Sorry, I may wrong, just to throw some ideas for brainstorming: The program u are looking at (arch/x86/kernel/pci-calgary_64.c) is a watchdog. Check this: http://en.wikipedia.org/wiki/Watchdog_timer#cite_note-0 So accordingly, u are supposed to reset the timer of the watchdog, if not the system will behave in a different way. (eg, For certain motherboard, the watchdog timer will reboot the whole system if instructions streams are not detected at the hard-reset, but it will not be triggered at soft-reset, or in Cisco router, the watchdog is to continuously check that the router did not go into an infinite loop, if timer not reset, it will reboot the whole OS). but i did saw another patch where the mod_timer() is used to emulate missing interrupts. that was just a temporary workaround. but i think the usage of mod_timer() is very diverse, and hardware knowledge comes into play in major way. On Mon, Mar 31, 2008 at 5:19 PM, Mulyadi Santosa <mulyadi.santosa@xxxxxxxxx> wrote: > Hi Robert > > On Mon, Mar 31, 2008 at 3:45 PM, Robert P. J. Day <rpjday@xxxxxxxxxxxxxx> wrote: > > sure, that's what i thought as well, but depending on the semantics > > of how timers work, can't you imagine an implementation that checks > > first and realizes right away that you're *already* at that time and > > just invokes the handler immediately? i mean, are people making that > > kind of call to mod_timer() *assuming* that there will be a time lag, > > even if it's a tiny one. that strikes me as a dangerous assumption to > > make, don't you think? > > this is the point where I think sometimes we can't interpret what the > original code developer(s) really wanted to do just by looking at the > codes. It could be "that way", "this way" or none of both. In short > word, we have less justifications > > > the only other reason i can think of for the above is that someone > > wants to call a handler *right now* or as soon as possible, but wants > > to continue execution anyway. in other words, it just happens to be a > > way to asynchronously invoke a handler -- it just *looks* weird. > > in some degree, I tend to agree on your above "speculation". -- Regards, Peter Teoh -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
