[Question] the check of ioeventfd collision in kvm_*assign_ioeventfd_idx

[Question] the check of ioeventfd collision in kvm_*assign_ioeventfd_idx

[Zhenyu Ye]

Hi all,

There are checks of ioeventfd collision in both kvm_assign_ioeventfd_idx()
and kvm_deassign_ioeventfd_idx(), however, with different logic.

In kvm_assign_ioeventfd_idx(), this is done by ioeventfd_check_collision():
---8<---
	if (_p->bus_idx == p->bus_idx &&
	    _p->addr == p->addr &&
	    (!_p->length || !p->length ||
	     (_p->length == p->length &&
	      (_p->wildcard || p->wildcard ||
	       _p->datamatch == p->datamatch))))
		// then we consider the two are the same
---8<---

The logic in kvm_deassign_ioeventfd_idx() is as follows:
---8<---
	if (p->bus_idx != bus_idx ||
	    p->eventfd != eventfd  ||
	    p->addr != args->addr  ||
	    p->length != args->len ||
	    p->wildcard != wildcard)
		continue;

	if (!p->wildcard && p->datamatch != args->datamatch)
		continue;

	// then we consider the two are the same
---8<---

As we can see, there is extra check of p->eventfd in

().  Why we don't check p->eventfd
in kvm_assign_ioeventfd_idx()? Or should we delete this in
kvm_deassign_ioeventfd_idx()?


Thanks,
Zhenyu

Re: [Question] the check of ioeventfd collision in kvm_*assign_ioeventfd_idx

[Zhenyu Ye]

On 2020/7/31 2:03, Paolo Bonzini wrote:
> Yes, I think it's not needed. Probably the deassign check can be turned into an assertion?
> 
> Paolo
> 

I think we can do this in the same function, and turnt he check of
p->eventfd into assertion in kvm_deassign_ioeventfd_idx(). Just like:

---8<---
static inline struct _ioeventfd *
get_ioeventfd(struct kvm *kvm, enum kvm_bus bus_idx,
              struct kvm_ioeventfd *args)
{
        static struct _ioeventfd *_p;
        bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH);

        list_for_each_entry(_p, &kvm->ioeventfds, list)
                if (_p->bus_idx == bus_idx &&
                    _p->addr == args->addr &&
                    (!_p->length || !args->len ||
                     (_p->length == args->len &&
                      (_p->wildcard || wildcard ||
                       _p->datamatch == args->datamatch))))
                        return _p;

        return NULL;
}

kvm_deassign_ioeventfd_idx() {
	...
	p = get_ioeventfd(kvm, bus_idx, args);
	if (p) {
		assert(p->eventfd == eventfd);
		...
	}

---8<----

This may be easier to understand (keep the same logic in assign/deassign).

I will send a formal patch soon.

Thanks,
Zhenyu


> Il gio 30 lug 2020, 16:36 Zhenyu Ye <yezhenyu2@huawei.com <mailto:yezhenyu2@huawei.com>> ha scritto:
> 
>     Hi all,
> 
>     There are checks of ioeventfd collision in both kvm_assign_ioeventfd_idx()
>     and kvm_deassign_ioeventfd_idx(), however, with different logic.
> 
>     In kvm_assign_ioeventfd_idx(), this is done by ioeventfd_check_collision():
>     ---8<---
>             if (_p->bus_idx == p->bus_idx &&
>                 _p->addr == p->addr &&
>                 (!_p->length || !p->length ||
>                  (_p->length == p->length &&
>                   (_p->wildcard || p->wildcard ||
>                    _p->datamatch == p->datamatch))))
>                     // then we consider the two are the same
>     ---8<---
> 
>     The logic in kvm_deassign_ioeventfd_idx() is as follows:
>     ---8<---
>             if (p->bus_idx != bus_idx ||
>                 p->eventfd != eventfd  ||
>                 p->addr != args->addr  ||
>                 p->length != args->len ||
>                 p->wildcard != wildcard)
>                     continue;
> 
>             if (!p->wildcard && p->datamatch != args->datamatch)
>                     continue;
> 
>             // then we consider the two are the same
>     ---8<---
> 
>     As we can see, there is extra check of p->eventfd in
> 
>     ().  Why we don't check p->eventfd
>     in kvm_assign_ioeventfd_idx()? Or should we delete this in
>     kvm_deassign_ioeventfd_idx()?
> 
> 
>     Thanks,
>     Zhenyu
> 

