Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[Qian Cai]

On Fri, 2020-09-04 at 17:32 +0200, Ahmed S. Darwish wrote:
> Hi,
> 
> Changelog-v2
> ============
> 
>   - Standardize on seqcount_LOCKNAME_t as the canonical reference for
>     sequence counters with associated locks, instead of v1
>     seqcount_LOCKTYPE_t.
> 
>   - Use unique prefix "seqprop_*" for all seqcount_t/seqcount_LOCKNAME_t
>     property accessors.
> 
>   - Touch-up the lock-unlock rationale for more clarity. Enforce writer
>     non-preemitiblity using "__seq_enforce_writer_non_preemptibility()".
> 
> Cover letter (v1)
> =================
> 
> https://lkml.kernel.org/r/20200828010710.5407-1-a.darwish@linutronix.de
> 
> Preemption must be disabled before entering a sequence counter write
> side critical section.  Otherwise the read side section can preempt the
> write side section and spin for the entire scheduler tick.  If that
> reader belongs to a real-time scheduling class, it can spin forever and
> the kernel will livelock.
> 
> Disabling preemption cannot be done for PREEMPT_RT though: it can lead
> to higher latencies, and the write side sections will not be able to
> acquire locks which become sleeping locks (e.g. spinlock_t).
> 
> To remain preemptible, while avoiding a possible livelock caused by the
> reader preempting the writer, use a different technique: let the reader
> detect if a seqcount_LOCKNAME_t writer is in progress. If that's the
> case, acquire then release the associated LOCKNAME writer serialization
> lock. This will allow any possibly-preempted writer to make progress
> until the end of its writer serialization lock critical section.
> 
> Implement this lock-unlock technique for all seqcount_LOCKNAME_t with
> an associated (PREEMPT_RT) sleeping lock, and for seqlock_t.

Reverting this patchset [1] from today's linux-next fixed a splat below. The
splat looks like a false positive anyway because the existing locking dependency
chains from the task #1 here:

&s->seqcount#2 ---> pidmap_lock

[  528.078061][ T7867] -> #1 (pidmap_lock){....}-{2:2}:
[  528.078078][ T7867]        lock_acquire+0x10c/0x560
[  528.078089][ T7867]        _raw_spin_lock_irqsave+0x64/0xb0
[  528.078108][ T7867]        free_pid+0x5c/0x160
free_pid at kernel/pid.c:131
[  528.078127][ T7867]        release_task.part.40+0x59c/0x7f0
__unhash_process at kernel/exit.c:76
(inlined by) __exit_signal at kernel/exit.c:147
(inlined by) release_task at kernel/exit.c:198
[  528.078145][ T7867]        do_exit+0x77c/0xda0
exit_notify at kernel/exit.c:679
(inlined by) do_exit at kernel/exit.c:826
[  528.078163][ T7867]        kthread+0x148/0x1d0
[  528.078182][ T7867]        ret_from_kernel_thread+0x5c/0x80

It is write_seqlock(&sig->stats_lock) in __exit_signal(), but the &s->seqcount#2 
in read_mems_allowed_begin() is read_seqcount_begin(&current->mems_allowed_seq), 
so there should be no deadlock?

[1] git revert --no-edit 0c9794c8b678..1909760f5fc3

[  528.077900][ T7867] WARNING: possible circular locking dependency detected
[  528.077912][ T7867] 5.9.0-rc5-next-20200914 #1 Not tainted
[  528.077921][ T7867] ------------------------------------------------------
[  528.077931][ T7867] runc:[1:CHILD]/7867 is trying to acquire lock:
[  528.077942][ T7867] c000001fce5570c8 (&s->seqcount#2){....}-{0:0}, at: __slab_alloc+0x34/0xf0
[  528.077972][ T7867] 
[  528.077972][ T7867] but task is already holding lock:
[  528.077983][ T7867] c0000000056b0198 (pidmap_lock){....}-{2:2}, at: alloc_pid+0x258/0x590
[  528.078009][ T7867] 
[  528.078009][ T7867] which lock already depends on the new lock.
[  528.078009][ T7867] 
[  528.078031][ T7867] 
[  528.078031][ T7867] the existing dependency chain (in reverse order) is:
[  528.078061][ T7867] 
[  528.078061][ T7867] -> #1 (pidmap_lock){....}-{2:2}:
[  528.078078][ T7867]        lock_acquire+0x10c/0x560
[  528.078089][ T7867]        _raw_spin_lock_irqsave+0x64/0xb0
[  528.078108][ T7867]        free_pid+0x5c/0x160
free_pid at kernel/pid.c:131
[  528.078127][ T7867]        release_task.part.40+0x59c/0x7f0
__unhash_process at kernel/exit.c:76
(inlined by) __exit_signal at kernel/exit.c:147
(inlined by) release_task at kernel/exit.c:198
[  528.078145][ T7867]        do_exit+0x77c/0xda0
exit_notify at kernel/exit.c:679
(inlined by) do_exit at kernel/exit.c:826
[  528.078163][ T7867]        kthread+0x148/0x1d0
[  528.078182][ T7867]        ret_from_kernel_thread+0x5c/0x80
[  528.078208][ T7867] 
[  528.078208][ T7867] -> #0 (&s->seqcount#2){....}-{0:0}:
[  528.078241][ T7867]        check_prevs_add+0x1c4/0x1120
check_prev_add at kernel/locking/lockdep.c:2820
(inlined by) check_prevs_add at kernel/locking/lockdep.c:2944
[  528.078260][ T7867]        __lock_acquire+0x176c/0x1c00
validate_chain at kernel/locking/lockdep.c:3562
(inlined by) __lock_acquire at kernel/locking/lockdep.c:4796
[  528.078278][ T7867]        lock_acquire+0x10c/0x560
[  528.078297][ T7867]        ___slab_alloc+0xa40/0xb40
seqcount_lockdep_reader_access at include/linux/seqlock.h:103
(inlined by) read_mems_allowed_begin at include/linux/cpuset.h:135
(inlined by) get_any_partial at mm/slub.c:2035
(inlined by) get_partial at mm/slub.c:2078
(inlined by) new_slab_objects at mm/slub.c:2577
(inlined by) ___slab_alloc at mm/slub.c:2745
[  528.078324][ T7867]        __slab_alloc+0x34/0xf0
[  528.078342][ T7867]        kmem_cache_alloc+0x2d4/0x470
[  528.078362][ T7867]        create_object+0x74/0x430
[  528.078381][ T7867]        slab_post_alloc_hook+0xa4/0x670
[  528.078399][ T7867]        kmem_cache_alloc+0x1b4/0x470
[  528.078418][ T7867]        radix_tree_node_alloc.constprop.19+0xe4/0x160
[  528.078438][ T7867]        idr_get_free+0x298/0x360
[  528.078456][ T7867]        idr_alloc_u32+0x84/0x130
[  528.078474][ T7867]        idr_alloc_cyclic+0x7c/0x150
[  528.078493][ T7867]        alloc_pid+0x27c/0x590
[  528.078511][ T7867]        copy_process+0xc90/0x1930
copy_process at kernel/fork.c:2104
[  528.078529][ T7867]        kernel_clone+0x120/0xa10
[  528.078546][ T7867]        __do_sys_clone+0x88/0xd0
[  528.078565][ T7867]        system_call_exception+0xf8/0x1d0
[  528.078592][ T7867]        system_call_common+0xe8/0x218
[  528.078609][ T7867] 
[  528.078609][ T7867] other info that might help us debug this:
[  528.078609][ T7867] 
[  528.078650][ T7867]  Possible unsafe locking scenario:
[  528.078650][ T7867] 
[  528.078670][ T7867]        CPU0                    CPU1
[  528.078695][ T7867]        ----                    ----
[  528.078713][ T7867]   lock(pidmap_lock);
[  528.078730][ T7867]                                lock(&s->seqcount#2);
[  528.078751][ T7867]                                lock(pidmap_lock);
[  528.078770][ T7867]   lock(&s->seqcount#2);
[  528.078788][ T7867] 
[  528.078788][ T7867]  *** DEADLOCK ***
[  528.078788][ T7867] 
[  528.078800][ T7867] 2 locks held by runc:[1:CHILD]/7867:
[  528.078808][ T7867]  #0: c000001ffea6f4f0 (lock#2){+.+.}-{2:2}, at: __radix_tree_preload+0x8/0x370
__radix_tree_preload at lib/radix-tree.c:322
[  528.078844][ T7867]  #1: c0000000056b0198 (pidmap_lock){....}-{2:2}, at: alloc_pid+0x258/0x590
[  528.078870][ T7867] 
[  528.078870][ T7867] stack backtrace:
[  528.078890][ T7867] CPU: 46 PID: 7867 Comm: runc:[1:CHILD] Not tainted 5.9.0-rc5-next-20200914 #1
[  528.078921][ T7867] Call Trace:
[  528.078940][ T7867] [c000001ff07eefc0] [c00000000063f8c8] dump_stack+0xec/0x144 (unreliable)
[  528.078964][ T7867] [c000001ff07ef000] [c00000000013f44c] print_circular_bug.isra.43+0x2dc/0x350
[  528.078978][ T7867] [c000001ff07ef0a0] [c00000000013f640] check_noncircular+0x180/0x1b0
[  528.079000][ T7867] [c000001ff07ef170] [c000000000140b84] check_prevs_add+0x1c4/0x1120
[  528.079022][ T7867] [c000001ff07ef280] [c0000000001446ec] __lock_acquire+0x176c/0x1c00
[  528.079043][ T7867] [c000001ff07ef3a0] [c00000000014578c] lock_acquire+0x10c/0x560
[  528.079066][ T7867] [c000001ff07ef490] [c0000000003565f0] ___slab_alloc+0xa40/0xb40
[  528.079079][ T7867] [c000001ff07ef590] [c000000000356724] __slab_alloc+0x34/0xf0
[  528.079100][ T7867] [c000001ff07ef5e0] [c000000000356ab4] kmem_cache_alloc+0x2d4/0x470
[  528.079122][ T7867] [c000001ff07ef670] [c000000000397e14] create_object+0x74/0x430
[  528.079144][ T7867] [c000001ff07ef720] [c000000000351944] slab_post_alloc_hook+0xa4/0x670
[  528.079165][ T7867] [c000001ff07ef7e0] [c000000000356994] kmem_cache_alloc+0x1b4/0x470
[  528.079187][ T7867] [c000001ff07ef870] [c00000000064e004] radix_tree_node_alloc.constprop.19+0xe4/0x160
radix_tree_node_alloc at lib/radix-tree.c:252
[  528.079219][ T7867] [c000001ff07ef8e0] [c00000000064f2b8] idr_get_free+0x298/0x360
idr_get_free at lib/radix-tree.c:1507
[  528.079249][ T7867] [c000001ff07ef970] [c000000000645db4] idr_alloc_u32+0x84/0x130
idr_alloc_u32 at lib/idr.c:46 (discriminator 4)
[  528.079271][ T7867] [c000001ff07ef9e0] [c000000000645f8c] idr_alloc_cyclic+0x7c/0x150
idr_alloc_cyclic at lib/idr.c:126 (discriminator 1)
[  528.079301][ T7867] [c000001ff07efa40] [c0000000000e48ac] alloc_pid+0x27c/0x590
[  528.079342][ T7867] [c000001ff07efb20] [c0000000000acc60] copy_process+0xc90/0x1930
[  528.079404][ T7867] [c000001ff07efc40] [c0000000000adc00] kernel_clone+0x120/0xa10
[  528.079499][ T7867] [c000001ff07efd00] [c0000000000ae578] __do_sys_clone+0x88/0xd0
[  528.079579][ T7867] [c000001ff07efdc0] [c000000000029c48] system_call_exception+0xf8/0x1d0
[  528.079691][ T7867] [c000001ff07efe20] [c00000000000d0a8] system_call_common+0xe8/0x218

