]> www.infradead.org Git - users/jedix/linux-maple.git/commit
KVM: arm64: Don't retire aborted MMIO instruction
authorOliver Upton <oliver.upton@linux.dev>
Fri, 25 Oct 2024 20:31:03 +0000 (20:31 +0000)
committerOliver Upton <oliver.upton@linux.dev>
Sat, 26 Oct 2024 14:37:49 +0000 (14:37 +0000)
commite735a5da64420a86be370b216c269b5dd8e830e2
treef0510a11ac22de68ba2fee31a8a36db2b303f5a2
parent8e929cb546ee42c9a61d24fae60605e9e3192354
KVM: arm64: Don't retire aborted MMIO instruction

Returning an abort to the guest for an unsupported MMIO access is a
documented feature of the KVM UAPI. Nevertheless, it's clear that this
plumbing has seen limited testing, since userspace can trivially cause a
WARN in the MMIO return:

  WARNING: CPU: 0 PID: 30558 at arch/arm64/include/asm/kvm_emulate.h:536 kvm_handle_mmio_return+0x46c/0x5c4 arch/arm64/include/asm/kvm_emulate.h:536
  Call trace:
   kvm_handle_mmio_return+0x46c/0x5c4 arch/arm64/include/asm/kvm_emulate.h:536
   kvm_arch_vcpu_ioctl_run+0x98/0x15b4 arch/arm64/kvm/arm.c:1133
   kvm_vcpu_ioctl+0x75c/0xa78 virt/kvm/kvm_main.c:4487
   __do_sys_ioctl fs/ioctl.c:51 [inline]
   __se_sys_ioctl fs/ioctl.c:893 [inline]
   __arm64_sys_ioctl+0x14c/0x1c8 fs/ioctl.c:893
   __invoke_syscall arch/arm64/kernel/syscall.c:35 [inline]
   invoke_syscall+0x98/0x2b8 arch/arm64/kernel/syscall.c:49
   el0_svc_common+0x1e0/0x23c arch/arm64/kernel/syscall.c:132
   do_el0_svc+0x48/0x58 arch/arm64/kernel/syscall.c:151
   el0_svc+0x38/0x68 arch/arm64/kernel/entry-common.c:712
   el0t_64_sync_handler+0x90/0xfc arch/arm64/kernel/entry-common.c:730
   el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:598

The splat is complaining that KVM is advancing PC while an exception is
pending, i.e. that KVM is retiring the MMIO instruction despite a
pending synchronous external abort. Womp womp.

Fix the glaring UAPI bug by skipping over all the MMIO emulation in
case there is a pending synchronous exception. Note that while userspace
is capable of pending an asynchronous exception (SError, IRQ, or FIQ),
it is still safe to retire the MMIO instruction in this case as (1) they
are by definition asynchronous, and (2) KVM relies on hardware support
for pending/delivering these exceptions instead of the software state
machine for advancing PC.

Cc: stable@vger.kernel.org
Fixes: da345174ceca ("KVM: arm/arm64: Allow user injection of external data aborts")
Reported-by: Alexander Potapenko <glider@google.com>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Link: https://lore.kernel.org/r/20241025203106.3529261-2-oliver.upton@linux.dev
Signed-off-by: Oliver Upton <oliver.upton@linux.dev>
arch/arm64/kvm/mmio.c