在 2022/4/6 上午3:18, Si-Wei Liu 写道:
On 4/1/2022 7:00 PM, Jason Wang wrote:
On Sat, Apr 2, 2022 at 4:37 AM Si-Wei Liu <si-wei....@oracle.com> wrote:
On 3/31/2022 1:36 AM, Jason Wang wrote:
On Thu, Mar 31, 2022 at 12:41 AM Si-Wei Liu <si-wei....@oracle.com>
wrote:
On 3/30/2022 2:14 AM, Jason Wang wrote:
On Wed, Mar 30, 2022 at 2:33 PM Si-Wei Liu
<si-wei....@oracle.com> wrote:
Previous commit prevents vhost-user and vhost-vdpa from using
userland vq handler via disable_ioeventfd_handler. The same
needs to be done for host notifier cleanup too, as the
virtio_queue_host_notifier_read handler still tends to read
pending event left behind on ioeventfd and attempts to handle
outstanding kicks from QEMU userland vq.
If vq handler is not disabled on cleanup, it may lead to sigsegv
with recursive virtio_net_set_status call on the control vq:
0 0x00007f8ce3ff3387 in raise () at /lib64/libc.so.6
1 0x00007f8ce3ff4a78 in abort () at /lib64/libc.so.6
2 0x00007f8ce3fec1a6 in __assert_fail_base () at /lib64/libc.so.6
3 0x00007f8ce3fec252 in () at /lib64/libc.so.6
4 0x0000558f52d79421 in vhost_vdpa_get_vq_index (dev=<optimized
out>, idx=<optimized out>) at ../hw/virtio/vhost-vdpa.c:563
5 0x0000558f52d79421 in vhost_vdpa_get_vq_index (dev=<optimized
out>, idx=<optimized out>) at ../hw/virtio/vhost-vdpa.c:558
6 0x0000558f52d7329a in vhost_virtqueue_mask
(hdev=0x558f55c01800, vdev=0x558f568f91f0, n=2, mask=<optimized
out>) at ../hw/virtio/vhost.c:1557
I feel it's probably a bug elsewhere e.g when we fail to start
vhost-vDPA, it's the charge of the Qemu to poll host notifier and we
will fallback to the userspace vq handler.
Apologies, an incorrect stack trace was pasted which actually came
from
patch #1. I will post a v2 with the corresponding one as below:
0 0x000055f800df1780 in qdev_get_parent_bus (dev=0x0) at
../hw/core/qdev.c:376
1 0x000055f800c68ad8 in virtio_bus_device_iommu_enabled
(vdev=vdev@entry=0x0) at ../hw/virtio/virtio-bus.c:331
2 0x000055f800d70d7f in vhost_memory_unmap (dev=<optimized out>) at
../hw/virtio/vhost.c:318
3 0x000055f800d70d7f in vhost_memory_unmap (dev=<optimized out>,
buffer=0x7fc19bec5240, len=2052, is_write=1, access_len=2052) at
../hw/virtio/vhost.c:336
4 0x000055f800d71867 in vhost_virtqueue_stop
(dev=dev@entry=0x55f8037ccc30, vdev=vdev@entry=0x55f8044ec590,
vq=0x55f8037cceb0, idx=0) at ../hw/virtio/vhost.c:1241
5 0x000055f800d7406c in vhost_dev_stop
(hdev=hdev@entry=0x55f8037ccc30,
vdev=vdev@entry=0x55f8044ec590) at ../hw/virtio/vhost.c:1839
6 0x000055f800bf00a7 in vhost_net_stop_one (net=0x55f8037ccc30,
dev=0x55f8044ec590) at ../hw/net/vhost_net.c:315
7 0x000055f800bf0678 in vhost_net_stop
(dev=dev@entry=0x55f8044ec590,
ncs=0x55f80452bae0, data_queue_pairs=data_queue_pairs@entry=7,
cvq=cvq@entry=1)
at ../hw/net/vhost_net.c:423
8 0x000055f800d4e628 in virtio_net_set_status (status=<optimized
out>,
n=0x55f8044ec590) at ../hw/net/virtio-net.c:296
9 0x000055f800d4e628 in virtio_net_set_status
(vdev=vdev@entry=0x55f8044ec590, status=15 '\017') at
../hw/net/virtio-net.c:370
I don't understand why virtio_net_handle_ctrl() call
virtio_net_set_stauts()...