> 
> 8<--------------
> 
> Ahmed S. Darwish (5):
>   seqlock: seqcount_LOCKNAME_t: Standardize naming convention
>   seqlock: Use unique prefix for seqcount_t property accessors
>   seqlock: seqcount_t: Implement all read APIs as statement expressions
>   seqlock: seqcount_LOCKNAME_t: Introduce PREEMPT_RT support
>   seqlock: PREEMPT_RT: Do not starve seqlock_t writers
> 
>  include/linux/seqlock.h | 281 ++++++++++++++++++++++++----------------
>  1 file changed, 167 insertions(+), 114 deletions(-)
> 
> base-commit: f75aef392f869018f78cfedf3c320a6b3fcfda6b
> --
> 2.28.0

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[Boqun Feng]

On Mon, Sep 14, 2020 at 08:20:53PM -0400, Qian Cai wrote:
> On Fri, 2020-09-04 at 17:32 +0200, Ahmed S. Darwish wrote:
> > Hi,
> > 
> > Changelog-v2
> > ============
> > 
> >   - Standardize on seqcount_LOCKNAME_t as the canonical reference for
> >     sequence counters with associated locks, instead of v1
> >     seqcount_LOCKTYPE_t.
> > 
> >   - Use unique prefix "seqprop_*" for all seqcount_t/seqcount_LOCKNAME_t
> >     property accessors.
> > 
> >   - Touch-up the lock-unlock rationale for more clarity. Enforce writer
> >     non-preemitiblity using "__seq_enforce_writer_non_preemptibility()".
> > 
> > Cover letter (v1)
> > =================
> > 
> > https://lkml.kernel.org/r/20200828010710.5407-1-a.darwish@linutronix.de
> > 
> > Preemption must be disabled before entering a sequence counter write
> > side critical section.  Otherwise the read side section can preempt the
> > write side section and spin for the entire scheduler tick.  If that
> > reader belongs to a real-time scheduling class, it can spin forever and
> > the kernel will livelock.
> > 
> > Disabling preemption cannot be done for PREEMPT_RT though: it can lead
> > to higher latencies, and the write side sections will not be able to
> > acquire locks which become sleeping locks (e.g. spinlock_t).
> > 
> > To remain preemptible, while avoiding a possible livelock caused by the
> > reader preempting the writer, use a different technique: let the reader
> > detect if a seqcount_LOCKNAME_t writer is in progress. If that's the
> > case, acquire then release the associated LOCKNAME writer serialization
> > lock. This will allow any possibly-preempted writer to make progress
> > until the end of its writer serialization lock critical section.
> > 
> > Implement this lock-unlock technique for all seqcount_LOCKNAME_t with
> > an associated (PREEMPT_RT) sleeping lock, and for seqlock_t.
> 
> Reverting this patchset [1] from today's linux-next fixed a splat below. The
> splat looks like a false positive anyway because the existing locking dependency
> chains from the task #1 here:
> 
> &s->seqcount#2 ---> pidmap_lock
> 
> [  528.078061][ T7867] -> #1 (pidmap_lock){....}-{2:2}:
> [  528.078078][ T7867]        lock_acquire+0x10c/0x560
> [  528.078089][ T7867]        _raw_spin_lock_irqsave+0x64/0xb0
> [  528.078108][ T7867]        free_pid+0x5c/0x160
> free_pid at kernel/pid.c:131
> [  528.078127][ T7867]        release_task.part.40+0x59c/0x7f0
> __unhash_process at kernel/exit.c:76
> (inlined by) __exit_signal at kernel/exit.c:147
> (inlined by) release_task at kernel/exit.c:198
> [  528.078145][ T7867]        do_exit+0x77c/0xda0
> exit_notify at kernel/exit.c:679
> (inlined by) do_exit at kernel/exit.c:826
> [  528.078163][ T7867]        kthread+0x148/0x1d0
> [  528.078182][ T7867]        ret_from_kernel_thread+0x5c/0x80
> 
> It is write_seqlock(&sig->stats_lock) in __exit_signal(), but the &s->seqcount#2 
> in read_mems_allowed_begin() is read_seqcount_begin(&current->mems_allowed_seq), 
> so there should be no deadlock?
> 

