Athlon doesn't like Athlon optimisation?

Athlon doesn't like Athlon optimisation?

[Jan Niehusmann]

I have a computer with a duron 600 which doesn't like current athlon
optimised kernels: It runs fairly well with an old 2.4.0-test7 kernel
(but I had some unexplained crashes during the last months),
but crashes after a few minutes after booting 2.4.9-ac3 or 2.4.7. 

If I don't build the kernels for athlon, but for i386 only, the 
system seems to be stable. (Not tested for more than 20 minutes, 
but definitely longer than the athlon optimised kernel was able to run)

Does anybody know these symptoms and has an idea what may be wrong?
Is it likely to be a broken CPU? 
The board is an A7V with the infamous via chipset, but I don't think
this looks like the typical via problems, does it?

Jan

-- 
OpenPGP-signierte bzw. -verschlüsselte Mail erwünscht
EMail-Key: 1024D/F12DA065 (=> Keyserver oder auf Anfrage)

Re: Athlon doesn't like Athlon optimisation?

[David Hollister]

Jan Niehusmann wrote:
> I have a computer with a duron 600 which doesn't like current athlon
> optimised kernels: It runs fairly well with an old 2.4.0-test7 kernel
> (but I had some unexplained crashes during the last months),
> but crashes after a few minutes after booting 2.4.9-ac3 or 2.4.7. 
> 
> If I don't build the kernels for athlon, but for i386 only, the 
> system seems to be stable. (Not tested for more than 20 minutes, 
> but definitely longer than the athlon optimised kernel was able to run)
> 
> Does anybody know these symptoms and has an idea what may be wrong?
> Is it likely to be a broken CPU? 
> The board is an A7V with the infamous via chipset, but I don't think
> this looks like the typical via problems, does it?
> 
> Jan

This has apparently been a source of frustration for many an Athlon user, myself 
included.  I can't even get my system to finish the init process before it 
oopses and locks up on me.

It seems to work somewhat better for some if you set your BIOS to the 
conservative settings, but that didn't help me.  I have an Epox 8KTA3+ (Via 
KT133A) w/ a 1.4GHz Athlon and 512MB memory.  If you can't get it to work that 
way, just stick with the K6 setting.  The point is, your hardware is likely fine 
(fine being relative, I suppose)
If there are other tricks, I'm all ears.

-- 
David Hollister
Driversoft Engineering:  http://devicedrivers.com
Digital Audio Resources: http://digitalaudioresources.org

Re: Athlon doesn't like Athlon optimisation?

[Albert D. Cahalan]

Richard, the FAQ could use an entry about this and the other VIA problem.

David Hollister writes:
> Jan Niehusmann wrote:

>> I have a computer with a duron 600 which doesn't like current athlon
>> optimised kernels: It runs fairly well with an old 2.4.0-test7 kernel
...
>> Is it likely to be a broken CPU? 
>> The board is an A7V with the infamous via chipset, but I don't think
>> this looks like the typical via problems, does it?
...
> point is, your hardware is likely fine (fine being relative, I
> suppose) If there are other tricks, I'm all ears.

There are highly optimized memory copy/clear operations that
run twice as fast on the Athlon, thus demanding more from the
motherboard and power supply. You have a VIA chipset and most
likely have a relatively weak power supply.

Don't go blaming Linux when power supply upgrades sometimes
make this problem go away. You could also try one of the
recent SiS or ALi chipsets.

I just saw a reference (maybe www.tomshardware.com) to AMD's new
chips having trouble on VIA boards -- I'd guess that the Palimino
core can push the motherboard too hard without fancy Athlon code.

Re: Athlon doesn't like Athlon optimisation?

[Dan Hollis]

On Thu, 30 Aug 2001, Albert D. Cahalan wrote:
> Don't go blaming Linux when power supply upgrades sometimes
> make this problem go away. You could also try one of the
> recent SiS or ALi chipsets.
> I just saw a reference (maybe www.tomshardware.com) to AMD's new
> chips having trouble on VIA boards -- I'd guess that the Palimino
> core can push the motherboard too hard without fancy Athlon code.

So what happens when someone is able to duplicate the problem on say AMD
760MP chipset with registered ECC PC2100 ram and 450W power supply?

Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
but where would the finger start pointing then?

-Dan

-- 
[-] Omae no subete no kichi wa ore no mono da. [-]

Re: Athlon doesn't like Athlon optimisation?

[Chris Abbey]

Today, Dan Hollis wrote:
> but where would the finger start pointing then?

hmm... *compiler optimizations* for a specific family cause
problems on that family, but *compiler optimizations* for
a lesser family don't... I'll admit my kernel h4x0|^ 5k1!!s
aren't on par with most on this list, but has anyone thought
to take a look at the *compiler optimizations* that are
generated? It sure wouldn't be a first if the combination
of agressive optimizations and complex kernel code exposed
a subtle and/or complex bug in one, the other, or both...
and different levels of compiler might explain why some
have the problems, and others don't.

Having spent way too many hours this week looking at highly
optimized 64bit ppc assembly I can only say that <drawl
type=hick> them thar compiler hackors is devious lil
twerps when yas ask fer all the bells n wistles. </drawl>

This of course is all assuming that one or more folks do
recreate it on obviously good hardware. ;)

-- 
now the forces of openness have a powerful and
  unexpected new ally - http://ibm.com/linux

Re: Athlon doesn't like Athlon optimisation?

