Re: [PATCH] delte PAGE_ORDER_1G in pod

[George Dunlap <george.dunlap@citrix.com>]

On 26/04/16 12:05, Zhang, Chunyu wrote:
> 
>> On 26/04/16 11:49, Zhang, Chunyu wrote:
>>>
>>>> On 26/04/16 08:27, zhangcy wrote:
>>>>> PoD does not have cache list for 1GB pages.
>>>>>
>>>>> Signed-off-by: zhangcy <zhangcy@cn.fujitsu.com>
>>>>
>>>> Thanks for the patch.  FYI we normally tag the area in the title in a
>>>> structured way; I probably would have used something like the following:
>>>>
>>>> xen/pod: Remove code handling PAGE_ORDER_1G from p2m_pod_cache_add
>>> got it, thanks.
>>>>
>>>> But with regards to the patch itself: The question isn't whether we have
>>>> a cache list for 1G pages; the question is whether p2m_pod_cache_add()
>>>> will ever be called with order == PAGE_ORDER_1G.
>>>>
>>>> Taking a quick glance around, it looks like in theory if a guest called
>>>> decrease_reservation with order == PAGE_ORDER_1G, you could conceivably
>>>> get to p2m_pod_cache_add() with order == PAGE_ORDER_1G.
>>> i just think like this:
>>>
>>> p2m_pod_decrease_reservation
>>>   - if ( steal_for_cache && p2m_is_ram(t) )
>>>      - p2m_pod_cache_add(p2m, page, cur_order)
>>> i think p2m_is_ram(t) , ram also from pod cache,
>>
>> No, that's memory from the guest's p2m table.  The p2m table can have 1G
> right..
> sorry , i did not write clearly.
> i mean: ram come like this:
> - pod cache is 4K or 2M
> - ram get from pod cache
> - set ram to p2m table.
> so i think p2m table is 4K or 2M.

Oh, right, I see -- a guest booted in PoD mode would normally only have
2M or 4k entries, since that's how they get filled in.

But there's nothing preventing someone coming up with a new domain
builder that comes with some 1G entries filled in already.  Nor is there
anything stopping a guest ballooning out a 1G region, then ballooning it
back in (hoping to get a full 1G entry), and then ballooning it out
again, causing Xen to potentially leak memory.

I haven't checked to see whether any of that is actually feasible or
not, but four lines of code is a small price to pay to not have to worry
about it. :-)

 -George


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