Re: [libvirt] [PATCH] xen config: No vfb in HVM guest configuration

2012/4/6 Daniel Veillard <veillard(a)redhat.com&gt;: > On Thu, Apr 05, 2012 at 06:44:35PM +0200, Stefan Bader wrote: >> This causes an implicit vkbd device to be added which takes >> 6min to finally fail being initialized in the guest. >> >> http://lists.xen.org/archives/html/xen-devel/2012-04/msg00409.html >> >> Signed-off-by: Stefan Bader <stefan.bader(a)canonical.com&gt; >> --- >>  src/xenxs/xen_sxpr.c |   11 ++++------- >>  src/xenxs/xen_xm.c   |    2 +- >>  2 files changed, 5 insertions(+), 8 deletions(-) >> >> diff --git a/src/xenxs/xen_sxpr.c b/src/xenxs/xen_sxpr.c >> index b26b2bc..b2f98f6 100644 >> --- a/src/xenxs/xen_sxpr.c >> +++ b/src/xenxs/xen_sxpr.c >> @@ -2464,9 +2464,8 @@ xenFormatSxpr(virConnectPtr conn, >>              } >>          } >> >> -        /* PV graphics for xen <= 3.0.4, or HVM graphics for xen <= 3.1.0 */ >> -        if ((!hvm && xendConfigVersion < XEND_CONFIG_MIN_VERS_PVFB_NEWCONF) || >> -            (hvm && xendConfigVersion < XEND_CONFIG_VERSION_3_1_0)) { >> +        /* PV graphics for xen <= 3.0.4, or HVM graphics */ >> +        if (hvm || (xendConfigVersion < XEND_CONFIG_MIN_VERS_PVFB_NEWCONF)) { >>              if ((def->ngraphics == 1) && >>                  xenFormatSxprGraphicsOld(def->graphics[0], >>                                           &buf, xendConfigVersion) < 0) >> @@ -2578,10 +2577,8 @@ xenFormatSxpr(virConnectPtr conn, >>      if (xenFormatSxprAllPCI(def, &buf) < 0) >>          goto error; >> >> -    /* New style PV graphics config xen >= 3.0.4, >> -     * or HVM graphics config xen >= 3.0.5 */ >> -    if ((xendConfigVersion >= XEND_CONFIG_MIN_VERS_PVFB_NEWCONF && !hvm) || >> -        (xendConfigVersion >= XEND_CONFIG_VERSION_3_1_0 && hvm)) { >> +    /* New style PV graphics config xen >= 3.0.4 */ >> +    if (!hmv && (xendConfigVersion >= XEND_CONFIG_MIN_VERS_PVFB_NEWCONF)) { > >            ^^^ typo, it's hvm >    did you really compiled with the patch applied ? No, but no, too. Sorry, just saw this. I must admit I did not compile this again. I tried but it would have required a bit more fiddling to get over the autogen stage. And it was getting later the day, so I just ported what I got for our package and sent it. But the !hvm would be correct here. This is emitting the vfb part in sxpr which comes later with the devices sections. It should only be done for PVM not HVM. The patch basically removes the lower part of the if (so I probably should have left the order of config version test and then looking at hvm or not untouched, but I thought it be clearer to have all tests in the same pattern).