[Arjan van de Ven]

Chris Abbey wrote:
> 
> Today, Dan Hollis wrote:
> > but where would the finger start pointing then?
> 
> hmm... *compiler optimizations* for a specific family cause
> problems on that family, but *compiler optimizations* for
> a lesser family don't... I'll admit my kernel h4x0|^ 5k1!!s
> aren't on par with most on this list, but has anyone thought
> to take a look at the *compiler optimizations* that are
> generated? 

It's not the compiler options. (or at least not alone).
I have proof for this, let me explain:

For the upcomming Red Hat Linux release an athlon kernel
will be included, and due to the people who have this
problem, I added a kernel commandline option to disable
the optimized page_copy() and clear_page() functions.
The use of this option makes the machines, of the people
who had this problem, happy again.

Now I also wrote the 2 functions in question, and I am
very convinced that they are correct. They also work on
the vast majority of motherboards, and most of the failure
cases are cheaper motherboards (or cheap PSU's).

The net effect of using the optimized functions is that
the memory bandwidth the CPU uses effecively doubles during
COW and page_clear() operations. This puts additional load
on the motherboard it seems.... I don't know if it's the
voltage regulators or borderline ram chips that give up, 
but there are people who bought 25 identical machines (for
a classroom) and only 1 failed, reproducable.

Oh and btw, having these functions is the main reason for
enabling the "Athlon" CPU type; that's basically the real 
difference between a PII and Athlon compiled kernel.

Greetings,
    Arjan van de Ven

Re: Athlon doesn't like Athlon optimisation?

[Nicholas Knight]

On Friday 31 August 2001 01:27 am, Arjan van de Ven wrote:

<snip>

> Now I also wrote the 2 functions in question, and I am
> very convinced that they are correct. They also work on
> the vast majority of motherboards, and most of the failure
> cases are cheaper motherboards (or cheap PSU's).

Want cheap hardware?
Slot-A Athlon 800Mhz non-tbird
Soyo K7VIA motherboard, VIA KX133 chipset
Generic 300W power supply bought from mwave

and, for the other hardware:

1 DVD-ROM drive
1 Plextor PlexWriter CD-RW drive
1 IBM 75GXP 7200RPM 45GB ATA/100 drive
1 Promise ATA/100 HDD controller
1 350MB Western Digital Caviar drive, but this powers down completely 
when not in use
2 Generic 80mm case fans
1 low-end HSF unit with three small fans
1 GeForce2MX
1 stick of generic 256MB PC133 SDRAM that refuses to run at 133Mhz
1 ISA NIC
1 PCI NIC
1 floppy drive, sony I think, otherwise generic

I have never experienced problems with Athlon optimizations, and I 
*always* compile with them enabled.

Prehaps the key lies in Thunderbird vs Non-Thunderbird? When did these 
problems start showing up, and at what clock speeds? What motherboards? 
What chipsets?

>Greetings,
>     Arjan van de Ven

Re: Athlon doesn't like Athlon optimisation?

[David Hollister]

Arjan van de Ven wrote:
> 
> For the upcomming Red Hat Linux release an athlon kernel
> will be included, and due to the people who have this
> problem, I added a kernel commandline option to disable
> the optimized page_copy() and clear_page() functions.
> The use of this option makes the machines, of the people
> who had this problem, happy again.
> 
> Now I also wrote the 2 functions in question, and I am
> very convinced that they are correct. They also work on
> the vast majority of motherboards, and most of the failure
> cases are cheaper motherboards (or cheap PSU's).

Hey look, folks.  I didn't point a finger and try to blame anybody or anything. 
  I'm as much a Linux advocate as the next guy.  Granted, I have not tried every 
trick under the sun to get it to work.  I don't really care that much.  I can 
live with my memory accesses taking a few microseconds longer.  My point, and my 
only point, to all this was just to add data.  If there's something relatively 
easy I can try, I will.  Otherwise, life goes on.
-- 
David Hollister
Driversoft Engineering:  http://devicedrivers.com
Digital Audio Resources: http://digitalaudioresources.org

Re: Athlon doesn't like Athlon optimisation?

[Paul G. Allen]

Dan Hollis wrote:
> 
> On Thu, 30 Aug 2001, Albert D. Cahalan wrote:
> > Don't go blaming Linux when power supply upgrades sometimes
> > make this problem go away. You could also try one of the
> > recent SiS or ALi chipsets.
> > I just saw a reference (maybe www.tomshardware.com) to AMD's new
> > chips having trouble on VIA boards -- I'd guess that the Palimino
> > core can push the motherboard too hard without fancy Athlon code.
> 
> So what happens when someone is able to duplicate the problem on say AMD
> 760MP chipset with registered ECC PC2100 ram and 450W power supply?
> 
> Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
> but where would the finger start pointing then?
> 

My Dual Athlon (1.4GHz) works just fine (with the exception of the ATA 100 - I have to disable DMA).

It's been running 24/7 for weeks at 100% CPU usage (on both CPUs).

My A7V133 is another story (also a 1.4GHz). It used to work with RH 7.1, now I can't even get the OS to install.

PGA

-- 
Paul G. Allen
UNIX Admin II/Programmer
Akamai Technologies, Inc.
www.akamai.com
Work: (858)909-3630
Cell: (858)395-5043

Re: Athlon doesn't like Athlon optimisation?

