2022-06-15 14:58:21

by Juri Lelli

[permalink] [raw]
Subject: [RT] WARNING at hrtimer_start_range_ns

Hi,

QE reported splats with signatures similar to the following (edited)
when running blktests/nvmeof-mp [1] cases.

--->8---
[ 329.284321] WARNING: CPU: 0 PID: 4830 at kernel/time/hrtimer.c:1294 hrtimer_start_range_ns+0xa0/0xb0
[ 329.286298] Modules linked in: nvme_rdma(E) nvme_fabrics(E) nvme(E) nvmet_rdma(E) ...
[ 329.286354] Unloaded tainted modules: intel_uncore_frequency(E):2 isst_if_mbox_msr(E):2 ...
[ 329.296991] CPU: 0 PID: 4830 Comm: multipathd Kdump: loaded Tainted: G E -------- --- 5.19.0-0.rc1.9886142c7a22.16.test.eln120.x86_64+rt #2
[ 329.298861] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
[ 329.299798] RIP: 0010:hrtimer_start_range_ns+0xa0/0xb0
[ 329.300533] Code: 41 5d 41 5e 41 5f e9 df b4 94 00 be 01 00 00 00 48 89 df e8 82 f0 ff ff eb d8 49 8b 7d 00 48 89 c6 e8 c4 b4 94 00 f3 90 eb 91 <0f> 0b eb 8d 66 66 2e 0f 1f 84 00 00 00 00 00 90 0f 1f 44 00 00 80
[ 329.303087] RSP: 0018:ffffb4c180a1bd18 EFLAGS: 00010202
[ 329.303817] RAX: 0000000000000001 RBX: ffffb4c180a1bd80 RCX: 0000000000000001
[ 329.304828] RDX: 000000000000c300 RSI: 00000000000003e8 RDI: ffffb4c180a1bd80
[ 329.305759] RBP: 00000000000003e8 R08: 00000000000003e8 R09: 0000000000000000
[ 329.306637] R10: 0000000000000000 R11: 0000000000000001 R12: 000000000000c350
[ 329.307510] R13: 00007efd462b3540 R14: 0000000000000001 R15: 0000000000000000
[ 329.308384] FS: 00007efd462b5600(0000) GS:ffff8d30bbc00000(0000) knlGS:0000000000000000
[ 329.309368] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 329.310071] CR2: 00007efd4628f000 CR3: 000000011ba62004 CR4: 00000000007706f0
[ 329.310939] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 329.311811] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 329.312680] PKRU: 55555554
[ 329.313022] Call Trace:
[ 329.313352] <TASK>
[ 329.313632] read_events+0xd5/0x180
[ 329.314086] ? fget+0x79/0xa0
[ 329.314519] ? __hrtimer_init+0xe0/0xe0
[ 329.315056] do_io_getevents+0x71/0xd0
[ 329.315580] __x64_sys_io_getevents+0x58/0xc0
[ 329.316130] do_syscall_64+0x5b/0x80
[ 329.316590] ? syscall_exit_to_user_mode+0x12/0x30
[ 329.317192] ? do_syscall_64+0x67/0x80
[ 329.317669] ? do_syscall_64+0x67/0x80
[ 329.318139] ? do_syscall_64+0x67/0x80
[ 329.318666] ? syscall_exit_to_user_mode+0x12/0x30
[ 329.319337] ? do_syscall_64+0x67/0x80
[ 329.319851] ? do_syscall_64+0x67/0x80
[ 329.320396] ? syscall_exit_to_user_mode+0x12/0x30
[ 329.321062] ? do_syscall_64+0x67/0x80
[ 329.321540] ? __irq_exit_rcu+0x72/0x160
[ 329.322042] entry_SYSCALL_64_after_hwframe+0x46/0xb0
[ 329.322671] RIP: 0033:0x7efd477feecd
[ 329.323128] Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 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 8b 0d 23 7f 1b 00 f7 d8 64 89 01 48
[ 329.325462] RSP: 002b:00007efd462b34b8 EFLAGS: 00000246 ORIG_RAX: 00000000000000d0
[ 329.326388] RAX: ffffffffffffffda RBX: 00007efd462b5300 RCX: 00007efd477feecd
[ 329.326962] nvme nvme0: Removing ctrl: NQN "nvme-test"
[ 329.327885] RDX: 0000000000000080 RSI: 0000000000000001 RDI: 00007efd4628f000
[ 329.328760] RBP: 00007efd4628f000 R08: 00007efd462b45a0 R09: 0000000000000000
[ 329.329631] R10: 00007efd462b3540 R11: 0000000000000246 R12: 0000000000000001
[ 329.330506] R13: 0000000000000000 R14: 0000000000000080 R15: 00007efd462b3540
[ 329.331390] </TASK>
--->8---

