[PATCH] x86/entry: Fix AC assertion

[PATCH] x86/entry: Fix AC assertion

[peterz]

The WARN added in commit 3c73b81a9164 ("x86/entry, selftests: Further
improve user entry sanity checks") unconditionally triggers on my IVB
machine because it does not support SMAP.

For !SMAP hardware we patch out CLAC/STAC instructions and thus if
userspace sets AC, we'll still have it set after entry.

Fixes: 3c73b81a9164 ("x86/entry, selftests: Further improve user entry sanity checks")
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Andy Lutomirski <luto@kernel.org>
---
 arch/x86/include/asm/entry-common.h |   11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

--- a/arch/x86/include/asm/entry-common.h
+++ b/arch/x86/include/asm/entry-common.h
@@ -18,8 +18,15 @@ static __always_inline void arch_check_u
 		 * state, not the interrupt state as imagined by Xen.
 		 */
 		unsigned long flags = native_save_fl();
-		WARN_ON_ONCE(flags & (X86_EFLAGS_AC | X86_EFLAGS_DF |
-				      X86_EFLAGS_NT));
+		unsigned long mask = X86_EFLAGS_DF | X86_EFLAGS_NT;
+
+		/*
+		 * For !SMAP hardware we patch out CLAC on entry.
+		 */
+		if (boot_cpu_has(X86_FEATURE_SMAP))
+			mask |= X86_EFLAGS_AC;
+
+		WARN_ON_ONCE(flags & mask);
 
 		/* We think we came from user mode. Make sure pt_regs agrees. */
 		WARN_ON_ONCE(!user_mode(regs));

Re: [PATCH] x86/entry: Fix AC assertion

[Andrew Cooper]

On 24/08/2020 11:14, peterz@infradead.org wrote:
> The WARN added in commit 3c73b81a9164 ("x86/entry, selftests: Further
> improve user entry sanity checks") unconditionally triggers on my IVB
> machine because it does not support SMAP.
>
> For !SMAP hardware we patch out CLAC/STAC instructions and thus if
> userspace sets AC, we'll still have it set after entry.

Technically, you don't patch in, rather than patch out.

>
> Fixes: 3c73b81a9164 ("x86/entry, selftests: Further improve user entry sanity checks")
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> Acked-by: Andy Lutomirski <luto@kernel.org>
> ---
>  arch/x86/include/asm/entry-common.h |   11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
>
> --- a/arch/x86/include/asm/entry-common.h
> +++ b/arch/x86/include/asm/entry-common.h
> @@ -18,8 +18,15 @@ static __always_inline void arch_check_u
>  		 * state, not the interrupt state as imagined by Xen.
>  		 */
>  		unsigned long flags = native_save_fl();
> -		WARN_ON_ONCE(flags & (X86_EFLAGS_AC | X86_EFLAGS_DF |
> -				      X86_EFLAGS_NT));
> +		unsigned long mask = X86_EFLAGS_DF | X86_EFLAGS_NT;
> +
> +		/*
> +		 * For !SMAP hardware we patch out CLAC on entry.
> +		 */
> +		if (boot_cpu_has(X86_FEATURE_SMAP))
> +			mask |= X86_EFLAGS_AC;

The Xen PV ABI clears AC on entry for 64bit guests, because Linux is
actually running in Ring 3, and therefore susceptible to #AC's which
wouldn't occur natively.

~Andrew

Re: [PATCH] x86/entry: Fix AC assertion

[peterz]

On Mon, Aug 24, 2020 at 03:22:06PM +0100, Andrew Cooper wrote:
> On 24/08/2020 11:14, peterz@infradead.org wrote:
> > The WARN added in commit 3c73b81a9164 ("x86/entry, selftests: Further
> > improve user entry sanity checks") unconditionally triggers on my IVB
> > machine because it does not support SMAP.
> >
> > For !SMAP hardware we patch out CLAC/STAC instructions and thus if
> > userspace sets AC, we'll still have it set after entry.
> 
> Technically, you don't patch in, rather than patch out.

True.

> > Fixes: 3c73b81a9164 ("x86/entry, selftests: Further improve user entry sanity checks")
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > Acked-by: Andy Lutomirski <luto@kernel.org>
> > ---
> >  arch/x86/include/asm/entry-common.h |   11 +++++++++--
> >  1 file changed, 9 insertions(+), 2 deletions(-)
> >
> > --- a/arch/x86/include/asm/entry-common.h
> > +++ b/arch/x86/include/asm/entry-common.h
> > @@ -18,8 +18,15 @@ static __always_inline void arch_check_u
> >  		 * state, not the interrupt state as imagined by Xen.
> >  		 */
> >  		unsigned long flags = native_save_fl();
> > -		WARN_ON_ONCE(flags & (X86_EFLAGS_AC | X86_EFLAGS_DF |
> > -				      X86_EFLAGS_NT));
> > +		unsigned long mask = X86_EFLAGS_DF | X86_EFLAGS_NT;
> > +
> > +		/*
> > +		 * For !SMAP hardware we patch out CLAC on entry.
> > +		 */
> > +		if (boot_cpu_has(X86_FEATURE_SMAP))
> > +			mask |= X86_EFLAGS_AC;
> 
> The Xen PV ABI clears AC on entry for 64bit guests, because Linux is
> actually running in Ring 3, and therefore susceptible to #AC's which
> wouldn't occur natively.

So do you then want it to be something like:

	if (boot_cpu_has(X86_FEATURE_SMAP) ||
	    (IS_ENABLED(CONFIG_64_BIT) && boot_cpu_has(X86_FEATURE_XENPV)))

? Or are you fine with the proposed?

