Re: [Xen-devel] [PATCH] xen/p2m: Remove np2m-specific filter from generic p2m_flush_table

[George Dunlap <george.dunlap@xxxxxxxxxx>]

On 31/01/17 10:44, Jan Beulich wrote:
>>>> On 30.01.17 at 16:17, <george.dunlap@xxxxxxxxxx> wrote:
>> Commit 71bb7304e7a7a35ea6df4b0cedebc35028e4c159 added flushing of
>> nested p2m tables whenever the host p2m table changed.  Unfortunately
>> in the process, it added a filter to the generic p2m_flush_table()
>> function so that the p2m would only be flushed if it was being used as
>> a nested p2m.  This meant that the p2m was not being flushed at all
>> for altp2m callers.
>>
>> Instead do the nested p2m filtering in p2m_flush_nestedp2m().
>>
>> NB that this is not a security issue: The only time this codepath is
>> called is in cases where either nestedp2m or altp2m is enabled, and
>> neither of them are in security support.
>>
>> Reported-by: Matt Leinhos <matt@xxxxxxxxxx>
>> Signed-off-by: George Dunlap <george.dunlap@xxxxxxxxxx>
>> ---
>> I've smoke-tested this with nested virt and it seems to work fine.
>> Matt / Tamas, could you test this with altp2m and see if it fixes your
>> issue?
>>
>>
>> CC: Liang Li <liang.z.li@xxxxxxxxx>
>> CC: Yang Zhang <yang.z.zhang@xxxxxxxxx>
>> CC: Tim Deegan <tim@xxxxxxx>
>> CC: Tamas K Lengyel <tamas.lengyel@xxxxxxxxxxxx>
>> CC: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
>> CC: Jan Beulich <jbeulich@xxxxxxxx>
>> CC: Matt Leinhos <matt@xxxxxxxxxx>
>> ---
>>  xen/arch/x86/mm/p2m.c | 12 +++++-------
>>  1 file changed, 5 insertions(+), 7 deletions(-)
>>
>> diff --git a/xen/arch/x86/mm/p2m.c b/xen/arch/x86/mm/p2m.c
>> index aa627d8..0849c6e 100644
>> --- a/xen/arch/x86/mm/p2m.c
>> +++ b/xen/arch/x86/mm/p2m.c
>> @@ -2048,12 +2048,6 @@ p2m_flush_table(struct p2m_domain *p2m)
>>      ASSERT(page_list_empty(&p2m->pod.super));
>>      ASSERT(page_list_empty(&p2m->pod.single));
>>  
>> -    if ( p2m->np2m_base == P2M_BASE_EADDR )
>> -    {
>> -        p2m_unlock(p2m);
>> -        return;
>> -    }
>> -
>>      /* This is no longer a valid nested p2m for any address space */
>>      p2m->np2m_base = P2M_BASE_EADDR;
>>      
>> @@ -2088,7 +2082,11 @@ p2m_flush_nestedp2m(struct domain *d)
>>  {
>>      int i;
>>      for ( i = 0; i < MAX_NESTEDP2M; i++ )
>> -        p2m_flush_table(d->arch.nested_p2m[i]);
>> +    {
>> +        struct p2m_domain *p2m = d->arch.nested_p2m[i];
>> +        if ( p2m->np2m_base != P2M_BASE_EADDR )
>> +            p2m_flush_table(p2m);
>> +    }
>>  }
> 
> So the use of p2m_flush_table() in p2m_get_nestedp2m() is fine
> as is because the new np2m_base can't be P2M_BASE_EADDR (as
> a comment there says slightly indirectly). I think this may be worth
> clarifying in the commit message.
> 
> What about p2m_flush()'es use of p2m_flush_table() though?
> There in particular are uses from vvmx.c and hap.c, both of which
> suggest nested-virt context.

I think the "filter" is only an optimization: If it's not there you'll
just end up "clearing" an already clear table.  That's the way things
were before 71bb730.

We could add an nestedp2m-specific wrapper function do to the test
instead, and then have all nestedp2m-specific callers call it.  Might be
a worthwhile change.

 -George

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