On Tue, Feb 13, 2024 at 02:59:11PM -0800, Suren Baghdasaryan wrote:

On Tue, Feb 13, 2024 at 2:50 PM Kent Overstreet
[off-list ref] wrote:

quoted

On Tue, Feb 13, 2024 at 11:48:41PM +0100, David Hildenbrand wrote:

quoted

On 13.02.24 23:30, Suren Baghdasaryan wrote:

quoted

On Tue, Feb 13, 2024 at 2:17 PM David Hildenbrand [off-list ref] wrote:

If you think you can easily achieve what Michal requested without all that,
good.

He requested something?

Yes, a cleaner instrumentation. Unfortunately the cleanest one is not
possible until the compiler feature is developed and deployed. And it
still would require changes to the headers, so don't think it's worth
delaying the feature for years.

Hang on, let's look at the actual code.

This is what instrumenting an allocation function looks like:

#define krealloc_array(...)                     alloc_hooks(krealloc_array_noprof(__VA_ARGS__))

IOW, we have to:
 - rename krealloc_array to krealloc_array_noprof
 - replace krealloc_array with a one wrapper macro call

Is this really all we're getting worked up over?

The renaming we need regardless, because the thing that makes this
approach efficient enough to run in production is that we account at
_one_ point in the callstack, we don't save entire backtraces.

And thus we need to explicitly annotate which one that is; which means
we need _noprof() versions of functions for when the accounting is done
by an outer wraper (e.g. mempool).

And, as I keep saying: that alloc_hooks() macro will also get us _per
callsite fault injection points_, and we really need that because - if
you guys have been paying attention to other threads - whenever moving
more stuff to PF_MEMALLOC_* flags comes up (including adding
PF_MEMALLOC_NORECLAIM), the issue of small allocations not failing and
not being testable keeps coming up.