TEST: Sleep patch #6

TEST: Sleep patch #6

[Benjamin Herrenschmidt]

(As usual, I'm cross posting several lists, please don't reply to all of them,
and CC me as I'm not subscribed to all of them neither)

Ok, here's the 6th version of the sleep patch for ATI based albooks &
iBook G4. Other machine users, please test too as it may cause
regressions (or improvements) as well.

This one fixes finally the cache flush problem that made cpufreq (on machines
using the PMU for the switch) and sleep/wake unreliable occasionally (memory
corruption would occur on some CPUs at least). It improves bits here or there
too, and adds the sungem wake-on-lan feature. 

There are still pending issues, like cpufreq on some machines will
"think" it's running at full speed on wakeup while it's in fact running
at slow speed (thankfully not the opposite). I don't think I'll fix this one
in 2.6.9, but rather in 2.6.10.

http://gate.crashing.org/~benh/albook-ibookg4-sleep-6.diff

I'll post a 2.6.10-rc2-bk* based version of the patch soon.

Ben.

Re: TEST: Sleep patch #6

[Colin Leroy]

On 01 Dec 2004 at 16h12, Benjamin Herrenschmidt wrote:

Hi Ben, 

> This one fixes finally the cache flush problem that made cpufreq (on machines
> using the PMU for the switch) and sleep/wake unreliable occasionally (memory
> corruption would occur on some CPUs at least). It improves bits here or there
> too, and adds the sungem wake-on-lan feature. 

Compiling that. btw, you didn't add the "&& !wol" before sungem's suspend call...

-- 
Colin

Re: TEST: Sleep patch #6

[Benjamin Herrenschmidt]

On Wed, 2004-12-01 at 08:36 +0100, Colin Leroy wrote:
> On 01 Dec 2004 at 16h12, Benjamin Herrenschmidt wrote:
> 
> Hi Ben, 
> 
> > This one fixes finally the cache flush problem that made cpufreq (on machines
> > using the PMU for the switch) and sleep/wake unreliable occasionally (memory
> > corruption would occur on some CPUs at least). It improves bits here or there
> > too, and adds the sungem wake-on-lan feature. 
> 
> Compiling that. btw, you didn't add the "&& !wol" before sungem's suspend call...


What do you mean ? the whole code block is now in !wol...

Note that I'm still having problems with the cache ... I'm asking
Moto/Freescale...

Ben.

Re: TEST: Sleep patch #6

[Colin Leroy]

On 01 Dec 2004 at 20h12, Benjamin Herrenschmidt wrote:

Hi, 

> > Compiling that. btw, you didn't add the "&& !wol" before sungem's suspend call...
> 
> What do you mean ? the whole code block is now in !wol...

I'm just being stupid, i didn't see that :)
So much for trying to read patches before having had my coffee...

> Note that I'm still having problems with the cache ... I'm asking
> Moto/Freescale...

With which CPUs? I did notice zero problems with patch #5 (without freq scaling).

-- 
Colin

Re: TEST: Sleep patch #6

[Benjamin Herrenschmidt]

On Wed, 2004-12-01 at 10:11 +0100, Colin Leroy wrote:
> On 01 Dec 2004 at 20h12, Benjamin Herrenschmidt wrote:
> 
> Hi, 
> 
> > > Compiling that. btw, you didn't add the "&& !wol" before sungem's suspend call...
> > 
> > What do you mean ? the whole code block is now in !wol...
> 
> I'm just being stupid, i didn't see that :)
> So much for trying to read patches before having had my coffee...
> 
> > Note that I'm still having problems with the cache ... I'm asking
> > Moto/Freescale...
> 
> With which CPUs? I did notice zero problems with patch #5 (without freq scaling).

On my 7455 rev 3.3 at least. cpufreq on mine uses the PMU as well, thus
approx. the same code path as sleep. By doing freq. switches in a loop
from a shell script, I can make it crash in a few iterations. It seem to
be solid if I add an ugly displacement flush before the cache flush
code, but that should not be necessary and thus makes me think something
is wrong with the code.

Ben.

Re: TEST: Sleep patch #6

[Colin Leroy]

On 01 Dec 2004 at 16h12, Benjamin Herrenschmidt wrote:

Hi, 

> Ok, here's the 6th version of the sleep patch for ATI based albooks &
> iBook G4. Other machine users, please test too as it may cause
> regressions (or improvements) as well.

Works ok here on iBook G4.

