On (Tue) Apr 20 2010 [13:32:18], Gerd Hoffmann wrote:
>> static void tcp_chr_connect(void *opaque)
>> {
>
>> + chr->write_blocked = false;
>> s->connected = 1;
>> qemu_set_fd_handler2(s->fd, tcp_chr_read_poll,
>> - tcp_chr_read, NULL, chr);
>> + tcp_chr_read, tcp_chr_write_unblocked, chr);
>
> This is wrong. You want register tcp_chr_write_unblocked only for the
> chr->write_blocked == true (i.e. output buffers are full) case.
> Otherwise qemu will burn cpu calling tcp_chr_write_unblocked.
>
> Yes, you'll have to call qemu_set_fd_handler2 each time write_blocked
> changes state.
Agreed. That's coming in the next round when I make it generic enough
for all the backends to call the common code.
> Also implementing the whole logic at the individual chardev drivers
> level feels somewhat wrong as it will identical for most (all?) unix
> chardev drivers.
Yes, both these come together, I guess.
I sent out this series as a "feeler" to see if the approach was
acceptable.
Paul didn't reply to my reply addressing his concern, so I take that as
he's OK with the approach as well :-)
Amit
