Skip to contentSkip to conversation

seqcount: Add lockdep functionality to seqcount/seqlock structures

RFC PATCH In review John Stultz 3 participants 6 messages Sep 14, 2013 – Sep 17, 2013

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

RFC PATCH v2 1/2 seqcount: Add lockdep functionality to seqcount/seqlock structures

Currently seqlocks and seqcounts don't support lockdep.

After running across a seqcount related deadlock in the timekeeping code, I used a less-refined and more focused varient of this patch to narrow down the cause of the issue.

This is a first-pass attempt to properly enable lockdep functionality on seqlocks and seqcounts.

Since seqcounts are used in the vdso gettimeofday code, I've provided lockdep accessors.

I've also handled one cases where there were nested seqlock writers and there may be more edge cases.

Comments and feedback would be appreciated!

Cc Mathieu Desnoyers <[email protected]>
Cc Li Zefan <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>

---
v2:
* Update to new simplified lockdep.h
* vdso accessor simplifications
* removed needless preempt_disable
* removed unneeded ifdefs

 arch/x86/vdso/vclock_gettime.c |  8 ++---
 fs/dcache.c                    |  4 +--
 fs/fs_struct.c                 |  2 +-
 include/linux/init_task.h      |  8 ++---
 include/linux/lockdep.h        |  8 +++--
 include/linux/seqlock.h        | 79 ++++++++++++++++++++++++++++++++++++++----
 mm/filemap_xip.c               |  2 +-
 7 files changed, 91 insertions(+), 20 deletions(-)
arch/x86/vdso/vclock_gettime.c +4 -4
index 72074d5..2ada505 100644
@@ -178,7 +178,7 @@ notrace static int __always_inline do_realtime(struct timespec *ts)
 
 	ts->tv_nsec = 0;
 	do {
-		seq = read_seqcount_begin(&gtod->seq);
+		seq = read_seqcount_begin_no_lockdep(&gtod->seq);
 		mode = gtod->clock.vclock_mode;
 		ts->tv_sec = gtod->wall_time_sec;
 		ns = gtod->wall_time_snsec;
@@ -198,7 +198,7 @@ notrace static int do_monotonic(struct timespec *ts)
 
 	ts->tv_nsec = 0;
 	do {
-		seq = read_seqcount_begin(&gtod->seq);
+		seq = read_seqcount_begin_no_lockdep(&gtod->seq);
 		mode = gtod->clock.vclock_mode;
 		ts->tv_sec = gtod->monotonic_time_sec;
 		ns = gtod->monotonic_time_snsec;
@@ -214,7 +214,7 @@ notrace static int do_realtime_coarse(struct timespec *ts)
 {
 	unsigned long seq;
 	do {
-		seq = read_seqcount_begin(&gtod->seq);
+		seq = read_seqcount_begin_no_lockdep(&gtod->seq);
 		ts->tv_sec = gtod->wall_time_coarse.tv_sec;
 		ts->tv_nsec = gtod->wall_time_coarse.tv_nsec;
 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
@@ -225,7 +225,7 @@ notrace static int do_monotonic_coarse(struct timespec *ts)
 {
 	unsigned long seq;
 	do {
-		seq = read_seqcount_begin(&gtod->seq);
+		seq = read_seqcount_begin_no_lockdep(&gtod->seq);
 		ts->tv_sec = gtod->monotonic_time_coarse.tv_sec;
 		ts->tv_nsec = gtod->monotonic_time_coarse.tv_nsec;
 	} while (unlikely(read_seqcount_retry(&gtod->seq, seq)));
fs/dcache.c +2 -2
index ca8e9cd..5e59bd3 100644
@@ -2396,7 +2396,7 @@ static void __d_move(struct dentry * dentry, struct dentry * target)
 	dentry_lock_for_move(dentry, target);
 
 	write_seqcount_begin(&dentry->d_seq);
-	write_seqcount_begin(&target->d_seq);
+	write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
 
