Am 15.03.2012 15:42, schrieb Stefan Hajnoczi:
> On Thu, Mar 15, 2012 at 1:37 PM, Anthony Liguori <aligu...@us.ibm.com> wrote:
>> + sock = socket(PF_UNIX, SOCK_STREAM, 0);
>> + g_assert_no_errno(sock);
>> +
>> + addr.sun_family = AF_UNIX;
>> + snprintf(addr.sun_path, sizeof(addr.sun_path), "%s", socket_path);
>> +
>> + pid = fork();
>> + if (pid == 0) {
>> + command = g_strdup_printf("%s "
>> + "-qtest unix:%s,server,nowait "
>> + "-qtest-log /dev/null "
>> + "-pidfile %s "
>> + "-machine accel=qtest "
>> + "%s", qemu_binary, socket_path,
>> + pid_file,
>> + extra_args ?: "");
>> +
>> + ret = system(command);
>> + exit(ret);
>> + g_free(command);
>> + }
>> +
>> + do {
>> + sleep(1);
>> + ret = connect(sock, (struct sockaddr *)&addr, sizeof(addr));
>> + } while (ret == -1);
>
> I believe Kevin suggested using -qtest unix:%s and creating the listen
> socket in the test program rather than inside QEMU. The advantage is
> that we never sleep(3), instead we accept(2) the connection from QEMU
> and get going right away.
>
> This can be added as a patch later.
>
> Kevin: Do you already have a patch or does someone need to write it?
I just complained (and reduced the delay locally). Paolo suggested a
real solution, which may have been what you describe, I don't remember
the details.
Kevin
