Hi Reinhard,
Le 22/01/2011 12:00, Reinhard Meyer a écrit :
Dear Albert ARIBAUD,
this is not an ARM local issue.
Well, there *is* an ARM specific side of it (use of gd variables during relocation), and that is what prompted me to start the RFC, but generalization to U-boot is welcome if it leads to a satisfactory solution. If not, at least I'll adopt a solution for ARM.
The timeouts are used in generic drivers all around u-boot.
Have a grep for get_timer, reset_timer...
The most ugly use is with reset_timer involved, where the internal pseudo-tick is reset to zero, so all calls to get_timer are relative to that moment.
We are looking at replacing all those occurrences of reset_timer and get_timer with "better" methods.
Seems to me this replacement is quite straightforward, as most uses of reset_timer() and subsequent get_timer are actually loops, functionally the same as our proposals, only instead of using a local to store the start or end time, they reset ticks to zero -- which, I concur, is a Bad Thing.
But 'reset_timer()' calls just need to be replaced with your 'timeout = timeout_init(N)' or my 'then = now()' and 'get_timer() > N' or 'get_timer_masked() > N' by your 'time_out(timeout)' or my 'ms_elapsed(N)'. Seems to me like a tedious effort, possibly involving occasional time-out value conversions, but not a difficulty.
What do I miss?
Best Regards, Reinhard
Amicalement,