From: Ionut Nechita ionut.nechita@windriver.com
This series addresses two critical performance regressions in the block layer multiqueue (blk-mq) subsystem when running on PREEMPT_RT kernels.
On RT kernels, regular spinlocks are converted to sleeping rt_mutex locks, which can cause severe performance degradation in the I/O hot path. This series converts two problematic locking patterns to prevent IRQ threads from sleeping during I/O operations.
Testing on MegaRAID 12GSAS controller with 8 MSI-X vectors shows: - v6.6.52-rt (before regression): 640 MB/s sequential read - v6.6.64-rt (regression introduced): 153 MB/s (-76% regression) - v6.6.68-rt with queue_lock fix only: 640 MB/s (performance restored) - v6.6.69-rt with both fixes: expected similar or better performance
The first patch replaces queue_lock with memory barriers in the I/O completion hot path, eliminating the contention that caused IRQ threads to sleep. The second patch converts the global blk_mq_cpuhp_lock from mutex to raw_spinlock to prevent sleeping during CPU hotplug operations.
Both conversions are safe because the protected code paths only perform fast, non-blocking operations (memory barriers, list/hlist manipulation, flag checks).
Ionut Nechita (2): block/blk-mq: fix RT kernel regression with queue_lock in hot path block/blk-mq: convert blk_mq_cpuhp_lock to raw_spinlock for RT
block/blk-mq.c | 33 +++++++++++++++------------------ 1 file changed, 15 insertions(+), 18 deletions(-)
-- 2.52.0
From: Ionut Nechita ionut.nechita@windriver.com
Hi Ming,
Thank you for the feedback!
You're absolutely right - blk_mq_cpuhp_lock is only acquired in the slow path (setup/cleanup operations during queue initialization/teardown), not in the fast I/O path.
Looking at my testing results more carefully: - The queue_lock patch (PATCH 1/2) alone restores performance to 640 MB/s - The cpuhp_lock conversion (PATCH 2/2) doesn't contribute to fixing the I/O regression
The cpuhp_lock is used in: - blk_mq_remove_cpuhp() - queue cleanup - blk_mq_add_hw_queues_cpuhp() - queue setup - blk_mq_remove_hw_queues_cpuhp() - queue cleanup
These are indeed slow path operations with no contention in the I/O hot path.
I'll drop the second patch (cpuhp_lock conversion) and send v2 with only the queue_lock fix, which addresses the actual bottleneck: removing the sleeping lock from blk_mq_run_hw_queue() that was causing IRQ threads to serialize and enter D-state during I/O completion.
Best regards, Ionut
From: Ionut Nechita ionut.nechita@windriver.com
Commit 679b1874eba7 ("block: fix ordering between checking QUEUE_FLAG_QUIESCED request adding") introduced queue_lock acquisition in blk_mq_run_hw_queue() to synchronize QUEUE_FLAG_QUIESCED checks.
On RT kernels (CONFIG_PREEMPT_RT), regular spinlocks are converted to rt_mutex (sleeping locks). When multiple MSI-X IRQ threads process I/O completions concurrently, they contend on queue_lock in the hot path, causing all IRQ threads to enter D (uninterruptible sleep) state. This serializes interrupt processing completely.
Test case (MegaRAID 12GSAS with 8 MSI-X vectors on RT kernel): - Good (v6.6.52-rt): 640 MB/s sequential read - Bad (v6.6.64-rt): 153 MB/s sequential read (-76% regression) - 6-8 out of 8 MSI-X IRQ threads stuck in D-state waiting on queue_lock
The original commit message mentioned memory barriers as an alternative approach. Use full memory barriers (smp_mb) instead of queue_lock to provide the same ordering guarantees without sleeping in RT kernel.
Memory barriers ensure proper synchronization: - CPU0 either sees QUEUE_FLAG_QUIESCED cleared, OR - CPU1 sees dispatch list/sw queue bitmap updates
This maintains correctness while avoiding lock contention that causes RT kernel IRQ threads to sleep in the I/O completion path.
