[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [Bug 1646610] Re: "Assertion `!r->req.sg' failed." during l
From: |
Thomas Huth |
Subject: |
[Qemu-devel] [Bug 1646610] Re: "Assertion `!r->req.sg' failed." during live migration with VirtIO |
Date: |
Tue, 06 Dec 2016 21:06:55 -0000 |
OK. QEMU version 2.3 is not maintained by the QEMU project any more. Can
you reproduce it with the latest version (release candidate of 2.8
preferably, or at least 2.7)?
** Changed in: qemu
Status: New => Incomplete
--
You received this bug notification because you are a member of qemu-
devel-ml, which is subscribed to QEMU.
https://bugs.launchpad.net/bugs/1646610
Title:
"Assertion `!r->req.sg' failed." during live migration with VirtIO
Status in QEMU:
Incomplete
Bug description:
We've hit this issue twice so far, but don't have an obvious repro
yet. It's pretty rare for us to hit it but I'm still trying so I can
get a core and backtrace. The guest was Windows running a constant
workload. We were using VirtIO SCSI drivers in both cases.
In both cases we hit the assert here:
hw/scsi/scsi-generic.c:
static void scsi_generic_save_request(QEMUFile *f, SCSIRequest *req)
{
SCSIGenericReq *r = DO_UPCAST(SCSIGenericReq, req, req);
qemu_put_sbe32s(f, &r->buflen);
if (r->buflen && r->req.cmd.mode == SCSI_XFER_TO_DEV) {
*** assert(!r->req.sg);
qemu_put_buffer(f, r->buf, r->req.cmd.xfer);
}
}
From code inspection, it seems that this will always happen if
scsi_req_enqueue_internal in hw/scsi/scsi-bus.c is ever invoked.
static void scsi_req_enqueue_internal(SCSIRequest *req)
{
assert(!req->enqueued);
scsi_req_ref(req);
if (req->bus->info->get_sg_list) {
req->sg = req->bus->info->get_sg_list(req);
} else {
req->sg = NULL;
}
req->enqueued = true;
QTAILQ_INSERT_TAIL(&req->dev->requests, req, next);
}
req->bus->info->get_sg_list will return &req->qsgl for a virtio-scsi
bus since it's a method stored on the SCSIBusInfo struct. I didn't see
anything that would clear the req->sg if scsi_req_enqueue_internal is
called at least once.
I think this can only happen if scsi_dma_restart_bh in hw/scsi/scsi-
bus.c is called. The only other location I see
scsi_req_enqueue_internal is on the load side for the destination of a
migration.
static void scsi_dma_restart_bh(void *opaque)
{
SCSIDevice *s = opaque;
SCSIRequest *req, *next;
qemu_bh_delete(s->bh);
s->bh = NULL;
QTAILQ_FOREACH_SAFE(req, &s->requests, next, next) {
scsi_req_ref(req);
if (req->retry) {
req->retry = false;
switch (req->cmd.mode) {
case SCSI_XFER_FROM_DEV:
case SCSI_XFER_TO_DEV:
scsi_req_continue(req);
break;
case SCSI_XFER_NONE:
scsi_req_dequeue(req);
scsi_req_enqueue(req); // *** this calls
scsi_req_enqueue_internal
break;
}
}
scsi_req_unref(req);
}
}
Finally when put_scsi_requests is called for migration, it seems like
it will call both virtio_scsi_save_request (from
bus->info->save_request) and scsi_generic_save_request (from
req->ops->save_request) and trigger the assert.
I searched for a bit, but didn't find anyone else reporting this. Has
anyone else seen this? It seems to me like that assert should check
that the sg list is empty instead of checking that it exists. Is this
an appropriate assessment? Assuming I find a repro, I'll try to test
this solution.
Thanks!
To manage notifications about this bug go to:
https://bugs.launchpad.net/qemu/+bug/1646610/+subscriptions