evtchn_resize_ring() serialization

evtchn_resize_ring() serialization

[Jan Beulich]

David,

in the course of putting together the fix just sent I got puzzled by the
apparent raciness of the initial part of evtchn_resize_ring(). Is it
intentional that the function relies on being implicitly serialized via the
bind mutex? If so, it would seem that this should be stated in a
comment. If not, I think that part of the function would need some
adjustment.

Thanks, Jan


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

Re: evtchn_resize_ring() serialization

[David Vrabel]

On 04/05/16 14:08, Jan Beulich wrote:
> David,
> 
> in the course of putting together the fix just sent I got puzzled by the
> apparent raciness of the initial part of evtchn_resize_ring(). Is it
> intentional that the function relies on being implicitly serialized via the
> bind mutex?

Yes.

Although, I'm not sure I'd call it "implicit", as I think the bind_mutex
rather explicitly serializes bind operations (which may include resizing
the ring).

David

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

Re: evtchn_resize_ring() serialization

[Jan Beulich]

>>> On 04.05.16 at 15:14, <david.vrabel@citrix.com> wrote:
> On 04/05/16 14:08, Jan Beulich wrote:
>> David,
>> 
>> in the course of putting together the fix just sent I got puzzled by the
>> apparent raciness of the initial part of evtchn_resize_ring(). Is it
>> intentional that the function relies on being implicitly serialized via the
>> bind mutex?
> 
> Yes.
> 
> Although, I'm not sure I'd call it "implicit", as I think the bind_mutex
> rather explicitly serializes bind operations (which may include resizing
> the ring).

Well, that's only if you assume that re-sizing is something which can
only ever happen in the context of binding. As said - I had looked at
the function in isolation first; nothing says it's a helper for just one
very specific operation.

Jan


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

Re: evtchn_resize_ring() serialization

[David Vrabel]

On 04/05/16 14:38, Jan Beulich wrote:
>>>> On 04.05.16 at 15:14, <david.vrabel@citrix.com> wrote:
>> On 04/05/16 14:08, Jan Beulich wrote:
>>> David,
>>>
>>> in the course of putting together the fix just sent I got puzzled by the
>>> apparent raciness of the initial part of evtchn_resize_ring(). Is it
>>> intentional that the function relies on being implicitly serialized via the
>>> bind mutex?
>>
>> Yes.
>>
>> Although, I'm not sure I'd call it "implicit", as I think the bind_mutex
>> rather explicitly serializes bind operations (which may include resizing
>> the ring).
> 
> Well, that's only if you assume that re-sizing is something which can
> only ever happen in the context of binding. As said - I had looked at
> the function in isolation first; nothing says it's a helper for just one
> very specific operation.

You could add a comment saying "u->bind_mutex must be locked" if you
think it will help.

David

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