[PATCH 2/2] arm64: uaccess: disable preempt during uaccess PSTATE.{PAN,TCO}

[Pingfan Liu]

On Wed, Dec 22, 2021 at 01:02:44PM +0000, Catalin Marinas wrote:
> On Tue, Dec 21, 2021 at 09:42:42PM +0800, Pingfan Liu wrote:
> > For the pair of
> > uaccess_enable_privileged()/uaccess_disable_privileged(), if preemption
> > happens, the exposure of the access to user space by another code piece
> > is unexpected.
> 
> How can this happen? We set PAN on exception entry.
> 

Yes, I make the some mistake as [1/2]

> > For the pair of
> > __uaccess_enable_tco_async()/__uaccess_disable_tco_async(), the change
> > of PSTATE.TCO is exposed unexpectedly.
> 
> As above, we disable TCO on exception entry.
> 

Yes.

> > For the two cases, simply disable preemption to fix the issue.
> > 
> > Signed-off-by: Pingfan Liu <kernelfans at gmail.com>
> > Cc: Catalin Marinas <catalin.marinas at arm.com>
> > Cc: Will Deacon <will at kernel.org>
> > Cc: Mark Rutland <mark.rutland at arm.com>
> > Cc: Vincenzo Frascino <vincenzo.frascino at arm.com>
> > Cc: Andrey Konovalov <andreyknvl at gmail.com>
> > Cc: Arnd Bergmann <arnd at arndb.de>
> > To: linux-arm-kernel at lists.infradead.org
> > ---
> >  arch/arm64/include/asm/uaccess.h | 15 +++++++++++----
> >  1 file changed, 11 insertions(+), 4 deletions(-)
> > 
> > diff --git a/arch/arm64/include/asm/uaccess.h b/arch/arm64/include/asm/uaccess.h
> > index 406888877bbd..37d5531f210d 100644
> > --- a/arch/arm64/include/asm/uaccess.h
> > +++ b/arch/arm64/include/asm/uaccess.h
> > @@ -193,14 +193,18 @@ static inline void __uaccess_enable_tco(void)
> >   */
> >  static inline void __uaccess_disable_tco_async(void)
> >  {
> > -	if (system_uses_mte_async_or_asymm_mode())
> > -		 __uaccess_disable_tco();
> > +	if (system_uses_mte_async_or_asymm_mode()) {
> > +		__uaccess_disable_tco();
> > +		preempt_enable();
> > +	}
> >  }
> >  
> >  static inline void __uaccess_enable_tco_async(void)
> >  {
> > -	if (system_uses_mte_async_or_asymm_mode())
> > +	if (system_uses_mte_async_or_asymm_mode()) {
> > +		preempt_disable();
> >  		__uaccess_enable_tco();
> > +	}
> >  }
> 
> As with the previous patch, disabling preemption prior to a
> might_sleep() is wrong.
> 

Right.
I will make clear the kernel_{entry/_exit} part. Appreciate for your
help again.

Thanks,

	Pingfan