[lkml]   [2022]   [Feb]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v2] KVM: Move VM's worker kthreads back to the original cgroups before exiting.
Hi Paolo, Michal

Will you accept a patch which uses real_parent in
kvm_vm_worker_thread() as suggested by Sean, while I figure out the
recommendation from Michal about making kthread_stop() wait on
cgroup_attach_task_all(current->real_parent, current)


On Thu, Jan 20, 2022 at 7:05 AM Michal Koutný <> wrote:
> On Wed, Jan 19, 2022 at 08:30:43AM -1000, Tejun Heo <> wrote:
> > It'd be nicer if we can make kthread_stop() waiting more regular but I
> > couldn't find a good existing place and routing the usual parent
> > signaling might be too complicated. Anyone has better ideas?
> The regular way is pictured in Paolo's diagram already, the
> exit_notify/do_signal_parent -> wait4 path.
> Actually, I can see that there exists already kernel_wait() and is used
> by a UMH wrapper kthread. kthreadd issues ignore_signals() so (besides
> no well defined point of signalling a kthread) the signal notification
> is moot and only waking up the waiter is relevant. So kthread_stop()
> could wait via kernel_wait() based on pid (extracted from task_struct).
> Have I missed an obstacle?

I must admit I do not have a good understanding of kernel_wait() and
kthread_stop() APIs. I tried making some changes in the kthread_stop()
but I was not able to successfully use the API. I tested it by a
writing a test module, where during the init I start a kthread which
prints some message every few seconds and during the module exit I
call kernel_stop(). This module worked as intended without the
kernel_wait() changes in the kthread_stop() API.

My changes were basically replacing wait_for_completion() with
kernel_wait() call.

@@ -645,8 +645,9 @@ int kthread_stop(struct task_struct *k)
set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
- wait_for_completion(&kthread->exited);
- ret = k->exit_code;
+ kernel_wait(k->pid, &ret);
+// kernel_wait(task_pid_vnr(k), &ret);
+// wait_for_completion(&kthread->exited);
+// ret = k->exit_code;

I used few other combination where I put kernel_wait() call after
put_task_struct(k) call.

Every time during the module exit, kernel was crashing like:

[ 285.014612] RIP: 0010:0xffffffffc04ed074
[ 285.018537] RSP: 0018:ffff9ccdc8365ee8 EFLAGS: 00010246
[ 285.023761] RAX: 0000000000000000 RBX: 0000000000000012 RCX: 0000000000000001
[ 285.030896] RDX: 0000000000000000 RSI: 0000000000000286 RDI: ffff9cce3f7d9cc0
[ 285.038028] RBP: ffff9ccdc8365ef8 R08: 0000000000000000 R09: 0000000000015504
[ 285.045160] R10: 000000000000004b R11: ffffffff8dd92880 R12: 0000000000000012
[ 285.052293] R13: ffff9ccdc813db90 R14: ffff9ccdc7e66240 R15: ffffffffc04ed000
[ 285.059425] FS: 0000000000000000(0000) GS:ffff9cce3f7c0000(0000)
[ 285.067510] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 285.073258] CR2: ffffffffc04ed074 CR3: 000000c07f20e002 CR4: 0000000000362ef0
[ 285.080390] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 285.087522] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 285.094656] Call Trace:
[ 285.097112] kthread+0x148/0x1b0
[ 285.100343] ? kthread_blkcg+0x30/0x30
[ 285.104096] ret_from_fork+0x3a/0x60
[ 285.107671] Code: Bad RIP value.
[ 285.107671] IP: 0xffffffffc04ecff4:

Crash is not observed if I keep wait_for_completion(&kthread->exited)
along with kernel_wait(), but I guess the kernel_wait() should be
sufficient by itself if I figure out proper way to use it.

Do you have any suggestions what might be the right way to use this API?


 \ /
  Last update: 2022-02-16 18:39    [W:0.235 / U:0.416 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site