RE: apm suspend broken in 2.4.10

RE: apm suspend broken in 2.4.10

[Alex Cruise]

From: Randy.Dunlap [mailto:rddunlap@osdlab.org]
 
> > Here's the complete list of modules which might typically 
> > be loaded at boot:
> > [ ad nausaeum ]

> Unload some of these (that you don't really need to run)
> and try "apm -s".
> If that fails, unload some more of them and try again...
> That would at least narrow down the search for us.

I already tried that... Maybe my message didn't get through :)  

>AC> Just for fun, I tried removing all of my loaded 2.4.10 modules one by
one,
>AC> and attempting 'apm --suspend' in between, and still had the same
problem
>AC> when I got down to the bare minimum (ext3 and jbd)

Anyway, it looks like something keyboard- or A20-related is vetoing my
suspend request.  Did you get my "the plot thickens" message?  It's not
appearing in the lkml archives, maybe it got lost last night.

-0xe1a

Re: apm suspend broken in 2.4.10

[Randy.Dunlap]

> > Unload some of these (that you don't really need to run)
> > and try "apm -s".
> > If that fails, unload some more of them and try again...
> > That would at least narrow down the search for us.
> 
> I already tried that... Maybe my message didn't get through :)
> 
> >AC> Just for fun, I tried removing all of my loaded 2.4.10 modules one by
> one,
> >AC> and attempting 'apm --suspend' in between, and still had the same
> problem
> >AC> when I got down to the bare minimum (ext3 and jbd)
> 
> Anyway, it looks like something keyboard- or A20-related is vetoing my
> suspend request.  Did you get my "the plot thickens" message?  It's not
> appearing in the lkml archives, maybe it got lost last night.

I saw the keyboard message, and someone else's A20 reference.
Not one from you IIRC.

~Randy

RE: apm suspend broken in 2.4.10

[Lewin A.R.W. Edwards]

>Sep 27 19:39:09 onus kernel: awc: APM Suspend vetoed by:
>Sep 27 19:39:09 onus kernel:   type = 1 id = 1104151299 callback =
>-1072064644 data = 0 flags = 0 state = 0 prev_state = 0
>
>1104151299 is 0x41D00303, which if you consult your include/linux/pm.h
>(PM_SYS_DEV = 1, PM_SYS_KBC = 0x41d00303), would seem AFAICT to indicate
>that it's the keyboard driver--or something upstream of it--who's vetoing my
>suspend.  Am I crazy?

Maybe something to do with A20 switching?


-- Lewin A.R.W. Edwards
Embedded Engineer, Digi-Frame Inc.
Work: http://www.digi-frame.com/
Tel (914) 937-4090 9am-6:30pm M-F ET
Personal: http://www.larwe.com/ http://www.zws.com/

RE: apm suspend broken in 2.4.10

[Alex Cruise]

The plot thickens!  Here's what I did:

In kernel/pm.c, at around line 242, I inserted these lines:

	if (status) {
		// inserted by awc

		printk( "awc: APM Suspend vetoed by: \n");
		printk( "  type = %d", dev->type );
		printk( " id = %d", dev->id );
		printk( " callback = %d", dev->callback );
		printk( " data = %d", dev->data );
		printk( " flags = %d", dev->flags );
		printk( " state = %d", dev->state );
		printk( " prev_state = %d\n", dev->prev_state );
	
		// we now return to your regularly scheduled kernel...

...and after compiling, installing, rebooting and entering "apm --suspend"
at a vc, I found this in syslog:

Sep 27 19:39:09 onus kernel: awc: APM Suspend vetoed by: 
Sep 27 19:39:09 onus kernel:   type = 1 id = 1104151299 callback =
-1072064644 data = 0 flags = 0 state = 0 prev_state = 0

1104151299 is 0x41D00303, which if you consult your include/linux/pm.h
(PM_SYS_DEV = 1, PM_SYS_KBC = 0x41d00303), would seem AFAICT to indicate
that it's the keyboard driver--or something upstream of it--who's vetoing my
suspend.  Am I crazy?

