Re: [PATCH] bcache: Make gc wakeup sane, remove set_task_state()

[Greg KH <greg@kroah.com>]

On Mon, Jan 23, 2017 at 10:45:47PM +0800, Coly Li wrote:
> On 2017/1/23 下午10:16, Greg KH wrote:
> > On Mon, Jan 23, 2017 at 09:20:12PM +0800, colyli@suse.de wrote:
> >> Hi stable maintainers,
> >>
> >> This patch is from Kent, upstream commit ID is be628be09563.
> >> Olav Reinert <seroton10@gmail.com> reports a kerenl crash from
> >> bcache (boo#1021260) and Oliver Nuekum points out this patch fixes the problem. 
> > 
> > "boo"?
> > 
> 
> Hi Greg,
> 
> 
> "boo" is abbreviation of bugzilla.opensuse.org, I paste the original bug
> report here,
> ==== start of bug report ==========
> I have starting seeing errors like the one quoted below in the system
> log. It occurs infrequently, but quite regularly, about 1-3 times a
> week, on a server running 24x7.
> 
> Around the time it began, I started running a beta version of Leap 42.2,
> upgraded from 42.1. Also, I enabled the "discard" option (SSD TRIM) on
> the bcache cache about 3-6 months ago. I believe one of those two events
> caused the bug to appear.
> 
> Not sure what other info is useful, please ask for whatever you need.
> 
> 
> Oct 10 00:00:02 blackbox kernel: ------------[ cut here ]------------
> Oct 10 00:00:02 blackbox kernel: WARNING: CPU: 4 PID: 1269 at
> ../kernel/sched/core.c:7891 __might_sleep+0x76/0x80()
> Oct 10 00:00:02 blackbox kernel: do not call blocking ops when
> !TASK_RUNNING; state=1 set at [<ffffffffa09e2325>]
> bch_gc_thread+0x25/0x100 [
> Oct 10 00:00:02 blackbox kernel: Modules linked in: vhost_net vhost
> macvtap macvlan fuse ebt_arp ebt_ip ebtable_nat ebtable_filter ebtables
> Oct 10 00:00:02 blackbox kernel:  mxm_wmi
> Oct 10 00:00:02 blackbox kernel:  bcache aesni_intel raid1
> snd_hda_codec_realtek aes_x86_64 lrw snd_hda_codec_generic gf128mul
> md_mod glue_h
> Oct 10 00:00:02 blackbox kernel:
> Oct 10 00:00:02 blackbox kernel: CPU: 4 PID: 1269 Comm: bcache_gc Not
> tainted 4.4.21-2-default #1
> Oct 10 00:00:02 blackbox kernel: Hardware name: To be filled by O.E.M.
> To be filled by O.E.M./M5A99X EVO R2.0, BIOS 2301 01/06/2014
> Oct 10 00:00:02 blackbox kernel:  0000000000000000 ffffffff81326967
> ffff8800b605be10 ffffffff81a5e431
> Oct 10 00:00:02 blackbox kernel:  ffffffff8107e7d1 ffffffff81a5f54f
> ffff8800b605be60 0000000000000061
> Oct 10 00:00:02 blackbox kernel:  0000000000000000
> Oct 10 00:00:02 blackbox kernel:  0000000000000000 ffffffff8107e84c
> ffffffff81a4ef88
> Oct 10 00:00:02 blackbox kernel: Call Trace:
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff81019e69>] dump_trace+0x59/0x320
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8101a22a>]
> show_stack_log_lvl+0xfa/0x180
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8101afd1>] show_stack+0x21/0x40
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff81326967>] dump_stack+0x5c/0x85
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8107e7d1>]
> warn_slowpath_common+0x81/0xb0
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8107e84c>]
> warn_slowpath_fmt+0x4c/0x50
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff810a3026>]
> __might_sleep+0x76/0x80
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff81605cac>] mutex_lock+0x1c/0x38
> Oct 10 00:00:02 blackbox kernel:  [<ffffffffa09e2365>]
> bch_gc_thread+0x65/0x100 [bcache]
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8109d268>] kthread+0xc8/0xe0
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8160828f>]
> ret_from_fork+0x3f/0x70
> Oct 10 00:00:02 blackbox kernel: DWARF2 unwinder stuck at
> ret_from_fork+0x3f/0x70
> Oct 10 00:00:02 blackbox kernel:
> Oct 10 00:00:02 blackbox kernel: Leftover inexact backtrace:
> Oct 10 00:00:02 blackbox kernel:  [<ffffffff8109d1a0>] ?
> kthread_park+0x50/0x50
> Oct 10 00:00:02 blackbox kernel: ---[ end trace c63abcb6c473e79b ]---
> ==== end of bug report ==========
> 
> 
> # journalctl|grep "blocking ops"
> Oct 10 00:00:02 blackbox kernel: do not call blocking ops when
> !TASK_RUNNING; state=1 set at [<ffffffffa09e2325>]
> bch_gc_thread+0x25/0x100 [bcache]
> [snip repeated lines]
> 
> 
> 
> >> I send this patch to stable@kernel.vger.org, hope this patch can be taken care
> >> in stable kernels.
> >>
> >> Thanks in advance.
> >>
> >> Coly Li
> >>
> >> Here I attach the original patch, just FYI.
> >> ---
> >> From: Kent Overstreet <kent.overstreet@gmail.com>
> >> Date: Wed, 26 Oct 2016 20:31:17 -0700
> >> Subject: [PATCH] bcache: Make gc wakeup sane, remove set_task_state()
> >>
> >> Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
> > 
> > No changelog text?  Worst short changelog description ever?
> 
> There is no change log from original patch, I am not the author, and it
> is in upstream already. So I think I am not the right person to change
> its commit log.

Oh, I didn't mean to complain to you, my complain was to Kent.

Kent, please go read the section, "The canonical patch format" in the
Documentation/SubmittingPatches file for how to do this properly.

> This is the first time I encounter this situation, that send a patch to
> stable which is not from me. I guess Kent does not notice that this
> patch indeed fixes a kernel oops. But it does fix a bug report for Leap
> 42.2 and SLE12-SP2.
> > 
> > This gives me no context of what is going on here.  Why does this fix a
> > bug?  What kernel(s) should it be backported to?
> > 
> 
> The bug is reported on Linux 4.4 based kernel, so at least all kernels
> since Linux 4.4 should have the fix. Maybe Kent can provide more
> accurate suggestion.

Kent, any hints?

thanks,

greg k-h