Re: [PATCH V2 1/7] trace/events: Add gup trace events

[Shi, Yang]

On 12/2/2015 8:07 PM, Steven Rostedt wrote:
> On Wed,  2 Dec 2015 14:53:27 -0800
> Yang Shi <yang.shi@xxxxxxxxxx> wrote:
>
> > page-faults events record the invoke to handle_mm_fault, but the invoke
> > may come from do_page_fault or gup. In some use cases, the finer event count
> > mey be needed, so add trace events support for:
> >
> > __get_user_pages
> > __get_user_pages_fast
> > fixup_user_fault
> >
> > Signed-off-by: Yang Shi <yang.shi@xxxxxxxxxx>
> > ---
> >   include/trace/events/gup.h | 71 ++++++++++++++++++++++++++++++++++++++++++++++
> >   1 file changed, 71 insertions(+)
> >   create mode 100644 include/trace/events/gup.h
> >
> > diff --git a/include/trace/events/gup.h b/include/trace/events/gup.h
> > new file mode 100644
> > index 0000000..03a4674
> > --- /dev/null
> > +++ b/include/trace/events/gup.h
> > @@ -0,0 +1,71 @@
> > +#undef TRACE_SYSTEM
> > +#define TRACE_SYSTEM gup
> > +
> > +#if !defined(_TRACE_GUP_H) || defined(TRACE_HEADER_MULTI_READ)
> > +#define _TRACE_GUP_H
> > +
> > +#include <linux/types.h>
> > +#include <linux/tracepoint.h>
> > +
> > +TRACE_EVENT(gup_fixup_user_fault,
> > +
> > +	TP_PROTO(struct task_struct *tsk, struct mm_struct *mm,
> > +			unsigned long address, unsigned int fault_flags),
> > +
> > +	TP_ARGS(tsk, mm, address, fault_flags),
>
> Arges added and not used by TP_fast_assign(), this will slow down the
> code while tracing is enabled, as they need to be added to the trace
> function call.
>
> > +
> > +	TP_STRUCT__entry(
> > +		__field(	unsigned long,	address		)
> > +	),
> > +
> > +	TP_fast_assign(
> > +		__entry->address	= address;
> > +	),
> > +
> > +	TP_printk("address=%lx",  __entry->address)
> > +);
> > +
> > +TRACE_EVENT(gup_get_user_pages,
> > +
> > +	TP_PROTO(struct task_struct *tsk, struct mm_struct *mm,
> > +			unsigned long start, unsigned long nr_pages),
> > +
> > +	TP_ARGS(tsk, mm, start, nr_pages),
>
> Here too but this is worse. See below.
>
> > +
> > +	TP_STRUCT__entry(
> > +		__field(	unsigned long,	start		)
> > +		__field(	unsigned long,	nr_pages	)
> > +	),
> > +
> > +	TP_fast_assign(
> > +		__entry->start		= start;
> > +		__entry->nr_pages	= nr_pages;
> > +	),
> > +
> > +	TP_printk("start=%lx nr_pages=%lu", __entry->start, __entry->nr_pages)
> > +);
> > +
> > +TRACE_EVENT(gup_get_user_pages_fast,
> > +
> > +	TP_PROTO(unsigned long start, int nr_pages, int write,
> > +			struct page **pages),
> > +
> > +	TP_ARGS(start, nr_pages, write, pages),
>
> This and the above "gup_get_user_pages" have the same entry field,
> assign and printk. They should be combined into a DECLARE_EVENT_CLASS()
> and two DEFINE_EVENT()s. That will save on size as the
> DECLARE_EVENT_CLASS() is the biggest part of each TRACE_EVENT().

Thanks for the suggestion, will fix them in V3.

Regards,
Yang

> -- Steve
>
>
> > +
> > +	TP_STRUCT__entry(
> > +		__field(	unsigned long,	start		)
> > +		__field(	unsigned long,	nr_pages	)
> > +	),
> > +
> > +	TP_fast_assign(
> > +		__entry->start  	= start;
> > +		__entry->nr_pages	= nr_pages;
> > +	),
> > +
> > +	TP_printk("start=%lx nr_pages=%lu",  __entry->start, __entry->nr_pages)
> > +);
> > +
> > +#endif /* _TRACE_GUP_H */
> > +
> > +/* This part must be outside protection */
> > +#include <trace/define_trace.h>

--
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/