This patch renames vhost_work_dev_flush to just vhost_dev_flush to
relfect that it flushes everything on the device and that drivers
don't know/care that polls are based on vhost_works. Drivers just
flush the entire device and polls, and works for vhost-scsi
management TMFs and IO net virtqueues, etc all are flushed.
Signed-off-by: Mike Christie <michael.christie@oracle.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220517180850.198915-9-michael.christie@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost_poll_flush() is a simple wrapper around vhost_work_dev_flush().
It gives wrong impression that we are doing some work over vhost_poll,
while in fact it flushes vhost_poll->dev.
It only complicate understanding of the code and leads to mistakes
like flushing the same vhost_dev several times in a row.
Just remove vhost_poll_flush() and call vhost_work_dev_flush() directly.
Signed-off-by: Andrey Ryabinin <arbn@yandex-team.com>
[merge vhost_poll_flush removal from Stefano Garzarella]
Signed-off-by: Mike Christie <michael.christie@oracle.com>
Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Message-Id: <20220517180850.198915-2-michael.christie@oracle.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch extends the vhost-vdpa to support ASID based IOTLB API. The
vhost-vdpa device will allocated multiple IOTLBs for vDPA device that
supports multiple address spaces. The IOTLBs and vDPA device memory
mappings is determined and maintained through ASID.
Note that we still don't support vDPA device with more than one
address spaces that depends on platform IOMMU. This work will be done
by moving the IOMMU logic from vhost-vDPA to vDPA device driver.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Gautam Dawar <gdawar@xilinx.com>
Message-Id: <20220330180436.24644-16-gdawar@xilinx.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Includes fixup:
vhost-vdpa: Fix some error handling path in vhost_vdpa_process_iotlb_msg()
In the error paths introduced by the original patch, a mutex may be left locked.
Add the correct goto instead of a direct return.
Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr>
Message-Id: <89ef0ae4c26ac3cfa440c71e97e392dcb328ac1b.1653227924.git.christophe.jaillet@wanadoo.fr>
Acked-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patches allows userspace to send ASID based IOTLB message to
vhost. This idea is to use the reserved u32 field in the existing V2
IOTLB message. Vhost device should advertise this capability via
VHOST_BACKEND_F_IOTLB_ASID backend feature.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Gautam Dawar <gdawar@xilinx.com>
Message-Id: <20220330180436.24644-10-gdawar@xilinx.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
In vhost_enable_notify() we enable the notifications and we read
the avail index to check if new buffers have become available in
the meantime.
We are not caching the avail index, so when the device will call
vhost_get_vq_desc(), it will find the old value in the cache and
it will read the avail index again.
It would be better to refresh the cache every time we read avail
index, so let's change vhost_enable_notify() caching the value in
`avail_idx` and compare it with `last_avail_idx` to check if there
are new buffers available.
We don't expect a significant performance boost because
the above path is not very common, indeed vhost_enable_notify()
is often called with unlikely(), expecting that avail index has
not been updated.
We ran virtio-test/vhost-test and noticed minimal improvement as
expected. To stress the patch more, we modified vhost_test.ko to
call vhost_enable_notify()/vhost_disable_notify() on every cycle
when calling vhost_get_vq_desc(); in this case we observed a more
evident improvement, with a reduction of the test execution time
of about 3.7%.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Link: https://lore.kernel.org/r/20220121153108.187291-1-sgarzare@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Commit e2ae38cf3d ("vhost: fix hung thread due to erroneous iotlb
entries") tries to reject the IOTLB message whose size is zero. But
the size is not necessarily meaningful, one example is the batching
hint, so the commit breaks that.
Fixing this be reject zero size message only if the message is used to
update/invalidate the IOTLB.
Fixes: e2ae38cf3d ("vhost: fix hung thread due to erroneous iotlb entries")
Reported-by: Eli Cohen <elic@nvidia.com>
Cc: Anirudh Rayabharam <mail@anirudhrb.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20220310075211.4801-1-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Tested-by: Eli Cohen <elic@nvidia.com>
In vhost_iotlb_add_range_ctx(), range size can overflow to 0 when
start is 0 and last is ULONG_MAX. One instance where it can happen
is when userspace sends an IOTLB message with iova=size=uaddr=0
(vhost_process_iotlb_msg). So, an entry with size = 0, start = 0,
last = ULONG_MAX ends up in the iotlb. Next time a packet is sent,
iotlb_access_ok() loops indefinitely due to that erroneous entry.
Call Trace:
<TASK>
iotlb_access_ok+0x21b/0x3e0 drivers/vhost/vhost.c:1340
vq_meta_prefetch+0xbc/0x280 drivers/vhost/vhost.c:1366
vhost_transport_do_send_pkt+0xe0/0xfd0 drivers/vhost/vsock.c:104
vhost_worker+0x23d/0x3d0 drivers/vhost/vhost.c:372
kthread+0x2e9/0x3a0 kernel/kthread.c:377
ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:295
</TASK>
Reported by syzbot at:
https://syzkaller.appspot.com/bug?extid=0abd373e2e50d704db87
To fix this, do two things:
1. Return -EINVAL in vhost_chr_write_iter() when userspace asks to map
a range with size 0.
2. Fix vhost_iotlb_add_range_ctx() to handle the range [0, ULONG_MAX]
by splitting it into two entries.
Fixes: 0bbe30668d ("vhost: factor out IOTLB")
Reported-by: syzbot+0abd373e2e50d704db87@syzkaller.appspotmail.com
Tested-by: syzbot+0abd373e2e50d704db87@syzkaller.appspotmail.com
Signed-off-by: Anirudh Rayabharam <mail@anirudhrb.com>
Link: https://lore.kernel.org/r/20220305095525.5145-1-mail@anirudhrb.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vhost_reset_is_le() is vhost_init_is_le(), and in the case of
cross-endian legacy, vhost_init_is_le() depends on vq->user_be.
vq->user_be is set by vhost_disable_cross_endian().
But in vhost_vq_reset(), we have:
vhost_reset_is_le(vq);
vhost_disable_cross_endian(vq);
And so user_be is used before being set.
To fix that, reverse the lines order as there is no other dependency
between them.
Signed-off-by: Laurent Vivier <lvivier@redhat.com>
Link: https://lore.kernel.org/r/20210312140913.788592-1-lvivier@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
When the IOTLB device is enabled, the log_guest_addr that is passed by
userspace to the VHOST_SET_VRING_ADDR ioctl, and which is then written
to vq->log_addr, is a GIOVA. All writes to this address are translated
by log_user() to writes to an HVA, and then ultimately logged through
the corresponding GPAs in log_write_hva(). No logging will ever occur
with vq->log_addr in this case. It is thus wrong to pass vq->log_addr
and log_guest_addr to log_access_vq() which assumes they are actual
GPAs.
Introduce a new vq_log_used_access_ok() helper that only checks accesses
to the log for the used structure when there isn't an IOTLB device around.
Signed-off-by: Greg Kurz <groug@kaod.org>
Link: https://lore.kernel.org/r/160171933385.284610.10189082586063280867.stgit@bahia.lan
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
"enable" should be "disable" when the function name is
vhost_disable_notify(), which does the disabling work.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Acked-by: Jason Wang <jasowang@redhat.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull virtio updates from Michael Tsirkin:
- IRQ bypass support for vdpa and IFC
- MLX5 vdpa driver
- Endianness fixes for virtio drivers
- Misc other fixes
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost: (71 commits)
vdpa/mlx5: fix up endian-ness for mtu
vdpa: Fix pointer math bug in vdpasim_get_config()
vdpa/mlx5: Fix pointer math in mlx5_vdpa_get_config()
vdpa/mlx5: fix memory allocation failure checks
vdpa/mlx5: Fix uninitialised variable in core/mr.c
vdpa_sim: init iommu lock
virtio_config: fix up warnings on parisc
vdpa/mlx5: Add VDPA driver for supported mlx5 devices
vdpa/mlx5: Add shared memory registration code
vdpa/mlx5: Add support library for mlx5 VDPA implementation
vdpa/mlx5: Add hardware descriptive header file
vdpa: Modify get_vq_state() to return error code
net/vdpa: Use struct for set/get vq state
vdpa: remove hard coded virtq num
vdpasim: support batch updating
vhost-vdpa: support IOTLB batching hints
vhost-vdpa: support get/set backend features
vhost: generialize backend features setting/getting
vhost-vdpa: refine ioctl pre-processing
vDPA: dont change vq irq after DRIVER_OK
...
