[Xen-devel] Migration: deadbeef?

[Andres Lagar Cavilla <andreslc@xxxxxxxxxxxxxx>]

Hi list,
I've been getting a strange warning on xfrd.log lately, when 
live-migrating a domain. At a very eraly stage of the initial full 
memory transfer, the following ominous message surfaces
%FNI 982 : [1000007e,1007] pte=00be2063, mfn=00000be2, pfn=ffffffff 
[mfn]=deadbeef
A comment on xc_linux_save.c code implies that this should never happen

/* entering this loop, pfn_type is now in pfns (Not mfns) */
for( j = 0; j < batch; j++ ){
........
        for ( k = 0;
                k < (((pfn_type[j] & LTABTYPE_MASK) == L2TAB) ?
                (HYPERVISOR_VIRT_START >> L2_PAGETABLE_SHIFT) :
                1024);
                 k++ ){
                 ........

                      if ( !MFN_IS_IN_PSEUDOPHYS_MAP(mfn) )
                       {
                           /* I don't think this should ever happen */
                           printf("FNI %d : [%08lx,%d] pte=%08lx, "
                                  "mfn=%08lx, pfn=%08lx [mfn]=%08lx\n",
                                  j, pfn_type[j], k,
                                  page[k], mfn, live_mfn_to_pfn_table[mfn],
                                  (live_mfn_to_pfn_table[mfn]<nr_pfns)?
                                  live_pfn_to_mfn_table[
                                      live_mfn_to_pfn_table[mfn]] :
                                  0xdeadbeef);

                           pfn = 0; /* be suspicious */
                       }

Afetr migration finishes, I usually notice internal misbehavior of the 
domain after receiving this deadbeef warning, particularly when accesing 
block devices, as in "journal commit I/O error" and the like
Any help will be greatly appreciated.
Thanks a lot
Andres

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel