vhost-vdpa: uAPI to suspend the device

The ioctl adds support for suspending the device from userspace.

This is a must before getting virtqueue indexes (base) for live migration,
since the device could modify them after userland gets them. There are
individual ways to perform that action for some devices
(VHOST_NET_SET_BACKEND, VHOST_VSOCK_SET_RUNNING, ...) but there was no
way to perform it for any vhost device (and, in particular, vhost-vdpa).

After a successful return of the ioctl call the device must not process
more virtqueue descriptors. The device can answer to read or writes of
config fields as if it were not suspended. In particular, writing to
"queue_enable" with a value of 1 will not make the device start
processing buffers of the virtqueue.

Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Message-Id: <20220810171512.2343333-4-eperezma@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vhost-vdpa: introduce SUSPEND backend feature bit

Userland knows if it can suspend the device or not by checking this feature
bit.

It's only offered if the vdpa driver backend implements the suspend()
operation callback, and to offer it or userland to ack it if the backend
does not offer that callback is an error.

Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Message-Id: <20220810171512.2343333-3-eperezma@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vdpa: Add suspend operation

This operation is optional: It it's not implemented, backend feature bit
will not be exposed.

Signed-off-by: Eugenio Pérez <eperezma@redhat.com>
Message-Id: <20220810171512.2343333-2-eperezma@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio-blk: Avoid use-after-free on suspend/resume

hctx->user_data is set to vq in virtblk_init_hctx().  However, vq is
freed on suspend and reallocated on resume.  So, hctx->user_data is
invalid after resume, and it will cause use-after-free accessing which
will result in the kernel crash something like below:

[   22.428391] Call Trace:
[   22.428899]  <TASK>
[   22.429339]  virtqueue_add_split+0x3eb/0x620
[   22.430035]  ? __blk_mq_alloc_requests+0x17f/0x2d0
[   22.430789]  ? kvm_clock_get_cycles+0x14/0x30
[   22.431496]  virtqueue_add_sgs+0xad/0xd0
[   22.432108]  virtblk_add_req+0xe8/0x150
[   22.432692]  virtio_queue_rqs+0xeb/0x210
[   22.433330]  blk_mq_flush_plug_list+0x1b8/0x280
[   22.434059]  __blk_flush_plug+0xe1/0x140
[   22.434853]  blk_finish_plug+0x20/0x40
[   22.435512]  read_pages+0x20a/0x2e0
[   22.436063]  ? folio_add_lru+0x62/0xa0
[   22.436652]  page_cache_ra_unbounded+0x112/0x160
[   22.437365]  filemap_get_pages+0xe1/0x5b0
[   22.437964]  ? context_to_sid+0x70/0x100
[   22.438580]  ? sidtab_context_to_sid+0x32/0x400
[   22.439979]  filemap_read+0xcd/0x3d0
[   22.440917]  xfs_file_buffered_read+0x4a/0xc0
[   22.441984]  xfs_file_read_iter+0x65/0xd0
[   22.442970]  __kernel_read+0x160/0x2e0
[   22.443921]  bprm_execve+0x21b/0x640
[   22.444809]  do_execveat_common.isra.0+0x1a8/0x220
[   22.446008]  __x64_sys_execve+0x2d/0x40
[   22.446920]  do_syscall_64+0x37/0x90
[   22.447773]  entry_SYSCALL_64_after_hwframe+0x63/0xcd

This patch fixes this issue by getting vq from vblk, and removes
virtblk_init_hctx().

Fixes: 4e0400525691 ("virtio-blk: support polling I/O")
Cc: "Suwan Kim" <suwan.kim027@gmail.com>
Signed-off-by: Shigeru Yoshida <syoshida@redhat.com>
Message-Id: <20220810160948.959781-1-syoshida@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_vdpa: support the arg sizes of find_vqs()

Virtio vdpa support the new parameter sizes of find_vqs().

Signed-off-by: Bo Liu <liubo03@inspur.com>
Message-Id: <20220810085151.7251-1-liubo03@inspur.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vhost-vdpa: Call ida_simple_remove() when failed

In function vhost_vdpa_probe(), when code execution fails, we should
call ida_simple_remove() to free ida.

Signed-off-by: Bo Liu <liubo03@inspur.com>
Message-Id: <20220805091254.20026-1-liubo03@inspur.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vDPA: fix 'cast to restricted le16' warnings in vdpa.c

This commit fixes spars warnings: cast to restricted __le16
in function vdpa_dev_net_config_fill() and
vdpa_fill_stats_rec()

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Reviewed-by: Parav Pandit <parav@nvidia.com>
Message-Id: <20220722115309.82746-7-lingshan.zhu@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vDPA: !FEATURES_OK should not block querying device config space

Users may want to query the config space of a vDPA device,
to choose a appropriate one for a certain guest. This means the
users need to read the config space before FEATURES_OK, and
the existence of config space contents does not depend on
FEATURES_OK.

The spec says:
The device MUST allow reading of any device-specific configuration
field before FEATURES_OK is set by the driver. This includes
fields which are conditional on feature bits, as long as those
feature bits are offered by the device.

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Message-Id: <20220722115309.82746-5-lingshan.zhu@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vDPA/ifcvf: support userspace to query features and MQ of a management device

Adapting to current netlink interfaces, this commit allows userspace
to query feature bits and MQ capability of a management device.

Currently both the vDPA device and the management device are the VF itself,
thus this ifcvf should initialize the virtio capabilities in probe() before
setting up the struct vdpa_mgmt_dev.