The pending request left over on the ctrl vq was a VIRTIO_NET_CTRL_MQ
command, i.e. in virtio_net_handle_mq():
Completely forget that the code was actually written by me :\
1413 n->curr_queue_pairs = queue_pairs;
1414 /* stop the backend before changing the number of queue_pairs
to avoid handling a
1415 * disabled queue */
1416 virtio_net_set_status(vdev, vdev->status);
1417 virtio_net_set_queue_pairs(n);
Noted before the vdpa multiqueue support, there was never a vhost_dev
for ctrl_vq exposed, i.e. there's no host notifier set up for the
ctrl_vq on vhost_kernel as it is emulated in QEMU software.
10 0x000055f800d534d8 in virtio_net_handle_ctrl (iov_cnt=<optimized
out>, iov=<optimized out>, cmd=0 '\000', n=0x55f8044ec590) at
../hw/net/virtio-net.c:1408
11 0x000055f800d534d8 in virtio_net_handle_ctrl (vdev=0x55f8044ec590,
vq=0x7fc1a7e888d0) at ../hw/net/virtio-net.c:1452
12 0x000055f800d69f37 in virtio_queue_host_notifier_read
(vq=0x7fc1a7e888d0) at ../hw/virtio/virtio.c:2331
13 0x000055f800d69f37 in virtio_queue_host_notifier_read
(n=n@entry=0x7fc1a7e8894c) at ../hw/virtio/virtio.c:3575
14 0x000055f800c688e6 in virtio_bus_cleanup_host_notifier
(bus=<optimized out>, n=n@entry=14) at ../hw/virtio/virtio-bus.c:312
15 0x000055f800d73106 in vhost_dev_disable_notifiers
(hdev=hdev@entry=0x55f8035b51b0, vdev=vdev@entry=0x55f8044ec590)
at ../../../include/hw/virtio/virtio-bus.h:35
16 0x000055f800bf00b2 in vhost_net_stop_one (net=0x55f8035b51b0,
dev=0x55f8044ec590) at ../hw/net/vhost_net.c:316
17 0x000055f800bf0678 in vhost_net_stop
(dev=dev@entry=0x55f8044ec590,
ncs=0x55f80452bae0, data_queue_pairs=data_queue_pairs@entry=7,
cvq=cvq@entry=1)
at ../hw/net/vhost_net.c:423
18 0x000055f800d4e628 in virtio_net_set_status (status=<optimized
out>,
n=0x55f8044ec590) at ../hw/net/virtio-net.c:296
19 0x000055f800d4e628 in virtio_net_set_status (vdev=0x55f8044ec590,
status=15 '\017') at ../hw/net/virtio-net.c:370
20 0x000055f800d6c4b2 in virtio_set_status (vdev=0x55f8044ec590,
val=<optimized out>) at ../hw/virtio/virtio.c:1945
21 0x000055f800d11d9d in vm_state_notify
(running=running@entry=false,
state=state@entry=RUN_STATE_SHUTDOWN) at ../softmmu/runstate.c:333
22 0x000055f800d04e7a in do_vm_stop
(state=state@entry=RUN_STATE_SHUTDOWN,
send_stop=send_stop@entry=false)
at ../softmmu/cpus.c:262
23 0x000055f800d04e99 in vm_shutdown () at ../softmmu/cpus.c:280
24 0x000055f800d126af in qemu_cleanup () at ../softmmu/runstate.c:812
25 0x000055f800ad5b13 in main (argc=<optimized out>, argv=<optimized
out>, envp=<optimized out>) at ../softmmu/main.c:51
From the trace pending read only occurs in stop path. The
recursive
virtio_net_set_status from virtio_net_handle_ctrl doesn't make
sense to me.
Yes, we need to figure this out to know the root cause.
I think it has something to do with the virtqueue unready issue that
the
vhost_reset_device() refactoring series attempt to fix. If that is
fixed
we should not see this sigsegv with mlx5_vdpa. However I guess we both
agreed that the vq_unready support would need new uAPI (some flag) to
define, hence this fix applies to the situation where uAPI doesn't
exist
on the kernel or the vq_unready is not supported by vdpa vendor driver.
Yes.
The code should work for the case when vhost-vdp fails to start.
Unlike the other datapath queues for net vdpa, the events left
behind in
the control queue can't be processed by the userspace, as unlike
vhost-kernel we don't have a fallback path in the userspace.
So that's the question, we should have a safe fallback.
To ignore
the pending event and let vhost vdpa process it on resume/start is
perhaps the best thing to do. This is even true for datapath queues for
other vdpa devices than of network.
Not sure I got the reason why we need to handle pending host
notification in userland vq, can you elaborate?
