Fix kvm migration with empty virtio-scsi controller (bz #1032208)

0001-init_virtio_scsi-reset-the-HBA-before-freeing-its-vi.patch

@@ -0,0 +1,54 @@
+From 5f2d17d35b2339526f3b3d580b279ea78e406a25 Mon Sep 17 00:00:00 2001
+Message-Id: <5f2d17d35b2339526f3b3d580b279ea78e406a25.1394734933.git.crobinso@redhat.com>
+From: Laszlo Ersek <lersek@redhat.com>
+Date: Wed, 15 Jan 2014 02:48:40 +0100
+Subject: [PATCH] init_virtio_scsi(): reset the HBA before freeing its virtio
+ ring
+
+When init_virtio_scsi() finds no SCSI targets connected to the HBA, it
+frees the virtio ring. Other code in SeaBIOS proceeds to overwrite the
+area. However, the ring is in use by qemu at that point -- not only did we
+report the (ACK|DRIVER|DRIVER_OK) status earlier, we even communicated
+over the ring.
+
+Of course SeaBIOS doesn't "kick" the HBA ever again, hence qemu has no
+reason to look at the ring. However, when qemu uses KVM acceleration, and
+ioeventfd is enabled for the HBA, then a vmstate change to "running"
+(including stop->cont monitor commands and incoming migration) "forces" a
+kick (see qemu commit 25db9ebe). Qemu then tries to interpret whatever
+unrelated guest data is in the HBA's original ring area, as virtio
+protocol. Qemu exits upon seeing the garbage.
+
+init_virtio_scsi() should reset the HBA before allowing the virtio ring
+memory to be reused. Device reset causes the hypervisor to drop its
+references.
+
+This change is justified / underpinned by pure virtio-spec compliance as
+well.
+
+Related RHBZ: https://bugzilla.redhat.com/show_bug.cgi?id=1013418
+
+Signed-off-by: Laszlo Ersek <lersek@redhat.com>
+---
+ src/hw/virtio-scsi.c | 4 +++-
+ 1 file changed, 3 insertions(+), 1 deletion(-)
+
+diff --git a/src/hw/virtio-scsi.c b/src/hw/virtio-scsi.c
+index 4b4ec7b..48fb3e1 100644
+--- a/src/hw/virtio-scsi.c
++++ b/src/hw/virtio-scsi.c
+@@ -158,8 +158,10 @@ init_virtio_scsi(struct pci_device *pci)
+     for (tot = 0, i = 0; i < 256; i++)
+         tot += virtio_scsi_scan_target(pci, ioaddr, vq, i);
+ 
+-    if (!tot)
++    if (!tot) {
++        vp_reset(ioaddr);
+         goto fail;
++    }
+ 
+     return;
+ 
+-- 
+1.8.5.3
+

seabios.spec

@@ -1,6 +1,6 @@
 Name:           seabios
 Version:        1.7.4
-Release:        1%{?dist}
+Release:        2%{?dist}
 Summary:        Open-source legacy BIOS implementation
 
 Group:          Applications/Emulators
@@ -15,6 +15,9 @@ Source12:       config.vga.qxl
 Source13:       config.vga.stdvga
 Source14:       config.vga.vmware
 
+# Fix kvm migration with empty virtio-scsi controller (bz #1032208)
+Patch0001: 0001-init_virtio_scsi-reset-the-HBA-before-freeing-its-vi.patch
+
 BuildRequires: python iasl
 BuildRequires: binutils-x86_64-linux-gnu gcc-x86_64-linux-gnu
 
@@ -67,6 +70,9 @@ SeaVGABIOS is an open-source VGABIOS implementation.
 %prep
 %setup -q
 
+# Fix kvm migration with empty virtio-scsi controller (bz #1032208)
+%patch0001 -p1
+
 # Makefile changes version to include date and buildhost
 sed -i 's,VERSION=%{version}.*,VERSION=%{version},g' Makefile
 
@@ -147,6 +153,9 @@ install -m 0644 binaries/vgabios*.bin $RPM_BUILD_ROOT%{_datadir}/seavgabios
 
 
 %changelog
+* Thu Mar 13 2014 Cole Robinson <crobinso@redhat.com> - 1.7.4-2
+- Fix kvm migration with empty virtio-scsi controller (bz #1032208)
+
 * Mon Jan 06 2014 Cole Robinson <crobinso@redhat.com> - 1.7.4-1
 - Rebased to version 1.7.4
 - Support for obtaining ACPI tables directly from QEMU.