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Message-Id: <20220722115309.82746-3-lingshan.zhu@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vDPA/ifcvf: get_config_size should return a value no greater than dev implementation

Drivers must not access a BAR outside the capability length,
and for a virtio device, ifcvf driver should not report any non-standard
capability contents to the upper layers.

Function ifcvf_get_config_size() is introduced here to return a safe value
of the device config capability size.

Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Message-Id: <20220722115309.82746-2-lingshan.zhu@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vhost scsi: Allow user to control num virtqueues

We are currently hard coded to always create 128 IO virtqueues, so this
adds a modparam to control it. For large systems where we are ok with
using memory for virtqueues it allows us to add up to 1024. This limit
was just selected becuase that's qemu's limit.

Signed-off-by: Mike Christie <michael.christie@oracle.com>
Message-Id: <20220708030525.5065-3-michael.christie@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vhost-scsi: Fix max number of virtqueues

Qemu takes it's num_queues limit then adds the fixed queues (control and
event) to the total it will request from the kernel. So when a user
requests 128 (or qemu does it's num_queues calculation based on vCPUS
and other system limits), we hit errors due to userspace trying to setup
130 queues when vhost-scsi has a hard coded limit of 128.

This has vhost-scsi adjust it's max so we can do a total of 130 virtqueues
(128 IO and 2 fixed). For the case where the user has 128 vCPUs the guest
OS can then nicely map each IO virtqueue to a vCPU and not have the odd case
where 2 vCPUs share a virtqueue.

Signed-off-by: Mike Christie <michael.christie@oracle.com>
Message-Id: <20220708030525.5065-2-michael.christie@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vdpa/mlx5: Support different address spaces for control and data

Partition virtqueues to two different address spaces: one for control
virtqueue which is implemented in software, and one for data virtqueues.

Based-on: <20220526124338.36247-1-eperezma@redhat.com>
Signed-off-by: Eli Cohen <elic@nvidia.com>
Message-Id: <20220714113927.85729-3-elic@nvidia.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vdpa/mlx5: Implement susupend virtqueue callback

Implement the suspend callback allowing to suspend the virtqueues so
they stop processing descriptors. This is required to allow to query a
consistent state of the virtqueue while live migration is taking place.

Signed-off-by: Eli Cohen <elic@nvidia.com>
Message-Id: <20220714113927.85729-2-elic@nvidia.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vduse: Support querying information of IOVA regions

This introduces a new ioctl: VDUSE_IOTLB_GET_INFO to
support querying some information of IOVA regions.

Now it can be used to query whether the IOVA region
supports userspace memory registration.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Message-Id: <20220803045523.23851-6-xieyongji@bytedance.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vduse: Support registering userspace memory for IOVA regions

Introduce two ioctls: VDUSE_IOTLB_REG_UMEM and
VDUSE_IOTLB_DEREG_UMEM to support registering
and de-registering userspace memory for IOVA
regions.

Now it only supports registering userspace memory
for bounce buffer region in virtio-vdpa case.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220803045523.23851-5-xieyongji@bytedance.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vduse: Support using userspace pages as bounce buffer

Introduce two APIs: vduse_domain_add_user_bounce_pages()
and vduse_domain_remove_user_bounce_pages() to support
adding and removing userspace pages for bounce buffers.
During adding and removing, the DMA data would be copied
from the kernel bounce pages to the userspace bounce pages
and back.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220803045523.23851-4-xieyongji@bytedance.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vduse: Use memcpy_{to,from}_page() in do_bounce()

kmap_atomic() is being deprecated in favor of kmap_local_page().

The use of kmap_atomic() in do_bounce() is all thread local therefore
kmap_local_page() is a sufficient replacement.

Convert to kmap_local_page() but, instead of open coding it,
use the helpers memcpy_to_page() and memcpy_from_page().

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Message-Id: <20220803045523.23851-3-xieyongji@bytedance.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vduse: Remove unnecessary spin lock protection

Now we use domain->iotlb_lock to protect two different
variables: domain->bounce_maps->bounce_page and
domain->iotlb. But for domain->bounce_maps->bounce_page,
we actually don't need any synchronization between
vduse_domain_get_bounce_page() and vduse_domain_free_bounce_pages()
since vduse_domain_get_bounce_page() will only be called in
page fault handler and vduse_domain_free_bounce_pages() will
be called during file release.

So let's remove the unnecessary spin lock protection in
vduse_domain_get_bounce_page(). Then the usage of
domain->iotlb_lock could be more clear: the lock will be
only used to protect the domain->iotlb.

Signed-off-by: Xie Yongji <xieyongji@bytedance.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220803045523.23851-2-xieyongji@bytedance.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

net: virtio_net: notifications coalescing support

New VirtIO network feature: VIRTIO_NET_F_NOTF_COAL.

Control a Virtio network device notifications coalescing parameters
using the control virtqueue.

A device that supports this fetature can receive
VIRTIO_NET_CTRL_NOTF_COAL control commands.

- VIRTIO_NET_CTRL_NOTF_COAL_TX_SET:
  Ask the network device to change the following parameters:
  - tx_usecs: Maximum number of usecs to delay a TX notification.
  - tx_max_packets: Maximum number of packets to send before a
    TX notification.

- VIRTIO_NET_CTRL_NOTF_COAL_RX_SET:
  Ask the network device to change the following parameters:
  - rx_usecs: Maximum number of usecs to delay a RX notification.
  - rx_max_packets: Maximum number of packets to receive before a
    RX notification.

