HV: io: refactoring vmexit handler on I/O instruction

This patch refactors how I/O instructions are emulated, in order for a unify the
I/O emulation path. The major control flow includes:

1. pio_instr_vmexit_handler (entry point for handling vmexit on I/O instruction):

    Extract port address, register size, direction and value (for write only),
    fill in an I/O request (of type io_request), invokes do_io to handle that
    and update the guest registers if the request has been successfully handled
    when do_io returns.

2. emulate_io:

    Handle the given I/O request. The request is handled or sent to VHM if it
    returns 0 (the actual status can be found in io_req->processed). On errors a
    negative error code is returned.

3. emulate_pio_by_handler:

    Look for the PIO handler for the given request and invoke that
    handler. Return 0 if a proper handler is found and invoked (the status of
    the emulation can be found in io_req->processed), -EIO when the request
    spans across devices, and -ENODEV when no handler is found.

4. emulate_pio_post:

    Update guest registers after the emulation is done. Currently this can
    happen either right after do_io() or after the vcpu is resumed. Status check
    on the I/O request and follow-up actions on failure will also go here.

Note:

Currently do_io can return 0 with io_req->processed being REQ_STATE_PENDING if
the request is sent to VHM for further processing. In this case the current vcpu
will be paused after handling this vm_exit, and dm_emulate_pio_post will be
invoked to do the rest after this vcpu is resumed. When vcpus are scheduled back
to exactly where they are scheduled out later, do_io should be responsible for
the post_work and the processing of do_io results shall be mostly the same.

v2 -> v3:

    * Rename: emulate_pio_by_handler -> hv_emulate_pio.
    * Properly mask the value passed to port I/O handler.

v1 -> v2:

    * Rename: do_io -> emulate_io.
    * Rename io_instr_vmexit_handler -> pio_instr_vmexit_handler to reflect the
      fact that it handles port I/O only.

Signed-off-by: Junjie Mao <junjie.mao@intel.com>
Acked-by: Eddie Dong <eddie.dong@intel.com>

hypervisor/include/arch/x86/ioreq.h

@@ -108,7 +108,7 @@ struct mem_io_node {
 };
 
 /* External Interfaces */
-int io_instr_vmexit_handler(struct vcpu *vcpu);
+int32_t pio_instr_vmexit_handler(struct vcpu *vcpu);
 void   setup_io_bitmap(struct vm *vm);
 void   free_io_emulation_resource(struct vm *vm);
 void   allow_guest_io_access(struct vm *vm, uint32_t address_arg, uint32_t nbytes);