[Steven Spence]

Paul G. Allen wrote:

>Dan Hollis wrote:
>
>>On Thu, 30 Aug 2001, Albert D. Cahalan wrote:
>>
>>>Don't go blaming Linux when power supply upgrades sometimes
>>>make this problem go away. You could also try one of the
>>>recent SiS or ALi chipsets.
>>>I just saw a reference (maybe www.tomshardware.com) to AMD's new
>>>chips having trouble on VIA boards -- I'd guess that the Palimino
>>>core can push the motherboard too hard without fancy Athlon code.
>>>
>>So what happens when someone is able to duplicate the problem on say AMD
>>760MP chipset with registered ECC PC2100 ram and 450W power supply?
>>
>>Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
>>but where would the finger start pointing then?
>>
>
>My Dual Athlon (1.4GHz) works just fine (with the exception of the ATA 100 - I have to disable DMA).
>
>It's been running 24/7 for weeks at 100% CPU usage (on both CPUs).
>
>My A7V133 is another story (also a 1.4GHz). It used to work with RH 7.1, now I can't even get the OS to install.
>
>PGA
>
My dual 1.2 Athlon MPs  work great as well.  I have my DMA on without 
any troubles though.  Actually I have never had
a problem with Athlons and Linux.  I have a 650MHz and a 1GHz as well 
and they are rock solid.  I would look at your
check your hardware before blaming the software.  I have compiled kernel 
after kernel, made a couple graphic posters of
the kernel, and run many simulations on it without incident.

I did however have one of those mammoth 460W power supplies fail on me 
though.  As long as you don't use the AGP Pro
slot on the board you should have more then enough power from it.  I 
have 4 hard drives on it, two cdroms working, and
about 8 fans without any problems.   

Steve

Re: Athlon doesn't like Athlon optimisation?

[Alan Cox]

> My Dual Athlon (1.4GHz) works just fine (with the exception of the ATA 100 -
> I have to disable DMA).

There is a known errata with the dual athlon chipset where prefetching and
IDE DMA together hang the box. You might want to scan the chip docs/errata
and try turning that bit off and see if it helps. If so its one for pci
quirks

Re: Athlon doesn't like Athlon optimisation?

[Phillip Susi]

I have just such a system and thought I'd note that the stock smp kernels 
that redhat 7.1 and slackware 8.0 installed would lock up intermitently.  I 
couldn't get through a build of the kernel.  I ended up booting into the UP 
kernel, got and built the 2.4.9 kernel, and have not had any trouble since.  
For the processor type, I selected athlon, that enables the optimizations you 
are talking about right?

System specs: dual Athlon MP 1.2 Ghz on Tyan's mobo with the dual ethernet 
and u160 scsi, 512 megs corsair cas 2.5 registered pc2100 ddr sdram, and 
seagate's second generation 15,000 rpm cheetah, using the NMB 460 watt power 
supply that tyan recomended.  

On Friday 31 August 2001 04:20 am, Dan Hollis wrote:
>
> So what happens when someone is able to duplicate the problem on say AMD
> 760MP chipset with registered ECC PC2100 ram and 450W power supply?
>
> Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
> but where would the finger start pointing then?
>
> -Dan

Re: Athlon doesn't like Athlon optimisation? [PATCH]

[Radu-Adrian Feurdean]

wolfram@schlich.org wrote:
> Dan Hollis <goemon@anime.net> wrote:
> 
>>On Thu, 30 Aug 2001, Albert D. Cahalan wrote:
>>
>>>Don't go blaming Linux when power supply upgrades sometimes
>>>make this problem go away. You could also try one of the
>>>recent SiS or ALi chipsets.
>>>I just saw a reference (maybe www.tomshardware.com) to AMD's new
>>>chips having trouble on VIA boards -- I'd guess that the Palimino
>>>core can push the motherboard too hard without fancy Athlon code.
>>>
> 
>>So what happens when someone is able to duplicate the problem on say AMD
>>760MP chipset with registered ECC PC2100 ram and 450W power supply?
>>
> 
>>Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
>>
> 
> it has happened to me. dual athlon mp 1.2ghz w/ crucial reg. ddr sdram.
> after setting CONFIG_MK7 to CONFIG_MK6 all works fine.

The following worked for me (Duron@800, Epox 8KTA2L)

root@WormHole:/usr/src# diff -u linux-2.4.9/arch/i386/lib/mmx.c{~,}
--- linux-2.4.9/arch/i386/lib/mmx.c~	Tue May 22 19:23:16 2001
+++ linux-2.4.9/arch/i386/lib/mmx.c	Fri Aug 31 15:51:58 2001
@@ -97,7 +97,7 @@
  	return p;
  }

-#ifdef CONFIG_MK7
+#if 0

  /*
   *	The K7 has streaming cache bypass load/store. The Cyrix III, K6 and

-- 
Radu-Adrian Feurdean
mailto: raf@chez.com
------------------------------------------------------------------------------
+#if defined(__alpha__) && defined(CONFIG_PCI)
+       /*
+        * The meaning of life, the universe, and everything. Plus
+        * this makes the year come out right.
+        */
+       year -= 42;
+#endif
     -- From the patch for 1.3.2: (kernel/time.c), submitted by Marcus Meissner

Re: Athlon doesn't like Athlon optimisation?

[Alan Cox]

> So what happens when someone is able to duplicate the problem on say AMD
> 760MP chipset with registered ECC PC2100 ram and 450W power supply?
> 
> Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
> but where would the finger start pointing then?

That would make it a lot more complex. There were a few cases much earlier
on with AMD chipset lockups but those have been cured (and were an Athlon
processor errata where a prefetch of an uncacheable line made a very very
nasty mess).

Alan

Re: Athlon doesn't like Athlon optimisation?

