pckbd: move KBDState from pckbd.c to i8042.h

This allows the QOM types in pckbd.c to be used elsewhere by simply including
i8042.h.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-20-mark.cave-ayland@ilande.co.uk>

pckbd: checkpatch fixes

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-19-mark.cave-ayland@ilande.co.uk>

lasips2: rename ps2dev_update_irq() to lasips2_port_set_irq()

This better reflects that the IRQ input opaque is a LASIPS2Port structure
and not a PS2_DEVICE.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-18-mark.cave-ayland@ilande.co.uk>

lasips2: spacing fixes

This helps improve the readability of lasips2.c.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-17-mark.cave-ayland@ilande.co.uk>

pl050: split pl050_update_irq() into separate pl050_set_irq() and pl050_update_irq() functions

This will soon allow pl050_set_irq() to be used as a GPIO input function.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-16-mark.cave-ayland@ilande.co.uk>

pl050: checkpatch fixes

This patch also includes a couple of minor spacing updates.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-15-mark.cave-ayland@ilande.co.uk>

ps2: don't use vmstate_register() in ps2_mouse_init()

Since PS2_MOUSE_DEVICE is a qdev device then vmstate_ps2_mouse can be registered
using the DeviceClass vmsd field instead. There is no need to use
qdev_set_legacy_instance_id() to ensure migration compatibility since the first 2
parameters to vmstate_register() are NULL and 0 respectively.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-14-mark.cave-ayland@ilande.co.uk>

ps2: don't use vmstate_register() in ps2_kbd_init()

Since PS2_KBD_DEVICE is a qdev device then vmstate_ps2_keyboard can be registered
using the DeviceClass vmsd field instead. There is no need to use
qdev_set_legacy_instance_id() to ensure migration compatibility since the first 2
parameters to vmstate_register() are NULL and 0 respectively.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-13-mark.cave-ayland@ilande.co.uk>

ps2: implement ps2_mouse_realize() and use it to register ps2_mouse_handler

Move the registration of ps2_mouse_handler from ps2_mouse_init() to a new
ps2_mouse_realize() function. Since the abstract PS2_DEVICE parent class doesn't
have a realize() function then it is not necessary to store the reference to
it in PS2DeviceClass and use device_class_set_parent_realize().

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-12-mark.cave-ayland@ilande.co.uk>

ps2: implement ps2_kbd_realize() and use it to register ps2_keyboard_handler

Move the registration of ps2_keyboard_handler from ps2_kbd_init() to a new
ps2_kbd_realize() function. Since the abstract PS2_DEVICE parent class doesn't
have a realize() function then it is not necessary to store the reference to
it in PS2DeviceClass and use device_class_set_parent_realize().

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-11-mark.cave-ayland@ilande.co.uk>

ps2: remove duplicate setting of scancode_set in ps2_kbd_init()

The default value for scancode_set is already set in ps2_kbd_reset() so there is no
need to duplicate this in ps2_kbd_init().

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-10-mark.cave-ayland@ilande.co.uk>

ps2: implement ps2_reset() for the PS2_DEVICE QOM type based upon ps2_common_reset()

The functionality of ps2_common_reset() can be moved into a new ps2_reset() function
for the PS2_DEVICE QOM type. Update PS2DeviceClass to hold a reference to the parent
reset function and update the PS2_KBD_DEVICE and PS2_MOUSE_DEVICE types to use
device_class_set_parent_reset() accordingly.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-9-mark.cave-ayland@ilande.co.uk>

ps2: introduce PS2DeviceClass

This is in preparation for allowing the new PS2_KBD_DEVICE and PS2_MOUSE_DEVICE
QOM types to reference the parent PS2_DEVICE device reset() function.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-8-mark.cave-ayland@ilande.co.uk>

ps2: improve function prototypes in ps2.c and ps2.h

With the latest changes it is now possible to improve some of the function
prototypes in ps2.c and ps.h to use the appropriate PS2KbdState or
PS2MouseState type instead of being a void opaque.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-7-mark.cave-ayland@ilande.co.uk>

ps2: move QOM type definitions from ps2.c to ps2.h

Move the QOM type definitions into the ps2.h header file to allow the new QOM
types to be used by other devices.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-6-mark.cave-ayland@ilande.co.uk>

ps2: QOMify PS2MouseState

Make PS2MouseState into a new PS2_MOUSE_DEVICE QOM type which inherits from the
abstract PS2_DEVICE type.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-5-mark.cave-ayland@ilande.co.uk>

ps2: QOMify PS2KbdState

Make PS2KbdState into a new PS2_KBD_DEVICE QOM type which inherits from the
abstract PS2_DEVICE type.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-4-mark.cave-ayland@ilande.co.uk>

ps2: QOMify PS2State

Make PS2State a new abstract PS2_DEVICE QOM type to represent the common
functionality shared between PS2 keyboard and mouse devices.

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-3-mark.cave-ayland@ilande.co.uk>

ps2: checkpatch fixes

Signed-off-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
Acked-by: Helge Deller <deller@gmx.de>
Message-Id: <20220624134109.881989-2-mark.cave-ayland@ilande.co.uk>

Merge tag 'for-upstream' of git://repo.or.cz/qemu/kevin into staging

Block layer patches

- Add vduse-blk export
- Dirty bitmaps: Fix and improve bitmap merge
- gluster: correctly set max_pdiscard
- rbd: report a better error when namespace does not exist
- aio_wait_kick: add missing memory barrier
- Code cleanups

# -----BEGIN PGP SIGNATURE-----
#
# iQJFBAABCAAvFiEE3D3rFZqa+V09dFb+fwmycsiPL9YFAmK10yURHGt3b2xmQHJl
# ZGhhdC5jb20ACgkQfwmycsiPL9YnXg/+J9PKsGl7hbJUFVIb4RZ3pMQ0o4xC1TbS
# OjA63RRwdJWt1gGsgLNH6kdsgz04FXiXk7XNDTe8MYn21bvznCYGFsI2EFJiit+x
# HkOYXJwAGsrfKBViq8QznosBQbwf9fOrDUlOdbnixFH/dx6DXUZM2F9ud5shiuiC
# vrP76zjUgSlHhPwL+0xsJSNxxNISv845uN9SYNpiGbZDaYLBtNbEc8NA2GMOYW45
# sCbxJDxNSFNrTT5gaZyo4KGFH9393qJSmYHJ+q53elbhje5dpmcn2c/AkIG7/UM4
# HsEWd7Gj1YbLQV6kvkefJzF4BOuZmN/b8sG+LRDjrEMHLeOw5rh2ETmkQfIdy2H3
# 2U52mHqw9bMGWtds0ocOJFKw3vPZuOCJaA+ql/A0aWGO9Fjivx1cQcOVXn3jKgj5
# 7riSxigK9TCTd92Qhi3j4do86mJi9NpjgawbQ5oOvKqUiq29moOwcv1hit0WVzwB
# hJp2qsrM6w00wFseyp+10vHigW7ZMNEIh4TKYLfZ15Gvy6YIG/m/oyCYF4oVM5y5
# knbsnJBRpTjLnggag86Pk8Ji+Lxi0cbv7zrbn++/S9OLQ5VMV5OpxKXHmyX2HyXa
# RCV0rMaTC3s4hy8gElDSwSeor6NBaJ9TECcit67uJ/kw1eK+omoa4hKMCgKf6qYC
# GY+NRPELJGA=
# =OuiN
# -----END PGP SIGNATURE-----
# gpg: Signature made Fri 24 Jun 2022 08:07:17 AM PDT
# gpg:                using RSA key DC3DEB159A9AF95D3D7456FE7F09B272C88F2FD6
# gpg:                issuer "kwolf@redhat.com"
# gpg: Good signature from "Kevin Wolf <kwolf@redhat.com>" [full]