VirtIO spec. patch:
https://lists.oasis-open.org/archives/virtio-comment/202206/msg00100.html

Signed-off-by: Alvaro Karsz <alvaro.karsz@solid-run.com>
Message-Id: <20220718091102.498774-1-alvaro.karsz@solid-run.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Jakub Kicinski <kuba@kernel.org>
Acked-by: Jason Wang <jasowang@redhat.com>

virtio: Check dev_set_name() return value

It's possible that dev_set_name() returns -ENOMEM, catch and handle this.

Signed-off-by: Bo Liu <liubo03@inspur.com>
Message-Id: <20220707031751.4802-1-liubo03@inspur.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

tools/virtio: fix build

Fix the build caused by the following changes:
- phys_addr_t is now defined in tools/include/linux/types.h
- dev_warn_once() is used in drivers/virtio/virtio_ring.c
- linux/uio.h included by vringh.h use INT_MAX defined in limits.h

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220705072249.7867-1-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Peng Fan <peng.fan@nxp.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vDPA/ifcvf: remove duplicated assignment to pointer cfg

The assignment to pointer cfg is duplicated, the second assignment
is redundant and can be removed.

Signed-off-by: Colin Ian King <colin.i.king@gmail.com>
Message-Id: <20220704190456.593464-1-colin.i.king@gmail.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vdpa: ifcvf: Fix spelling mistake in comments

There is a typo(does't) in comments.
It maybe 'doesn't' instead of 'does't'.

Signed-off-by: Zhang Jiaming <jiaming@nfschina.com>
Message-Id: <20220704024104.15535-1-jiaming@nfschina.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vdpa/mlx5: Use eth_broadcast_addr() to assign broadcast address

Using eth_broadcast_addr() to assign broadcast address instead
of memset().

Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Xu Qiang <xuqiang36@huawei.com>
Message-Id: <20220704021405.64545-1-xuqiang36@huawei.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vdpa_sim: use max_iotlb_entries as a limit in vhost_iotlb_init

Commit bda324fd037a ("vdpasim: control virtqueue support") changed
the allocation of iotlbs calling vhost_iotlb_init() for each address
space, instead of vhost_iotlb_alloc().

With this change we forgot to use the limit we had introduced with
the `max_iotlb_entries` module parameter.

Fixes: bda324fd037a ("vdpasim: control virtqueue support")
Cc: gautam.dawar@xilinx.com
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220621151208.189959-1-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Eugenio Pérez <eperezma@redhat.com>

vdpa_sim_blk: set number of address spaces and virtqueue groups

Commit bda324fd037a ("vdpasim: control virtqueue support") added two
new fields (nas, ngroups) to vdpasim_dev_attr, but we forgot to
initialize them for vdpa_sim_blk.

When creating a new vdpa_sim_blk device this causes the kernel
to panic in this way:
    $ vdpa dev add mgmtdev vdpasim_blk name blk0
    BUG: kernel NULL pointer dereference, address: 0000000000000030
    ...
    RIP: 0010:vhost_iotlb_add_range_ctx+0x41/0x220 [vhost_iotlb]
    ...
    Call Trace:
     <TASK>
     vhost_iotlb_add_range+0x11/0x800 [vhost_iotlb]
     vdpasim_map_range+0x91/0xd0 [vdpa_sim]
     vdpasim_alloc_coherent+0x56/0x90 [vdpa_sim]
     ...

This happens because vdpasim->iommu[0] is not initialized when
dev_attr.nas is 0.

Let's fix this issue by initializing both (nas, ngroups) to 1 for
vdpa_sim_blk.

Fixes: bda324fd037a ("vdpasim: control virtqueue support")
Cc: gautam.dawar@xilinx.com
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220621151323.190431-1-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Eugenio Pérez <eperezma@redhat.com>

vdpa_sim_blk: call vringh_complete_iotlb() also in the error path

Call vringh_complete_iotlb() even when we encounter a serious error
that prevents us from writing the status in the "in" header
(e.g. the header length is incorrect, etc.).

The guest is misbehaving, so maybe the ring is in a bad state, but
let's avoid making things worse.

Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220630153221.83371-4-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vdpa_sim_blk: limit the number of request handled per batch

Limit the number of requests (4 per queue as for vdpa_sim_net) handled
in a batch to prevent the worker from using the CPU for too long.

Suggested-by: Eugenio Pérez <eperezma@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220630153221.83371-3-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

vdpa_sim_blk: use dev_dbg() to print errors

Use dev_dbg() instead of dev_err()/dev_warn() to avoid flooding the
host with prints, when the guest driver is misbehaving.
In this way, prints can be dynamically enabled when the vDPA block
simulator is used to validate a driver.

Suggested-by: Jason Wang <jasowang@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220630153221.83371-2-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_net: support set_ringparam

Support set_ringparam based on virtio queue reset.

Users can use ethtool -G eth0 <ring_num> to modify the ring size of
virtio-net.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-43-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_net: support tx queue resize

This patch implements the resize function of the tx queues.
Based on this function, it is possible to modify the ring num of the
queue.

Inludes fixup:

virtio_net: fix for stuck when change tx ring size with dev down

When dev is set to DOWN state, napi has been disabled, if we modify the
ring size at this time, we should not call napi_disable() again, which
will cause stuck.

And all operations are under the protection of rtnl_lock, so there is no
need to consider concurrency issues.

Message-Id: <20220801063902.129329-42-xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220811080258.79398-3-xuanzhuo@linux.alibaba.com>
Reported-by: Kangjie Xu <kangjie.xu@linux.alibaba.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_net: support rx queue resize

This patch implements the resize function of the rx queues.
Based on this function, it is possible to modify the ring num of the
queue.

