Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Sean Christopherson]

On Wed, Mar 25, 2020 at 08:30:00AM -0700, Randy Dunlap wrote:
> On 3/25/20 1:53 AM, Stephen Rothwell wrote:
> > Hi all,
> > 
> > Changes since 20200324:
> > 
> 
> 
> on i386 randconfig build:
> and gcc 7.5.0:
> 
> 24 (only showing one of them here) BUILD_BUG() errors in arch/x86/kvm/cpuid.h
> function __cpuid_entry_get_reg(), for the default: case.

I'll take a gander at this.

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Paolo Bonzini]

On 25/03/20 16:30, Randy Dunlap wrote:
> 24 (only showing one of them here) BUILD_BUG() errors in arch/x86/kvm/cpuid.h
> function __cpuid_entry_get_reg(), for the default: case.
> 
> 
>   CC      arch/x86/kvm/cpuid.o
> In file included from ../include/linux/export.h:43:0,
>                  from ../include/linux/linkage.h:7,
>                  from ../include/linux/preempt.h:10,
>                  from ../include/linux/hardirq.h:5,
>                  from ../include/linux/kvm_host.h:7,
>                  from ../arch/x86/kvm/cpuid.c:12:
> In function ‘__cpuid_entry_get_reg’,
>     inlined from ‘kvm_cpu_cap_mask’ at ../arch/x86/kvm/cpuid.c:272:25,
>     inlined from ‘kvm_set_cpu_caps’ at ../arch/x86/kvm/cpuid.c:292:2:
> ../include/linux/compiler.h:394:38: error: call to ‘__compiletime_assert_114’ declared with attribute error: BUILD_BUG failed
>   _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
>                                       ^
> ../include/linux/compiler.h:375:4: note: in definition of macro ‘__compiletime_assert’
>     prefix ## suffix();    \
>     ^~~~~~
> ../include/linux/compiler.h:394:2: note: in expansion of macro ‘_compiletime_assert’
>   _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
>   ^~~~~~~~~~~~~~~~~~~
> ../include/linux/build_bug.h:39:37: note: in expansion of macro ‘compiletime_assert’
>  #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
>                                      ^~~~~~~~~~~~~~~~~~
> ../include/linux/build_bug.h:59:21: note: in expansion of macro ‘BUILD_BUG_ON_MSG’
>  #define BUILD_BUG() BUILD_BUG_ON_MSG(1, "BUILD_BUG failed")
>                      ^~~~~~~~~~~~~~~~
> ../arch/x86/kvm/cpuid.h:114:3: note: in expansion of macro ‘BUILD_BUG’
>    BUILD_BUG();
>    ^~~~~~~~~
> 

Looks like the compiler is not smart enough to figure out the constant 
expressions in BUILD_BUG.  I think we need to do something like this:

diff --git a/arch/x86/kvm/cpuid.h b/arch/x86/kvm/cpuid.h
index 23b4cd1ad986..8f711b0cdec0 100644
--- a/arch/x86/kvm/cpuid.h
+++ b/arch/x86/kvm/cpuid.h
@@ -40,6 +40,7 @@ struct cpuid_reg {
 	int reg;
 };
 
+/* Update reverse_cpuid_check as well when adding an entry.  */
 static const struct cpuid_reg reverse_cpuid[] = {
 	[CPUID_1_EDX]         = {         1, 0, CPUID_EDX},
 	[CPUID_8000_0001_EDX] = {0x80000001, 0, CPUID_EDX},
@@ -68,12 +69,21 @@ static const struct cpuid_reg reverse_cpuid[] = {
  */
 static __always_inline void reverse_cpuid_check(unsigned int x86_leaf)
 {
-	BUILD_BUG_ON(x86_leaf == CPUID_LNX_1);
-	BUILD_BUG_ON(x86_leaf == CPUID_LNX_2);
-	BUILD_BUG_ON(x86_leaf == CPUID_LNX_3);
-	BUILD_BUG_ON(x86_leaf == CPUID_LNX_4);
-	BUILD_BUG_ON(x86_leaf >= ARRAY_SIZE(reverse_cpuid));
-	BUILD_BUG_ON(reverse_cpuid[x86_leaf].function == 0);
+	BUILD_BUG_ON(x86_leaf != CPUID_1_EDX &&
+	             x86_leaf != CPUID_8000_0001_EDX &&
+	             x86_leaf != CPUID_8086_0001_EDX &&
+	             x86_leaf != CPUID_1_ECX &&
+	             x86_leaf != CPUID_C000_0001_EDX &&
+	             x86_leaf != CPUID_8000_0001_ECX &&
+	             x86_leaf != CPUID_7_0_EBX &&
+	             x86_leaf != CPUID_D_1_EAX &&
+	             x86_leaf != CPUID_8000_0008_EBX &&
+	             x86_leaf != CPUID_6_EAX &&
+	             x86_leaf != CPUID_8000_000A_EDX &&
+	             x86_leaf != CPUID_7_ECX &&
+	             x86_leaf != CPUID_8000_0007_EBX &&
+	             x86_leaf != CPUID_7_EDX &&
+	             x86_leaf != CPUID_7_1_EAX);
 }
 
 /*

Randy, can you test it with your compiler?

