Michal Hocko wrote: > On Fri 15-09-17 17:58:49, wang Yu wrote: > > From: "yuwang.yuwang" <yuwang.yuwang@xxxxxxxxxxxxxxx> > > > > I found a softlockup when running some stress testcase in 4.9.x, > > but i think the mainline have the same problem. > > > > call trace: > > [365724.502896] NMI watchdog: BUG: soft lockup - CPU#31 stuck for 22s! > > [jbd2/sda3-8:1164] > > ... > > ... > > [365724.503258] Call Trace: > > [365724.503260] [<ffffffff811ace5f>] warn_alloc+0x13f/0x170 > > [365724.503264] [<ffffffff811ad8c2>] __alloc_pages_slowpath+0x9b2/0xc10 > > [365724.503265] [<ffffffff811add43>] __alloc_pages_nodemask+0x223/0x2a0 > > [365724.503268] [<ffffffff811fe838>] alloc_pages_current+0x88/0x120 > > [365724.503270] [<ffffffff811a3644>] __page_cache_alloc+0xb4/0xc0 > > [365724.503272] [<ffffffff811a49e9>] pagecache_get_page+0x59/0x230 > > [365724.503275] [<ffffffff8126b2db>] __getblk_gfp+0xfb/0x2f0 > > [365724.503281] [<ffffffffa00f9cee>] > > jbd2_journal_get_descriptor_buffer+0x5e/0xe0 [jbd2] > > [365724.503286] [<ffffffffa00f2a01>] > > jbd2_journal_commit_transaction+0x901/0x1880 [jbd2] > > [365724.503291] [<ffffffff8102d6a5>] ? __switch_to+0x215/0x730 > > [365724.503294] [<ffffffff810f962d>] ? lock_timer_base+0x7d/0xa0 > > [365724.503298] [<ffffffffa00f7cda>] kjournald2+0xca/0x260 [jbd2] > > [365724.503300] [<ffffffff810cfb00>] ? prepare_to_wait_event+0xf0/0xf0 > > [365724.503304] [<ffffffffa00f7c10>] ? commit_timeout+0x10/0x10 [jbd2] > > [365724.503307] [<ffffffff810a8d66>] kthread+0xe6/0x100 > > [365724.503309] [<ffffffff810a8c80>] ? kthread_park+0x60/0x60 > > [365724.503313] [<ffffffff816f3795>] ret_from_fork+0x25/0x30 > > > > we can limit the warn_alloc caller to workaround it. > > __alloc_pages_slowpath only call once warn_alloc each time. > > similar attempts to add a lock there were tried in the past and refused. Wang already read that thread before proposing this patch. > Anyway using a normal lock would be preferred over a bit lock. But the > most important part is to identify _why_ we see the lockup trigerring in > the first place. And try to fix it rather than workaround it here. The bitlock is what Andrew thought at http://lkml.kernel.org/r/20170601151022.b17716472adbf0e6d51fb011@xxxxxxxxxxxxxxxxxxxx . I'm OK with using a normal lock. This patch does not make callers of warn_alloc() sleep. This is different from past attempt. You said "identify _why_ we see the lockup trigerring in the first place" without providing means to identify it. Unless you provide means to identify it (in a form which can be immediately and easily backported to 4.9 kernels; that is, backporting not-yet-accepted printk() offloading patchset is not a choice), this patch cannot be refused. > > > Signed-off-by: yuwang.yuwang <yuwang.yuwang@xxxxxxxxxxxxxxx> > > Suggested-by: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> > > --- > > mm/page_alloc.c | 7 +++++-- > > 1 file changed, 5 insertions(+), 2 deletions(-) > > > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > > index 2abf8d5..8b86686 100644 > > --- a/mm/page_alloc.c > > +++ b/mm/page_alloc.c > > @@ -3525,6 +3525,7 @@ bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) > > unsigned long alloc_start = jiffies; > > unsigned int stall_timeout = 10 * HZ; > > unsigned int cpuset_mems_cookie; > > + static unsigned long stall_warn_lock; > > > > /* > > * In the slowpath, we sanity check order to avoid ever trying to > > @@ -3698,11 +3699,13 @@ bool gfp_pfmemalloc_allowed(gfp_t gfp_mask) > > goto nopage; > > > > /* Make sure we know about allocations which stall for too long */ > > - if (time_after(jiffies, alloc_start + stall_timeout)) { > > + if (time_after(jiffies, alloc_start + stall_timeout) && > > + !test_and_set_bit_lock(0, &stall_warn_lock)) { > > warn_alloc(gfp_mask, > > "page allocation stalls for %ums, order:%u", > > jiffies_to_msecs(jiffies-alloc_start), order); > > - stall_timeout += 10 * HZ; > > + stall_timeout = jiffies - alloc_start + 10 * HZ; > > + clear_bit_unlock(0, &stall_warn_lock); > > } > > > > if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags, > > -- > > 1.8.3.1 > > -- > Michal Hocko > SUSE Labs -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
