Re: Perf lockups / stack overflows on v3.17-rc6, x86_64, arm, arm64

From: Vince Weaver
Date: Sun Oct 05 2014 - 01:07:37 EST


On Thu, 25 Sep 2014, Mark Rutland wrote:

> Log 1, x86_64 lockup
> [ 223.007005] [<ffffffff81168910>] ? poll_select_copy_remaining+0x130/0x130
> [ 223.007005] [<ffffffff811600ea>] ? getname_flags+0x4a/0x1a0
> [ 223.007005] [<ffffffff8116007d>] ? final_putname+0x1d/0x40
> [ 223.007005] [<ffffffff811602f4>] ? putname+0x24/0x40
> [ 223.007005] [<ffffffff8116581a>] ? user_path_at_empty+0x5a/0x90
> [ 223.007005] [<ffffffff810701c0>] ? wake_up_state+0x10/0x10
> [ 223.007005] [<ffffffff81198078>] ? eventfd_read+0x38/0x60
> [ 223.007005] [<ffffffff810a1e75>] ? ktime_get_ts64+0x45/0xf0
> [ 223.007005] [<ffffffff81169f00>] SyS_poll+0x60/0xf0

I have seen issues similar to this before, where the problem appeared
to be in poll/hrtimer. Never managed to track down anything useful about
the bug.

> Log 2, x86_64 stack overflow

> [ 346.641345] divide error: 0000 [#1] SMP
> [ 346.642010] Modules linked in:
> [ 346.642010] CPU: 0 PID: 4076 Comm: perf_fuzzer Not tainted 3.17.0-rc6hark-perf-lockup+ #1
> [ 346.642010] Hardware name: LENOVO 7484A3G/LENOVO, BIOS 5CKT54AUS 09/07/2009
> [ 346.642010] task: ffff8801ac449a70 ti: ffff8801ac574000 task.ti: ffff8801ac574000
> [ 346.642010] RIP: 0010:[<ffffffff81078bce>] [<ffffffff81078bce>] find_busiest_group+0x28e/0x8a0
> [ 346.642010] RSP: 0018:ffff8801ac577760 EFLAGS: 00010006
> [ 346.642010] RAX: 00000000000003ff RBX: 0000000000000000 RCX: 00000000ffff8801
> [ 346.642010] RDX: 0000000000000000 RSI: 0000000000000001 RDI: 0000000000000001
> [ 346.642010] RBP: ffff8801ac577890 R08: 0000000000000000 R09: 0000000000000000
> [ 346.704010] ------------[ cut here ]------------
> [ 346.704017] WARNING: CPU: 2 PID: 5 at arch/x86/kernel/irq_64.c:70 handle_irq+0x141/0x150()
> [ 346.704019] do_IRQ(): has overflown the kernel stack (cur:1,sp:ffff8801b653fe88,irq stk top-bottom:ffff8801bed00080-ffff8801bed03fc0,exception stk top-bottom:ffff8801bed04080-ffff8801bed0a000)

weird, have not seen this before. Though I was hitting a reboot issue
that would give really strange crash messages that was possibly fixed by
a patch that went into 3.17-rc7.

> Log 3, arm64 lockup
> ---->8----

> Seeding random number generator with 1411488270
> /proc/sys/kernel/perf_event_max_sample_rate currently: 285518974/s
> /proc/sys/kernel/perf_event_paranoid currently: 1142898651

Those last two lines are suspect. Is my fuzzer broken on arm64 somehow?
I do try to test on arm occasionally, but my pandaboard suffered massive
SD card failure and I haven't had a chance to get it back running yet.
I have trouble getting any of my other flock of arm machines running
recent upstream kernels.


Sorry that I don't have good answers for these bugs, but I will stick them
in my perf_fuzzer outstanding bugs list.

Vince
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/