Make use of the flex_array_size() helper to calculate the size of a
flexible array member within an enclosing structure.
This helper offers defense-in-depth against potential integer
overflows, while at the same time makes it explicitly clear that
we are dealing with a flexible array member.
Signed-off-by: Gustavo A. R. Silva <gustavoars@kernel.org>
Link: https://lore.kernel.org/r/20200731130956.GA30525@embeddedor
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Since commit 76ebbe78f7 ("locking/barriers: Add implicit
smp_read_barrier_depends() to READ_ONCE()"), there is no need to use
smp_read_barrier_depends() outside of the Alpha architecture code.
Unfortunately, there is precisely _one_ user in the vhost code, and
there isn't an obvious READ_ONCE() access making the barrier
redundant. However, on closer inspection (thanks, Jason), it appears
that vring synchronisation between the producer and consumer occurs via
the 'avail_idx' field, which is followed up by an rmb() in
vhost_get_vq_desc(), making the read_barrier_depends() redundant on
Alpha.
Jason says:
| I'm also confused about the barrier here, basically in driver side
| we did:
|
| 1) allocate pages
| 2) store pages in indirect->addr
| 3) smp_wmb()
| 4) increase the avail idx (somehow a tail pointer of vring)
|
| in vhost we did:
|
| 1) read avail idx
| 2) smp_rmb()
| 3) read indirect->addr
| 4) read from indirect->addr
|
| It looks to me even the data dependency barrier is not necessary
| since we have rmb() which is sufficient for us to the correct
| indirect->addr and driver are not expected to do any writing to
| indirect->addr after avail idx is increased
Remove the redundant barrier invocation.
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Suggested-by: Jason Wang <jasowang@redhat.com>
Acked-by: Paul E. McKenney <paulmck@kernel.org>
Signed-off-by: Will Deacon <will@kernel.org>
Merge some more updates from Andrew Morton:
- various hotfixes and minor things
- hch's use_mm/unuse_mm clearnups
Subsystems affected by this patch series: mm/hugetlb, scripts, kcov,
lib, nilfs, checkpatch, lib, mm/debug, ocfs2, lib, misc.
* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
kernel: set USER_DS in kthread_use_mm
kernel: better document the use_mm/unuse_mm API contract
kernel: move use_mm/unuse_mm to kthread.c
kernel: move use_mm/unuse_mm to kthread.c
stacktrace: cleanup inconsistent variable type
lib: test get_count_order/long in test_bitops.c
mm: add comments on pglist_data zones
ocfs2: fix spelling mistake and grammar
mm/debug_vm_pgtable: fix kernel crash by checking for THP support
lib: fix bitmap_parse() on 64-bit big endian archs
checkpatch: correct check for kernel parameters doc
nilfs2: fix null pointer dereference at nilfs_segctor_do_construct()
lib/lz4/lz4_decompress.c: document deliberate use of `&'
kcov: check kcov_softirq in kcov_remote_stop()
scripts/spelling: add a few more typos
khugepaged: selftests: fix timeout condition in wait_for_scan()
Pull virtio updates from Michael Tsirkin:
- virtio-mem: paravirtualized memory hotplug
- support doorbell mapping for vdpa
- config interrupt support in ifc
- fixes all over the place
* tag 'for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost: (40 commits)
vhost/test: fix up after API change
virtio_mem: convert device block size into 64bit
virtio-mem: drop unnecessary initialization
ifcvf: implement config interrupt in IFCVF
vhost: replace -1 with VHOST_FILE_UNBIND in ioctls
vhost_vdpa: Support config interrupt in vdpa
ifcvf: ignore continuous setting same status value
virtio-mem: Don't rely on implicit compiler padding for requests
virtio-mem: Try to unplug the complete online memory block first
virtio-mem: Use -ETXTBSY as error code if the device is busy
virtio-mem: Unplug subblocks right-to-left
virtio-mem: Drop manual check for already present memory
virtio-mem: Add parent resource for all added "System RAM"
virtio-mem: Better retry handling
virtio-mem: Offline and remove completely unplugged memory blocks
mm/memory_hotplug: Introduce offline_and_remove_memory()
virtio-mem: Allow to offline partially unplugged memory blocks
mm: Allow to offline unmovable PageOffline() pages via MEM_GOING_OFFLINE
virtio-mem: Paravirtualized memory hotunplug part 2
virtio-mem: Paravirtualized memory hotunplug part 1
...
