[ Upstream commit bf959931ddb88c4e4366e96dd22e68fa0db9527c ]
The following program (simplified version of generated by syzkaller)
#include <pthread.h>
#include <unistd.h>
#include <sys/ptrace.h>
#include <stdio.h>
#include <signal.h>
void *thread_func(void *arg)
{
ptrace(PTRACE_TRACEME, 0,0,0);
return 0;
}
int main(void)
{
pthread_t thread;
if (fork())
return 0;
while (getppid() != 1)
;
pthread_create(&thread, NULL, thread_func, NULL);
pthread_join(thread, NULL);
return 0;
}
creates an unreapable zombie if /sbin/init doesn't use __WALL.
This is not a kernel bug, at least in a sense that everything works as
expected: debugger should reap a traced sub-thread before it can reap the
leader, but without __WALL/__WCLONE do_wait() ignores sub-threads.
Unfortunately, it seems that /sbin/init in most (all?) distributions
doesn't use it and we have to change the kernel to avoid the problem.
Note also that most init's use sys_waitid() which doesn't allow __WALL, so
the necessary user-space fix is not that trivial.
This patch just adds the "ptrace" check into eligible_child(). To some
degree this matches the "tsk->ptrace" in exit_notify(), ->exit_signal is
mostly ignored when the tracee reports to debugger. Or WSTOPPED, the
tracer doesn't need to set this flag to wait for the stopped tracee.
This obviously means the user-visible change: __WCLONE and __WALL no
longer have any meaning for debugger. And I can only hope that this won't
break something, but at least strace/gdb won't suffer.
We could make a more conservative change. Say, we can take __WCLONE into
account, or !thread_group_leader(). But it would be nice to not
complicate these historical/confusing checks.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: Jan Kratochvil <jan.kratochvil@redhat.com>
Cc: "Michael Kerrisk (man-pages)" <mtk.manpages@gmail.com>
Cc: Pedro Alves <palves@redhat.com>
Cc: Roland McGrath <roland@hack.frob.com>
Cc: <syzkaller@googlegroups.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 59643d1535eb220668692a5359de22545af579f6 ]
If the size passed to ring_buffer_resize() is greater than MAX_LONG - BUF_PAGE_SIZE
then the DIV_ROUND_UP() will return zero.
Here's the details:
# echo 18014398509481980 > /sys/kernel/debug/tracing/buffer_size_kb
tracing_entries_write() processes this and converts kb to bytes.
18014398509481980 << 10 = 18446744073709547520
and this is passed to ring_buffer_resize() as unsigned long size.
size = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
Where DIV_ROUND_UP(a, b) is (a + b - 1)/b
BUF_PAGE_SIZE is 4080 and here
18446744073709547520 + 4080 - 1 = 18446744073709551599
where 18446744073709551599 is still smaller than 2^64
2^64 - 18446744073709551599 = 17
But now 18446744073709551599 / 4080 = 4521260802379792
and size = size * 4080 = 18446744073709551360
This is checked to make sure its still greater than 2 * 4080,
which it is.
Then we convert to the number of buffer pages needed.
nr_page = DIV_ROUND_UP(size, BUF_PAGE_SIZE)
but this time size is 18446744073709551360 and
2^64 - (18446744073709551360 + 4080 - 1) = -3823
Thus it overflows and the resulting number is less than 4080, which makes
3823 / 4080 = 0
an nr_pages is set to this. As we already checked against the minimum that
nr_pages may be, this causes the logic to fail as well, and we crash the
kernel.
There's no reason to have the two DIV_ROUND_UP() (that's just result of
historical code changes), clean up the code and fix this bug.
Cc: stable@vger.kernel.org # 3.5+
Fixes: 83f40318dab00 ("ring-buffer: Make removal of ring buffer pages atomic")
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 9b94a8fba501f38368aef6ac1b30e7335252a220 ]
The size variable to change the ring buffer in ftrace is a long. The
nr_pages used to update the ring buffer based on the size is int. On 64 bit
machines this can cause an overflow problem.