Thanks,

Paolo

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Randy Dunlap]

On 3/25/20 8:46 AM, Paolo Bonzini wrote:
> On 25/03/20 16:30, Randy Dunlap wrote:
>> 24 (only showing one of them here) BUILD_BUG() errors in arch/x86/kvm/cpuid.h
>> function __cpuid_entry_get_reg(), for the default: case.
>>
>>
>>   CC      arch/x86/kvm/cpuid.o
>> In file included from ../include/linux/export.h:43:0,
>>                  from ../include/linux/linkage.h:7,
>>                  from ../include/linux/preempt.h:10,
>>                  from ../include/linux/hardirq.h:5,
>>                  from ../include/linux/kvm_host.h:7,
>>                  from ../arch/x86/kvm/cpuid.c:12:
>> In function ‘__cpuid_entry_get_reg’,
>>     inlined from ‘kvm_cpu_cap_mask’ at ../arch/x86/kvm/cpuid.c:272:25,
>>     inlined from ‘kvm_set_cpu_caps’ at ../arch/x86/kvm/cpuid.c:292:2:
>> ../include/linux/compiler.h:394:38: error: call to ‘__compiletime_assert_114’ declared with attribute error: BUILD_BUG failed
>>   _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
>>                                       ^
>> ../include/linux/compiler.h:375:4: note: in definition of macro ‘__compiletime_assert’
>>     prefix ## suffix();    \
>>     ^~~~~~
>> ../include/linux/compiler.h:394:2: note: in expansion of macro ‘_compiletime_assert’
>>   _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
>>   ^~~~~~~~~~~~~~~~~~~
>> ../include/linux/build_bug.h:39:37: note: in expansion of macro ‘compiletime_assert’
>>  #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
>>                                      ^~~~~~~~~~~~~~~~~~
>> ../include/linux/build_bug.h:59:21: note: in expansion of macro ‘BUILD_BUG_ON_MSG’
>>  #define BUILD_BUG() BUILD_BUG_ON_MSG(1, "BUILD_BUG failed")
>>                      ^~~~~~~~~~~~~~~~
>> ../arch/x86/kvm/cpuid.h:114:3: note: in expansion of macro ‘BUILD_BUG’
>>    BUILD_BUG();
>>    ^~~~~~~~~
>>
> 
> Looks like the compiler is not smart enough to figure out the constant 
> expressions in BUILD_BUG.  I think we need to do something like this:
> 
> diff --git a/arch/x86/kvm/cpuid.h b/arch/x86/kvm/cpuid.h
> index 23b4cd1ad986..8f711b0cdec0 100644
> --- a/arch/x86/kvm/cpuid.h
> +++ b/arch/x86/kvm/cpuid.h
> @@ -40,6 +40,7 @@ struct cpuid_reg {
>  	int reg;
>  };
>  
> +/* Update reverse_cpuid_check as well when adding an entry.  */
>  static const struct cpuid_reg reverse_cpuid[] = {
>  	[CPUID_1_EDX]         = {         1, 0, CPUID_EDX},
>  	[CPUID_8000_0001_EDX] = {0x80000001, 0, CPUID_EDX},
> @@ -68,12 +69,21 @@ static const struct cpuid_reg reverse_cpuid[] = {
>   */
>  static __always_inline void reverse_cpuid_check(unsigned int x86_leaf)
>  {
> -	BUILD_BUG_ON(x86_leaf == CPUID_LNX_1);
> -	BUILD_BUG_ON(x86_leaf == CPUID_LNX_2);
> -	BUILD_BUG_ON(x86_leaf == CPUID_LNX_3);
> -	BUILD_BUG_ON(x86_leaf == CPUID_LNX_4);
> -	BUILD_BUG_ON(x86_leaf >= ARRAY_SIZE(reverse_cpuid));
> -	BUILD_BUG_ON(reverse_cpuid[x86_leaf].function == 0);
> +	BUILD_BUG_ON(x86_leaf != CPUID_1_EDX &&
> +	             x86_leaf != CPUID_8000_0001_EDX &&
> +	             x86_leaf != CPUID_8086_0001_EDX &&
> +	             x86_leaf != CPUID_1_ECX &&
> +	             x86_leaf != CPUID_C000_0001_EDX &&
> +	             x86_leaf != CPUID_8000_0001_ECX &&
> +	             x86_leaf != CPUID_7_0_EBX &&
> +	             x86_leaf != CPUID_D_1_EAX &&
> +	             x86_leaf != CPUID_8000_0008_EBX &&
> +	             x86_leaf != CPUID_6_EAX &&
> +	             x86_leaf != CPUID_8000_000A_EDX &&
> +	             x86_leaf != CPUID_7_ECX &&
> +	             x86_leaf != CPUID_8000_0007_EBX &&
> +	             x86_leaf != CPUID_7_EDX &&
> +	             x86_leaf != CPUID_7_1_EAX);
>  }
>  
>  /*
> 
> Randy, can you test it with your compiler?

