Hi Ravi and List, [I maybe completely wrong as this interrupt thing is quite dangerous to talk about ;) .. apologies upfront] On Mon, Jul 7, 2008 at 10:03 AM, ravikumar <ravikumar.vallabhu@xxxxxxxxx> wrote: > Hello All , > I've a few questions. > 1. Lets suppose i've some critical section and it is synchronized with > locks. A thread came and it's execution is in the middle of critical > section, at this time an interrupt was occurred which also executes > critical section. As interrupts can't wait on any lock it will get lock and > try to execute the critical section and above thread was not yet completed. > So inconsistency . How this was handled in Linux. What I understand is that If in critical section (like spin_lock) interrupts would be disabled (at least on local CPU) and hence till the working kernel thread releases the lock (and thus enabling interrupts) no interrupt can break this kernel thread. In case this is a multiprocessor system, probably the interrupts (from other CPU) would take have to wait for the critical section lock to be released - which I think can be a candidate for bad design (if that critical section has any code for sleeping). > 2.Some process is running in kernel mode, at this a kill was sent to this > process, does this signal will delivered to the process while it is in > kernel mode. I don't think so - not until a proper signal handling check is there in the kernel control path in which process is right now - which I think would seldom be the case (may be if that is your own IOCTL handler, for example, in which checking has been done - or may be a kernel daemon with signal checking). But, if a signal is sent from user-space (with appropriate credentials) it should be queued with the process task_struct (this 'should' is my understanding). Please feel free to point out mistakes in this answer - so that I can correct myself as well. > > And also please provide some links how threads ,signals and interrupts will > be mixed. I find Understanding Linux kernel book too good in this respect (but yes, I had to read each topic many times to understand substantially). [Also, Scott had already answered even before I could press enter - but still I took the liberties of expressing my understanding] -- Shreyansh -- To unsubscribe from this list: send an email with "unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx Please read the FAQ at http://kernelnewbies.org/FAQ
