PROBLEM: Long Workqueue delays.

From: Jim Baxter
Date: Mon Aug 17 2020 - 07:48:18 EST


We have issues with the workqueue of the kernel overloading the CPU 0
when we we disconnect a USB stick.

This results in other items on the shared workqueue being delayed by
around 6.5 seconds with a default kernel configuration and 2.3 seconds
on a config tailored for our RCar embedded platform.

I am aware there will be delays on the shared workqueue, are the delays
we are seeing considered normal?



We first noticed this issue on custom hardware and we have recreated it
on an RCar Starter Kit using a test module [1] to replicate the
behaviour, the test module outputs any delays of greater then 9ms.

To run the test we have a 4GB random file on a USB stick and perform
the following test:


- Load the Module:
# taskset -c 0 modprobe latency-mon

- Copy large amount of data from the stick:
# dd if=/run/media/sda1/sample.txt of=/dev/zero
[ 1437.517603] DELAY: 10
8388607+1 records in
8388607+1 records out


- Disconnect the USB stick:
[ 1551.796792] usb 2-1: USB disconnect, device number 2
[ 1558.625517] DELAY: 6782


The Delay output 6782 is in milliseconds.

Thank you for you help.
Jim Baxter

[1] Test Module:
// SPDX-License-Identifier: GPL-2.0
/*
* Simple WQ latency monitoring
*
* Copyright (C) 2020 Advanced Driver Information Technology.
*/

#include <linux/init.h>
#include <linux/ktime.h>
#include <linux/module.h>

#define PERIOD_MS 100

static struct delayed_work wq;
static u64 us_save;

static void wq_cb(struct work_struct *work)
{
u64 us = ktime_to_us(ktime_get());
u64 us_diff = us - us_save;
u64 us_print = 0;

if (!us_save)
goto skip_print;


us_print = us_diff / 1000 - PERIOD_MS;
if (us_print > 9)
pr_crit("DELAY: %lld\n", us_print);

skip_print:
us_save = us;
schedule_delayed_work(&wq, msecs_to_jiffies(PERIOD_MS));
}

static int latency_mon_init(void)
{
us_save = 0;
INIT_DELAYED_WORK(&wq, wq_cb);
schedule_delayed_work(&wq, msecs_to_jiffies(PERIOD_MS));

return 0;
}

static void latency_mon_exit(void)
{
cancel_delayed_work_sync(&wq);
pr_info("%s\n", __func__);
}

module_init(latency_mon_init);
module_exit(latency_mon_exit);
MODULE_AUTHOR("Eugeniu Rosca <erosca@xxxxxxxxxxxxxx>");
MODULE_LICENSE("GPL");