Re: [Xen-ia64-devel] [Patch][RFC] Avoid to domain_put_page pte of INVALID_M2P_ENTRY

[Akio Takebe <takebe_akio@xxxxxxxxxxxxxx>]

Hi, Isaku

>
>On Wed, Apr 23, 2008 at 01:54:29PM +0900, Akio Takebe wrote:
>Content-Description: Mail message body
>> Hi,
>
>Hi Akio.
>
>
>> This log show that domain_put_page is called 2 times for the same mfn.
>> But the mfn has different mpaddrs.
>> I guess the follwing case is occured:
>> 1. at 1st time, domain_put_page is called.
>> 2. it sets INMALID_M2P_ENTRY to the mfn.
>> 3. at 2nd time, domain_put_page is called.
>> 4. it hits BUG() becasuse of "get_gpfn_from_mfn(mfn) == INVALID_M2P_ENTRY"
>> 
>> The attached patch(avoid_to_domain_put_page_INVALID_M2P_ENTRY.patch) 
>> avoids to
>> domain_put_page ptes of INVALID_M2P_ENTRY.
>> After the patch was applied, I didn't get this issue.
>> But I'm not sure yet that the patch is the correct way to fix this issue.
>> I'll debug more, but if you have any comments, please tell me.
>
>
>The BUG() check is correct. It correctly detected the inconsistency
>between p2m table and m2p table.
>
>It looks like xen/ia64's grant table implementation's bug.
>Between 2 and 3, the page of mfn=a6a1f is assigned to
>mpaddr=1ee44. So its m2p entry should be set to 1ee44.
>But the entry isn't updated. It's the real issue.
>
>Let me look into it.
>Your stack trace didn't say about what grant table operation.
>It should GNTTABOP_unmap_grant_ref or GNTTABOP_unmap_and_replace.
>Do you know which it is?
Thank you for you investigating.
Both of them called GNTTABOP_unmap_and_replace.

Best Regards,

Akio Takebe


_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel