2.6.0 modules, hotplug, PCMCIA

2.6.0 modules, hotplug, PCMCIA

[Norman Diamond]

In 2.6.0-test days, with SuSE 8.2, I reported a number of problems with
modules.  Most of these remain unchanged in 2.6.0 release.  Although SuSE
8.2 has a version of modutils which should be new enough (newer than the
Post Halloween document was at the time that someone first pointed me to a
working link for that document), I upgraded anyway.  I also updated hotplug
scripts.

1.  "rpm -Uvh modutils-2.4.25-1.i386.rpm" gave an error message that the
sensors package depends on the modules package.  This makes no sense to me,
what is there in the upgrade that could break this dependency?  I added
the --nodeps flag and did the upgrade.

2.  I also installed module-init-utils 0.9.12 from the bz2 download.  Sorry,
I noticed that 0.9.14 is available without a "pre" on it, but didn't
download it yet.  If the items in this paragraph have been fixed then please
kindly ignore this paragraph but still consider the others.  Anyway I
followed the instructions in the README:
   ./configure --prefix=/
   make moveold
   make
   make install
   ./generate-modprobe.conf /etc/modprobe.conf
   depmod 2.6.0
Sorry I am not copying the dozens of error messages from
./generate-modprobe.conf, but anyone with SuSE 8.2 can reproduce them.  When
there was no improvement, I looked at the instructions in the FAQ.  I'm not
running RedHat and I did upgrade to later than 0.9.9.  But things still
don't load properly, for which the second answer in the FAQ recommends using
non-existent file "generate-module.conf".  Presumably if that command
existed then it would overwrite the existing /etc/module.conf file.  So I
only obeyed the README not the FAQ.

3.  I also installed hotplug 2003_08_05 from the bz2 download.  I followed
the instructions in the README and they did resolve some boot-time problems.
However, the README says:  "There's more hotplug-related work yet to be done
as part of the 2.5 series; a 2.6 update will be required."  I did not find a
newer hotplug package or 2.6 update on ftp.jp.kernel.org.

4.  SuSE 8.2 defaults to using the kernel PCMCIA package rather than the
external PCMCIA package.  This is fine with me so kernel 2.6.0 also uses its
own compiled PCMCIA drivers instead of trying to make an external PCMCIA
package work with two kernels.  It seems to me that it should be OK to
compile PCMCIA as modules instead of built-in, but there were boot-time
errors, so I had to change PCMCIA and Yenta to built-in.  (This is the
opposite of the change that I had to make to mice, described in a separate
e-mail message.)  Now with PCMCIA compiled built-in, the low-level drivers
get loaded, but cardmgr still doesn't run automatically.  I can do "su" and
"cardmgr &" and then PCMCIA starts working enough to do modprobes when cards
are inserted.

5.  However, file /etc/pcmcia/serial.opts is still getting ignored under
2.6.0.  The modem is detected as containing a TI 16750 UART, and whatever
the serial driver does then, it causes the modem to hang up.  The serial
driver in 2.4.20 defaults to the same thing but 2.4.20 reads file
/etc/pcmcia/serial.opts, obeys the line SERIAL_OPTS="uart 16550A", and lets
the modem operate at 33% of its rated speed instead of hanging up.

6.  Boot messages remain schizophrenic about whether modules are enabled or
not:
   Inspecting /boot/System.map-2.6.0
   Loaded 28865 symbols from /boot/System.map-2.6.0.
   Symbols match kernel version 2.6.0.
   No module symbols loaded - kernel modules not enabled.
But lsmod shows that some modules are loaded, and there are other messages
in the system log showing that (after hacks described above) PCMCIA
correctly loaded some modules.

7.  Linux 2.4.20 was willing to drive the sound chip to some degree, a
Neomagic NM256.  KDE produces a short sound during KDE's startup under
2.4.20 but not under 2.6.0.  Under 2.6.0 the driver sometimes gives errors
about the sound chip not existing, sometimes it recommends loading the
module with option force_ac97=1, if I use option force_ac97=1 then it says
that option doesn't exist but then it tells me to try that option again,
etc.  But even under 2.4.20, the only sound I got was KDE's startup sound,
XMMS and other stuff never produced sound.