Re: [Question] the check of ioeventfd collision in kvm_*assign_ioeventfd_idx

[Paolo Bonzini]

On 31/07/20 08:39, Zhenyu Ye wrote:
> On 2020/7/31 2:03, Paolo Bonzini wrote:
>> Yes, I think it's not needed. Probably the deassign check can be turned into an assertion?
>>
>> Paolo
>>
> 
> I think we can do this in the same function, and turnt he check of
> p->eventfd into assertion in kvm_deassign_ioeventfd_idx(). Just like:
> 
> ---8<---
> static inline struct _ioeventfd *
> get_ioeventfd(struct kvm *kvm, enum kvm_bus bus_idx,
>               struct kvm_ioeventfd *args)
> {
>         static struct _ioeventfd *_p;
>         bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH);
> 
>         list_for_each_entry(_p, &kvm->ioeventfds, list)
>                 if (_p->bus_idx == bus_idx &&
>                     _p->addr == args->addr &&
>                     (!_p->length || !args->len ||
>                      (_p->length == args->len &&
>                       (_p->wildcard || wildcard ||
>                        _p->datamatch == args->datamatch))))
>                         return _p;
> 
>         return NULL;
> }
> 
> kvm_deassign_ioeventfd_idx() {
> 	...
> 	p = get_ioeventfd(kvm, bus_idx, args);
> 	if (p) {
> 		assert(p->eventfd == eventfd);
> 		...
> 	}
> 
> ---8<----
> 
> This may be easier to understand (keep the same logic in assign/deassign).

I think you should also warn if:

1) p->length != args->len

2) p->wildcard != args->wildcard if p->length

3) p->datamatch != args->datamatch if p->length && !p->wildcard

but yeah it sounds like a plan.

Paolo

Re: [Question] the check of ioeventfd collision in kvm_*assign_ioeventfd_idx

[Zhenyu Ye]

On 2020/7/31 14:44, Paolo Bonzini wrote:
> On 31/07/20 08:39, Zhenyu Ye wrote:
>> On 2020/7/31 2:03, Paolo Bonzini wrote:
>>> Yes, I think it's not needed. Probably the deassign check can be turned into an assertion?
>>>
>>> Paolo
>>>
>>
>> I think we can do this in the same function, and turnt he check of
>> p->eventfd into assertion in kvm_deassign_ioeventfd_idx(). Just like:
>>
>> ---8<---
>> static inline struct _ioeventfd *
>> get_ioeventfd(struct kvm *kvm, enum kvm_bus bus_idx,
>>               struct kvm_ioeventfd *args)
>> {
>>         static struct _ioeventfd *_p;
>>         bool wildcard = !(args->flags & KVM_IOEVENTFD_FLAG_DATAMATCH);
>>
>>         list_for_each_entry(_p, &kvm->ioeventfds, list)
>>                 if (_p->bus_idx == bus_idx &&
>>                     _p->addr == args->addr &&
>>                     (!_p->length || !args->len ||
>>                      (_p->length == args->len &&
>>                       (_p->wildcard || wildcard ||
>>                        _p->datamatch == args->datamatch))))
>>                         return _p;
>>
>>         return NULL;
>> }
>>
>> kvm_deassign_ioeventfd_idx() {
>> 	...
>> 	p = get_ioeventfd(kvm, bus_idx, args);
>> 	if (p) {
>> 		assert(p->eventfd == eventfd);
>> 		...
>> 	}
>>
>> ---8<----
>>
>> This may be easier to understand (keep the same logic in assign/deassign).
> 
> I think you should also warn if:
> 
> 1) p->length != args->len
>
> 2) p->wildcard != args->wildcard if p->length
>
> 3) p->datamatch != args->datamatch if p->length && !p->wildcard
> 
> but yeah it sounds like a plan.
> 

I will try to do this. :)

Zhenyu