On Thu, Feb 13, 2003 at 10:31:12AM -0500, Derrick 'dman' Hudson wrote: > On Wed, Feb 12, 2003 at 09:53:15PM +0100, martin f krafft wrote: > | i have a process waiting for data on a device that doesn't exist > | anymore (USB). now the process is listed as uninterruptibly sleeping. > | i want to get rid of it, but kill -9 doesn't do anything, the process > | remains. > | > | what must i do to kill this process? it can't be that i can get > | a process to lock into the scheduler so as to not get out anymore... > > The process isn't "locked into the scheduler" -- the (CPU) scheduler > always bypasses it because the process doesn't want any CPU right now. > The process' PC (Program Counter) is currently pointing to > instructions that are part of the kernel. It is waiting for some I/O > activity. The processs can't be killed outright because that would > harm the integrity of the running kernel. Once the process is back in > user space it can be killed with SIGKILL. Unfortunately, it isn't > likely the process will ever become unblocked.
The normal unix-way is to deliver signals at the transition from kernel to user space, just before the kernel reliquishes control of the cpu to the process. When a process gets stuck in kernel space, usually waiting for an interupt that doesn't arrive due to some hardware related problem, it never comes to the point where the kernel checks for the presense of signals. A non running process is always in kernel space. It is either waiting for some event or it is waiting for the cpu that the scheduler has taken away from it. In any case it is in kernel space. When a signal is delivered, a bit is set in a signal mask in the receiving process control block. In a normal one-processor system the recipient cannot be running, because the sender is, so the signal is delivered later when the scheduler decides to to give the cpu to the recipient. With such a system there is really no way of dragging a recalcitrant process out of kernel mode. If it is stuck in there, it is stuck. Signals are registered, but not delivered, because the transition to user-space never happens. If Linux does as Unix, changing it will be very difficult. It will be a major change in kernel logic. -- René Seindal ([EMAIL PROTECTED]) http://sights.seindal.dk/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