Fixes: 679b1874eba7 ("block: fix ordering between checking QUEUE_FLAG_QUIESCED request adding") Cc: stable@vger.kernel.org Signed-off-by: Ionut Nechita ionut.nechita@windriver.com --- block/blk-mq.c | 19 ++++++++----------- 1 file changed, 8 insertions(+), 11 deletions(-)
diff --git a/block/blk-mq.c b/block/blk-mq.c index 5da948b07058..5fb8da4958d0 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -2292,22 +2292,19 @@ void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async)
might_sleep_if(!async && hctx->flags & BLK_MQ_F_BLOCKING);
+ /* + * First lockless check to avoid unnecessary overhead. + * Memory barrier below synchronizes with blk_mq_unquiesce_queue(). + */ need_run = blk_mq_hw_queue_need_run(hctx); if (!need_run) { - unsigned long flags; - - /* - * Synchronize with blk_mq_unquiesce_queue(), because we check - * if hw queue is quiesced locklessly above, we need the use - * ->queue_lock to make sure we see the up-to-date status to - * not miss rerunning the hw queue. - */ - spin_lock_irqsave(&hctx->queue->queue_lock, flags); + /* Synchronize with blk_mq_unquiesce_queue() */ + smp_mb(); need_run = blk_mq_hw_queue_need_run(hctx); - spin_unlock_irqrestore(&hctx->queue->queue_lock, flags); - if (!need_run) return; + /* Ensure dispatch list/sw queue updates visible before execution */ + smp_mb(); }
if (async || !cpumask_test_cpu(raw_smp_processor_id(), hctx->cpumask)) {
From: Ionut Nechita ionut.nechita@windriver.com
Commit 58bf93580fec ("blk-mq: move cpuhp callback registering out of q->sysfs_lock") introduced a global mutex blk_mq_cpuhp_lock to avoid lockdep warnings between sysfs_lock and CPU hotplug lock.
On RT kernels (CONFIG_PREEMPT_RT), regular mutexes are converted to rt_mutex (sleeping locks). When block layer operations need to acquire blk_mq_cpuhp_lock, IRQ threads processing I/O completions may sleep, causing additional contention on top of the queue_lock issue from commit 679b1874eba7 ("block: fix ordering between checking QUEUE_FLAG_QUIESCED request adding").
Test case (MegaRAID 12GSAS with 8 MSI-X vectors on RT kernel): - v6.6.68-rt with queue_lock fix: 640 MB/s (queue_lock fixed) - v6.6.69-rt: still exhibits contention due to cpuhp_lock mutex
The functions protected by blk_mq_cpuhp_lock only perform fast, non-sleeping operations: - hlist_unhashed() checks - cpuhp_state_add_instance_nocalls() - just hlist manipulation - cpuhp_state_remove_instance_nocalls() - just hlist manipulation - INIT_HLIST_NODE() initialization
The _nocalls variants do not invoke state callbacks and only manipulate data structures, making them safe to call under raw_spinlock.
Convert blk_mq_cpuhp_lock from mutex to raw_spinlock to prevent it from becoming a sleeping lock in RT kernel. This eliminates the contention bottleneck while maintaining the lockdep fix's original intent.
Fixes: 58bf93580fec ("blk-mq: move cpuhp callback registering out of q->sysfs_lock") Cc: stable@vger.kernel.org Signed-off-by: Ionut Nechita ionut.nechita@windriver.com --- block/blk-mq.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/block/blk-mq.c b/block/blk-mq.c index 5fb8da4958d0..3982e24b1081 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -43,7 +43,7 @@
static DEFINE_PER_CPU(struct llist_head, blk_cpu_done); static DEFINE_PER_CPU(call_single_data_t, blk_cpu_csd); -static DEFINE_MUTEX(blk_mq_cpuhp_lock); +static DEFINE_RAW_SPINLOCK(blk_mq_cpuhp_lock);
static void blk_mq_insert_request(struct request *rq, blk_insert_t flags); static void blk_mq_request_bypass_insert(struct request *rq, @@ -3641,9 +3641,9 @@ static void __blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx)
static void blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx) { - mutex_lock(&blk_mq_cpuhp_lock); + raw_spin_lock(&blk_mq_cpuhp_lock); __blk_mq_remove_cpuhp(hctx); - mutex_unlock(&blk_mq_cpuhp_lock); + raw_spin_unlock(&blk_mq_cpuhp_lock); }
static void __blk_mq_add_cpuhp(struct blk_mq_hw_ctx *hctx) @@ -3683,9 +3683,9 @@ static void blk_mq_remove_hw_queues_cpuhp(struct request_queue *q) list_splice_init(&q->unused_hctx_list, &hctx_list); spin_unlock(&q->unused_hctx_lock);
- mutex_lock(&blk_mq_cpuhp_lock); + raw_spin_lock(&blk_mq_cpuhp_lock); __blk_mq_remove_cpuhp_list(&hctx_list); - mutex_unlock(&blk_mq_cpuhp_lock); + raw_spin_unlock(&blk_mq_cpuhp_lock);
spin_lock(&q->unused_hctx_lock); list_splice(&hctx_list, &q->unused_hctx_list); @@ -3702,10 +3702,10 @@ static void blk_mq_add_hw_queues_cpuhp(struct request_queue *q) struct blk_mq_hw_ctx *hctx; unsigned long i;
- mutex_lock(&blk_mq_cpuhp_lock); + raw_spin_lock(&blk_mq_cpuhp_lock); queue_for_each_hw_ctx(q, hctx, i) __blk_mq_add_cpuhp(hctx); - mutex_unlock(&blk_mq_cpuhp_lock); + raw_spin_unlock(&blk_mq_cpuhp_lock); }
/*
On Sat, Dec 20, 2025 at 01:02:41PM +0200, Ionut Nechita (WindRiver) wrote:
From: Ionut Nechita ionut.nechita@windriver.com
Commit 58bf93580fec ("blk-mq: move cpuhp callback registering out of q->sysfs_lock") introduced a global mutex blk_mq_cpuhp_lock to avoid lockdep warnings between sysfs_lock and CPU hotplug lock.
On RT kernels (CONFIG_PREEMPT_RT), regular mutexes are converted to rt_mutex (sleeping locks). When block layer operations need to acquire blk_mq_cpuhp_lock, IRQ threads processing I/O completions may sleep, causing additional contention on top of the queue_lock issue from commit 679b1874eba7 ("block: fix ordering between checking QUEUE_FLAG_QUIESCED request adding").
Test case (MegaRAID 12GSAS with 8 MSI-X vectors on RT kernel):
- v6.6.68-rt with queue_lock fix: 640 MB/s (queue_lock fixed)
- v6.6.69-rt: still exhibits contention due to cpuhp_lock mutex
The functions protected by blk_mq_cpuhp_lock only perform fast, non-sleeping operations:
- hlist_unhashed() checks
- cpuhp_state_add_instance_nocalls() - just hlist manipulation
- cpuhp_state_remove_instance_nocalls() - just hlist manipulation
- INIT_HLIST_NODE() initialization
The _nocalls variants do not invoke state callbacks and only manipulate data structures, making them safe to call under raw_spinlock.
Convert blk_mq_cpuhp_lock from mutex to raw_spinlock to prevent it from becoming a sleeping lock in RT kernel. This eliminates the contention bottleneck while maintaining the lockdep fix's original intent.
What is the contention bottleneck? blk_mq_cpuhp_lock is only acquired in slow code path, and it isn't required in fast io path.
Fixes: 58bf93580fec ("blk-mq: move cpuhp callback registering out of q->sysfs_lock")
With the 1st patch, the perf becomes 640MB/s, same with before regression.
So can you share what you try to fix with this patch?
