Re: [PATCH v1 00/14] x86/alternatives: Nest them

[Borislav Petkov]

On June 11, 2024 3:26:09 PM GMT+02:00, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>On Fri, Jun 07, 2024 at 01:16:47PM +0200, Borislav Petkov wrote:
>
>> Borislav Petkov (AMD) (13):
>>   x86/alternative: Zap alternative_ternary()
>>   x86/alternative: Convert alternative()
>>   x86/alternative: Convert alternative_2()
>>   x86/alternative: Convert alternative_input()
>>   x86/alternative: Convert alternative_io()
>>   x86/alternative: Convert alternative_call()
>>   x86/alternative: Convert alternative_call_2()
>>   x86/alternative: Convert ALTERNATIVE_TERNARY()
>>   x86/alternative: Convert ALTERNATIVE_3()
>>   x86/alternative: Convert the asm ALTERNATIVE() macro
>>   x86/alternative: Convert the asm ALTERNATIVE_2() macro
>>   x86/alternative: Convert the asm ALTERNATIVE_3() macro
>>   x86/alternative: Replace the old macros
>> 
>> Peter Zijlstra (1):
>>   x86/alternatives: Add nested alternatives macros
>> 
>>  arch/x86/include/asm/alternative.h | 225 +++++++++--------------------
>>  arch/x86/kernel/alternative.c      |  20 ++-
>>  arch/x86/kernel/fpu/xstate.h       |  14 +-
>>  tools/objtool/arch/x86/special.c   |  23 +++
>>  tools/objtool/special.c            |  16 +-
>>  5 files changed, 125 insertions(+), 173 deletions(-)
>
>The whole back and forth with the n_foo things ia weird, but sure, have
>at.
>
>Acked-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>

Thanks, yeah it is in the 0th mail:

"For ease of bisection, the old macros are converted to the new, nested
variants in a step-by-step manner so that in case an issue is
encountered during testing, one can pinpoint the place where it fails
easier. Because debugging alternatives is a serious pain."

If you remember, your original patch had this issue where it would confuse two back-to-back .altinstr entries for nested because you were comparing only the offsets and they happened to be equal, by l

Now debugging *that* was insane and I managed to find the issue only when I started converting each place one-by-one and booting each one step.

That's why this superficial split is all about. We will appreciate it later. I hope not but if something explodes, bisection will be a lot easier this way.

I hope this makes sense.


-- 
Sent from a small device: formatting sucks and brevity is inevitable.