My understanding is that we have

fs/aio.c::read_events
wait_event_interruptible_hrtimeout
__wait_event_hrtimeout
hrtimer_init_sleeper_on_stack <- this might mode |=HRTIMER_MODE_HARD on RT
hrtimer_start_range_ns
WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard)
fires since the latter doesn't see the change of mode done by
init_sleeper

The attached patch seems to cure the problem, although I'm not entirely
sure it's the cleanest fix (if I'm right at all about it :).

Thanks,
Juri

1 - https://github.com/yizhanglinux/blktests/tree/nvme-rdma-tcp/tests/nvmeof-mp

--->8---
From c7bb856eef8293c3a027f63ec3dc4fec93807227 Mon Sep 17 00:00:00 2001
From: Juri Lelli <[email protected]>
Date: Tue, 14 Jun 2022 10:58:51 +0200
Subject: [PATCH] time/hrtimer: Embed hrtimer mode into hrtimer_sleeper

Changes to hrtimer mode (potentially made by __hrtimer_init_sleeper() on
PREEMPT_RT) are not visible to do_nanosleep(), thus not accounted for
by hrtimer_start_expires() call paths.

Embed hrtimer mode into hrtimer_sleeper struct, so that the same mode is
used by code following hrtimer_sleeper initialization.

Signed-off-by: Juri Lelli <[email protected]>
---
block/blk-mq.c | 2 +-
include/linux/hrtimer.h | 1 +
include/linux/wait.h | 2 +-
kernel/futex/pi.c | 2 +-
kernel/time/hrtimer.c | 7 ++++---
net/core/pktgen.c | 2 +-
6 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 03235ed3a792..429d5f08a7c4 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -4672,7 +4672,7 @@ static bool blk_mq_poll_hybrid(struct request_queue *q, blk_qc_t qc)
if (blk_mq_rq_state(rq) == MQ_RQ_COMPLETE)
break;
set_current_state(TASK_UNINTERRUPTIBLE);
- hrtimer_sleeper_start_expires(&hs, mode);
+ hrtimer_sleeper_start_expires(&hs, hs.mode);
if (hs.task)
io_schedule();
hrtimer_cancel(&hs.timer);
diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index d22418264bec..90f79b4e20c8 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -140,6 +140,7 @@ struct hrtimer {
struct hrtimer_sleeper {
struct hrtimer timer;
struct task_struct *task;
+ enum hrtimer_mode mode;
};

#ifdef CONFIG_64BIT
diff --git a/include/linux/wait.h b/include/linux/wait.h
index 2d0df57c9902..b7499818f2c7 100644
--- a/include/linux/wait.h
+++ b/include/linux/wait.h
@@ -521,7 +521,7 @@ do { \
if ((timeout) != KTIME_MAX) \
hrtimer_start_range_ns(&__t.timer, timeout, \
current->timer_slack_ns, \
- HRTIMER_MODE_REL); \
+ __t.mode); \
\
__ret = ___wait_event(wq_head, condition, state, 0, 0, \
if (!__t.task) { \
diff --git a/kernel/futex/pi.c b/kernel/futex/pi.c
index 183b28c32c83..c576e84d96c9 100644
--- a/kernel/futex/pi.c
+++ b/kernel/futex/pi.c
@@ -1034,7 +1034,7 @@ int futex_lock_pi(u32 __user *uaddr, unsigned int flags, ktime_t *time, int tryl
}

if (unlikely(to))
- hrtimer_sleeper_start_expires(to, HRTIMER_MODE_ABS);
+ hrtimer_sleeper_start_expires(to, to->mode);

ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter);

diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 0ea8702eb516..0019c802e2e8 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -1996,6 +1996,7 @@ static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
__hrtimer_init(&sl->timer, clock_id, mode);
sl->timer.function = hrtimer_wakeup;
sl->task = current;
+ sl->mode = mode;
}

/**
@@ -2075,7 +2076,7 @@ static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
HRTIMER_MODE_ABS);
hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
- ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
+ ret = do_nanosleep(&t, t.mode);
destroy_hrtimer_on_stack(&t.timer);
return ret;
}
@@ -2094,7 +2095,7 @@ long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,

hrtimer_init_sleeper_on_stack(&t, clockid, mode);
hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
- ret = do_nanosleep(&t, mode);
+ ret = do_nanosleep(&t, t.mode);
if (ret != -ERESTART_RESTARTBLOCK)
goto out;

@@ -2299,7 +2300,7 @@ schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,

hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
- hrtimer_sleeper_start_expires(&t, mode);
+ hrtimer_sleeper_start_expires(&t, t.mode);

if (likely(t.task))
schedule();
diff --git a/net/core/pktgen.c b/net/core/pktgen.c
index 82c016e8c196..51eddba28c03 100644
--- a/net/core/pktgen.c
+++ b/net/core/pktgen.c
@@ -2169,7 +2169,7 @@ static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
} else {
do {
set_current_state(TASK_INTERRUPTIBLE);
- hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS);
+ hrtimer_sleeper_start_expires(&t, t.mode);

