__z_clock_nanosleep function was getting current time in cycles, via
k_cycle_get_32(), to perform its time calculations. However, when calling
k_sleep() to actually sleep, times are measured in ticks.
This causes a problem when there's a big skew between the uptime
measured in cycles vs uptime measured in ticks: in some platforms, the
system clock maybe up for a long time already when Zephyr starts
counting ticks, for instance, while downloading an image via PXE. In
this case, the calculations done inside __z_clock_nanosleep end up
measuring a much bigger current time than expected, thus sleeping too
much, basically all the time since system clock initialization.
This patch fixes that by avoiding the cycle trip: stick to ticks,
instead. They start counting from Zephyr initialization instead, which
is the expected uptime.
Fixes#69608
Signed-off-by: Ederson de Souza <ederson.desouza@intel.com>
d83a649a67