Obviously, apart from the final CR after typing "apm --suspend", I'm not
touching the keyboard before it fails.  I tried asking KLaptop (the KDE
power applet) to suspend using the mouse, and I found exactly the same debug
output repeated in syslog.

-0xe1a

RE: apm suspend broken in 2.4.10

[Alex Cruise]

From: Randy.Dunlap [mailto:rddunlap@osdlab.org]

> > I'm gonna go try the 2.4.7 from RH's "Roswell" beta now.
> OK, thanks for testing that.

2.4.7 has the apm=on bug and won't suspend either... I'm beginning to
suspect that something else is afoot, but I might as well try 2.4.2... 

Funny, with 2.4.2, "apm=on" seems to work as expected, and the first time I
tried "apm --suspend", the screen blanked but nothing else appeared to
happen.  When I hit a key, the suspend command appeared to still be running.
I switched to a different vc and did a "ps ax", it looked like it was  stuck
in the middle of trying to unmount an smbfs filesystem.

Next time I booted up, I umounted the smbfs system and did an "apm
--suspend" that appears to have shut everything down... And when I hit the
power button, it comes up again, albeit with an incredibly dim screen (a
common problem on this Dell Latitude C600).

So, from the point of view of my laptop, something important seems to have
changed between 2.4.2 and 2.4.7.

> I suspect that it's something like a single driver change (not apm,
> but PM-support in a driver).  How many I/O-device drivers do you
> use?  Would it be difficult to try to isolate which one may be
> faulty?

How would I find out what driver(s) might be vetoing my suspend request?

Here's the complete list of modules which might typically be loaded at boot:

3c59x
smbfs
lns_cp437
vfat
fat
ide-cd
cdrom
md
usb-uhci
usbcore
maestro3
sound
soundcore
ac97_codec
r128
agpgart
usb-uhci
usbcore

-0xe1a

Re: apm suspend broken in 2.4.10

[Randy.Dunlap]

Alex Cruise wrote:
> 
> 2.4.7 has the apm=on bug and won't suspend either... I'm beginning to
> suspect that something else is afoot, but I might as well try 2.4.2...
> 
> Funny, with 2.4.2, "apm=on" seems to work as expected, and the first time I
> tried "apm --suspend", the screen blanked but nothing else appeared to
> happen.  When I hit a key, the suspend command appeared to still be running.
> I switched to a different vc and did a "ps ax", it looked like it was  stuck
> in the middle of trying to unmount an smbfs filesystem.
> 
> Next time I booted up, I umounted the smbfs system and did an "apm
> --suspend" that appears to have shut everything down... And when I hit the
> power button, it comes up again, albeit with an incredibly dim screen (a
> common problem on this Dell Latitude C600).
> 
> So, from the point of view of my laptop, something important seems to have
> changed between 2.4.2 and 2.4.7.
> 
> > I suspect that it's something like a single driver change (not apm,
> > but PM-support in a driver).  How many I/O-device drivers do you
> > use?  Would it be difficult to try to isolate which one may be
> > faulty?
> 
> How would I find out what driver(s) might be vetoing my suspend request?
> 
> Here's the complete list of modules which might typically be loaded at boot:
> 
> 3c59x
> smbfs
> lns_cp437
> vfat
> fat
> ide-cd
> cdrom
> md
> usb-uhci
> usbcore
> maestro3
> sound
> soundcore
> ac97_codec
> r128
> agpgart
> usb-uhci
> usbcore


Unload some of these (that you don't really need to run)
and try "apm -s".

If that fails, unload some more of them and try again...

That would at least narrow down the search for us.

