32/64 resotre broken

32/64 resotre broken

[Zhai, Edwin]

keir,

i'm almost sure following change set break 32b HVM guest save/restore on 
64b hypvisor:
r13594: Make domctl/sysctl interfaces 32-/64-bit invariant.

restoring 32b linux cause a guest kernel panic, but restoring saved 
image from previous changeset is okay. i.e. saved image become damaged 
in r13594.

according the log, hvm context seems to be okay. i doubt whether vcpu 
context saving process is changed.

do you have any idea?

thanks,

-- 
best rgds,
edwin

Cs 13594 - broken?

[Petersson, Mats]

> i'm almost sure following change set break 32b HVM guest 
> save/restore on 
> 64b hypvisor:
> r13594: Make domctl/sysctl interfaces 32-/64-bit invariant.
> 
> restoring 32b linux cause a guest kernel panic, but restoring saved 
> image from previous changeset is okay. i.e. saved image 
> become damaged 
> in r13594.

Looking at that patch (and the current changeset matches the patch for
the below code with about +30 lines offset), I think this code is
broken:
@@ -976,10 +976,16 @@ int xc_linux_save(int xc_handle, int io_
                 goto out;
             }
 
-            if (xc_get_pfn_type_batch(xc_handle, dom, batch, pfn_type))
{
+            for ( j = 0; j < batch; j++ )
+                ((uint32_t *)pfn_type)[i] = pfn_type[i];
Shouldn't that be [j] in both places?
+            if ( xc_get_pfn_type_batch(xc_handle, dom, batch,
+                                       (uint32_t *)pfn_type) )
+            {
                 ERROR("get_pfn_type_batch failed");
                 goto out;
             }
+            for ( j = batch-1; j >= 0; j-- )
+                pfn_type[i] = ((uint32_t *)pfn_type)[i];
And here, as well?


Admittedly, this probably won't fix the 32/64 HVM restore being broken,
but I doubt that the code ACTUALLY works as it stands today. 

I'm still trying to find a reason for the restore not working... 

--
Mats

Re: Cs 13594 - broken?

[Tim Deegan]

At 13:55 +0100 on 31 Jan (1170251725), Petersson, Mats wrote:
> Shouldn't that be [j] in both places?

Yep; that's fixed in the staging tree, but hasn't made it to the public
tree yet.

Cheers,

Tim.

RE: Cs 13594 - broken?

[Petersson, Mats]

> -----Original Message-----
> From: Tim Deegan [mailto:Tim.Deegan@xensource.com] 
> Sent: 31 January 2007 13:01
> To: Petersson, Mats
> Cc: edwin.zhai@intel.com; Keir; xen-devel@lists.xensource.com
> Subject: Re: [Xen-devel] Cs 13594 - broken?
> 
> At 13:55 +0100 on 31 Jan (1170251725), Petersson, Mats wrote:
> > Shouldn't that be [j] in both places?
> 
> Yep; that's fixed in the staging tree, but hasn't made it to 
> the public
> tree yet.

So it is - sorry for the duplicated error report... :-(

--
Mats
> 
> Cheers,
> 
> Tim.
> 
> 
> 
>