if (likely(t.task))
schedule();
--
2.36.1


2022-06-22 08:27:01

by Juri Lelli

[permalink] [raw]
Subject: Re: [RT] WARNING at hrtimer_start_range_ns

Hi,

On 15/06/22 16:53, Juri Lelli wrote:
> Hi,
>
> QE reported splats with signatures similar to the following (edited)
> when running blktests/nvmeof-mp [1] cases.
>
> --->8---
> [ 329.284321] WARNING: CPU: 0 PID: 4830 at kernel/time/hrtimer.c:1294 hrtimer_start_range_ns+0xa0/0xb0
> [ 329.286298] Modules linked in: nvme_rdma(E) nvme_fabrics(E) nvme(E) nvmet_rdma(E) ...
> [ 329.286354] Unloaded tainted modules: intel_uncore_frequency(E):2 isst_if_mbox_msr(E):2 ...
> [ 329.296991] CPU: 0 PID: 4830 Comm: multipathd Kdump: loaded Tainted: G E -------- --- 5.19.0-0.rc1.9886142c7a22.16.test.eln120.x86_64+rt #2
> [ 329.298861] Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
> [ 329.299798] RIP: 0010:hrtimer_start_range_ns+0xa0/0xb0
> [ 329.300533] Code: 41 5d 41 5e 41 5f e9 df b4 94 00 be 01 00 00 00 48 89 df e8 82 f0 ff ff eb d8 49 8b 7d 00 48 89 c6 e8 c4 b4 94 00 f3 90 eb 91 <0f> 0b eb 8d 66 66 2e 0f 1f 84 00 00 00 00 00 90 0f 1f 44 00 00 80
> [ 329.303087] RSP: 0018:ffffb4c180a1bd18 EFLAGS: 00010202
> [ 329.303817] RAX: 0000000000000001 RBX: ffffb4c180a1bd80 RCX: 0000000000000001
> [ 329.304828] RDX: 000000000000c300 RSI: 00000000000003e8 RDI: ffffb4c180a1bd80
> [ 329.305759] RBP: 00000000000003e8 R08: 00000000000003e8 R09: 0000000000000000
> [ 329.306637] R10: 0000000000000000 R11: 0000000000000001 R12: 000000000000c350
> [ 329.307510] R13: 00007efd462b3540 R14: 0000000000000001 R15: 0000000000000000
> [ 329.308384] FS: 00007efd462b5600(0000) GS:ffff8d30bbc00000(0000) knlGS:0000000000000000
> [ 329.309368] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> [ 329.310071] CR2: 00007efd4628f000 CR3: 000000011ba62004 CR4: 00000000007706f0
> [ 329.310939] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
> [ 329.311811] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
> [ 329.312680] PKRU: 55555554
> [ 329.313022] Call Trace:
> [ 329.313352] <TASK>
> [ 329.313632] read_events+0xd5/0x180
> [ 329.314086] ? fget+0x79/0xa0
> [ 329.314519] ? __hrtimer_init+0xe0/0xe0
> [ 329.315056] do_io_getevents+0x71/0xd0
> [ 329.315580] __x64_sys_io_getevents+0x58/0xc0
> [ 329.316130] do_syscall_64+0x5b/0x80
> [ 329.316590] ? syscall_exit_to_user_mode+0x12/0x30
> [ 329.317192] ? do_syscall_64+0x67/0x80
> [ 329.317669] ? do_syscall_64+0x67/0x80
> [ 329.318139] ? do_syscall_64+0x67/0x80
> [ 329.318666] ? syscall_exit_to_user_mode+0x12/0x30
> [ 329.319337] ? do_syscall_64+0x67/0x80
> [ 329.319851] ? do_syscall_64+0x67/0x80
> [ 329.320396] ? syscall_exit_to_user_mode+0x12/0x30
> [ 329.321062] ? do_syscall_64+0x67/0x80
> [ 329.321540] ? __irq_exit_rcu+0x72/0x160
> [ 329.322042] entry_SYSCALL_64_after_hwframe+0x46/0xb0
> [ 329.322671] RIP: 0033:0x7efd477feecd
> [ 329.323128] Code: ff c3 66 2e 0f 1f 84 00 00 00 00 00 90 f3 0f 1e fa 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 8b 0d 23 7f 1b 00 f7 d8 64 89 01 48
> [ 329.325462] RSP: 002b:00007efd462b34b8 EFLAGS: 00000246 ORIG_RAX: 00000000000000d0
> [ 329.326388] RAX: ffffffffffffffda RBX: 00007efd462b5300 RCX: 00007efd477feecd
> [ 329.326962] nvme nvme0: Removing ctrl: NQN "nvme-test"
> [ 329.327885] RDX: 0000000000000080 RSI: 0000000000000001 RDI: 00007efd4628f000
> [ 329.328760] RBP: 00007efd4628f000 R08: 00007efd462b45a0 R09: 0000000000000000
> [ 329.329631] R10: 00007efd462b3540 R11: 0000000000000246 R12: 0000000000000001
> [ 329.330506] R13: 0000000000000000 R14: 0000000000000080 R15: 00007efd462b3540
> [ 329.331390] </TASK>
> --->8---
>
> My understanding is that we have
>
> fs/aio.c::read_events
> wait_event_interruptible_hrtimeout
> __wait_event_hrtimeout
> hrtimer_init_sleeper_on_stack <- this might mode |=HRTIMER_MODE_HARD on RT
> hrtimer_start_range_ns
> WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard)
> fires since the latter doesn't see the change of mode done by
> init_sleeper
>
> The attached patch seems to cure the problem, although I'm not entirely
> sure it's the cleanest fix (if I'm right at all about it :).

Gentle ping about this one.