Includes fixup:

virtio_net: fix for stuck when change rx ring size with dev down

When dev is set to DOWN state, napi has been disabled, if we modify the
ring size at this time, we should not call napi_disable() again, which
will cause stuck.

And all operations are under the protection of rtnl_lock, so there is no
need to consider concurrency issues.

Message-Id: <20220801063902.129329-41-xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220811080258.79398-2-xuanzhuo@linux.alibaba.com>
Reported-by: Kangjie Xu <kangjie.xu@linux.alibaba.com>
Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_net: split free_unused_bufs()

This patch separates two functions for freeing sq buf and rq buf from
free_unused_bufs().

When supporting the enable/disable tx/rq queue in the future, it is
necessary to support separate recovery of a sq buf or a rq buf.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-40-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_net: get ringparam by virtqueue_get_vring_max_size()

Use virtqueue_get_vring_max_size() in virtnet_get_ringparam() to set
tx,rx_max_pending.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-39-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_net: set the default max ring size by find_vqs()

Use virtio_find_vqs_ctx_size() to specify the maximum ring size of tx,
rx at the same time.

                         | rx/tx ring size
-------------------------------------------
speed == UNKNOWN or < 10G| 1024
speed < 40G              | 4096
speed >= 40G             | 8192

Call virtnet_update_settings() once before calling init_vqs() to update
speed.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-38-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: add helper virtio_find_vqs_ctx_size()

Introduce helper virtio_find_vqs_ctx_size() to call find_vqs and specify
the maximum size of each vq ring.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-37-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_mmio: support the arg sizes of find_vqs()

Virtio MMIO support the new parameter sizes of find_vqs().

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-36-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pci: support the arg sizes of find_vqs()

Virtio PCI supports new parameter sizes of find_vqs().

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-35-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: find_vqs() add arg sizes

find_vqs() adds a new parameter sizes to specify the size of each vq
vring.

NULL as sizes means that all queues in find_vqs() use the maximum size.
A value in the array is 0, which means that the corresponding queue uses
the maximum size.

In the split scenario, the meaning of size is the largest size, because
it may be limited by memory, the virtio core will try a smaller size.
And the size is power of 2.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-34-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pci: support VIRTIO_F_RING_RESET

This patch implements virtio pci support for QUEUE RESET.

Performing reset on a queue is divided into these steps:

 1. notify the device to reset the queue
 2. recycle the buffer submitted
 3. reset the vring (may re-alloc)
 4. mmap vring to device, and enable the queue

This patch implements virtio_reset_vq(), virtio_enable_resetq() in the
pci scenario.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-33-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pci: extract the logic of active vq for modern pci

Introduce vp_active_vq() to configure vring to backend after vq attach
vring. And configure vq vector if necessary.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-32-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pci: introduce helper to get/set queue reset

Introduce new helpers to implement queue reset and get queue reset
status.

 https://github.com/oasis-tcs/virtio-spec/issues/124
 https://github.com/oasis-tcs/virtio-spec/issues/139

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-31-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pci: struct virtio_pci_common_cfg add queue_reset

Add queue_reset in virtio_pci_modern_common_cfg.

 https://github.com/oasis-tcs/virtio-spec/issues/124
 https://github.com/oasis-tcs/virtio-spec/issues/139

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-30-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: struct virtqueue introduce reset

Introduce a new member reset to the structure virtqueue to determine
whether the current vq is in the reset state. Subsequent patches will
use it.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-29-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: queue_reset: add VIRTIO_F_RING_RESET

Added VIRTIO_F_RING_RESET, it came from here

https://github.com/oasis-tcs/virtio-spec/issues/124
https://github.com/oasis-tcs/virtio-spec/issues/139

This feature indicates that the driver can reset a queue individually.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-28-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: allow to unbreak/break virtqueue individually

This patch allows the new introduced
__virtqueue_break()/__virtqueue_unbreak() to break/unbreak the
virtqueue.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-27-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pci: struct virtio_pci_common_cfg add queue_notify_data

Add queue_notify_data in struct virtio_pci_common_cfg, which comes from
here https://github.com/oasis-tcs/virtio-spec/issues/89

In order not to affect the API, add a dedicated structure struct
virtio_pci_modern_common_cfg to virtio_pci_modern.h.

Since I want to add queue_reset after queue_notify_data, I submitted
this patch first.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-26-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: introduce virtqueue_resize()

Introduce virtqueue_resize() to implement the resize of vring.
Based on these, the driver can dynamically adjust the size of the vring.
For example: ethtool -G.

virtqueue_resize() implements resize based on the vq reset function. In
case of failure to allocate a new vring, it will give up resize and use
the original vring.

During this process, if the re-enable reset vq fails, the vq can no
longer be used. Although the probability of this situation is not high.

The parameter recycle is used to recycle the buffer that is no longer
used.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-25-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: introduce virtqueue_resize_packed()

virtio ring packed supports resize.

Only after the new vring is successfully allocated based on the new num,
we will release the old vring. In any case, an error is returned,
indicating that the vring still points to the old vring.

In the case of an error, re-initialize(by virtqueue_reinit_packed()) the
virtqueue to ensure that the vring can be used.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-24-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: introduce virtqueue_reinit_packed()

Introduce a function to initialize vq without allocating new ring,
desc_state, desc_extra.

Subsequent patches will call this function after reset vq to
reinitialize vq.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-23-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: extract the logic of attach vring

Separate the logic of attach vring, the subsequent patch will call it
separately.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-22-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: extract the logic of vring init

Separate the logic of initializing vring, and subsequent patches will
call it separately.

