Re: [PATCH v2] Show kernel stack usage to /proc/meminfo and OOM log

From: KOSAKI Motohiro
Date: Tue Jun 30 2009 - 22:52:39 EST


> On Wed, 1 Jul 2009, KOSAKI Motohiro wrote:
>
> > Subject: [PATCH] Show kernel stack usage to /proc/meminfo and OOM log
> >
> > if the system have a lot of thread, kernel stack consume unignorable large size
> > memory.
> > IOW, it make a lot of unaccountable memory.
> >
> > Tons unaccountable memory bring to harder analyse memory related trouble.
> >
> > Then, kernel stack account is useful.
> >
> >
>
> I know this is the second revision of the patch, apologies for not
> responding to the first.

Thanks, good review.

>
> > Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@xxxxxxxxxxxxxx>
> > ---
> > fs/proc/meminfo.c | 2 ++
> > include/linux/mmzone.h | 3 ++-
> > kernel/fork.c | 12 ++++++++++++
> > mm/page_alloc.c | 6 ++++--
> > mm/vmstat.c | 1 +
> > 5 files changed, 21 insertions(+), 3 deletions(-)
> >
> > Index: b/fs/proc/meminfo.c
> > ===================================================================
> > --- a/fs/proc/meminfo.c
> > +++ b/fs/proc/meminfo.c
> > @@ -85,6 +85,7 @@ static int meminfo_proc_show(struct seq_
> > "SReclaimable: %8lu kB\n"
> > "SUnreclaim: %8lu kB\n"
> > "PageTables: %8lu kB\n"
> > + "KernelStack %8lu kB\n"
>
> Missing :.

Grr, thanks. Will fix.

>
> > #ifdef CONFIG_QUICKLIST
> > "Quicklists: %8lu kB\n"
> > #endif
> > @@ -129,6 +130,7 @@ static int meminfo_proc_show(struct seq_
> > K(global_page_state(NR_SLAB_RECLAIMABLE)),
> > K(global_page_state(NR_SLAB_UNRECLAIMABLE)),
> > K(global_page_state(NR_PAGETABLE)),
> > + K(global_page_state(NR_KERNEL_STACK)),
> > #ifdef CONFIG_QUICKLIST
> > K(quicklist_total_size()),
> > #endif
> > Index: b/include/linux/mmzone.h
> > ===================================================================
> > --- a/include/linux/mmzone.h
> > +++ b/include/linux/mmzone.h
> > @@ -94,10 +94,11 @@ enum zone_stat_item {
> > NR_SLAB_RECLAIMABLE,
> > NR_SLAB_UNRECLAIMABLE,
> > NR_PAGETABLE, /* used for pagetables */
> > + NR_KERNEL_STACK,
> > + /* Second 128 byte cacheline */
> > NR_UNSTABLE_NFS, /* NFS unstable pages */
> > NR_BOUNCE,
> > NR_VMSCAN_WRITE,
> > - /* Second 128 byte cacheline */
> > NR_WRITEBACK_TEMP, /* Writeback using temporary buffers */
> > #ifdef CONFIG_NUMA
> > NUMA_HIT, /* allocated in intended node */
> > Index: b/kernel/fork.c
> > ===================================================================
> > --- a/kernel/fork.c
> > +++ b/kernel/fork.c
> > @@ -137,9 +137,18 @@ struct kmem_cache *vm_area_cachep;
> > /* SLAB cache for mm_struct structures (tsk->mm) */
> > static struct kmem_cache *mm_cachep;
> >
> > +static void account_kernel_stack(struct thread_info *ti, int on)
> > +{
> > + struct zone *zone = page_zone(virt_to_page(ti));
> > + int pages = THREAD_SIZE / PAGE_SIZE;
> > +
> > + mod_zone_page_state(zone, NR_KERNEL_STACK, on ? pages : -pages);
> > +}
> > +
> > void free_task(struct task_struct *tsk)
> > {
> > prop_local_destroy_single(&tsk->dirties);
> > + account_kernel_stack(tsk->stack, 0);
>
> I think it would be better to do
>
> #define THREAD_PAGES (THREAD_SIZE / PAGE_SIZE)
>
> since it's currently unused and then
>
> struct zone *zone = page_zone(virt_to_page(tsk->stack));
> mod_zone_page_state(zone, NR_KERNEL_STACK, THREAD_PAGES);
>
> in free_task() and
>
> struct zone *zone = page_zone(virt_to_page(ti));
> mod_zone_page_state(zone, NR_KERNEL_STACK, -THREAD_PAGES);
>
> in dup_task_struct().

maybe, gcc makes same code. then I keep current code. because
"struct zone *zone = page_zone(virt_to_page(tsk->stack))" line is a bit
complicate statement and I don't hope sprinkle it.

>
> > free_thread_info(tsk->stack);
> > rt_mutex_debug_task_free(tsk);
> > ftrace_graph_exit_task(tsk);
> > @@ -255,6 +264,9 @@ static struct task_struct *dup_task_stru
> > tsk->btrace_seq = 0;
> > #endif
> > tsk->splice_pipe = NULL;
> > +
> > + account_kernel_stack(ti, 1);
> > +
> > return tsk;
> >
> > out:
> > Index: b/mm/page_alloc.c
> > ===================================================================
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -2119,7 +2119,8 @@ void show_free_areas(void)
> > " inactive_file:%lu"
> > " unevictable:%lu"
> > " dirty:%lu writeback:%lu unstable:%lu\n"
> > - " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
> > + " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n"
> > + " kernel_stack:%lu\n",
>
> Does kernel_stack really need to be printed on its own line?

Well, my another patch (Makes slab pages field in show_free_areas() separate two field)
already used full space of previous line. new line is really needed.





--
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/