lkml.org 
[lkml]   [2016]   [Jul]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [PATCH 1/2] mm: page_alloc.c: Add tracepoints for slowpath
On Wed 27-07-16 10:47:59, Janani Ravichandran wrote:
> Add tracepoints to the slowpath code to gather some information.
> The tracepoints can also be used to find out how much time was spent in
> the slowpath.

I do not think this is a right thing to measure. __alloc_pages_slowpath
is more a code organization thing. The fast path might perform an
expensive operations like zone reclaim (if node_reclaim_mode > 0) so
these trace point would miss it.

__alloc_pages_nodemask already has a trace point after the allocation
request is done. This alone is not sufficient to measure the allocation
latency which is the main point of this patch AFAIU. Wouldn't it be
better to add another trace point when we enter __alloc_pages_nodemask?

> Signed-off-by: Janani Ravichandran <janani.rvchndrn@gmail.com>
> ---
> include/trace/events/kmem.h | 40 ++++++++++++++++++++++++++++++++++++++++
> mm/page_alloc.c | 5 +++++
> 2 files changed, 45 insertions(+)
>
> diff --git a/include/trace/events/kmem.h b/include/trace/events/kmem.h
> index 6b2e154..c19ab9f 100644
> --- a/include/trace/events/kmem.h
> +++ b/include/trace/events/kmem.h
> @@ -169,6 +169,46 @@ TRACE_EVENT(mm_page_free,
> __entry->order)
> );
>
> +TRACE_EVENT(mm_slowpath_begin,
> +
> + TP_PROTO(gfp_t gfp_mask, int order),
> +
> + TP_ARGS(gfp_mask, order),
> +
> + TP_STRUCT__entry(
> + __field(gfp_t, gfp_mask)
> + __field(int, order)
> + ),
> +
> + TP_fast_assign(
> + __entry->gfp_mask = gfp_mask;
> + __entry->order = order;
> + ),
> +
> + TP_printk("gfp_mask:%s order=%d",
> + show_gfp_flags(__entry->gfp_mask),
> + __entry->order)
> +);
> +
> +TRACE_EVENT(mm_slowpath_end,
> +
> + TP_PROTO(struct page *page),
> +
> + TP_ARGS(page),
> +
> + TP_STRUCT__entry(
> + __field(unsigned long, pfn)
> + ),
> +
> + TP_fast_assign(
> + __entry->pfn = page ? page_to_pfn(page) : -1UL;
> + ),
> +
> + TP_printk("page=%p pfn=%lu",
> + __entry->pfn != -1UL ? pfn_to_page(__entry->pfn) : NULL,
> + __entry->pfn != -1UL ? __entry->pfn : 0)
> +);
> +
> TRACE_EVENT(mm_page_free_batched,
>
> TP_PROTO(struct page *page, int cold),
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 8b3e134..be9c688 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3595,6 +3595,8 @@ __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
> (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
> gfp_mask &= ~__GFP_ATOMIC;
>
> + trace_mm_slowpath_begin(gfp_mask, order);
> +
> retry:
> if (gfp_mask & __GFP_KSWAPD_RECLAIM)
> wake_all_kswapds(order, ac);
> @@ -3769,6 +3771,9 @@ noretry:
> nopage:
> warn_alloc_failed(gfp_mask, order, NULL);
> got_pg:
> +
> + trace_mm_slowpath_end(page);
> +
> return page;
> }
>
> --
> 2.7.0
>

--
Michal Hocko
SUSE Labs

\
 
 \ /
  Last update: 2016-07-27 19:01    [W:0.088 / U:0.128 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site