I think this happened because seqcount_##lockname##_init() is defined at
function rather than macro, so when the seqcount_init() gets expand in
that function, the lock_class_key of seqcount will be a static variable
of seqcount_##lockname##_init() function, as a result, all
seqcount_##lockname##_t in the same compile unit (in this case it's
kernel/fork.c) share the same lock class key, and lockdep thought they
are the same lock ;-)

Regards,
Boqun

> [1] git revert --no-edit 0c9794c8b678..1909760f5fc3
> 
> [  528.077900][ T7867] WARNING: possible circular locking dependency detected
> [  528.077912][ T7867] 5.9.0-rc5-next-20200914 #1 Not tainted
> [  528.077921][ T7867] ------------------------------------------------------
> [  528.077931][ T7867] runc:[1:CHILD]/7867 is trying to acquire lock:
> [  528.077942][ T7867] c000001fce5570c8 (&s->seqcount#2){....}-{0:0}, at: __slab_alloc+0x34/0xf0
> [  528.077972][ T7867] 
> [  528.077972][ T7867] but task is already holding lock:
> [  528.077983][ T7867] c0000000056b0198 (pidmap_lock){....}-{2:2}, at: alloc_pid+0x258/0x590
> [  528.078009][ T7867] 
> [  528.078009][ T7867] which lock already depends on the new lock.
> [  528.078009][ T7867] 
> [  528.078031][ T7867] 
> [  528.078031][ T7867] the existing dependency chain (in reverse order) is:
> [  528.078061][ T7867] 
> [  528.078061][ T7867] -> #1 (pidmap_lock){....}-{2:2}:
> [  528.078078][ T7867]        lock_acquire+0x10c/0x560
> [  528.078089][ T7867]        _raw_spin_lock_irqsave+0x64/0xb0
> [  528.078108][ T7867]        free_pid+0x5c/0x160
> free_pid at kernel/pid.c:131
> [  528.078127][ T7867]        release_task.part.40+0x59c/0x7f0
> __unhash_process at kernel/exit.c:76
> (inlined by) __exit_signal at kernel/exit.c:147
> (inlined by) release_task at kernel/exit.c:198
> [  528.078145][ T7867]        do_exit+0x77c/0xda0
> exit_notify at kernel/exit.c:679
> (inlined by) do_exit at kernel/exit.c:826
> [  528.078163][ T7867]        kthread+0x148/0x1d0
> [  528.078182][ T7867]        ret_from_kernel_thread+0x5c/0x80
> [  528.078208][ T7867] 
> [  528.078208][ T7867] -> #0 (&s->seqcount#2){....}-{0:0}:
> [  528.078241][ T7867]        check_prevs_add+0x1c4/0x1120
> check_prev_add at kernel/locking/lockdep.c:2820
> (inlined by) check_prevs_add at kernel/locking/lockdep.c:2944
> [  528.078260][ T7867]        __lock_acquire+0x176c/0x1c00
> validate_chain at kernel/locking/lockdep.c:3562
> (inlined by) __lock_acquire at kernel/locking/lockdep.c:4796
> [  528.078278][ T7867]        lock_acquire+0x10c/0x560
> [  528.078297][ T7867]        ___slab_alloc+0xa40/0xb40
> seqcount_lockdep_reader_access at include/linux/seqlock.h:103
> (inlined by) read_mems_allowed_begin at include/linux/cpuset.h:135
> (inlined by) get_any_partial at mm/slub.c:2035
> (inlined by) get_partial at mm/slub.c:2078
> (inlined by) new_slab_objects at mm/slub.c:2577
> (inlined by) ___slab_alloc at mm/slub.c:2745
> [  528.078324][ T7867]        __slab_alloc+0x34/0xf0
> [  528.078342][ T7867]        kmem_cache_alloc+0x2d4/0x470
> [  528.078362][ T7867]        create_object+0x74/0x430
> [  528.078381][ T7867]        slab_post_alloc_hook+0xa4/0x670
> [  528.078399][ T7867]        kmem_cache_alloc+0x1b4/0x470
> [  528.078418][ T7867]        radix_tree_node_alloc.constprop.19+0xe4/0x160
> [  528.078438][ T7867]        idr_get_free+0x298/0x360
> [  528.078456][ T7867]        idr_alloc_u32+0x84/0x130
> [  528.078474][ T7867]        idr_alloc_cyclic+0x7c/0x150
> [  528.078493][ T7867]        alloc_pid+0x27c/0x590
> [  528.078511][ T7867]        copy_process+0xc90/0x1930
> copy_process at kernel/fork.c:2104
> [  528.078529][ T7867]        kernel_clone+0x120/0xa10
> [  528.078546][ T7867]        __do_sys_clone+0x88/0xd0
> [  528.078565][ T7867]        system_call_exception+0xf8/0x1d0
> [  528.078592][ T7867]        system_call_common+0xe8/0x218
> [  528.078609][ T7867] 
> [  528.078609][ T7867] other info that might help us debug this:
> [  528.078609][ T7867] 