~Randy

RE: apm suspend broken in 2.4.10

[Alex Cruise]

From: Randy.Dunlap [mailto:rddunlap@osdlab.org]

> > -0xe1a
> {little-endian n[iy]bbles ?}

It's just the closest I can get to my own name in hex. ;)

> Sounds like our 2.4.10's are different then.  :)

It's possible... I got mine from kernel.org, applied the preemptible-kernel
and ext3fs patches, and  compiled with RH's "kgcc" 

> Without this patch, mine didn't create /proc/apm, register as a
> misc device, or create the kapmd-idle kernel thread.
> Must be a distro thingy.

Did you have apm=on set before, or nothing at all?  Here's what I've seen so
far:

In all cases, I've got apm compiled into the kernel, not a module.

- With 2.4.10, Before your patch, with no apm= option in the kernel command
line, APM in general works, but suspend doesn't.  When I append apm=on or
apm=off to my kernel command line, APM is disabled.
- With 2.4.10, After applying your patch, apm=on no longer disables APM, but
suspend still doesn't work.

> Return of EAGAIN from the SUSPEND ioctl means that
> send_event() failed, which means that some device driver
> didn't want suspend to happen...which means that some
> device driver got changed. :(

Just for fun, I tried removing all of my loaded 2.4.10 modules one by one,
and attempting 'apm --suspend' in between, and still had the same problem
when I got down to the bare minimum (ext3 and jbd)

> What was the last working kernel AFAUK (for this APM stuff)?

I just checked, and the RH-compiled 2.4.9-0.5 doesn't suspend either.  It
appears to suffer from the same "apm=on" command-line bug too.  I'm gonna go
try the 2.4.7 from RH's "Roswell" beta now.

-0xe1a

Re: apm suspend broken in 2.4.10

[Randy.Dunlap]

Alex Cruise wrote:
> 
> From: Randy.Dunlap [mailto:rddunlap@osdlab.org]
> 
> > Sounds like our 2.4.10's are different then.  :)
> 
> It's possible... I got mine from kernel.org, applied the preemptible-kernel
> and ext3fs patches, and  compiled with RH's "kgcc"
> 
> > Without this patch, mine didn't create /proc/apm, register as a
> > misc device, or create the kapmd-idle kernel thread.
> > Must be a distro thingy.
> 
> Did you have apm=on set before, or nothing at all?  Here's what I've seen so
> far:

I have "apm=on apm=debug".

> In all cases, I've got apm compiled into the kernel, not a module.

Same here.

> - With 2.4.10, Before your patch, with no apm= option in the kernel command
> line, APM in general works, but suspend doesn't.  When I append apm=on or
> apm=off to my kernel command line, APM is disabled.
> - With 2.4.10, After applying your patch, apm=on no longer disables APM, but
> suspend still doesn't work.
> 
> > Return of EAGAIN from the SUSPEND ioctl means that
> > send_event() failed, which means that some device driver
> > didn't want suspend to happen...which means that some
> > device driver got changed. :(
> 
> Just for fun, I tried removing all of my loaded 2.4.10 modules one by one,
> and attempting 'apm --suspend' in between, and still had the same problem
> when I got down to the bare minimum (ext3 and jbd)
> 
> > What was the last working kernel AFAUK (for this APM stuff)?
> 
> I just checked, and the RH-compiled 2.4.9-0.5 doesn't suspend either.  It
> appears to suffer from the same "apm=on" command-line bug too.  I'm gonna go
> try the 2.4.7 from RH's "Roswell" beta now.

OK, thanks for testing that.

I suspect that it's something like a single driver change (not apm,
but PM-support in a driver).  How many I/O-device drivers do you
use?  Would it be difficult to try to isolate which one may be
faulty?

