Re: [PATCH] reboot: Fix variable assignments in type_store

From: Matteo Croce <mcroce@linux.microsoft.com>
To: Petr Mladek <pmladek@suse.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	Nathan Chancellor <natechancellor@gmail.com>,
	Arnd Bergmann <arnd@arndb.de>, Kees Cook <keescook@chromium.org>,
	linux-kernel@vger.kernel.org, Guenter Roeck <linux@roeck-us.net>,
	Pavel Tatashin <pasha.tatashin@soleen.com>,
	Mike Rapoport <rppt@kernel.org>,
	Tyler Hicks <tyhicks@linux.microsoft.com>,
	Nick Desaulniers <ndesaulniers@google.com>,
	clang-built-linux@googlegroups.com
Subject: Re: [PATCH] reboot: Fix variable assignments in type_store
Date: Fri, 13 Nov 2020 22:28:18 +0100	[thread overview]
Message-ID: <CAFnufp3NBHUQ8fT1TEw5BvazGQjP9KmsU3y7ioJScWQ9qGiq1A@mail.gmail.com> (raw)
In-Reply-To: <20201113162043.GO1602@alley>

On Fri, Nov 13, 2020 at 9:06 PM Petr Mladek <pmladek@suse.com> wrote:
>
> On Fri 2020-11-13 03:58:49, Matteo Croce wrote:
> > On Fri, Nov 13, 2020 at 3:46 AM Andrew Morton <akpm@linux-foundation.org> wrote:
> > >
> > > On Fri, 13 Nov 2020 02:38:18 +0100 Matteo Croce <mcroce@linux.microsoft.com> wrote:
> > >
> > > > At this point, since 'pci' enables BOOT_CF9_FORCE type and
> > > > BOOT_CF9_SAFE is not user selectable, should I simply leave only
> > > > 'pci'?
> > > >
> > > > This way, we'll have the same set of options for both sysfs and kernel cmdline.
> > >
> > > Well, you're the reboot expert ;)
> > >
> >
> > So honored! :)
> >
> > > But my $0.02 is yes, let's keep the command-line and sysfs interfaces
> > > in sync and cover it all in documentation.  It would of course be
> > > problematic to change the existing reboot= interface.
> > >
> > > I assume that means doing this?
> > >
> > > - #define BOOT_CF9_FORCE_STR     "cf9_force"
> > > + #define BOOT_CF9_FORCE_STR     "pci"
> > > - #define BOOT_CF9_SAFE_STR      "cf9_safe"
> >
> > Either BOOT_PCI_STR or BOOT_CF9_FORCE_STR, I have no strong preference.
> >
> > The syntax is 'pci' while the enum BOOT_CF9_FORCE, so we can't please both.
>
> The question is whether we should modify/allow to set these values at
> all.
>
> Anyway, we must prevent them on non-x86 architectures because
> the reboot behavior would be undefined there. They could probably
> make a mess even on many x86-architectures.
>

That's right, but the same can be obtained by passing 'reboot=pci'  on
non x86 machines: the cmdline parsing is generic and will set
reboot_type on all arches.

> I have to admit it has become much more complicated than I thought.
> It brings back Andrew's original question whether this interface is
> really needed. Are you going to use in the real life?
>

Yes, there are some cases.
Not to mention complex use cases like let persist some memory regions,
or change the page size,
if a network driver fails to rmmod with the infamous
"unregistered_netdevice: waiting for wlan0 to become free",
enabling force on the fly allows to reboot the machine.

> The interface might do more harm then good when it allows to set
> reboot_type that is not normally accessible or disable it when
> it is strictly needed.
>

I looked at the reboot_type usage, there isn't any reference outside
arch/x86. In fact, the parameter is just ignored:

# uname -m
aarch64
# cat /proc/cmdline
console=ttyS0,115200n8 reboot=pci
# reboot -ff
Rebooting.
[   43.893833] reboot: Restarting system

The same applies for reboot_force, the only flags available on
different architectures are reboot_mode and reboot_cpu.
We could hide some handlers for some architectures. We save some
space, and avoid letting the user set flags which do nothing.

> Anyway, we should get input from some x86-experts about the BOOT_CF9
> values.
>

Sure, x86@kernel.org ?

Regards,
-- 
per aspera ad upstream