Thanks!
Juri

> 1 - https://github.com/yizhanglinux/blktests/tree/nvme-rdma-tcp/tests/nvmeof-mp
>
> --->8---
> From c7bb856eef8293c3a027f63ec3dc4fec93807227 Mon Sep 17 00:00:00 2001
> From: Juri Lelli <[email protected]>
> Date: Tue, 14 Jun 2022 10:58:51 +0200
> Subject: [PATCH] time/hrtimer: Embed hrtimer mode into hrtimer_sleeper
>
> Changes to hrtimer mode (potentially made by __hrtimer_init_sleeper() on
> PREEMPT_RT) are not visible to do_nanosleep(), thus not accounted for
> by hrtimer_start_expires() call paths.
>
> Embed hrtimer mode into hrtimer_sleeper struct, so that the same mode is
> used by code following hrtimer_sleeper initialization.
>
> Signed-off-by: Juri Lelli <[email protected]>
> ---
> block/blk-mq.c | 2 +-
> include/linux/hrtimer.h | 1 +
> include/linux/wait.h | 2 +-
> kernel/futex/pi.c | 2 +-
> kernel/time/hrtimer.c | 7 ++++---
> net/core/pktgen.c | 2 +-
> 6 files changed, 9 insertions(+), 7 deletions(-)
>
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 03235ed3a792..429d5f08a7c4 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -4672,7 +4672,7 @@ static bool blk_mq_poll_hybrid(struct request_queue *q, blk_qc_t qc)
> if (blk_mq_rq_state(rq) == MQ_RQ_COMPLETE)
> break;
> set_current_state(TASK_UNINTERRUPTIBLE);
> - hrtimer_sleeper_start_expires(&hs, mode);
> + hrtimer_sleeper_start_expires(&hs, hs.mode);
> if (hs.task)
> io_schedule();
> hrtimer_cancel(&hs.timer);
> diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
> index d22418264bec..90f79b4e20c8 100644
> --- a/include/linux/hrtimer.h
> +++ b/include/linux/hrtimer.h
> @@ -140,6 +140,7 @@ struct hrtimer {
> struct hrtimer_sleeper {
> struct hrtimer timer;
> struct task_struct *task;
> + enum hrtimer_mode mode;
> };
>
> #ifdef CONFIG_64BIT
> diff --git a/include/linux/wait.h b/include/linux/wait.h
> index 2d0df57c9902..b7499818f2c7 100644
> --- a/include/linux/wait.h
> +++ b/include/linux/wait.h
> @@ -521,7 +521,7 @@ do { \
> if ((timeout) != KTIME_MAX) \
> hrtimer_start_range_ns(&__t.timer, timeout, \
> current->timer_slack_ns, \
> - HRTIMER_MODE_REL); \
> + __t.mode); \
> \
> __ret = ___wait_event(wq_head, condition, state, 0, 0, \
> if (!__t.task) { \
> diff --git a/kernel/futex/pi.c b/kernel/futex/pi.c
> index 183b28c32c83..c576e84d96c9 100644
> --- a/kernel/futex/pi.c
> +++ b/kernel/futex/pi.c
> @@ -1034,7 +1034,7 @@ int futex_lock_pi(u32 __user *uaddr, unsigned int flags, ktime_t *time, int tryl
> }
>
> if (unlikely(to))
> - hrtimer_sleeper_start_expires(to, HRTIMER_MODE_ABS);
> + hrtimer_sleeper_start_expires(to, to->mode);
>
> ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter);
>
> diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
> index 0ea8702eb516..0019c802e2e8 100644
> --- a/kernel/time/hrtimer.c
> +++ b/kernel/time/hrtimer.c
> @@ -1996,6 +1996,7 @@ static void __hrtimer_init_sleeper(struct hrtimer_sleeper *sl,
> __hrtimer_init(&sl->timer, clock_id, mode);
> sl->timer.function = hrtimer_wakeup;
> sl->task = current;
> + sl->mode = mode;
> }
>
> /**
> @@ -2075,7 +2076,7 @@ static long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
> hrtimer_init_sleeper_on_stack(&t, restart->nanosleep.clockid,
> HRTIMER_MODE_ABS);
> hrtimer_set_expires_tv64(&t.timer, restart->nanosleep.expires);
> - ret = do_nanosleep(&t, HRTIMER_MODE_ABS);
> + ret = do_nanosleep(&t, t.mode);
> destroy_hrtimer_on_stack(&t.timer);
> return ret;
> }
> @@ -2094,7 +2095,7 @@ long hrtimer_nanosleep(ktime_t rqtp, const enum hrtimer_mode mode,
>
> hrtimer_init_sleeper_on_stack(&t, clockid, mode);
> hrtimer_set_expires_range_ns(&t.timer, rqtp, slack);
> - ret = do_nanosleep(&t, mode);
> + ret = do_nanosleep(&t, t.mode);
> if (ret != -ERESTART_RESTARTBLOCK)
> goto out;
>
> @@ -2299,7 +2300,7 @@ schedule_hrtimeout_range_clock(ktime_t *expires, u64 delta,
>
> hrtimer_init_sleeper_on_stack(&t, clock_id, mode);
> hrtimer_set_expires_range_ns(&t.timer, *expires, delta);
> - hrtimer_sleeper_start_expires(&t, mode);
> + hrtimer_sleeper_start_expires(&t, t.mode);
>
> if (likely(t.task))
> schedule();
> diff --git a/net/core/pktgen.c b/net/core/pktgen.c
> index 82c016e8c196..51eddba28c03 100644
> --- a/net/core/pktgen.c
> +++ b/net/core/pktgen.c
> @@ -2169,7 +2169,7 @@ static void spin(struct pktgen_dev *pkt_dev, ktime_t spin_until)
> } else {
> do {
> set_current_state(TASK_INTERRUPTIBLE);
> - hrtimer_sleeper_start_expires(&t, HRTIMER_MODE_ABS);
> + hrtimer_sleeper_start_expires(&t, t.mode);
>
> if (likely(t.task))
> schedule();
> --
> 2.36.1