Because vhost-vDPA fails to start, we will "fallback" to a dummy
userspace.
Is the dummy userspace working or operational? What's the use case of
this "fallback" dummy if what guest user can get is a busted NIC?
The problem is we can't do better in this case now. Such fallack (e.g
for vhost-user) has been used for years. Or do you have any better
ideas?
In my opinion if vhost-vdpa or vhost-user fails to start, maybe we
should try to disable the device via virtio_error(), which would set
broken to true and set NEEDS_RESET in case of VERSION_1. That way the
device won't move forward further and the guest may get the indication
via config interrupt that something had gone wrong underneath. If
device reset is well supported there the guest driver would retry.
Note that the NEEDS_RESET is not implemented in the current Linux drivers.
This can at least give the backend some chance to recover if running
into intermittent error. The worst result would be the device keeps
resetting repeatedly, for which we may introduce tunable to control
the rate if seeing reset occurs too often.. Did this ever get
considered before?
I don't know, but we manage to survive with such fallback for years. We
can do this, but can virtio_error() fix the issue you describe here?
Noted that the dummy userspace can't handle any control vq command
effectively once the vhost backend fails, for e.g. how does it handle
those guest offload, rx mode, MAC or VLAN filter changes without
sending the request down to the backend?
It should be no difference compared to the real hardware. The device is
just malfunction. The driver can detect this in many ways. E.g in the
past I suggest to implement the device watchdog for virtio-net, the
prototype is posted but for some reason it was stalled. Maybe we can
consider to continue the work.
This could easily get inconsistent state to the guest if somehow we
are able to resume the virtqueue without a reset. Even so, I suspect
the device reset eventually is still needed on the other part, which
is subject to the specific failure. It looks to me at least for
vhost-vdpa, it might be the safest fallback so far to ignore pending
event in ctrl_vq, and disable the device from moving forward in case
of backend start failure.
I don't get here, if we fail to start vhost-vdpa, the Qemu should do a
safe rewind otherwise it would be a bug.
It doesn't differ too much of the two approaches:
1) dummy fallback which can do even cvq
and
2) disable host notifiers
Especially consider 2) requires more changes.
I'm not clear if 2) really needs more changes, as it seems to me that
it would take more unwarranted changes to make dummy fallback to work
on cvq? And suppose we can fallback to disabling device via
virtio_error(), we don't even need to change any code on cvq?
So let me explain my points:
1) we use dummy receive as a fallback as vhost-user
2) the code should safely fallback to that otherwise it could be a bug
elsewhere
3) if we think the dummy fallback doesn't make sense, we can improve,
but we need to figure out why we can crash for 2) since the code could
be used in other path.
On the other hand, for the specific code path this patch tries to fix,
it is not due to failure to start vhost-vdpa backend, but more of a
control flow flaw in the stop path due to lack of VQ stop uAPI. Let
alone dummy or host notifier, considering currently it's in the stop
path followed by a reset, I feel it should be pretty safe to just
ignore the pending event on the control vq rather than process it
prematurely in userspace. What do you think? I can leave without the
host notifier handler change for sure.
I wonder how vhost-user deal with this.
I
think this is very different from the vhost-kernel case in that once
vhost fails, we can fallback to userspace to emulate the network
through
the tap fd in a good way. However, there's no equivalent yet for
vhost-vdpa...
As said previously, technically we can have vhost-vDPA network backend
as a fallback.
But this is working as yet. And how do you envision the datapath may
work given that we don't have a fallback tap fd?
I mean we can treat vhost-vdpa as a kind of general networking backend
that could be used by all NIC model like e1000e. Then we can use that as
a fallback.
But I'm not sure it's worth to bother.
Thanks
-Siwei
(So did for vhost-user).