* tag 'for-upstream' of git://repo.or.cz/qemu/kevin:
  vduse-blk: Add name option
  vduse-blk: Add serial option
  nbd: Drop dead code spotted by Coverity
  aio_wait_kick: add missing memory barrier
  block/gluster: correctly set max_pdiscard
  block/rbd: report a better error when namespace does not exist
  qsd: document vduse-blk exports
  libvduse: Add support for reconnecting
  vduse-blk: Add vduse-blk resize support
  vduse-blk: Implement vduse-blk export
  libvduse: Add VDUSE (vDPA Device in Userspace) library
  linux-headers: Add vduse.h
  block/export: Abstract out the logic of virtio-blk I/O process
  block/export: Fix incorrect length passed to vu_queue_push()
  block: Support passing NULL ops to blk_set_dev_ops()
  block: simplify handling of try to merge different sized bitmaps
  block: improve block_dirty_bitmap_merge(): don't allocate extra bitmap
  block: block_dirty_bitmap_merge(): fix error path
  block: get rid of blk->guest_block_size
  block: drop unused bdrv_co_drain() API

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

Merge tag 'linux-user-for-7.1-pull-request' of https://gitlab.com/laurent_vivier/qemu into staging

linux-user pull request 20220624

# -----BEGIN PGP SIGNATURE-----
#
# iQJGBAABCAAwFiEEzS913cjjpNwuT1Fz8ww4vT8vvjwFAmK1iJISHGxhdXJlbnRA
# dml2aWVyLmV1AAoJEPMMOL0/L748NYIP/R11Ztk4XH8LNDszg/s0y+D3FJm3XAkj
# 5LHaJdRGu9KsCO1zVkZDbhm81LIzY8yEzoQKZhV0vy6A071A2QNp8Pu6JZ4fMt6H
# kIwKclwW7WOiI8Ox+cLW4YYPJvY0tbe9yGoHJnLQjjYOOPeoUQKJ/0wK6sBgnK4y
# R8u2x4AvE5hgFfQvI0HRF1Q1e4y5MQt1WmLIxsGl2yCKzKWA0LrPuLgLROLQPbFj
# R/7fNwQkFW0PGnyyd8MMzRTEV6mdNZ9rQDA8/n2Nqgf7uQQFsOpHfXnhoJPz/CGi
# njhixJXingozKb/mX8OA5/NO0Ps7n7/WgugB36QvaoG9M4+E+9vjOkKTFo0LtfeA
# r75++/gNcJhQrMRU6/jSRx25dlWjfuLmEbUVvvGYtgQJ1hsAs6UwXO9EF5n4ZOr8
# qzgbW0GssvFO8+Ow5HQvWOHENR6ylJSSGc0bvX61ZH+h7cJRbELCiO3wkpyGe8zN
# OjG4tHbAnqKhN52k2uannNnPPqXI4f7GMV91pPXqM5oppY0BNMKRC6+nIS6Ew1D3
# FnNDwYJcDBXRY+Abqdrxfx2BSC35cl7Bfl+OVkXal9tGZTmFB+NOaeFWs9fdHXFA
# QE0Ct3xMut6SfJCNLvHwvBHwYq5yDCV/dngGonqgacn+F/epkbdY9sdlsYeRbud2
# h1tremHRgq9e
# =6mTR
# -----END PGP SIGNATURE-----
# gpg: Signature made Fri 24 Jun 2022 02:49:06 AM PDT
# gpg:                using RSA key CD2F75DDC8E3A4DC2E4F5173F30C38BD3F2FBE3C
# gpg:                issuer "laurent@vivier.eu"
# gpg: Good signature from "Laurent Vivier <lvivier@redhat.com>" [undefined]
# gpg:                 aka "Laurent Vivier <laurent@vivier.eu>" [undefined]
# gpg:                 aka "Laurent Vivier (Red Hat) <lvivier@redhat.com>" [undefined]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: CD2F 75DD C8E3 A4DC 2E4F  5173 F30C 38BD 3F2F BE3C

* tag 'linux-user-for-7.1-pull-request' of https://gitlab.com/laurent_vivier/qemu:
  linux-user: Adjust child_tidptr on set_tid_address() syscall
  linux-user: Add partial support for MADV_DONTNEED
  linux-user/x86_64: Fix ELF_PLATFORM

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

vduse-blk: Add name option

Currently we use 'id' option as the name of VDUSE device.
It's a bit confusing since we use one value for two different
purposes: the ID to identfy the export within QEMU (must be
distinct from any other exports in the same QEMU process, but
can overlap with names used by other processes), and the VDUSE
name to uniquely identify it on the host (must be distinct from
other VDUSE devices on the same host, but can overlap with other
export types like NBD in the same process). To make it clear,
this patch adds a separate 'name' option to specify the VDUSE
name for the vduse-blk export instead.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220614051532.92-7-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vduse-blk: Add serial option

Add a 'serial' option to allow user to specify this value
explicitly. And the default value is changed to an empty
string as what we did in "hw/block/virtio-blk.c".

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220614051532.92-6-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

nbd: Drop dead code spotted by Coverity

CID 1488362 points out that the second 'rc >= 0' check is now dead
code.

Reported-by: Peter Maydell <peter.maydell@linaro.org>
Fixes: 172f5f1a40(nbd: remove peppering of nbd_client_connected)
Signed-off-by: Eric Blake <eblake@redhat.com>
Message-Id: <20220516210519.76135-1-eblake@redhat.com>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

aio_wait_kick: add missing memory barrier

It seems that aio_wait_kick always required a memory barrier
or atomic operation in the caller, but nobody actually
took care of doing it.

Let's put the barrier in the function instead, and pair it
with another one in AIO_WAIT_WHILE. Read aio_wait_kick()
comment for further explanation.

Suggested-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Message-Id: <20220524173054.12651-1-eesposit@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/gluster: correctly set max_pdiscard

On 64-bit platforms, assigning SIZE_MAX to the int64_t max_pdiscard
results in a negative value, and the following assertion would trigger
down the line (it's not the same max_pdiscard, but computed from the
other one):
qemu-system-x86_64: ../block/io.c:3166: bdrv_co_pdiscard: Assertion
`max_pdiscard >= bs->bl.request_alignment' failed.

On 32-bit platforms, it's fine to keep using SIZE_MAX.

The assertion in qemu_gluster_co_pdiscard() is checking that the value
of 'bytes' can safely be passed to glfs_discard_async(), which takes a
size_t for the argument in question, so it is kept as is. And since
max_pdiscard is still <= SIZE_MAX, relying on max_pdiscard is still
fine.