Nope, no help.  That's the wrong location.
Need a patch for this:
>> 24 (only showing one of them here) BUILD_BUG() errors in arch/x86/kvm/cpuid.h
>> function __cpuid_entry_get_reg(), for the default: case.


-- 
~Randy

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Paolo Bonzini]

On 25/03/20 16:57, Randy Dunlap wrote:
>> Randy, can you test it with your compiler?
> Nope, no help.  That's the wrong location.
> Need a patch for this:
>>> 24 (only showing one of them here) BUILD_BUG() errors in arch/x86/kvm/cpuid.h
>>> function __cpuid_entry_get_reg(), for the default: case.

Doh, right.  I think the only solution for that one is to degrade it to
WARN_ON(1).

Paolo

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Sean Christopherson]

On Wed, Mar 25, 2020 at 05:08:03PM +0100, Paolo Bonzini wrote:
> On 25/03/20 16:57, Randy Dunlap wrote:
> >> Randy, can you test it with your compiler?
> > Nope, no help.  That's the wrong location.
> > Need a patch for this:
> >>> 24 (only showing one of them here) BUILD_BUG() errors in arch/x86/kvm/cpuid.h
> >>> function __cpuid_entry_get_reg(), for the default: case.
> 
> Doh, right.  I think the only solution for that one is to degrade it to
> WARN_ON(1).

I reproduced the error, give me a bit to play with the code to see if the
BUILD_BUG can be preserved.  I'm curious as to why kvm_cpu_cap_mask() is
special, and why it only fails with this config.

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Paolo Bonzini]

On 25/03/20 17:14, Sean Christopherson wrote:
>> Doh, right.  I think the only solution for that one is to degrade it to
>> WARN_ON(1).
> I reproduced the error, give me a bit to play with the code to see if the
> BUILD_BUG can be preserved.  I'm curious as to why kvm_cpu_cap_mask() is
> special, and why it only fails with this config.
> 

I could not reproduce it, but I would not be surprised if there are
other configurations where the compiler cannot constant-propagate from
the reverse_cpuid struct into __cpuid_entry_get_reg.

Paolo

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Sean Christopherson]

On Wed, Mar 25, 2020 at 05:26:20PM +0100, Paolo Bonzini wrote:
> On 25/03/20 17:14, Sean Christopherson wrote:
> >> Doh, right.  I think the only solution for that one is to degrade it to
> >> WARN_ON(1).
> > I reproduced the error, give me a bit to play with the code to see if the
> > BUILD_BUG can be preserved.  I'm curious as to why kvm_cpu_cap_mask() is
> > special, and why it only fails with this config.
> > 
> 
> I could not reproduce it, but I would not be surprised if there are
> other configurations where the compiler cannot constant-propagate from
> the reverse_cpuid struct into __cpuid_entry_get_reg.

The error is related to UBSAN.  There is at least one legitimate (but benign)
underlying issue.  I'm chasing down a second instance of the BUILD_BUG.

Assuming all issues can be fixed, I think it'd make sense to keep the
BUILD_BUG, especially if it's teasing out actual weirdness, even if the
weirdness is benign.

Re: linux-next: Tree for Mar 25 (arch/x86/kvm/)

[Sean Christopherson]

On Wed, Mar 25, 2020 at 09:46:06AM -0700, Sean Christopherson wrote:
> On Wed, Mar 25, 2020 at 05:26:20PM +0100, Paolo Bonzini wrote:
> > On 25/03/20 17:14, Sean Christopherson wrote:
> > >> Doh, right.  I think the only solution for that one is to degrade it to
> > >> WARN_ON(1).
> > > I reproduced the error, give me a bit to play with the code to see if the
> > > BUILD_BUG can be preserved.  I'm curious as to why kvm_cpu_cap_mask() is
> > > special, and why it only fails with this config.
> > > 
> > 
> > I could not reproduce it, but I would not be surprised if there are
> > other configurations where the compiler cannot constant-propagate from
> > the reverse_cpuid struct into __cpuid_entry_get_reg.
> 
> The error is related to UBSAN.  There is at least one legitimate (but benign)
> underlying issue.  I'm chasing down a second instance of the BUILD_BUG.

Argh, red herring.  There is no underlying issue other than gcc tripping up
when -fsanitize=alignment is enabled by UBSAN.  Good news is that the build
error can be fixed without resorting to a hack.