Cc: stable@vger.kernel.org Signed-off-by: Ionut Nechita ionut.nechita@windriver.com
block/blk-mq.c | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/block/blk-mq.c b/block/blk-mq.c index 5fb8da4958d0..3982e24b1081 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -43,7 +43,7 @@ static DEFINE_PER_CPU(struct llist_head, blk_cpu_done); static DEFINE_PER_CPU(call_single_data_t, blk_cpu_csd); -static DEFINE_MUTEX(blk_mq_cpuhp_lock); +static DEFINE_RAW_SPINLOCK(blk_mq_cpuhp_lock); static void blk_mq_insert_request(struct request *rq, blk_insert_t flags); static void blk_mq_request_bypass_insert(struct request *rq, @@ -3641,9 +3641,9 @@ static void __blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx) static void blk_mq_remove_cpuhp(struct blk_mq_hw_ctx *hctx) {
- mutex_lock(&blk_mq_cpuhp_lock);
- raw_spin_lock(&blk_mq_cpuhp_lock); __blk_mq_remove_cpuhp(hctx);
- mutex_unlock(&blk_mq_cpuhp_lock);
- raw_spin_unlock(&blk_mq_cpuhp_lock);
__blk_mq_remove_cpuhp() ->cpuhp_state_remove_instance_nocalls() ->__cpuhp_state_remove_instance ->cpus_read_lock ->percpu_down_read ->percpu_down_read_internal ->might_sleep()
Thanks, Ming
syzbot ci has tested the following series
[v1] block/blk-mq: fix RT kernel performance regressions https://lore.kernel.org/all/20251220110241.8435-1-ionut.nechita@windriver.co... * [PATCH 1/2] block/blk-mq: fix RT kernel regression with queue_lock in hot path * [PATCH 2/2] block/blk-mq: convert blk_mq_cpuhp_lock to raw_spinlock for RT
and found the following issues: * BUG: sleeping function called from invalid context in __cpuhp_state_add_instance * BUG: sleeping function called from invalid context in __cpuhp_state_remove_instance
Full report is available here: https://ci.syzbot.org/series/632f4721-6256-44fd-83f5-bf439d5f33f9
***
BUG: sleeping function called from invalid context in __cpuhp_state_add_instance
tree: torvalds URL: https://kernel.googlesource.com/pub/scm/linux/kernel/git/torvalds/linux base: dd9b004b7ff3289fb7bae35130c0a5c0537266af arch: amd64 compiler: Debian clang version 20.1.8 (++20250708063551+0c9f909b7976-1~exp1~20250708183702.136), Debian LLD 20.1.8 config: https://ci.syzbot.org/builds/9ad1c682-13b1-4626-b61a-a2156384698d/config C repro: https://ci.syzbot.org/findings/f999a055-07f3-4d7a-acfd-8bc0be61e2ec/c_repro syz repro: https://ci.syzbot.org/findings/f999a055-07f3-4d7a-acfd-8bc0be61e2ec/syz_repr...
BUG: sleeping function called from invalid context at ./include/linux/percpu-rwsem.h:51 in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 5982, name: syz.0.17 preempt_count: 1, expected: 0 RCU nest depth: 0, expected: 0 INFO: lockdep is turned off. Preemption disabled at: [<0000000000000000>] 0x0 CPU: 1 UID: 0 PID: 5982 Comm: syz.0.17 Tainted: G W syzkaller #0 PREEMPT(full) Tainted: [W]=WARN Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.2-debian-1.16.2-1 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x189/0x250 lib/dump_stack.c:120 __might_resched+0x495/0x610 kernel/sched/core.c:8827 percpu_down_read_internal include/linux/percpu-rwsem.h:51 [inline] percpu_down_read include/linux/percpu-rwsem.h:77 [inline] cpus_read_lock+0x1b/0x160 kernel/cpu.c:491 __cpuhp_state_add_instance+0x19/0x40 kernel/cpu.c:2454 cpuhp_state_add_instance_nocalls include/linux/cpuhotplug.h:401 [inline] __blk_mq_add_cpuhp block/blk-mq.c:3858 [inline] blk_mq_add_hw_queues_cpuhp+0x19a/0x250 block/blk-mq.c:3906 blk_mq_realloc_hw_ctxs block/blk-mq.c:4611 [inline] blk_mq_init_allocated_queue+0x366/0x1350 block/blk-mq.c:4635 blk_mq_alloc_queue block/blk-mq.c:4416 [inline] __blk_mq_alloc_disk+0x1f0/0x340 block/blk-mq.c:4459 loop_add+0x411/0xad0 drivers/block/loop.c:2050 loop_control_ioctl+0x128/0x5a0 drivers/block/loop.c:2216 vfs_ioctl fs/ioctl.c:51 [inline] __do_sys_ioctl fs/ioctl.c:597 [inline] __se_sys_ioctl+0xfc/0x170 fs/ioctl.c:583 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline] do_syscall_64+0xfa/0xf80 arch/x86/entry/syscall_64.c:94 entry_SYSCALL_64_after_hwframe+0x77/0x7f RIP: 0033:0x7f1dc598f7c9 Code: ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 a8 ff ff ff f7 d8 64 89 01 48 RSP: 002b:00007ffff2134d08 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 00007f1dc5be5fa0 RCX: 00007f1dc598f7c9 RDX: 00000000004080f9 RSI: 0000000000004c80 RDI: 0000000000000003 RBP: 00007f1dc59f297f R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00007f1dc5be5fa0 R14: 00007f1dc5be5fa0 R15: 0000000000000003 </TASK>
***
BUG: sleeping function called from invalid context in __cpuhp_state_remove_instance
tree: torvalds URL: https://kernel.googlesource.com/pub/scm/linux/kernel/git/torvalds/linux base: dd9b004b7ff3289fb7bae35130c0a5c0537266af arch: amd64 compiler: Debian clang version 20.1.8 (++20250708063551+0c9f909b7976-1~exp1~20250708183702.136), Debian LLD 20.1.8 config: https://ci.syzbot.org/builds/9ad1c682-13b1-4626-b61a-a2156384698d/config C repro: https://ci.syzbot.org/findings/f39691bc-570a-4163-9791-31ce10e18fb6/c_repro syz repro: https://ci.syzbot.org/findings/f39691bc-570a-4163-9791-31ce10e18fb6/syz_repr...