Fixes: 0c8022876f ("block: use int64_t instead of int in driver discard handlers")
Cc: qemu-stable@nongnu.org
Signed-off-by: Fabian Ebner <f.ebner@proxmox.com>
Message-Id: <20220520075922.43972-1-f.ebner@proxmox.com>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Reviewed-by: Vladimir Sementsov-Ogievskiy <vsementsov@yandex-team.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/rbd: report a better error when namespace does not exist

If the namespace does not exist, rbd_create() fails with -ENOENT and
QEMU reports a generic "error rbd create: No such file or directory":

    $ qemu-img create rbd:rbd/namespace/image 1M
    Formatting 'rbd:rbd/namespace/image', fmt=raw size=1048576
    qemu-img: rbd:rbd/namespace/image: error rbd create: No such file or directory

Unfortunately rados_ioctx_set_namespace() does not fail if the namespace
does not exist, so let's use rbd_namespace_exists() in qemu_rbd_connect()
to check if the namespace exists, reporting a more understandable error:

    $ qemu-img create rbd:rbd/namespace/image 1M
    Formatting 'rbd:rbd/namespace/image', fmt=raw size=1048576
    qemu-img: rbd:rbd/namespace/image: namespace 'namespace' does not exist

Reported-by: Tingting Mao <timao@redhat.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220517071012.6120-1-sgarzare@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

qsd: document vduse-blk exports

Document vduse-blk exports in qemu-storage-daemon --help and the
qemu-storage-daemon(1) man page.

Based-on: <20220523084611.91-1-xieyongji@bytedance.com>
Cc: Xie Yongji <xieyongji@bytedance.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220525121947.859820-1-stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

libvduse: Add support for reconnecting

To support reconnecting after restart or crash, VDUSE backend
might need to resubmit inflight I/Os. This stores the metadata
such as the index of inflight I/O's descriptors to a shm file so
that VDUSE backend can restore them during reconnecting.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-9-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vduse-blk: Add vduse-blk resize support

To support block resize, this uses vduse_dev_update_config()
to update the capacity field in configuration space and inject
config interrupt on the block resize callback.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-8-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

vduse-blk: Implement vduse-blk export

This implements a VDUSE block backends based on
the libvduse library. We can use it to export the BDSs
for both VM and container (host) usage.

The new command-line syntax is:

$ qemu-storage-daemon \
    --blockdev file,node-name=drive0,filename=test.img \
    --export vduse-blk,node-name=drive0,id=vduse-export0,writable=on

After the qemu-storage-daemon started, we need to use
the "vdpa" command to attach the device to vDPA bus:

$ vdpa dev add name vduse-export0 mgmtdev vduse

Also the device must be removed via the "vdpa" command
before we stop the qemu-storage-daemon.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-7-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

libvduse: Add VDUSE (vDPA Device in Userspace) library

VDUSE [1] is a linux framework that makes it possible to implement
software-emulated vDPA devices in userspace. This adds a library
as a subproject to help implementing VDUSE backends in QEMU.

[1] https://www.kernel.org/doc/html/latest/userspace-api/vduse.html

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-6-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

linux-headers: Add vduse.h

This adds vduse header to linux headers so that the
relevant VDUSE API can be used in subsequent patches.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-5-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/export: Abstract out the logic of virtio-blk I/O process

Abstract the common logic of virtio-blk I/O process to a function
named virtio_blk_process_req(). It's needed for the following commit.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-4-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block/export: Fix incorrect length passed to vu_queue_push()

Now the req->size is set to the correct value only
when handling VIRTIO_BLK_T_GET_ID request. This patch
fixes it.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220523084611.91-3-xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: Support passing NULL ops to blk_set_dev_ops()

This supports passing NULL ops to blk_set_dev_ops()
so that we can remove stale ops in some cases.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220523084611.91-2-xieyongji@bytedance.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: simplify handling of try to merge different sized bitmaps

We have too much logic to simply check that bitmaps are of the same
size. Let's just define that hbitmap_merge() and
bdrv_dirty_bitmap_merge_internal() require their argument bitmaps be of
same size, this simplifies things.

Let's look through the callers:

For backup_init_bcs_bitmap() we already assert that merge can't fail.

In bdrv_reclaim_dirty_bitmap_locked() we gracefully handle the error
that can't happen: successor always has same size as its parent, drop
this logic.

In bdrv_merge_dirty_bitmap() we already has assertion and separate
check. Make the check explicit and improve error message.

Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Reviewed-by: Nikita Lapshin <nikita.lapshin@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220517111206.23585-4-v.sementsov-og@mail.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: improve block_dirty_bitmap_merge(): don't allocate extra bitmap

We don't need extra bitmap. All we need is to backup the original
bitmap when we do first merge. So, drop extra temporary bitmap and work
directly with target and backup.

Still to keep old semantics, that on failure target is unchanged and
user don't need to restore, we need a local_backup variable and do
restore ourselves on failure path.

Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Message-Id: <20220517111206.23585-3-v.sementsov-og@mail.ru>
Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: block_dirty_bitmap_merge(): fix error path

At the end we ignore failure of bdrv_merge_dirty_bitmap() and report
success. And still set errp. That's wrong.

Signed-off-by: Vladimir Sementsov-Ogievskiy <v.sementsov-og@mail.ru>
Reviewed-by: Nikita Lapshin <nikita.lapshin@virtuozzo.com>
Reviewed-by: Kevin Wolf <kwolf@redhat.com>
Message-Id: <20220517111206.23585-2-v.sementsov-og@mail.ru>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: get rid of blk->guest_block_size

Commit 1b7fd729559c ("block: rename buffer_alignment to
guest_block_size") noted:

  At this point, the field is set by the device emulation, but completely
  ignored by the block layer.

The last time the value of buffer_alignment/guest_block_size was
actually used was before commit 339064d50639 ("block: Don't use guest
sector size for qemu_blockalign()").

This value has not been used since 2013. Get rid of it.

Cc: Xie Yongji <xieyongji@bytedance.com>
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220518130945.2657905-1-stefanha@redhat.com>
Reviewed-by: Paul Durrant <paul@xen.org>
Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Alberto Faria <afaria@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

block: drop unused bdrv_co_drain() API

bdrv_co_drain() has not been used since commit 9a0cec664eef ("mirror:
use bdrv_drained_begin/bdrv_drained_end") in 2016. Remove it so there
are fewer drain scenarios to worry about.

Use bdrv_drained_begin()/bdrv_drained_end() instead. They are "mixed"
functions that can be called from coroutine context. Unlike
bdrv_co_drain(), these functions provide control of the length of the
drained section, which is usually the right thing.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-Id: <20220521122714.3837731-1-stefanha@redhat.com>
Reviewed-by: Emanuele Giuseppe Esposito <eesposit@redhat.com>
Reviewed-by: Alberto Faria <afaria@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>

linux-user: Adjust child_tidptr on set_tid_address() syscall

Keep track of the new child tidptr given by a set_tid_address() syscall.

Do not call the host set_tid_address() syscall because we are emulating
the behaviour of writing to child_tidptr in the exit() path.

Signed-off-by: Helge Deller<deller@gmx.de>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <YpH+2sw1PCRqx/te@p100>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>

linux-user: Add partial support for MADV_DONTNEED

Currently QEMU ignores madvise(MADV_DONTNEED), which break apps that
rely on this for zeroing out memory [1]. Improve the situation by doing
a passthrough when the range in question is a host-page-aligned
anonymous mapping.

