On 2/7/06, Hareesh Nagarajan <hareesh.nagarajan@xxxxxxxxx> wrote: > Hi Gaurav, > > Thank you for your detailed reply. Some comments appear below. > > On 2/7/06, Gaurav Dhiman <gauravd.chd@xxxxxxxxx> wrote: > > I dont think you need to use the global falg variable to sync the > > access to shared data structure; INTERRUPTED or IN_PROGRESS flags. > > This sync can be handled directly by spinlock which you use to sync > > the access to global flag variable. > > > > So remove the flag variable and directly sync the access with spinlock. > > I used the flag variable as an example. Basically, if sys_foo() gets > holds of the spin lock first, then it sets the flag (which is embedded > elsewhere; it is rather fine grain and not as coarse grain as I've > mentioned) to FOO_IN_PROGRESS (among states the flag can be in). > > In due course, sys_foo(), will release the lock, perform operations > that sleep and grab it again, so on and so forth and finally it will > set the flag to FOO_COMPLETED while holding the lock and soon after > which it will release the spin lock. > > Now once sys_foo() releases the lock, and the flag isn't set to > FOO_COMPLETED as yet, so when the lock the grabbed in interrupt > context, the code there checks to see if FOO_IN_PROGRESS has been set. > If it has, then it sets the FOO_ABORT_INTERRUPTED flag and goes away. > > But if the timer code, gets the lock first, then it does a bunch of > (non-sleeping) stuff on the shared data, sets the status to > FOO_ABORT_COMPLETED and releases the lock. So now, when sys_foo() gets > called, it sees this new state and does something else. > > Summary: If sys_foo() is in progress (after it has grabbed the spin > lock for the first time and FOO_ABORT_COMPLETED) the timer interrupt > code musn't do certain things. If sys_foo() hasn't started yet, then > the timer interrupt can do certain other things and as a consequence > sys_foo() must do certain other things. ok, I was taking the simpler case. Basically you dont want the timer to do certain things once the sys_foo() has been started and not completed yet. Somewhat same mechanisum has been there to sync the access between multiple CPUs on SMP machine to "irq_desc" interrupt datasructure regards, -Gaurav > > Thanks! > > Hareesh > -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
