Re: system slow since ~ 2.6.7

Re: system slow since ~ 2.6.7

[Len Brown]

On Wed, 2004-11-17 at 19:25, Guennadi Liakhovetski wrote:
> "Slow" means just running top alone in a vt it takes 1.6% CPU. Under
> 2.6.3 it takes 0.2% (Duron 900MHz). Another peculiarity with 2.6.7 and
> 2.6.9 is that the power LED is blinking with about 1Hz frequency. It's
> an ASUS A7VI-VM motherboard. In the manual there's nothing about
> error-codes. I played with various APIC settings - no change. Related
> or not - if running with LAPIC enabled (2.6.7), I get quite a few ERR
> in /proc/interrupts.

PCI: Disabling Via external APIC routing

Curiously, this line appears in 2.6.3, but not in 2.6.7 or 2.6.9 dmesg
-- even though all the configs build in IOAPIC support.

Can you forward the /proc/interrupts from 2.6.3, and from 2.6.9 with and
without acpi=off?  do you see a significant change in /proc/interrupts
before and after the sensor-provoked slowness starts?

if you build 2.6.9 w/o the CONFIG_ACPI_PROCESSOR and boot w/o cmdline
params, do you still see slowness?

if you boot 2.6.9 with these parameters, do you see any additional dmesg
lines?

acpi_dbg_level=0xF acpi_dbg_layer=0xFFFF3FFF


thanks,
-Len




-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: system slow since ~ 2.6.7

[Guennadi Liakhovetski]

Thanks for the reply

On Tue, 23 Nov 2004, Len Brown wrote:

> On Wed, 2004-11-17 at 19:25, Guennadi Liakhovetski wrote:
> > "Slow" means just running top alone in a vt it takes 1.6% CPU. Under
> > 2.6.3 it takes 0.2% (Duron 900MHz). Another peculiarity with 2.6.7 and
> > 2.6.9 is that the power LED is blinking with about 1Hz frequency. It's
> > an ASUS A7VI-VM motherboard. In the manual there's nothing about
> > error-codes. I played with various APIC settings - no change. Related
> > or not - if running with LAPIC enabled (2.6.7), I get quite a few ERR
> > in /proc/interrupts.
> 
> PCI: Disabling Via external APIC routing
> 
> Curiously, this line appears in 2.6.3, but not in 2.6.7 or 2.6.9 dmesg
> -- even though all the configs build in IOAPIC support.

Well, I think, there's just a local APIC on the system, and that is 
disabled in BIOS (there's no way to enable it). 2.6.3 disables VIA 
external APIC routing, as you noticed, whereas 2.6.7 sais 

Local APIC disabled by BIOS -- reenabling.
Found and enabled local APIC!

2.6.9 respects BIOS decision (I think, there was a thread on LKML on 
this?):

No local APIC present or hardware disabled

> Can you forward the /proc/interrupts from 2.6.3, and from 2.6.9 with and
> without acpi=off?  do you see a significant change in /proc/interrupts
> before and after the sensor-provoked slowness starts?
> 
> if you build 2.6.9 w/o the CONFIG_ACPI_PROCESSOR and boot w/o cmdline
> params, do you still see slowness?
> 
> if you boot 2.6.9 with these parameters, do you see any additional dmesg
> lines?
> 
> acpi_dbg_level=0xF acpi_dbg_layer=0xFFFF3FFF

I'll try to do all this tomorrow and report results.

Thanks
Guennadi
---
Guennadi Liakhovetski

Re: [sensors] system slow since ~ 2.6.7

[Guennadi Liakhovetski]

(added Andrew to CC as he also answered my original email. Don't know if 
sensors@stimpy.netroedge.com allows non-subscribers)

On Tue, 23 Nov 2004, Len Brown wrote:

> On Wed, 2004-11-17 at 19:25, Guennadi Liakhovetski wrote:
> > "Slow" means just running top alone in a vt it takes 1.6% CPU. Under
> > 2.6.3 it takes 0.2% (Duron 900MHz). Another peculiarity with 2.6.7 and
> > 2.6.9 is that the power LED is blinking with about 1Hz frequency. It's
> > an ASUS A7VI-VM motherboard. In the manual there's nothing about
> 
> PCI: Disabling Via external APIC routing
> 
> Curiously, this line appears in 2.6.3, but not in 2.6.7 or 2.6.9 dmesg
> -- even though all the configs build in IOAPIC support.
> 
> Can you forward the /proc/interrupts from 2.6.3, and from 2.6.9 with and
> without acpi=off?  do you see a significant change in /proc/interrupts
> before and after the sensor-provoked slowness starts?
> 
> if you build 2.6.9 w/o the CONFIG_ACPI_PROCESSOR and boot w/o cmdline
> params, do you still see slowness?
> 
> if you boot 2.6.9 with these parameters, do you see any additional dmesg
> lines?
> 
> acpi_dbg_level=0xF acpi_dbg_layer=0xFFFF3FFF

