[DjDeveloperr]

What changed

• add the opt-in Aura PV discovery client (CONFIG_AURA_PV) and ABI 0.1 UAPI
• expose bounded discovery and arithmetic self-test ioctls through /dev/aura-pv
• add the executable selftest and document the ABI and explicit limitations

Validation

• drivers/virt/aura_pv.o builds with CONFIG_AURA_PV=m
• selftest compiles and exits 4 when /dev/aura-pv is unavailable, as expected
• m1n1 host endpoint default-off/opt-in build and contract checks pass

This PR intentionally does not claim /dev/kvm guest-run, EL1 entry, stage-2 translation, virtual interrupts, timers, or nested guest context switching. The current m1n1 host has one global stage-2/VTTBR context, so a safe per-VM protocol is still required before ABI 0.2 guest execution.

[RyanTheTide]

Review, not merged yet. It builds in a test merge on integration/aurora 86cb1c3 (with 731 and 856), and the header passes HDRTEST. A few things to fix first:

1. include/uapi/linux/aura_pv.h uses BIT_ULL() from <linux/bits.h>. The uapi linux/bits.h only has __GENMASK, so userspace that uses AURA_PV_FEATURE_* or AURA_PV_OUTPUT_* won't compile. HDRTEST doesn't catch it because nothing expands the macros. _BITULL() from <linux/const.h> works in both.
2. aura_pv_hvc() is hand-rolled and doesn't list x4 to x17 as clobbered. That's only safe if Aura promises SMCCC 1.1 register preservation for these function IDs. arm_smccc_1_1_hvc() would cover it, or the asm needs the clobbers. If the x19/x20 sentinels are there to check that Aura preserves callee-saved registers, the client never compares them after the call.
3. .compat_ioctl = aura_pv_ioctl passes a 32-bit pointer through unchanged. compat_ptr_ioctl is the usual fix.
4. There's no module_exit, so the module can't be unloaded, and with no modalias it never autoloads. That's fine for a diagnostic, but the docs should say it has to be loaded by hand.
5. asahi.config turns it on as =m for every distro kernel. It's harmless without the DT node, but the Aura side that publishes aurora,aura-kvm-pv-0 isn't merged yet (aura PR 33 is still a draft). Until it is, this would ship a module nothing can bind.

I'd hold it until the Aura host side lands and one guest run has passed on the Neo. Happy to re-run the build once it's updated.

[DjDeveloperr]

Heads-up: #1363 (merged as 8484356) removed APIF from integration/aurora, including tools/testing/selftests/aurora-apif/. This PR still adds a test under that directory, so it needs a small fixup on rebase: move the test, or drop it if it's APIF-only.