On Wed, Jul 31, 2024 at 5:50 AM Yunsheng Lin [off-list ref] wrote:

Basing on the lib/objpool.c, change it to something like a
ptrpool, so that we can utilize that to test the correctness
and performance of the page_frag.

The testing is done by ensuring that the fragment allocated
from a frag_frag_cache instance is pushed into a ptrpool
instance in a kthread binded to a specified cpu, and a kthread
binded to a specified cpu will pop the fragment from the
ptrpool and free the fragment.

We may refactor out the common part between objpool and ptrpool
if this ptrpool thing turns out to be helpful for other place.

This isn't a patch where you should be introducing stuff you hope to
refactor out and reuse later. Your objpoo/ptrpool stuff is just going
to add bloat and overhead as you are going to have to do pointer
changes to get them in and out of memory and you are having to scan
per-cpu lists. You would be better served using a simple array as your
threads should be stick to a consistent CPU anyway in terms of
testing.

I would suggest keeping this much more simple. Trying to pattern this
after something like the dmapool_test code would be a better way to go
for this. We don't need all this extra objpool overhead getting in the
way of testing the code you should be focused on. Just allocate your
array on one specific CPU and start placing and removing your pages
from there instead of messing with the push/pop semantics.

Lastly something that is a module only tester that always fails to
probe doesn't sound like it really makes sense as a standard kernel
module. I still think it would make more sense to move it to the
selftests tree and just have it build there as a module instead of
trying to force it into the mm tree. The example of dmapool_test makes
sense as it could be run at early boot to run the test and then it
just goes quiet. This module won't load and will always just return
-EAGAIN which doesn't sound like a valid kernel module to me.