Re: [PATCH] x86/cpufeature: drop *_MASK_CEHCK

[Dave Hansen]

On 8/27/19 12:41 AM, Borislav Petkov wrote:
>>  #define DISABLED_MASK_BIT_SET(feature_bit)				\
>> @@ -104,7 +103,6 @@ extern const char * const x86_bug_flags[NBUGINTS*32];
>>  	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 16, feature_bit) ||	\
>>  	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 17, feature_bit) ||	\
>>  	   CHECK_BIT_IN_MASK_WORD(DISABLED_MASK, 18, feature_bit) ||	\
>> -	   DISABLED_MASK_CHECK					  ||	\
>>  	   BUILD_BUG_ON_ZERO(NCAPINTS != 19))
>>  
>>  #define cpu_has(c, bit)							\
> If you do a little bit of git archeology:
> 
> $ git annotate arch/x86/include/asm/cpufeature.h
> 
> after a while, you'll see that this was added in:
> 
> 1e61f78baf89 ("x86/cpufeature: Make sure DISABLED/REQUIRED macros are updated")
> 
> and then you could Cc Dave and ask what he was thinking then?
> 
> Leaving in the rest for reference.

The point was that there are 5 files in the code that need to be changed
if you change NCAPINTS:

1. arch/x86/include/asm/required-features.h
2. arch/x86/include/asm/disabled-features.h
3. tools/arch/x86/include/asm/disabled-features.h
4. tools/arch/x86/include/asm/required-features.h
5. arch/x86/include/asm/cpufeature.h (2 sites)

Each of those sites has a compile-time check for NCAPINTS.  The problem
is that the *-features.h code doesn't get compiled directly so a
BUILD_BUG_ON() doesn't work by itself.  So, for the sites there, we put
it somewhere that *will* get compiled: the macros that actually check
the bits.

It looks weird, but the end effect is good: If you change NCAPINTS, you
get compile errors in 5 files and have to go edit those 5 files to fix
it.  Your patch makes it easier to introduce errors and miss one of
those sites.