This is based on the patches from Simon Hausmann [2] and Chris Fallin
[3]. The structure is taken from Simon's patch. The PAGE_MAP_ANONYMOUS
bits are superseded by commit 26bab757d41b ("linux-user: Introduce
PAGE_ANON"). In the end the patch acts like the one from Chris: we
either pass-through the entire syscall, or do nothing, since doing this
only partially would not help the affected applications much. Finally,
add some extra checks to match the behavior of the Linux kernel [4].

[1] https://gitlab.com/qemu-project/qemu/-/issues/326
[2] https://patchew.org/QEMU/20180827084037.25316-1-simon.hausmann@qt.io/
[3] https://github.com/bytecodealliance/wasmtime/blob/v0.37.0/ci/qemu-madvise.patch
[4] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/mm/madvise.c?h=v5.19-rc3#n1368

Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Laurent Vivier <laurent@vivier.eu>
Message-Id: <20220621144205.158452-1-iii@linux.ibm.com>
Signed-off-by: Laurent Vivier <laurent@vivier.eu>

Merge tag 'nvme-next-pull-request' of git://git.infradead.org/qemu-nvme into staging

hw/nvme updates

- sriov functionality
- odd fixes

# -----BEGIN PGP SIGNATURE-----
#
# iQEzBAABCAAdFiEEUigzqnXi3OaiR2bATeGvMW1PDekFAmK02wUACgkQTeGvMW1P
# DenNPwgAwhQCXXacTb+6vEdxN30QoWygzQj5BLm//SiXlj7hBX7P/JqCxYF5vUDU
# EaZkl4n3ry5T1xqlUWIBFdIAmKyrsWz2eKTrX41g64i/L+/nfJXZ+IgQc3WkM/FK
# 5NwwAE8q/JGiRczLesF/9QvQq/90L6QtyC48bsS8AIcl5IcqHCKGwEJS7LErltex
# YZDJyTNU4wB2XFophylJUL43GrHa/kUFA2ZHgs9iuH0p5LGG6UM3KoinBKcbwn47
# iEWKccvsHSyfE8VpJJS5STMEeGGaBPziZ654ElLmzVq6EXDKMCoX03naQ9Q8oSpl
# FiktbxllCYdmECb44PNBEd/nLdpCdQ==
# =o54a
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 23 Jun 2022 02:28:37 PM PDT
# gpg:                using RSA key 522833AA75E2DCE6A24766C04DE1AF316D4F0DE9
# gpg: Good signature from "Klaus Jensen <its@irrelevant.dk>" [unknown]
# gpg:                 aka "Klaus Jensen <k.jensen@samsung.com>" [unknown]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: DDCA 4D9C 9EF9 31CC 3468  4272 63D5 6FC5 E55D A838
#      Subkey fingerprint: 5228 33AA 75E2 DCE6 A247  66C0 4DE1 AF31 6D4F 0DE9

* tag 'nvme-next-pull-request' of git://git.infradead.org/qemu-nvme:
  hw/nvme: clear aen mask on reset
  Revert "hw/block/nvme: add support for sgl bit bucket descriptor"
  hw/nvme: clean up CC register write logic
  hw/acpi: Make the PCI hot-plug aware of SR-IOV
  hw/nvme: Update the initalization place for the AER queue
  docs: Add documentation for SR-IOV and Virtualization Enhancements
  hw/nvme: Add support for the Virtualization Management command
  hw/nvme: Initialize capability structures for primary/secondary controllers
  hw/nvme: Calculate BAR attributes in a function
  hw/nvme: Remove reg_size variable and update BAR0 size calculation
  hw/nvme: Make max_ioqpairs and msix_qsize configurable in runtime
  hw/nvme: Implement the Function Level Reset
  hw/nvme: Add support for Secondary Controller List
  hw/nvme: Add support for Primary Controller Capabilities
  hw/nvme: Add support for SR-IOV

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

hw/nvme: clear aen mask on reset

The internally maintained AEN mask is not cleared on reset. Fix this.

Reviewed-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

Revert "hw/block/nvme: add support for sgl bit bucket descriptor"

This reverts commit d97eee64fef35655bd06f5c44a07fdb83a6274ae.

The emulated controller correctly accounts for not including bit buckets
in the controller-to-host data transfer, however it doesn't correctly
account for the holes for the on-disk data offsets.

Reported-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: clean up CC register write logic

The SRIOV series exposed an issued with how CC register writes are
handled and how CSTS is set in response to that. Specifically, after
applying the SRIOV series, the controller could end up in a state with
CC.EN set to '1' but with CSTS.RDY cleared to '0', causing drivers to
expect CSTS.RDY to transition to '1' but timing out.

Clean this up.

Reviewed-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/acpi: Make the PCI hot-plug aware of SR-IOV

PCI device capable of SR-IOV support is a new, still-experimental
feature with only a single working example of the Nvme device.

This patch in an attempt to fix a double-free problem when a
SR-IOV-capable Nvme device is hot-unplugged in the following scenario:

Qemu CLI:
---------
-device pcie-root-port,slot=0,id=rp0
-device nvme-subsys,id=subsys0
-device nvme,id=nvme0,bus=rp0,serial=deadbeef,subsys=subsys0,sriov_max_vfs=1,sriov_vq_flexible=2,sriov_vi_flexible=1

Guest OS:
---------
sudo nvme virt-mgmt /dev/nvme0 -c 0 -r 1 -a 1 -n 0
sudo nvme virt-mgmt /dev/nvme0 -c 0 -r 0 -a 1 -n 0
echo 1 > /sys/bus/pci/devices/0000:01:00.0/reset
sleep 1
echo 1 > /sys/bus/pci/devices/0000:01:00.0/sriov_numvfs
nvme virt-mgmt /dev/nvme0 -c 1 -r 1 -a 8 -n 1
nvme virt-mgmt /dev/nvme0 -c 1 -r 0 -a 8 -n 2
nvme virt-mgmt /dev/nvme0 -c 1 -r 0 -a 9 -n 0
sleep 2
echo 01:00.1 > /sys/bus/pci/drivers/nvme/bind

Qemu monitor:
-------------
device_del nvme0

Explanation of the problem and the proposed solution:

1) The current SR-IOV implementation assumes it’s the PhysicalFunction
   that creates and deletes VirtualFunctions.
2) It’s a design decision (the Nvme device at least) for the VFs to be
   of the same class as PF. Effectively, they share the dc->hotpluggable
   value.
3) When a VF is created, it’s added as a child node to PF’s PCI bus
   slot.
4) Monitor/device_del triggers the ACPI mechanism. The implementation is
   not aware of SR/IOV and ejects PF’s PCI slot, directly unrealizing all
   hot-pluggable (!acpi_pcihp_pc_no_hotplug) children nodes.
5) VFs are unrealized directly, and it doesn’t work well with (1).
   SR/IOV structures are not updated, so when it’s PF’s turn to be
   unrealized, it works on stale pointers to already-deleted VFs.

The proposed fix is to make the PCI ACPI code aware of SR/IOV.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Update the initalization place for the AER queue

This patch updates the initialization place for the AER queue, so it’s
initialized once, at controller initialization, and not every time
controller is enabled.

