Re: [Xen-devel] [PATCH] xen: privcmd: schedule() after private hypercall when non CONFIG_PREEMPT

[Luis R. Rodriguez]

On Mon, Dec 01, 2014 at 03:42:33PM +0100, Juergen Gross wrote:
> On 12/01/2014 02:32 PM, Luis R. Rodriguez wrote:
>> On Mon, Dec 01, 2014 at 11:01:18AM +0000, David Vrabel wrote:
>>> On 28/11/14 04:49, Juergen Gross wrote:
>>>> On 11/27/2014 07:50 PM, Andrew Cooper wrote:
>>>>>
>>>>> XenServer uses
>>>>>
>>>>> https://github.com/xenserver/linux-3.x.pg/blob/master/master/0001-x86-xen-allow-privcmd-hypercalls-to-be-preempted.patch
>>>>>
>>>>>
>>>>> to deal with these issues.  That patch is based on 3.10.
>>>>
>>>> Clever. :-)
>>>>
>>>>>
>>>>> I can remember whether this has been submitted upstream before (and
>>>>> there were outstanding issues), or whether it fell at an inconvenient
>>>>> time with our development cycles.
>>>>
>>>> I found
>>>>
>>>> http://lists.xen.org/archives/html/xen-devel/2014-02/msg02540.html
>>>>
>>>> and nothing else.
>>>
>>> I dropped it because it copy-and-paste a bunch of otherwise generic x86
>>> assembler and looked unlikely to get an x86 maintainer ack.  If you
>>> think otherwise, feel free to pick it up and run with it.
>>
>> I was trying to run with it, but my biggest gripe with this was
>> the use of preempt_schedule_irq(), but we can review that on the
>> other thread.

So much for the other thread :)

> I think this can be handled completely inside xen_evtchn_do_upcall().
>
> xen_preemptible_hcall_begin() (possibly with another name) could take
> the pointer of a function as parameter which is used as continuation
> point after an asynchronous interrupt in the critical section.

Oh so we'd only preempt to one specific task?

> Parameter for that function would be the exception frame of the
> original interrupt to be able to continue at the interrupted position
> after e.g. calling schedule().

Interesting... if reasonable I wonder if this should be generalized.
Are there other CONFIG_PREEMPT=n use cases for such excemptions?

  Luis
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/