PCMCIA breaks suspend-to-(disk|ram) with 2.6.11

[Sebastian Kügler]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Dominik,

On Wednesday 02 March 2005 23:21, Dominik Brodowski wrote:
> [moving this to linux-pcmcia list where it is better suited]

Good point, I'm subscribed now.

> On Wed, Mar 02, 2005 at 07:24:37PM +0100, Sebastian Kügler wrote:
> > Unloading orincoco_cs and then yenta_socket hangs rmmod in D state.
>
> Does issuing a "cardctl eject" in between help? Or do you do this before?
> I'd think orinoco_cs has usage count 1 during normal use (even with
> interface down).

Exactly. I didn't use cardctl before, now I do and unloading of the modules is 
much more reliable, which in turn makes suspend reliable. (Still not sure why 
keeping the modules loaded breaks at least S4. (S3 seems wo work, but not a 
100% good, as some brief testing revealed.)

> > Both, S3 and S4 work fine without the PCMCIA adapter inserted. My suspend
> > scripts unload PCMCIA and related stuff and are supposed to stop cardmgr.
> > (I have problems resuming without the PCMCIA card plugged in when I
> > suspended with it inserted, making swsusp hang about at the end of
> > resuming with "orinoco_lock() called with hw_unavailable
> > (dev=de08f800)".)
>
> Is this a regression, too?

WRT to S4, this is probably a regression (I think I could resume with earlier 
versions without having all the stuff unloaded, but I'm not a 100% sure. I 
can do some testing with these kernels if you'd like to be sure, just let me 
know  in that case.) In my memory PCMCIA worked fine without script magic up 
to 2.6.9, for 2.6.10 I added removing and rmmodding PCMCIA stuff. (I can't 
recall every single state of my suspend scripts over the last year, so I 
might aswell be wrong.)

I still have problems resuming with all the PCMCIA stuff loaded, but cardctl 
helps a lot in getting the modules unloaded, so I this is getting better 
(although not having to unload would be even better, of course.)

> > I am getting this oops if I try to rmmod -f orinoco_cs while the card is
> > inserted, the problem seems to boil down to that (This oops obviously
> > renders the PCMCIA card unusable until I reboot):
>
> cardctl eject or
>
> echo "42" > /sys/class/pcmcia_socket/pcmcia_socket%n/card_eject
>
> should allow you to rmmod orinoco_cs and then yenta_socket. Actually, you
> won't need to rmmod orinoco_cs then as the kernel will behave as if there
> were no card inserted to the socket then.

Only cardctl eject'ing does still produce hangs on resuming, it seems like 
powermanagement of one of the modules is broken, probably pcmcia as some 
earlier testing with - I think - 2.6.11-rc3 revealed. Do you know of any good 
reason why a /etc/init.d/pcmcia stop doesn't do a cardctl eject equivalent, 
or should I bug the package maintainers (Debian Sid, btw)?

> On a sidenote, I intend to allow to do a "rmmod MODULE" when a matching
> card is inserted in a few weeks, but only after some more cleanups have
> gone in. Currently, it'd mean nasty little races.

Sounds like I stumbled into that by force-unloading the modules...

> > On a different note, is there a split out version of the PCMCIA hotplug
> > stuff recently merged into -mm? I'd like to try it and see if it fixes my
> > issues, but I'd rather not introduce too much new deltas.
>
> everything is at
> http://kernel.org/pub/linux/kernel/people/brodo/patches/2.6.11/
> [actually, it isn't yet, but should appear there any minute]
>
> and patches are applied in the order of the numbering [-02b- after -02-].
> However, I don't think these patches change anything with regard to the
> problems you're experiencing with PCMCIA. Nonetheless, these patches enjoy
> being tested ;)

I will give them a shot and report back, hotplug seems to be the way these 
kinds of devices are handled (it works fine for me for USB).

>  Dominik

Thanks, 2.6.11 now seems much more usable to me, and I gained some more 
experience to help users to get their system sleeping (and resuming!) while 
powermanagement in some linux drivers is still a little flaky.

Cheers,

sebas
- -- 
  http://vizZzion.org   |   GPG Key ID: 9119 0EF9 
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Accident, n.: A condition in which presence of mind is good, but absence of 
body is better.  - Foolish Dictionary

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.0 (GNU/Linux)

iQEVAwUBQiZtlWdNh9WRGQ75AQLFQwgAoegX4bnssInOGXF6avvy7O5mNSIFj4+C
PsOm6dmhph8WVVVaRY0gas9gyPyJRQZBe4ju2P1IF7voB6lItmK6gcx19xBpBYiB
V8mkVlV0LAt/igNJRS6oVQjEww30vP+4/ILpYzokSGEOERfM5Eo91F5pB9SIThtr
Uciryip7nt8Km1g6j0r1IyTT/iJV3H6of7PY82M44rZjbD6O4ZeUJtqfFXstFssD
3pmkThK3b9io0HaWAbBs0IBNgzcF9P9cxgTg7w+gl65S3JU8MbagTJzgYF/Mntea
6qGdWprrGShQ9HM2yiyOrYHr39zu9tGWVekIcDXG4ZlVlbp/FyrnLA==
=AzAJ
-----END PGP SIGNATURE-----