lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Wed, 11 Sep 2019 20:23:33 +0200
From:   Johannes Berg <johannes@...solutions.net>
To:     Kalle Valo <kvalo@...eaurora.org>
Cc:     Linus Torvalds <torvalds@...ux-foundation.org>,
        "David S. Miller" <davem@...emloft.net>,
        linux-wireless@...r.kernel.org, Netdev <netdev@...r.kernel.org>,
        Linux List Kernel Mailing <linux-kernel@...r.kernel.org>,
        ath10k@...ts.infradead.org
Subject: Re: WARNING at net/mac80211/sta_info.c:1057
 (__sta_info_destroy_part2())

On Wed, 2019-09-11 at 21:19 +0300, Kalle Valo wrote:
> > Looks like indeed the driver gives the device at least *3 seconds* for
> > every command, see ath10k_wmi_cmd_send(), so most likely this would
> > eventually have finished, but who knows how many firmware commands it
> > would still have attempted to send...
> 
> 3 seconds is a bit short but in normal cases it should be enough. Of
> course we could increase the delay but I'm skeptic it would help here.

I was thinking 3 seconds is way too long :-)

> > Perhaps the driver should mark the device as dead and fail quickly once
> > it timed out once, or so, but I'll let Kalle comment on that.
> 
> Actually we do try to restart the device when a timeout happens in
> ath10k_wmi_cmd_send():
> 
>         if (ret == -EAGAIN) {
>                 ath10k_warn(ar, "wmi command %d timeout, restarting hardware\n",
>                             cmd_id);
>                 queue_work(ar->workqueue, &ar->restart_work);
>         }

Yeah, and this is the problem, in a sense, I'd think. It seems to me
that at this point the code needs to tag the device as "dead" and
immediately return from any further commands submitted to it with an
error (e.g. -EIO). You can can actually see in the initial report that
while the restart was triggered, it too is waiting to acquire the RTNL:

>    Workqueue: events_freezable ieee80211_restart_work [mac80211]
>    Call Trace:
>     schedule+0x39/0xa0
>     schedule_preempt_disabled+0xa/0x10
>     __mutex_lock.isra.0+0x263/0x4b0
>     ieee80211_restart_work+0x54/0xe0 [mac80211]
>     process_one_work+0x1cf/0x370
>     worker_thread+0x4a/0x3c0
>     kthread+0xfb/0x130
>     ret_from_fork+0x35/0x40


So basically all this delay is mac80211 and the driver doing stuff with
the device, but every single thing has to time out and probably some
stuff loops etc., and then it just takes long enough with the RTNL held
that everything goes south.

johannes

Powered by blists - more mailing lists