Ok, I started debugging the problem closely, and after booting into 2.6.9 
with acpi=off I still could reproduce the problem by starting sensors... 
So, I guess, there's no need to do all the acpi debugging you are 
suggesting above, right? As for /proc/interrupts with / without acpi and 
before / after sensors I don't see any difference. Notice also, that the 
slowness doesn't necessarily start immediately after starting sensors, it 
can start later, and it can spontaneously stop later. Just now while 
typing this email I saw the power LED stopped blinking and the speed went 
back to normal.

This reminds me: about a year ago my CPU fan burnt down. Then too, shortly 
after booting the PC, it slowed down. Then by accident I noticed in BIOS 
CPU temperature 98 deg C. With a new fan problem disappeared.

So, can it be, that the BIOS automatically slows down (throttles) the CPU 
at high temperature. And after ~ 2.6.7 sensors program the sensor 
interface with some (wrong) coefficient, and then it throttles the CPU 
wrongly? Yes, some coefficients are definitely wrong. Here are a couple of 
snapshots:

via686a-isa-e200
Adapter: ISA adapter
CPU core:  +1.09 V  (min =  +2.00 V, max =  +2.50 V)   ALARM
+2.5V:     +1.16 V  (min =  +3.10 V, max =  +1.57 V)   ALARM
I/O:       +3.40 V  (min =  +4.13 V, max =  +4.13 V)   ALARM
+5V:       +5.55 V  (min =  +6.44 V, max =  +6.44 V)   ALARM
+12V:      +4.81 V  (min = +15.60 V, max = +15.60 V)   ALARM
CPU Fan:  5443 RPM  (min =    0 RPM, div = 2)          
P/S Fan:     0 RPM  (min =    0 RPM, div = 2)          
SYS Temp:  +45.4 C  (high =   +45 C, hyst =   +40 C)   ALARM
CPU Temp:  +34.5 C  (high =   +60 C, hyst =   +55 C)   
SBr Temp:  +28.4 C  (high =   +65 C, hyst =   +60 C)   

via686a-isa-e200
Adapter: ISA adapter
CPU core:  +1.09 V  (min =  +2.00 V, max =  +2.50 V)   ALARM
+2.5V:     +1.16 V  (min =  +3.10 V, max =  +1.57 V)   ALARM
I/O:       +3.40 V  (min =  +4.13 V, max =  +4.13 V)   ALARM
+5V:       +5.55 V  (min =  +6.44 V, max =  +6.44 V)   ALARM
+12V:      +4.81 V  (min = +15.60 V, max = +15.60 V)   ALARM
CPU Fan:  5487 RPM  (min =    0 RPM, div = 2)          
P/S Fan:     0 RPM  (min =    0 RPM, div = 2)          
SYS Temp:  +45.2 C  (high =   +91 C, hyst =   +40 C)   ALARM
CPU Temp:  +34.4 C  (high =   +60 C, hyst =   +55 C)   
SBr Temp:  +28.4 C  (high =   +65 C, hyst =   +60 C)   

Notice how SYS Temp high changed... Can my guesses be correct and how 
can the situation be fixed? Again - no problems with 2.6.3.

Thanks
Guennadi
---
Guennadi Liakhovetski

Re: [sensors] system slow since ~ 2.6.7

[Jean Delvare]

> (added Andrew to CC as he also answered my original email. Don't know
> if sensors-LT64U7CwzWEZMC/lefXSXFaTQe2KTcn/@public.gmane.org allows non-subscribers)

Yes it does :) Now I hope acpi-devel does too.

> This reminds me: about a year ago my CPU fan burnt down. Then too,
> shortly after booting the PC, it slowed down. Then by accident I
> noticed in BIOS CPU temperature 98 deg C. With a new fan problem
> disappeared.

Wow, 98, no less. You're lucky it didn't catch on fire, you know.

> So, can it be, that the BIOS automatically slows down (throttles) the
> CPU at high temperature.

Yes this is certainly what happens. Blinking power led usually means
that some kinf of thermal or power management is in effect.

> And after ~ 2.6.7 sensors program the sensor 
> interface with some (wrong) coefficient, and then it throttles the CPU
> wrongly?

There were some significant changes to the via686a driver in 2.6.6 and
2.6.7.

2.6.6: Limit initialization was removed from the driver. The same was
done for most other drivers. The limits have to be set by either the
BIOS or user-space, not kernel drivers. Also, chip initialization is now
less agressive (previous version would possibly arbitrarily overwrite
BIOS settings).

