Il 12/02/2013 21:49, Michael S. Tsirkin ha scritto:
> On Tue, Feb 12, 2013 at 09:08:27PM +0100, Paolo Bonzini wrote:
>> Il 12/02/2013 19:23, Michael S. Tsirkin ha scritto:
>>> On Tue, Feb 12, 2013 at 07:04:27PM +0100, Paolo Bonzini wrote:
>>>>>> Perhaps, but 3 or 4 arguments (in/out/nsg or in/out/nsg_in/nsg_out) just
>>>>>> for this are definitely too many and make the API harder to use.
>>>>>>
>>>>>> You have to find a balance. Having actually used the API, the
>>>>>> possibility of mixing in/out buffers by mistake never even occurred to
>>>>>> me, much less happened in practice, so I didn't consider it a problem.
>>>>>> Mixing in/out buffers in a single call wasn't a necessity, either.
>>>>>
>>>>> It is useful for virtqueue_add_buf implementation.
>>>>
>>>> ret = virtqueue_start_buf(vq, data, out + in, !!out + !!in,
>>>> gfp);
>>>> if (ret < 0)
>>>> return ret;
>>>>
>>>> if (out)
>>>> virtqueue_add_sg(vq, sg, out, DMA_TO_DEVICE);
>>>> if (in)
>>>> virtqueue_add_sg(vq, sg + out, in, DMA_FROM_DEVICE);
>>>>
>>>> virtqueue_end_buf(vq);
>>>> return 0;
>>>>
>>>> How can it be simpler and easier to understand than that?
>>>
>>> Like this:
>>>
>>> ret = virtqueue_start_buf(vq, data, in, out, gfp);
>>> if (ret < 0)
>>> return ret;
>>>
>>> virtqueue_add_sg(vq, sg, in, out);
>>>
>>> virtqueue_end_buf(vq);
>>
>> It's out/in, not in/out... I know you wrote it in a hurry, but it kind
>> of shows that the new API is easier to use. Check out patch 8, it's a
>> real improvement in readability.
>
> That's virtqueue_add_buf_single, that's a separate matter.
> Another option for _single is just two wrappers:
> virtqueue_add_buf_in
> virtqueue_add_buf_out
I like it less, but yes this one would be ok (no driver uses a variable
for the enum parameter).
>> Plus you haven't solved the problem of alternating to/from-device
>> elements (which is also harder to spot with in/out than with the enum).
>
> Yes it does, if add_sg does not have in/out at all there's no way to
> request the impossible to/from mix.
In your example above it does have it. I assume you meant
ret = virtqueue_start_buf(vq, data, out, in, gfp);
if (ret < 0)
return ret;
virtqueue_add_sg(vq, sg, out + in);
virtqueue_end_buf(vq);
>>>> virtqueue_add_buf and virtqueue_add_sg are very different, despite the
>>>> similar name.
>>>
>>> True. The similarity is between _start and _add_buf.
>>> And this is confusing too. Maybe this means
>>> _start and _add_sg should be renamed.
>>
>> Maybe. If you have any suggestions it's fine.
>>
>> BTW I tried using out/in for start_buf, and the code in virtio-blk gets
>> messier, it has to do all the math twice.
>
> I'm pretty sure we can do this without duplication, if we want to.
Indeed, if you remove the out/in arguments from _sg there is no
duplication in virtio-blk. That's because it places data-out at the end
and data-in at the beginning (so data is always after the request header
and before the response footer).
>> Perhaps we just need to
>> acknowledge that the API is different and thus the optimal choice of
>> arguments is different. C doesn't have keyword arguments, there not
>> much that we can do.
>
> Yea, maybe. I'm not the API guru here anyway, it's Rusty's street.
Let's wait for him.
Paolo
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