For example, the following will cause the ring buffer to crash:
# cd /sys/kernel/debug/tracing
# echo 10 > buffer_size_kb
# echo 8556384240 > buffer_size_kb
Then you get the warning of:
WARNING: CPU: 1 PID: 318 at kernel/trace/ring_buffer.c:1527 rb_update_pages+0x22f/0x260
Which is:
RB_WARN_ON(cpu_buffer, nr_removed);
Note each ring buffer page holds 4080 bytes.
This is because:
1) 10 causes the ring buffer to have 3 pages.
(10kb requires 3 * 4080 pages to hold)
2) (2^31 / 2^10 + 1) * 4080 = 8556384240
The value written into buffer_size_kb is shifted by 10 and then passed
to ring_buffer_resize(). 8556384240 * 2^10 = 8761737461760
3) The size passed to ring_buffer_resize() is then divided by BUF_PAGE_SIZE
which is 4080. 8761737461760 / 4080 = 2147484672
4) nr_pages is subtracted from the current nr_pages (3) and we get:
2147484669. This value is saved in a signed integer nr_pages_to_update
5) 2147484669 is greater than 2^31 but smaller than 2^32, a signed int
turns into the value of -2147482627
6) As the value is a negative number, in update_pages_handler() it is
negated and passed to rb_remove_pages() and 2147482627 pages will
be removed, which is much larger than 3 and it causes the warning
because not all the pages asked to be removed were removed.
Link: https://bugzilla.kernel.org/show_bug.cgi?id=118001
Cc: stable@vger.kernel.org # 2.6.28+
Fixes: 7a8e76a3829f1 ("tracing: unified trace buffer")
Reported-by: Hao Qin <QEver.cn@gmail.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 58a09ec6e3ec88c9c7e061479f1ef7fe93324a87 ] Instead of using a global per_cpu variable to perform the recursive checks into the ring buffer, use the already existing per_cpu descriptor that is part of the ring buffer itself. Not only does this simplify the code, it also allows for one ring buffer to be used within the guts of the use of another ring buffer. For example trace_printk() can now be used within the ring buffer to record changes done by an instance into the main ring buffer. The recursion checks will prevent the trace_printk() itself from causing recursive issues with the main ring buffer (it is just ignored), but the recursive checks wont prevent the trace_printk() from recording other ring buffers. Signed-off-by: Steven Rostedt <rostedt@goodmis.org> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 3205f8063b6cc54b20d5080fb79dfcbd9c39e93d ]
I was running the trace_event benchmark and noticed that the times
to record a trace_event was all over the place. I looked at the assembly
of the ring_buffer_lock_reserver() and saw this:
<ring_buffer_lock_reserve>:
31 c0 xor %eax,%eax
48 83 3d 76 47 bd 00 cmpq $0x1,0xbd4776(%rip) # ffffffff81d10d60 <ring_buffer_flags>
01
55 push %rbp
48 89 e5 mov %rsp,%rbp
75 1d jne ffffffff8113c60d <ring_buffer_lock_reserve+0x2d>
65 ff 05 69 e3 ec 7e incl %gs:0x7eece369(%rip) # a960 <__preempt_count>
8b 47 08 mov 0x8(%rdi),%eax
85 c0 test %eax,%eax
+---- 74 12 je ffffffff8113c610 <ring_buffer_lock_reserve+0x30>
| 65 ff 0d 5b e3 ec 7e decl %gs:0x7eece35b(%rip) # a960 <__preempt_count>
| 0f 84 85 00 00 00 je ffffffff8113c690 <ring_buffer_lock_reserve+0xb0>
| 31 c0 xor %eax,%eax
| 5d pop %rbp
| c3 retq
| 90 nop
+---> 65 44 8b 05 48 e3 ec mov %gs:0x7eece348(%rip),%r8d # a960 <__preempt_count>
7e
41 81 e0 ff ff ff 7f and $0x7fffffff,%r8d
b0 08 mov $0x8,%al
65 8b 0d 58 36 ed 7e mov %gs:0x7eed3658(%rip),%ecx # fc80 <current_context>
41 f7 c0 00 ff 1f 00 test $0x1fff00,%r8d
74 1e je ffffffff8113c64f <ring_buffer_lock_reserve+0x6f>
41 f7 c0 00 00 10 00 test $0x100000,%r8d
b0 01 mov $0x1,%al
75 13 jne ffffffff8113c64f <ring_buffer_lock_reserve+0x6f>
41 81 e0 00 00 0f 00 and $0xf0000,%r8d
49 83 f8 01 cmp $0x1,%r8
19 c0 sbb %eax,%eax
83 e0 02 and $0x2,%eax
83 c0 02 add $0x2,%eax
85 c8 test %ecx,%eax
75 ab jne ffffffff8113c5fe <ring_buffer_lock_reserve+0x1e>
09 c8 or %ecx,%eax
65 89 05 24 36 ed 7e mov %eax,%gs:0x7eed3624(%rip) # fc80 <current_context>
The arrow is the fast path.
After adding the unlikely's, the fast path looks a bit better:
<ring_buffer_lock_reserve>:
31 c0 xor %eax,%eax
48 83 3d 76 47 bd 00 cmpq $0x1,0xbd4776(%rip) # ffffffff81d10d60 <ring_buffer_flags>
01
55 push %rbp
48 89 e5 mov %rsp,%rbp
75 7b jne ffffffff8113c66b <ring_buffer_lock_reserve+0x8b>
65 ff 05 69 e3 ec 7e incl %gs:0x7eece369(%rip) # a960 <__preempt_count>
8b 47 08 mov 0x8(%rdi),%eax
85 c0 test %eax,%eax
0f 85 9f 00 00 00 jne ffffffff8113c6a1 <ring_buffer_lock_reserve+0xc1>
65 8b 0d 57 e3 ec 7e mov %gs:0x7eece357(%rip),%ecx # a960 <__preempt_count>
81 e1 ff ff ff 7f and $0x7fffffff,%ecx
b0 08 mov $0x8,%al
65 8b 15 68 36 ed 7e mov %gs:0x7eed3668(%rip),%edx # fc80 <current_context>
f7 c1 00 ff 1f 00 test $0x1fff00,%ecx
75 50 jne ffffffff8113c670 <ring_buffer_lock_reserve+0x90>
85 d0 test %edx,%eax
75 7d jne ffffffff8113c6a1 <ring_buffer_lock_reserve+0xc1>
09 d0 or %edx,%eax
65 89 05 53 36 ed 7e mov %eax,%gs:0x7eed3653(%rip) # fc80 <current_context>
65 8b 05 fc da ec 7e mov %gs:0x7eecdafc(%rip),%eax # a130 <cpu_number>
89 c2 mov %eax,%edx
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 20878232c52329f92423d27a60e48b6a6389e0dd ]
Systems show a minimal load average of 0.00, 0.01, 0.05 even when they
have no load at all.
Uptime and /proc/loadavg on all systems with kernels released during the
last five years up until kernel version 4.6-rc5, show a 5- and 15-minute
minimum loadavg of 0.01 and 0.05 respectively. This should be 0.00 on
idle systems, but the way the kernel calculates this value prevents it
from getting lower than the mentioned values.
Likewise but not as obviously noticeable, a fully loaded system with no
processes waiting, shows a maximum 1/5/15 loadavg of 1.00, 0.99, 0.95
(multiplied by number of cores).
Once the (old) load becomes 93 or higher, it mathematically can never
get lower than 93, even when the active (load) remains 0 forever.
This results in the strange 0.00, 0.01, 0.05 uptime values on idle
systems. Note: 93/2048 = 0.0454..., which rounds up to 0.05.
It is not correct to add a 0.5 rounding (=1024/2048) here, since the
result from this function is fed back into the next iteration again,
so the result of that +0.5 rounding value then gets multiplied by
(2048-2037), and then rounded again, so there is a virtual "ghost"
load created, next to the old and active load terms.
By changing the way the internally kept value is rounded, that internal
value equivalent now can reach 0.00 on idle, and 1.00 on full load. Upon
increasing load, the internally kept load value is rounded up, when the
load is decreasing, the load value is rounded down.
The modified code was tested on nohz=off and nohz kernels. It was tested
on vanilla kernel 4.6-rc5 and on centos 7.1 kernel 3.10.0-327. It was
tested on single, dual, and octal cores system. It was tested on virtual
hosts and bare hardware. No unwanted effects have been observed, and the
problems that the patch intended to fix were indeed gone.
