On 13.03.2012 21:44, Paolo Bonzini wrote:
> Il 12/03/2012 20:14, Michael Tokarev ha scritto:
>> + for (i = 0, done = 0; done < bytes && i < iov_cnt; i++) {
>> + if (offset < iov[i].iov_len) {
>> + size_t len = MIN(iov[i].iov_len - offset, bytes - done);
>> + memcpy(iov[i].iov_base + offset, buf + done, len);
>> + done += len;
>> + offset = 0;
>> + } else {
>> + offset -= iov[i].iov_len;
>> }
>> - iovec_off += iov[i].iov_len;
>> }
>> - return buf_off;
>> + assert(offset == 0);
>
> This needs to be assert(offset == 0 || done == 0).
Nope. All these functions actually allow to specify very
large value for `bytes' (and I suggested using -1 for it
to mean "up to the end"). Only offset must be within
the iovec.
>> + return done;
>
> Otherwise looks good, but I must say I do not like changing the API
> *and* the implementation in the same patch. It seems like a
> bisectability nightmare (and reviewing nightmare, too). I do prefer
> your new code though.
I changed it all to be the same and actually verified.
Sure I can split it into two patches -- that should be
easy, will do that in a moment... Just like I did with
qemu_sendv_recvv(). The whole thing is just trivial but.. ;)
Thank you for the review!
> Paolo
/mjt
