Very serious cooling issues CURRENT/STABLE

[Ian Smith]

On Sun, 21 Dec 2008, Martin wrote:
 > Am Sun, 21 Dec 2008 00:54:35 -0500
 > schrieb Nathan Lay <nslay at comcast.net>:
 > 
 > > Hi list(s)
 > > Early in the year I noticed CURRENT failed to cool my frankenstein 
 > > Thinkpad T40 (built from T40/40p parts) under load.  The system would 
 > > shut down from critically high temperatures while building a port.  I 
 > > did nothing special with ACPI.  I thought nothing of it since, after 
 > > all, I rebuilt a broken T40 from T40p parts.  Today, however, I
 > > noticed that a very recent STABLE build failed to keep my T43 cool
 > > while building ports.  It reached temperatures of 97, 98, 99C.

Nathan, have you cleaned out airpaths, heatsinks, thermal paste etc 
recently?  Just checking; people have reported drastic improvements.

 > Hallo Nathan,
 > 
 > I can confirm this on a T60p. My notebooks shuts down since I use 7.0
 > at 101°C. I have written a script that throttles the CPU if it gets
 > above 75°C. I have to use it when I'm building world or ports.
 > 
 > Btw, I have reported the problem here a some time ago [1]. I have
 > received some feedback that IBM/Lenovo uses wrong cooling strategy
 > placing the CPU and the VGA under one single cooled surface and using
 > too much thermal grease.

We keep hearing that, but that shouldn't affect the T40 or T43 .. so 
addressing the maximum fan speed issue below is more a 'last resort'; 
overriding and lowering the _PSV value for .tz0 might help via p4tcc?

 > On the other hand I have also emphasized that the fan should be able to
 > have about 800rpm more at the temperature of 80°C, but this has been
 > ignored so far. This is why I have never had any problems running "the
 > other common operating systems" and I have put them under heavy load.
 > They don't exceed 85°C.

http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/dev/acpi_support/acpi_ibm.c?rev=1.17

Comments in acpi_ibm_sysctl_get, case ACPI_IBM_METHOD_FANLEVEL: say that 
setting bit 6 (IBM_EC_MASK_FANDISENGAGED) of IBM_EC_FANSTATUS overrides 
the fan speed limit, which linux allows, 'doze too apparently.

acpi_ibm_sysctl_set currently rejects fan levels higher than 7, ie does 
not allow running the fan disengaged, but that's something you could fix 
at your own risk .. it's been suggested that this could wear the fan out 
more quickly, so I guess acpi_ibm is tending to be conservative here.

It doesn't look hard to at least allow values with bit 6 set (ie 64-127) 
in acpi_ibm_sysctl_set, case ACPI_IBM_METHOD_FANLEVEL: and a bit more 
work to preserve that bit in the _get values.  It looks like you might
need to set the fan in manual mode too (dev.acpi_ibm.0.fan=0) but test!

 > > It
 > > used to reach a max temperature of 80C while building a port (idling
 > > temperature is ~ 45C).  It behaved normally from 5.3 all the way
 > > through 7.0.  To remedy both issues, I had to underclock the
 > > processor (via dev.cpu) and force the fan to its highest (I'm running
 > > 533MHz shy of the processor's full potential, with a fan level of 7
 > > to keep the temperature at ~75C).
 > 
 > I have one further hint for you. The VGA is running very hot on
 > FreeBSD (~75°C when idle), because it is not able to throttle the GPU
 > and its voltage. Maybe you've had a working setup for your GPU before
 > and you even did not notice? Did you change your Xorg drivers?
 > 
 > > I've always suspected FreeBSD's
 > > ACPI didn't work properly on Thinkpad T series laptops (I have T40,
 > > T41, T42 and T43) as it could never fully use the fan (Windows XP can
 > > rev the fan far higher than acpi_ibm's level 7).  Between STABLE and
 > > CURRENT...something seems very wrong.  These systems never had
 > > cooling issues with FreeBSD before.
 > 
 > The fan is running definitely too slow. On my system, too.

I guess if I was having overheating issues with my T23 and was confident 
with C and its debugging, I'd have a go at this .. but I am neither ..

 > Thanks,
 > Martin
 > 
 > 
 > References:
 > [1]
 > http://lists.freebsd.org/pipermail/freebsd-stable/2008-June/042812.html

cheers, Ian