Hi community,
Any suggestions on how should I proceed with my issue?
Description of my issue:
- My board kernel prints: "sched: RT throttling activated", then reboot
(there is no back-trace in UART console).
- I have sc16is752 (I2C based UART) on my board.
- As Atheros AR9331 does not have dedicated I2C controller, I am using
bitbang I2C.
- This is likely related to below line of sc16is752 driver
(drivers//tty/serial/sc16is7xx.c)
ret = devm_request_threaded_irq(dev, irq, NULL, sc16is7xx_ist,
- './ps -eo pid,nice,policy,cmd' confirms that above is a realtime
kernel thread (with SCHED_FIFO scheduling policy)
- Occurrence rate is about once or several times a day, after I stress
I2C/UART traffic.
- Occurrence rate is much lower if I do *NOT* stress I2C/UART traffic.
My setup:
- Atheros AR9331 chip, which is MIPS32 24kc
- OpenWRT tag 15.05
- Linux kernel v3.18.29
- Not preempt kernel (CONFIG_PREEMPT_NONE=y)
- As AR9331 does not have dedicated I2C controller, I am using bitbang
I2C as below:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/tree/drivers/i2c/algos/i2c-algo-bit.c?h=linux-3.18.y
- source code of sc16is752 driver sc16is7xx.c is here:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/tree/drivers/tty/serial/sc16is7xx.c?h=linux-3.18.y
Question 1): There seems a fundamental question about OpenWRT mips linux
kernel?
- OpenWRT tag 15.05
- In OpenWRT file target/linux/ar71xx/Makefile, there is CPU_TYPE=34kc
- Atheros AR9331 datasheet says it is 24kc (rather than 34kc)
- With CPU_TYPE=34kc, it uses gcc options "-mips32r2 -mtune=34kc"
Question 2): I have no idea what causes the reboot? It is a chip
hardware reboot, or function similar to emergency_restart() is called
somewhere in the kernel? Kernel console only shows "sched: RT throttling
activated" and reboot, there is no back-trace.
If I enable preempt kernel (CONFIG_PREEMPT=y), "RT throttling activated"
still happens, and there is no reboot.
- After "RT throttling activated" happens, I use "top -d 1" to check
CPU usage, I observed at least once sc16is752 threaded-irq handler
kernel thread occupies 95% of CPU for some seconds (not sure how many
seconds), then sc16is752 threaded-irq handler CPU usage drops to normal,
and the whole OpenWRT/Linux system is normal.
- I did observe once that sc16is752 threaded-irq handler CPU usage
stays at 95% forever, it seems like a dead-lock inside sc16is752
threaded-irq handler.
- It seems a dead-lock similar (in some way?) to below patch:
https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/commit/?h=linux-4.4.y&id=d78006d2345f87889918a8a7aa3764628ff84263
Question 3) Is there any explanation why board does not reboot when I
enable preempt kernel (CONFIG_PREEMPT=y)?
Question 4) I am not sure if this is related to MIPS32r2 34kc (with
9-stage pipeline multi-threaded processor core), while Atheros AR9331 is
24kc (with 8-stage pipeline processor core)?
https://www.imgtec.com/mips/classic/
I have a temporary work-around:
- In bitbang I2C driver, use hrtimer-backed usleep_range() (rather than
busywait udelay()), then this issue is gone (verified by stress test).
- As I do not know how long it takes to arm hrtimer on AR9331 platform,
I use usleep_range(10us, 20us), this greatly reduces bitbang I2C
throughput, and reduces sc16is752 (I2C based UART) baudrate.
Thanks.
Xuebing Wang