Thanks
Thanks,
-Siwei
Thanks
Thanks,
-Siwei
Thanks
7 0x0000558f52c6b89a in virtio_pci_set_guest_notifier
(d=d@entry=0x558f568f0f60, n=n@entry=2,
assign=assign@entry=true, with_irqfd=with_irqfd@entry=false)
at ../hw/virtio/virtio-pci.c:974
8 0x0000558f52c6c0d8 in virtio_pci_set_guest_notifiers
(d=0x558f568f0f60, nvqs=3, assign=true) at
../hw/virtio/virtio-pci.c:1019
9 0x0000558f52bf091d in vhost_net_start
(dev=dev@entry=0x558f568f91f0, ncs=0x558f56937cd0,
data_queue_pairs=data_queue_pairs@entry=1, cvq=cvq@entry=1)
at ../hw/net/vhost_net.c:361
10 0x0000558f52d4e5e7 in virtio_net_set_status
(status=<optimized out>, n=0x558f568f91f0) at
../hw/net/virtio-net.c:289
11 0x0000558f52d4e5e7 in virtio_net_set_status
(vdev=0x558f568f91f0, status=15 '\017') at
../hw/net/virtio-net.c:370
12 0x0000558f52d6c4b2 in virtio_set_status
(vdev=vdev@entry=0x558f568f91f0, val=val@entry=15 '\017') at
../hw/virtio/virtio.c:1945
13 0x0000558f52c69eff in virtio_pci_common_write
(opaque=0x558f568f0f60, addr=<optimized out>, val=<optimized
out>, size=<optimized out>) at ../hw/virtio/virtio-pci.c:1292
14 0x0000558f52d15d6e in memory_region_write_accessor
(mr=0x558f568f19d0, addr=20, value=<optimized out>, size=1,
shift=<optimized out>, mask=<optimized out>, attrs=...)
at ../softmmu/memory.c:492
15 0x0000558f52d127de in access_with_adjusted_size
(addr=addr@entry=20, value=value@entry=0x7f8cdbffe748,
size=size@entry=1, access_size_min=<optimized out>,
access_size_max=<optimized out>, access_fn=0x558f52d15cf0
<memory_region_write_accessor>, mr=0x558f568f19d0, attrs=...) at
../softmmu/memory.c:554
16 0x0000558f52d157ef in memory_region_dispatch_write
(mr=mr@entry=0x558f568f19d0, addr=20, data=<optimized out>,
op=<optimized out>, attrs=attrs@entry=...)
at ../softmmu/memory.c:1504
17 0x0000558f52d078e7 in flatview_write_continue
(fv=fv@entry=0x7f8accbc3b90, addr=addr@entry=103079215124,
attrs=..., ptr=ptr@entry=0x7f8ce6300028, len=len@entry=1,
addr1=<optimized out>, l=<optimized out>, mr=0x558f568f19d0) at
../../../include/qemu/host-utils.h:165
18 0x0000558f52d07b06 in flatview_write (fv=0x7f8accbc3b90,
addr=103079215124, attrs=..., buf=0x7f8ce6300028, len=1) at
../softmmu/physmem.c:2822
19 0x0000558f52d0b36b in address_space_write (as=<optimized
out>, addr=<optimized out>, attrs=...,
buf=buf@entry=0x7f8ce6300028, len=<optimized out>)
at ../softmmu/physmem.c:2914
20 0x0000558f52d0b3da in address_space_rw (as=<optimized out>,
addr=<optimized out>, attrs=...,
attrs@entry=..., buf=buf@entry=0x7f8ce6300028,
len=<optimized out>, is_write=<optimized out>) at
../softmmu/physmem.c:2924
21 0x0000558f52dced09 in kvm_cpu_exec
(cpu=cpu@entry=0x558f55c2da60) at ../accel/kvm/kvm-all.c:2903
22 0x0000558f52dcfabd in kvm_vcpu_thread_fn
(arg=arg@entry=0x558f55c2da60) at ../accel/kvm/kvm-accel-ops.c:49
23 0x0000558f52f9f04a in qemu_thread_start (args=<optimized
out>) at ../util/qemu-thread-posix.c:556
24 0x00007f8ce4392ea5 in start_thread () at /lib64/libpthread.so.0
25 0x00007f8ce40bb9fd in clone () at /lib64/libc.so.6
Fixes: 4023784 ("vhost-vdpa: multiqueue support")
Cc: Jason Wang <jasow...@redhat.com>
Signed-off-by: Si-Wei Liu <si-wei....@oracle.com>
---
hw/virtio/virtio-bus.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/hw/virtio/virtio-bus.c b/hw/virtio/virtio-bus.c
index 0f69d1c..3159b58 100644
--- a/hw/virtio/virtio-bus.c
+++ b/hw/virtio/virtio-bus.c
@@ -311,7 +311,8 @@ void
virtio_bus_cleanup_host_notifier(VirtioBusState *bus, int n)
/* Test and clear notifier after disabling event,
* in case poll callback didn't have time to run.
*/
- virtio_queue_host_notifier_read(notifier);
+ if (!vdev->disable_ioeventfd_handler)
+ virtio_queue_host_notifier_read(notifier);
event_notifier_cleanup(notifier);
}
--
1.8.3.1
