ia64: fix timer cleanup regression
authorArnd Bergmann <arnd@arndb.de>
Sat, 19 Dec 2020 09:05:45 +0000 (10:05 +0100)
committerArnd Bergmann <arnd@arndb.de>
Tue, 12 Jan 2021 16:01:42 +0000 (17:01 +0100)
commit796130b1de29575e2e3fc3b0da4bda162b750db7
tree0cd0887b05497c5a8f3a9c445ee1e20e179182b8
parent5c8fe583cce542aa0b84adc939ce85293de36e5e
ia64: fix timer cleanup regression

A cleanup patch from my legacy timer series broke ia64 and led
to RCU stall errors and a fast system clock:

[  909.360108] INFO: task systemd-sysv-ge:200 blocked for more than 127 seconds.
[  909.360108]       Not tainted 5.10.0+ #130
[  909.360108] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  909.360108] task:systemd-sysv-ge state:D stack:    0 pid:  200 ppid:   189 flags:0x00000000
[  909.364108]
[  909.364108] Call Trace:
[  909.364423]  [<a00000010109b210>] __schedule+0x890/0x21e0
[  909.364423]                                 sp=e0000100487d7b70 bsp=e0000100487d1748
[  909.368423]  [<a00000010109cc00>] schedule+0xa0/0x240
[  909.368423]                                 sp=e0000100487d7b90 bsp=e0000100487d16e0
[  909.368558]  [<a00000010109ce70>] io_schedule+0x70/0xa0
[  909.368558]                                 sp=e0000100487d7b90 bsp=e0000100487d16c0
[  909.372290]  [<a00000010109e1c0>] bit_wait_io+0x20/0xe0
[  909.372290]                                 sp=e0000100487d7b90 bsp=e0000100487d1698
[  909.374168] rcu: INFO: rcu_sched detected stalls on CPUs/tasks:
[  909.376290]  [<a00000010109d860>] __wait_on_bit+0xc0/0x1c0
[  909.376290]                                 sp=e0000100487d7b90 bsp=e0000100487d1648
[  909.374168] rcu:     3-....: (2 ticks this GP) idle=19e/1/0x4000000000000002 softirq=1581/1581 fqs=2
[  909.374168]  (detected by 0, t=5661 jiffies, g=1089, q=3)
[  909.376290]  [<a00000010109da80>] out_of_line_wait_on_bit+0x120/0x140
[  909.376290]                                 sp=e0000100487d7b90 bsp=e0000100487d1610
[  909.374168] Task dump for CPU 3:
[  909.374168] task:khungtaskd      state:R  running task

Revert most of my patch to make this work again, including the extra
update_process_times()/profile_tick() and the local_irq_enable() in the
loop that I expected not to be needed here.

I have not found out exactly what goes wrong, and would suggest that
someone with hardware access tries to convert this code into a singleshot
clockevent driver, which should give better behavior in all cases.

Reported-by: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Fixes: 2b49ddcef297 ("ia64: convert to legacy_timer_tick")
Cc: John Stultz <john.stultz@linaro.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Stephen Boyd <sboyd@kernel.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
arch/ia64/kernel/time.c