 	/* __d_drop does write_seqcount_barrier, but they're OK to nest. */
 
@@ -2528,7 +2528,7 @@ static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
 	dentry_lock_for_move(anon, dentry);
 
 	write_seqcount_begin(&dentry->d_seq);
-	write_seqcount_begin(&anon->d_seq);
+	write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
 
 	dparent = dentry->d_parent;
 
fs/fs_struct.c +1 -1
index d8ac61d..7dca743 100644
@@ -161,6 +161,6 @@ EXPORT_SYMBOL(current_umask);
 struct fs_struct init_fs = {
 	.users		= 1,
 	.lock		= __SPIN_LOCK_UNLOCKED(init_fs.lock),
-	.seq		= SEQCNT_ZERO,
+	.seq		= SEQCNT_ZERO(init_fs.seq),
 	.umask		= 0022,
 };
include/linux/init_task.h +4 -4
index 5cd0f09..b0ed422 100644
@@ -32,10 +32,10 @@ extern struct fs_struct init_fs;
 #endif
 
 #ifdef CONFIG_CPUSETS
-#define INIT_CPUSET_SEQ							\
-	.mems_allowed_seq = SEQCNT_ZERO,
+#define INIT_CPUSET_SEQ(tsk)							\
+	.mems_allowed_seq = SEQCNT_ZERO(tsk.mems_allowed_seq),
 #else
-#define INIT_CPUSET_SEQ
+#define INIT_CPUSET_SEQ(tsk)
 #endif
 
 #define INIT_SIGNALS(sig) {						\
@@ -220,7 +220,7 @@ extern struct task_group root_task_group;
 	INIT_FTRACE_GRAPH						\
 	INIT_TRACE_RECURSION						\
 	INIT_TASK_RCU_PREEMPT(tsk)					\
-	INIT_CPUSET_SEQ							\
+	INIT_CPUSET_SEQ(tsk)						\
 	INIT_VTIME(tsk)							\
 }
 
include/linux/lockdep.h +6 -2
index cfc2f11..92b1bfc 100644
@@ -497,6 +497,10 @@ static inline void print_irqtrace_events(struct task_struct *curr)
 #define rwlock_acquire_read(l, s, t, i)		lock_acquire_shared_recursive(l, s, t, NULL, i)
 #define rwlock_release(l, n, i)			lock_release(l, n, i)
 
+#define seqcount_acquire(l, s, t, i)		lock_acquire_exclusive(l, s, t, NULL, i)
+#define seqcount_acquire_read(l, s, t, i)	lock_acquire_shared_recursive(l, s, t, NULL, i)
+#define seqcount_release(l, n, i)		lock_release(l, n, i)
+
 #define mutex_acquire(l, s, t, i)		lock_acquire_exclusive(l, s, t, NULL, i)
 #define mutex_acquire_nest(l, s, t, n, i)	lock_acquire_exclusive(l, s, t, n, i)
 #define mutex_release(l, n, i)			lock_release(l, n, i)
@@ -504,11 +508,11 @@ static inline void print_irqtrace_events(struct task_struct *curr)
 #define rwsem_acquire(l, s, t, i)		lock_acquire_exclusive(l, s, t, NULL, i)
 #define rwsem_acquire_nest(l, s, t, n, i)	lock_acquire_exclusive(l, s, t, n, i)
 #define rwsem_acquire_read(l, s, t, i)		lock_acquire_shared(l, s, t, NULL, i)
-# define rwsem_release(l, n, i)			lock_release(l, n, i)
+#define rwsem_release(l, n, i)			lock_release(l, n, i)
 
 #define lock_map_acquire(l)			lock_acquire_exclusive(l, 0, 0, NULL, _THIS_IP_)
 #define lock_map_acquire_read(l)		lock_acquire_shared_recursive(l, 0, 0, NULL, _THIS_IP_)
-# define lock_map_release(l)			lock_release(l, 1, _THIS_IP_)
+#define lock_map_release(l)			lock_release(l, 1, _THIS_IP_)
 
 #ifdef CONFIG_PROVE_LOCKING
 # define might_lock(lock) 						\
include/linux/seqlock.h +73 -6

127 lines not on this page · raw message

mm/filemap_xip.c +1 -1
index 28fe26b..d8d9fe3 100644
@@ -26,7 +26,7 @@
  * of ZERO_PAGE(), such as /dev/zero
  */
 static DEFINE_MUTEX(xip_sparse_mutex);
-static seqcount_t xip_sparse_seq = SEQCNT_ZERO;
+static seqcount_t xip_sparse_seq = SEQCNT_ZERO(xip_sparse_seq);
 static struct page *__xip_sparse_page;
 
 /* called under xip_sparse_mutex */
Signature · 1 line
1.8.1.2
#1 lore raw

RFC PATCH 2/2 cpuset: Fix potential deadlock w/ set_mems_allowed

After adding lockdep support to seqlock/seqcount structures, I started seeing the following warning:

