* [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize
@ 2026-10-23 16:19 Denis V. Lunev
2026-10-23 16:19 ` [PATCH 1/1] " Denis V. Lunev
` (2 more replies)
0 siblings, 3 replies; 5+ messages in thread
From: Denis V. Lunev @ 2026-10-23 16:19 UTC (permalink / raw)
To: qemu-devel; +Cc: den, Paolo Bonzini, Fam Zheng, Michael S. Tsirkin
A disk or CD-ROM hot-plugged into virtio-scsi sometimes never shows up
in the guest. QEMU has the device, the guest has consumed the RESCAN
event, and a manual bus rescan finds the LUN at once. We see this in CI
with Windows Server guests (vioscsi) and with Linux guests, always with
the controller on an iothread.
The event is pushed from the plug() callback, before the device is
realized, and the iothread request path does not see unrealized devices
since 8ddf958e8d62. A guest that scans right away gets BAD_TARGET and
does not retry. The patch moves the event to a main-loop bottom half.
The alternative, letting the request path see a plugged but unrealized
device, would undo what 8ddf958e8d62 protects against.
There is no test in the patch. The window lies inside a single
device_add under the BQL, and a qtest or iotests client is served by the
same main loop, so it cannot send a request in between. To reproduce, I
booted a Linux guest with virtio-scsi on an iothread, ran QEMU under gdb
in non-stop mode and stalled the main thread for 3 s in
scsi_bus_set_ua(), right after the event is pushed, then hot-plugged a
scsi-cd:
break scsi_bus_set_ua
commands
call (int)usleep(3000000)
continue
end
Without the patch the guest never enumerated the LUN (3 of 3 runs).
With the patch the same stall falls after realize and the LUN appeared
every time (3 of 3). Without the stall both builds pass, which is why
this only shows up on loaded hosts. tests/qtest virtio-scsi passes.
Every release since 5.2 has the bug, hence qemu-stable.
Signed-off-by: Denis V. Lunev <den@openvz.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Fam Zheng <fam@euphon.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
Denis V. Lunev (1):
virtio-scsi: fix hot-plug event racing device realize
hw/scsi/virtio-scsi.c | 38 +++++++++++++++++++++++++++-----------
1 file changed, 27 insertions(+), 11 deletions(-)
base-commit: b6c4b7e86dec12d07f5cac65265b986d38ee9002
--
2.53.0
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH 1/1] virtio-scsi: fix hot-plug event racing device realize
2026-10-23 16:19 [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize Denis V. Lunev
@ 2026-10-23 16:19 ` Denis V. Lunev
2026-11-02 23:51 ` [PATCH 0/1] " Denis V. Lunev
2026-11-17 19:53 ` Denis V. Lunev
2 siblings, 0 replies; 5+ messages in thread
From: Denis V. Lunev @ 2026-10-23 16:19 UTC (permalink / raw)
To: qemu-devel
Cc: den, qemu-stable, Ilya Rudakov, Paolo Bonzini, Fam Zheng,
Michael S. Tsirkin
From: Denis V. Lunev <den@openvz.org>
virtio_scsi_hotplug() is the HotplugHandler plug() callback, which
device_set_realized() calls before it sets dev->realized. The callback
pushes VIRTIO_SCSI_T_TRANSPORT_RESET/RESCAN to the guest at once, but
until realized is set the request path ignores the new device:
virtio_scsi_device_get() -> scsi_device_get() skips unrealized devices,
and so does REPORT LUNS. A guest that reacts to the event quickly, on a
vCPU while the request queue is served by an iothread, scans the target
and gets VIRTIO_SCSI_S_BAD_TARGET.
Neither Linux nor Windows retries that scan, so the hot-plugged disk or
CD-ROM stays invisible until somebody rescans the bus by hand. Linux
calls scsi_scan_target() once from the event handler. Windows storport
clears its rescan flag before enumerating the bus and never sets it
again for a failed target.
Push the event, and the REPORTED LUNS CHANGED unit attention that goes
with it, from a main-loop bottom half instead. It runs under the BQL,
which device_add holds until device_set_realized() returns, so the guest
is only told about a device it can reach. The iothread context would not
do: its bottom half could run before realized is set, and the realized
check and scsi_bus_set_ua() rely on the BQL as plug() did. The bottom
half is only scheduled when the guest negotiated VIRTIO_SCSI_F_HOTPLUG,
as before, so cold-plugged devices do not get one. It still checks that
the device is realized: nothing orders it against a device_del that
follows device_add, and an unplugged device has no parent bus.