While the original version works for a non-SR-IOV device, as it’s hard
to interact with the controller if it’s not enabled, the multiple
reinitialization is not necessarily correct.

With the SR/IOV feature enabled a segfault can happen: a VF can have its
controller disabled, while a namespace can still be attached to the
controller through the parent PF. An event generated in such case ends
up on an uninitialized queue.

While it’s an interesting question whether a VF should support AER in
the first place, I don’t think it must be answered today.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

docs: Add documentation for SR-IOV and Virtualization Enhancements

Documentation describes 5 new parameters being added regarding SR-IOV:
sriov_max_vfs
sriov_vq_flexible
sriov_vi_flexible
sriov_max_vi_per_vf
sriov_max_vq_per_vf

The description also includes the simplest possible QEMU invocation
and the series of NVMe commands required to enable SR-IOV support.

Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Add support for the Virtualization Management command

With the new command one can:
 - assign flexible resources (queues, interrupts) to primary and
   secondary controllers,
 - toggle the online/offline state of given controller.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Initialize capability structures for primary/secondary controllers

With four new properties:
 - sriov_v{i,q}_flexible,
 - sriov_max_v{i,q}_per_vf,
one can configure the number of available flexible resources, as well as
the limits. The primary and secondary controller capability structures
are initialized accordingly.

Since the number of available queues (interrupts) now varies between
VF/PF, BAR size calculation is also adjusted.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Calculate BAR attributes in a function

An NVMe device with SR-IOV capability calculates the BAR size
differently for PF and VF, so it makes sense to extract the common code
to a separate function.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Remove reg_size variable and update BAR0 size calculation

The n->reg_size parameter unnecessarily splits the BAR0 size calculation
in two phases; removed to simplify the code.

With all the calculations done in one place, it seems the pow2ceil,
applied originally to reg_size, is unnecessary. The rounding should
happen as the last step, when BAR size includes Nvme registers, queue
registers, and MSIX-related space.

Finally, the size of the mmio memory region is extended to cover the 1st
4KiB padding (see the map below). Access to this range is handled as
interaction with a non-existing queue and generates an error trace, so
actually nothing changes, while the reg_size variable is no longer needed.

    --------------------
    |      BAR0        |
    --------------------
    [Nvme Registers    ]
    [Queues            ]
    [power-of-2 padding] - removed in this patch
    [4KiB padding (1)  ]
    [MSIX TABLE        ]
    [4KiB padding (2)  ]
    [MSIX PBA          ]
    [power-of-2 padding]

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Make max_ioqpairs and msix_qsize configurable in runtime

The NVMe device defines two properties: max_ioqpairs, msix_qsize. Having
them as constants is problematic for SR-IOV support.

SR-IOV introduces virtual resources (queues, interrupts) that can be
assigned to PF and its dependent VFs. Each device, following a reset,
should work with the configured number of queues. A single constant is
no longer sufficient to hold the whole state.

This patch tries to solve the problem by introducing additional
variables in NvmeCtrl’s state. The variables for, e.g., managing queues
are therefore organized as:
 - n->params.max_ioqpairs – no changes, constant set by the user
 - n->(mutable_state) – (not a part of this patch) user-configurable,
                        specifies number of queues available _after_
                        reset
 - n->conf_ioqpairs - (new) used in all the places instead of the ‘old’
                      n->params.max_ioqpairs; initialized in realize()
                      and updated during reset() to reflect user’s
                      changes to the mutable state

Since the number of available i/o queues and interrupts can change in
runtime, buffers for sq/cqs and the MSIX-related structures are
allocated big enough to handle the limits, to completely avoid the
complicated reallocation. A helper function (nvme_update_msixcap_ts)
updates the corresponding capability register, to signal configuration
changes.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Implement the Function Level Reset

This patch implements the Function Level Reset, a feature currently not
implemented for the Nvme device, while listed as a mandatory ("shall")
in the 1.4 spec.

The implementation reuses FLR-related building blocks defined for the
pci-bridge module, and follows the same logic:
    - FLR capability is advertised in the PCIE config,
    - custom pci_write_config callback detects a write to the trigger
      register and performs the PCI reset,
    - which, eventually, calls the custom dc->reset handler.

Depending on reset type, parts of the state should (or should not) be
cleared. To distinguish the type of reset, an additional parameter is
passed to the reset function.

This patch also enables advertisement of the Power Management PCI
capability. The main reason behind it is to announce the no_soft_reset=1
bit, to signal SR-IOV support where each VF can be reset individually.

The implementation purposedly ignores writes to the PMCS.PS register,
as even such naïve behavior is enough to correctly handle the D3->D0
transition.

It’s worth to note, that the power state transition back to to D3, with
all the corresponding side effects, wasn't and stil isn't handled
properly.

Signed-off-by: Łukasz Gieryk <lukasz.gieryk@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Add support for Secondary Controller List

Introduce handling for Secondary Controller List (Identify command with
CNS value of 15h).

Secondary controller ids are unique in the subsystem, hence they are
reserved by it upon initialization of the primary controller to the
number of sriov_max_vfs.

ID reservation requires the addition of an intermediate controller slot
state, so the reserved controller has the address 0xFFFF.
A secondary controller is in the reserved state when it has no virtual
function assigned, but its primary controller is realized.
Secondary controller reservations are released to NULL when its primary
controller is unregistered.

Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Add support for Primary Controller Capabilities

Implementation of Primary Controller Capabilities data
structure (Identify command with CNS value of 14h).

Currently, the command returns only ID of a primary controller.
Handling of remaining fields are added in subsequent patches
implementing virtualization enhancements.

Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

hw/nvme: Add support for SR-IOV

This patch implements initial support for Single Root I/O Virtualization
on an NVMe device.

Essentially, it allows to define the maximum number of virtual functions
supported by the NVMe controller via sriov_max_vfs parameter.

Passing a non-zero value to sriov_max_vfs triggers reporting of SR-IOV
capability by a physical controller and ARI capability by both the
physical and virtual function devices.

NVMe controllers created via virtual functions mirror functionally
the physical controller, which may not entirely be the case, thus
consideration would be needed on the way to limit the capabilities of
the VF.

NVMe subsystem is required for the use of SR-IOV.

Signed-off-by: Lukasz Maniak <lukasz.maniak@linux.intel.com>
Reviewed-by: Klaus Jensen <k.jensen@samsung.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>

Merge tag 'pull-migration-20220623b' of https://gitlab.com/dagrh/qemu into staging

[v3] Migration pull 2022-06-23

This replaces my and Juan's earlier pulls over the last 2 days;
4th time lucky?

