[AxxzyWasTaken]

has_apple_apflg() and has_apple_afp() (#868, #869) ask the hypervisor whether it emulates ASPSR_EL1 / AFPCR_EL0 whenever the kernel isn't in hyp mode. That's right under KVM, which traps IMPDEF registers through HCR_EL2.TIDCP. But on M4 machines aura stays at EL2 as a monitor and enters Linux at EL1 with PSCI over SMC (monitor_required() in kboot.c, SMCCC 1.2), or with no PSCI and conduit NONE on older aura builds. The vendor services are only discovered from KVM, so kvm_arm_hyp_service_available() is always false, and APFLG is never detected, although nothing traps ASPSR_EL1 there.

This asks only over the HVC conduit, which is what KVM guests use. Anywhere else there's no hypervisor to answer, so the existing AIDR_EL1 and "enable bit sticks" checks decide. Guests with an HVC conduit (KVM) behave as before. AMX keeps its check: at EL1 it also needs EL2 to set AMX_CTL_EL2.EN_EL1.

Testing (J616C, M4 Max, EL1 under aura's monitor, kvm: HYP mode not available, no psci node)

The same check on our 7.2.10 J616C tree (the merged #868 plus the first version of this, which tested for conduit NONE). That boot used an older aura with no /psci, where both versions give the same answer:

• CPU features: detected: x86 parity and adjust flags (Apple APFLG), apple_apflg on all 14 CPUs in /proc/cpuinfo, AT_HWCAP3: 0x8000000000000000. Without it the hypervisor check fails here (no vendor services are discovered), and HWCAP3 stays 0.
• A test that turns APFLG on with PR_AURORA_SET_X86_FLAGS in 64 threads (14 CPUs), writes PF/AF through NZCV, then spins or makes a syscall and reads them back: 0 mismatches in 12.8M rounds. Threads that didn't turn it on read PF/AF as 0.

AFP isn't testable here: the ACTLR enable bit doesn't stick on this M4 Max, so it stays off either way.

Not booted: current aura main, with its PSCI over SMC. That case comes from reading psci.c and kboot.c in aura, and it's why the check is now != SMCCC_CONDUIT_HVC: the first version would have left APFLG off there.

On integration/aurora: builds with W=1 and no warnings with APFLG/AFP on and off, and checkpatch is clean.

[RyanTheTide]

Merge train (MTL, 06/11). Gates passed on test tree 0412501d (integration/aurora 2661d71 + this batch):

• Signed-off-by from the author on every commit. Adversarial review done.
• W=1 on the touched objects: 0 new warnings. KUnit: drm/apple 112/113 (1 skip), hid 19/19, SEP 158/163 (the 4 known apple_sep_host_store failures, profile tests all pass). dt_binding_check clean. apple-dcp host selftests 86/86, run-host-tests at base parity (with Merge train: G17 hang confinement (#928) plus its host-test fixture #989's fixture).
• Regression boots of 7.3.0-aurora-mtl1106a, compared with the fleet16/fleet17 rows: M2, M1 Pro, J700 and J616s all clean (0 err-level lines, 0 AVCs, displays, Wi-Fi, BT, Touch ID, camera, speakers at minimal volume with speakerguardd, s2idle, perf at parity).
• arm64: apple: Ask a hypervisor about IMPDEF registers only over HVC #910: with it the J700, M2 and J616s now detect APFLG; all three run stable through s2idle. We have no J616C, so your own M4 Max testing is the proof there.