Re: [PATCH 0/3] firmware: Fix usermodehelper deadlock at shutdown

[Takashi Iwai <tiwai@suse.de>]

At Wed, 8 May 2013 20:46:43 +0200,
Kay Sievers wrote:
> 
> On Wed, May 8, 2013 at 6:26 PM, Takashi Iwai <tiwai@suse.de> wrote:
> > At Thu, 9 May 2013 00:07:17 +0800,
> > Ming Lei wrote:
> 
> >> On Wed, May 8, 2013 at 2:56 PM, Takashi Iwai <tiwai@suse.de> wrote:
> >> > Hi,
> >> >
> >> > this is a series of patches for the issue we faced in the firmware
> >> > loader code during debugging the problem with dell_rbu driver with
> >> > 3.9 kernel.
> >> >
> >> > The original problem was that the shutdown gets stuck when DELL BIOS
> >> > update is performed.  This turned out to be a problem in the firmware
> >> > loader.  Although the reason of dell_rbu driver breakage is still
> >>
> >> Sorry, from these patchset, I can't see why it is a problem in firmware.
> >>
> >> > unclear, we should fix the firmware loader side, at least, not to
> >> > stall during shutdown.
> >>
> >> Firstly you need to describe what/why is the stall? In fact, firmware
> >> loading can't stall forever and it will timeout, but the current 60sec
> >> timeout might be too long.
> >
> > The timeout check is activated only when uevent flag is set, and
> > dell_rbu driver doesn't set it explicitly (because it's not supposed
> > to be handled via udev or whatever).
> 
> These use the firmware loader not in the way the interface was intended:
>   drivers/misc/lattice-ecp3-config.c
>   drivers/firmware/dell_rbu.c
> 
> They just use the mechanism without any of the usual userspace setup.
> It's really a nasty hack to hijack the interface that way.
> 
> The commit 6e3eaab02028c4087a92711b20abb9e72cc803a7 is a pretty broken
> idea to start with. If something triggers uevents during runtime which
> is not uncommon, these on-demand silently created firmware devices
> would get really confused and race against the udev firmware loader
> which cancels the events.
> 
> As if the userspace firmware loading in general wasn't a bad enough
> idea already. :)

Luckily there are only few users, so conversion of such known drivers
should be easy.  But the major question is whether we are allowed to
drop the non-hotplug f/w loading right now after years.  It means a
clear user-space breakage, and we don't want it as much as possible.


Takashi