INFO: rcu_preempt detected expedited stalls on CPUs/tasks (6.1.0-rc3): in cat /sys/kernel/debug/kmemleak

[Mirsad Goran Todorovac <mirsad.todorovac@alu.unizg.hr>]

[-- Attachment #1: Type: text/plain, Size: 5019 bytes --]

Dear all,

When investigating thermald kmemleak, it occurred that the "cat 
/sys/kernel/debug/kmemleak"
and "tail -20 /sys/kernel/debug/kmemleak" commands take unusual amount 
of time.

Dmesg output showed expedited stalls that the commands caused NMIs and 
NMI backtraces:

[ 8123.263464] rcu: INFO: rcu_preempt detected expedited stalls on 
CPUs/tasks: { 0-.... } 26 jiffies s: 3829 root: 0x1/.
[ 8123.263500] rcu: blocking rcu_node structures (internal RCU debug):
[ 8123.263508] Sending NMI from CPU 7 to CPUs 0:
[ 8123.263528] NMI backtrace for cpu 0
[ 8123.263539] CPU: 0 PID: 27898 Comm: cat Not tainted 6.1.0-rc3 #1
[ 8123.263552] Hardware name: LENOVO 82H8/LNVNB161216, BIOS GGCN34WW 
03/08/2022
[ 8123.263557] RIP: 0010:kmemleak_seq_start+0x41/0x80
[ 8123.263579] Code: 55 04 a6 00 4c 63 e0 85 c0 78 40 e8 b9 80 db ff 48 
8b 05 92 fb 88 01 4c 8d 60 f8 48 3d 30
                                     62 63 92 75 17 eb 32 49 8b 44 24 08 
<48> 83 eb 01 4c 8d 60 f8 48 3d 30 62 63 92 74 1d
                                     48 85 db 7f e6 4c
[ 8123.263588] RSP: 0018:ffff9968e400fc30 EFLAGS: 00000206
[ 8123.263598] RAX: ffff8963b7005b58 RBX: 0000000000011cb8 RCX: 
0000000000000001
[ 8123.263604] RDX: ffff8963b09c4000 RSI: ffff8963856de028 RDI: 
ffffffff926361c0
[ 8123.263608] RBP: ffff9968e400fc40 R08: 0000000000020000 R09: 
ffff896380592a80
[ 8123.263613] R10: 0000000000020000 R11: 0000000000000000 R12: 
ffff8964114c2390
[ 8123.263617] R13: ffff89639fa25b00 R14: ffff8963856de000 R15: 
ffff9968e400fe30
[ 8123.263622] FS:  00007ff217c15740(0000) GS:ffff896528800000(0000) 
knlGS:0000000000000000
[ 8123.263630] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 8123.263636] CR2: 00007f19bd6a8000 CR3: 000000028e392001 CR4: 
0000000000770ef0
[ 8123.263642] PKRU: 55555554
[ 8123.263646] Call Trace:
[ 8123.263649]  <TASK>
[ 8123.263656]  seq_read_iter+0x169/0x420
[ 8123.263671]  seq_read+0xad/0xe0
[ 8123.263685]  full_proxy_read+0x59/0x90
[ 8123.263701]  vfs_read+0xb2/0x2e0
[ 8123.263718]  ksys_read+0x61/0xe0
[ 8123.263730]  __x64_sys_read+0x1a/0x20
[ 8123.263741]  do_syscall_64+0x58/0x80
[ 8123.263754]  ? do_syscall_64+0x67/0x80
[ 8123.263767]  ? exit_to_user_mode_prepare+0x15d/0x190
[ 8123.263781]  ? syscall_exit_to_user_mode+0x1b/0x30
[ 8123.263791]  ? do_syscall_64+0x67/0x80
[ 8123.263804]  ? syscall_exit_to_user_mode+0x1b/0x30
[ 8123.263813]  ? do_syscall_64+0x67/0x80
[ 8123.263823]  ? do_syscall_64+0x67/0x80
[ 8123.263833]  ? do_syscall_64+0x67/0x80
[ 8123.263844]  entry_SYSCALL_64_after_hwframe+0x63/0xcd
[ 8123.263857] RIP: 0033:0x7ff217914992
[ 8123.263867] Code: c0 e9 b2 fe ff ff 50 48 8d 3d fa b2 0c 00 e8 c5 1d 
02 00 0f 1f 44 00 00 f3 0f
                                     1e fa 64 8b 04 25 18 00 00 00 85 c0 
75 10 0f 05 <48> 3d 00 f0 ff ff 77 56
                                     c3 0f 1f 44 00 00 48 83 ec 28 48 89 
54 24
[ 8123.263875] RSP: 002b:00007ffdfcbc3e28 EFLAGS: 00000246 ORIG_RAX: 
0000000000000000
[ 8123.263894] RAX: ffffffffffffffda RBX: 0000000000020000 RCX: 
00007ff217914992
[ 8123.263899] RDX: 0000000000020000 RSI: 00007ff217b9d000 RDI: 
0000000000000003
[ 8123.263904] RBP: 00007ff217b9d000 R08: 00007ff217b9c010 R09: 
00007ff217b9c010
[ 8123.263909] R10: 0000000000000022 R11: 0000000000000246 R12: 
0000000000022000
[ 8123.263914] R13: 0000000000000003 R14: 0000000000020000 R15: 
0000000000020000
[ 8123.263927]  </TASK>

To reproduce:

Enable CONFIG_DEBUG_KMEMLEAK=y in linux_stable 6.1.0-rc3 build.

Then, a stress test on a service known from the previous report is required:

for a in {1..1000}; doRIP: 0010:kmemleak_seq_start+0x41/0x80
          echo $a
          systemctl stop thermald
          sleep 0.5
          systemctl start thermald
          sleep 0.5
done

After that, /sys/kernel/debug/kmemleak indicated 1413 unreferenced objects.

However, dmesg had shown the stalls on CPUs while executing "cat" or 
"tail -40" of /sys/kernel/debug/kmemleak.

I've read the 
https://www.kernel.org/doc/Documentation/RCU/stallwarn.txt, but I was 
unable to understand
what is going on except to locate the stall to "RIP: 
0010:kmemleak_seq_start+0x41/0x80" and probable
maintainer/developer.

Please find attached the kernel build config, the output of "cat 
/sys/kernel/debug/kmemleak" and
dmesg ouput about the expedited stalls. The number of jiffies is not so 
large, but IMHO they possibly indicate
greater problems or deadlocks in RCU code in kmemleak.

NOTE:
Please add:

Reported-By: Mirsad Goran Todorovac <mirsad.todorovac@alu.unizg.hr>

if (or when) the bug is fixed.

Thank you.

-mt

--
Mirsad Goran Todorovac
Sistem inženjer
Grafički fakultet | Akademija likovnih umjetnosti
Sveučilište u Zagrebu
-- 
System engineer
Faculty of Graphic Arts | Academy of Fine Arts
University of Zagreb, Republic of Croatia
The European Union

[-- Attachment #2: config-6.1.0-rc3.gz --]
[-- Type: application/gzip, Size: 65705 bytes --]

[-- Attachment #3: expedited_stalls_cat_tail_kmemleak.dmesg.gz --]
[-- Type: application/gzip, Size: 24792 bytes --]

[-- Attachment #4: kmemleak-thermald.txt.gz --]
[-- Type: application/gzip, Size: 31677 bytes --]