Re: [Qemu-devel] [PATCH 4/4] migration: simplify writev vs. non-writev logic

[Juan Quintela <quintela@redhat.com>]

Paolo Bonzini <pbonzini@redhat.com> wrote:
> Il 09/04/2013 13:43, Juan Quintela ha scritto:
>>> > @@ -687,12 +685,10 @@ void qemu_put_byte(QEMUFile *f, int v)
>>> >      f->bytes_xfer++;
>>> >      if (f->ops->writev_buffer) {
>>> >          add_to_iovec(f, f->buf + f->buf_index, 1);
>>> > -        f->buf_index++;
>>> > -    } else {
>>> > -        f->buf_index++;
>>> > -        if (f->buf_index == IO_BUF_SIZE) {
>>> > -            qemu_fflush(f);
>>> > -        }
>>> > +    }
>>> > +    f->buf_index++;
>>> > +    if (f->buf_index == IO_BUF_SIZE) {
>>> > +        qemu_fflush(f);
>>> >      }
>>> >  }
>> If you follow my advice of moving the call to add_to_iovec() you get
>> this one simplified and only one place to do this.
>
> Moving what call?  The apparent complication is because the old logic
> was a bit more involute than necessary.  If you look at the code after
> the patches, not the patches themselves, you'll see for yourself.
>
> The logic now is:
>
>    add byte
>    if using iovs
>        add byte to iov list
>    if buffer full
>        flush
>
> add_to_iovec has no business checking the buffer.  Why should
> qemu_put_buffer_async() check the buffer?

We can rename the function.  I agree with that.

>
> The duplication between qemu_put_byte and qemu_put_buffer is a different
> topic.  I think it's acceptable in the name of performance, but perhaps
> you can just call qemu_put_buffer(f, &c, 1).

Right now,  it is important because all the integer types are sent as
callas to qemu_put_byte(),  but Orit had a patch to rewrote those using
qemu_put_buffer() dirrectly,  and then we don't use so many
qemu_put_byte()'s anymore.

Later,  Juan.