Re: [Xen-devel] [PATCH] xen/arm: p2m: Restrict preemption check in apply_p2m_changes

[Julien Grall <julien.grall@xxxxxxxxxx>]

Hi Ian,

On 06/05/15 09:31, Ian Campbell wrote:
> On Tue, 2015-05-05 at 17:31 +0100, Julien Grall wrote:
>> Hi Konrad,
>>
>> On 05/05/15 16:50, Konrad Rzeszutek Wilk wrote:
>>>> ---
>>>>  xen/arch/arm/p2m.c | 4 +++-
>>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/xen/arch/arm/p2m.c b/xen/arch/arm/p2m.c
>>>> index 65efa94..59dd23a 100644
>>>> --- a/xen/arch/arm/p2m.c
>>>> +++ b/xen/arch/arm/p2m.c
>>>> @@ -948,6 +948,7 @@ static int apply_p2m_changes(struct domain *d,
>>>>      const unsigned long sgfn = paddr_to_pfn(start_gpaddr),
>>>>                          egfn = paddr_to_pfn(end_gpaddr);
>>>>      const unsigned int preempt_count_limit = (op == MEMACCESS) ? 1 : 
>>>> 0x2000;
>>>> +    const bool_t preempt = !is_idle_vcpu(current);
>>>>      bool_t flush = false;
>>>>      bool_t flush_pt;
>>>>  
>>>> @@ -980,7 +981,8 @@ static int apply_p2m_changes(struct domain *d,
>>>>           * always make at least one pass as long as preempt_count_limit is
>>>>           * initialized with a value >= 1.
>>>>           */
>>>> -        if ( count >= preempt_count_limit && hypercall_preempt_check() )
>>>> +        if ( preempt && count >= preempt_count_limit
>>>> +             && hypercall_preempt_check() )
>>>
>>> Could you  use the softirq_pending() check to deal when there are no 
>>> domains?
>>
>> apply_p2m_changes will only be called with no current domain (i.e idle
>> vcpu) during DOM0 construction.
> 
> I wonder if we could/should try and switch to d0:v0 sooner during the
> build process? Is that even feasible?

We would hide out the problem and rely on hypercall_preempt_check to
always return 0 in non-hypercall context (i.e during DOM0 construction).

The problem may come back if we decide to change the behavior of
local_events_need_delivery or the interrupt subsystem.

I think this should be properly fix by calling hypercall_preempt_check
only when it's necessary.

Regards,

-- 
Julien Grall

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