Fixes: 8ddf958e8d62 ("scsi/scsi-bus: scsi_device_find: don't return unrealized devices")
Cc: qemu-stable@nongnu.org
Reported-by: Ilya Rudakov <irudakov@virtuozzo.com>
Signed-off-by: Denis V. Lunev <den@openvz.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Fam Zheng <fam@euphon.net>
Cc: "Michael S. Tsirkin" <mst@redhat.com>
---
hw/scsi/virtio-scsi.c | 38 +++++++++++++++++++++++++++-----------
1 file changed, 27 insertions(+), 11 deletions(-)
diff --git a/hw/scsi/virtio-scsi.c b/hw/scsi/virtio-scsi.c
index d7e2662aa7..9ea1acb84c 100644
--- a/hw/scsi/virtio-scsi.c
+++ b/hw/scsi/virtio-scsi.c
@@ -1151,6 +1151,29 @@ static void virtio_scsi_pre_hotplug(const HotplugHandler *hotplug_dev,
sd->hba_supports_iothread = true;
}
+static void virtio_scsi_hotplug_notify_bh(void *opaque)
+{
+ SCSIDevice *sd = opaque;
+ DeviceState *dev = DEVICE(sd);
+
+ /* Nothing orders us against a following device_del */
+ if (qdev_is_realized(dev)) {
+ VirtIOSCSI *s = VIRTIO_SCSI(dev->parent_bus->parent);
+ VirtIOSCSIEventInfo info = {
+ .event = VIRTIO_SCSI_T_TRANSPORT_RESET,
+ .reason = VIRTIO_SCSI_EVT_RESET_RESCAN,
+ .address = {
+ .id = sd->id,
+ .lun = sd->lun,
+ },
+ };
+
+ virtio_scsi_push_event(s, &info);
+ scsi_bus_set_ua(&s->bus, SENSE_CODE(REPORTED_LUNS_CHANGED));
+ }
+ object_unref(OBJECT(sd));
+}
+
static void virtio_scsi_hotplug(const HotplugHandler *hotplug_dev, DeviceState *dev,
Error **errp)
{
@@ -1169,17 +1192,10 @@ static void virtio_scsi_hotplug(const HotplugHandler *hotplug_dev, DeviceState *
}
if (virtio_vdev_has_feature(vdev, VIRTIO_SCSI_F_HOTPLUG)) {
- VirtIOSCSIEventInfo info = {
- .event = VIRTIO_SCSI_T_TRANSPORT_RESET,
- .reason = VIRTIO_SCSI_EVT_RESET_RESCAN,
- .address = {
- .id = sd->id,
- .lun = sd->lun,
- },
- };
-
- virtio_scsi_push_event(s, &info);
- scsi_bus_set_ua(&s->bus, SENSE_CODE(REPORTED_LUNS_CHANGED));
+ /* The request path ignores @sd until it is realized, so notify after */
+ object_ref(OBJECT(sd));
+ aio_bh_schedule_oneshot(qemu_get_aio_context(),
+ virtio_scsi_hotplug_notify_bh, sd);
}
}
base-commit: b6c4b7e86dec12d07f5cac65265b986d38ee9002
--
2.53.0
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize
2026-10-23 16:19 [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize Denis V. Lunev
2026-10-23 16:19 ` [PATCH 1/1] " Denis V. Lunev
@ 2026-11-02 23:51 ` Denis V. Lunev
2026-11-17 19:53 ` Denis V. Lunev
2 siblings, 0 replies; 5+ messages in thread
From: Denis V. Lunev @ 2026-11-02 23:51 UTC (permalink / raw)
To: Denis V. Lunev, qemu-devel; +Cc: Paolo Bonzini, Fam Zheng, Michael S. Tsirkin
On 10/23/26 18:19, Denis V. Lunev wrote:
> This email originated from an IP that might not be authorized by the domain it was sent from.
> Do not click links or open attachments unless it is an email you expected to receive.
