Copy stable/9 to releng/9.3 as part of the 9.3-RELEASE cycle.

Approved by: re (implicit)
Sponsored by: The FreeBSD Foundation

MFC r255111, r255112, r255131, r255167, r256066, r261150, r261151,
r261164, r261166, r261167, r261168, r261394, r261395:

This updates the network driver to support multiple queues, and several bug
fixes. Note that multiqueue support is not compiled in by default since that
would change ALTQ behavior.

- Sync VirtIO net device header file from recent Linux
- Import multiqueue VirtIO net driver
- Fix build with gcc
- Do not hold the vtnet Rx queue lock when calling up into the stack
- Read and write the MAC address in the config space byte by byte
- Also include the mbuf's csum_flags in an assert message
- Remove stray space
- Move duplicated transmit start code into a single function
- Avoid queue unlock followed by relock when the enable interrupt race is
lost
- Check for a full virtqueue in the multiqueue transmit path
- Do not place the sglist used for Rx/Tx on the stack
- Use m_defrag() instead of m_collapse() to compact a long mbuf chain

MFC VirtIO r252702,r252703,r252704,r252706,r252707,r252708,r252709

This brings in several bug fixes and better conformance with various
aspects of the spec that have changed.

r252702: Convert VirtIO to use ithreads instead of taskqueues
r252703: Block driver bug fixes and spec changes
r252704: Balloon driver improvements
r252706: Minor network driver improvements
r252707: Minor misc VirtIO changes
r252708: PCI driver bug fixes and cleanup
r252709: Fix SCSI driver lock not owned panic

MFC r227652, r228301, r234270, r234349

Sync Bryan Venteicher's virtio base, PCI front-end, and net/block/balloon
drivers from HEAD.

Import virtio base, PCI front-end, and net/block/balloon drivers.
Tested on Qemu/KVM, VirtualBox, and BHyVe.

Currently built as modules-only on i386/amd64. Man pages not yet hooked
up, pending review.

Submitted by: Bryan Venteicher bryanv at daemoninthecloset dot org
Reviewed by: bz
MFC after: 4 weeks or so