Keep per-timer interrupt data together

[andrew]

Summary

Eliminate a redundant resource array allow possible use by bhyve later.

Sponsored by: Arm Ltd
Sponsored by: The FreeBSD Foundation

[andrew]

Rebase and rework as a possible alternative to D39074 to remove redundant information

[kevans]

IMO let's go one step further and remove the significance of the index into sc->irqs

sys/arm/arm/generic_timer.c

659

This is guaranteed by virtue of the fact that GT_VIRT is not RF_OPTIONAL, we can do this unconditionally for arm64 (and the other block unconditionally for arm)

675

Iterate over all nitems(sc->irqs), filter on actual rid

683

0 to i

691

Ditto above; we can do this unconditionally, we know it's present.

694–695

These could be reduced to just sc->physical && HAS_PHYS or even just HAS_PHYS; we're guaranteed at least the phys timer (if not the sec-phys timer)

[kevans]

I'm actually happy with this as-is whether we go the extra step right now or not, forgot to click the button.

[andrew]

I'll go with this for now to be a mechanical change, then plan to clean up in a future change.

There are some complications with virtual timer interrupts and bhyve. The easiest way to handle it is to not enable them when the hypervisor is enabled & use the physical timer in the kernel.

sys/arm/arm/generic_timer.c

659

I'll replace with HAS_HYP as we need to use the physical timer in the kernel with bhyve.

694–695

I'll replace the resource checks with a KASSERT. sc->physical is incorrect here because firmware may leave the physical timer enabled so this is careful to disable the physical timer if present.