Re: [PATCH v1 9/9] mm: Make sure pages are scrubbed

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 29/03/17 17:45, Wei Liu wrote:
> On Wed, Mar 29, 2017 at 12:35:25PM -0400, Boris Ostrovsky wrote:
>> On 03/29/2017 12:25 PM, Wei Liu wrote:
>>> On Fri, Mar 24, 2017 at 01:05:04PM -0400, Boris Ostrovsky wrote:
>>>> +static void check_one_page(struct page_info *pg)
>>>> +{
>>>> +    mfn_t mfn = _mfn(page_to_mfn(pg));
>>>> +    uint64_t *ptr;
>>>> +
>>>> +    ptr  = map_domain_page(mfn);
>>>> +    ASSERT(*ptr != PAGE_POISON);
>>> Should be ASSERT(*ptr == 0) here.
>> We can't assume it will be zero --- see scrub_one_page().
> It's still possible a value is leaked from the guest, and that value has
> rather high probability to be != PAGE_POISON.
>
> In that case there should be a ifdef to match the debug and non-debug
> builds.

The only sensible thing to do is check that the entire page is zeroes. 
This is a debug option after all.

We don't want to waste time poisoning zero pages we hand back to the
free pool.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel