Timer takes longer to expire on ESP32

[zephyrbot]

Reported by Leandro Pereira:

As noted by [~anasnashif], it seems that timers are taking longer to expire than expected. A k_sleep(1000) is taking longer than 1s, as can be evidenced by executing samples such as synchronization or drivers/gpio.

(Imported from Jira ZEP-2456)

[zephyrbot]

by Andy Ross:

Just got far enough along to see this for real. Yeah, a simple loop doing k_sleep(1000) followed by a printk shows that the system timer is operating at 1/4 speed, to within the precision of my finger on a stopwatch app.

A quick check of xternsa_sys_timer.c shows that it didn't get modified for the esp32 port. My guess is that it's hard-configured for the clock configuration of some other board. Shouldn't be hard to fix, though I need to read through the datasheet a bit.

[zephyrbot]

by Andy Ross:

It was actually the CPU clock that was wrong, not the driver. Marketing material for ESP-32 says "160 MHz", but it turns out that it boots at 40 MHz instead. Happily the cycle counter used by the timer has a really simple interface (very similar to the TSC on x86) and it was really easy to validate the hypothesis.

Long term, we're going to want a CPU frequency driver for this board given the 4x difference between default and maximum.

[zephyrbot]

by Ruud Derwig:

(sorry, wrong button)

[zephyrbot]

by Mark Linkmeyer:

Leandro Pereira , please verify this bug is fixed.

[zephyrbot]

by Leandro Pereira:

Works for me.

[zephyrbot]

Related to GH-3930