Tested-by: Damien Wyart <damien.wyart@free.fr>
Signed-off-by: Vik Heyndrickx <vik.heyndrickx@veribox.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <stable@vger.kernel.org>
Cc: Doug Smythies <dsmythies@telus.net>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Fixes: 0f004f5a696a ("sched: Cure more NO_HZ load average woes")
Link: http://lkml.kernel.org/r/e8d32bff-d544-7748-72b5-3c86cc71f09f@veribox.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit f7c17d26f43d5cc1b7a6b896cd2fa24a079739b9 ] ------------[ cut here ]------------ WARNING: CPU: 0 PID: 16 at kernel/workqueue.c:4559 rebind_workers+0x1c0/0x1d0 Modules linked in: CPU: 0 PID: 16 Comm: cpuhp/0 Not tainted 4.6.0-rc4+ #31 Hardware name: IBM IBM System x3550 M4 Server -[7914IUW]-/00Y8603, BIOS -[D7E128FUS-1.40]- 07/23/2013 0000000000000000 ffff881037babb58 ffffffff8139d885 0000000000000010 0000000000000000 0000000000000000 0000000000000000 ffff881037babba8 ffffffff8108505d ffff881037ba0000 000011cf3e7d6e60 0000000000000046 Call Trace: dump_stack+0x89/0xd4 __warn+0xfd/0x120 warn_slowpath_null+0x1d/0x20 rebind_workers+0x1c0/0x1d0 workqueue_cpu_up_callback+0xf5/0x1d0 notifier_call_chain+0x64/0x90 ? trace_hardirqs_on_caller+0xf2/0x220 ? notify_prepare+0x80/0x80 __raw_notifier_call_chain+0xe/0x10 __cpu_notify+0x35/0x50 notify_down_prepare+0x5e/0x80 ? notify_prepare+0x80/0x80 cpuhp_invoke_callback+0x73/0x330 ? __schedule+0x33e/0x8a0 cpuhp_down_callbacks+0x51/0xc0 cpuhp_thread_fun+0xc1/0xf0 smpboot_thread_fn+0x159/0x2a0 ? smpboot_create_threads+0x80/0x80 kthread+0xef/0x110 ? wait_for_completion+0xf0/0x120 ? schedule_tail+0x35/0xf0 ret_from_fork+0x22/0x50 ? __init_kthread_worker+0x70/0x70 ---[ end trace eb12ae47d2382d8f ]--- notify_down_prepare: attempt to take down CPU 0 failed This bug can be reproduced by below config w/ nohz_full= all cpus: CONFIG_BOOTPARAM_HOTPLUG_CPU0=y CONFIG_DEBUG_HOTPLUG_CPU0=y CONFIG_NO_HZ_FULL=y As Thomas pointed out: | If a down prepare callback fails, then DOWN_FAILED is invoked for all | callbacks which have successfully executed DOWN_PREPARE. | | But, workqueue has actually two notifiers. One which handles | UP/DOWN_FAILED/ONLINE and one which handles DOWN_PREPARE. | | Now look at the priorities of those callbacks: | | CPU_PRI_WORKQUEUE_UP = 5 | CPU_PRI_WORKQUEUE_DOWN = -5 | | So the call order on DOWN_PREPARE is: | | CB 1 | CB ... | CB workqueue_up() -> Ignores DOWN_PREPARE | CB ... | CB X ---> Fails | | So we call up to CB X with DOWN_FAILED | | CB 1 | CB ... | CB workqueue_up() -> Handles DOWN_FAILED | CB ... | CB X-1 | | So the problem is that the workqueue stuff handles DOWN_FAILED in the up | callback, while it should do it in the down callback. Which is not a good idea | either because it wants to be called early on rollback... | | Brilliant stuff, isn't it? The hotplug rework will solve this problem because | the callbacks become symetric, but for the existing mess, we need some | workaround in the workqueue code. The boot CPU handles housekeeping duty(unbound timers, workqueues, timekeeping, ...) on behalf of full dynticks CPUs. It must remain online when nohz full is enabled. There is a priority set to every notifier_blocks: workqueue_cpu_up > tick_nohz_cpu_down > workqueue_cpu_down So tick_nohz_cpu_down callback failed when down prepare cpu 0, and notifier_blocks behind tick_nohz_cpu_down will not be called any more, which leads to workers are actually not unbound. Then hotplug state machine will fallback to undo and online cpu 0 again. Workers will be rebound unconditionally even if they are not unbound and trigger the warning in this progress. This patch fix it by catching !DISASSOCIATED to avoid rebind bound workers. Cc: Tejun Heo <tj@kernel.org> Cc: Lai Jiangshan <jiangshanlai@gmail.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Frédéric Weisbecker <fweisbec@gmail.com> Cc: stable@vger.kernel.org Suggested-by: Lai Jiangshan <jiangshanlai@gmail.com> Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 9f448cd3cbcec8995935e60b27802ae56aac8cc0 ] When the PMU driver reports a truncated AUX record, it effectively means that there is no more usable room in the event's AUX buffer (even though there may still be some room, so that perf_aux_output_begin() doesn't take action). At this point the consumer still has to be woken up and the event has to be disabled, otherwise the event will just keep spinning between perf_aux_output_begin() and perf_aux_output_end() until its context gets unscheduled. Again, for cpu-wide events this means never, so once in this condition, they will be forever losing data. Fix this by disabling the event and waking up the consumer in case of a truncated AUX record. Reported-by: Markus Metzger <markus.t.metzger@intel.com> Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: <stable@vger.kernel.org> Cc: Arnaldo Carvalho de Melo <acme@infradead.org> Cc: Arnaldo Carvalho de Melo <acme@redhat.com> Cc: Borislav Petkov <bp@alien8.de> Cc: Jiri Olsa <jolsa@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Stephane Eranian <eranian@google.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Vince Weaver <vincent.weaver@maine.edu> Cc: vince@deater.net Link: http://lkml.kernel.org/r/1462886313-13660-3-git-send-email-alexander.shishkin@linux.intel.com Signed-off-by: Ingo Molnar <mingo@kernel.org> Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 854145e0a8e9a05f7366d240e2f99d9c1ca6d6dd ]
Currently register functions for events will be called
through the 'reg' field of event class directly without
any check when seting up triggers.
Triggers for events that don't support register through
debug fs (events under events/ftrace are for trace-cmd to
read event format, and most of them don't have a register
function except events/ftrace/functionx) can't be enabled
at all, and an oops will be hit when setting up trigger
for those events, so just not creating them is an easy way
to avoid the oops.
Link: http://lkml.kernel.org/r/1462275274-3911-1-git-send-email-chuhu@redhat.com
Cc: stable@vger.kernel.org # 3.14+
Fixes: 85f2b08268c01 ("tracing: Add basic event trigger framework")
Signed-off-by: Chunyu Hu <chuhu@redhat.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>
[ Upstream commit 346c09f80459a3ad97df1816d6d606169a51001a ]
The bug in a workqueue leads to a stalled IO request in MQ ctx->rq_list
with the following backtrace:
[ 601.347452] INFO: task kworker/u129:5:1636 blocked for more than 120 seconds.
