On 05. 03. 26, 12:53, Jiri Slaby wrote:
On 05. 03. 26, 8:00, Jiri Slaby wrote:
On 02. 03. 26, 12:46, Peter Zijlstra wrote:
On Mon, Mar 02, 2026 at 06:28:38AM +0100, Jiri Slaby wrote:
The state of the lock:
crash> struct rq.__lock -x ffff8d1a6fd35dc0
__lock = {
raw_lock = {
{
val = {
counter = 0x40003
},
{
locked = 0x3,
pending = 0x0
},
{
locked_pending = 0x3,
tail = 0x4
}
}
}
},
That had me remember the below patch that never quite made it. I've
rebased it to something more recent so it applies.
If you stick that in, we might get a clue as to who is owning that lock.
Provided it all wants to reproduce well enough.
Thanks, I applied it, but to date it is still not accepted yet:
https://build.opensuse.org/requests/1335893
OK, I have a first dump with the patch applied:
__lock = {
raw_lock = {
{
val = {
counter = 0x2c0003
},
{
locked = 0x3,
pending = 0x0
},
{
locked_pending = 0x3,
tail = 0x2c
}
}
}
},
I am not sure if it is of any help?
BUT: I have another dump with LOCKDEP (but NOT the patch above). The
kernel is again spinning in mm_get_cid(), presumably waiting for a free
bit in the map as before [1]:
[ 162.660584] rcu: INFO: rcu_preempt detected stalls on CPUs/tasks:
...
[ 162.661378] Sending NMI from CPU 3 to CPUs 1:
[ 162.661398] NMI backtrace for cpu 1
...
[ 162.661411] RIP: 0010:mm_get_cid+0x54/0xc0
7680 is active on CPU 1:
PID: 7680 TASK: ffff8cc4038525c0 CPU: 1 COMMAND: "asm"
CPU3 is waiting for the CPU1's rq_lock:
RDX: 0000000000000000 RSI: 0000000000000003 RDI: ffff8cc72fcb8500
...
#3 [ffffd2e9c0083da0] raw_spin_rq_lock_nested+0x20 at ffffffff9339e700
crash> struct rq.__lock -x ffff8cc72fcb8500
__lock = {
raw_lock = {
{
val = {
counter = 0x100003
},
{
locked = 0x3,
pending = 0x0
},
{
locked_pending = 0x3,
tail = 0x10
}
}
},
magic = 0xdead4ead,
owner_cpu = 0x1,
owner = 0xffff8cc4038b8000,
dep_map = {
key = 0xffffffff96245970 <__key.7>,
class_cache = {0xffffffff9644b488 <lock_classes+10600>, 0x0},
name = 0xffffffff94ba3ab3 "&rq->__lock",
wait_type_outer = 0x0,
wait_type_inner = 0x2,
lock_type = 0x0
}
},
owner_cpu is 1, owner is:
PID: 7508 TASK: ffff8cc4038b8000 CPU: 1 COMMAND: "compile"
But as you can see above, CPU1 is occupied with a different task:
crash> bt -sxc 1
PID: 7680 TASK: ffff8cc4038525c0 CPU: 1 COMMAND: "asm"
spinning in mm_get_cid() as I wrote. See the objdump of mm_get_cid below.
You might be interested in mm_cid dumps:
====== PID 7508 (sleeping, holding the rq lock) ======
crash> task -R mm_cid -x 7508
PID: 7508 TASK: ffff8cc4038b8000 CPU: 1 COMMAND: "compile"
mm_cid = {
active = 0x1,
cid = 0x40000003
},
crash> p ((struct task_struct *)(0xffff8cc4038b8000))->mm->mm_cid|head -4
$6 = {
pcpu = 0x66222619df40,
mode = 1073741824,
max_cids = 4,
====== PID 7680 (spinning in mm_get_cid()) ======
crash> task -R mm_cid -x 7680
PID: 7680 TASK: ffff8cc4038525c0 CPU: 1 COMMAND: "asm"
mm_cid = {
active = 0x1,
cid = 0x80000000
},
crash> p ((struct task_struct *)(0xffff8cc4038b8000))->mm->mm_cid|head -4
$8 = {
pcpu = 0x66222619df40,
mode = 1073741824,
max_cids = 4,
====== per-cpu for CPU1 ======
crash> struct mm_cid_pcpu -x fffff2e9bfc89f40
struct mm_cid_pcpu {
cid = 0x40000003
}
Dump of any other's mm_cids needed?