[Dan Hollis]

On Fri, 31 Aug 2001, Alan Cox wrote:
> > So what happens when someone is able to duplicate the problem on say AMD
> > 760MP chipset with registered ECC PC2100 ram and 450W power supply?
> > Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
> > but where would the finger start pointing then?
> That would make it a lot more complex. There were a few cases much earlier
> on with AMD chipset lockups but those have been cured (and were an Athlon
> processor errata where a prefetch of an uncacheable line made a very very
> nasty mess).

Can you define a hardware configuration that if it fails under athlon
optimizations, you would consider a falsification of the "marginal
hardware" theory?

I want a hardware config that will get everyones attention if it fails.

-Dan

-- 
[-] Omae no subete no kichi wa ore no mono da. [-]

Re: Athlon doesn't like Athlon optimisation?

[Alan Cox]

> optimizations, you would consider a falsification of the "marginal
> hardware" theory?

Not trivially. 

The current theory is
	VIA chipset + Athlon + [unknown factors]

So seeing it on SiS, AMD or Ali chipsets would be significant

Re: Athlon doesn't like Athlon optimisation?

[Eric W. Biederman]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

> > optimizations, you would consider a falsification of the "marginal
> > hardware" theory?
> 
> Not trivially. 
> 
> The current theory is
> 	VIA chipset + Athlon + [unknown factors]
> 
> So seeing it on SiS, AMD or Ali chipsets would be significant

Would it help with the tracking if someone had a board that reliably
crashes before /etc/rc.d/rc finishes running?  And were willing to
help with the investigation?

Eric

Re: Athlon doesn't like Athlon optimisation?

[Mike Fedyk]

On Fri, Aug 31, 2001 at 01:37:16PM -0700, Dan Hollis wrote:
> On Fri, 31 Aug 2001, Alan Cox wrote:
> > > So what happens when someone is able to duplicate the problem on say AMD
> > > 760MP chipset with registered ECC PC2100 ram and 450W power supply?
> > > Not to say it has happened yet (I havent got my dual Tyan Tiger MP yet :-)
> > > but where would the finger start pointing then?
> > That would make it a lot more complex. There were a few cases much earlier
> > on with AMD chipset lockups but those have been cured (and were an Athlon
> > processor errata where a prefetch of an uncacheable line made a very very
> > nasty mess).
> 
> Can you define a hardware configuration that if it fails under athlon
> optimizations, you would consider a falsification of the "marginal
> hardware" theory?
> 
> I want a hardware config that will get everyones attention if it fails.
> 

>From other posts, that would be boards based on AMD chipsets, and possibly
SIS or ALI.

Re: Athlon doesn't like Athlon optimisation?

[Richard Gooch]

Albert D. Cahalan writes:
> Richard, the FAQ could use an entry about this and the other VIA
> problem.

How about sending me a patch to the HTML file? That's less work for me
:-)

				Regards,

					Richard....
Permanent: rgooch@atnf.csiro.au
Current:   rgooch@ras.ucalgary.ca

Re: Athlon doesn't like Athlon optimisation?

[Joerg Plate]

> It seems to work somewhat better for some if you set your BIOS to the
> conservative settings, but that didn't help me. I have an Epox 8KTA3+
> (Via KT133A) w/ a 1.4GHz Athlon and 512MB memory.

I'm using an Epox 8KT3+ (VIA KT133A) motherboard, too. The processor is a
1.333GHz Athlon and I never had problems with the optimisation (CONFIG_MK7=y).

-- 
"I'm working on it."  <http://www.psyche.kn-bremen.de/>

1998: U.S.Congress abolished Free Speech by replacing First Amendment with DMCA.

Re: Athlon doesn't like Athlon optimisation?

[Tim Moore]

> It seems to work somewhat better for some if you set your BIOS to the
> conservative settings, but that didn't help me.  I have an Epox 8KTA3+ (Via
> KT133A) w/ a 1.4GHz Athlon and 512MB memory.  If you can't get it to work that
> way, just stick with the K6 setting.  The point is, your hardware is likely fine
> (fine being relative, I suppose)
> If there are other tricks, I'm all ears.

The i686 setting works perfectly.

rgds,
tim.
--

Re: Athlon doesn't like Athlon optimisation?

[Bobby D. Bryant]

Tim Moore wrote:

> > It seems to work somewhat better for some if you set your BIOS to the
> > conservative settings, but that didn't help me.  I have an Epox 8KTA3+ (Via
> > KT133A) w/ a 1.4GHz Athlon and 512MB memory.  If you can't get it to work that
> > way, just stick with the K6 setting.  The point is, your hardware is likely fine
> > (fine being relative, I suppose)
> > If there are other tricks, I'm all ears.
>
> The i686 setting works perfectly.

For some people.  I have an 8KTA3+ that will boot as an i686, but starts oopsing its
shorts off after it has been up a while.

I posted some of the oopsen a few months ago, and to my feeble mind they all looked
memory related.  (Several were "bug in slab.c" kind of thing, IIRC.)

Bobby Bryant
Austin, Texas

Re: Athlon doesn't like Athlon optimisation?

