Re: [PATCH v2 5/8] evtchn: drop acquiring of per-channel lock from send_guest_{global,vcpu}_virq()

[Jan Beulich <jbeulich@xxxxxxxx>]

On 30.10.2020 12:15, Jürgen Groß wrote:
> On 30.10.20 11:57, Julien Grall wrote:
>>
>>
>> On 30/10/2020 10:49, Jan Beulich wrote:
>>> On 30.10.2020 11:38, Julien Grall wrote:
>>>> On 22/10/2020 17:17, Jan Beulich wrote:
>>>>> On 22.10.2020 18:00, Roger Pau Monné wrote:
>>>>>> On Tue, Oct 20, 2020 at 04:10:09PM +0200, Jan Beulich wrote:
>>>>>>> --- a/xen/include/xen/event.h
>>>>>>> +++ b/xen/include/xen/event.h
>>>>>>> @@ -177,9 +177,16 @@ int evtchn_reset(struct domain *d, bool
>>>>>>>     * Low-level event channel port ops.
>>>>>>>     *
>>>>>>>     * All hooks have to be called with a lock held which prevents 
>>>>>>> the channel
>>>>>>> - * from changing state. This may be the domain event lock, the 
>>>>>>> per-channel
>>>>>>> - * lock, or in the case of sending interdomain events also the 
>>>>>>> other side's
>>>>>>> - * per-channel lock. Exceptions apply in certain cases for the PV 
>>>>>>> shim.
>>>>>>> + * from changing state. This may be
>>>>>>> + * - the domain event lock,
>>>>>>> + * - the per-channel lock,
>>>>>>> + * - in the case of sending interdomain events the other side's 
>>>>>>> per-channel
>>>>>>> + *   lock,
>>>>>>> + * - in the case of sending non-global vIRQ-s the per-vCPU 
>>>>>>> virq_lock (in
>>>>>>> + *   combination with the ordering enforced through how the vCPU's
>>>>>>> + *   virq_to_evtchn[] gets updated),
>>>>>>> + * - in the case of sending global vIRQ-s vCPU 0's virq_lock.
>>>>>>> + * Exceptions apply in certain cases for the PV shim.
>>>>>>
>>>>>> Having such a wide locking discipline looks dangerous to me, it's easy
>>>>>> to get things wrong without notice IMO.
>>>>>
>>>>> It is effectively only describing how things are (or were before
>>>>> XSA-343, getting restored here).
>>>>
>>>> I agree with Roger here, the new/old locking discipline is dangerous and
>>>> it is only a matter of time before it will bite us again.
>>>>
>>>> I think we should consider Juergen's series because the locking for the
>>>> event channel is easier to understand.
>>>
>>> We should, yes. The one thing I'm a little uneasy with is the
>>> new lock "variant" that gets introduced. Custom locking methods
>>> also are a common source of problems (which isn't to say I see
>>> any here).
>>
>> I am also unease with a new lock "variant". However, this is the best 
>> proposal I have seen so far to unblock the issue.
>>
>> I am open to other suggestion with simple locking discipline.
> 
> In theory my new lock variant could easily be replaced by a rwlock and
> using the try-variant for the readers only.

Well, only until we would add check_lock() there, which I think
we should really have (not just on the slow paths, thanks to
the use of spin_lock() there). The read-vs-write properties
you're utilizing aren't applicable in the general case afaict,
and hence such checking would get in the way.

> The disadvantage of that approach would be a growth of struct evtchn.

Wasn't it you who had pointed out to me the aligned(64) attribute
on the struct (in a different context), which afaict would subsume
any possible growth?

Jan