Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
docs: RCU: Convert lockdep-splat.txt to ReST
- Add a SPDX header;
- Add a document title;
- Some whitespace fixes and new line breaks;
- Mark literal blocks as such;
- Add it to RCU/index.rst.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Signed-off-by: Paul E. McKenney <paulmck@kernel.org>
authored by
Mauro Carvalho Chehab
and committed by
Paul E. McKenney
a3b0a79f
6b05dfac
+53
-47
+1
Documentation/RCU/index.rst
+1
Documentation/RCU/index.rst
+52
-47
Documentation/RCU/lockdep-splat.txt
Documentation/RCU/lockdep-splat.rst
+52
-47
Documentation/RCU/lockdep-splat.txt
Documentation/RCU/lockdep-splat.rst
···
1
+
.. SPDX-License-Identifier: GPL-2.0
2
+
3
+
=================
4
+
Lockdep-RCU Splat
5
+
=================
6
+
1
7
Lockdep-RCU was added to the Linux kernel in early 2010
2
8
(http://lwn.net/Articles/371986/). This facility checks for some common
3
9
misuses of the RCU API, most notably using one of the rcu_dereference()
···
18
12
being the real world and all that.
19
13
20
14
So let's look at an example RCU lockdep splat from 3.0-rc5, one that
21
-
has long since been fixed:
15
+
has long since been fixed::
22
16
23
-
=============================
24
-
WARNING: suspicious RCU usage
25
-
-----------------------------
26
-
block/cfq-iosched.c:2776 suspicious rcu_dereference_protected() usage!
17
+
=============================
18
+
WARNING: suspicious RCU usage
19
+
-----------------------------
20
+
block/cfq-iosched.c:2776 suspicious rcu_dereference_protected() usage!
27
21
28
-
other info that might help us debug this:
22
+
other info that might help us debug this::
29
23
24
+
rcu_scheduler_active = 1, debug_locks = 0
25
+
3 locks held by scsi_scan_6/1552:
26
+
#0: (&shost->scan_mutex){+.+.}, at: [<ffffffff8145efca>]
27
+
scsi_scan_host_selected+0x5a/0x150
28
+
#1: (&eq->sysfs_lock){+.+.}, at: [<ffffffff812a5032>]
29
+
elevator_exit+0x22/0x60
30
+
#2: (&(&q->__queue_lock)->rlock){-.-.}, at: [<ffffffff812b6233>]
31
+
cfq_exit_queue+0x43/0x190
30
32
31
-
rcu_scheduler_active = 1, debug_locks = 0
32
-
3 locks held by scsi_scan_6/1552:
33
-
#0: (&shost->scan_mutex){+.+.}, at: [<ffffffff8145efca>]
34
-
scsi_scan_host_selected+0x5a/0x150
35
-
#1: (&eq->sysfs_lock){+.+.}, at: [<ffffffff812a5032>]
36
-
elevator_exit+0x22/0x60
37
-
#2: (&(&q->__queue_lock)->rlock){-.-.}, at: [<ffffffff812b6233>]
38
-
cfq_exit_queue+0x43/0x190
33
+
stack backtrace:
34
+
Pid: 1552, comm: scsi_scan_6 Not tainted 3.0.0-rc5 #17
35
+
Call Trace:
36
+
[<ffffffff810abb9b>] lockdep_rcu_dereference+0xbb/0xc0
37
+
[<ffffffff812b6139>] __cfq_exit_single_io_context+0xe9/0x120
38
+
[<ffffffff812b626c>] cfq_exit_queue+0x7c/0x190
39
+
[<ffffffff812a5046>] elevator_exit+0x36/0x60
40
+
[<ffffffff812a802a>] blk_cleanup_queue+0x4a/0x60
41
+
[<ffffffff8145cc09>] scsi_free_queue+0x9/0x10
42
+
[<ffffffff81460944>] __scsi_remove_device+0x84/0xd0
43
+
[<ffffffff8145dca3>] scsi_probe_and_add_lun+0x353/0xb10
44
+
[<ffffffff817da069>] ? error_exit+0x29/0xb0
45
+
[<ffffffff817d98ed>] ? _raw_spin_unlock_irqrestore+0x3d/0x80
46
+
[<ffffffff8145e722>] __scsi_scan_target+0x112/0x680
47
+
[<ffffffff812c690d>] ? trace_hardirqs_off_thunk+0x3a/0x3c
48
+
[<ffffffff817da069>] ? error_exit+0x29/0xb0
49
+
[<ffffffff812bcc60>] ? kobject_del+0x40/0x40
50
+
[<ffffffff8145ed16>] scsi_scan_channel+0x86/0xb0
51
+
[<ffffffff8145f0b0>] scsi_scan_host_selected+0x140/0x150
52
+
[<ffffffff8145f149>] do_scsi_scan_host+0x89/0x90