~Randy

Re: apm suspend broken in 2.4.10

[Alan Cox]

> It's possible... I got mine from kernel.org, applied the preemptible-kernel
> and ext3fs patches, and  compiled with RH's "kgcc" 

Don't build later 2.4.x with egcs-1.1.2. At least some of the network
drivers stop working mysteriously with < gcc 2.95.3, RH 2.96-74+ or 
gcc 3.0

Its not the cause of the APM problem but be aware of it

Re: apm suspend broken in 2.4.10

[jc]

On vendredi 28 septembre, 2001 à 01:52:40 +0100, Alan Cox wrote:
> Don't build later 2.4.x with egcs-1.1.2. At least some of the network
> drivers stop working mysteriously with < gcc 2.95.3, RH 2.96-74+ or 
> gcc 3.0
> 
> Its not the cause of the APM problem but be aware of it


how could i get the list of files that changed between 2.4.9 and 2.4.10
concerning apm ?

Re: apm suspend broken in 2.4.10

[Keith Owens]

On Fri, 28 Sep 2001 11:09:23 +0200, 
jc <jcb@jcb.yi.org> wrote:
>how could i get the list of files that changed between 2.4.9 and 2.4.10
>concerning apm ?

diff -urN -I '$[ABD-Z].*\$' 2.4.9 2.4.10 > /var/tmp/patch-2.4.9-2.4.10
rm -rf /var/tmp/patches
split_patch /var/tmp/patch-2.4.9-2.4.10

Look through /var/tmp/patches for files and directories related to apm.
split_patch is

#!/usr/bin/perl -w
$out = "";
while (<>) {
	next if (/^Only/);
	next if (/^Binary/);
	if (/^diff/ || /^Index/) {
		if ($out) {
			close OUT;
		}
		(@out) = split(' ', $_);
		shift(@out) if (/^diff/);
		$out = pop(@out);
		$out =~ s:/*usr/:/:;
		$out =~ s:/*src/:/:;
		$out =~ s:^/*linux[^/]*::;
		$out =~ s:\(w\)::;
		next if ($out eq "");
		$out = "/var/tmp/patches/$out";
		$dir = $out;
		$dir =~ s:/[^/]*$::;
		print STDERR "$out\n";
		system("mkdir -p $dir");
		open(OUT, ">$out") || die("cannot open $out");
	}
	if ($out) {
		print OUT $_;
	}
}

RE: apm suspend broken in 2.4.10

[Alex Cruise]

I should, however, mention that "apm --standby" seems to work as expected.
It's only the "suspend" option that says "Resource temporarily unavailable."

-0xe1a

RE: apm suspend broken in 2.4.10

[Alex Cruise]

From: Randy.Dunlap [mailto:rddunlap@osdlab.org]

> Verified here.
> APM doesn't install if apm=on or apm=off is used in 2.4.10.
> 
> Here's a small patch for it.  With this patch, apm thread,
> /proc/apm, misc apm_bios device etc. are created.

Thanks... apm=on works now, but APM functionality itself still suffers from
the same failure as before (Resource temporarily unavailble.)

I should mention that before your patch, /dev/misc/apm_bios, /dev/apm_bios
and /proc/apm were already being created by the driver; it's going through
the motions but not delivering the goods.

-0xe1a

Re: apm suspend broken in 2.4.10

[Randy.Dunlap]

Alex Cruise wrote:
> 
> From: Randy.Dunlap [mailto:rddunlap@osdlab.org]
> 
> > Verified here.
> > APM doesn't install if apm=on or apm=off is used in 2.4.10.
> >
> > Here's a small patch for it.  With this patch, apm thread,
> > /proc/apm, misc apm_bios device etc. are created.
> 
> Thanks... apm=on works now, but APM functionality itself still suffers from
> the same failure as before (Resource temporarily unavailble.)
> 
> I should mention that before your patch, /dev/misc/apm_bios, /dev/apm_bios
> and /proc/apm were already being created by the driver; it's going through
> the motions but not delivering the goods.
> 
> -0xe1a

{little-endian n[iy]bbles ?}


Sounds like our 2.4.10's are different then.  :)
Without this patch, mine didn't create /proc/apm, register as a
misc device, or create the kapmd-idle kernel thread.
Must be a distro thingy.

Return of EAGAIN from the SUSPEND ioctl means that
send_event() failed, which means that some device driver
didn't want suspend to happen...which means that some
device driver got changed. :(

What was the last working kernel AFAUK (for this APM stuff)?

