On Wed, May 24, 2017 at 16:06:56 +0100, Daniel Berrange wrote: > On Wed, May 24, 2017 at 04:45:57PM +0200, Michal Privoznik wrote: > > https://bugzilla.redhat.com/show_bug.cgi?id=1450349 > > > > Problem is, qemu fails to load guest memory image if these > > attribute change on migration/restore from an image. > > [snip] > > > diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c > > index 35fd79de8..c1ff8ca8a 100644 > > --- a/src/qemu/qemu_domain.c > > +++ b/src/qemu/qemu_domain.c > > @@ -5797,6 +5797,46 @@ qemuDomainUpdateMemoryDeviceInfo(virQEMUDriverPtr driver, > > } > > > > > > +static bool > > +qemuDomainABIStabilityCheck(const virDomainDef *src, > > + const virDomainDef *dst) > > +{ > > + if (src->mem.source != dst->mem.source) { > > + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, > > + _("Target memoryBacking source '%s' doesn't " > > + "match source memoryBacking source'%s'"), > > + virDomainMemorySourceTypeToString(dst->mem.source), > > + virDomainMemorySourceTypeToString(src->mem.source)); > > + return false; > > + } > > + > > + if (src->mem.access != dst->mem.access) { > > + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, > > + _("Target memoryBacking access '%s' doesn't " > > + "match access memoryBacking access'%s'"), > > + virDomainMemoryAccessTypeToString(dst->mem.access), > > + virDomainMemoryAccessTypeToString(src->mem.access)); > > + return false; > > + } > > + > > + if (src->mem.allocation != dst->mem.allocation) { > > + virReportError(VIR_ERR_CONFIG_UNSUPPORTED, > > + _("Target memoryBacking allocation '%s' doesn't " > > + "match allocation memoryBacking allocation'%s'"), > > + virDomainMemoryAllocationTypeToString(dst->mem.allocation), > > + virDomainMemoryAllocationTypeToString(src->mem.allocation)); > > + return false; > > + } > > > Do we really need to blacklist all of these changes. I can understand that > changing the memory source would affect migration ABI, as it causes us to > use the memory backend command line config differently. > > Assuming that matches though, I'm sceptical that changing 'access' or > 'allocation' affects ABI. If it does, a comment should state when it's required. (e.g. access may need to be checked, since shared access may force usage of memory-backend-file). Allocation is indeed weird.
Attachment:
signature.asc
Description: PGP signature
-- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list