This function completes the variable initialization of packed vring. It
together with the logic of atatch constitutes the initialization of
vring.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-21-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: extract the logic of alloc state and extra

Separate the logic for alloc desc_state and desc_extra, which will
be called separately by subsequent patches.

Use struct vring_packed to pass desc_state, desc_extra.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-20-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: extract the logic of alloc queue

Separate the logic of packed to create vring queue.

This feature is required for subsequent virtuqueue reset vring.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-19-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: packed: introduce vring_free_packed

Free the structure struct vring_vritqueue_packed.

Subsequent patches require it.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-18-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: introduce virtqueue_resize_split()

virtio ring split supports resize.

Only after the new vring is successfully allocated based on the new num,
we will release the old vring. In any case, an error is returned,
indicating that the vring still points to the old vring.

In the case of an error, re-initialize(virtqueue_reinit_split()) the
virtqueue to ensure that the vring can be used.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-17-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: reserve vring_align, may_reduce_num

In vring_alloc_queue_split() save vring_align, may_reduce_num to
structure vring_virtqueue_split. Used to create a new vring when
implementing resize.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-16-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: introduce virtqueue_reinit_split()

Introduce a function to initialize vq without allocating new ring,
desc_state, desc_extra.

Subsequent patches will call this function after reset vq to
reinitialize vq.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-15-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: extract the logic of attach vring

Separate the logic of attach vring, subsequent patches will call it
separately.

virtqueue_vring_init_split() completes the initialization of other
variables of vring split. We can directly use
vq->split = *vring_split to complete attach.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-14-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: extract the logic of vring init

Separate the logic of initializing vring, and subsequent patches will
call it separately.

This function completes the variable initialization of split vring. It
together with the logic of atatch constitutes the initialization of
vring.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-13-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: extract the logic of alloc state and extra

Separate the logic of creating desc_state, desc_extra, and subsequent
patches will call it independently.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-12-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: extract the logic of alloc queue

Separate the logic of split to create vring queue.

This feature is required for subsequent virtuqueue reset vring.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-11-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: introduce vring_free_split()

Free the structure struct vring_vritqueue_split.

Subsequent patches require it.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-10-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: __vring_new_virtqueue() accept struct vring_virtqueue_split

__vring_new_virtqueue() instead accepts struct vring_virtqueue_split.

The purpose of this is to pass more information into
__vring_new_virtqueue() to make the code simpler and the structure
cleaner.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-9-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split: stop __vring_new_virtqueue as export symbol

There is currently only one place to reference __vring_new_virtqueue()
directly from the outside of virtio core. And here vring_new_virtqueue()
can be used instead.

Subsequent patches will modify __vring_new_virtqueue, so stop it as an
export symbol for now.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-8-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: introduce virtqueue_init()

Separate the logic of virtqueue initialization. These variables should
be reset during reset.

This logic can be called independently when implementing resize/reset
later.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-7-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: split vring_virtqueue

Separate the two inline structures(split and packed) from the structure
vring_virtqueue.

In this way, we can use these two structures later to pass parameters
and retain temporary variables.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-6-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: extract the logic of freeing vring

Introduce vring_free() to free the vring of vq.

Subsequent patches will use vring_free() alone.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-5-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: update the document of the virtqueue_detach_unused_buf for queue reset

Added documentation for virtqueue_detach_unused_buf, allowing it to be
called on queue reset.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-4-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: struct virtio_config_ops add callbacks for queue_reset

reset can be divided into the following four steps (example):
 1. transport: notify the device to reset the queue
 2. vring:     recycle the buffer submitted
 3. vring:     reset/resize the vring (may re-alloc)
 4. transport: mmap vring to device, and enable the queue

In order to support queue reset, add two callbacks in struct
virtio_config_ops to implement steps 1 and 4.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-3-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio: record the maximum queue num supported by the device.

virtio-net can display the maximum (supported by hardware) ring size in
ethtool -g eth0.

When the subsequent patch implements vring reset, it can judge whether
the ring size passed by the driver is legal based on this.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220801063902.129329-2-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

drivers/virtio: Clarify CONFIG_VIRTIO_MEM for unsupported architectures

Let's make it clearer that simply unlocking CONFIG_VIRTIO_MEM on an
architecture is most probably not sufficient to have it working as
expected.

Cc: "Michael S. Tsirkin" <mst@redhat.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Signed-off-by: David Hildenbrand <david@redhat.com>
Message-Id: <20220610094737.65254-1-david@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_mmio: add support to set IRQ of a virtio device as wakeup source

According to virtio_mmio wakeup flag in device trees, set its IRQ
as wakeup source in virtqueue initialization.

Signed-off-by: Minghao Xue <quic_mingxue@quicinc.com>
Message-Id: <1654851507-13891-3-git-send-email-quic_mingxue@quicinc.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

dt-bindings: virtio: mmio: add optional wakeup-source property

Some systems want to set the interrupt of virtio_mmio device
as a wakeup source. On such systems, we'll use the existence
of the "wakeup-source" property as a signal of requirement.

Signed-off-by: Minghao Xue <quic_mingxue@quicinc.com>
Reviewed-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Message-Id: <1654851507-13891-2-git-send-email-quic_mingxue@quicinc.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vdpa: Use device_iommu_capable()

Use the new interface to check the capability for our device
specifically.

Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Message-Id: <548e316fa282ce513fabb991a4c4d92258062eb5.1654688822.git.robin.murphy@arm.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

virtio: VIRTIO_HARDEN_NOTIFICATION is broken

This option doesn't really work and breaks too many drivers.
Not yet sure what's the right thing to do, for now
let's make sure randconfig isn't broken by this.

Fixes: c346dae4f3fb ("virtio: disable notification hardening by default")
Cc: "Jason Wang" <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>

virtio_pmem: set device ready in probe()

The NVDIMM region could be available before the virtio_device_ready()
that is called by virtio_dev_probe(). This means the driver tries to
use device before DRIVER_OK which violates the spec, fixing this by
set device ready before the nvdimm_pmem_region_create().

Note that this means the virtio_pmem_host_ack() could be triggered
before the creation of the nd region, this is safe since the pmem_lock
has been initialized and whether or not any available buffer is added
before is validated by virtio_pmem_host_ack().

Fixes 6e84200c0a29 ("virtio-pmem: Add virtio pmem driver")
Acked-by: Pankaj Gupta <pankaj.gupta@amd.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220628083430.61856-2-jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_pmem: initialize provider_data through nd_region_desc

We used to initialize the provider_data manually after
nvdimm_pemm_region_create(). This seems to be racy if the flush is
issued before the initialization of provider_data[1]. Fixing this by
initializing the provider_data through nd_region_desc to make sure the
provider_data is ready after the pmem is created.

[1]:

[   80.152281] nd_pmem namespace0.0: unable to guarantee persistence of writes
[   92.393956] BUG: kernel NULL pointer dereference, address: 0000000000000318
[   92.394551] #PF: supervisor read access in kernel mode
[   92.394955] #PF: error_code(0x0000) - not-present page
[   92.395365] PGD 0 P4D 0
[   92.395566] Oops: 0000 [#1] PREEMPT SMP PTI
[   92.395867] CPU: 2 PID: 506 Comm: mkfs.ext4 Not tainted 5.19.0-rc1+ #453
[   92.396365] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009),
BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
[   92.397178] RIP: 0010:virtio_pmem_flush+0x2f/0x1f0
[   92.397521] Code: 55 41 54 55 53 48 81 ec a0 00 00 00 65 48 8b 04
25 28 00 00 00 48 89 84 24 98 00 00 00 31 c0 48 8b 87 78 03 00 00 48
89 04 24 <48> 8b 98 18 03 00 00 e8 85 bf 6b 00 ba 58 00 00 00 be c0 0c
00 00
[   92.398982] RSP: 0018:ffff9a7380aefc88 EFLAGS: 00010246
[   92.399349] RAX: 0000000000000000 RBX: ffff8e77c3f86f00 RCX: 0000000000000000
[   92.399833] RDX: ffffffffad4ea720 RSI: ffff8e77c41e39c0 RDI: ffff8e77c41c5c00
[   92.400388] RBP: ffff8e77c41e39c0 R08: ffff8e77c19f0600 R09: 0000000000000000
[   92.400874] R10: 0000000000000000 R11: 0000000000000000 R12: ffff8e77c0814e28
[   92.401364] R13: 0000000000000000 R14: 0000000000000000 R15: ffff8e77c41e39c0
[   92.401849] FS:  00007f3cd75b2780(0000) GS:ffff8e7937d00000(0000)
knlGS:0000000000000000
[   92.402423] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   92.402821] CR2: 0000000000000318 CR3: 0000000103c80002 CR4: 0000000000370ee0
[   92.403307] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[   92.403793] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[   92.404278] Call Trace:
[   92.404481]  <TASK>
[   92.404654]  ? mempool_alloc+0x5d/0x160
[   92.404939]  ? terminate_walk+0x5f/0xf0
[   92.405226]  ? bio_alloc_bioset+0xbb/0x3f0
[   92.405525]  async_pmem_flush+0x17/0x80
[   92.405806]  nvdimm_flush+0x11/0x30
[   92.406067]  pmem_submit_bio+0x1e9/0x200
[   92.406354]  __submit_bio+0x80/0x120
[   92.406621]  submit_bio_noacct_nocheck+0xdc/0x2a0
[   92.406958]  submit_bio_wait+0x4e/0x80
[   92.407234]  blkdev_issue_flush+0x31/0x50
[   92.407526]  ? punt_bios_to_rescuer+0x230/0x230
[   92.407852]  blkdev_fsync+0x1e/0x30
[   92.408112]  do_fsync+0x33/0x70
[   92.408354]  __x64_sys_fsync+0xb/0x10
[   92.408625]  do_syscall_64+0x43/0x90
[   92.408895]  entry_SYSCALL_64_after_hwframe+0x46/0xb0
[   92.409257] RIP: 0033:0x7f3cd76c6c44

Fixes 6e84200c0a29 ("virtio-pmem: Add virtio pmem driver")
Acked-by: Pankaj Gupta <pankaj.gupta@amd.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220628083430.61856-1-jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

vringh: iterate on iotlb_translate to handle large translations

iotlb_translate() can return -ENOBUFS if the bio_vec is not big enough
to contain all the ranges for translation.
This can happen for example if the VMM maps a large bounce buffer,
without using hugepages, that requires more than 16 ranges to translate
the addresses.

To handle this case, let's extend iotlb_translate() to also return the
number of bytes successfully translated.
In copy_from_iotlb()/copy_to_iotlb() loops by calling iotlb_translate()
several times until we complete the translation.

Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220624075656.13997-1-sgarzare@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

virtio_ring: remove the arg vq of vring_alloc_desc_extra()

The parameter vq of vring_alloc_desc_extra() is useless. This patch
removes this parameter.

Subsequent patches will call this function to avoid passing useless
arguments.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Message-Id: <20220624025621.128843-6-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

remoteproc: rename len of rpoc_vring to num

Rename the member len in the structure rpoc_vring to num. And remove 'in
bytes' from the comment of it. This is misleading. Because this actually
refers to the size of the virtio vring to be created. The unit is not
bytes.

Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Message-Id: <20220624025621.128843-2-xuanzhuo@linux.alibaba.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

Linux 5.19

Merge tag 'clk-fixes-for-linus' of git://git./linux/kernel/git/clk/linux

Pull clk fix from Stephen Boyd:
 "One-liner fix of a NULL pointer deref in the Allwinner clk driver"

* tag 'clk-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/clk/linux:
  clk: sunxi-ng: Fix H6 RTC clock definition

Merge tag 'x86_urgent_for_v5.19' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Borislav Petkov:

 - Update the 'mitigations=' kernel param documentation

 - Check the IBPB feature flag before enabling IBPB in firmware calls
   because cloud vendors' fantasy when it comes to creating guest
   configurations is unlimited

 - Unexport sev_es_ghcb_hv_call() before 5.19 releases now that HyperV
   doesn't need it anymore

 - Remove dead CONFIG_* items

* tag 'x86_urgent_for_v5.19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  docs/kernel-parameters: Update descriptions for "mitigations=" param with retbleed
  x86/bugs: Do not enable IBPB at firmware entry when IBPB is not available
  Revert "x86/sev: Expose sev_es_ghcb_hv_call() for use by HyperV"
  x86/configs: Update configs in x86_debug.config

Merge tag 'locking_urgent_for_v5.19' of git://git./linux/kernel/git/tip/tip

Pull locking fix from Borislav Petkov:

 - Avoid rwsem lockups in certain situations when handling the handoff
   bit

* tag 'locking_urgent_for_v5.19' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  locking/rwsem: Allow slowpath writer to ignore handoff bit if not set by first waiter

Merge tag 'edac_urgent_for_v5.19' of git://git./linux/kernel/git/ras/ras

Pull EDAC fixes from Borislav Petkov:

 - Relax the condition under which the DIMM label in ghes_edac is set in
   order to accomodate an HPE BIOS which sets only the device but not
   the bank

 - Two forgotten fixes to synopsys_edac when handling error interrupts

* tag 'edac_urgent_for_v5.19' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras:
  EDAC/ghes: Set the DIMM label unconditionally
  EDAC/synopsys: Re-enable the error interrupts on v3 hw
  EDAC/synopsys: Use the correct register to disable the error interrupt on v3 hw

Merge tag 'for-linus' of git://git.armlinux.org.uk/~rmk/linux-arm

Pull ARM fixes from Russell King:
 "Last set of ARM fixes for 5.19:

   - fix for MAX_DMA_ADDRESS overflow

   - fix for find_*_bit performing an out of bounds memory access"

* tag 'for-linus' of git://git.armlinux.org.uk/~rmk/linux-arm:
  ARM: findbit: fix overflowing offset
  ARM: 9216/1: Fix MAX_DMA_ADDRESS overflow

locking/rwsem: Allow slowpath writer to ignore handoff bit if not set by first waiter

With commit d257cc8cb8d5 ("locking/rwsem: Make handoff bit handling more
consistent"), the writer that sets the handoff bit can be interrupted
out without clearing the bit if the wait queue isn't empty. This disables
reader and writer optimistic lock spinning and stealing.

Now if a non-first writer in the queue is somehow woken up or a new
waiter enters the slowpath, it can't acquire the lock.  This is not the
case before commit d257cc8cb8d5 as the writer that set the handoff bit
will clear it when exiting out via the out_nolock path. This is less
efficient as the busy rwsem stays in an unlock state for a longer time.

In some cases, this new behavior may cause lockups as shown in [1] and
[2].

This patch allows a non-first writer to ignore the handoff bit if it
is not originally set or initiated by the first waiter. This patch is
shown to be effective in fixing the lockup problem reported in [1].

[1] https://lore.kernel.org/lkml/20220617134325.GC30825@techsingularity.net/
[2] https://lore.kernel.org/lkml/3f02975c-1a9d-be20-32cf-f1d8e3dfafcc@oracle.com/

Fixes: d257cc8cb8d5 ("locking/rwsem: Make handoff bit handling more consistent")
Signed-off-by: Waiman Long <longman@redhat.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: John Donnelly <john.p.donnelly@oracle.com>
Tested-by: Mel Gorman <mgorman@techsingularity.net>
Link: https://lore.kernel.org/r/20220622200419.778799-1-longman@redhat.com

Merge tag 'mm-hotfixes-stable-2022-07-29' of git://git./linux/kernel/git/akpm/mm

Pull misc fixes from Andrew Morton:
 "Two hotfixes, both cc:stable"

* tag 'mm-hotfixes-stable-2022-07-29' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  mm/hmm: fault non-owner device private entries
  page_alloc: fix invalid watermark check on a negative value

Merge tag 'block-5.19-2022-07-29' of git://git.kernel.dk/linux-block

Pull block fix from Jens Axboe:
 "Just a single fix for NVMe, yet another quirk addition"

* tag 'block-5.19-2022-07-29' of git://git.kernel.dk/linux-block:
  nvme-pci: Crucial P2 has bogus namespace ids

Merge tag 'drm-fixes-2022-07-30' of git://anongit.freedesktop.org/drm/drm

Pull more drm fixes from Dave Airlie:
 "Maxime had the dog^Wmailing list server eat his homework^Wmisc pull
  request.

