From: Nic Ferrier <nferrier@ta...> - 2006-02-07 23:43:36
Maybe related to Adam's problem.
With certain simple disc accesses I am getting:
soft lockup detected on CPU0
When it happens is very strange... right now I'm getting this when I'm
redirecting stdin to a file (namely the ssh authorized_keys file) like
cat > .ssh/authorized_keys
I get about 100 bytes into the file and then the UML hangs for a while
and then spits this out:
BUG: soft lockup detected on CPU#0!
EIP: 0073:[<a0198e0a>] CPU: 0 Not tainted ESP: 007b:a08b3984 EFLAGS: 00000246
EAX: 00000000 EBX: 00000001 ECX: a08b39b4 EDX: a08b39b4
ESI: 00000008 EDI: 00000001 EBP: a08b398c DS: 007b ES: 007b
a08b3618: [<a0031bf3>] show_regs+0x113/0x140
a08b3638: [<a0056bb7>] softlockup_tick+0x57/0x60
a08b3658: [<a0041627>] do_timer+0x47/0xd0
a08b3668: [<a001a627>] boot_timer_handler+0x17/0x20
a08b36a8: [<ffffe420>] _etext+0x5fe18976/0x0
a08b3990: [<a0019755>] set_signals+0x45/0x130
a08b3a50: [<a00492ec>] __queue_work+0x5c/0x70
a08b3a70: [<a004932b>] queue_work+0x2b/0x70
a08b3a90: [<a0049b72>] schedule_work+0x12/0x20
a08b3aa0: [<a0020aa5>] chan_interrupt+0x135/0x140
a08b3ad0: [<a0020fb0>] line_interrupt+0x30/0x40
a08b3af0: [<a0020fe7>] line_timer_cb+0x27/0x30
a08b3b10: [<a0049642>] worker_thread+0x212/0x2f0
a08b3bc0: [<a004d7d9>] kthread+0xb9/0xc0
a08b3c00: [<a002d969>] run_kernel_thread+0x49/0x50
a08b3cd0: [<a001d5cb>] new_thread_handler+0x14b/0x180
a08b3d20: [<ffffe420>] _etext+0x5fe18976/0x0
However, I can do all sorts of other things, I quite happily installed
rsync via apt for example.
At first it looks like a disc IO error, but it occurs to me as I write
this that it might be an console handling error. My console is on
From: Jeff Dike <jdike@ad...> - 2006-02-08 18:20:41
On Tue, Feb 07, 2006 at 11:22:12PM +0000, Nic Ferrier wrote:
> Maybe related to Adam's problem.
I don't think so. I see this on I/O workloads, but not a total hang.
> BUG: soft lockup detected on CPU#0!
> Any ideas?
As far as I've been able to tell, UML is spending so much time in the host
kernel that > 10 seconds can pass between timer ticks (which are
supposed to happen at 100 HZ while UML is running in userspace).
The host is doing heavy I/O at the time, but I don't see that excusing
a 1000-fold slowdown in timer ticks.
BB made a proposal a while ago to use SIGPROF instead of SIGVTALRM
since SIGPROF ticks while in the kernel. I think this is a good idea,
but at this point, I think it would just cover over the real problem.