[ 601.347574] Tainted: G O 4.4.5-1-storage+ #6
[ 601.347651] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[ 601.348142] kworker/u129:5 D ffff880803077988 0 1636 2 0x00000000
[ 601.348519] Workqueue: ibnbd_server_fileio_wq ibnbd_dev_file_submit_io_worker [ibnbd_server]
[ 601.348999] ffff880803077988 ffff88080466b900 ffff8808033f9c80 ffff880803078000
[ 601.349662] ffff880807c95000 7fffffffffffffff ffffffff815b0920 ffff880803077ad0
[ 601.350333] ffff8808030779a0 ffffffff815b01d5 0000000000000000 ffff880803077a38
[ 601.350965] Call Trace:
[ 601.351203] [<ffffffff815b0920>] ? bit_wait+0x60/0x60
[ 601.351444] [<ffffffff815b01d5>] schedule+0x35/0x80
[ 601.351709] [<ffffffff815b2dd2>] schedule_timeout+0x192/0x230
[ 601.351958] [<ffffffff812d43f7>] ? blk_flush_plug_list+0xc7/0x220
[ 601.352208] [<ffffffff810bd737>] ? ktime_get+0x37/0xa0
[ 601.352446] [<ffffffff815b0920>] ? bit_wait+0x60/0x60
[ 601.352688] [<ffffffff815af784>] io_schedule_timeout+0xa4/0x110
[ 601.352951] [<ffffffff815b3a4e>] ? _raw_spin_unlock_irqrestore+0xe/0x10
[ 601.353196] [<ffffffff815b093b>] bit_wait_io+0x1b/0x70
[ 601.353440] [<ffffffff815b056d>] __wait_on_bit+0x5d/0x90
[ 601.353689] [<ffffffff81127bd0>] wait_on_page_bit+0xc0/0xd0
[ 601.353958] [<ffffffff81096db0>] ? autoremove_wake_function+0x40/0x40
[ 601.354200] [<ffffffff81127cc4>] __filemap_fdatawait_range+0xe4/0x140
[ 601.354441] [<ffffffff81127d34>] filemap_fdatawait_range+0x14/0x30
[ 601.354688] [<ffffffff81129a9f>] filemap_write_and_wait_range+0x3f/0x70
[ 601.354932] [<ffffffff811ced3b>] blkdev_fsync+0x1b/0x50
[ 601.355193] [<ffffffff811c82d9>] vfs_fsync_range+0x49/0xa0
[ 601.355432] [<ffffffff811cf45a>] blkdev_write_iter+0xca/0x100
[ 601.355679] [<ffffffff81197b1a>] __vfs_write+0xaa/0xe0
[ 601.355925] [<ffffffff81198379>] vfs_write+0xa9/0x1a0
[ 601.356164] [<ffffffff811c59d8>] kernel_write+0x38/0x50
The underlying device is a null_blk, with default parameters:
queue_mode = MQ
submit_queues = 1
Verification that nullb0 has something inflight:
root@pserver8:~# cat /sys/block/nullb0/inflight
0 1
root@pserver8:~# find /sys/block/nullb0/mq/0/cpu* -name rq_list -print -exec cat {} \;
...
/sys/block/nullb0/mq/0/cpu2/rq_list
CTX pending:
ffff8838038e2400
...
During debug it became clear that stalled request is always inserted in
the rq_list from the following path:
save_stack_trace_tsk + 34
blk_mq_insert_requests + 231
blk_mq_flush_plug_list + 281
blk_flush_plug_list + 199
wait_on_page_bit + 192
__filemap_fdatawait_range + 228
filemap_fdatawait_range + 20
filemap_write_and_wait_range + 63
blkdev_fsync + 27
vfs_fsync_range + 73
blkdev_write_iter + 202
__vfs_write + 170
vfs_write + 169
kernel_write + 56
So blk_flush_plug_list() was called with from_schedule == true.
If from_schedule is true, that means that finally blk_mq_insert_requests()
offloads execution of __blk_mq_run_hw_queue() and uses kblockd workqueue,
i.e. it calls kblockd_schedule_delayed_work_on().
That means, that we race with another CPU, which is about to execute
__blk_mq_run_hw_queue() work.
Further debugging shows the following traces from different CPUs:
CPU#0 CPU#1
---------------------------------- -------------------------------
reqeust A inserted
STORE hctx->ctx_map[0] bit marked
kblockd_schedule...() returns 1
<schedule to kblockd workqueue>
request B inserted
STORE hctx->ctx_map[1] bit marked
kblockd_schedule...() returns 0
*** WORK PENDING bit is cleared ***
flush_busy_ctxs() is executed, but
bit 1, set by CPU#1, is not observed
As a result request B pended forever.
This behaviour can be explained by speculative LOAD of hctx->ctx_map on
CPU#0, which is reordered with clear of PENDING bit and executed _before_
actual STORE of bit 1 on CPU#1.
The proper fix is an explicit full barrier <mfence>, which guarantees
that clear of PENDING bit is to be executed before all possible
speculative LOADS or STORES inside actual work function.
Signed-off-by: Roman Pen <roman.penyaev@profitbricks.com>
Cc: Gioh Kim <gi-oh.kim@profitbricks.com>
Cc: Michael Wang <yun.wang@profitbricks.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org
Signed-off-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Sasha Levin <sasha.levin@oracle.com>