[Tim Moore]

"Bobby D. Bryant" wrote:
> 
> Tim Moore wrote:
> 
> > > It seems to work somewhat better for some if you set your BIOS to the
> > > conservative settings, but that didn't help me.  I have an Epox 8KTA3+ (Via
> > > KT133A) w/ a 1.4GHz Athlon and 512MB memory.  If you can't get it to work that
> > > way, just stick with the K6 setting.  The point is, your hardware is likely fine
> > > (fine being relative, I suppose)
> > > If there are other tricks, I'm all ears.
> >
> > The i686 setting works perfectly.
> 
> For some people.  I have an 8KTA3+ that will boot as an i686, but starts oopsing its
> shorts off after it has been up a while.
> 
> I posted some of the oopsen a few months ago, and to my feeble mind they all looked
> memory related.  (Several were "bug in slab.c" kind of thing, IIRC.

I hadn't heard a case of 686 not working given the compatability between
Athlon/PIII.  To be clear, CONFIG_M686=y has worked perfectly with
850MHz Athlon + VT82C686.

rgds,
tim.
--

Re: Athlon doesn't like Athlon optimisation?

[Jim Roland]

Which kernel are you gentlemen using?  I have a Athlon 1.2GHz (not
overclocked), 512MB PC133, and also an EPoX 8KTA3+, and have had no problems
whatsoever (using kernel 2.4.2-2).

----- Original Message -----
From: "David Hollister" <david@digitalaudioresources.org>
To: "Jan Niehusmann" <jan@gondor.com>
Cc: <linux-kernel@vger.kernel.org>
Sent: Thursday, August 30, 2001 10:07 PM
Subject: Re: Athlon doesn't like Athlon optimisation?


> Jan Niehusmann wrote:
> > I have a computer with a duron 600 which doesn't like current athlon
> > optimised kernels: It runs fairly well with an old 2.4.0-test7 kernel
> > (but I had some unexplained crashes during the last months),
> > but crashes after a few minutes after booting 2.4.9-ac3 or 2.4.7.
> >
> > If I don't build the kernels for athlon, but for i386 only, the
> > system seems to be stable. (Not tested for more than 20 minutes,
> > but definitely longer than the athlon optimised kernel was able to run)
> >
> > Does anybody know these symptoms and has an idea what may be wrong?
> > Is it likely to be a broken CPU?
> > The board is an A7V with the infamous via chipset, but I don't think
> > this looks like the typical via problems, does it?
> >
> > Jan
>
> This has apparently been a source of frustration for many an Athlon user,
myself
> included.  I can't even get my system to finish the init process before it
> oopses and locks up on me.
>
> It seems to work somewhat better for some if you set your BIOS to the
> conservative settings, but that didn't help me.  I have an Epox 8KTA3+
(Via
> KT133A) w/ a 1.4GHz Athlon and 512MB memory.  If you can't get it to work
that
> way, just stick with the K6 setting.  The point is, your hardware is
likely fine
> (fine being relative, I suppose)
> If there are other tricks, I'm all ears.
>
> --
> David Hollister
> Driversoft Engineering:  http://devicedrivers.com
> Digital Audio Resources: http://digitalaudioresources.org
>
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: Athlon doesn't like Athlon optimisation?

[Joerg Plate]

> Which kernel are you gentlemen using?
No problems with: 2.4.5, 2.4.5ac24, 2.4.6, 2.4.7, 2.4.8ac4, 2.4.8ac7, 2.4.9ac5

-- 
"I'm working on it."  <http://www.psyche.kn-bremen.de/>

1998: U.S.Congress abolished Free Speech by replacing First Amendment with DMCA.

Re: Athlon doesn't like Athlon optimisation?

[David Hollister]

Jim Roland wrote:
> Which kernel are you gentlemen using?  I have a Athlon 1.2GHz (not
> overclocked), 512MB PC133, and also an EPoX 8KTA3+, and have had no problems
> whatsoever (using kernel 2.4.2-2).

I'm on 2.4.9.  No overclocking.  I applied the patch that somebody (sorry, 
forgot who) posted yesterday for arch/i386/lib/mmx.c and rebuilt the kernel with 
Athlon optimization.  It now works.
-- 
David Hollister
Driversoft Engineering:  http://devicedrivers.com
Digital Audio Resources: http://digitalaudioresources.org

Re: Athlon doesn't like Athlon optimisation?

[Alan Cox]

> I'm on 2.4.9.  No overclocking.  I applied the patch that somebody (sorry, 
> forgot who) posted yesterday for arch/i386/lib/mmx.c and rebuilt the kernel with 
> Athlon optimization.  It now works.

Well not really. The patch posted turns off athlon optimisation even though
you selected it

Re: Athlon doesn't like Athlon optimisation?

[David Hollister]

Alan Cox wrote:
>>I'm on 2.4.9.  No overclocking.  I applied the patch that somebody (sorry, 
>>forgot who) posted yesterday for arch/i386/lib/mmx.c and rebuilt the kernel with 
>>Athlon optimization.  It now works.
>>
> 
> Well not really. The patch posted turns off athlon optimisation even though
> you selected it

Well, that's what I thought, too, since that was the only file in the kernel 
source where CONFIG_MK7 was defined.  Somebody pointed out to me, though, that 
there are other defines that are turned on if you use Athlon settings, like 
(from arch/i386/config.in):

CONFIG_X86_GOOD_APIC
CONFIG_X86_USE_3DNOW
CONFIG_X86_PGE

Although, looking at arch/i386/config.in again right now, I see none of these 
are Athlon specific.  That's what I get for trusting somebody else instead of 
looking into it myself.

Thanks.
-- 
David Hollister
Driversoft Engineering:  http://devicedrivers.com
Digital Audio Resources: http://digitalaudioresources.org

Re: Athlon doesn't like Athlon optimisation?

[Jim Roland]

Hmm...missed that patch, but I'm not on 2.4.9 yet.  Can someone email that
to me directly or repost to the list?

