Skip to contentSkip to conversation

On resume from suspend only CPU 0 comes back on-line [REGRESSION][BISECTED]

DISCUSSION Doug Smythies 3 participants 6 messages May 14, 2015 – May 15, 2015

lore
Share & tools
lore
#0 lore raw

As of, or about, Kernel 4.1RC1 on resume from suspend only CPU 0 comes back on-line. The issue persists through Kernel 4.1RC3.
This is on my test computer with an i7-2600K.
I do not normally use suspend on this computer, but was doing so while working on a bug report.

The kernel was bisected, and this is the result:

3c18d447b3b36a8d3c90dc37dfbd363cdb685d0a is the first bad commit commit 3c18d447b3b36a8d3c90dc37dfbd363cdb685d0a

Author: Juri Lelli <[email protected]>
Date:   Tue Mar 31 09:53:37 2015 +0100

sched/core: Check for available DL bandwidth in cpuset_cpu_inactive()

Hotplug operations are destructive w.r.t. cpusets. In case such an operation is performed on a CPU belonging to an exlusive cpuset, the DL bandwidth information associated with the corresponding root domain is gone even if the operation fails (in sched_cpu_inactive()).

For this reason we need to move the check we currently have in sched_cpu_inactive() to cpuset_cpu_inactive() to prevent useless cpusets reconfiguration in the CPU_DOWN_FAILED path.

:040000 040000 10f8d81afdc8e625f8e6720883d3eb42c28d452b c08264528890941bad35d5d4cc134c03f259c534 M kernel

Since I sometimes mess up using git bisect, and end up at some random result, the above was double checked manually:

3c18d447b3b36a8d3c90dc37dfbd363cdb685d0a has the issue. 4cd57f97135840f637431c92380c8da3edbe44ed (the previous commit) does not have the issue.

#1 lore raw

Hi Doug,

Quoted 7 lines · Doug Smythies

As of, or about, Kernel 4.1RC1 on resume from suspend only CPU 0 comes back on-line. The issue persists through Kernel 4.1RC3.
This is on my test computer with an i7-2600K.
I do not normally use suspend on this computer, but was doing so while working on a bug report.

The kernel was bisected, and this is the result:

Does commit 533445c6e533 "sched/core: Fix regression in cpuset_cpu_inactive() for suspend" on tip/sched/core fix the bug?

Thanks,

- Juri

Quoted 31 lines

3c18d447b3b36a8d3c90dc37dfbd363cdb685d0a is the first bad commit commit 3c18d447b3b36a8d3c90dc37dfbd363cdb685d0a

Author: Juri Lelli <[email protected]>
Date:   Tue Mar 31 09:53:37 2015 +0100

sched/core: Check for available DL bandwidth in cpuset_cpu_inactive()

Hotplug operations are destructive w.r.t. cpusets. In case such an operation is performed on a CPU belonging to an exlusive cpuset, the DL bandwidth information associated with the corresponding root domain is gone even if the operation fails (in sched_cpu_inactive()).

For this reason we need to move the check we currently have in sched_cpu_inactive() to cpuset_cpu_inactive() to prevent useless cpusets reconfiguration in the CPU_DOWN_FAILED path.

:040000 040000 10f8d81afdc8e625f8e6720883d3eb42c28d452b c08264528890941bad35d5d4cc134c03f259c534 M kernel

Since I sometimes mess up using git bisect, and end up at some random result, the above was double checked manually:

3c18d447b3b36a8d3c90dc37dfbd363cdb685d0a has the issue. 4cd57f97135840f637431c92380c8da3edbe44ed (the previous commit) does not have the issue.

#2 lore raw
Quoted 8 lines · Juri Lelli

Hi Doug,

Quoted 5 lines · Doug Smythies

As of, or about, Kernel 4.1RC1 on resume from suspend only CPU 0 comes back on-line. The issue persists through Kernel 4.1RC3.
This is on my test computer with an i7-2600K.
I do not normally use suspend on this computer, but was doing so while working on a bug report.

Quoted 6 lines
Quoted 2 lines

The kernel was bisected, and this is the result:

Does commit 533445c6e533 "sched/core: Fix regression in cpuset_cpu_inactive() for suspend" on tip/sched/core fix the bug?

That would be sched/urgent primarily, not sched/core:

git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent

I'll get this fix to Linus ASAP.

Thanks,

Ingo

#3 lore raw
Quoted 10 lines · On 2015.05.14 10:48 Ingo Molnar wrote:
Quoted 9 lines · Juri Lelli
Quoted 8 lines · Doug Smythies

As of, or about, Kernel 4.1RC1 on resume from suspend only CPU 0 comes back on-line. The issue persists through Kernel 4.1RC3.
This is on my test computer with an i7-2600K.
I do not normally use suspend on this computer, but was doing so while working on a bug report.

The kernel was bisected, and this is the result:

Quoted 7 lines
Quoted 3 lines

Does commit 533445c6e533 "sched/core: Fix regression in cpuset_cpu_inactive() for suspend" on tip/sched/core fix the bug?

That would be sched/urgent primarily, not sched/core:
git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent I'll get this fix to Linus ASAP.

I could not seem to figure out how to get the patch from tip, and any version I could find on internet would not apply properly.

Juri kindly sent me the patch.

With the patch applied to an otherwise unmodified Kernel 4.1RC3, Resume from suspend is working fine again.

Thanks.

#4 lore raw
Quoted 16 lines · Doug Smythies
Quoted 13 lines · On 2015.05.14 10:48 Ingo Molnar wrote:
Quoted 12 lines · Juri Lelli
Quoted 8 lines · Doug Smythies

As of, or about, Kernel 4.1RC1 on resume from suspend only CPU 0 comes back on-line. The issue persists through Kernel 4.1RC3.
This is on my test computer with an i7-2600K.
I do not normally use suspend on this computer, but was doing so while working on a bug report.

The kernel was bisected, and this is the result:

Does commit 533445c6e533 "sched/core: Fix regression in cpuset_cpu_inactive() for suspend" on tip/sched/core fix the bug?

Quoted 6 lines
Quoted 3 lines

That would be sched/urgent primarily, not sched/core:
git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent I'll get this fix to Linus ASAP.

I could not seem to figure out how to get the patch from tip, and any version I could find on internet would not apply properly.

So the way you could have picked it up is something like:

  # pick up 533445c6e533
  git fetch git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent
  git cherry-pick 533445c6e533

or pick up all pending scheduler fixes, amongst them 533445c6e533:

git pull git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent

Or to export just that single one as a patch:

  git fetch git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent
  git log --pretty=email --stat -p -1 533445c6e533
Quoted 4 lines

Juri kindly sent me the patch.

With the patch applied to an otherwise unmodified Kernel 4.1RC3, Resume from suspend is working fine again.

Great, thanks for testing!

Ingo

#5 lore raw
In reply to Doug Smythies
Quoted 20 lines · Doug Smythies
Quoted 10 lines · On 2015.05.14 10:48 Ingo Molnar wrote:
Quoted 9 lines · Juri Lelli
Quoted 8 lines · Doug Smythies

As of, or about, Kernel 4.1RC1 on resume from suspend only CPU 0 comes back on-line. The issue persists through Kernel 4.1RC3.
This is on my test computer with an i7-2600K.
I do not normally use suspend on this computer, but was doing so while working on a bug report.

The kernel was bisected, and this is the result:

Quoted 7 lines
Quoted 3 lines

Does commit 533445c6e533 "sched/core: Fix regression in cpuset_cpu_inactive() for suspend" on tip/sched/core fix the bug?

That would be sched/urgent primarily, not sched/core:
git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git sched/urgent I'll get this fix to Linus ASAP.

Quoted 8 lines

I could not seem to figure out how to get the patch from tip, and any version I could find on internet would not apply properly.

Juri kindly sent me the patch.

With the patch applied to an otherwise unmodified Kernel 4.1RC3, Resume from suspend is working fine again.

Great! Thanks for testing :).

Best,

- Juri