[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [libvirt] [PATCH 3/4] xen: fix PyGrub boot device order

On 10/12/2011 02:26 AM, Philipp Hahn wrote:
When PyGrub is used as the bootloader in Xen, it gets passed the first
bootable disk. Xend supports a "bootable"-flag for this, which isn't
explicitly supported by libvirt.

Hmm, the XML has been enhanced in the meantime; we can now mark various disks as bootable (and in fact, in which order they should be attempted at boot) when targeting qemu. Why can't the same xml changes be made to affect xen? But that can be done as a separate patch.

When converting libvirt-xml to xen-sxpr the "bootable"-flag gets
implicitly set by xen.xend.XenConfig.device_add() for the first disk
(marked as "Compat hack -- mark first disk bootable").
When converting back xen-sxpr to libvirt-xml, the disks are returned in
the internal order used by Xend ignoring the "bootable"-flag, which
looses the original order. When the domain is then re-defined, the order


of disks is changed, which breaks PyGrub, since a different disk gets

When converting xen-sxpt to libvirt-xml, use the "bootable"-flag to


determine the first disk.

This isn't perfect, since several disks can be marked as bootable using
the Xend-API, but that is not supported by libvirt. In all known cases
relevant to libvirt exactly one disk is marked as bootable.

Signed-off-by: Philipp Hahn<hahn univention de>
valgrind might complain about an uninitialized read access when copying
disks[0] to disks[0], but that is overwritten again with the following

No problem from valgrind; VIR_ALLOC_N guarantees that disks starts life initialized at 0 (that is, calloc, not malloc).

@@ -481,7 +485,13 @@ xenParseSxprDisks(virDomainDefPtr def,
              if (VIR_REALLOC_N(def->disks, def->ndisks+1)<  0)
                  goto no_memory;

-            def->disks[def->ndisks++] = disk;
+            /* re-order disks if there is a bootable device */
+            if (STREQ_NULLABLE(bootable, "1")) {
+                def->disks[def->ndisks++] = def->disks[0];
+                def->disks[0] = disk;
+            } else {
+                def->disks[def->ndisks++] = disk;
+            }

Feels a bit hacky, compared to a memmove to slide all disks; but if, as you say, xend already returns disks in an arbitrary order, then it doesn't matter if we preserve order.

ACK and pushed.

Eric Blake   eblake redhat com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]