  Two more small fixes, one in nouveau svm code and the other in
  simpledrm.

  nouveau:
   - page migration fix

  simpledrm:
   - fix mode_valid return value"

* tag 'drm-fixes-2022-07-30' of git://anongit.freedesktop.org/drm/drm:
  nouveau/svm: Fix to migrate all requested pages
  drm/simpledrm: Fix return type of simpledrm_simple_display_pipe_mode_valid()

Merge tag 'drm-misc-fixes-2022-07-29' of git://anongit.freedesktop.org/drm/drm-misc into drm-fixes

One fix to fix simpledrm mode_valid return value, and one for page
migration in nouveau

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20220729094514.sfzhc3gqjgwgal62@penduick

Merge tag 'scsi-fixes' of git://git./linux/kernel/git/jejb/scsi

Pull SCSI fixes from James Bottomley:
 "Four fixes, three in drivers.

  The two biggest fixes are ufs and the remaining driver and core fix
  are small and obvious (and the core fix is low risk)"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  scsi: ufs: core: Fix a race condition related to device management
  scsi: core: Fix warning in scsi_alloc_sgtables()
  scsi: ufs: host: Hold reference returned by of_parse_phandle()
  scsi: mpt3sas: Stop fw fault watchdog work item during system shutdown

docs/kernel-parameters: Update descriptions for "mitigations=" param with retbleed

Updates descriptions for "mitigations=off" and "mitigations=auto,nosmt"
with the respective retbleed= settings.

Signed-off-by: Eiichi Tsukata <eiichi.tsukata@nutanix.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: corbet@lwn.net
Link: https://lore.kernel.org/r/20220728043907.165688-1-eiichi.tsukata@nutanix.com

mm/hmm: fault non-owner device private entries

If hmm_range_fault() is called with the HMM_PFN_REQ_FAULT flag and a
device private PTE is found, the hmm_range::dev_private_owner page is used
to determine if the device private page should not be faulted in.
However, if the device private page is not owned by the caller,
hmm_range_fault() returns an error instead of calling migrate_to_ram() to
fault in the page.

For example, if a page is migrated to GPU private memory and a RDMA fault
capable NIC tries to read the migrated page, without this patch it will
get an error.  With this patch, the page will be migrated back to system
memory and the NIC will be able to read the data.

Link: https://lkml.kernel.org/r/20220727000837.4128709-2-rcampbell@nvidia.com
Link: https://lkml.kernel.org/r/20220725183615.4118795-2-rcampbell@nvidia.com
Fixes: 08ddddda667b ("mm/hmm: check the device private page owner in hmm_range_fault()")
Signed-off-by: Ralph Campbell <rcampbell@nvidia.com>
Reported-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Cc: Philip Yang <Philip.Yang@amd.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

page_alloc: fix invalid watermark check on a negative value

There was a report that a task is waiting at the
throttle_direct_reclaim. The pgscan_direct_throttle in vmstat was
increasing.

This is a bug where zone_watermark_fast returns true even when the free
is very low. The commit f27ce0e14088 ("page_alloc: consider highatomic
reserve in watermark fast") changed the watermark fast to consider
highatomic reserve. But it did not handle a negative value case which
can be happened when reserved_highatomic pageblock is bigger than the
actual free.

If watermark is considered as ok for the negative value, allocating
contexts for order-0 will consume all free pages without direct reclaim,
and finally free page may become depleted except highatomic free.

Then allocating contexts may fall into throttle_direct_reclaim. This
symptom may easily happen in a system where wmark min is low and other
reclaimers like kswapd does not make free pages quickly.

Handle the negative case by using MIN.

Link: https://lkml.kernel.org/r/20220725095212.25388-1-jaewon31.kim@samsung.com
Fixes: f27ce0e14088 ("page_alloc: consider highatomic reserve in watermark fast")
Signed-off-by: Jaewon Kim <jaewon31.kim@samsung.com>
Reported-by: GyeongHwan Hong <gh21.hong@samsung.com>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Baoquan He <bhe@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Yong-Taek Lee <ytk.lee@samsung.com>
Cc: <stable@vger.kerenl.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

Merge tag 'perf-tools-fixes-for-v5.19-2022-07-29' of git://git./linux/kernel/git/acme/linux

Pull perf tools fixes from Arnaldo Carvalho de Melo:

 - Fix addresses for bss symbols, describing variables used in resolving
   data access in tools such as 'perf c2c' and 'perf mem'.

 - Skip symbols if SHF_ALLOC flag is not set, a technique used for
   listing deprecated symbols, its addresses are zeros, so not useful.

 - Remove undefined behavior from bpf_perf_object__next() when dealing
   with an empty bpf_objects_list list.

 - Make a ARM CoreSight disasm script work with both python2 and
   python3.

 - Sync x86's cpufeatures header with with the kernel sources.

* tag 'perf-tools-fixes-for-v5.19-2022-07-29' of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux:
  perf bpf: Remove undefined behavior from bpf_perf_object__next()
  perf symbol: Skip symbols if SHF_ALLOC flag is not set
  perf symbol: Correct address for bss symbols
  perf scripts python: Let script to be python2 compliant
  tools headers cpufeatures: Sync with the kernel sources

Merge tag 'wq-for-5.19-rc8-fixes' of git://git./linux/kernel/git/tj/wq

Pull workqueue fix from Tejun Heo:
 "Just one commit to suppress a spurious warning added during the 5.19
  cycle"

* tag 'wq-for-5.19-rc8-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq:
  workqueue: Avoid a false warning in unbind_workers()

Merge tag 'pm-5.19-rc9' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fix from Rafael Wysocki:
 "Make some false positive RCU splats resulting from a recent intel_idle
  driver change go away (Waiman Long)"

* tag 'pm-5.19-rc9' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  intel_idle: Fix false positive RCU splats due to incorrect hardirqs state