offline some processors, say:
$ printf '0\n' | sudo tee /sys/devices/system/cpu/cpu{8..31}/online
Suspend the system:
$ systemctl suspend
And resume. The system should resume normally.
-
[ Where problem could occur ]
From the discussion in the mail thread, this would be a general
scheduler code change. Ideally would need test coverage from other
platforms.
-
[ Additional information ]
The offending commit 89802ca36c96 ("lib/group_cpus: make group CPU
cluster aware") was included in 7.0 kernel, so older kernel shouldn't
get impacted.
[1]
https://lore.kernel.org/all/[email protected]/
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2166862
Title:
[SRU] System hang after S0ix if enough of CPUs are offlined beforehand
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2166862/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
