(In reply to Kevin Wolf from comment #0)
> When qemu is configured with a block device that has aio=native set, but the
> cache mode doesn't use O_DIRECT (i.e. isn't cache=none/directsync or any
> unnamed
> mode with explicit cache.direct=on), then the raw-posix block driver for
> local
> files and block devices will silently fall back to aio=threads.
> 
> The blockdev-add interface rejects such combinations, but qemu can't change
> the
> existing legacy interfaces that libvirt uses today, so the behaviour will
> stay there. Still, in the common case such behaviour is surprising and
> management should protect the user against it by rejecting or (if at all
> possible on the libvirt level) warning against it.
> 
> danpb suggested on IRC that libvirt doesn't have such a check yet and I
> should
> file a bug.
FYI: We have such a bug 1086502 for qemu-kvm component which caused QEMU core dumped.

Best Regards,
sluo

This patch seems to be enough for showing a warning and it is the solution I am in favour of:

diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c
index 4829176..6af76ce 100644
--- a/src/qemu/qemu_command.c
+++ b/src/qemu/qemu_command.c
@@ -3449,6 +3449,14 @@ qemuBuildDriveStr(virConnectPtr conn,
             mode = qemuDiskCacheV1TypeToString(disk->cachemode);
         }
 
+        if (disk->iomode == VIR_DOMAIN_DISK_IO_NATIVE &&
+            disk->cachemode != VIR_DOMAIN_DISK_CACHE_DIRECTSYNC) {
+            VIR_WARN(_("native I/O needs either no disk cache "
+                       "or directsync cache mode, QEMU will fallback "
+                       "to aio=threads"));
+            goto error;
+        }
+
         virBufferAsprintf(&opt, ",cache=%s", mode);
     } else if (disk->shared && !disk->readonly) {
         virBufferAddLit(&opt, ",cache=off");

is a warning enough in this case or should VIR_WARN be promoted to virReportError and deny the execution of the VM?

fixed upstream by:

commit 058384003db776c580d0e5a3016a6384e8eb7b92
Author: Giuseppe Scrivano <gscrivan>
Date:   Tue Jul 8 16:08:57 2014 +0200

    qemu: raise an eror when using aio=native without cache=none
    
    Qemu will fallback to aio=threads when the cache mode doesn't use
    O_DIRECT, even if aio=native was explictly set.
    
    Closes: https://bugzilla.redhat.com/show_bug.cgi?id=1086704
    
    Signed-off-by: Giuseppe Scrivano <gscrivan>


Fine to close this as UPSTREAM?

No, we need to make sure this is properly tested by QA.

Test with the following package version, this bug is verified.

Package version:
libvirt-1.2.8-8.el7.x86_64
qemu-kvm-rhev-2.1.2-12.el7.x86_64
kernel-3.10.0-208.el7.x86_64

Testing Matrix:
1. combine option testing "io='native'" and "cache='writethrough'", then start guest.
#virsh dumpxml rhel7.1
......
 <disk type='block' device='disk'>
      <driver name='qemu' type='raw' cache='writethrough' io='native'/>
      <source dev='/dev/sdc'/>
      <target dev='vdb' bus='virtio'/>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x09' function='0x0'/>
    </disk>
......
# virsh start rhel7.1
error: Failed to start domain rhel7.1
error: unsupported configuration: native I/O needs either no disk cache or directsync cache mode, QEMU will fallback to aio=threads


2. combine option testing "io='native'" and "cache='writeback'", then start guest.
#virsh dumpxml rhel7.1
......
 <disk type='block' device='disk'>
      <driver name='qemu' type='raw' cache='writeback' io='native'/>
      <source dev='/dev/sdc'/>
      <target dev='vdb' bus='virtio'/>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x09' function='0x0'/>
    </disk>
......
# virsh start rhel7.1
error: Failed to start domain rhel7.1
error: unsupported configuration: native I/O needs either no disk cache or directsync cache mode, QEMU will fallback to aio=threads

3. combine option testing "io='native'" and "cache='unsafe'", then start guest.
#virsh dumpxml rhel7.1
......
 <disk type='block' device='disk'>
      <driver name='qemu' type='raw' cache='unsafe' io='native'/>
      <source dev='/dev/sdc'/>
      <target dev='vdb' bus='virtio'/>
      <address type='pci' domain='0x0000' bus='0x00' slot='0x09' function='0x0'/>
    </disk>
......
# virsh start rhel7.1
error: Failed to start domain rhel7.1
error: unsupported configuration: native I/O needs either no disk cache or directsync cache mode, QEMU will fallback to aio=threads


4. as for other cache type and io type combination testing, the guest can be started up as expected. Following is the option combination testing matrix for your reference:
4.1 "io='native'" and "cache='default'" ----> guest can be started up as expected.
4.2 "io='native'" and "cache='none'" ----> guest can be started up as expected.
4.3 "io='native'" and "cache='directsync'" ----> guest can be started up as 
4.4 "io='threads'" and "cache='writethrough'" ----> guest can be started up as expected.
4.5 "io='threads'" and "cache='none'" ----> guest can be started up as expected.
4.6 "io='threads'" and "cache='directsync'" ----> guest can be started up as expected.
4.7 "io='threads'" and "cache='default'" ----> guest can be started up as expected.
4.8 "io='threads'" and "cache='writeback'" ----> guest can be started up as expected.
4.9 "io='threads'" and "cache='unsafe'" ----> guest can be started up as expected.

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://rhn.redhat.com/errata/RHSA-2015-0323.html

*** Bug 1259398 has been marked as a duplicate of this bug. ***