BUG: sleeping function called from invalid context at ./include/linux/percpu-rwsem.h:51 in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 5975, name: syz.0.17 preempt_count: 1, expected: 0 RCU nest depth: 0, expected: 0 INFO: lockdep is turned off. Preemption disabled at: [<0000000000000000>] 0x0 CPU: 0 UID: 0 PID: 5975 Comm: syz.0.17 Tainted: G W syzkaller #0 PREEMPT(full) Tainted: [W]=WARN Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.16.2-debian-1.16.2-1 04/01/2014 Call Trace: <TASK> dump_stack_lvl+0x189/0x250 lib/dump_stack.c:120 __might_resched+0x495/0x610 kernel/sched/core.c:8827 percpu_down_read_internal include/linux/percpu-rwsem.h:51 [inline] percpu_down_read include/linux/percpu-rwsem.h:77 [inline] cpus_read_lock+0x1b/0x160 kernel/cpu.c:491 __cpuhp_state_remove_instance+0x77/0x2e0 kernel/cpu.c:2565 cpuhp_state_remove_instance_nocalls include/linux/cpuhotplug.h:502 [inline] __blk_mq_remove_cpuhp+0x140/0x1a0 block/blk-mq.c:3835 blk_mq_remove_cpuhp block/blk-mq.c:3844 [inline] blk_mq_exit_hw_queues block/blk-mq.c:3974 [inline] blk_mq_exit_queue+0xe8/0x380 block/blk-mq.c:4670 __del_gendisk+0x832/0x9e0 block/genhd.c:774 del_gendisk+0xe8/0x160 block/genhd.c:823 loop_remove+0x42/0xc0 drivers/block/loop.c:2121 loop_control_remove drivers/block/loop.c:2180 [inline] loop_control_ioctl+0x4ac/0x5a0 drivers/block/loop.c:2218 vfs_ioctl fs/ioctl.c:51 [inline] __do_sys_ioctl fs/ioctl.c:597 [inline] __se_sys_ioctl+0xfc/0x170 fs/ioctl.c:583 do_syscall_x64 arch/x86/entry/syscall_64.c:63 [inline] do_syscall_64+0xfa/0xf80 arch/x86/entry/syscall_64.c:94 entry_SYSCALL_64_after_hwframe+0x77/0x7f RIP: 0033:0x7f911d78f7c9 Code: ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 0f 1f 40 00 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 a8 ff ff ff f7 d8 64 89 01 48 RSP: 002b:00007ffd4f0e1fb8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 RAX: ffffffffffffffda RBX: 00007f911d9e5fa0 RCX: 00007f911d78f7c9 RDX: 0000000000000006 RSI: 0000000000004c81 RDI: 0000000000000003 RBP: 00007f911d7f297f R08: 0000000000000000 R09: 0000000000000000 R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000 R13: 00007f911d9e5fa0 R14: 00007f911d9e5fa0 R15: 0000000000000003 </TASK>
***
If these findings have caused you to resend the series or submit a separate fix, please add the following tag to your commit message: Tested-by: syzbot@syzkaller.appspotmail.com
--- This report is generated by a bot. It may contain errors. syzbot ci engineers can be reached at syzkaller@googlegroups.com.
linux-stable-mirror@lists.linaro.org
-
Ionut Nechita (WindRiver) -
Ming Lei -
syzbot ci