8.  Connecting a USB hard disk is recognized.  Under 2.4.20 some agent
automatically adds entries to /etc/fstab and the user can mount the
partitions.  Under 2.6.0 there is no automatic update to /etc/fstab and only
the superuser can mount the partitions.

9.  There are still other boot-time errors under 2.6.0 which do not occur
under 2.4.20.  They do not get logged to /var/log/messages or dmesg.  There
are fewer than they used to be but still they scroll off the screen very
quickly and I cannot pause the display to copy them down.  There are 17
fatal errors about non-existent devices 4-72 to 4-87 and 212-something.  The
cat command must be working a double shift because its 18th life boots
Linux.

Re: 2.6.0 modules, hotplug, PCMCIA

[Russell King]

On Sun, Dec 28, 2003 at 11:33:04AM +0900, Norman Diamond wrote:
> 4.  SuSE 8.2 defaults to using the kernel PCMCIA package rather than the
> external PCMCIA package.  This is fine with me so kernel 2.6.0 also uses its
> own compiled PCMCIA drivers instead of trying to make an external PCMCIA
> package work with two kernels.  It seems to me that it should be OK to
> compile PCMCIA as modules instead of built-in, but there were boot-time
> errors, so I had to change PCMCIA and Yenta to built-in.

What were these errors?

> (This is the
> opposite of the change that I had to make to mice, described in a separate
> e-mail message.)  Now with PCMCIA compiled built-in, the low-level drivers
> get loaded, but cardmgr still doesn't run automatically.  I can do "su" and
> "cardmgr &" and then PCMCIA starts working enough to do modprobes when cards
> are inserted.

It sounds like the SuSE init scripts are being clever and probably only
know about how their 2.4 situation works.  What we need is a SuSE person
to comment on this behaviour; I don't have access to any SuSE based
systems to investigate their quirks.

> 5.  However, file /etc/pcmcia/serial.opts is still getting ignored under
> 2.6.0.

"still" ?  This is news to me (as the guy who seems to be handling both
PCMCIA and serial.)

> The modem is detected as containing a TI 16750 UART, and whatever
> the serial driver does then, it causes the modem to hang up.  The serial
> driver in 2.4.20 defaults to the same thing but 2.4.20 reads file
> /etc/pcmcia/serial.opts, obeys the line SERIAL_OPTS="uart 16550A", and lets
> the modem operate at 33% of its rated speed instead of hanging up.

"hang up"?  Do you mean "on-hook" or do you mean "stop working"?  Is
there anything in /var/log/messages about this?

On my RH systems, cardmgr logs a fair amount to the system messages log,
which includes details of any commands run and any failures.  It would
be really useful to see this.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core

Re: 2.6.0 modules, hotplug, PCMCIA

[Norman Diamond]

Russell King replied to me:

> > 4.  SuSE 8.2 defaults to using the kernel PCMCIA package rather than the
> > external PCMCIA package.  This is fine with me so kernel 2.6.0 also uses its
> > own compiled PCMCIA drivers instead of trying to make an external PCMCIA
> > package work with two kernels.  It seems to me that it should be OK to
> > compile PCMCIA as modules instead of built-in, but there were boot-time
> > errors, so I had to change PCMCIA and Yenta to built-in.
>
> What were these errors?

I think it was that the PCMCIA core and Yenta modules didn't even get
loaded, I had to modprobe them.  The need to type "cardmgr &" by hand
remained necessary even after recompiling with them built-in and rebooting.

> It sounds like the SuSE init scripts are being clever and probably only
> know about how their 2.4 situation works.

I agree.

> > 5.  However, file /etc/pcmcia/serial.opts is still getting ignored under
> > 2.6.0.
>
> "still" ?  This is news to me (as the guy who seems to be handling both
> PCMCIA and serial.)

"still" as in "even after compiling the drivers as Y instead of M, and
typing the cardmgr command by hand".  Not "still" as in "[not] same as 2.4"
because this bug does not occur in 2.4.

