Re: [Xen-devel] [PATCH v2] x86/mm: Add mem access rights to NPT

[Razvan Cojocaru <rcojocaru@xxxxxxxxxxxxxxx>]

On 06/28/2018 05:53 PM, Alexandru Stefan ISAILA wrote:
> On Jo, 2018-06-28 at 08:40 -0600, Jan Beulich wrote:
>>>
>>>>
>>>>>
>>>>> On 28.06.18 at 16:10, <aisaila@xxxxxxxxxxxxxxx> wrote:
>>> On Vi, 2018-06-22 at 09:51 -0600, Jan Beulich wrote:
>>>>
>>>>>
>>>>>>
>>>>>>>
>>>>>>> On 18.06.18 at 17:17, <aisaila@xxxxxxxxxxxxxxx> wrote:
>>>>> From: Isaila Alexandru <aisaila@xxxxxxxxxxxxxxx>
>>>>>
>>>>> This patch adds access rights for the NPT pages. The access
>>>>> rights
>>>>> are
>>>>> saved in a radix tree with the root saved in p2m_domain.
>>>> Sounds resource intensive. How many nodes would such a radix tree
>>>> have
>>>> on average?
>>> The average is around 1478890 for a machine with 4GB of ram.
>> Is this with ...
>>
>>>
>>>>
>>>>>
>>>>> +static void p2m_set_access(struct p2m_domain *p2m, unsigned
>>>>> long
>>>>> gfn,
>>>>> +                                      p2m_access_t a)
>>>>> +{
>>>>> +    int rc;
>>>>> +
>>>>> +    if ( p2m_access_rwx == a )
>>>>> +        radix_tree_delete(&p2m->mem_access_settings, gfn);
>>>>> +
>>>>> +    rc = radix_tree_insert(&p2m->mem_access_settings, gfn,
>>>>> +                           radix_tree_int_to_ptr(a));
>>>> Is there an "else" missing above here? Otherwise why would you
>>>> delete the node first?
>>> Yes it needs a else or a return. We plan not to have the rwx in the
>>> tree so we ca save up some space.
>> ... this corrected? Otherwise I'm tempted to say that the creation of
>> this radix tree needs to be avoided by all means, as long as it's not
>> really needed.
> 
> The number was with this patch so no correction. This was done with
> xen-access write and I don't think it will make a difference if you
> change the access to all the mem pages.

Right, so the average is the average between runs of "xen-access <DOMID>
write", which sets all of the domain's pages to r-x. A typical
introspection application will not set all of the domain's pages up this
way - so the number is an average of maximums.

AFAIK ARM uses this model - are there fewer pages in a typical ARM
guest, or are we missing something about the way they are using radix
trees for this purpose? Or perhaps the ARM code should be changed as well?


Thanks,
Razvan

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel