Date: Fri, 17 May 2024 21:37:44 +0300
From: Kalle Valo <kvalo@...nel.org>
To: Dave Hansen <dave.hansen@...el.com>
Cc: Borislav Petkov <bp@...en8.de>,  Pawan Gupta
 <pawan.kumar.gupta@...ux.intel.com>,  Thomas Gleixner
 <tglx@...utronix.de>,  Ingo Molnar <mingo@...hat.com>,  Dave Hansen
 <dave.hansen@...ux.intel.com>,  "Rafael J. Wysocki" <rafael@...nel.org>,
  x86@...nel.org,  linux-pm@...r.kernel.org,  linux-kernel@...r.kernel.org,
  regressions@...ts.linux.dev,  Jeff Johnson <quic_jjohnson@...cinc.com>
Subject: Re: [regression] suspend stress test stalls within 30 minutes

Dave Hansen <dave.hansen@...el.com> writes:

> On 5/17/24 10:15, Kalle Valo wrote:
>> Borislav Petkov <bp@...en8.de> writes:
>>> There might be some #GP or so in the logs in case we've managed to f*ck
>>> up microcode application which emulates that IBRS MSR bit and the
>>> actual toggling or so when suspending...
>> So the weird part is that when the bug happens (ie. suspend stalls) I
>> can access the box normally using ssh and I don't see anything special
>> in dmesg. Below is a full copy of dmesg output after the suspend
>> stalled. Do note that I copied this dmesg before I updated microcode so
>> it will still show the old microcode version.
>> 
>> Let me know if you need more info.
>
> Kalle, could you remind us what we're seeing here?  Does this show 30
> working rtcwake tests followed by a failure at "rtcwake test 31" where
> the system failed to suspend?

Correct. So basically what I do is that I start the nuc box, ssh into it
and run:

sudo su
for i in {1..400}; do echo "rtcwake test $i" > /dev/kmsg; rtcwake -m mem -s 10; sleep 10; done

Here's the start of first loop:

[   54.945105] rtcwake test 1
[   55.162603] PM: suspend entry (deep)
[   55.168875] Filesystems sync: 0.006 seconds
[   55.182427] Freezing user space processes
[   55.191498] Freezing user space processes completed (elapsed 0.008 seconds)
[   55.191711] OOM killer disabled.
[   55.191805] Freezing remaining freezable tasks
[   55.193507] Freezing remaining freezable tasks completed (elapsed 0.001 seconds)
[   55.194056] printk: Suspending console(s) (use no_console_suspend to debug)
[   55.244962] e1000e: EEE TX LPI TIMER: 00000011

Now I leave the box to run it's test. I come back later to see that the
for loop has stalled and the box is not going into suspend gain. I ssh
into the machine and see this in dmesg:

[  449.061525] rtcwake test 31
[  449.176854] PM: suspend entry (deep)
[  449.179072] Filesystems sync: 0.002 seconds
[  632.961545] loop0: detected capacity change from 0 to 8
[  637.003835] gpu-manager (6735) used greatest stack depth: 23808 bytes left
[  738.799026] kworker/dying (87) used greatest stack depth: 23488 bytes left
[  932.951032] loop0: detected capacity change from 0 to 8
[ 1232.962610] loop0: detected capacity change from 0 to 8

The system tried go into suspend but after the "Filesystems sync:"
message nothing happened for 10 minutes. I assume loop0 messages are
from some Ubuntu daemon, maybe snapd or similar. I have seen them
always, they are not specific to this issue.

And now comes the really strange part: if I run  'shutdown -h now' the
suspend continues apparently normally. Afterwards I checked from
/var/log/kern.log and didn't see any errors:

May 17 13:34:38 nuc2 kernel: [  449.176854] PM: suspend entry (deep)
May 17 13:34:38 nuc2 kernel: [  449.179072] Filesystems sync: 0.002 seconds
May 17 13:37:42 nuc2 kernel: [  632.961545] loop0: detected capacity change from 0 to 8
May 17 13:37:46 nuc2 kernel: [  637.003835] gpu-manager (6735) used greatest stack depth: 23808 bytes left
May 17 13:39:28 nuc2 kernel: [  738.799026] kworker/dying (87) used greatest stack depth: 23488 bytes left
May 17 13:42:42 nuc2 kernel: [  932.951032] loop0: detected capacity change from 0 to 8
May 17 13:47:42 nuc2 kernel: [ 1232.962610] loop0: detected capacity change from 0 to 8
May 17 13:52:45 nuc2 kernel: [ 1527.307800] Freezing user space processes
May 17 13:52:45 nuc2 kernel: [ 1527.334585] Freezing user space processes completed (elapsed 0.024 seconds)
May 17 13:52:45 nuc2 kernel: [ 1527.336094] OOM killer disabled.
May 17 13:52:45 nuc2 kernel: [ 1527.337562] Freezing remaining freezable tasks
May 17 13:52:45 nuc2 kernel: [ 1527.340324] Freezing remaining freezable tasks completed (elapsed 0.001 seconds)
May 17 13:52:45 nuc2 kernel: [ 1527.342596] printk: Suspending console(s) (use no_console_suspend to debug)
May 17 13:52:45 nuc2 kernel: [ 1527.380121] e1000e: EEE TX LPI TIMER: 00000011
May 17 13:52:45 nuc2 kernel: [ 1527.474981] ACPI: EC: interrupt blocked
May 17 13:52:45 nuc2 kernel: [ 1527.540696] ACPI: PM: Preparing to enter system sleep state S3
May 17 13:52:45 nuc2 kernel: [ 1527.567302] ACPI: EC: event blocked
May 17 13:52:45 nuc2 kernel: [ 1527.567307] ACPI: EC: EC stopped
May 17 13:52:45 nuc2 kernel: [ 1527.567311] ACPI: PM: Saving platform NVS memory
May 17 13:52:45 nuc2 kernel: [ 1527.567412] Disabling non-boot CPUs ...

While writing this email I found another way to continue the suspend
after a stall: terminate rtcwake with CTRL-C in the ssh session running
the for loop. That explains why 'sudo shutdown -h now' makes the suspend
go forward, it most likely kills the stalled rtcwake process.

Powered by blists - more mailing lists