> A disk or CD-ROM hot-plugged into virtio-scsi sometimes never shows up
> in the guest. QEMU has the device, the guest has consumed the RESCAN
> event, and a manual bus rescan finds the LUN at once. We see this in CI
> with Windows Server guests (vioscsi) and with Linux guests, always with
> the controller on an iothread.
>
> The event is pushed from the plug() callback, before the device is
> realized, and the iothread request path does not see unrealized devices
> since 8ddf958e8d62. A guest that scans right away gets BAD_TARGET and
> does not retry. The patch moves the event to a main-loop bottom half.
> The alternative, letting the request path see a plugged but unrealized
> device, would undo what 8ddf958e8d62 protects against.
>
> There is no test in the patch. The window lies inside a single
> device_add under the BQL, and a qtest or iotests client is served by the
> same main loop, so it cannot send a request in between. To reproduce, I
> booted a Linux guest with virtio-scsi on an iothread, ran QEMU under gdb
> in non-stop mode and stalled the main thread for 3 s in
> scsi_bus_set_ua(), right after the event is pushed, then hot-plugged a
> scsi-cd:
>
> break scsi_bus_set_ua
> commands
> call (int)usleep(3000000)
> continue
> end
>
> Without the patch the guest never enumerated the LUN (3 of 3 runs).
> With the patch the same stall falls after realize and the LUN appeared
> every time (3 of 3). Without the stall both builds pass, which is why
> this only shows up on loaded hosts. tests/qtest virtio-scsi passes.
>
> Every release since 5.2 has the bug, hence qemu-stable.
>
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Fam Zheng <fam@euphon.net>
> Cc: "Michael S. Tsirkin" <mst@redhat.com>
>
> Denis V. Lunev (1):
> virtio-scsi: fix hot-plug event racing device realize
>
> hw/scsi/virtio-scsi.c | 38 +++++++++++++++++++++++++++-----------
> 1 file changed, 27 insertions(+), 11 deletions(-)
>
>
> base-commit: b6c4b7e86dec12d07f5cac65265b986d38ee9002
ping
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize
2026-10-23 16:19 [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize Denis V. Lunev
2026-10-23 16:19 ` [PATCH 1/1] " Denis V. Lunev
2026-11-02 23:51 ` [PATCH 0/1] " Denis V. Lunev
@ 2026-11-17 19:53 ` Denis V. Lunev
2026-11-19 7:57 ` Michael S. Tsirkin
2 siblings, 1 reply; 5+ messages in thread
From: Denis V. Lunev @ 2026-11-17 19:53 UTC (permalink / raw)
To: Denis V. Lunev, qemu-devel; +Cc: Paolo Bonzini, Fam Zheng, Michael S. Tsirkin
On 10/23/26 18:19, Denis V. Lunev wrote:
> This email originated from an IP that might not be authorized by the domain it was sent from.
> Do not click links or open attachments unless it is an email you expected to receive.
> A disk or CD-ROM hot-plugged into virtio-scsi sometimes never shows up
> in the guest. QEMU has the device, the guest has consumed the RESCAN
> event, and a manual bus rescan finds the LUN at once. We see this in CI
> with Windows Server guests (vioscsi) and with Linux guests, always with
> the controller on an iothread.
>
> The event is pushed from the plug() callback, before the device is
> realized, and the iothread request path does not see unrealized devices
> since 8ddf958e8d62. A guest that scans right away gets BAD_TARGET and
> does not retry. The patch moves the event to a main-loop bottom half.
> The alternative, letting the request path see a plugged but unrealized
> device, would undo what 8ddf958e8d62 protects against.
>
> There is no test in the patch. The window lies inside a single
> device_add under the BQL, and a qtest or iotests client is served by the
> same main loop, so it cannot send a request in between. To reproduce, I
> booted a Linux guest with virtio-scsi on an iothread, ran QEMU under gdb
> in non-stop mode and stalled the main thread for 3 s in
> scsi_bus_set_ua(), right after the event is pushed, then hot-plugged a
> scsi-cd:
>
> break scsi_bus_set_ua
> commands
> call (int)usleep(3000000)
> continue
> end
>
> Without the patch the guest never enumerated the LUN (3 of 3 runs).
> With the patch the same stall falls after realize and the LUN appeared
> every time (3 of 3). Without the stall both builds pass, which is why
> this only shows up on loaded hosts. tests/qtest virtio-scsi passes.
