lkml.org 
[lkml]   [2009]   [Aug]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 1/4] tracing, page-allocator: Add trace events for page allocation and page freeing
On Fri, Aug 07, 2009 at 10:17:57AM +0900, KOSAKI Motohiro wrote:
>
> > > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > > index d052abb..843bdec 100644
> > > > --- a/mm/page_alloc.c
> > > > +++ b/mm/page_alloc.c
> > > > @@ -1905,6 +1905,7 @@ __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
> > > > zonelist, high_zoneidx, nodemask,
> > > > preferred_zone, migratetype);
> > > >
> > > > + trace_mm_page_alloc(_RET_IP_, page, order, gfp_mask, migratetype);
> > > > return page;
> > > > }
> > >
> > > In almost case, __alloc_pages_nodemask() is called from alloc_pages_current().
> > > Can you add call_site argument? (likes slab_alloc)
> > >
> >
> > In the NUMA case, this will be true but addressing it involves passing down
> > an additional argument in the non-tracing case which I wanted to avoid.
> > As the stacktrace option is available to ftrace, I think I'll drop call_site
> > altogether as anyone who really needs that information has options.
>
> Insted, can we move this tracepoint to alloc_pages_current(), alloc_pages_node() et al ?
> On page tracking case, call_site information is one of most frequently used one.
> if we need multiple trace combination, it become hard to use and reduce usefulness a bit.
>

Ok, lets think about that. The potential points that would need
annotation are

o alloc_pages_current
o alloc_page_vma
o alloc_pages_node
o alloc_pages_exact_node

The inlined functions that call those and should preserve the call_site
are

o alloc_pages

The slightly lower functions they call are as follows. These cannot
trigger a tracepoint event because it would look like a duplicate.

o __alloc_pages_nodemask
- called by __alloc_pages
o __alloc_pages
- called by alloc_page_interleave() but event logged
- called by alloc_pages_node but event logged
- called by alloc_pages_exact_node but event logged

The more problematic ones are

o __get_free_pages
o get_zeroed_page
o alloc_pages_exact

The are all real functions that call down to functions that would log
events already based on your suggestion - alloc_pages_current() in
particularly.

Looking at it, it would appear the page allocator API would need a fair
amount of reschuffling to preserve call_site and not duplicate events or
else to pass call_site down through the API even in the non-tracing case.
Minimally, that makes it a standalone patch but it would also need a good
explanation as to why capturing the stack trace on the event is not enough
to track the page for things like catching memory leaks.

--
Mel Gorman
Part-time Phd Student Linux Technology Center
University of Limerick IBM Dublin Software Lab


\
 
 \ /
  Last update: 2009-08-07 19:33    [from the cache]
©2003-2011 Jasper Spaans