Lock used in interrupt handlers.

[Calvin Lee]

Because of safety the PIC and KEYBOARD static interfaces are spin::Mutex-es. The concepts of locks are very problematic in an interrupt because if we run an isr when the lock is being used, the isr will block forever. This can very easily happen, for example I have to use a lock to access the PICs to signal an end of interrupt. If the pic immediately sent another interrupt (which could happen) the isr called would not be able to even send an end of interrupt. We need to use some sort of safety for these types, but a lock won't work.

[Calvin Lee]

Readings on the subject: (from our gracious friends on #rust-osdev)

• Paul E. McKenny
  • and his work on RCU
• Writing Lock Free Code
• Threadable NAPI polling, softirqs, and proper fixes (LWN)

I'll go through these and think about what we should do for interrupts in the kernel.

[Calvin Lee]

I believe that the biggest problem in this was the WRITER lock, and that is now solved. I will close this soon unless there is any objection. Specific locking structs that are creating problems should probably have their own issues or pull requests.