Re: [v3.19-rc1] read_events => aio_read_events => WARNING: "do not call blocking ops when !TASK_RUNNING"
From: Fam Zheng
Date: Wed Dec 24 2014 - 01:50:32 EST
On Mon, 12/22 13:12, Bjørn Mork wrote:
> Hello,
>
> I got this warning on v3.19-rc1:
>
> [ 16.325814] ------------[ cut here ]------------
> [ 16.325832] WARNING: CPU: 0 PID: 3368 at kernel/sched/core.c:7303 __might_sleep+0x55/0x94()
> [ 16.325839] do not call blocking ops when !TASK_RUNNING; state=1 set at [<ffffffff81071360>] prepare_to_wait_event+0x8c/0xd8
> [ 16.325843] Modules linked in: xt_multiport xt_addrtype xt_conntrack ipt_MASQUERADE nf_nat_masquerade_ipv4 iptable_nat nf_conntrack_ipv4 nf_defrag_ipv4 nf_nat_ipv4 nf_nat nf_conntrack bridge dm_thin_pool dm_persistent_data crc32c_generic libcrc32c dm_bio_prison dm_bufio dm_mod iptable_filter ip_tables bnep xt_hl nf_log_ipv6 nf_log_common xt_LOG binfmt_misc ip6table_filter ip6_tables x_tables nfsd nfs_acl nfs lockd grace fscache sunrpc 8021q garp mrp stp llc tun fuse loop iTCO_wdt iTCO_vendor_support snd_hda_codec_conexant snd_hda_codec_generic coretemp arc4 kvm_intel kvm ecb evdev psmouse serio_raw uvcvideo videobuf2_vmalloc videobuf2_memops cdc_mbim cdc_wdm cdc_ncm videobuf2_core v4l2_common usbnet mii cdc_acm videodev iwlmvm btusb mac80211 bluetooth iwlwifi snd_hda_intel i2c_i801 snd_hda_controller
> [ 16.325980] lpc_ich mfd_core cfg80211 snd_hda_codec snd_hwdep snd_pcm_oss snd_mixer_oss snd_pcm snd_timer thinkpad_acpi wmi nvram snd soundcore rfkill ac i915 battery i2c_algo_bit drm_kms_helper video drm button acpi_cpufreq processor ext4 crc16 jbd2 mbcache nbd microcode sg sr_mod sd_mod cdrom ahci libahci libata scsi_mod e1000e ehci_pci uhci_hcd ehci_hcd ptp usbcore usb_common pps_core thermal thermal_sys
> [ 16.326070] CPU: 0 PID: 3368 Comm: mysqld Not tainted 3.19.0-rc1 #270
> [ 16.326074] Hardware name: LENOVO 2776LEG/2776LEG, BIOS 6EET55WW (3.15 ) 12/19/2011
> [ 16.326077] 0000000000000009 ffff88022a66fc58 ffffffff813cf430 0000000000000000
> [ 16.326087] ffff88022a66fca8 ffff88022a66fc98 ffffffff8104117c 00000000000092e8
> [ 16.326095] ffffffff81060e84 ffffffff815a5241 000000000000026d 0000000000000000
> [ 16.326104] Call Trace:
> [ 16.326112] [<ffffffff813cf430>] dump_stack+0x4c/0x65
> [ 16.326119] [<ffffffff8104117c>] warn_slowpath_common+0x9c/0xb6
> [ 16.326126] [<ffffffff81060e84>] ? __might_sleep+0x55/0x94
> [ 16.326132] [<ffffffff8104122a>] warn_slowpath_fmt+0x41/0x43
> [ 16.326141] [<ffffffff8100a3cd>] ? native_sched_clock+0x35/0x37
> [ 16.326147] [<ffffffff81071360>] ? prepare_to_wait_event+0x8c/0xd8
> [ 16.326153] [<ffffffff81071360>] ? prepare_to_wait_event+0x8c/0xd8
> [ 16.326159] [<ffffffff81060e84>] __might_sleep+0x55/0x94
> [ 16.326166] [<ffffffff813d3205>] mutex_lock_nested+0x2a/0x39e
> [ 16.326173] [<ffffffff81070db9>] ? spin_unlock_irqrestore+0x9/0xb
> [ 16.326180] [<ffffffff81067f61>] ? local_clock+0x19/0x22
> [ 16.326187] [<ffffffff81076446>] ? lock_release_holdtime.part.26+0x71/0x7f
> [ 16.326194] [<ffffffff813d514e>] ? _raw_spin_unlock_irqrestore+0x3a/0x48
> [ 16.326201] [<ffffffff8117691f>] aio_read_events+0x45/0x282
> [ 16.326208] [<ffffffff81176fe0>] read_events+0x1c0/0x23c
> [ 16.326214] [<ffffffff8107124d>] ? abort_exclusive_wait+0x95/0x95
> [ 16.326221] [<ffffffff81095193>] ? hrtimer_get_res+0x47/0x47
> [ 16.326227] [<ffffffff81095c9e>] ? hrtimer_start_range_ns+0xf/0x11
> [ 16.326234] [<ffffffff81178d62>] SyS_io_getevents+0x4c/0x76
> [ 16.326241] [<ffffffff813d5492>] system_call_fastpath+0x12/0x17
> [ 16.326246] ---[ end trace aa79768e2e6ba68a ]---
>
>
> And the mysqld process just sits there burning CPU:
>
> top - 13:01:23 up 33 min, 15 users, load average: 10.07, 10.23, 9.28
> Tasks: 175 total, 2 running, 173 sleeping, 0 stopped, 0 zombie
> %Cpu(s): 4.0 us, 95.5 sy, 0.0 ni, 0.2 id, 0.1 wa, 0.0 hi, 0.2 si, 0.0 st
> KiB Mem: 8069208 total, 4414640 used, 3654568 free, 475752 buffers
> KiB Swap: 9525244 total, 0 used, 9525244 free. 2063192 cached Mem
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 3343 mysql 20 0 549944 47456 10492 S 171.9 0.6 60:21.11 mysqld
>
>
> Looking at it, it just loops over the io_getevents syscall:
>
>
> nemi:/tmp# strace -fp 3343
> Process 3343 attached with 17 threads
