Headers problems

Wit WitlessIdiot at triad.rr.com
Sat Jun 9 07:26:01 PDT 2007


Tijnema wrote:
> On 6/8/07, Ken Moffat <ken at linuxfromscratch.org> wrote:
>   
>> On Fri, Jun 08, 2007 at 10:46:11AM +0200, Tijnema wrote:
>>     
>>>>  <snip>

>>> ... I'm only at a good old single core AMD
>>> Athlon XP system, which is clocked down to 1.15Ghz with 512MB SD
>>>       
Having followed the complete thread, I thought I would throw this in.

First, feeling the CPU by hand (how when the HSF is on top?) is not a
really reliable thing. At the time you are feeling it, it could have
cooled *substantiall* from the point at which heat was driven too high
by heavy load.

I suggest doing your kernel with CONFIG_ACPI=y and CONFIG_ACPI_THERMAL=y
if you mobo support acpi. Then you can cat
/proc/acpi/thermal_zone/THRM/temperature
in a loop to get a more accurate reading.

Have you ever upgraded your BIOS? Often a BIOS upgrade fixes some boot
issues related to modern OSs.

Have tried booting with acpi-off or noapic? This might let you boot at
full speed. However, if you have cooling issues at the reduced clocking,
I would expect them to be exacerbated at the higher clocking.

PS may be undersized. I had a SIS-based mobo that ran fine with a 300
wat PS, but the EPOX I tried (VIA chipset) would not work with that.
User's manual warned that 300 was minimum, 350 recommended. Since it was
reasonably loaded, I threw in a 575 watt and problem disappeared.

This *should* be most noticable at boot time. That's when the drives are
spinning up, etc. Since it booted at reduced clocking (indicating the
power requirements are reduced), this seems a possibility. One way might
be to boot at full clocking, it freezes (or whatever) and then hit the
reset button. Or hit reset before that. Since HDs will be up to full
speed, maybe load is reduced enough to run full speed for a short while.

Another test: remove everything you can and see if boot succeeds at full
clocking. Then add items back in, one at time. The power requirements
listed on various devices may let you calculate the maximum load with
doing all this remove/re-insert.

Have you checked that all memory, PCI cards, power connectors are
well-seated?

Have you tried hooking "high-load" items (like disks) on different
"legs" of the PS? Have you looked at the PS label to see what the limits
are for 12/5 and that-other-thing-I-can never-remember-what-its-called?
That other thing is often very undersized relative to the total PS capacity.

Does your BIOS have settings to delay spin-up of the HDs to reduce peak
load?

>>> <snip>

> I hate old versions :P I want to keep my system up to date, so when I
> want to install a small app, i don't need to reinstall my complete
> system because glibc is outdated :P
>   

Keep in mind that glibc provides the ABI between kernel and outside. You
only don't have a new feature if the kernel/userspace change requires a
new ABI definition to support the feature. This is a relatively seldom
occurrence. As pointed out elsewhere, the glibc must remain the one with
which the system is built to avoid breaking everything that depends on
the ABI. Of course, a point release may be save, but one would need to
watch the glibc lists and comprehend the nature of the changes to make
an informed decision as to the safety of the upgrade.

That's more effort/knowledge required than most of us would find
worthwhile (IMO) for the very small gains.

> <snip the stuff about resets: addressed above>
>   

>> ... which regularly reports temperatures around 48°C even on
>> a cool day).

# cat /proc/acpi/thermal_zone/THRM/temperature
temperature: 36 C

But this is on an AMD PR2200+ with a nice little Zallman FHS setup. On
my 3200+ with just a standard HSF rated for the CPU, I get 40 C. Both of
these are very lightly loaded units in a room temperature of about 75 F.
The slow one has the covers on, the other has covers off right now.

If it is heat, try running with covers on for awhile.

Reinforcing what was said by another, I stress that the heat sink *must*
have sufficient colling area and the fan must move enough air.

Keep in mind about your CPU speed rating: these are determined not by
how fast the CPU can go, but haw fast it can go and still dissipate heat
sufficiently fast enough to not overheat. As the units are tested, the
expected capacity of the HSF combination is taken into consideration and
the CPU is then rated for the speed at which it can run and remain cool
enough to be reliable.

Also, use a quality thermal paste. Quicksilver or equivalent. Must have
even application with just enough to fill the micropeaks and valleys.

>> <snip>

HTH
--
Wit



More information about the lfs-support mailing list