Re: [PATCH v3 1/3] x86/ldt: Make modify_ldt synchronous

[Andy Lutomirski <luto@amacapital.net>]

On Fri, Jul 24, 2015 at 9:13 PM, Boris Ostrovsky
<boris.ostrovsky@oracle.com> wrote:
>
>
> On 07/22/2015 06:20 PM, Boris Ostrovsky wrote:
>>
>> On 07/22/2015 03:23 PM, Andy Lutomirski wrote:
>>>
>>>
>>> +    error = -ENOMEM;
>>> +    new_ldt = alloc_ldt_struct(newsize);
>>> +    if (!new_ldt)
>>>           goto out_unlock;
>>> -    }
>>>   -    fill_ldt(&ldt, &ldt_info);
>>> -    if (oldmode)
>>> -        ldt.avl = 0;
>>> +    if (old_ldt) {
>>> +        memcpy(new_ldt->entries, old_ldt->entries,
>>> +               oldsize * LDT_ENTRY_SIZE);
>>> +    }
>>> +    memset(new_ldt->entries + oldsize * LDT_ENTRY_SIZE, 0,
>>> +           (newsize - oldsize) * LDT_ENTRY_SIZE);
>>
>>
>> We need to zero out full page (probably better in alloc_ldt_struct() with
>> vmzalloc/__GFP_ZERO) --- Xen checks whole page that is assigned to  G/LDT
>> and gets unhappy if an invalid descriptor is found there.
>>
>> This fixes one problem. There is something else that Xen gets upset about,
>> I haven't figured what it is yet (and I am out tomorrow so it may need to
>> wait until Friday).
>>
>
>
> What I thought was another problem turned out not to be one so both 64- and
> 32-bit tests passed on 64-bit PV (when allocated LDT is zeroed out)
>
> However, on 32-bit kernel the test is failing multicpu test, I don't know
> yet what it is.

Test case bug or unrelated kernel bug depending on your point of view.
I forgot that x86_32 and x86_64 have very different handling of IRET
faults.  Wait for v2 :)

--Andy