Skip to contentSkip to conversation

libxc: failure to save/restore 32bit HVM guest

PATCH In review Mukesh Rathor 2 participants 5 messages Nov 23, 2009 – Nov 25, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 libxc: failure to save/restore 32bit HVM guest

Following appears to be an obvious bug. Save/restore of HVM guests works fine after the following fix. Including the patch, please let me know if you agree. I'll make parallel fix in our previous versions here.

thanks,
Mukesh

tools/libxc/xc_core.c Mon Nov 23 15:10:33 2009 -0800 renamed from tools/libxc/xc_core.c Mon Nov 23 08:06:54 2009 +0000 +1 -2
diff -r c8caa281a75a tools/libxc/xc_core.c
@@ -628,7 +628,7 @@
         PERROR("Could not get section header for .xen_prstatus");
         goto out;
     }
-    filesz = sizeof(ctxt[0].c) * nr_vcpus;
+    filesz = sizeof(*ctxt) * nr_vcpus;
     sts = xc_core_shdr_set(shdr, strtab, XEN_DUMPCORE_SEC_PRSTATUS,
                            SHT_PROGBITS, offset, filesz,
                            __alignof__(ctxt[0].c), sizeof(ctxt[0].c));
@@ -755,7 +755,7 @@
         goto out;
 
     /* prstatus: .xen_prstatus */
-    sts = dump_rtn(args, (char *)&ctxt[0].c, sizeof(ctxt[0].c) *

nr_vcpus);

+    sts = dump_rtn(args, (char *)&ctxt[0].c, sizeof(*ctxt) * nr_vcpus);
     if ( sts != 0 )
         goto out;
#1 lore raw
Quoted 4 lines · Mukesh Rathor

Following appears to be an obvious bug. Save/restore of HVM guests works fine after the following fix. Including the patch, please let me know if you agree. I'll make parallel fix in our previous versions here.

Save/restore of HVM guests should work fine before this fix too, since nothing in xc_core.c is used for guest save/restore. It's used for core dumps, as the name implies, and which is a rather different thing.

Furthermore I think the code in xc_core.c is correct as is.

-- Keir

#2 lore raw
Quoted 7 lines · On Tue, 24 Nov 2009 08:48:16 +0000 Keir Fraser <[email protected]>…
Quoted 4 lines · Mukesh Rathor

Following appears to be an obvious bug. Save/restore of HVM guests works fine after the following fix. Including the patch, please let me know if you agree. I'll make parallel fix in our previous versions here.

Quoted 9 lines

Save/restore of HVM guests should work fine before this fix too, since nothing in xc_core.c is used for guest save/restore. It's used for core dumps, as the name implies, and which is a rather different thing.

Furthermore I think the code in xc_core.c is correct as is.

-- Keir

Sorry, my bad, I meant dump core. I was simultaneously looking at save/restore, and accidentally said that.

Anyways, since 32bit HVM returns 64bit contexts, it appears the way it is, is buggy, since the sizeof(ctxt[0].c) will return 32bit context size in 32 bit libxc. Infact, after making the fix, user is able to get proper core of a Java HVM guest. Can you please quickly explain why you think it's correct in such case?

thanks,
Mukesh

#3 lore raw
Quoted 5 lines · Mukesh Rathor

Anyways, since 32bit HVM returns 64bit contexts, it appears the way it is, is buggy, since the sizeof(ctxt[0].c) will return 32bit context size in 32 bit libxc. Infact, after making the fix, user is able to get proper core of a Java HVM guest. Can you please quickly explain why you think it's correct in such case?

Ah, maybe you are right after all. But shouldn't you change the last two arguments in the call:
sts = xc_core_shdr_set(shdr, strtab, XEN_DUMPCORE_SEC_PRSTATUS,
SHT_PROGBITS, offset, filesz, __alignof__(ctxt[0].c), sizeof(ctxt[0].c));
...to reference *ctxt instead of ctxt[0].c, as well?

If you agree, send an updated patch and I'll apply it.

-- Keir

#4 lore raw
Quoted 15 lines · On Tue, 24 Nov 2009 19:58:48 +0000 Keir Fraser <[email protected]>…
Quoted 5 lines · Mukesh Rathor

Anyways, since 32bit HVM returns 64bit contexts, it appears the way it is, is buggy, since the sizeof(ctxt[0].c) will return 32bit context size in 32 bit libxc. Infact, after making the fix, user is able to get proper core of a Java HVM guest. Can you please quickly explain why you think it's correct in such case?

Ah, maybe you are right after all. But shouldn't you change the last two arguments in the call:
sts = xc_core_shdr_set(shdr, strtab, XEN_DUMPCORE_SEC_PRSTATUS,
SHT_PROGBITS, offset, filesz, __alignof__(ctxt[0].c), sizeof(ctxt[0].c));
...to reference *ctxt instead of ctxt[0].c, as well?

Quoted 4 lines

If you agree, send an updated patch and I'll apply it.

-- Keir

Yup, attached please find updated patch. I had the user test it out also. Since, ctxt is a union, the penultimate argument should be fine.

thanks,
mukesh

tools/libxc/xc_core.c +3 -3
diff -r c8caa281a75a tools/libxc/xc_core.c
@@ -628,10 +628,10 @@
         PERROR("Could not get section header for .xen_prstatus");
         goto out;
     }
-    filesz = sizeof(ctxt[0].c) * nr_vcpus;
+    filesz = sizeof(*ctxt) * nr_vcpus;
     sts = xc_core_shdr_set(shdr, strtab, XEN_DUMPCORE_SEC_PRSTATUS,
                            SHT_PROGBITS, offset, filesz,
-                           __alignof__(ctxt[0].c), sizeof(ctxt[0].c));
+                           __alignof__(ctxt[0].c), sizeof(*ctxt));
     if ( sts != 0 )
         goto out;
     offset += filesz;
@@ -755,7 +755,7 @@
         goto out;
 
     /* prstatus: .xen_prstatus */
-    sts = dump_rtn(args, (char *)&ctxt[0].c, sizeof(ctxt[0].c) * nr_vcpus);
+    sts = dump_rtn(args, (char *)&ctxt[0].c, sizeof(*ctxt) * nr_vcpus);
     if ( sts != 0 )
         goto out;
 

_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel