Regression: w1_therm: sysfs w1_slave sometimes report 85 degrees Celsius

[Stefan Wahren <stefan.wahren@i2se.com>]

Hi,

recently we switch on our Tarragon board (i.MX6ULL) to Linux 6.1 and 
noticed that the connected 1-wire temperature sensors 
(w1_therm.w1_strong_pull=0) sometimes (~ 1 of 20 times) report 85 
degrees Celsius, which is AFAIK the only way to report errors to the 
1-wire master:

sys/bus/w1/devices/28-04168158faff# cat w1_slave
50 05 4b 46 7f ff 0c 10 1c : crc=1c YES
50 05 4b 46 7f ff 0c 10 1c t=85000

I wasn't able to reproduce this issue with the old kernel 4.9.

After that i successfully bisected the issue to this commit:
67b392f7b8ed ("w1_therm: optimizing temperature read timings")

Unfortunately this commit contains a lot of independent changes, which 
makes it hard to figured out the cause of this issue. So i tried to 
split this patch in seven independent changes [1]. Now i was able to 
bisect the cause further to this change [2] which seems to rework the 
pullup handling within read_therm().

Looking closer at the code change and verify it some debug messages, the 
change inverted the locking behavior (before: no pullup -> keep lock, 
after: no pullup -> release lock during sleep).

Before:
	if (external_power) {
		mutex_unlock(&dev_master->bus_mutex);

		sleep_rem = msleep_interruptible(tm);
		if (sleep_rem != 0) {
			ret = -EINTR;
			goto dec_refcnt;
		}

		ret = mutex_lock_interruptible(&dev_master->bus_mutex);
		if (ret != 0)
			goto dec_refcnt;
	} else if (!w1_strong_pullup) {
		sleep_rem = msleep_interruptible(tm);
		if (sleep_rem != 0) {
			ret = -EINTR;
			goto mt_unlock;
		}
	}

After:
	if (strong_pullup) { /*some device need pullup */
		sleep_rem = msleep_interruptible(tm);
		if (sleep_rem != 0) {
			ret = -EINTR;
			goto mt_unlock;
		}
	} else { /*no device need pullup */
		mutex_unlock(&dev_master->bus_mutex);

		sleep_rem = msleep_interruptible(tm);
		if (sleep_rem != 0) {
			ret = -EINTR;
			goto dec_refcnt;
		}

		ret = mutex_lock_interruptible(&dev_master->bus_mutex);
		if (ret != 0)
			goto dec_refcnt;
	}

I don't believe this is intended. After inverting the strong_pullup 
check, the issue wasn't reproducible on our platform anymore. But i'm 
not sure this is clean.

Best regards

#regzbot introduced: 67b392f7b8ed

[1] - https://github.com/chargebyte/linux/commits/v6.1-tarragon_w1
[2] - 
https://github.com/chargebyte/linux/commit/17ca863a32a6a1bdd376959f05c954bef12fc1b5