Re: [PATCH 4.9 18/22] x86/fpu: Disable bottom halves while loading FPU registers

[Sasha Levin <sashal@kernel.org>]

On Fri, Jul 24, 2020 at 07:44:37PM +0200, Greg Kroah-Hartman wrote:
>On Fri, Jul 24, 2020 at 07:07:06PM +0200, Jan Kiszka wrote:
>> On 28.12.18 12:52, Greg Kroah-Hartman wrote:
>> > 4.9-stable review patch.  If anyone has any objections, please let me know.
>> >
>> > ------------------
>> >
>> > From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
>> >
>> > commit 68239654acafe6aad5a3c1dc7237e60accfebc03 upstream.
>> >
>> > The sequence
>> >
>> >    fpu->initialized = 1;		/* step A */
>> >    preempt_disable();		/* step B */
>> >    fpu__restore(fpu);
>> >    preempt_enable();
>> >
>> > in __fpu__restore_sig() is racy in regard to a context switch.
>> >
>> > For 32bit frames, __fpu__restore_sig() prepares the FPU state within
>> > fpu->state. To ensure that a context switch (switch_fpu_prepare() in
>> > particular) does not modify fpu->state it uses fpu__drop() which sets
>> > fpu->initialized to 0.
>> >
>> > After fpu->initialized is cleared, the CPU's FPU state is not saved
>> > to fpu->state during a context switch. The new state is loaded via
>> > fpu__restore(). It gets loaded into fpu->state from userland and
>> > ensured it is sane. fpu->initialized is then set to 1 in order to avoid
>> > fpu__initialize() doing anything (overwrite the new state) which is part
>> > of fpu__restore().
>> >
>> > A context switch between step A and B above would save CPU's current FPU
>> > registers to fpu->state and overwrite the newly prepared state. This
>> > looks like a tiny race window but the Kernel Test Robot reported this
>> > back in 2016 while we had lazy FPU support. Borislav Petkov made the
>> > link between that report and another patch that has been posted. Since
>> > the removal of the lazy FPU support, this race goes unnoticed because
>> > the warning has been removed.
>> >
>> > Disable bottom halves around the restore sequence to avoid the race. BH
>> > need to be disabled because BH is allowed to run (even with preemption
>> > disabled) and might invoke kernel_fpu_begin() by doing IPsec.
>> >
>> >   [ bp: massage commit message a bit. ]
>> >
>> > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
>> > Signed-off-by: Borislav Petkov <bp@suse.de>
>> > Acked-by: Ingo Molnar <mingo@kernel.org>
>> > Acked-by: Thomas Gleixner <tglx@linutronix.de>
>> > Cc: Andy Lutomirski <luto@kernel.org>
>> > Cc: Dave Hansen <dave.hansen@linux.intel.com>
>> > Cc: "H. Peter Anvin" <hpa@zytor.com>
>> > Cc: "Jason A. Donenfeld" <Jason@zx2c4.com>
>> > Cc: kvm ML <kvm@vger.kernel.org>
>> > Cc: Paolo Bonzini <pbonzini@redhat.com>
>> > Cc: Radim Krčmář <rkrcmar@redhat.com>
>> > Cc: Rik van Riel <riel@surriel.com>
>> > Cc: stable@vger.kernel.org
>> > Cc: x86-ml <x86@kernel.org>
>> > Link: http://lkml.kernel.org/r/20181120102635.ddv3fvavxajjlfqk@linutronix.de
>> > Link: https://lkml.kernel.org/r/20160226074940.GA28911@pd.tnic
>> > Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
>> > Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
>> > ---
>> >   arch/x86/kernel/fpu/signal.c |    4 ++--
>> >   1 file changed, 2 insertions(+), 2 deletions(-)
>> >
>> > --- a/arch/x86/kernel/fpu/signal.c
>> > +++ b/arch/x86/kernel/fpu/signal.c
>> > @@ -342,10 +342,10 @@ static int __fpu__restore_sig(void __use
>> >   			sanitize_restored_xstate(tsk, &env, xfeatures, fx_only);
>> >   		}
>> > +		local_bh_disable();
>> >   		fpu->fpstate_active = 1;
>> > -		preempt_disable();
>> >   		fpu__restore(fpu);
>> > -		preempt_enable();
>> > +		local_bh_enable();
>> >   		return err;
>> >   	} else {
>> >
>> >
>>
>> Any reason why the backport stopped back than at 4.9? I just debugged this
>> out of a 4.4 kernel, and it is needed there as well. I'm happy to propose a
>> backport, would just appreciate a hint if the BH protection is needed also
>> there (my case was without BH).
>
>You are asking about something we did back in 2018.  I can't remember
>what I did last week :)
>
>If you provide a backport that works, I'll be glad to take it.  The
>current patch does not apply cleanly there at all.

The conflict was due to a missing rename back in 4.4: e4a81bfcaae1
("x86/fpu: Rename fpu::fpstate_active to fpu::initialized").

I've fixed up the patch and queued it for 4.4, thanks for pointing it
out Jan!

-- 
Thanks,
Sasha