Re: [PATCH v3] ring.h: introduce macros to handle monodirectional rings with multiple req sizes

[Oleksandr Andrushchenko <Oleksandr_Andrushchenko@epam.com>]

Hi, Stefano!

On 02/22/2017 07:10 PM, Stefano Stabellini wrote:
> On Wed, 22 Feb 2017, Oleksandr Andrushchenko wrote:
>> Hi, Stefano, Jan!
>> 1. Stefano, are you still NOT considering adding
>> functionality to avoid memory copying? We discussed
>> this a little bit here [1].
> Hi Oleksandr,
>
> these macros are the generic versions of what pvcalls and xen-9pfs need,
> and these protocols use memory copies.
These macros will live in generic ring.h. It means that
they can be used not only by pvcalls/9pfs, but others are
allowed to use them too (kbdif/fbif/displif/??).
That being said, I am not convinced that this is a good idea
to introduce a memory copying while dealing with the packets
in an IRQ handler, for example.
So, my point is that we should give a possibility to directly
access ring's contents, which will not be used in your case.
>   I cannot introduce memory sharing
> interfaces as part of this patch, because they wouldn't comply to
> pvcalls or xen-9pfs
Again, you are thinking of pvcalls/9pfs, but none of the macros
say they are for these use-cases: anyone can use them
>   and I don't have another test case for them.
> But if you had a patch to introduce them in ring.h, I would be happy to
> help you review it.
>
No, I don't have any, sorry
>> 2. Will you also provide macros/inlines for fixed sized packets?
>> So, others do not reinvent the wheel again on top of your code.
> I thought I already did: you can read/write fixed sized packets using
> the two read/write_packet functions.
I was thinking of something like we have for req/resp
     DEFINE_RING_TYPES(fsif, struct fsif_request, struct fsif_response);
so you don't need to care of sizeof(req/resp/event)

Thank you,
Oleksandr

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