Re: [hylafax-users] Request For Linux Kernel Help: Was 2.6.26 Breaks USB Modem

[Frank Peters <frank.peters@xxxxxxxxxxx>]

On Mon, 25 Aug 2008 08:58:23 -0700
Lee Howard <lee.howard@xxxxxxxxxxxx> wrote:

> 
> Are you sure that the problem is limited to USB modems?  Have you tested 
> with a different hardware modem?  Who acknowledged the problem as a 
> flow-control issue in the kernel?
> 

After applying some patches that were provided by the kernel
developers, I (and the list) received this terse message from
Oliver Neukum, the maintainer of the cdc-acm module:

"Yes, I broke flow control. Would it bother you a lot to
bisect this problem with git?"

Not being at all intimate with the details of the kernel
all I can do is relate these messages.

No, I haven't tried any other modem, although I do have 
access to a PCI-based serial modem.  This would require
a complete re-configure of my current kernel to include
serial drivers.  Legacy serial ports are essentially obsolete
on the newest hardware.

My intention was just to report the problem and to leave
the fixing to people more qualified for the purpose than
I.  Actually, the problem was first noticed by me several
weeks ago, and only after failing to notice any mention
of it on any lists of forums did I think of informing
the kernel developers.


> The reason I ask:
> 
> http://www.mail-archive.com/linux-kernel@xxxxxxxxxxxxxxx/msg188348.php
> 
> If you read through that entire thread you may find that there is some 
> degree of ambiguity in the presumed diagnosis of "flow control 
> problem"... because it actually seemed more-likely that the buffer 
> overrun was due to interrupt latency caused by the heavy-used ATA driver 
> getting preferential treatment on its interrupts.  Note that the thread 
> eventually got Alan Cox's attention - who I believe is responsible for 
> the serial driver these days - however the problem was never 
> pinpointed... or, rather, a solution was never identified (even after 
> spending a lot of effort with real-time kernel support, see: 
> http://hylafax.sourceforge.net/howto/install.php#ss2.6).
> 

I recall that this serial driver issue was mentioned on the HylaFAX
list some months back.  Since I was transitioning to USB at about the
same time I considered it unimportant to my needs.  It is important
to know that is has not been resolved.


> My advice to you would be step back and stay with 2.6.25 until a larger 
> mass of users are running 2.6.26 such that it could attract the 
> attention of someone capable of actually fixing the problem head-on (and 
> they would most-likely be those you can see responding to serial 
> questions on linux-serial or linux-kernel mailing lists).
> 

That is essentially what I am doing now.  My intention was only
to report the problem.  Since I have no experience in kernel
development there is little I could do help fix the issue.  If
asked to apply a patch or do some kind of test I would certainly
be willing, but providing insight into the nature of the problem
is simply beyond my current capabilities.


> In my experience bug fixes to the Linux kernel either come very quickly 
> in response to the bug report or they become very lengthy and laborious 
> processes which really require the assistance from, if not advocacy of, 
> someone with a good deal of Linux kernel development experience.
> 
> Sometimes issues such as interrupt latency become very complicated, and 
> it's really impossible to undo a commit because it was part of a whole 
> transformation of internal kernel API.  In those cases a fix has to come 
> from direct inspection and creative problem-solving, rather than trying 
> to back-step through commits in order to "fix" something that isn't 
> really broken per the contributor.
> 

I can certainly appreciate this.  In fact, I had reported a different
problem with the cdc-acm USB modem driver after an earlier kernel
release.  That problem was fixed within a few days.  But, as you say,
the nature of the issue could be so intertwined with everything else
that a quick solution may be impossible.

But at least HylaFAX users are adequately forewarned.

Frank Peters


____________________ HylaFAX(tm) Users Mailing List _______________________
  To subscribe/unsubscribe, click http://lists.hylafax.org/cgi-bin/lsg2.cgi
 On UNIX: mail -s unsubscribe hylafax-users-request@xxxxxxxxxxx < /dev/null
  *To learn about commercial HylaFAX(tm) support, mail sales@xxxxxxxxx*