Re: [PATCH 2/7] x86/wait: prevent duplicated assembly labels

[Jan Beulich <jbeulich@xxxxxxxx>]

On 13.03.2025 16:30, Roger Pau Monne wrote:
> When enabling UBSAN with clang, the following error is triggered during the
> build:
> 
> common/wait.c:154:9: error: symbol '.L_wq_resume' is already defined
>   154 |         "push %%rbx; push %%rbp; push %%r12;"
>       |         ^
> <inline asm>:1:121: note: instantiated into assembly here
>     1 |         push %rbx; push %rbp; push %r12;push %r13; push %r14; push 
> %r15;sub %esp,%ecx;cmp $4096, %ecx;ja .L_skip;mov %rsp,%rsi;.L_wq_resume: rep 
> movsb;mov %rsp,%rsi;.L_skip:pop %r15; pop %r14; pop %r13;pop %r12; pop %rbp; 
> pop %rbx
>       |                                                                       
>                                                          ^
> common/wait.c:154:9: error: symbol '.L_skip' is already defined
>   154 |         "push %%rbx; push %%rbp; push %%r12;"
>       |         ^
> <inline asm>:1:159: note: instantiated into assembly here
>     1 |         push %rbx; push %rbp; push %r12;push %r13; push %r14; push 
> %r15;sub %esp,%ecx;cmp $4096, %ecx;ja .L_skip;mov %rsp,%rsi;.L_wq_resume: rep 
> movsb;mov %rsp,%rsi;.L_skip:pop %r15; pop %r14; pop %r13;pop %r12; pop %rbp; 
> pop %rbx
>       |                                                                       
>                                                                               
>                  ^
> 2 errors generated.
> 
> The inline assembly block in __prepare_to_wait() is duplicated, thus
> leading to multiple definitions of the otherwise unique labels inside the
> assembly block.  GCC extended-asm documentation notes the possibility of
> duplicating asm blocks:
> 
>> Under certain circumstances, GCC may duplicate (or remove duplicates of)
>> your assembly code when optimizing. This can lead to unexpected duplicate
>> symbol errors during compilation if your asm code defines symbols or
>> labels. Using ‘%=’ (see AssemblerTemplate) may help resolve this problem.
> 
> Move the assembly blocks that deal with saving and restoring the current
> CPU context into it's own explicitly non-inline functions.  This prevents
> clang from duplicating the assembly blocks.  Just using noinline attribute
> seems to be enough to prevent assembly duplication, in the future noclone
> might also be required if asm block duplication issues arise again.

Wouldn't it be a far easier / less intrusive change to simply append %= to
the label names?

Jan