On Thu, Apr 21, 2016 at 6:30 AM, Boris Ostrovsky
<boris.ostrov...@oracle.com> wrote:
>
>
> On 04/15/2016 06:03 PM, Thomas Garnier wrote:
>>
>> +void __init kernel_randomize_memory(void)
>> +{
>> + size_t i;
>> + unsigned long addr = memory_rand_start;
>> + unsigned long padding, rand, mem_tb;
>> + struct rnd_state rnd_st;
>> + unsigned long remain_padding = memory_rand_end -
>> memory_rand_start;
>> +
>> + if (!kaslr_enabled())
>> + return;
>> +
>> + /* Take the additional space when Xen is not active. */
>> + if (!xen_domain())
>> + page_offset_base -= __XEN_SPACE;
>
>
> This should be !xen_pv_domain(). Xen HVM guests are no different from bare
> metal as far as address ranges are concerned. (Technically it's probably
> !xen_pv_domain() && !xen_pvh_domain() but we can ignore PVH for now since it
> is being replaced by an HVM-type guest)
>
In my test KASLR was disabled on Xen so I should just remove this
check. I kept it in case it might change in the future.
> Having said that, I am not sure I understand why page_offset_base is
> shifted. I thought 0xffff800000000000 - 0xffff87ffffffffff is not supposed
> to be used by anyone, whether we are running under a hypervisor or not.
>
It is shifted to get the most space possible, it increases the entropy
available. Do you know why we should not use 0xffff800000000000 -
0xffff87ffffffffff?
> -boris
>
>
