2.4.x parport_serial link order bug (only works as a module)

2.4.x parport_serial link order bug (only works as a module)

[Marek Michalkiewicz]

Hello,

I've just looked at the 2.4.x changelog (up to 2.4.23-rc2) and still
don't see any fix for the parport_serial link order bug fix.  Without
it, the driver (which handles various PCI multi I/O serial+parallel
cards) only works as a module (broken when compiled into the kernel,
because parport_serial must be initialised after serial).

I've tried to submit the fix a few times since 2.4.19 or so, with
no success so far.  Is there any hope that it would go into 2.4.23?

The patch is here (can be updated to 2.4.23-rc if you are interested):

http://www.amelek.gda.pl/linux-patches/2.4.21/00_parport_serial

Quite big, but the largest part of it simply moves parport_serial.c
from drivers/parport/ to drivers/char/ without changing a single line
inside the file.  Really, no 2-line local root backdoors inserted ;-)

In the same directory, you can also find the NetMos patch, which should
be applied after the parport_serial link order bugfix patch.  Yes, I'm
still using a few NM9835 cards, no problems except having to patch each
kernel version forever.  This boring task would be easier for me if at
least the simple parport_serial link order fix would be accepted...

Thanks,
Marek

Re: 2.4.x parport_serial link order bug (only works as a module)

[Marek Michalkiewicz]

Hi,

On Thu, Nov 20, 2003 at 10:09:39AM -0800, Ed Vance wrote:
> 
> I think it should go into 2.4. It fixes this particular ordering issue
> without mucking about with the ordering mechanism or even adding code. I'm
> sorry to say that I had forgotten about it. It's a good minimum-change
> brute-force umm... solution. No bloat or complexity issues. 
> 
> At your convenience, please move the parport_serial.c file and apply the
> corresponding Makefile changes. 

Now that 2.4.23 is out, could you try to help me getting this (and
possibly NetMos changes) into 2.4.24?  I've just read that 2.4.24
is the last release for which new drivers are accepted, and it's
not even a new driver...  Tim Waugh has been silent about NetMos
problems, and the current version of my patch (made for 2.4.21,
but should apply cleanly to 2.4.23) makes NetMos support a config
option, conditional on CONFIG_EXPERIMENTAL.

Thanks,
Marek

RE: 2.4.x parport_serial link order bug (only works as a module)

[Ed Vance]

Hi Marcelo,

on Thu, November 20, 2003 at 7:23 AM, Marek Michalkiewicz wrote:
> I've just looked at the 2.4.x changelog (up to 2.4.23-rc2) and still
> don't see any fix for the parport_serial link order bug fix.  Without
> it, the driver (which handles various PCI multi I/O serial+parallel
> cards) only works as a module (broken when compiled into the kernel,
> because parport_serial must be initialised after serial).
> 
> I've tried to submit the fix a few times since 2.4.19 or so, with
> no success so far.  Is there any hope that it would go into 2.4.23?
> 
> The patch is here (can be updated to 2.4.23-rc if you are interested):
> 
> http://www.amelek.gda.pl/linux-patches/2.4.21/00_parport_serial
> 
> Quite big, but the largest part of it simply moves parport_serial.c
> from drivers/parport/ to drivers/char/ without changing a single line
> inside the file.  Really, no 2-line local root backdoors inserted ;-)
> 
> In the same directory, you can also find the NetMos patch, 
> which should
> be applied after the parport_serial link order bugfix patch.  Yes, I'm
> still using a few NM9835 cards, no problems except having to 
> patch each
> kernel version forever.  This boring task would be easier for me if at
> least the simple parport_serial link order fix would be accepted...

Russell King looked at this about a year ago and found no specific flaw with
moving the parport_serial.c file from drivers/parport/ to drivers/char to
solve the ordering issue: 

On Mon, Sep 30, 2002 at 2:49 AM, Russell King wrote:
[snip] 
> Other than it's a gross hack rather than a fix.  However, for 2.4, I
> think this is probably the best solution without creating a risk of
> other init ordering problems.  Ed, any comments?
> 
> In 2.5, its easier to solve; we just need to make sure serial is
> initialised before parport.  This is easy, since serial now has its
> own drivers/serial subdirectory.

I think it should go into 2.4. It fixes this particular ordering issue
without mucking about with the ordering mechanism or even adding code. I'm
sorry to say that I had forgotten about it. It's a good minimum-change
brute-force umm... solution. No bloat or complexity issues. 

At your convenience, please move the parport_serial.c file and apply the
corresponding Makefile changes. 

cheers,
Ed Vance