> [  528.078650][ T7867]  Possible unsafe locking scenario:
> [  528.078650][ T7867] 
> [  528.078670][ T7867]        CPU0                    CPU1
> [  528.078695][ T7867]        ----                    ----
> [  528.078713][ T7867]   lock(pidmap_lock);
> [  528.078730][ T7867]                                lock(&s->seqcount#2);
> [  528.078751][ T7867]                                lock(pidmap_lock);
> [  528.078770][ T7867]   lock(&s->seqcount#2);
> [  528.078788][ T7867] 
> [  528.078788][ T7867]  *** DEADLOCK ***
> [  528.078788][ T7867] 
> [  528.078800][ T7867] 2 locks held by runc:[1:CHILD]/7867:
> [  528.078808][ T7867]  #0: c000001ffea6f4f0 (lock#2){+.+.}-{2:2}, at: __radix_tree_preload+0x8/0x370
> __radix_tree_preload at lib/radix-tree.c:322
> [  528.078844][ T7867]  #1: c0000000056b0198 (pidmap_lock){....}-{2:2}, at: alloc_pid+0x258/0x590
> [  528.078870][ T7867] 
> [  528.078870][ T7867] stack backtrace:
> [  528.078890][ T7867] CPU: 46 PID: 7867 Comm: runc:[1:CHILD] Not tainted 5.9.0-rc5-next-20200914 #1
> [  528.078921][ T7867] Call Trace:
> [  528.078940][ T7867] [c000001ff07eefc0] [c00000000063f8c8] dump_stack+0xec/0x144 (unreliable)
> [  528.078964][ T7867] [c000001ff07ef000] [c00000000013f44c] print_circular_bug.isra.43+0x2dc/0x350
> [  528.078978][ T7867] [c000001ff07ef0a0] [c00000000013f640] check_noncircular+0x180/0x1b0
> [  528.079000][ T7867] [c000001ff07ef170] [c000000000140b84] check_prevs_add+0x1c4/0x1120
> [  528.079022][ T7867] [c000001ff07ef280] [c0000000001446ec] __lock_acquire+0x176c/0x1c00
> [  528.079043][ T7867] [c000001ff07ef3a0] [c00000000014578c] lock_acquire+0x10c/0x560
> [  528.079066][ T7867] [c000001ff07ef490] [c0000000003565f0] ___slab_alloc+0xa40/0xb40
> [  528.079079][ T7867] [c000001ff07ef590] [c000000000356724] __slab_alloc+0x34/0xf0
> [  528.079100][ T7867] [c000001ff07ef5e0] [c000000000356ab4] kmem_cache_alloc+0x2d4/0x470
> [  528.079122][ T7867] [c000001ff07ef670] [c000000000397e14] create_object+0x74/0x430
> [  528.079144][ T7867] [c000001ff07ef720] [c000000000351944] slab_post_alloc_hook+0xa4/0x670
> [  528.079165][ T7867] [c000001ff07ef7e0] [c000000000356994] kmem_cache_alloc+0x1b4/0x470
> [  528.079187][ T7867] [c000001ff07ef870] [c00000000064e004] radix_tree_node_alloc.constprop.19+0xe4/0x160
> radix_tree_node_alloc at lib/radix-tree.c:252
> [  528.079219][ T7867] [c000001ff07ef8e0] [c00000000064f2b8] idr_get_free+0x298/0x360
> idr_get_free at lib/radix-tree.c:1507
> [  528.079249][ T7867] [c000001ff07ef970] [c000000000645db4] idr_alloc_u32+0x84/0x130
> idr_alloc_u32 at lib/idr.c:46 (discriminator 4)
> [  528.079271][ T7867] [c000001ff07ef9e0] [c000000000645f8c] idr_alloc_cyclic+0x7c/0x150
> idr_alloc_cyclic at lib/idr.c:126 (discriminator 1)
> [  528.079301][ T7867] [c000001ff07efa40] [c0000000000e48ac] alloc_pid+0x27c/0x590
> [  528.079342][ T7867] [c000001ff07efb20] [c0000000000acc60] copy_process+0xc90/0x1930
> [  528.079404][ T7867] [c000001ff07efc40] [c0000000000adc00] kernel_clone+0x120/0xa10
> [  528.079499][ T7867] [c000001ff07efd00] [c0000000000ae578] __do_sys_clone+0x88/0xd0
> [  528.079579][ T7867] [c000001ff07efdc0] [c000000000029c48] system_call_exception+0xf8/0x1d0
> [  528.079691][ T7867] [c000001ff07efe20] [c00000000000d0a8] system_call_common+0xe8/0x218
> 
> > 
> > 8<--------------
> > 
> > Ahmed S. Darwish (5):
> >   seqlock: seqcount_LOCKNAME_t: Standardize naming convention
> >   seqlock: Use unique prefix for seqcount_t property accessors
> >   seqlock: seqcount_t: Implement all read APIs as statement expressions
> >   seqlock: seqcount_LOCKNAME_t: Introduce PREEMPT_RT support
> >   seqlock: PREEMPT_RT: Do not starve seqlock_t writers
> > 
> >  include/linux/seqlock.h | 281 ++++++++++++++++++++++++----------------
> >  1 file changed, 167 insertions(+), 114 deletions(-)
> > 
> > base-commit: f75aef392f869018f78cfedf3c320a6b3fcfda6b
> > --
> > 2.28.0
> 

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[Boqun Feng]

