Blame SOURCES/kvm-virtio-scsi-restart-DMA-after-iothread.patch

8b1478
From 64aa185d19f4e4afacd2501831049d6e615b5a84 Mon Sep 17 00:00:00 2001
8b1478
From: Stefan Hajnoczi <stefanha@redhat.com>
8b1478
Date: Tue, 16 Jul 2019 13:22:15 +0200
8b1478
Subject: [PATCH 22/23] virtio-scsi: restart DMA after iothread
8b1478
8b1478
RH-Author: Stefan Hajnoczi <stefanha@redhat.com>
8b1478
Message-id: <20190716132215.18503-4-stefanha@redhat.com>
8b1478
Patchwork-id: 89535
8b1478
O-Subject: [RHEL-7.8 RHEL-7.7.z qemu-kvm-rhev PATCH 3/3] virtio-scsi: restart DMA after iothread
8b1478
Bugzilla: 1673546
8b1478
RH-Acked-by: Paolo Bonzini <pbonzini@redhat.com>
8b1478
RH-Acked-by: John Snow <jsnow@redhat.com>
8b1478
RH-Acked-by: Kevin Wolf <kwolf@redhat.com>
8b1478
8b1478
When the 'cont' command resumes guest execution the vm change state
8b1478
handlers are invoked.  Unfortunately there is no explicit ordering
8b1478
between classic qemu_add_vm_change_state_handler() callbacks.  When two
8b1478
layers of code both use vm change state handlers, we don't control which
8b1478
handler runs first.
8b1478
8b1478
virtio-scsi with iothreads hits a deadlock when a failed SCSI command is
8b1478
restarted and completes before the iothread is re-initialized.
8b1478
8b1478
This patch uses the new qdev_add_vm_change_state_handler() API to
8b1478
guarantee that virtio-scsi's virtio change state handler executes before
8b1478
the SCSI bus children.  This way DMA is restarted after the iothread has
8b1478
re-initialized.
8b1478
8b1478
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
8b1478
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
8b1478
(cherry picked from commit 1a8c091c4ea5db3126514e3f7df678c9ee328802)
8b1478
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
8b1478
Signed-off-by: Miroslav Rezanina <mrezanin@redhat.com>
8b1478
---
8b1478
 hw/scsi/scsi-bus.c | 4 ++--
8b1478
 hw/virtio/virtio.c | 4 ++--
8b1478
 2 files changed, 4 insertions(+), 4 deletions(-)
8b1478
8b1478
diff --git a/hw/scsi/scsi-bus.c b/hw/scsi/scsi-bus.c
8b1478
index 5905f6b..ee4c449 100644
8b1478
--- a/hw/scsi/scsi-bus.c
8b1478
+++ b/hw/scsi/scsi-bus.c
8b1478
@@ -206,8 +206,8 @@ static void scsi_qdev_realize(DeviceState *qdev, Error **errp)
8b1478
         error_propagate(errp, local_err);
8b1478
         return;
8b1478
     }
8b1478
-    dev->vmsentry = qemu_add_vm_change_state_handler(scsi_dma_restart_cb,
8b1478
-                                                     dev);
8b1478
+    dev->vmsentry = qdev_add_vm_change_state_handler(DEVICE(dev),
8b1478
+            scsi_dma_restart_cb, dev);
8b1478
 }
8b1478
 
8b1478
 static void scsi_qdev_unrealize(DeviceState *qdev, Error **errp)
8b1478
diff --git a/hw/virtio/virtio.c b/hw/virtio/virtio.c
8b1478
index 3492b20..08a4332 100644
8b1478
--- a/hw/virtio/virtio.c
8b1478
+++ b/hw/virtio/virtio.c
8b1478
@@ -2306,8 +2306,8 @@ void virtio_init(VirtIODevice *vdev, const char *name,
8b1478
     } else {
8b1478
         vdev->config = NULL;
8b1478
     }
8b1478
-    vdev->vmstate = qemu_add_vm_change_state_handler(virtio_vmstate_change,
8b1478
-                                                     vdev);
8b1478
+    vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
8b1478
+            virtio_vmstate_change, vdev);
8b1478
     vdev->device_endian = virtio_default_endian();
8b1478
     vdev->use_guest_notifier_mask = true;
8b1478
 }
8b1478
-- 
8b1478
1.8.3.1
8b1478