Mike Frysinger wrote: > On Friday 02 October 2009 08:30:51 Wolfgang Denk wrote: > >> Ingo van Lil wrote: >> >>> The CFI driver does not reset the device's watchdog, so long-running >>> flash operations will cause the watchdog timer to expire. A comment in >>> flash_status_check() suggests that udelay() is expected to reset the >>> watchdog, but I can't find any architecture where it does. >>> >> Please have a closer look, then. On PowerPC, udelay() >> ["lib_ppc/time.c"] calls wait_ticks(), which in turn >> ["lib_ppc/ticks.S"] calls WATCHDOG_RESET >> >> If this is missing in other architectures, it should be fixed at the >> root cause, i. e. in udelay() or in the respective support routines. >> > > Blackfin is missing it as well as i really had no idea it was supposed to be > there. certainly no doc states this requirement. perhaps it'd make sense to > break apart the common stuff to a common udelay() that does things like call > the watchdog and then call the implementation __udelay(). there should be at > least a doc/README.arch that includes these kind of details ... > -mike > > At this point it might be appropriate to ask if including such a reset in udelay() is a good idea. The way it is, no "infinite loop" in u-boot that contains a udelay() will ever allow the watchdog to time out. That restriction is somewhat non-obvious. I would argue that there are basically two classes of commands in u-boot, those that should execute so fast that there is no way the watchdog should be triggered, and commands that may take "a long time" during which the watchdog may erroneously be triggered if there is no provision made to reset it. In the second case, the resetting of the watchdog must be placed where "measurable progress" towards command completion is being made, i.e. there is a certainty that we are getting closer to a command complete. Just because the program invokes udelay, there is no assurance that measurable progress is being made. The udelay may be part of a process that will never complete. Therefore, having a watchdog reset in udelay seems like a less than optimal idea in general. Maybe now would be a good time to look at removing it?
Bill Campbell > ------------------------------------------------------------------------ > > _______________________________________________ > U-Boot mailing list > U-Boot@lists.denx.de > http://lists.denx.de/mailman/listinfo/u-boot > _______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