[    1.070907] ======================================================
[    1.072015] [ INFO: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected ]
[    1.073181] 3.11.0+ #67 Not tainted
[    1.073801] ------------------------------------------------------
[    1.074882] kworker/u4:2/708 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
[    1.076088]  (&p->mems_allowed_seq){+.+...}, at: [<ffffffff81187d7f>] new_slab+0x5f/0x280
[    1.077572]
[    1.077572] and this task is already holding:
[    1.078593]  (&(&q->__queue_lock)->rlock){..-...}, at: [<ffffffff81339f03>] blk_execute_rq_nowait+0x53/0xf0
[    1.080042] which would create a new lock dependency:
[    1.080042]  (&(&q->__queue_lock)->rlock){..-...} -> (&p->mems_allowed_seq){+.+...}
[    1.080042]
[    1.080042] but this new dependency connects a SOFTIRQ-irq-safe lock:
[    1.080042]  (&(&q->__queue_lock)->rlock){..-...}
[    1.080042] ... which became SOFTIRQ-irq-safe at:
[    1.080042]   [<ffffffff810ec179>] __lock_acquire+0x5b9/0x1db0
[    1.080042]   [<ffffffff810edfe5>] lock_acquire+0x95/0x130
[    1.080042]   [<ffffffff818968a1>] _raw_spin_lock+0x41/0x80
[    1.080042]   [<ffffffff81560c9e>] scsi_device_unbusy+0x7e/0xd0
[    1.080042]   [<ffffffff8155a612>] scsi_finish_command+0x32/0xf0
[    1.080042]   [<ffffffff81560e91>] scsi_softirq_done+0xa1/0x130
[    1.080042]   [<ffffffff8133b0f3>] blk_done_softirq+0x73/0x90
[    1.080042]   [<ffffffff81095dc0>] __do_softirq+0x110/0x2f0
[    1.080042]   [<ffffffff81095fcd>] run_ksoftirqd+0x2d/0x60
[    1.080042]   [<ffffffff810bc506>] smpboot_thread_fn+0x156/0x1e0
[    1.080042]   [<ffffffff810b3916>] kthread+0xd6/0xe0
[    1.080042]   [<ffffffff818980ac>] ret_from_fork+0x7c/0xb0
[    1.080042]
[    1.080042] to a SOFTIRQ-irq-unsafe lock:
[    1.080042]  (&p->mems_allowed_seq){+.+...}
[    1.080042] ... which became SOFTIRQ-irq-unsafe at:
[    1.080042] ...  [<ffffffff810ec1d3>] __lock_acquire+0x613/0x1db0
[    1.080042]   [<ffffffff810edfe5>] lock_acquire+0x95/0x130
[    1.080042]   [<ffffffff810b3df2>] kthreadd+0x82/0x180
[    1.080042]   [<ffffffff818980ac>] ret_from_fork+0x7c/0xb0
[    1.080042]
[    1.080042] other info that might help us debug this:
[    1.080042]
[    1.080042]  Possible interrupt unsafe locking scenario:
[    1.080042]
[    1.080042]        CPU0                    CPU1
[    1.080042]        ----                    ----
[    1.080042]   lock(&p->mems_allowed_seq);
[    1.080042]                                local_irq_disable();
[    1.080042]                                lock(&(&q->__queue_lock)->rlock);
[    1.080042]                                lock(&p->mems_allowed_seq);
[    1.080042]   <Interrupt>
[    1.080042]     lock(&(&q->__queue_lock)->rlock);
[    1.080042]
[    1.080042]  *** DEADLOCK ***

The issue stems from the kthreadd() function calling set_mems_allowed with irqs enabled. While its possibly unlikely for the actual deadlock to trigger, a fix is fairly simple: disable irqs before taking the mems_allowed_seq lock.

Let me know if you have any other suggestions or alternative fixes you'd prefer.

