Re: [PATCH 0/7] remove SLOB and allow kfree() with kmem_cache_alloc()

[Mike Rapoport]

On Mon, Mar 13, 2023 at 12:31:47PM -0400, Steven Rostedt wrote:
> On Sun, 12 Mar 2023 11:51:29 +0200
> Mike Rapoport <mike.rapoport@xxxxxxxxx> wrote:
> 
> > git grep -in slob still gives a couple of matches. I've dropped the
> > irrelevant ones it it left me with these:
> > 
> > CREDITS:14:D: SLOB slab allocator
> > kernel/trace/ring_buffer.c:358: * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
> > mm/Kconfig:251:    SLOB allocator and is not recommended for systems with more than
> > mm/Makefile:25:KCOV_INSTRUMENT_slob.o := n
> >  
> > Except the comment in kernel/trace/ring_buffer.c all are trivial.
> > 
> > As for the comment in ring_buffer.c, it looks completely irrelevant at this
> > point.
> > 
> > @Steve?
> 
> You want me to remember something I wrote almost 15 years ago?

I just wanted to make sure you don't have a problem with removing this
comment :)

> I think I understand that comment as much as you do. Yeah, that was when
> I was still learning to write comments for my older self to understand,
> and I failed miserably!
>
> But git history comes to the rescue. The commit that added that comment was:
> 
> ed56829cb3195 ("ring_buffer: reset buffer page when freeing")
> 
> This was at a time when it was suggested to me to use the struct page
> directly in the ring buffer and where we could do fun "tricks" for
> "performance". (I was never really for this, but I wasn't going to argue).
> 
> And the code in question then had:
> 
> /*
>  * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
>  * this issue out.
>  */
> static inline void free_buffer_page(struct buffer_page *bpage)
> {
>         reset_page_mapcount(&bpage->page);
>         bpage->page.mapping = NULL;
>         __free_page(&bpage->page);
> }
> 
> 
> But looking at commit: e4c2ce82ca27 ("ring_buffer: allocate buffer page
> pointer")
> 
> It was finally decided that method was not safe, and we should not be using
> struct page but just allocate an actual page (much safer!).
> 
> I never got rid of the comment, which was more about that
> "reset_page_mapcount()", and should have been deleted back then.

Yeah, I did the same analysis, just was too lazy to post it.
 
> Just remove that comment. And you could even add:
> 
> Suggested-by: Steven Rostedt (Google) <rostedt@xxxxxxxxxxx>
> Fixes: e4c2ce82ca27 ("ring_buffer: allocate buffer page pointer")
> 
> -- Steve

-- 
Sincerely yours,
Mike.