I have never used these kind of boards.It seems to me
NMI is implemented by interrupt controller,to cpu,it is a
normal interrupt source.If 'NMI' in Adeel's board is like
what you repeated,request_irq could just use cpu's pin number
as the 'irq' parameter. I think NMI has been used in many
boards that only means 'non-maskable' to interrupt controller,
not cpu. If it is this case, NMI interrupt is a normal case.
From: Ralf Baechle [mailto:email@example.com]
Sent: Tuesday, November 11, 2003 11:40 AM
To: Liu Hongming (Alan)
Cc: Adeel Malik; firstname.lastname@example.org
Subject: Re: How to request an IRQ for NMI on MIPS Processor
On Tue, Nov 11, 2003 at 10:51:50AM +0800, Liu Hongming (Alan) wrote:
> when using request_irq(...),the parameter irq is a value specified by you,
> of course,when porting linux for your board,you should have specified value
> for every IRQ. 0--5 only means CPU pin for interrupt,unless that only one
> may occur on this pin,you will use other value in request_irq,instead of
> all in all, when we touch request_irq,it is board specific.When your board
> been made out,all interrupts have specific route to cpu(unless you have IRQ
> router,since embedded,need this??).If you have more external interrupts than
> cpu pins,maybe you have cascaded many interrupt using one cpu pin.So,
> the parameter irq in request_irq is determined by your board and your
> for interrupt handling.Just ask that guy that ported linux.He will tell
> you.If you
> are using linux ported by others,have a look at BSP codes.
your answer is correct for normal interrupts, not the NMI. The NMI goes
through the firmware and none of the board support code so far bothered
to make it available via request_irq as it has several severe limitations.
To repeat one of my prior postings about the NMI:
NMI on MIPS is pretty much miss-designed for use in application code; it's
use should be limited to debugging and recovery from catastrophical
failure. The reason for this is the way this exception is handled:
- the BEV, TS, SR, NMI and ERL bits in c0_status are overwritten - that is
their old state is lost.
- c0_errorepc is overwritten - again that means the old value is lost so
in case the NMI interrupts an exception handler that uses this register
such as the cache error handler you can not resume execution.
- the program counter is set to 0xbfc00000. Most likely a slow flash
memory is mapped at this address but in any case it's an uncached
segment of the address space so execution will be even slower.
- execution will pass through the firmware. That means you can only
use the NMI at all if firmware provides some kind of hook.
It seems pretty clear to me that the MIPS designers never intended the
NMI for anything else than catastrophic events.