-- 
Colin

Re: TEST: Sleep patch #6

[David Woodhouse]

On Wed, 2004-12-01 at 16:03 +1100, Benjamin Herrenschmidt wrote:
> (As usual, I'm cross posting several lists, please don't reply to all of them,
> and CC me as I'm not subscribed to all of them neither)
> 
> Ok, here's the 6th version of the sleep patch for ATI based albooks &
> iBook G4. Other machine users, please test too as it may cause
> regressions (or improvements) as well.

kernel-2.6.9-1.691_FC3.dwmw2 in same place as before.

-- 
dwmw2

Re: TEST: Sleep patch #6

[Timo Reimerdes]

Ok,

I mentioned those weird problems about the computer not falling asleep?
It keeps happening. I compiled a kernel with the config from Alexander
Wirt and tried alot of combinations. 

The logs really wont help a lot.
USB-Mouse plugged in/out wont change anything, AC plugged in/out wont
change anything, restarting X wont fix it. 

Is there a possibility that pbbuttonsd is somehow related to that
problem? I wonder. Might check that out later. 

This problem I must fix before I can call sleep working. It's really
annoying. 
Do you have any hint on "what needs to be running" or "needs to be set
to X" or anything I could look for?

There are two things I do get during boot:
_________________________________________________
Dec  2 01:41:38 localhost kernel: pmac_zilog: 0.6 (Benjamin
Herrenschmidt <benh@kernel.crashing.org>)
Dec  2 01:41:38 localhost kernel: pmac_zilog: Error registering serial
device, disabling pmac_zilog.
Dec  2 01:41:38 localhost kernel: pmac_zilog: Did another serial driver
already claim the minors?
Dec  2 01:41:38 localhost kernel: RAMDISK driver initialized: 16 RAM
disks of 8192K size 1024 blocksize
Dec  2 01:41:38 localhost kernel: divert: not allocating divert_blk for
non-ethernet device lo
Dec  2 01:41:38 localhost kernel: MacIO PCI driver attached to Intrepid
chipset
Dec  2 01:41:38 localhost kernel: input: Macintosh mouse button
emulation
.
.
.
_________________________________________________
Dec  2 01:41:38 localhost kernel: device-mapper: 4.1.0-ioctl
(2003-12-10) initialised: dm@uk.sistina.com
Dec  2 01:41:38 localhost kernel: md: md driver 0.90.0 MAX_MD_DEVS=256,
MD_SB_DISKS=27
Dec  2 01:41:38 localhost kernel: hdc: packet command error: status=0x51
{ DriveReady SeekComplete Error }
Dec  2 01:41:38 localhost kernel: hdc: packet command error: error=0x54
Dec  2 01:41:38 localhost kernel: ide: failed opcode was 100
Dec  2 01:41:38 localhost kernel: cdrom: open failed.
Dec  2 01:41:38 localhost kernel: device-mapper: : dm-linear: Device
lookup failed
Dec  2 01:41:38 localhost kernel:
Dec  2 01:41:38 localhost kernel: device-mapper: error adding target to
table
Dec  2 01:41:38 localhost kernel: device-mapper: : dm-linear: Device
lookup failed
Dec  2 01:41:38 localhost kernel:
Dec  2 01:41:38 localhost kernel: device-mapper: error adding target to
table
.
. (20 times ".... failed"
.
Dec  2 01:41:38 localhost kernel: device-mapper: error adding target to
table
.
.
.
________________________________________________________

other than that my boot looks really clean and without any more errors.

If you could give me some hints I would really appreciate that.

Greetings

Timo Reimerdes

Re: TEST: Sleep patch #6

[Colin Leroy]

On 01 Dec 2004 at 16h12, Benjamin Herrenschmidt wrote:

Hi, 
 
> This one fixes finally the cache flush problem that made cpufreq (on machines
> using the PMU for the switch) and sleep/wake unreliable occasionally (memory
> corruption would occur on some CPUs at least). It improves bits here or there
> too, and adds the sungem wake-on-lan feature. 

By the way, I noticed a problem when the laptop is put to sleep by closing lid:
wake-on-lan still works, but as soon as the laptop's awake, pbbuttonsd sleeps 
it again. Combined with the fact that a closed iBook G4 not sleeping shutdowns 
by itself, it would be nice if sungem could disable wol when suspending and the 
lid is closed - if such info is available.