Compared to my pull earlier:
  Removed Hyman's dirty ring set

In this migration PULL request:
- Dainiel Berrangé - qemufileops cleanup
- Leonardo Bras  - cleanups for zero copy
- Juan Quintela  - RDMA cleanups

Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
# -----BEGIN PGP SIGNATURE-----
#
# iQIzBAABCAAdFiEERfXHG0oMt/uXep+pBRYzHrxb/ecFAmK0mnwACgkQBRYzHrxb
# /ecJYQ/8DdQBVYpJjpHj3mBx39aXodM7uM4Jt1okuSF92y9KRSNdIs3nvvwWAZbL
# dWAUHUZBNOfZF7Eqe6WWCIDNxUTz88RkMM16N3+a5sqBa0xU6rP6cvyw9vYrnsmx
# aHVQ1ESosTby2qcb1ofjYRXWNt7GhDtRIH55m3mSalP/WAgjMe3MsrAtz66T4w55
# 4paTVwgS/WMuLD9dqyESyofePnvtM8z3ye2a75JRscBQYmpO+XuX3IX5ah6m439s
# fI1BezQU2Q4YNDmCEWvdfZ2tqgcfi8dHnu0JB9QTfbkPVh9jw25VPpnymimMB7iW
# MlXAlDr7m9HQI6OjIkq8pXBcgWQpbVGMon1CcrDmGCReEjnQ5lTsb27fkXzf/Nwu
# 09iuNfYGcSGAbZ8GZa/lrRTGeINrSj99uOVxrTvVS0db2+1va3hkamGMULhsdX6O
# smOrje79pVLAr7JJSMH2bqFv9cKtLu77HndSVtswkRRMhtDU+VQI5FxYlwueuawJ
# toDM4DJMd3pJHIpPrUwxlo4D9dkdxPfqC1GATDPxw9/vYgbORn8fkt5g9EYxBzc0
# pWRY9SNuw0MC54JGEoFc77+VKJXK1A97j9GoF+Vyoh30yTgZ3q9tm2eElpYwtHDy
# t8zEVC9QadcgMdRAnJqgZgaWdfwKiHpjplSn5lOGDLOo7gfSmik=
# =ajVU
# -----END PGP SIGNATURE-----
# gpg: Signature made Thu 23 Jun 2022 09:53:16 AM PDT
# gpg:                using RSA key 45F5C71B4A0CB7FB977A9FA90516331EBC5BFDE7
# gpg: Good signature from "Dr. David Alan Gilbert (RH2) <dgilbert@redhat.com>" [full]

* tag 'pull-migration-20220623b' of https://gitlab.com/dagrh/qemu: (25 commits)
  migration: remove the QEMUFileOps abstraction
  migration: remove the QEMUFileOps 'get_return_path' callback
  migration: remove the QEMUFileOps 'writev_buffer' callback
  migration: remove the QEMUFileOps 'get_buffer' callback
  migration: remove the QEMUFileOps 'close' callback
  migration: remove the QEMUFileOps 'set_blocking' callback
  migration: remove the QEMUFileOps 'shut_down' callback
  migration: remove unused QEMUFileGetFD typedef / qemu_get_fd method
  migration: introduce new constructors for QEMUFile
  migration: hardcode assumption that QEMUFile is backed with QIOChannel
  migration: stop passing 'opaque' parameter to QEMUFile hooks
  migration: convert savevm to use QIOChannelBlock for VMState
  migration: introduce a QIOChannel impl for BlockDriverState VMState
  migration: rename qemu_file_update_transfer to qemu_file_acct_rate_limit
  migration: rename qemu_update_position to qemu_file_credit_transfer
  migration: rename qemu_ftell to qemu_file_total_transferred
  migration: rename 'pos' field in QEMUFile to 'bytes_processed'
  migration: rename rate limiting fields in QEMUFile
  migration: remove unreachble RDMA code in save_hook impl
  migration: switch to use QIOChannelNull for dummy channel
  ...

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

migration: remove the QEMUFileOps abstraction

Now that all QEMUFile callbacks are removed, the entire concept can be
deleted.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove the QEMUFileOps 'get_return_path' callback

This directly implements the get_return_path logic using QIOChannel APIs.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove the QEMUFileOps 'writev_buffer' callback

This directly implements the writev_buffer logic using QIOChannel APIs.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove the QEMUFileOps 'get_buffer' callback

This directly implements the get_buffer logic using QIOChannel APIs.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
  dgilbert: Fixup len = *-*EIO as spotted by Peter Xu

migration: remove the QEMUFileOps 'close' callback

This directly implements the close logic using QIOChannel APIs.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove the QEMUFileOps 'set_blocking' callback

This directly implements the set_blocking logic using QIOChannel APIs.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove the QEMUFileOps 'shut_down' callback

This directly implements the shutdown logic using QIOChannel APIs.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove unused QEMUFileGetFD typedef / qemu_get_fd method

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: introduce new constructors for QEMUFile

Prepare for the elimination of QEMUFileOps by introducing a pair of new
constructors. This lets us distinguish between an input and output file
object explicitly rather than via the existance of specific callbacks.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: hardcode assumption that QEMUFile is backed with QIOChannel

The only callers of qemu_fopen_ops pass 'true' for the 'has_ioc'
parameter, so hardcode this assumption in QEMUFile, by passing in
the QIOChannel object as a non-opaque parameter.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
   dgilbert: Fixed long line

migration: stop passing 'opaque' parameter to QEMUFile hooks

The only user of the hooks is RDMA which provides a QIOChannel backed
impl of QEMUFile. It can thus use the qemu_file_get_ioc() method.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: convert savevm to use QIOChannelBlock for VMState

With this change, all QEMUFile usage is backed by QIOChannel at
last.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
  dgilbert: Wrap long lines

migration: introduce a QIOChannel impl for BlockDriverState VMState

Introduce a QIOChannelBlock class that exposes the BlockDriverState
VMState region for I/O.

This is kept in the migration/ directory rather than io/, to avoid
a mutual dependancy between block/ <-> io/ directories. Also the
VMState should only be used by the migration code.

Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
  dgilbert: Fixed coding style in qio_channel_block_close

migration: rename qemu_file_update_transfer to qemu_file_acct_rate_limit

The qemu_file_update_transfer name doesn't give a clear guide on what
its purpose is, and how it differs from the qemu_file_credit_transfer
method. The latter is specifically for accumulating for total migration
traffic, while the former is specifically for accounting in thue rate
limit calculations. The new name give better guidance on its usage.

Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: rename qemu_update_position to qemu_file_credit_transfer

The qemu_update_position method name gives the misleading impression
that it is changing the current file offset. Most of the files are
just streams, however, so there's no concept of a file offset in the
general case.

What this method is actually used for is to report on the number of
bytes that have been transferred out of band from the main I/O methods.
This new name better reflects this purpose.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: rename qemu_ftell to qemu_file_total_transferred

The name 'ftell' gives the misleading impression that the QEMUFile
objects are seekable. This is not the case, as in general we just
have an opaque stream. The users of this method are only interested
in the total bytes processed. This switches to a new name that
reflects the intended usage.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
   dgilbert: Wrapped long line

migration: rename 'pos' field in QEMUFile to 'bytes_processed'

The field name 'pos' gives the misleading impression that the QEMUFile
objects are seekable. This is not the case, as in general we just
have an opaque stream. The users of this method are only interested
in the total bytes processed. This switches to a new name that
reflects the intended usage.

Every QIOChannel backed impl of QEMUFile is currently ignoring the
'pos' field.