> > The modem is detected as containing a TI 16750 UART, and whatever
> > the serial driver does then, it causes the modem to hang up.  The serial
> > driver in 2.4.20 defaults to the same thing but 2.4.20 reads file
> > /etc/pcmcia/serial.opts, obeys the line SERIAL_OPTS="uart 16550A", and lets
> > the modem operate at 33% of its rated speed instead of hanging up.
>
> "hang up"?  Do you mean "on-hook" or do you mean "stop working"?  Is
> there anything in /var/log/messages about this?

I think both.  "on-hook" happens immediately, and the only way to try again
is to eject the card and reinsert it.

> On my RH systems, cardmgr logs a fair amount to the system messages log,
> which includes details of any commands run and any failures.  It would
> be really useful to see this.

As far as I can tell, in SuSE 8.2 with SuSE's default version of the 2.4.20
kernel, this isn't happening, but in 2.6.0 it is happening even when I
didn't ask for it.  /var/log/messages gets around 20,000 lines of messages
every day that I experiment with 2.6.0, and 90% of it is PCMCIA stuff.  I
think I'm capable of e-mailing it to you if it will help.  Actually later
today I might have time to experiment with the modem and 2.6.0 again, and
might be able to grep for appropriate stuff in /var/log/messages.

Re: 2.6.0 modules, hotplug, PCMCIA

[Rusty Russell]

On Sun, 28 Dec 2003 11:33:04 +0900
"Norman Diamond" <ndiamond@wta.att.ne.jp> wrote:

> Sorry I am not copying the dozens of error messages from
> ./generate-modprobe.conf, but anyone with SuSE 8.2 can reproduce them.

Yes, it's kind of noisy.  I've put some silencers on it for the next release.

> When
> there was no improvement, I looked at the instructions in the FAQ.  I'm not
> running RedHat and I did upgrade to later than 0.9.9.  But things still
> don't load properly, for which the second answer in the FAQ recommends using
> non-existent file "generate-module.conf".  Presumably if that command
> existed then it would overwrite the existing /etc/module.conf file.  So I
> only obeyed the README not the FAQ.

I've fixed this too.  You've already done it as part of the README.

Good luck...
Rusty.
-- 
   there are those who do and those who hang on and you don't see too
   many doers quoting their contemporaries.  -- Larry McVoy

Re: 2.6.0 modules, hotplug, PCMCIA

[Norman Diamond]

I replied to Russell King:

> > > 5.  However, file /etc/pcmcia/serial.opts is still getting ignored under
> > > 2.6.0.
> > > The modem is detected as containing a TI 16750 UART, and whatever
> > > the serial driver does then, it causes the modem to hang up.  The serial
> > > driver in 2.4.20 defaults to the same thing but 2.4.20 reads file
> > > /etc/pcmcia/serial.opts, obeys the line SERIAL_OPTS="uart 16550A", and lets
> > > the modem operate at 33% of its rated speed instead of hanging up.
> >
> > "hang up"?  Do you mean "on-hook" or do you mean "stop working"?  Is
> > there anything in /var/log/messages about this
>
> I think both.  "on-hook" happens immediately, and the only way to try again
> is to eject the card and reinsert it.
>
> > On my RH systems, cardmgr logs a fair amount to the system messages log,
>
> As far as I can tell, in SuSE 8.2 with SuSE's default version of the 2.4.20
> kernel, this isn't happening, but in 2.6.0 it is happening even when I
> didn't ask for it.

Sorry for misstating that.  It's actually the other way around.  2.6.0 is
not giving me a ton of unrequested log syslog messages, but neither is it
giving any reason for the modem to stop working.  It only observes that the
modem hung up with no visible reason, and it doesn't even notice that the
modem cannot be retried until ejection and reinsertion.  In both 2.4.20 and
2.6.0, the serial_cs driver detects the UART as a TI 16750, which either
means that the UART really is a 16750 but the chip and the driver don't
agree on how the chip should be handled, or that the UART is a 16550A (or
something else) being misdetected.

SuSE's 2.4.20 PCMCIA configuration is writing around 6,000 syslog messages
every time I boot it.  I don't think you want that, especially when it's
from 2.4.20 which is obeying my addition in file /etc/pcmcia/serial.opts
saying  SERIAL_OPTS="uart 16550A"