----- Original Message -----
From: "David Hollister" <david@digitalaudioresources.org>
To: "Jim Roland" <jroland@roland.net>
Cc: "Jan Niehusmann" <jan@gondor.com>; <linux-kernel@vger.kernel.org>
Sent: Saturday, September 01, 2001 9:39 AM
Subject: Re: Athlon doesn't like Athlon optimisation?


> Jim Roland wrote:
> > Which kernel are you gentlemen using?  I have a Athlon 1.2GHz (not
> > overclocked), 512MB PC133, and also an EPoX 8KTA3+, and have had no
problems
> > whatsoever (using kernel 2.4.2-2).
>
> I'm on 2.4.9.  No overclocking.  I applied the patch that somebody (sorry,
> forgot who) posted yesterday for arch/i386/lib/mmx.c and rebuilt the
kernel with
> Athlon optimization.  It now works.
> --
> David Hollister
> Driversoft Engineering:  http://devicedrivers.com
> Digital Audio Resources: http://digitalaudioresources.org
>

Re: Athlon doesn't like Athlon optimisation?

[Dan Maas]

I now have three 1.2-1.4Ghz machines based on the Biostar M7MIA motherboard,
and none of them have had trouble with the Athlon optimizations... I was
very careful to select a motherboard based on the AMD760 northbridge, and to
avoid the non-AMD alternatives.

Dan

Re: Athlon doesn't like Athlon optimisation?

[Eric Olson]

I have found these Athlon problems are very interesting.

Is there a usermode memory testing program which uses the K7 MMX 3DNow
streaming cache bypass load/store instruction sequences that appear in
linux/arch/i386/lib/mmx.c ?

Could Robert Redelmeier's burnMMX at 
	http://users.ev1.net/~redelm/
be modified for the Athlon to detect these problems?

It would be usefull to have a Microsoft Windows program that could 
detect a faulty system without having to load Linux.  This would allow 
testing a system in a store before purchase, and quick testing of a 
new system shipped with Windows to determine whether it needs to be 
returned before reformatting the harddisk and installing Linux.

The reports I've know are for KT133 motherboards.  Have problems been 
reported with the KT266 DDR-SDRAM chipsets as well?

All the best, Eric Olson

Re: Athlon doesn't like Athlon optimisation?

[Ghozlane Toumi]

"Eric Olson" said not long ago:


> Could Robert Redelmeier's burnMMX at 
> http://users.ev1.net/~redelm/
> be modified for the Athlon to detect these problems?
> 
> This would allow 
> testing a system in a store before purchase, 

memtest86 could be extended for that 
as it's a stand alone mem tester, bootable from a floppy ...

www.memtest86.com

ghoz

Re: Athlon doesn't like Athlon optimisation?

[Dan Hollis]

On Mon, 3 Sep 2001, Ghozlane Toumi wrote:
> "Eric Olson" said not long ago:
> > Could Robert Redelmeier's burnMMX at
> > http://users.ev1.net/~redelm/
> > be modified for the Athlon to detect these problems?
> > This would allow
> > testing a system in a store before purchase,
> memtest86 could be extended for that
> as it's a stand alone mem tester, bootable from a floppy ...
> www.memtest86.com

Shouldnt that be "memtest86 *should* be extended for that".

It would be nice to have a bootable-floppy-testcase for
athlon-memory-optimization-failures.

-Dan

-- 
[-] Omae no subete no kichi wa ore no mono da. [-]

Re: Athlon doesn't like Athlon optimisation?

[Matthew S . Hallacy]

On Mon, Sep 03, 2001 at 01:13:51AM -0700, Eric Olson wrote:
> I have found these Athlon problems are very interesting.
> 
> Is there a usermode memory testing program which uses the K7 MMX 3DNow
> streaming cache bypass load/store instruction sequences that appear in
> linux/arch/i386/lib/mmx.c ?
> 
> Could Robert Redelmeier's burnMMX at 
> 	http://users.ev1.net/~redelm/
> be modified for the Athlon to detect these problems?
> 

Howdy, I've got a 1.3Ghz (non overclocked) running on the SiS 735 chipset
(yes, the first good chipset from SiS) with no problems under 2.4.9 with
athlon optimizations, the system has DDR ram, and is used a lot for video
capture/compression, the avifile library detects (and supposedly uses) the
optimizations, and I've had no problems whatsoever with the system. Perhaps
this is another VIA problem.

> It would be usefull to have a Microsoft Windows program that could 
> detect a faulty system without having to load Linux.  This would allow 
> testing a system in a store before purchase, and quick testing of a 
> new system shipped with Windows to determine whether it needs to be 
> returned before reformatting the harddisk and installing Linux.
> 
> The reports I've know are for KT133 motherboards.  Have problems been 
> reported with the KT266 DDR-SDRAM chipsets as well?

As I write this a friend is repairing a lost partition on his FIC AZ11E 
motherboard (KT133 chipset), apparently due to a BIOS bug..

> 
> All the best, Eric Olson

Re: Athlon doesn't like Athlon optimisation?

[Eric Olson]

Robert Redelmeier told me he has written a version of his burnMMX which 
uses K7 MMX 3DNow streaming cache bypass load/store instruction sequences
similar to what is used in linux/arch/i386/lib/mmx.c

He asked me to publicise that anyone who wanted to test it could contact
him though email.  His email address is available from the burnCPU website
	http://users.ev1.net/~redelm/
The program is 168 lines of assembler and compiles for Windows or Linux.  
It runs until it detects an error.  