2.6.7: Conversion formulas were reworked for a better accuracy. Errors
were previously introduced by incorrect rounding.

I think that the changes in 2.6.6 are the ones affecting you.

> Yes, some coefficients are definitely wrong. Here are a
> couple of snapshots:
> 
> via686a-isa-e200
> Adapter: ISA adapter
> CPU core:  +1.09 V  (min =  +2.00 V, max =  +2.50 V)   ALARM
> +2.5V:     +1.16 V  (min =  +3.10 V, max =  +1.57 V)   ALARM
> I/O:       +3.40 V  (min =  +4.13 V, max =  +4.13 V)   ALARM
> +5V:       +5.55 V  (min =  +6.44 V, max =  +6.44 V)   ALARM
> +12V:      +4.81 V  (min = +15.60 V, max = +15.60 V)   ALARM
> CPU Fan:  5443 RPM  (min =    0 RPM, div = 2)          
> P/S Fan:     0 RPM  (min =    0 RPM, div = 2)          
> SYS Temp:  +45.4 C  (high =   +45 C, hyst =   +40 C)   ALARM
> CPU Temp:  +34.5 C  (high =   +60 C, hyst =   +55 C)   
> SBr Temp:  +28.4 C  (high =   +65 C, hyst =   +60 C)   

Blame your BIOS! It did not properly configure voltage limits, among
others. BTW, Vcore, +2.5V and +12V look awfully wrong anyway. I/O and
+5V are acceptable but even +5V is a bit too high IMHO. Never heard of
your motherboard model before, seems to be a rare one. Maybe Asus didn't
put much support on it. 

Notice the ALARM in SYS Temp, which is probably causing the system to
throttle.

> via686a-isa-e200
> Adapter: ISA adapter
> CPU core:  +1.09 V  (min =  +2.00 V, max =  +2.50 V)   ALARM
> +2.5V:     +1.16 V  (min =  +3.10 V, max =  +1.57 V)   ALARM
> I/O:       +3.40 V  (min =  +4.13 V, max =  +4.13 V)   ALARM
> +5V:       +5.55 V  (min =  +6.44 V, max =  +6.44 V)   ALARM
> +12V:      +4.81 V  (min = +15.60 V, max = +15.60 V)   ALARM
> CPU Fan:  5487 RPM  (min =    0 RPM, div = 2)          
> P/S Fan:     0 RPM  (min =    0 RPM, div = 2)          
> SYS Temp:  +45.2 C  (high =   +91 C, hyst =   +40 C)   ALARM
> CPU Temp:  +34.4 C  (high =   +60 C, hyst =   +55 C)   
> SBr Temp:  +28.4 C  (high =   +65 C, hyst =   +60 C)   
> 
> Notice how SYS Temp high changed...

Did it change *on its own*? Weird. Note that 91 = 45 << 1 + 1. I wonder
if it could be some kind of read error. Will the value change 

> Can my guesses be correct and how can the situation be fixed?

Pick the latest default configuration file for sensors here:
http://www2.lm-sensors.nu/%7Elm78/cvs/lm_sensors2/etc/sensors.conf.eg
Save as /etc/sensors.conf, edit the via686a-* section, especially the
"set temp1_high" and set temp1_hyst" values. I'd suggest:

    set temp1_hyst 55
    set temp1_over 60

Save the changes and run "sensors -s". Your system should hopefully be
back to full speed right after that. So all you have to do is make sure
that "sensors -s" is called after you load the via686a driver at boot
time.

You should take a look at the hardware monitoring options in your BIOS
setup screen if it has any. Maybe you can configure the boot value of
temp1_high directly there, and it may provide hints about voltages as
well.

-- 
Jean Delvare
http://khali.linux-fr.org/


-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/

Re: [sensors] system slow since ~ 2.6.7

[Guennadi Liakhovetski]

Hello and thanks for the help!

On Thu, 25 Nov 2004, Jean Delvare wrote:

> > This reminds me: about a year ago my CPU fan burnt down. Then too,
> > shortly after booting the PC, it slowed down. Then by accident I
> > noticed in BIOS CPU temperature 98 deg C. With a new fan problem
> > disappeared.
> 
> Wow, 98, no less. You're lucky it didn't catch on fire, you know.

Yep, I do:-) I read some hot reports about hot AMDs right after that 
adventure:-)

> There were some significant changes to the via686a driver in 2.6.6 and
> 2.6.7.
> 
> 2.6.6: Limit initialization was removed from the driver. The same was
> done for most other drivers. The limits have to be set by either the
> BIOS or user-space, not kernel drivers. Also, chip initialization is now
> less agressive (previous version would possibly arbitrarily overwrite
> BIOS settings).
> 
> 2.6.7: Conversion formulas were reworked for a better accuracy. Errors
> were previously introduced by incorrect rounding.
> 
> I think that the changes in 2.6.6 are the ones affecting you.

