chris@dokein.co.uk wrote:

> That all makes sense, can you commit your fix?  Certianly for the
> original issue where interrupts stop entirely, the fix to always look up
> the mask is the best fix for now.

sys/arm/iomd/iomd_irq.S has the similar code.
I think we should also fix it, but is there anyone
who try recent -crrent on acorn32?

If we fix both shark/isa/isa_irq.S and arm/iomd/iomd_irq.S,
no other source refers the spl_mask variable (except debug printfs).
Should we also remove it especially from arm/arm32/spl.S?

> My lingering concern is that the IPL level will be raised and block
> interrupts for longer than it should.  Perhaps this is more a concern on
> slower hardware?

Overblocking might cause clock delays (if IPL_BIO interrupts
are handled at IPL_CLOCK), but it won't be fatal, I think.
---
Izumi Tsutsui