~Randy

Re: apm suspend broken in 2.4.10

[jc]

> 
> What was the last working kernel AFAUK (for this APM stuff)?
> 
2.4.9

Re: apm suspend broken in 2.4.10

[Alex Cruise]

Mine displays a similar failure, except my strace shows:

  ioctl(3, APM_IOC_SUSPEND, 0 ) = -1 EAGAIN (Resource temporarily
unavailable)

I also noticed (as reported by a previous poster) that whether you pass
"apm=on" or "apm=off" to the kernel, apm gets disabled.  When you don't
specify a setting, it's enabled.  I had a look at the arch/i386/kernel/apm.c
in 2.4.10 though, and it seemed to make sense.

--
Alex Cruise - Programmer, Iconoclast, Opinionated Bastard.
GCS/P v3.1a: d- s+:+ C++++$ UF++ UL++ P--- L++ E++ W++ N+ K w-- O- M 
PS++ PE- Y+ PGP-- t+@ 5- X-- R !tv b++ DI++ D++ G e* h--- r+++ y+++ 

Re: apm suspend broken in 2.4.10

[Randy.Dunlap]

[-- Attachment #1: Type: text/plain, Size: 639 bytes --]

Alex Cruise wrote:
> 
> Mine displays a similar failure, except my strace shows:
> 
>   ioctl(3, APM_IOC_SUSPEND, 0 ) = -1 EAGAIN (Resource temporarily
> unavailable)
> 
> I also noticed (as reported by a previous poster) that whether you pass
> "apm=on" or "apm=off" to the kernel, apm gets disabled.  When you don't
> specify a setting, it's enabled.  I had a look at the arch/i386/kernel/apm.c
> in 2.4.10 though, and it seemed to make sense.

Verified here.
APM doesn't install if apm=on or apm=off is used in 2.4.10.

Here's a small patch for it.  With this patch, apm thread,
/proc/apm, misc apm_bios device etc. are created.

~Randy

[-- Attachment #2: apmonoff.patch --]
[-- Type: text/plain, Size: 662 bytes --]

--- linux/arch/i386/kernel/apm.c.org	Mon Sep 17 22:52:35 2001
+++ linux/arch/i386/kernel/apm.c	Thu Sep 27 13:15:33 2001
@@ -1672,7 +1672,7 @@
 		apm_info.realmode_power_off = 1;
 	/* User can override, but default is to trust DMI */
 	if (apm_disabled != -1)
-		apm_info.disabled = 1;
+		apm_info.disabled = apm_disabled;
 
 	/*
 	 * Fix for the Compaq Contura 3/25c which reports BIOS version 0.1
@@ -1699,8 +1699,7 @@
 	}
 
 	if (apm_info.disabled) {
-		if(apm_disabled == 1)
-			printk(KERN_NOTICE "apm: disabled on user request.\n");
+		printk(KERN_NOTICE "apm: disabled on user request.\n");
 		return -ENODEV;
 	}
 	if ((smp_num_cpus > 1) && !power_off) {

apm suspend broken in 2.4.10

[jc]

it worked fine with 2.4.9

now with 2.4.10 :

strace apm -s 

...
open("/dev/apm_bios", O_RDWR)           = 3
time([1001455050])                      = 1001455050
sync()                                  = 0
ioctl(3, AGPIOC_RELEASE, 0)             = -1 EAGAIN (Resource temporarily unavailable)
...

Re: apm suspend broken in 2.4.10

[Matthias Andree]

On Wed, 26 Sep 2001, jc wrote:

> it worked fine with 2.4.9
> 
> now with 2.4.10 :
> 
> strace apm -s 

Works for me. Post more information about your system. Did you compile
APM? Did the machine show APM related boot log lines? Or did it use ACPI
perhaps?

BTW, please have your Message-ID contain a full domain name and not just
"athena" to avoid collisions.