Re: [PATCH V2 2/7] mm/gup: add gup trace points

[Shi, Yang]

On 12/2/2015 8:13 PM, Steven Rostedt wrote:
> On Wed, 2 Dec 2015 15:36:50 -0800
> Dave Hansen <dave.hansen@xxxxxxxxx> wrote:
>
> > On 12/02/2015 02:53 PM, Yang Shi wrote:
> > > diff --git a/mm/gup.c b/mm/gup.c
> > > index deafa2c..10245a4 100644
> > > --- a/mm/gup.c
> > > +++ b/mm/gup.c
> > > @@ -13,6 +13,9 @@
> > >   #include <linux/rwsem.h>
> > >   #include <linux/hugetlb.h>
> > >
> > > +#define CREATE_TRACE_POINTS
> > > +#include <trace/events/gup.h>
> > > +
> > >   #include <asm/pgtable.h>
> > >   #include <asm/tlbflush.h>
> >
> > This needs to be _the_ last thing that gets #included.  Otherwise, you
> > risk colliding with any other trace header that gets implicitly included
> > below.
>
> Agreed.
>
> > > @@ -1340,6 +1346,8 @@ int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
> > >   					start, len)))
> > >   		return 0;
> > >
> > > +	trace_gup_get_user_pages_fast(start, nr_pages, write, pages);
> > > +
> > >   	/*
> > >   	 * Disable interrupts.  We use the nested form as we can already have
> > >   	 * interrupts disabled by get_futex_key.
> >
> > It would be _really_ nice to be able to see return values from the
> > various gup calls as well.  Is that feasible?
>
> Only if you rewrite the functions to have a single return code path
> that we can add a tracepoint too. Or have a wrapper function that gets

Yes. My preliminary test just could cover the success case. gup could 
return errno from different a few code path.

> called directly that calls these functions internally and the tracepoint
> can trap the return value.

This will incur more changes in other subsystems (futex, kvm, etc), I'm 
not sure if it is worth making such changes to get return value.

> I can probably make function_graph tracer give return values, although
> it will give a return value for void functions as well. And it may give
> long long returns for int returns that may have bogus data in the
> higher bits.

If the return value requirement is not limited to gup, the approach 
sounds more reasonable.

Thanks,
Yang

> -- Steve

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/