When a VM is destroyed (bhyvectl --destroy) while a vCPU thread is
handling an exit in bhyve, every call on the VM fails. vm_loop()
handles that for vm_run() and ends bhyve with BHYVE_EXIT_ERROR, but a
vCPU in the middle of an exit hits one of the assert(error == 0) in
emulate_inout() after its register update, or reports "Failed to
emulate instruction sequence" and returns VMEXIT_ABORT: bhyve dies of
SIGABRT and leaves a core. With a guest doing a lot of port I/O
(an IDE or legacy device, a serial console) a forced stop hits this
regularly.

Return an error from emulate_inout() instead of asserting, and on the
failure paths of the I/O and instruction emulation exits check whether
the VM is still there; if it is gone, exit with BHYVE_EXIT_ERROR as
vm_loop() does. A failure on a VM that still exists is reported and
aborts as before.

Signed-off-by: Wanpeng Qian <wanpengqian@gmail.com>
Sponsored by: keelos.dev

Nested test machine, FreeBSD 16.0-CURRENT, bhyve built from main
04a3a6e3a with and without the change. Guest: FreeBSD, 2 vCPUs,
bhyveload, -s 31,lpc -l com1,tcp=...; in the guest two processes read
COM1's LSR (inb(0x3fd)) in a loop through /dev/io. Each round boots the
guest, starts the load, waits 3 s, runs bhyvectl --destroy and records
bhyve's exit status.

without the change: 9 of 10 rounds SIGABRT (status 134),
  "Assertion failed: (error == 0), function emulate_inout,
  file .../amd64/inout.c, line 228."; 1 round exit 4
with the change:    10 of 10 rounds exit 4 (one of them through
  vm_loop()'s "vm_run error -1, errno 6")

Compile-tested on fbsd-head (main 083b44ddc + this commit): usr.sbin/bhyve with -Werror, 0 warnings.