On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:
> On Mon, Sep 14, 2020 at 08:20:53PM -0400, Qian Cai wrote:
> > On Fri, 2020-09-04 at 17:32 +0200, Ahmed S. Darwish wrote:
> > > Hi,
> > > 
> > > Changelog-v2
> > > ============
> > > 
> > >   - Standardize on seqcount_LOCKNAME_t as the canonical reference for
> > >     sequence counters with associated locks, instead of v1
> > >     seqcount_LOCKTYPE_t.
> > > 
> > >   - Use unique prefix "seqprop_*" for all seqcount_t/seqcount_LOCKNAME_t
> > >     property accessors.
> > > 
> > >   - Touch-up the lock-unlock rationale for more clarity. Enforce writer
> > >     non-preemitiblity using "__seq_enforce_writer_non_preemptibility()".
> > > 
> > > Cover letter (v1)
> > > =================
> > > 
> > > https://lkml.kernel.org/r/20200828010710.5407-1-a.darwish@linutronix.de
> > > 
> > > Preemption must be disabled before entering a sequence counter write
> > > side critical section.  Otherwise the read side section can preempt the
> > > write side section and spin for the entire scheduler tick.  If that
> > > reader belongs to a real-time scheduling class, it can spin forever and
> > > the kernel will livelock.
> > > 
> > > Disabling preemption cannot be done for PREEMPT_RT though: it can lead
> > > to higher latencies, and the write side sections will not be able to
> > > acquire locks which become sleeping locks (e.g. spinlock_t).
> > > 
> > > To remain preemptible, while avoiding a possible livelock caused by the
> > > reader preempting the writer, use a different technique: let the reader
> > > detect if a seqcount_LOCKNAME_t writer is in progress. If that's the
> > > case, acquire then release the associated LOCKNAME writer serialization
> > > lock. This will allow any possibly-preempted writer to make progress
> > > until the end of its writer serialization lock critical section.
> > > 
> > > Implement this lock-unlock technique for all seqcount_LOCKNAME_t with
> > > an associated (PREEMPT_RT) sleeping lock, and for seqlock_t.
> > 
> > Reverting this patchset [1] from today's linux-next fixed a splat below. The
> > splat looks like a false positive anyway because the existing locking dependency
> > chains from the task #1 here:
> > 
> > &s->seqcount#2 ---> pidmap_lock
> > 
> > [  528.078061][ T7867] -> #1 (pidmap_lock){....}-{2:2}:
> > [  528.078078][ T7867]        lock_acquire+0x10c/0x560
> > [  528.078089][ T7867]        _raw_spin_lock_irqsave+0x64/0xb0
> > [  528.078108][ T7867]        free_pid+0x5c/0x160
> > free_pid at kernel/pid.c:131
> > [  528.078127][ T7867]        release_task.part.40+0x59c/0x7f0
> > __unhash_process at kernel/exit.c:76
> > (inlined by) __exit_signal at kernel/exit.c:147
> > (inlined by) release_task at kernel/exit.c:198
> > [  528.078145][ T7867]        do_exit+0x77c/0xda0
> > exit_notify at kernel/exit.c:679
> > (inlined by) do_exit at kernel/exit.c:826
> > [  528.078163][ T7867]        kthread+0x148/0x1d0
> > [  528.078182][ T7867]        ret_from_kernel_thread+0x5c/0x80
> > 
> > It is write_seqlock(&sig->stats_lock) in __exit_signal(), but the &s->seqcount#2 
> > in read_mems_allowed_begin() is read_seqcount_begin(&current->mems_allowed_seq), 
> > so there should be no deadlock?
> > 
> 
> I think this happened because seqcount_##lockname##_init() is defined at
> function rather than macro, so when the seqcount_init() gets expand in
> that function, the lock_class_key of seqcount will be a static variable
> of seqcount_##lockname##_init() function, as a result, all
> seqcount_##lockname##_t in the same compile unit (in this case it's
> kernel/fork.c) share the same lock class key, and lockdep thought they
> are the same lock ;-)
> 

Don't know how to fix this properly, but below is an ugly attemption,
only build test, just food for thoughts.

Regards,
Boqun

--------------->8
diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index f73c7eb68f27..938a5053def3 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -84,14 +84,18 @@ static inline void __seqcount_init(seqcount_t *s, const char *name,
 # define SEQCOUNT_DEP_MAP_INIT(lockname)				\
 		.dep_map = { .name = #lockname }
 
+# define MSIOCU 8 /* MAX SEQCOUNT IN ON COMPILE UNIT */
 /**
  * seqcount_init() - runtime initializer for seqcount_t
  * @s: Pointer to the seqcount_t instance
  */
 # define seqcount_init(s)						\
 	do {								\
-		static struct lock_class_key __key;			\
-		__seqcount_init((s), #s, &__key);			\
+		static struct lock_class_key __key[MSIOCU];		\
+		static int idx = 0;					\
+									\
+		BUG_ON(idx >= MSIOCU);					\
+		__seqcount_init((s), #s, &__key[idx++]);		\
 	} while (0)
 
 static inline void seqcount_lockdep_reader_access(const seqcount_t *s)

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[peterz]

On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:
> I think this happened because seqcount_##lockname##_init() is defined at
> function rather than macro, so when the seqcount_init() gets expand in

Bah! I hate all this :/

I suspect the below, while more verbose than I'd like is the best
option.