Subject: Re: [RT] WARNING at hrtimer_start_range_ns

On 2022-06-15 16:53:55 [+0200], Juri Lelli wrote:
> Hi,
Hi,

> My understanding is that we have
>
> fs/aio.c::read_events
> wait_event_interruptible_hrtimeout
> __wait_event_hrtimeout
> hrtimer_init_sleeper_on_stack <- this might mode |=HRTIMER_MODE_HARD on RT

will set HRTIMER_MODE_HARD of might set HRTIMER_MODE_HARD if the task has elevated priority
(RT/DL).

> hrtimer_start_range_ns
> WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard)
> fires since the latter doesn't see the change of mode done by
> init_sleeper

correct.

> The attached patch seems to cure the problem, although I'm not entirely
> sure it's the cleanest fix (if I'm right at all about it :).

The problem seems to be that __wait_event_hrtimeout() does
hrtimer_start_range_ns() instead of hrtimer_sleeper_start_expires(). So
if you switch that, then it should go away.

> Thanks,
> Juri

Sebastian

2022-06-23 14:00:08

by Juri Lelli

[permalink] [raw]
Subject: Re: [RT] WARNING at hrtimer_start_range_ns

On 23/06/22 12:46, Sebastian Andrzej Siewior wrote:
> On 2022-06-15 16:53:55 [+0200], Juri Lelli wrote:
> > Hi,
> Hi,
>
> > My understanding is that we have
> >
> > fs/aio.c::read_events
> > wait_event_interruptible_hrtimeout
> > __wait_event_hrtimeout
> > hrtimer_init_sleeper_on_stack <- this might mode |=HRTIMER_MODE_HARD on RT
>
> will set HRTIMER_MODE_HARD of might set HRTIMER_MODE_HARD if the task has elevated priority
> (RT/DL).
>
> > hrtimer_start_range_ns
> > WARN_ON_ONCE(!(mode & HRTIMER_MODE_HARD) ^ !timer->is_hard)
> > fires since the latter doesn't see the change of mode done by
> > init_sleeper
>
> correct.
>
> > The attached patch seems to cure the problem, although I'm not entirely
> > sure it's the cleanest fix (if I'm right at all about it :).
>
> The problem seems to be that __wait_event_hrtimeout() does
> hrtimer_start_range_ns() instead of hrtimer_sleeper_start_expires(). So
> if you switch that, then it should go away.

Ah, indeed. That looks simpler. :)

Will test and put a v2 together.

Thanks!
Juri