Skip to contentSkip to conversation

Kernel semaphores

DISCUSSION Pete Zaitcev 2 participants 2 messages Sep 14, 2004 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 4 lines · On Tue, 14 Sep 2004 15:25:24 +1000 James Roper <[email protected]> wrote:

[] So my question is, if my semaphore is
causing that error, what possible things could be triggering it? Could it be an interrupt while waiting to acquire the semaphore? I'm using the down_interruptible() to acquire and up() to release.

You have to use spinlocks to provide a mutual exclusion to interrupts. However, a process on CPU cannot sleep while holding a spinlock. So, sometimes it's needed to create a derivative locking scheme, based on spinlocks. A common trick is to combine semaphores and spinlocks. I cannot be more specific without knowing your code.

-- Pete

#2 lore raw
Quoted 6 lines · On Tue, 14 Sep 2004 15:25:24 +1000 James Roper <[email protected]> wrote:

[] So my question is, if my semaphore is
causing that error, what possible things could be triggering it? Could it be an interrupt while waiting to acquire the semaphore? I'm using the down_interruptible() to acquire and up() to release.

You use downXXX() and upXXX() to serialize user-access to your driver in the usual APIs like read() and write().

You need to use spin-locks to protect critical sections from being changed by interrupts. You need to do this even if you are using downXXX() and upXXX(). Basically downXXX() will cause a semaphore-contender to sleep. You can't do this in an interrupt.

Any code in an ISR or on the timer-queue must never execute downXXX(). It is possible, under extremely strange circumstances to execute upXXX() from an interrupt because it doesn't sleep, only releases the lock (if the resource is locked). You don't design anything to do this, though. It can be used as a work-around to release a resource that was acquired by a task that hung or died.

Cheers,
Dick Johnson
Penguin : Linux version 2.4.26 on an i686 machine (5570.56 BogoMips).
Note 96.31% of all statistics are fiction.