53
+
[<ffffffff8145f170>] do_scan_async+0x20/0x160
54
+
[<ffffffff8145f150>] ? do_scsi_scan_host+0x90/0x90
55
+
[<ffffffff810975b6>] kthread+0xa6/0xb0
56
+
[<ffffffff817db154>] kernel_thread_helper+0x4/0x10
57
+
[<ffffffff81066430>] ? finish_task_switch+0x80/0x110
58
+
[<ffffffff817d9c04>] ? retint_restore_args+0xe/0xe
59
+
[<ffffffff81097510>] ? __kthread_init_worker+0x70/0x70
60
+
[<ffffffff817db150>] ? gs_change+0xb/0xb
39
61
40
-
stack backtrace:
41
-
Pid: 1552, comm: scsi_scan_6 Not tainted 3.0.0-rc5 #17
42
-
Call Trace:
43
-
[<ffffffff810abb9b>] lockdep_rcu_dereference+0xbb/0xc0
44
-
[<ffffffff812b6139>] __cfq_exit_single_io_context+0xe9/0x120
45
-
[<ffffffff812b626c>] cfq_exit_queue+0x7c/0x190
46
-
[<ffffffff812a5046>] elevator_exit+0x36/0x60
47
-
[<ffffffff812a802a>] blk_cleanup_queue+0x4a/0x60
48
-
[<ffffffff8145cc09>] scsi_free_queue+0x9/0x10
49
-
[<ffffffff81460944>] __scsi_remove_device+0x84/0xd0
50
-
[<ffffffff8145dca3>] scsi_probe_and_add_lun+0x353/0xb10
51
-
[<ffffffff817da069>] ? error_exit+0x29/0xb0
52
-
[<ffffffff817d98ed>] ? _raw_spin_unlock_irqrestore+0x3d/0x80
53
-
[<ffffffff8145e722>] __scsi_scan_target+0x112/0x680
54
-
[<ffffffff812c690d>] ? trace_hardirqs_off_thunk+0x3a/0x3c
55
-
[<ffffffff817da069>] ? error_exit+0x29/0xb0
56
-
[<ffffffff812bcc60>] ? kobject_del+0x40/0x40
57
-
[<ffffffff8145ed16>] scsi_scan_channel+0x86/0xb0
58
-
[<ffffffff8145f0b0>] scsi_scan_host_selected+0x140/0x150
59
-
[<ffffffff8145f149>] do_scsi_scan_host+0x89/0x90
60
-
[<ffffffff8145f170>] do_scan_async+0x20/0x160
61
-
[<ffffffff8145f150>] ? do_scsi_scan_host+0x90/0x90
62
-
[<ffffffff810975b6>] kthread+0xa6/0xb0
63
-
[<ffffffff817db154>] kernel_thread_helper+0x4/0x10
64
-
[<ffffffff81066430>] ? finish_task_switch+0x80/0x110
65
-
[<ffffffff817d9c04>] ? retint_restore_args+0xe/0xe
66
-
[<ffffffff81097510>] ? __kthread_init_worker+0x70/0x70
67
-
[<ffffffff817db150>] ? gs_change+0xb/0xb
68
-
69
-
Line 2776 of block/cfq-iosched.c in v3.0-rc5 is as follows:
62
+
Line 2776 of block/cfq-iosched.c in v3.0-rc5 is as follows::
70
63
71
64
if (rcu_dereference(ioc->ioc_data) == cic) {
72
65
···
75
70
And maybe that lock really does protect this reference. If so, the fix
76
71
is to inform RCU, perhaps by changing __cfq_exit_single_io_context() to
77
72
take the struct request_queue "q" from cfq_exit_queue() as an argument,
78
-
which would permit us to invoke rcu_dereference_protected as follows:
73
+
which would permit us to invoke rcu_dereference_protected as follows::
79
74
80
75
if (rcu_dereference_protected(ioc->ioc_data,
81
76
lockdep_is_held(&q->queue_lock)) == cic) {
···
90
85
section. In this case, the critical section must span the use of the
91
86
return value from rcu_dereference(), or at least until there is some
92
87
reference count incremented or some such. One way to handle this is to
93
-
add rcu_read_lock() and rcu_read_unlock() as follows:
88
+
add rcu_read_lock() and rcu_read_unlock() as follows::
94
89
95
90
rcu_read_lock();
96
91
if (rcu_dereference(ioc->ioc_data) == cic) {
···
107
102
But in this particular case, we don't actually dereference the pointer
108
103
returned from rcu_dereference(). Instead, that pointer is just compared
109
104
to the cic pointer, which means that the rcu_dereference() can be replaced
110
-
by rcu_access_pointer() as follows:
105
+
by rcu_access_pointer() as follows::
111
106
112
107
if (rcu_access_pointer(ioc->ioc_data) == cic) {
113
108