Hi,
El Thu, Feb 25, 2010 at 08:22:59PM +0100 Alessandro Rubini ha dit:
i was/am working on a new version of the patch, taking into account your remarks about the unit of TIMER_FREQ and fixing some issues discussed with Alessandro Rubini off-list, who worked on a similar patch.
Actually, I checked the point we disagreed about, which is the unit of get_ticks() and get_tbclk(). You currently return hw-ticks in get_ticks, and CONFIG_SYS_HZ (i.e. 1000) in get_tbclk. However, these two functions are expected to be used together, so they must be consistent in their return value.
actually there is no disagreement between us, i totally agree with you that the return value of get_ticks() should be in CONFIG_SYS_HZ resolution and consistent with get_tbclk(). the patch i sent you yesterday off-list fixes exactly this.
It's true that the functions are little used (they are mostly used in ppc code, within cpu/*/interrupts), and that's why I didn't even provide them in cpu/arm926ejs/nomadik/timer.c. All few users assume they are consistent, but there is no documentation:
tornado% grep -qr get_tbclk README* doc || echo not found not found tornado% grep -qr get_ticks README* doc/* || echo not found not foundI've made a quick tour of all definitions in cpu/ and here is the result. As you see, at91 (which you used as reference, I understand) is wrong, while all the others use either hwticks or SYS_HZ consistently.
yes, i used precisely at91 as reference, i liked it's code structure and didn't notice that it is wrong in this point.
thanks for your research!