---
 include/linux/seqlock.h | 22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)

diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index f73c7eb68f27..76e44e6c0100 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -173,6 +173,19 @@ static inline void seqcount_lockdep_reader_access(const seqcount_t *s)
  * @lock:	Pointer to the associated lock
  */
 
+#define seqcount_LOCKNAME_init(s, _lock, lockname)			\
+	do {								\
+		seqcount_##lockname##_t *____s = (s);			\
+		seqcount_init(&____s->seqcount);			\
+		__SEQ_LOCK(____s->lock = (_lock));			\
+	} while (0)
+
+#define seqcount_raw_spinlock_init(s, lock)	seqcount_LOCKNAME_init(s, lock, raw_spinlock)
+#define seqcount_spinlock_init(s, lock)		seqcount_LOCKNAME_init(s, lock, spinlock)
+#define seqcount_rwlock_init(s, lock)		seqcount_LOCKNAME_init(s, lock, rwlock);
+#define seqcount_mutex_init(s, lock)		seqcount_LOCKNAME_init(s, lock, mutex);
+#define seqcount_ww_mutex_init(s, lock)		seqcount_LOCKNAME_init(s, lock, ww_mutex);
+
 /*
  * SEQCOUNT_LOCKNAME()	- Instantiate seqcount_LOCKNAME_t and helpers
  * seqprop_LOCKNAME_*()	- Property accessors for seqcount_LOCKNAME_t
@@ -190,13 +203,6 @@ typedef struct seqcount_##lockname {					\
 	__SEQ_LOCK(locktype	*lock);					\
 } seqcount_##lockname##_t;						\
 									\
-static __always_inline void						\
-seqcount_##lockname##_init(seqcount_##lockname##_t *s, locktype *lock)	\
-{									\
-	seqcount_init(&s->seqcount);					\
-	__SEQ_LOCK(s->lock = lock);					\
-}									\
-									\
 static __always_inline seqcount_t *					\
 __seqprop_##lockname##_ptr(seqcount_##lockname##_t *s)			\
 {									\
@@ -284,8 +290,8 @@ SEQCOUNT_LOCKNAME(ww_mutex,     struct ww_mutex, true,     &s->lock->base, ww_mu
 	__SEQ_LOCK(.lock	= (assoc_lock))				\
 }
 
-#define SEQCNT_SPINLOCK_ZERO(name, lock)	SEQCOUNT_LOCKNAME_ZERO(name, lock)
 #define SEQCNT_RAW_SPINLOCK_ZERO(name, lock)	SEQCOUNT_LOCKNAME_ZERO(name, lock)
+#define SEQCNT_SPINLOCK_ZERO(name, lock)	SEQCOUNT_LOCKNAME_ZERO(name, lock)
 #define SEQCNT_RWLOCK_ZERO(name, lock)		SEQCOUNT_LOCKNAME_ZERO(name, lock)
 #define SEQCNT_MUTEX_ZERO(name, lock)		SEQCOUNT_LOCKNAME_ZERO(name, lock)
 #define SEQCNT_WW_MUTEX_ZERO(name, lock) 	SEQCOUNT_LOCKNAME_ZERO(name, lock)

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[Qian Cai]

On Tue, 2020-09-15 at 16:30 +0200, peterz@infradead.org wrote:
> On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:
> > I think this happened because seqcount_##lockname##_init() is defined at
> > function rather than macro, so when the seqcount_init() gets expand in
> 
> Bah! I hate all this :/
> 
> I suspect the below, while more verbose than I'd like is the best
> option.

Stephen, can you add this patch for now until Peter beats you to it?

> 
> ---
>  include/linux/seqlock.h | 22 ++++++++++++++--------
>  1 file changed, 14 insertions(+), 8 deletions(-)
> 
> diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
> index f73c7eb68f27..76e44e6c0100 100644
> --- a/include/linux/seqlock.h
> +++ b/include/linux/seqlock.h
> @@ -173,6 +173,19 @@ static inline void seqcount_lockdep_reader_access(const
> seqcount_t *s)
>   * @lock:	Pointer to the associated lock
>   */
>  
> +#define seqcount_LOCKNAME_init(s, _lock, lockname)			\
> +	do {								\
> +		seqcount_##lockname##_t *____s = (s);			\
> +		seqcount_init(&____s->seqcount);			\
> +		__SEQ_LOCK(____s->lock = (_lock));			\
> +	} while (0)
> +
> +#define seqcount_raw_spinlock_init(s, lock)	seqcount_LOCKNAME_init(s, lock,
> raw_spinlock)
> +#define seqcount_spinlock_init(s, lock)		seqcount_LOCKNAME_init(s
> , lock, spinlock)
> +#define seqcount_rwlock_init(s, lock)		seqcount_LOCKNAME_init(s
> , lock, rwlock);
> +#define seqcount_mutex_init(s, lock)		seqcount_LOCKNAME_init(s, lock,
> mutex);
> +#define seqcount_ww_mutex_init(s, lock)		seqcount_LOCKNAME_init(s
> , lock, ww_mutex);
> +
>  /*
>   * SEQCOUNT_LOCKNAME()	- Instantiate seqcount_LOCKNAME_t and helpers
>   * seqprop_LOCKNAME_*()	- Property accessors for seqcount_LOCKNAME_t
> @@ -190,13 +203,6 @@ typedef struct seqcount_##lockname {				
> 	\
>  	__SEQ_LOCK(locktype	*lock);					\
>  } seqcount_##lockname##_t;						\
>  									\
> -static __always_inline void						\
> -seqcount_##lockname##_init(seqcount_##lockname##_t *s, locktype *lock)	
> \
> -{									\
> -	seqcount_init(&s->seqcount);					\
> -	__SEQ_LOCK(s->lock = lock);					\
> -}									\
> -									\
>  static __always_inline seqcount_t *					\
>  __seqprop_##lockname##_ptr(seqcount_##lockname##_t *s)			
> \
>  {									\
> @@ -284,8 +290,8 @@ SEQCOUNT_LOCKNAME(ww_mutex,     struct ww_mutex,
> true,     &s->lock->base, ww_mu
>  	__SEQ_LOCK(.lock	= (assoc_lock))				\
>  }
>  
> -#define SEQCNT_SPINLOCK_ZERO(name, lock)	SEQCOUNT_LOCKNAME_ZERO(name,
> lock)
>  #define SEQCNT_RAW_SPINLOCK_ZERO(name, lock)	SEQCOUNT_LOCKNAME_ZERO(name,
> lock)
> +#define SEQCNT_SPINLOCK_ZERO(name, lock)	SEQCOUNT_LOCKNAME_ZERO(name,
> lock)
>  #define SEQCNT_RWLOCK_ZERO(name, lock)		SEQCOUNT_LOCKNAME_ZERO(n
> ame, lock)
>  #define SEQCNT_MUTEX_ZERO(name, lock)		SEQCOUNT_LOCKNAME_ZERO(n
> ame, lock)
>  #define SEQCNT_WW_MUTEX_ZERO(name, lock) 	SEQCOUNT_LOCKNAME_ZERO(name,
> lock)
> 