It would be particularly interesting if someone with a problematic KT133A 
based motherboard would test it and report back.

All the best, -Eric

Re: Athlon doesn't like Athlon optimisation?

[Matthew S . Hallacy]

On Tue, Sep 04, 2001 at 10:21:58PM -0700, Eric Olson wrote:
> 
> Robert Redelmeier told me he has written a version of his burnMMX which 
> uses K7 MMX 3DNow streaming cache bypass load/store instruction sequences
> similar to what is used in linux/arch/i386/lib/mmx.c
> 

I'm happy to report that after leaving these (burnK7, and burnMMX) running for about 
30 minutes there were no problems, a slight increase in CPU/system temperature, but 
within safe limits, along with a nice load average.. FYI:

processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 4
model name      : AMD Athlon(tm) Processor
stepping        : 4
cpu MHz         : 1327.702
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 sep mtrr pge mca cmov pat pse36 mmx fxsr syscall mmxext 3dnowext 3dnow
bogomips        : 2647.65

note this is the "c" core chip,
256M DDR RAM,
PC Chips M830LR motherboard w/ the SiS 735 Chipset,
(note, my single-chip chipset requires no fan, unlike the VIA chipsets)

All rather well considering the CPU fan is missing two blades, and only has the little
pink patch of thermal grease.



				Matthew S. Hallacy

-- 

Re: Athlon doesn't like Athlon optimisation?

[David Hollister]

Eric Olson wrote:
> Robert Redelmeier told me he has written a version of his burnMMX which 
> uses K7 MMX 3DNow streaming cache bypass load/store instruction sequences
> similar to what is used in linux/arch/i386/lib/mmx.c
>  
> It would be particularly interesting if someone with a problematic KT133A 
> based motherboard would test it and report back.

Ran both burnK7 and burnMMX for 10+ minutes with no problems.  FWIW, burnMMX 
didn't do much to my system temp, but burnK7 raised it by a good 8-9 degrees.

-- 
David Hollister
Driversoft Engineering:  http://devicedrivers.com
Digital Audio Resources: http://digitalaudioresources.org

RE: Athlon doesn't like Athlon optimisation?

[noneuclidean]

I have an Iwill KK266 (VIAKT133A chipset, latest BIOS) with an unlocked but not overclocked Athlon 950 (AMD Thunderbird, A4, Model 4). The system DOES suffer the Athlon optimisation problem.

I ran burnK7, burnK7 in linux 2.4.8ac11 (optimised for K6) and WinME for over 3 hours with no problems.

For ?fun? I also tried running 50 mulitiple instances (in total) of a mix of burnK7, burnMMX, burnBX, burnP6 and burnK6 in linux (with different memory settings for burnBX and burnMMX), while accessing floppy, CD-ROM, 2xHDDs, my SBLive card and my Geforce 2 to try and load my voltages... but again completely stable, if a bit... well very... jerky!.

I think the burnK7 program does not test enough K7 specific instruction sets to find the problem.

Jamal Conway
I am not on this list, please CC replies.

RE: Athlon doesn't like Athlon optimisation?

[Simen Thoresen]

Jamal Conway wrote
>I think the burnK7 program does not test enough K7 specific instruction sets to find the problem.
>

I tend to agree; 30+ minutes on a Epox 8KTA3 motherboard with a processor that fails when running with Athlon-optimized fast_copy_page (now running without that), without a hitch.

-S
--
Simen Thoresen, Beowulf-cleaner and random artist - close and personal.

Er det ikke rart?
The gnu RART-project on http://valinor.dolphinics.no:1080/~simentt/rart

RE: Athlon doesn't like Athlon optimisation?

[Dan Hollis]

On Wed, 5 Sep 2001, noneuclidean wrote:
> I have an Iwill KK266 (VIAKT133A chipset, latest BIOS) with an unlocked but not overclocked Athlon 950 (AMD Thunderbird, A4, Model 4). The system DOES suffer the Athlon optimisation problem.
> I ran burnK7, burnK7 in linux 2.4.8ac11 (optimised for K6) and WinME for over 3 hours with no problems.
> For ?fun? I also tried running 50 mulitiple instances (in total) of a mix of burnK7, burnMMX, burnBX, burnP6 and burnK6 in linux (with different memory settings for burnBX and burnMMX), while accessing floppy, CD-ROM, 2xHDDs, my SBLive card and my Geforce 2 to try and load my voltages... but again completely stable, if a bit... well very... jerky!.
> I think the burnK7 program does not test enough K7 specific instruction sets to find the problem.

burnK7 doesnt test enough ram to expose the problem. burnK7 stays entirely
inside cache and doesnt touch main memory, which is where the problem
everyone is having is.

-Dan

-- 
[-] Omae no subete no kichi wa ore no mono da. [-]

Re: Athlon doesn't like Athlon optimisation?

[Eric Olson]

>Could you post Roberts email address? The one at the web page
>(redelm@ev1.net) doesn't seem to resolve. I would like to ask him
the
>source of the optimised burnK7
>
>thanks,
>
>-Kostas

I've been using the email address on the net, I think the link was 
down for most of the night until minutes ago.

Robert has a K7 version of burnMMX which is tentatively being called 
burnMMX2.  I think he would prefer to be directly in touch with people 
testing it.  Thus, it is not posted on the website but available only 
through email.

--Eric

Re: Athlon doesn't like Athlon optimisation?