The only QEMUFile impl using 'pos' as an offset for I/O is the block
device vmstate. A later patch is introducing a QIOChannel impl for the
vmstate, and to handle this it is tracking a file offset itself
internally to the QIOChannel impl. So when we later eliminate the
QEMUFileOps callbacks later, the 'pos' field will no longer be used
from any I/O read/write methods.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
  dgilbert: Fixed long line

migration: rename rate limiting fields in QEMUFile

This renames the following QEMUFile fields

 * bytes_xfer -> rate_limit_used
 * xfer_limit -> rate_limit_max

The intent is to make it clear that 'bytes_xfer' is specifically related
to rate limiting of data and applies to data queued, which need not have
been transferred on the wire yet if a flush hasn't taken place.

Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: remove unreachble RDMA code in save_hook impl

The QEMUFile 'save_hook' callback has a 'size_t size' parameter.

The RDMA impl of this has logic that takes different actions
depending on whether the value is zero or non-zero. It has
commented out logic that would have taken further actions
if the value was negative.

The only place where the 'save_hook' callback is invoked is
the ram_control_save_page() method, which passes 'size'
through from its caller. The only caller of this method is
in turn control_save_page(). This method unconditionally
passes the 'TARGET_PAGE_SIZE' constant for the 'size' parameter.

IOW, the only scenario for 'size' that can execute in the
qemu_rdma_save_page method is 'size > 0'. The remaining code
has been unreachable since RDMA support was first introduced
9 years ago.

Reviewed-by: Eric Blake <eblake@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: switch to use QIOChannelNull for dummy channel

This removes one further custom impl of QEMUFile, in favour of a
QIOChannel based impl.

Reviewed-by: Eric Blake <eblake@redhat.com>
Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

io: add a QIOChannelNull equivalent to /dev/null

This is for code which needs a portable equivalent to a QIOChannelFile
connected to /dev/null.

Signed-off-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

migration: Change zero_copy_send from migration parameter to migration capability

When originally implemented, zero_copy_send was designed as a Migration
paramenter.

But taking into account how is that supposed to work, and how
the difference between a capability and a parameter, it only makes sense
that zero-copy-send would work better as a capability.

Taking into account how recently the change got merged, it was decided
that it's still time to make it right, and convert zero_copy_send into
a Migration capability.

Signed-off-by: Leonardo Bras <leobras@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Acked-by: Markus Armbruster <armbru@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
  dgilbert: always define the capability, even on non-Linux but error if
set; avoids build problems with the capability

QIOChannelSocket: Fix zero-copy send so socket flush works

Somewhere between v6 and v7 the of the zero-copy-send patchset a crucial
part of the flushing mechanism got missing: incrementing zero_copy_queued.

Without that, the flushing interface becomes a no-op, and there is no
guarantee the buffer is really sent.

This can go as bad as causing a corruption in RAM during migration.

Fixes: 2bc58ffc2926 ("QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX")
Reported-by: 徐闯 <xuchuangxclwt@bytedance.com>
Signed-off-by: Leonardo Bras <leobras@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

QIOChannelSocket: Introduce assert and reduce ifdefs to improve readability

During implementation of MSG_ZEROCOPY feature, a lot of #ifdefs were
introduced, particularly at qio_channel_socket_writev().

Rewrite some of those changes so it's easier to read.

Also, introduce an assert to help detect incorrect zero-copy usage is when
it's disabled on build.

Signed-off-by: Leonardo Bras <leobras@redhat.com>
Reviewed-by: Daniel P. Berrangé <berrange@redhat.com>
Reviewed-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Peter Xu <peterx@redhat.com>
Signed-off-by: Juan Quintela <quintela@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
  dgilbert: Fixed up thinko'd g_assert_unreachable->g_assert_not_reached

migration: Remove RDMA_UNREGISTRATION_EXAMPLE

Nobody has ever showed up to unregister individual pages, and another
set of patches written by Daniel P. Berrangé <berrange@redhat.com>
just remove qemu_rdma_signal_unregister() function needed here.

Signed-off-by: Juan Quintela <quintela@redhat.com>
Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com>
Signed-off-by: Dr. David Alan Gilbert <dgilbert@redhat.com>

Merge tag 'pull-aspeed-20220622' of https://github.com/legoater/qemu into staging

aspeed queue:

* Extra avocado tests using buildroot images
* Conversion of the I2C model to the registerfield interface
* Support for the I2C new register interface on AST2600
* Various I2C enhancements
* I2C support for the AST1030
* Improvement of the Aspeed SMC and m25p80 qtest

# -----BEGIN PGP SIGNATURE-----
#
# iQIzBAABCAAdFiEEoPZlSPBIlev+awtgUaNDx8/77KEFAmKyzCYACgkQUaNDx8/7
# 7KG+Aw/+MM3BlQfzDhjETkznqFbvp+aDcnKYwd/brizUC1y3paPFGc0xzD05x3QC
# 2th44oYS934UwQ78EMkC0uNed/kHh+6aHaBrq/XylIg7Dbq5QeCBXwRGCNW6tgtc
# K3ZSM20QM+XRCjmo9ys792NYPC+8tYpw7idb2AOeum7ic/ZaeT3h1FX1Mr57I3XE
# PYwDEBEd4hJ3DroYzIP9YQvRBNu8/d5VoiNr3GLfNy3zrkhuJ4D4jUAEbHATG7Gb
# k0A6o6bVAL85AUSq/ksceHqzWAnizh1q1o/k9UP83HIt1S3ghgK6RsAu4+9HKlP4
# lZ6MFfx3Nzf8u2y/FlOiuABEBUNsngjNmLo6B/qe/cZk60/nS56qOWSvuzPxqVDO
# lI++SLY6R1D8q36H4eF/vq/AyLnXBxGqeq0DipPcnZVKdVVHUHppNly5efJ/7cWn
# VybobblU48BCgjc/EoMVEy8L/t/uRjY3wmoKkfKLCObRrlcPxSrLPUP8+j8nR0JG
# zDOh+CrxHTUbJGV6qRmZx9m2HQtbtH5k89UxskkUkscvVDqWhxqdFVnTWfXcmyP8
# LqTkEv7IV4ECM1zN5OVK9No46WCi5j24bxO3z7or4e04vgwjM41unV7HAFl8Z0/s
# tyFQUG4dFAKHH7quU0F3qSxnORNyCy5ssEpmobujeifbiFMpNss=
# =OsUe
# -----END PGP SIGNATURE-----
# gpg: Signature made Wed 22 Jun 2022 01:00:38 AM PDT
# gpg:                using RSA key A0F66548F04895EBFE6B0B6051A343C7CFFBECA1
# gpg: Good signature from "Cédric Le Goater <clg@kaod.org>" [undefined]
# gpg: WARNING: This key is not certified with a trusted signature!
# gpg:          There is no indication that the signature belongs to the owner.
# Primary key fingerprint: A0F6 6548 F048 95EB FE6B  0B60 51A3 43C7 CFFB ECA1

