Re: [PATCH 2/3] alloc_tag: uninline code gated by mem_alloc_profiling_key in slab allocator

[Peter Zijlstra]

On Sun, Jan 26, 2025 at 05:47:08PM +0100, Vlastimil Babka wrote:
> On 1/26/25 08:02, Suren Baghdasaryan wrote:
> > When a sizable code section is protected by a disabled static key, that
> > code gets into the instruction cache even though it's not executed and
> > consumes the cache, increasing cache misses. This can be remedied by
> > moving such code into a separate uninlined function. The improvement
> 
> Weird, I thought the static_branch_likely/unlikely/maybe was already
> handling this by the unlikely case being a jump to a block away from the
> fast-path stream of instructions, thus making it less likely to get cached.
> AFAIU even plain likely()/unlikely() should do this, along with branch
> prediction hints.

Very much depends on the compiler :-(

sometimes unlikely just moves it to the end of the function, sometimes
it's moved to .text.unlikely.

Some compilers have label attributes:

l_yes: __attribute__((cold));

but the same compilers utterly ignore it when it's combined with
asm-goto or something -- we could never get it to work reliably.


It's been a while since I looked at this, so I'm not entirely sure what
the current version of compilers do.