Hi,
I have read all recent mails about this misunderstooded change in 2.4
series. But as no one reports heating, I do :
for the AMD K6-2 on a DFI motherboard AT/ATX, using the AT power supply,
this option is buggy. I mean the cycles kapmidled works doesn't cool the
processor, they hot him.
This stops as soon as I uncheck the option "Send Halt command on idle" in
the APM settings of the kernel.
Jos?
P.S.:please CC to my address, as I am not subscriber of this list
On Tue, 9 Oct 2001 [email protected] wrote:
> for the AMD K6-2 on a DFI motherboard AT/ATX, using the AT power supply,
> this option is buggy. I mean the cycles kapmidled works doesn't cool the
> processor, they hot him.
Initially, I thought was odd. The spec seemed straight forward
enough, and doesn't say we have to do any special magic.
Just that "During the execution of the HLT instruction, the AMD-K6-2
processor executes a Halt special cycle."
The next bit is interesting however..
"After BRDY# is sampled asserted during this cycle, and then EWBE#
is also sampled asserted (if not masked off), the processor enters
the halt state in which the processor disables most of its internal
clock distribution."
EWBE is a feature that is enabled with bits 2-3 of the EFER MSR.
This controls the behaviour of the CPU with respect to ordering
of write cycles. Behaviour here can affect performance, and from
my interpretation of the above, the amount of power saving that
is possible.
You can control the EWBE register using powertweak
(http://www.powertweak.org), but if you don't want to/are unable
to build that, and want to do some further tests, let me know
and I'll hack something up.
If this feature is affecting temperature dramatically, it may
be worth us clearing this on boot up.
I've heard reports from Athlon users who also say HLT doesn't
do anything regarding temperature for their systems. I wonder
if it also has a similar feature tucked away in an MSR somewhere..
regards,
Dave.
--
| Dave Jones. http://www.suse.de/~davej
| SuSE Labs
On Tue, Oct 09, 2001 at 04:15:20PM +0100, [email protected] wrote:
> Hi,
>
> I have read all recent mails about this misunderstooded change in 2.4
> series. But as no one reports heating, I do :
>
> for the AMD K6-2 on a DFI motherboard AT/ATX, using the AT power supply,
> this option is buggy. I mean the cycles kapmidled works doesn't cool the
> processor, they hot him.
>
> This stops as soon as I uncheck the option "Send Halt command on idle" in
> the APM settings of the kernel.
I have a K6-2 too. When I'm using 100% of the CPU, the temp of
the CPU is around 42 - 44 ?C. When I don't use any CPU at all it
will lower with about 10 ?C.
>From my .config:
CONFIG_PM=y
# CONFIG_ACPI is not set
CONFIG_APM=y
# CONFIG_APM_IGNORE_USER_SUSPEND is not set
# CONFIG_APM_DO_ENABLE is not set
# CONFIG_APM_CPU_IDLE is not set
# CONFIG_APM_DISPLAY_BLANK is not set
# CONFIG_APM_RTC_IS_GMT is not set
# CONFIG_APM_ALLOW_INTS is not set
# CONFIG_APM_REAL_MODE_POWER_OFF is not set
The only reason I enable APM is because then I can do a
powerdown.
I'm willing to test if the temp goes up when I enable
CONFIG_APM_CPU_IDLE, and it's idle otherwise.
CONFIG_APM_CPU_IDLE is shown as "Make CPU Idle calls when idle",
I don't see a "Send Halt command on idle".
Kurt
> I've heard reports from Athlon users who also say HLT doesn't
> do anything regarding temperature for their systems. I wonder
> if it also has a similar feature tucked away in an MSR somewhere..
I think I can confirm this:
processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 4
model name : AMD Athlon(tm) Processor
stepping : 2
cpu MHz : 1009.015
cache size : 256 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr syscall mmxext 3dnowext 3dnow
bogomips : 2011.95
tranquil:~# uptime
09:57:59 up 4:42, 2 users, load average: 0.00, 0.00, 0.00
tranquil:~# vmstat 1
procs memory swap io system cpu
r b w swpd free buff cache si so bi bo in cs us sy id
0 0 0 0 601324 55064 57588 0 0 5 4 102 52 0 0 100
0 0 0 0 601324 55064 57588 0 0 0 0 106 66 0 0 100
0 0 0 0 601324 55064 57588 0 0 0 0 104 57 0 0 100
0 0 0 0 601324 55064 57588 0 0 0 0 106 63 0 0 100
0 0 0 0 601324 55064 57588 0 0 0 0 104 57 0 0 100
0 0 0 0 601324 55064 57588 0 0 0 0 104 65 0 0 100
tranquil:~# sensors | tail | egrep 'CPU|Mobo'
CPU fan: 6308 RPM (min = 3000 RPM, div = 2)
Mobo Temp: +30.0?C (limit = +60?C, hysteresis = +50?C)
CPU Temp: +49?C (limit = +67?C, hysteresis = +60?C) (beep)
With kernel compile it goes to 52 degrees, which makes 3 degrees
difference between almost 100% idle and almost 0.0% idle.
This is with kernel 2.4.10-ac10 if it matters.
Ookhoi
Hi!
> > for the AMD K6-2 on a DFI motherboard AT/ATX, using the AT power supply,
> > this option is buggy. I mean the cycles kapmidled works doesn't cool the
> > processor, they hot him.