>
> Every release since 5.2 has the bug, hence qemu-stable.
>
> Signed-off-by: Denis V. Lunev <den@openvz.org>
> Cc: Paolo Bonzini <pbonzini@redhat.com>
> Cc: Fam Zheng <fam@euphon.net>
> Cc: "Michael S. Tsirkin" <mst@redhat.com>
>
> Denis V. Lunev (1):
> virtio-scsi: fix hot-plug event racing device realize
>
> hw/scsi/virtio-scsi.c | 38 +++++++++++++++++++++++++++-----------
> 1 file changed, 27 insertions(+), 11 deletions(-)
>
>
> base-commit: b6c4b7e86dec12d07f5cac65265b986d38ee9002
ping
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize
2026-11-17 19:53 ` Denis V. Lunev
@ 2026-11-19 7:57 ` Michael S. Tsirkin
0 siblings, 0 replies; 5+ messages in thread
From: Michael S. Tsirkin @ 2026-11-19 7:57 UTC (permalink / raw)
To: Denis V. Lunev; +Cc: Denis V. Lunev, qemu-devel, Paolo Bonzini, Fam Zheng
On Tue, Nov 17, 2026 at 08:53:44PM +0100, Denis V. Lunev wrote:
> On 10/23/26 18:19, Denis V. Lunev wrote:
> > This email originated from an IP that might not be authorized by the domain it was sent from.
> > Do not click links or open attachments unless it is an email you expected to receive.
> > A disk or CD-ROM hot-plugged into virtio-scsi sometimes never shows up
> > in the guest. QEMU has the device, the guest has consumed the RESCAN
> > event, and a manual bus rescan finds the LUN at once. We see this in CI
> > with Windows Server guests (vioscsi) and with Linux guests, always with
> > the controller on an iothread.
> >
> > The event is pushed from the plug() callback, before the device is
> > realized, and the iothread request path does not see unrealized devices
> > since 8ddf958e8d62. A guest that scans right away gets BAD_TARGET and
> > does not retry. The patch moves the event to a main-loop bottom half.
> > The alternative, letting the request path see a plugged but unrealized
> > device, would undo what 8ddf958e8d62 protects against.
> >
> > There is no test in the patch. The window lies inside a single
> > device_add under the BQL, and a qtest or iotests client is served by the
> > same main loop, so it cannot send a request in between. To reproduce, I
> > booted a Linux guest with virtio-scsi on an iothread, ran QEMU under gdb
> > in non-stop mode and stalled the main thread for 3 s in
> > scsi_bus_set_ua(), right after the event is pushed, then hot-plugged a
> > scsi-cd:
> >
> > break scsi_bus_set_ua
> > commands
> > call (int)usleep(3000000)
> > continue
> > end
> >
> > Without the patch the guest never enumerated the LUN (3 of 3 runs).
> > With the patch the same stall falls after realize and the LUN appeared
> > every time (3 of 3). Without the stall both builds pass, which is why
> > this only shows up on loaded hosts. tests/qtest virtio-scsi passes.
> >
> > Every release since 5.2 has the bug, hence qemu-stable.
> >
> > Signed-off-by: Denis V. Lunev <den@openvz.org>
> > Cc: Paolo Bonzini <pbonzini@redhat.com>
> > Cc: Fam Zheng <fam@euphon.net>
> > Cc: "Michael S. Tsirkin" <mst@redhat.com>
> >
> > Denis V. Lunev (1):
> > virtio-scsi: fix hot-plug event racing device realize
> >
> > hw/scsi/virtio-scsi.c | 38 +++++++++++++++++++++++++++-----------
> > 1 file changed, 27 insertions(+), 11 deletions(-)
> >
> >
> > base-commit: b6c4b7e86dec12d07f5cac65265b986d38ee9002
> ping
pong. I'm not applying patches I'm not CC'd on. Paolo, wanna take
a look?
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-11-19 7:57 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-10-23 16:19 [PATCH 0/1] virtio-scsi: fix hot-plug event racing device realize Denis V. Lunev
2026-10-23 16:19 ` [PATCH 1/1] " Denis V. Lunev
2026-11-02 23:51 ` [PATCH 0/1] " Denis V. Lunev
2026-11-17 19:53 ` Denis V. Lunev
2026-11-19 7:57 ` Michael S. Tsirkin
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.