Cc Li Zefan <[email protected]>
Cc Mathieu Desnoyers <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
---
 include/linux/cpuset.h | 4 ++++
 1 file changed, 4 insertions(+)
include/linux/cpuset.h +4 -0
index cc1b01c..3fe661f 100644
@@ -110,10 +110,14 @@ static inline bool put_mems_allowed(unsigned int seq)
 
 static inline void set_mems_allowed(nodemask_t nodemask)
 {
+	unsigned long flags;
+
 	task_lock(current);
+	local_irq_save(flags);
 	write_seqcount_begin(&current->mems_allowed_seq);
 	current->mems_allowed = nodemask;
 	write_seqcount_end(&current->mems_allowed_seq);
+	local_irq_restore(flags);
 	task_unlock(current);
 }
 
Signature · 1 line
1.8.1.2
Acked-by on v1 2/2
#2 lore raw

Re: [PATCH 2/2] [RFC] cpuset: Fix potential deadlock w/ set_mems_allowed

Cc Mel, who added seqcount to cpuset.

Quoted 54 lines · John Stultz

After adding lockdep support to seqlock/seqcount structures, I started seeing the following warning:

[    1.070907] ======================================================
[    1.072015] [ INFO: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected ]
[    1.073181] 3.11.0+ #67 Not tainted
[    1.073801] ------------------------------------------------------
[    1.074882] kworker/u4:2/708 [HC0[0]:SC0[0]:HE0:SE1] is trying to acquire:
[    1.076088]  (&p->mems_allowed_seq){+.+...}, at: [<ffffffff81187d7f>] new_slab+0x5f/0x280
[    1.077572]
[    1.077572] and this task is already holding:
[    1.078593]  (&(&q->__queue_lock)->rlock){..-...}, at: [<ffffffff81339f03>] blk_execute_rq_nowait+0x53/0xf0
[    1.080042] which would create a new lock dependency:
[    1.080042]  (&(&q->__queue_lock)->rlock){..-...} -> (&p->mems_allowed_seq){+.+...}
[    1.080042]
[    1.080042] but this new dependency connects a SOFTIRQ-irq-safe lock:
[    1.080042]  (&(&q->__queue_lock)->rlock){..-...}
[    1.080042] ... which became SOFTIRQ-irq-safe at:
[    1.080042]   [<ffffffff810ec179>] __lock_acquire+0x5b9/0x1db0
[    1.080042]   [<ffffffff810edfe5>] lock_acquire+0x95/0x130
[    1.080042]   [<ffffffff818968a1>] _raw_spin_lock+0x41/0x80
[    1.080042]   [<ffffffff81560c9e>] scsi_device_unbusy+0x7e/0xd0
[    1.080042]   [<ffffffff8155a612>] scsi_finish_command+0x32/0xf0
[    1.080042]   [<ffffffff81560e91>] scsi_softirq_done+0xa1/0x130
[    1.080042]   [<ffffffff8133b0f3>] blk_done_softirq+0x73/0x90
[    1.080042]   [<ffffffff81095dc0>] __do_softirq+0x110/0x2f0
[    1.080042]   [<ffffffff81095fcd>] run_ksoftirqd+0x2d/0x60
[    1.080042]   [<ffffffff810bc506>] smpboot_thread_fn+0x156/0x1e0
[    1.080042]   [<ffffffff810b3916>] kthread+0xd6/0xe0
[    1.080042]   [<ffffffff818980ac>] ret_from_fork+0x7c/0xb0
[    1.080042]
[    1.080042] to a SOFTIRQ-irq-unsafe lock:
[    1.080042]  (&p->mems_allowed_seq){+.+...}
[    1.080042] ... which became SOFTIRQ-irq-unsafe at:
[    1.080042] ...  [<ffffffff810ec1d3>] __lock_acquire+0x613/0x1db0
[    1.080042]   [<ffffffff810edfe5>] lock_acquire+0x95/0x130
[    1.080042]   [<ffffffff810b3df2>] kthreadd+0x82/0x180
[    1.080042]   [<ffffffff818980ac>] ret_from_fork+0x7c/0xb0
[    1.080042]
[    1.080042] other info that might help us debug this:
[    1.080042]
[    1.080042]  Possible interrupt unsafe locking scenario:
[    1.080042]
[    1.080042]        CPU0                    CPU1
[    1.080042]        ----                    ----
[    1.080042]   lock(&p->mems_allowed_seq);
[    1.080042]                                local_irq_disable();
[    1.080042]                                lock(&(&q->__queue_lock)->rlock);
[    1.080042]                                lock(&p->mems_allowed_seq);
[    1.080042]   <Interrupt>
[    1.080042]     lock(&(&q->__queue_lock)->rlock);
[    1.080042]
[    1.080042]  *** DEADLOCK ***
Quoted 5 lines

The issue stems from the kthreadd() function calling set_mems_allowed with irqs enabled. While its possibly unlikely for the actual deadlock to trigger, a fix is fairly simple: disable irqs before taking the mems_allowed_seq lock.

