Re: [RFC PATCH] mm/madvise: remove redundant mmap_lock operations from process_madvise()

[Shakeel Butt <shakeel.butt@linux.dev>]

On Tue, Jan 14, 2025 at 06:47:15PM +0000, Lorenzo Stoakes wrote:
> On Tue, Jan 14, 2025 at 10:13:40AM -0800, Shakeel Butt wrote:
> > Ccing relevant folks.
> 
> Thanks Shakeel!
> 
> A side-note, I really wish there was a better way to get cc'd, since I
> fundamentally changed process_madvise() recently and was the main person
> changing this code lately, but on the other hand -
> scripts/get_maintainers.pl gets really really noisy if you try to use this
> kind of stat - so I in no way blame SJ for missing me.
> 
> Thankfully Shakeel kindly stepped in to make me aware :)
> 
> SJ - I will come back to you later, as it's late here and my brain is fried
> - but I was already thinking of doing something _like_ this, as I noticed
> for the purposes of self-process_madvise() operations (which I unrestricted
> for guard page purposes) - we are hammering locks in a way that we know we
> don't necessarily need to do.
> 
> So this is serendipitous for me! :) But I need to dig into your actual
> implementation to give feedback here.
> 
> Will come back to this in due course :)
> 

SJ is planning to do couple more optimizations like single tree
traversal (where possible) and batching TLB flushing for advices which
does TLB flushing. It is better to coordinate the work than orthogonally
repeating the work.

Thanks Lorenzo for your time.

Shakeel