>
> Initially, I thought was odd. The spec seemed straight forward
> enough, and doesn't say we have to do any special magic.
> Just that "During the execution of the HLT instruction, the AMD-K6-2
> processor executes a Halt special cycle."
>
> The next bit is interesting however..
>
> "After BRDY# is sampled asserted during this cycle, and then EWBE#
> is also sampled asserted (if not masked off), the processor enters
> the halt state in which the processor disables most of its internal
> clock distribution."
>
> EWBE is a feature that is enabled with bits 2-3 of the EFER MSR.
> This controls the behaviour of the CPU with respect to ordering
> of write cycles. Behaviour here can affect performance, and from
> my interpretation of the above, the amount of power saving that
> is possible.
>
> You can control the EWBE register using powertweak
> (http://www.powertweak.org), but if you don't want to/are unable
> to build that, and want to do some further tests, let me know
> and I'll hack something up.
If I don't want to build powertweak, are you willing to hack something up
for me? ;-). [My k6-2 is too hot to slow down CPU fan. I tried throttling
it using ACPI, but no success. I want to cool it down so that fan slows
and machine becomes quiet.]
Pavel
--
Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt,
details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html.
On Fri, 12 Oct 2001, Pavel Machek wrote:
> If I don't want to build powertweak, are you willing to hack something up
> for me? ;-). [My k6-2 is too hot to slow down CPU fan. I tried throttling
> it using ACPI, but no success. I want to cool it down so that fan slows
> and machine becomes quiet.]
Certainly :-)
http://www.codemonkey.org.uk/cruft/ewbe.c is a quick and dirty hack.
You'll need the msr driver loaded, and the /dev node there.
I've not tested this code as my K6-2 box is currently in bits, but
it should be ok. Bits 2 & 3 in 0xc0000080 do the magic, so theres not
much that could be wrong.. If all else fails, double check the spec.
Let me know if this does anything for you at all
(regarding both temperature, and performance)
Oh, and for everyone else, as the comment in the source says,
don't run this on anything that isn't a K6-2 / K6-3.
I was too lazy to add cpuid checking to this. Maybe I'll do that
later..
regards,
Dave.
--
| Dave Jones. http://www.codemonkey.org.uk
| SuSE Labs
When I say too hot, I mean as hot as when full-loaded when it is idle.
More precisely, less than 30?C when I am watching TV, so that I don't hear
the CPU fan!
Jos?
/----------------------------------------------------------------------------\
| Jos? Jorge <[email protected]> |
| TEKLYNX International http://www.teklynx.com |
\----------------------------------------------------------------------------/
Pavel Machek
<[email protected] To: Luigi Genoni <[email protected]>, Pavel Machek <[email protected]>
z> cc: Dave Jones <[email protected]>, [email protected], [email protected]
Subject: Re: kapmidled and AMD K6-2
19/10/2001
21:05
Hi!
> Excuse me, I ma quite curious. I had 50 K6-2, (from 350 to 500 Mhz), in
> the SNS CED, so I would like to know when you say it is too hot, what
> temperature are you talking about? In my experience I had some k6
> 500 working
> for 360 days continously fully loaded with a temperature of 55 degrees on
> the processor, and it was ever stable. On the other side, I saw a 450 Mhz
> processor to be unstable being ideld at 50 degrees (the fan was
> not properly working). I have a quite good idea when a K6 could be
> damnaged by temperature. Ohh, that is really important, did you make some
> burn in to your processors?
On my k6-2, cpu fan starts cooling on 30 celsia. But I do not know
where the sensor is ;-)
> Luigi
>
>
> On Fri, 12 Oct 2001, Pavel Machek wrote:
>
> > Hi!
> >
> > > > for the AMD K6-2 on a DFI motherboard AT/ATX, using the AT power
supply,
> > > > this option is buggy. I mean the cycles kapmidled works doesn't
cool the
> > > > processor, they hot him.
> > >
> > > Initially, I thought was odd. The spec seemed straight forward
> > > enough, and doesn't say we have to do any special magic.
> > > Just that "During the execution of the HLT instruction, the AMD-K6-2
> > > processor executes a Halt special cycle."
> > >
> > > The next bit is interesting however..
> > >
> > > "After BRDY# is sampled asserted during this cycle, and then EWBE#
> > > is also sampled asserted (if not masked off), the processor enters
> > > the halt state in which the processor disables most of its internal
> > > clock distribution."
> > >
> > > EWBE is a feature that is enabled with bits 2-3 of the EFER MSR.
> > > This controls the behaviour of the CPU with respect to ordering
> > > of write cycles. Behaviour here can affect performance, and from
> > > my interpretation of the above, the amount of power saving that
> > > is possible.
> > >
> > > You can control the EWBE register using powertweak
> > > (http://www.powertweak.org), but if you don't want to/are unable
> > > to build that, and want to do some further tests, let me know
> > > and I'll hack something up.
> >
> > If I don't want to build powertweak, are you willing to hack something
up
> > for me? ;-). [My k6-2 is too hot to slow down CPU fan. I tried
throttling
> > it using ACPI, but no success. I want to cool it down so that fan slows
> > and machine becomes quiet.]
--
STOP THE WAR! Someone killed innocent Americans. That does not give
U.S. right to kill people in Afganistan.