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[peterz]

On Wed, Sep 16, 2020 at 08:52:07AM -0400, Qian Cai wrote:
> On Tue, 2020-09-15 at 16:30 +0200, peterz@infradead.org wrote:
> > On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:
> > > I think this happened because seqcount_##lockname##_init() is defined at
> > > function rather than macro, so when the seqcount_init() gets expand in
> > 
> > Bah! I hate all this :/
> > 
> > I suspect the below, while more verbose than I'd like is the best
> > option.
> 
> Stephen, can you add this patch for now until Peter beats you to it?

Did you verify it works? I only wrote it..

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[Qian Cai]

----- Original Message -----
> On Wed, Sep 16, 2020 at 08:52:07AM -0400, Qian Cai wrote:
> > On Tue, 2020-09-15 at 16:30 +0200, peterz@infradead.org wrote:
> > > On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:
> > > > I think this happened because seqcount_##lockname##_init() is defined
> > > > at
> > > > function rather than macro, so when the seqcount_init() gets expand in
> > > 
> > > Bah! I hate all this :/
> > > 
> > > I suspect the below, while more verbose than I'd like is the best
> > > option.
> > 
> > Stephen, can you add this patch for now until Peter beats you to it?
> 
> Did you verify it works? I only wrote it..

Yes, I did.

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[peterz]

On Wed, Sep 16, 2020 at 09:00:59AM -0400, Qian Cai wrote:
> 
> 
> ----- Original Message -----
> > On Wed, Sep 16, 2020 at 08:52:07AM -0400, Qian Cai wrote:
> > > On Tue, 2020-09-15 at 16:30 +0200, peterz@infradead.org wrote:
> > > > On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:
> > > > > I think this happened because seqcount_##lockname##_init() is defined
> > > > > at
> > > > > function rather than macro, so when the seqcount_init() gets expand in
> > > > 
> > > > Bah! I hate all this :/
> > > > 
> > > > I suspect the below, while more verbose than I'd like is the best
> > > > option.
> > > 
> > > Stephen, can you add this patch for now until Peter beats you to it?
> > 
> > Did you verify it works? I only wrote it..
> 
> Yes, I did.

Excellent, I'll stick a Tested-by from you on then.

Re: [PATCH v2 0/5] seqlock: Introduce PREEMPT_RT support

[Stephen Rothwell]

[-- Attachment #1: Type: text/plain, Size: 1110 bytes --]

Hi all,

On Wed, 16 Sep 2020 15:02:33 +0200 peterz@infradead.org wrote:
>
> On Wed, Sep 16, 2020 at 09:00:59AM -0400, Qian Cai wrote:
> > 
> > 
> > ----- Original Message -----  
> > > On Wed, Sep 16, 2020 at 08:52:07AM -0400, Qian Cai wrote:  
> > > > On Tue, 2020-09-15 at 16:30 +0200, peterz@infradead.org wrote:  
> > > > > On Tue, Sep 15, 2020 at 08:48:17PM +0800, Boqun Feng wrote:  
> > > > > > I think this happened because seqcount_##lockname##_init() is defined
> > > > > > at
> > > > > > function rather than macro, so when the seqcount_init() gets expand in  
> > > > > 
> > > > > Bah! I hate all this :/
> > > > > 
> > > > > I suspect the below, while more verbose than I'd like is the best
> > > > > option.  
> > > > 
> > > > Stephen, can you add this patch for now until Peter beats you to it?  
> > > 
> > > Did you verify it works? I only wrote it..  
> > 
> > Yes, I did.  
> 
> Excellent, I'll stick a Tested-by from you on then.

I'll add this into the tip tree merge today (unless the tip tree is
updated in the mean time).

-- 
Cheers,
Stephen Rothwell

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 488 bytes --]