> [pid 3414] futex(0x7f08cbf43ba4, FUTEX_WAIT_PRIVATE, 95, NULL <unfinished ...>
> [pid 3410] rt_sigtimedwait([HUP QUIT ALRM TERM TSTP], NULL, NULL, 8 <unfinished ...>
> [pid 3396] futex(0x7f08cd669a0c, FUTEX_WAIT_PRIVATE, 1, NULL <unfinished ...>
> [pid 3395] restart_syscall(<... resuming interrupted call ...> <unfinished ...>
> [pid 3393] restart_syscall(<... resuming interrupted call ...> <unfinished ...>
> [pid 3375] io_getevents(139675742199808, 1, 256, <unfinished ...>
> [pid 3394] restart_syscall(<... resuming interrupted call ...> <unfinished ...>
> [pid 3343] restart_syscall(<... resuming interrupted call ...> <unfinished ...>
> [pid 3368] io_getevents(139675744239616, 1, 256, <unfinished ...>
> [pid 3370] io_getevents(139675744198656, 1, 256, <unfinished ...>
> [pid 3372] io_getevents(139675744157696, 1, 256, <unfinished ...>
> [pid 3369] io_getevents(139675744219136, 1, 256, <unfinished ...>
> [pid 3376] io_getevents(139675696115712, 1, 256, <unfinished ...>
> [pid 3377] io_getevents(139675696095232, 1, 256, <unfinished ...>
> [pid 3371] io_getevents(139675744178176, 1, 256, <unfinished ...>
> [pid 3374] io_getevents(139675742220288, 1, 256, <unfinished ...>
> [pid 3373] io_getevents(139675742240768, 1, 256, <unfinished ...>
> [pid 3375] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3368] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3370] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3370] io_getevents(139675744198656, 1, 256, <unfinished ...>
> [pid 3372] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3372] io_getevents(139675744157696, 1, 256, <unfinished ...>
> [pid 3369] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3375] io_getevents(139675742199808, 1, 256, <unfinished ...>
> [pid 3368] io_getevents(139675744239616, 1, 256, <unfinished ...>
> [pid 3376] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3369] io_getevents(139675744219136, 1, 256, <unfinished ...>
> [pid 3376] io_getevents(139675696115712, 1, 256, <unfinished ...>
> [pid 3377] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3377] io_getevents(139675696095232, 1, 256, <unfinished ...>
> [pid 3373] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3371] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3374] <... io_getevents resumed> {}{0, 500000000}) = 0
> [pid 3371] io_getevents(139675744178176, 1, 256, <unfinished ...>
> [pid 3374] io_getevents(139675742220288, 1, 256, <unfinished ...>
>
>
>
> I have not tried to repeat this yet, so I don't know if it is
> reproducable. FWIW, it never happened to me with v3.18. I send this to
> you because it looks like you touched this code path between v3.18 and
> v3.19-rc1, so I hope you have a clue what's going on here... Ref:
Could you trace down to the source code context of this io_getevents call in
mysqld? The change merely made io_getevents return a lot more quickly than
before, so it seems like that a polling loop is spinning too fast as a result.
Fam
>
>
>
> commit 5f785de588735306ec4d7c875caf9d28481c8b21
> Author: Fam Zheng <famz@xxxxxxxxxx>
> Date: Thu Nov 6 20:44:36 2014 +0800
>
> aio: Skip timer for io_getevents if timeout=0
>
> In this case, it is basically a polling. Let's not involve timer at all
> because that would hurt performance for application event loops.
>
> In an arbitrary test I've done, io_getevents syscall elapsed time
> reduces from 50000+ nanoseconds to a few hundereds.
>
> Signed-off-by: Fam Zheng <famz@xxxxxxxxxx>
> Signed-off-by: Benjamin LaHaise <bcrl@xxxxxxxxx>
>
> diff --git a/fs/aio.c b/fs/aio.c
> index bfab55607a4d..1b7893ecc296 100644
> --- a/fs/aio.c
> +++ b/fs/aio.c
> @@ -1253,8 +1253,12 @@ static long read_events(struct kioctx *ctx, long min_nr, long nr,
> * the ringbuffer empty. So in practice we should be ok, but it's
> * something to be aware of when touching this code.
> */
> - wait_event_interruptible_hrtimeout(ctx->wait,
> - aio_read_events(ctx, min_nr, nr, event, &ret), until);
> + if (until.tv64 == 0)
> + aio_read_events(ctx, min_nr, nr, event, &ret);
> + else
> + wait_event_interruptible_hrtimeout(ctx->wait,
> + aio_read_events(ctx, min_nr, nr, event, &ret),
> + until);
>
> if (!ret && signal_pending(current))
> ret = -EINTR;
>
>
>
>
>
> Bjørn
--
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/