For the device that doesn't use vhost worker and use_mm(), mmget() is
too heavy weight and it may brings troubles for implementing mmap()
support for vDPA device.
This is because, an reference to the address space was held via
mm_get() in vhost_dev_set_owner() and an reference to the file was
held in mmap(). This means when process exits, the mm can not be
released thus we can not release the file.
This patch tries to use mmgrab() instead of mmget(), which allows the
address space to be destroy in process exit without releasing the mm
structure itself. This is sufficient for vDPA device which pin user
pages and does not depend on the address space to work.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200529080303.15449-3-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
vDPA device currently relays the eventfd via vhost worker. This is
inefficient due the latency of wakeup and scheduling, so this patch
tries to introduce a use_worker attribute for the vhost device. When
use_worker is not set with vhost_dev_init(), vhost won't try to
allocate a worker thread and the vhost_poll will be processed directly
in the wakeup function.
This help for vDPA since it reduces the latency caused by vhost worker.
In my testing, it saves 0.2 ms in pings between VMs on a mutual host.
Signed-off-by: Zhu Lingshan <lingshan.zhu@intel.com>
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200529080303.15449-2-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The ring element addresses are passed between components with different
alignments assumptions. Thus, if guest/userspace selects a pointer and
host then gets and dereferences it, we might need to decrease the
compiler-selected alignment to prevent compiler on the host from
assuming pointer is aligned.
This actually triggers on ARM with -mabi=apcs-gnu - which is a
deprecated configuration, but it seems safer to handle this
generally.
Note that userspace that allocates the memory is actually OK and does
not need to be fixed, but userspace that gets it from guest or another
process does need to be fixed. The later doesn't generally talk to the
kernel so while it might be buggy it's not talking to the kernel in the
buggy way - it's just using the header in the buggy way - so fixing
header and asking userspace to recompile is the best we can do.
I verified that the produced kernel binary on x86 is exactly identical
before and after the change.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Acked-by: Jason Wang <jasowang@redhat.com>
sparse warns about converting void * to void __user *. This is not new
but only got noticed now that vhost is built on more systems.
This is just a question of __user tags missing in a couple of places,
so fix it up.
Fixes: f889491380 ("vhost: introduce O(1) vq metadata cache")
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This patch factors out IOTLB into a dedicated module in order to be
reused by other modules like vringh. User may choose to enable the
automatic retiring by specifying VHOST_IOTLB_FLAG_RETIRE flag to fit
for the case of vhost device IOTLB implementation.
Signed-off-by: Jason Wang <jasowang@redhat.com>
Link: https://lore.kernel.org/r/20200326140125.19794-4-jasowang@redhat.com
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
This reverts commit a89db445fb.
I was hasty to include this patch, and it breaks the build on 32 bit.
Defence in depth is good but let's do it properly.
Cc: stable@vger.kernel.org
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
The code assumes log_num < in_num everywhere, and that is true as long as
in_num is incremented by descriptor iov count, and log_num by 1. However
this breaks if there's a zero sized descriptor.
As a result, if a malicious guest creates a vring desc with desc.len = 0,
it may cause the host kernel to crash by overflowing the log array. This
bug can be triggered during the VM migration.
There's no need to log when desc.len = 0, so just don't increment log_num
in this case.
Fixes: 3a4d5c94e9 ("vhost_net: a kernel-level virtio server")
Cc: stable@vger.kernel.org
Reviewed-by: Lidong Chen <lidongchen@tencent.com>
Signed-off-by: ruippan <ruippan@tencent.com>
Signed-off-by: yongduan <yongduan@tencent.com>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Tyler Hicks <tyhicks@canonical.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
iovec addresses coming from vhost are assumed to be
pre-validated, but in fact can be speculated to a value
out of range.
Userspace address are later validated with array_index_nospec so we can
be sure kernel info does not leak through these addresses, but vhost
must also not leak userspace info outside the allowed memory table to
guests.