* tag 'pull-aspeed-20220622' of https://github.com/legoater/qemu:
  hw: m25p80: fixing individual test failure when tests are running in isolation
  aspeed/hace: Add missing newlines to unimp messages
  aspeed/i2c: Enable SLAVE_ADDR_RX_MATCH always
  hw/i2c/aspeed: add DEV_ADDR in old register mode
  hw/i2c/aspeed: rework raise interrupt trace event
  aspeed: Add I2C buses to AST1030 model
  aspeed/i2c: Add ast1030 controller models
  aspeed: i2c: Move regs and helpers to header file
  aspeed: i2c: Add PKT_DONE IRQ to trace
  aspeed: i2c: Add new mode support
  aspeed: i2c: Use reg array instead of individual vars
  aspeed: i2c: Migrate to registerfields API
  hw/registerfields: Add shared fields macros
  test/avocado/machine_aspeed.py: Add an I2C RTC test
  test/avocado/machine_aspeed.py: Add I2C tests to ast2600-evb
  test/avocado/machine_aspeed.py: Add I2C tests to ast2500-evb
  test/avocado/machine_aspeed.py: Add tests using buildroot images
  test/avocado/machine_aspeed.py: Move OpenBMC tests
  aspeed: Remove fake RTC device on ast2500-evb

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>

hw: m25p80: fixing individual test failure when tests are running in isolation

Signed-off-by: Iris Chen <irischenlj@fb.com>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed/hace: Add missing newlines to unimp messages

Signed-off-by: Joel Stanley <joel@jms.id.au>
Reviewed-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed/i2c: Enable SLAVE_ADDR_RX_MATCH always

There is no 'slave match interrupt' enable bit in the Interrupt
Control Register. Consider it is always enabled and extend the mask
value 'bus->regs[intr_ctrl_reg]' with the SLAVE_ADDR_RX_MATCH bit when
the interrupt is raised.

Signed-off-by: Cédric Le Goater <clg@kaod.org>

hw/i2c/aspeed: add DEV_ADDR in old register mode

Add support for writing and reading the device address register in old
register mode.

On the AST2400 (only 1 slave address)

  * no upper bits

On the AST2500 (2 possible slave addresses),

  * bit[31] : Slave Address match indicator
  * bit[30] : Slave Address Receiving pending

On the AST2600 (3 possible slave addresses),

  * bit[31-30] : Slave Address match indicator
  * bit[29] : Slave Address Receiving pending

The model could be more precise to take into account all fields but
since the Linux driver is masking the register value being set, it
should be fine. See commit 3fb2e2aeafb2 ("i2c: aspeed: disable
additional device addresses on ast2[56]xx") from Zeiv. This can be
addressed later.

Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
[ clg: add details to commit log ]
Message-Id: <20220601210831.67259-3-its@irrelevant.dk>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

hw/i2c/aspeed: rework raise interrupt trace event

Build a single string instead of having several parameters on the trace
event.

Suggested-by: Cédric Le Goater <clg@kaod.org>
Signed-off-by: Klaus Jensen <k.jensen@samsung.com>
[ clg: simplified trace buffer creation ]
Message-Id: <20220601210831.67259-2-its@irrelevant.dk>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed: Add I2C buses to AST1030 model

Instantiate the I2C buses in AST1030 model and create two slave device
for ast1030-evb.

Signed-off-by: Troy Lee <troy_lee@aspeedtech.com>
Signed-off-by: Jamin Lin <jamin_lin@aspeedtech.com>
Signed-off-by: Steven Lee <steven_lee@aspeedtech.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
[ clg : - adapted to current AST1030 upstream models
        - changed AST2600 to AST1030 in comment
        - fixed typo in commit log ]
Message-Id: <20220324100439.478317-3-troy_lee@aspeedtech.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed/i2c: Add ast1030 controller models

Based on :

  https://lore.kernel.org/qemu-devel/20220324100439.478317-2-troy_lee@aspeedtech.com/

Cc: Troy Lee <troy_lee@aspeedtech.com>
Cc: Jamin Lin <jamin_lin@aspeedtech.com>
Cc: Steven Lee <steven_lee@aspeedtech.com>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed: i2c: Move regs and helpers to header file

Moves register definitions and short commonly used inlined functiosn to
the header file to help tidy up the implementation file.

Signed-off-by: Joe Komlodi <komlodi@google.com>
Change-Id: I34dff7485b6bbe3c9482715ccd94dbd65dc5f324
Message-Id: <20220331043248.2237838-8-komlodi@google.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed: i2c: Add PKT_DONE IRQ to trace

Signed-off-by: Joe Komlodi <komlodi@google.com>
Change-Id: I566eb09f4b9016e24570572f367627f6594039f5
Message-Id: <20220331043248.2237838-7-komlodi@google.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed: i2c: Add new mode support

On AST2600, I2C has a secondary mode, called "new mode", which changes
the layout of registers, adds some minor behavior changes, and
introduces a new way to transfer data called "packet mode".

Most of the bit positions of the fields are the same between old and new
mode, so we use SHARED_FIELD_XX macros to reuse most of the code between
the different modes.

For packet mode, most of the command behavior is the same compared to
other modes, but there are some minor changes to how interrupts are
handled compared to other modes.

Signed-off-by: Joe Komlodi <komlodi@google.com>
Change-Id: I072f8301964f623afc74af1fe50c12e5caef199e
Message-Id: <20220331043248.2237838-6-komlodi@google.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed: i2c: Use reg array instead of individual vars

Using a register array will allow us to represent old-mode and new-mode
I2C registers by using the same underlying register array, instead of
adding an entire new set of variables to represent new mode.

As part of this, we also do additional cleanup to use ARRAY_FIELD_
macros instead of FIELD_ macros on registers.

Signed-off-by: Joe Komlodi <komlodi@google.com>
Change-Id: Ib94996b17c361b8490c042b43c99d8abc69332e3
[ clg: use of memset in aspeed_i2c_bus_reset() ]
Message-Id: <20220331043248.2237838-5-komlodi@google.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

aspeed: i2c: Migrate to registerfields API

This cleans up some of the field accessing, setting, and clearing
bitwise operations, and wraps them in macros instead.

Signed-off-by: Joe Komlodi <komlodi@google.com>
Change-Id: I33018d6325fa04376e7c29dc4a49ab389a8e333a
Message-Id: <20220331043248.2237838-4-komlodi@google.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

hw/registerfields: Add shared fields macros

Occasionally a peripheral will have different operating modes, where the
MMIO layout changes, but some of the register fields have the same offsets
and behaviors.

To help support this, we add SHARED_FIELD_XX macros that create SHIFT,
LENGTH, and MASK macros for the fields that are shared across registers,
and accessors for these fields.

An example use may look as follows:
There is a peripheral with registers REG_MODE1 and REG_MODE2 at
different addreses, and both have a field FIELD1 initialized by
SHARED_FIELD().

Depending on what mode the peripheral is operating in, the user could
extract FIELD1 via
SHARED_ARRAY_FIELD_EX32(s->regs, R_REG_MODE1, FIELD1)
or
SHARED_ARRAY_FIELD_EX32(s->regs, R_REG_MODE2, FIELD1)

Signed-off-by: Joe Komlodi <komlodi@google.com>
Change-Id: Id3dc53e7d2f8741c95697cbae69a81bb699fa3cb
Message-Id: <20220331043248.2237838-2-komlodi@google.com>
Signed-off-by: Cédric Le Goater <clg@kaod.org>

test/avocado/machine_aspeed.py: Add an I2C RTC test

Add an RTC device and check that the output of the hwclock command
matches the current year.

Signed-off-by: Cédric Le Goater <clg@kaod.org>
Reviewed-by: Joel Stanley <joel@jms.id.au>
Signed-off-by: Cédric Le Goater <clg@kaod.org>