[1] https://bugzilla.suse.com/show_bug.cgi?id=1258936#c17
ffffffff8139cd40 <mm_get_cid>:
mm_get_cid():
include/linux/cpumask.h:1020
ffffffff8139cd40: 8b 05 9a d7 40 02 mov
0x240d79a(%rip),%eax # ffffffff837aa4e0 <nr_cpu_ids>
kernel/sched/sched.h:3779
ffffffff8139cd46: 55 push %rbp
ffffffff8139cd47: 53 push %rbx
include/linux/mm_types.h:1477
ffffffff8139cd48: 48 8d 9f 80 0b 00 00 lea 0xb80(%rdi),%rbx
kernel/sched/sched.h:3780 (discriminator 2)
ffffffff8139cd4f: 8b b7 0c 01 00 00 mov 0x10c(%rdi),%esi
include/linux/cpumask.h:1020
ffffffff8139cd55: 83 c0 3f add $0x3f,%eax
ffffffff8139cd58: c1 e8 03 shr $0x3,%eax
kernel/sched/sched.h:3780 (discriminator 2)
ffffffff8139cd5b: 48 89 f5 mov %rsi,%rbp
include/linux/mm_types.h:1479 (discriminator 1)
ffffffff8139cd5e: 25 f8 ff ff 1f and $0x1ffffff8,%eax
include/linux/mm_types.h:1489 (discriminator 1)
ffffffff8139cd63: 48 8d 3c 43 lea (%rbx,%rax,2),%rdi
include/linux/find.h:393
ffffffff8139cd67: e8 44 d8 6e 00 call
ffffffff81a8a5b0 <_find_first_zero_bit>
kernel/sched/sched.h:3771
ffffffff8139cd6c: 39 e8 cmp %ebp,%eax
ffffffff8139cd6e: 73 7c jae
ffffffff8139cdec <mm_get_cid+0xac>
ffffffff8139cd70: 89 c1 mov %eax,%ecx
kernel/sched/sched.h:3773 (discriminator 1)
ffffffff8139cd72: 89 c2 mov %eax,%edx
include/linux/cpumask.h:1020
ffffffff8139cd74: 8b 05 66 d7 40 02 mov
0x240d766(%rip),%eax # ffffffff837aa4e0 <nr_cpu_ids>
ffffffff8139cd7a: 83 c0 3f add $0x3f,%eax
ffffffff8139cd7d: c1 e8 03 shr $0x3,%eax
include/linux/mm_types.h:1479 (discriminator 1)
ffffffff8139cd80: 25 f8 ff ff 1f and $0x1ffffff8,%eax
include/linux/mm_types.h:1489 (discriminator 1)
ffffffff8139cd85: 48 8d 04 43 lea (%rbx,%rax,2),%rax
arch/x86/include/asm/bitops.h:136
ffffffff8139cd89: f0 48 0f ab 10 lock bts %rdx,(%rax)
kernel/sched/sched.h:3773 (discriminator 2)
ffffffff8139cd8e: 73 4b jae
ffffffff8139cddb <mm_get_cid+0x9b>
ffffffff8139cd90: eb 5a jmp
ffffffff8139cdec <mm_get_cid+0xac>
arch/x86/include/asm/vdso/processor.h:13
ffffffff8139cd92: f3 90 pause
include/linux/cpumask.h:1020
ffffffff8139cd94: 8b 05 46 d7 40 02 mov
0x240d746(%rip),%eax # ffffffff837aa4e0 <nr_cpu_ids>
The CPU1 was caught by the NMI here ^^^^^^^^^^^^^^^^^^^^.
In the meantime, me and Michal K. did some digging into qemu dumps.
Details at (and a couple previous comments):
https://bugzilla.suse.com/show_bug.cgi?id=1258936#c17
tl;dr:
In one of the dumps, one process sits in
context_switch
-> mm_get_cid (before switch_to())
> 65 kworker/1:1 SP= 0xffffcf82c022fd98 -> __schedule+0x16ee
(ffffffff820f162e) -> call mm_get_cid
Michal extracted the vCPU's RIP and it turned out:
> Hm, I'd say the CPU could be spinning in mm_get_cid() waiting for a
free CID.
> ...
> ffff8a88458137c0: 000000000000000f 000000000000000f
> ^
> Hm, so indeed CIDs for all four CPUs are occupied.
To me (I don't know what CID is either), this might point as a
possible culprit to Thomas' "sched/mmcid: Cure mode transition woes" [1].
Funnily enough, 47ee94efccf6 ("sched/mmcid: Protect transition on
weakly ordered systems") spells:
> As a consequence the task will
> not drop the CID when scheduling out before the fixup is
completed, which
> means the CID space can be exhausted and the next task
scheduling in will
> loop in mm_get_cid() and the fixup thread can livelock on the
held runqueue
> lock as above.
Which sounds like what exactly happens here. Except the patch is from
the series above, so is already in 6.19 obviously.
I noticed there is also a 7.0-rc1 fix:
1e83ccd5921a sched/mmcid: Don't assume CID is CPU owned on mode switch
But that got into 6.19.1 already (we are at 6.19.3). So does not
improve the situation.
Any ideas?
[1] https://lore.kernel.org/all/[email protected]/
thanks,
--
js
suse labs