Following the defence in depth principle, make sure
the address is not validated out of node range.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Cc: stable@vger.kernel.org
Acked-by: Jason Wang <jasowang@redhat.com>
Tested-by: Jason Wang <jasowang@redhat.com>
This reverts commit 7f466032dc ("vhost: access vq metadata through
kernel virtual address"). The commit caused a bunch of issues, and
while commit 73f628ec9e ("vhost: disable metadata prefetch
optimization") disabled the optimization it's not nice to keep lots of
dead code around.
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
It is unnecessary to use ret variable to return the error
code, just return the error code directly.
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
Pull Documentation updates from Jonathan Corbet:
"It's been a relatively busy cycle for docs:
- A fair pile of RST conversions, many from Mauro. These create more
than the usual number of simple but annoying merge conflicts with
other trees, unfortunately. He has a lot more of these waiting on
the wings that, I think, will go to you directly later on.
- A new document on how to use merges and rebases in kernel repos,
and one on Spectre vulnerabilities.
- Various improvements to the build system, including automatic
markup of function() references because some people, for reasons I
will never understand, were of the opinion that
:c:func:``function()`` is unattractive and not fun to type.
- We now recommend using sphinx 1.7, but still support back to 1.4.
- Lots of smaller improvements, warning fixes, typo fixes, etc"
* tag 'docs-5.3' of git://git.lwn.net/linux: (129 commits)
docs: automarkup.py: ignore exceptions when seeking for xrefs
docs: Move binderfs to admin-guide
Disable Sphinx SmartyPants in HTML output
doc: RCU callback locks need only _bh, not necessarily _irq
docs: format kernel-parameters -- as code
Doc : doc-guide : Fix a typo
platform: x86: get rid of a non-existent document
Add the RCU docs to the core-api manual
Documentation: RCU: Add TOC tree hooks
Documentation: RCU: Rename txt files to rst
Documentation: RCU: Convert RCU UP systems to reST
Documentation: RCU: Convert RCU linked list to reST
Documentation: RCU: Convert RCU basic concepts to reST
docs: filesystems: Remove uneeded .rst extension on toctables
scripts/sphinx-pre-install: fix out-of-tree build
docs: zh_CN: submitting-drivers.rst: Remove a duplicated Documentation/
Documentation: PGP: update for newer HW devices
Documentation: Add section about CPU vulnerabilities for Spectre
Documentation: platform: Delete x86-laptop-drivers.txt
docs: Note that :c:func: should no longer be used
...
It was noticed that the copy_to/from_user() friends that was used to
access virtqueue metdata tends to be very expensive for dataplane
implementation like vhost since it involves lots of software checks,
speculation barriers, hardware feature toggling (e.g SMAP). The
extra cost will be more obvious when transferring small packets since
the time spent on metadata accessing become more significant.
This patch tries to eliminate those overheads by accessing them
through direct mapping of those pages. Invalidation callbacks is
implemented for co-operation with general VM management (swap, KSM,
THP or NUMA balancing). We will try to get the direct mapping of vq
metadata before each round of packet processing if it doesn't
exist. If we fail, we will simplely fallback to copy_to/from_user()
friends.
This invalidation and direct mapping access are synchronized through
spinlock and RCU. All matedata accessing through direct map is
protected by RCU, and the setup or invalidation are done under
spinlock.
This method might does not work for high mem page which requires
temporary mapping so we just fallback to normal
copy_to/from_user() and may not for arch that has virtual tagged cache
since extra cache flushing is needed to eliminate the alias. This will
result complex logic and bad performance. For those archs, this patch
simply go for copy_to/from_user() friends. This is done by ruling out
kernel mapping codes through ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE.
Note that this is only done when device IOTLB is not enabled. We
could use similar method to optimize IOTLB in the future.
Tests shows at most about 23% improvement on TX PPS when using
virtio-user + vhost_net + xdp1 + TAP on 2.6GHz Broadwell:
SMAP on | SMAP off
Before: 5.2Mpps | 7.1Mpps
After: 6.4Mpps | 8.2Mpps
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: James Bottomley <James.Bottomley@hansenpartnership.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David Miller <davem@davemloft.net>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: linux-mm@kvack.org
Cc: linux-arm-kernel@lists.infradead.org
Cc: linux-parisc@vger.kernel.org
Signed-off-by: Jason Wang <jasowang@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
