[libopencm3-devel] Proposed patch for RTC race condition

[Tomaž Š. <tom...@ij...>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi

First, I apologize for the amount of patches I'm sending to this list.
Please tell me, if you prefer some other way I can contribute them to
libopencm3.


I have some code that looks like this:

rtc_set_counter_val(0)

while(...) {
	int timestamp = rtc_get_counter_val()
	... some time consuming operation ...
}

The problem is that there is a race condition since
rtc_set_counter_val() doesn't wait for the RTC register write to finish
so timestamp might be set to the uninitialized value of the RTC counter
on the first iteration.

Although I can fix this in my code I think this is a somewhat surprising
result of using libopencm3 RTC functions.

rtc_set_counter_val() should wait until the RTC actually gets updated
before returning. There is code in rtc_exit_config_mode() that does just
that, but is commented out with "Unnecessary since we poll the bit on
config entry(?)". I think code above is one example, where this might be
necessary.

Best regards
Tomaž

- -- 
Tomaž Šolc, research assistant
SensorLab, Jožef Stefan Institute
http://sensorlab.ijs.si
mail: tom...@ij...
blog: http://www.tablix.org/~avian/blog
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iQEcBAEBAgAGBQJPYwr2AAoJED848b3tOBsIqtgH/1lmxb2BxbQLG1R9fwh9gLGW
Ng/CoSI5rOeMCGDnVcM2MwJ/bddNtQqAqobgA1+XEFha+V2rjg+7DahJ+9xo4wlj
R+IeRfihr/GtfEqMHl5a58q4IEFchC/xDLV8qkNGIN2VaIhVKfyhWCH0/hfxkhpY
wAjDJyMC4rDYfmkF59DMuDTYlJSLkliWSBaqQwIgfrcg4q/VM3JvAr84rmHPOGBr
j+uJTv0GZTXxmIEjFeNTvV26iSTZE9lYaGv/94I3HNlBqsVPR5D2Jt2GlxLWEtlo
MXy+gfU2ZQhIQwfhJDmusu+GZ4IxhiPeJdYNHnGkVrF3kEY+cDfuWkS/fCynQRA=
=A83y
-----END PGP SIGNATURE-----