The Integrators were using the .nr_irqs field of the machine
descriptor to pre-allocate a number of descriptors at boot. This
is not right: the irq chip implementations should allocate
their descriptors themselves, and as a result the simple
irqdomain code warns about it.
Get rid of this by just deleting the .nr_irq field from the
machine descriptors but take care: doing so makes the default
implementation hog the first 16 IRQ numbers, so these cannot
be used by the still static IRQ number assignments in the
ATAG boot case. So for these, bump the IRQ numbers to begin
at 64 and upward.
Introduce an offset to offset all IRQ numbers if need be,
though we don't expect to do that again as device tree comes
along.
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
This does two things to the FPGA IRQ controller in the versatile
family:
- Convert to MULTI_IRQ_HANDLER so we can drop the entry macro
from the Integrator. The C IRQ handler was inspired from
arch/arm/common/vic.c, recent bug discovered in this handler was
accounted for.
- Convert to using IRQ domains so we can get rid of the NO_IRQ
mess and proceed with device tree and such stuff.
As part of the exercise, bump all the low IRQ numbers on the
Integrator PIC to start from 1 rather than 0, since IRQ 0 is
now NO_IRQ. The Linux IRQ numbers are thus entirely decoupled
from the hardware IRQ numbers in this controller.
I was unable to split this patch. The main reason is the half-done
conversion to device tree in Versatile.
Tested on Integrator/AP and Integrator/CP.
Cc: Grant Likely <grant.likely@secretlab.ca>
Acked-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
This converts the Integrator AP/CP to use sparse IRQs.
Tested on both machines.
Acked-by: Rob Herring <rob.herring@calxeda.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
