Re: [hpoj-devel] 2.4.0 kernel issues

[Burkhard K. <bu...@bu...>]

David Paschal
> All I see in your 2.4 debug output is an open of the scan channel immediately
> followed by a close.  I assume this output corresponds to the SANE+PTAL
> output you sent on September 20.  In that case, we saw a successful open,
> and the write() of two bytes (the SCL "Esc-E" reset command) failed (returned
> -1), so the SANE backend had no choice but to give up and close the channel.
> I would suggest examining the code path taken by writes (see mlc_sendmsg()).

David, that set me onto the right track. The culprit was an old bug in
mlc_sendmsg which just happened to remain silent with 2.2 kernel version
because no flag in msghdr->flags was set from the socket layer:

    int nonblock = msg->msg_flags & MSG_DONTWAIT;

    /* XXX currently does not work with poll() */
    nonblock = 0;

    if (debug & 0x02)
        printk ("mlc_sendmsg: %p credit=%d\n", sock, so->mycredit);
    if (msg->msg_flags & ~MSG_DONTWAIT) 
                         ^^^^^^^^^^^^^^
        return = -EOPNOTSUPP;

The code checked for the supplement of MSG_DONTWAIT instead of the
flag itself. As long as no flag was set, all worked well but as 
soon as any flag besides MSG_DONTWAIT is set, the call will fail with
an error return of -95, which is -EOPNOTSUPP. That's exactly 
what happend with a 2.4.0-test kernel, msg_flags was set to 
MSG_EOR, thus revealing this bug.

I' ll be posting my patch seperately, asking everybody with 
a 2.4 kernel to test it out.

I see a lot of "already open" messages from the ptal layer - maybe
we have other problems as well, but "scanimage --test" runs on my 
box now successfully.

BTW: in ieee12844/Makefile.in the "cp" expression to move kernel
modules to the appropriate location was commented out. Was this
done for any particular purpose? I am asking because it meant
that the modules had to be installed by hand.

Burkhard
-- 
Burkhard Kohl			buk at/auf buks.ipn.de