In the meantime I tried 2.6.5 - same behaviour as 2.6.7.

> > Yes, some coefficients are definitely wrong. Here are a
> > couple of snapshots:
> > 
> > via686a-isa-e200
> > Adapter: ISA adapter
> > CPU core:  +1.09 V  (min =  +2.00 V, max =  +2.50 V)   ALARM
> > +2.5V:     +1.16 V  (min =  +3.10 V, max =  +1.57 V)   ALARM
> > I/O:       +3.40 V  (min =  +4.13 V, max =  +4.13 V)   ALARM
> > +5V:       +5.55 V  (min =  +6.44 V, max =  +6.44 V)   ALARM
> > +12V:      +4.81 V  (min = +15.60 V, max = +15.60 V)   ALARM
> > CPU Fan:  5443 RPM  (min =    0 RPM, div = 2)          
> > P/S Fan:     0 RPM  (min =    0 RPM, div = 2)          
> > SYS Temp:  +45.4 C  (high =   +45 C, hyst =   +40 C)   ALARM
> > CPU Temp:  +34.5 C  (high =   +60 C, hyst =   +55 C)   
> > SBr Temp:  +28.4 C  (high =   +65 C, hyst =   +60 C)   
> 
> Blame your BIOS! It did not properly configure voltage limits, among
> others. BTW, Vcore, +2.5V and +12V look awfully wrong anyway. I/O and
> +5V are acceptable but even +5V is a bit too high IMHO. Never heard of
> your motherboard model before, seems to be a rare one. Maybe Asus didn't
> put much support on it. 

Well, don't know how rare it is - I did find it on the ASUS site.

> Notice the ALARM in SYS Temp, which is probably causing the system to
> throttle.

Yep. However, I saw ALARM without throttling sometimes too.

> > via686a-isa-e200
> > Adapter: ISA adapter
> > CPU core:  +1.09 V  (min =  +2.00 V, max =  +2.50 V)   ALARM
> > +2.5V:     +1.16 V  (min =  +3.10 V, max =  +1.57 V)   ALARM
> > I/O:       +3.40 V  (min =  +4.13 V, max =  +4.13 V)   ALARM
> > +5V:       +5.55 V  (min =  +6.44 V, max =  +6.44 V)   ALARM
> > +12V:      +4.81 V  (min = +15.60 V, max = +15.60 V)   ALARM
> > CPU Fan:  5487 RPM  (min =    0 RPM, div = 2)          
> > P/S Fan:     0 RPM  (min =    0 RPM, div = 2)          
> > SYS Temp:  +45.2 C  (high =   +91 C, hyst =   +40 C)   ALARM
> > CPU Temp:  +34.4 C  (high =   +60 C, hyst =   +55 C)   
> > SBr Temp:  +28.4 C  (high =   +65 C, hyst =   +60 C)   
> > 
> > Notice how SYS Temp high changed...
> 
> Did it change *on its own*?

Hm, at least, I certainly didn't do anything to change it.

> Weird. Note that 91 = 45 << 1 + 1. I wonder
> if it could be some kind of read error. Will the value change 

Sorry? Did you mean "will it change again after that?" Well, not sure any 
more, but, I think, it did.

> > Can my guesses be correct and how can the situation be fixed?
> 
> Pick the latest default configuration file for sensors here:
> http://www2.lm-sensors.nu/%7Elm78/cvs/lm_sensors2/etc/sensors.conf.eg
> Save as /etc/sensors.conf, edit the via686a-* section, especially the
> "set temp1_high" and set temp1_hyst" values. I'd suggest:
> 
>     set temp1_hyst 55
>     set temp1_over 60
> 
> Save the changes and run "sensors -s". Your system should hopefully be
> back to full speed right after that. So all you have to do is make sure
> that "sensors -s" is called after you load the via686a driver at boot
> time.

Indeed! Just modifying these 2 values in my sensors.conf brought the 
system back to normal under 2.6.9! I'll get the latest config later. 
Thanks!

> You should take a look at the hardware monitoring options in your BIOS
> setup screen if it has any. Maybe you can configure the boot value of
> temp1_high directly there, and it may provide hints about voltages as
> well.

In BIOS one can only monitor sensors.

Thanks
Guennadi
---
Guennadi Liakhovetski



-------------------------------------------------------
SF email is sponsored by - The IT Product Guide
Read honest & candid reviews on hundreds of IT Products from real users.
Discover which products truly live up to the hype. Start reading now. 
http://productguide.itmanagersjournal.com/