Now I get it. I'm fine with this change.

Quoted 34 lines

Let me know if you have any other suggestions or alternative fixes you'd prefer.

Cc Li Zefan <[email protected]>
Cc Mathieu Desnoyers <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>
---
 include/linux/cpuset.h | 4 ++++
 1 file changed, 4 insertions(+)
include/linux/cpuset.h +4 -0
index cc1b01c..3fe661f 100644
@@ -110,10 +110,14 @@ static inline bool put_mems_allowed(unsigned int seq)
 
 static inline void set_mems_allowed(nodemask_t nodemask)
 {
+	unsigned long flags;
+
 	task_lock(current);
+	local_irq_save(flags);
 	write_seqcount_begin(&current->mems_allowed_seq);
 	current->mems_allowed = nodemask;
 	write_seqcount_end(&current->mems_allowed_seq);
+	local_irq_restore(flags);
 	task_unlock(current);
 }
 
#3 lore raw
In reply to John Stultz
Quoted 9 lines · John Stultz

Currently seqlocks and seqcounts don't support lockdep.

After running across a seqcount related deadlock in the timekeeping code, I used a less-refined and more focused varient of this patch to narrow down the cause of the issue.

This is a first-pass attempt to properly enable lockdep functionality on seqlocks and seqcounts.

Quoted 8 lines

Since seqcounts are used in the vdso gettimeofday code, I've provided lockdep accessors.

I've also handled one cases where there were nested seqlock writers and there may be more edge cases.

Comments and feedback would be appreciated!

Could you describe how seqlocks/seqcounts can lead to deadlock in the changelog?

Quoted 13 lines
Cc Mathieu Desnoyers <[email protected]>
Cc Li Zefan <[email protected]>
Cc Steven Rostedt <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Ingo Molnar <[email protected]>
Cc Thomas Gleixner <[email protected]>

---
v2:
* Update to new simplified lockdep.h
* vdso accessor simplifications
* removed needless preempt_disable
* removed unneeded ifdefs

#4 lore raw
In reply to John Stultz
Quoted 6 lines · John Stultz

Currently seqlocks and seqcounts don't support lockdep.

After running across a seqcount related deadlock in the timekeeping code, I used a less-refined and more focused varient of this patch to narrow down the cause of the issue.

Quoted 8 lines

This is a first-pass attempt to properly enable lockdep functionality on seqlocks and seqcounts.

Since seqcounts are used in the vdso gettimeofday code, I've provided lockdep accessors.

I've also handled one cases where there were nested seqlock writers and there may be more edge cases.

Oof.

So I just noticed there's a bunch of places in the network code that use fairly deeply embedded seqcounter: u64_stats_sync. There's almost never an explicit initialization, as they assume they're zeroed when allocated, but this causes trouble with the lockdep key initialization.

I'll have to go through each of these (about 25 cases) and make them call seqcount_init(), but since I'm heading to plumbers tomorrow I might not get to it until next week.

Anyway, let me know if you have any other thoughts on the patches.

thanks
-john

#5 lore raw
Quoted 23 lines · John Stultz
Quoted 6 lines · John Stultz

Currently seqlocks and seqcounts don't support lockdep.

After running across a seqcount related deadlock in the timekeeping code, I used a less-refined and more focused varient of this patch to narrow down the cause of the issue.

Quoted 8 lines

This is a first-pass attempt to properly enable lockdep functionality on seqlocks and seqcounts.

Since seqcounts are used in the vdso gettimeofday code, I've provided lockdep accessors.

I've also handled one cases where there were nested seqlock writers and there may be more edge cases.

Oof.

So I just noticed there's a bunch of places in the network code that use fairly deeply embedded seqcounter: u64_stats_sync. There's almost never an explicit initialization, as they assume they're zeroed when allocated, but this causes trouble with the lockdep key initialization.

Quoted 5 lines

I'll have to go through each of these (about 25 cases) and make them call seqcount_init(), but since I'm heading to plumbers tomorrow I might not get to it until next week.

Anyway, let me know if you have any other thoughts on the patches.

Explicit initialization is generally a bonus for readability, debuggability and ease of development, we enforce that for spinlocks as well.

Thanks,

Ingo