[Jim Blomo]

Hi, I am having a similar problem with my new board/chip. I am using
2.4.10-pre4, and when I compile with the Athlon/Thunderbird setting, the
kernel does absolutely nothing after being uncompressed by LILO. The
computer locks up and I must do a hard reboot to get going again. I have had
no errors (for about 1.5 days) when using the same kernel compiled as 386
and Pentium pro without any other changes. When I used make bzdisk and tried
to boot from that, it repeated the following message over and over until I
did a soft reset:

1007
AX:020C
BX:0000
CX:0007
DX:0000

When booting with the same disk with Athlon not selected as the chip, I had
no errors. Windows 200 has been running with no visible errors for several
days on the same hardware. Memory checks have turned up no errors with my
hardware. Here's some more specific info:

Motherboard + chipset:
 ECS K7AMA + Ali Magic 1645 & Magic 1535D
(http://www.ecs.com.tw/products/k7ama.htm)
FSB Speed:
 266
CPU + multiplier:
 Athlon Thunderbird 1.33Ghz not overclocked
cat /proc/cpuinfo:
 processor	: 0
 vendor_id	: AuthenticAMD
 cpu family	: 6
 model		: 4
 model name	: AMD Athlon(tm) Processor
 stepping	: 4
 cpu MHz		: 1333.382
 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 sep mtrr pge mca cmov pat pse36
mmx fxsr syscall mmxext 3dnowext 3dnow
 bogomips	: 2660.76
RAM type + speed:
 2 stick of 256MB DDR PC2100 (DDR CAS Select: 2.5/3)
Wattage:
 300 watt power supply

I've seen postings on this subject on the web archives, so is it general
opinion that this is a known issue with Athlons? Should I try to return the
chip? I would be happy with a work around, so what other CPU option gives
the best performance?

I hope I have provided enough information (this is my first post to this
list). Please CC me any replies as I am not subscribed to the list. Thanks
for any information!

Jim

Re: Athlon doesn't like Athlon optimisation?

[Dan Hollis]

On Fri, 7 Sep 2001, Jim Blomo wrote:
> Motherboard + chipset:
>  ECS K7AMA + Ali Magic 1645 & Magic 1535D

There goes the VIA-only theory out the window

-Dan

-- 
[-] Omae no subete no kichi wa ore no mono da. [-]

Re: Athlon doesn't like Athlon optimisation?

[Andrzej Krzysztofowicz]

"Jim Blomo wrote:"
> Hi, I am having a similar problem with my new board/chip. I am using
> 2.4.10-pre4, and when I compile with the Athlon/Thunderbird setting, the
> kernel does absolutely nothing after being uncompressed by LILO. The
> computer locks up and I must do a hard reboot to get going again. I have had
> no errors (for about 1.5 days) when using the same kernel compiled as 386
> and Pentium pro without any other changes. When I used make bzdisk and tried
> to boot from that, it repeated the following message over and over until I
> did a soft reset:
> 
> 1007
> AX:020C
> BX:0000
> CX:0007
> DX:0000

This has nothing to do with Athlon optimization. This is probably a broken
floppy disk.

This seem to be from the bootsector code from arch/i386/bootsect.S, it is
exactly the same for all x86 processors and called before any other code.

-- 
=======================================================================
  Andrzej M. Krzysztofowicz               ankry@mif.pg.gda.pl
  phone (48)(58) 347 14 61
Faculty of Applied Phys. & Math.,   Technical University of Gdansk

Re: Athlon doesn't like Athlon optimisation?

[Mike Fedyk]

On Fri, Sep 07, 2001 at 10:39:18AM +0200, Andrzej Krzysztofowicz wrote:
> "Jim Blomo wrote:"
> > Hi, I am having a similar problem with my new board/chip. I am using
> > 2.4.10-pre4, and when I compile with the Athlon/Thunderbird setting, the
> > kernel does absolutely nothing after being uncompressed by LILO. The
> > computer locks up and I must do a hard reboot to get going again. I have had
> > no errors (for about 1.5 days) when using the same kernel compiled as 386
> > and Pentium pro without any other changes. When I used make bzdisk and tried
> > to boot from that, it repeated the following message over and over until I
> > did a soft reset:
> > 
> > 1007
> > AX:020C
> > BX:0000
> > CX:0007
> > DX:0000
> 
> This has nothing to do with Athlon optimization. This is probably a broken
> floppy disk.
> 
> This seem to be from the bootsector code from arch/i386/bootsect.S, it is
> exactly the same for all x86 processors and called before any other code.

That still doesn't discount the report because booting was also tried from
the hard disk.

Re: Athlon doesn't like Athlon optimisation?

[Jim Blomo]

On Fri, 7 Sep 2001, Jim Blomo wrote: 
> Hi, I am having a similar problem with my new board/chip...
> Motherboard + chipset: 
>  ECS K7AMA + Ali Magic 1645 & Magic 1535D 
> CPU + multiplier: 
>  Athlon Thunderbird 1.33Ghz not overclocked 

Since a probable cause was found for the VIA chipset, is there any way for me 
to check if my board is having a similar problem?  I have tried using an 
older BIOS version, upgrading to 2.4.10-pre14, and even applying the VIA 
patch, with no effect on my problem.  When the kernel is compiled with the 
Athlon settings, absolutely nothing happens after LILO uncompresses the 
image. The keyboard and powerswitch also stop responding and I am forced to 
do a hard reboot.  k6, ppro, and i386 settings let Linux run without errors 
for days.  Thanks for CCing me the info as I am not subscribed to this list,

Jim