Or maybe such a thing should be done in the pbbuttonsd scripts by calling 
ethtool?
-- 
Colin

Re: TEST: Sleep patch #6

[Benjamin Herrenschmidt]

On Thu, 2004-12-02 at 12:08 +0100, Colin Leroy wrote:
> On 01 Dec 2004 at 16h12, Benjamin Herrenschmidt wrote:
> 
> Hi, 
>  
> > This one fixes finally the cache flush problem that made cpufreq (on machines
> > using the PMU for the switch) and sleep/wake unreliable occasionally (memory
> > corruption would occur on some CPUs at least). It improves bits here or there
> > too, and adds the sungem wake-on-lan feature. 
> 
> By the way, I noticed a problem when the laptop is put to sleep by closing lid:
> wake-on-lan still works, but as soon as the laptop's awake, pbbuttonsd sleeps 
> it again. 

Useland shit :)

> Combined with the fact that a closed iBook G4 not sleeping shutdowns 
> by itself, it would be nice if sungem could disable wol when suspending and the 
> lid is closed - if such info is available.

Userland shit (bis) :)

> Or maybe such a thing should be done in the pbbuttonsd scripts by calling 
> ethtool?
-- 
Benjamin Herrenschmidt <benh@kernel.crashing.org>

Re: TEST: Sleep patch #6

[Kjetil Ørbekk]

Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:

> http://gate.crashing.org/~benh/albook-ibookg4-sleep-6.diff

Worked like a charm.

-- 
Kjetil Orbekk

Re: TEST: Sleep patch #6

[Federico Gamio]

[-- Attachment #1: Type: text/plain, Size: 2042 bytes --]

With patch #6 in some random situations, my PowerBook G4 (ATI 9600)
doesn't goes to sleep.
One situation is:
 - Turn PB
 - Work
 - Sleep
 - 14 hours of sleep
 - Turn back from sleep (perfect, no problem at all)
 - Work for 4 hours (mails, www, no USB, Bluetooth, etc)
 - Try to sleep and problem...

When I try to make sleep my PB, nothing happens (in X), and I can't
switch from X to console.
Sorry, I can't help you more with this bug, I will keep trying to
establish which are the events that trigger this bug.

Federico

On Wed, 2004-12-01 at 16:03 +1100, Benjamin Herrenschmidt wrote:
> (As usual, I'm cross posting several lists, please don't reply to all of them,
> and CC me as I'm not subscribed to all of them neither)
> 
> Ok, here's the 6th version of the sleep patch for ATI based albooks &
> iBook G4. Other machine users, please test too as it may cause
> regressions (or improvements) as well.
> 
> This one fixes finally the cache flush problem that made cpufreq (on machines
> using the PMU for the switch) and sleep/wake unreliable occasionally (memory
> corruption would occur on some CPUs at least). It improves bits here or there
> too, and adds the sungem wake-on-lan feature. 
> 
> There are still pending issues, like cpufreq on some machines will
> "think" it's running at full speed on wakeup while it's in fact running
> at slow speed (thankfully not the opposite). I don't think I'll fix this one
> in 2.6.9, but rather in 2.6.10.
> 
> http://gate.crashing.org/~benh/albook-ibookg4-sleep-6.diff
> 
> I'll post a 2.6.10-rc2-bk* based version of the patch soon.
> 
> Ben.
> 
> 
> 
-- 
My software never has bugs.       | ASCII Ribbon Campaign         /"\
It just develops random features. | For Standards-Complaint Email \ /
Public GnuPG key available at: http://www.keyserver.net            X
1024D/203E154B 2003-10-03 Federico Gamio <federico@gamio.net>     / \
Key fingerprint = 0B9E 3A19 C88B EBAC 5422  C05F 76B5 B922 203E 154B
sub  2048g/32D2F465 2003-10-03

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

TEST: Sleep patch #6

[Benjamin Herrenschmidt]

(As usual, I'm cross posting several lists, please don't reply to all of them,
and CC me as I'm not subscribed to all of them neither)

Ok, here's the 7th version of the sleep patch for ATI based albooks &
iBook G4. Other machine users, please test too as it may cause
regressions (or improvements) as well.

The cache flush problem happens to not be completely fixed yet :( There is
something going on that I don't completely explain yet, might be related to
an errata of some CPU revisions. This version of the patch adds a load/flush
loop before the L2 HW flush on the 745x that appear to make the thing stable
on the machines I've tested on. It's a good enough workaround for now, I'm
working with freescale to find out what's really going on though.

http://gate.crashing.org/~benh/albook-ibookg4-sleep-7.diff


Ben.