Re: [PATCH] x86/entry: Fix AC assertion

[Andrew Cooper]

On 24/08/2020 16:21, peterz@infradead.org wrote:
> On Mon, Aug 24, 2020 at 03:22:06PM +0100, Andrew Cooper wrote:
>> On 24/08/2020 11:14, peterz@infradead.org wrote:
>>> The WARN added in commit 3c73b81a9164 ("x86/entry, selftests: Further
>>> improve user entry sanity checks") unconditionally triggers on my IVB
>>> machine because it does not support SMAP.
>>>
>>> For !SMAP hardware we patch out CLAC/STAC instructions and thus if
>>> userspace sets AC, we'll still have it set after entry.
>> Technically, you don't patch in, rather than patch out.
> True.
>
>>> Fixes: 3c73b81a9164 ("x86/entry, selftests: Further improve user entry sanity checks")
>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>> Acked-by: Andy Lutomirski <luto@kernel.org>
>>> ---
>>>  arch/x86/include/asm/entry-common.h |   11 +++++++++--
>>>  1 file changed, 9 insertions(+), 2 deletions(-)
>>>
>>> --- a/arch/x86/include/asm/entry-common.h
>>> +++ b/arch/x86/include/asm/entry-common.h
>>> @@ -18,8 +18,15 @@ static __always_inline void arch_check_u
>>>  		 * state, not the interrupt state as imagined by Xen.
>>>  		 */
>>>  		unsigned long flags = native_save_fl();
>>> -		WARN_ON_ONCE(flags & (X86_EFLAGS_AC | X86_EFLAGS_DF |
>>> -				      X86_EFLAGS_NT));
>>> +		unsigned long mask = X86_EFLAGS_DF | X86_EFLAGS_NT;
>>> +
>>> +		/*
>>> +		 * For !SMAP hardware we patch out CLAC on entry.
>>> +		 */
>>> +		if (boot_cpu_has(X86_FEATURE_SMAP))
>>> +			mask |= X86_EFLAGS_AC;
>> The Xen PV ABI clears AC on entry for 64bit guests, because Linux is
>> actually running in Ring 3, and therefore susceptible to #AC's which
>> wouldn't occur natively.
> So do you then want it to be something like:
>
> 	if (boot_cpu_has(X86_FEATURE_SMAP) ||
> 	    (IS_ENABLED(CONFIG_64_BIT) && boot_cpu_has(X86_FEATURE_XENPV)))
>
> ? Or are you fine with the proposed?

Dealers choice, but this option would be slightly better overall.

(Are there any other cases where Linux will be running in Ring 3?  I
haven't been paying attention to recent changes in PVOps.)

~Andrew

Re: [PATCH] x86/entry: Fix AC assertion

[Jürgen Groß]

On 24.08.20 17:58, Andrew Cooper wrote:
> On 24/08/2020 16:21, peterz@infradead.org wrote:
>> On Mon, Aug 24, 2020 at 03:22:06PM +0100, Andrew Cooper wrote:
>>> On 24/08/2020 11:14, peterz@infradead.org wrote:
>>>> The WARN added in commit 3c73b81a9164 ("x86/entry, selftests: Further
>>>> improve user entry sanity checks") unconditionally triggers on my IVB
>>>> machine because it does not support SMAP.
>>>>
>>>> For !SMAP hardware we patch out CLAC/STAC instructions and thus if
>>>> userspace sets AC, we'll still have it set after entry.
>>> Technically, you don't patch in, rather than patch out.
>> True.
>>
>>>> Fixes: 3c73b81a9164 ("x86/entry, selftests: Further improve user entry sanity checks")
>>>> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>>>> Acked-by: Andy Lutomirski <luto@kernel.org>
>>>> ---
>>>>   arch/x86/include/asm/entry-common.h |   11 +++++++++--
>>>>   1 file changed, 9 insertions(+), 2 deletions(-)
>>>>
>>>> --- a/arch/x86/include/asm/entry-common.h
>>>> +++ b/arch/x86/include/asm/entry-common.h
>>>> @@ -18,8 +18,15 @@ static __always_inline void arch_check_u
>>>>   		 * state, not the interrupt state as imagined by Xen.
>>>>   		 */
>>>>   		unsigned long flags = native_save_fl();
>>>> -		WARN_ON_ONCE(flags & (X86_EFLAGS_AC | X86_EFLAGS_DF |
>>>> -				      X86_EFLAGS_NT));
>>>> +		unsigned long mask = X86_EFLAGS_DF | X86_EFLAGS_NT;
>>>> +
>>>> +		/*
>>>> +		 * For !SMAP hardware we patch out CLAC on entry.
>>>> +		 */
>>>> +		if (boot_cpu_has(X86_FEATURE_SMAP))
>>>> +			mask |= X86_EFLAGS_AC;
>>> The Xen PV ABI clears AC on entry for 64bit guests, because Linux is
>>> actually running in Ring 3, and therefore susceptible to #AC's which
>>> wouldn't occur natively.
>> So do you then want it to be something like:
>>
>> 	if (boot_cpu_has(X86_FEATURE_SMAP) ||
>> 	    (IS_ENABLED(CONFIG_64_BIT) && boot_cpu_has(X86_FEATURE_XENPV)))
>>
>> ? Or are you fine with the proposed?
> 
> Dealers choice, but this option would be slightly better overall.
> 
> (Are there any other cases where Linux will be running in Ring 3?  I
> haven't been paying attention to recent changes in PVOps.)

I'm not aware of any other case running kernel code in ring 3.


Juergen