Re: SMP/HIGHMEM/7450 status ?

Re: SMP/HIGHMEM/7450 status ?

[benh]

> I was wondering if anyone has had any luck with the 7450 with SMP and
>HIGHMEM recently. There were problems with it in the past, and I was
>wondering if any fixes have been included in the current
>source.mvista.com::linuxppc_2_4 rsync tree. The reason why I ask is
>because the only dual 7450 machine I have is currently in production and I
>can't do any testing on it, and I need to turn on the second CPU some
>time soon.
>
> Any info would be much appreciated.

Well... It sorta works until you put really high load on it, then
usually, a CPU locks up in an unrecoverable way. We haven't yet been
able to find a workaround.

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[benh]

>>Well... It sorta works until you put really high load on it, then
>>usually, a CPU locks up in an unrecoverable way. We haven't yet been
>>able to find a workaround.
>
> Ouch. Thanks for the speedy response Ben. This doesn't look so good for
>Terra Soft if they want to be an OEM for the Xserve.

Xserve uses 7455 (like newer dual G4s), which doesn't seem to exhibit the
problem.

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

SMP/HIGHMEM/7450 status ?

[Christopher Murtagh]

Greetings,

 I was wondering if anyone has had any luck with the 7450 with SMP and
HIGHMEM recently. There were problems with it in the past, and I was
wondering if any fixes have been included in the current
source.mvista.com::linuxppc_2_4 rsync tree. The reason why I ask is
because the only dual 7450 machine I have is currently in production and I
can't do any testing on it, and I need to turn on the second CPU some
time soon.

 Any info would be much appreciated.

Cheers,

Chris



--

Christopher Murtagh
Webmaster / Sysadmin
Web Communications Group
McGill University
Montreal, Quebec
Canada

Tel.: (514) 398-3122
Fax:  (514) 398-2017


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Kevin B. Hendricks]

Hi,

I have been using SMP on my box for a long time.  I have not enabled
HIGHMEM though for the same fear you have.

Perhaps Ben Herrenschmidt can comment on the state of his trees as well.

Thanks,

Kevin

On June 6, 2002 11:51, Christopher Murtagh wrote:
> Greetings,
>
>  I was wondering if anyone has had any luck with the 7450 with SMP and
> HIGHMEM recently. There were problems with it in the past, and I was
> wondering if any fixes have been included in the current
> source.mvista.com::linuxppc_2_4 rsync tree. The reason why I ask is
> because the only dual 7450 machine I have is currently in production and
> I can't do any testing on it, and I need to turn on the second CPU some
> time soon.
>
>  Any info would be much appreciated.
>
> Cheers,
>
> Chris


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Christopher Murtagh]

On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
>Well... It sorta works until you put really high load on it, then
>usually, a CPU locks up in an unrecoverable way. We haven't yet been
>able to find a workaround.

 Ouch. Thanks for the speedy response Ben. This doesn't look so good for
Terra Soft if they want to be an OEM for the Xserve.

Cheers,

Chris

--

Christopher Murtagh
Webmaster / Sysadmin
Web Communications Group
McGill University
Montreal, Quebec
Canada

Tel.: (514) 398-3122
Fax:  (514) 398-2017


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Christopher Murtagh]

On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
>Xserve uses 7455 (like newer dual G4s), which doesn't seem to exhibit
>the problem.

 Damn... does this mean that the 7450 machines are lemons (wouldn't
surprise me if the answer was a resounding 'yes')?

 BTW, Ben, I'm not sure if you are aware of this (and I hope you don't
mind)...

 All of our production servers have been named after people that we admire
and/or we feel have made an important contribution in the computing world.
For example, our authentication machine (signin.mcgill.ca) is
turing.wcg.mcgill.ca from Alan Turing. Some of these include:

 lovelace.wcg.mcgill.ca
 babbage.wcg.mcgill.ca
 wall.wcg.mcgill.ca
 torvalds.wcg.mcgill.ca
 ritchie.wcg.mcgill.ca

 and recently, our 4 node briQ cluster:

 kernighan.wcg.mcgill.ca
 gosling.wcg.mcgill.ca
 wozniak.wcg.mcgil.ca
 herrenschmidt.wcg.mcgill.ca

 Currently, lovelace and babbage (which are web servers with a number of
v-hosts) have colophons if you go to their real names
(http://lovelace.wcg.mcgill.ca). We want to do something similar with all
of our machines as well some time this summer.

Cheers,

Chris


--

Christopher Murtagh
Webmaster / Sysadmin
Web Communications Group
McGill University
Montreal, Quebec
Canada

Tel.: (514) 398-3122
Fax:  (514) 398-2017


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[benh]

>
>On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
>>Xserve uses 7455 (like newer dual G4s), which doesn't seem to exhibit
>>the problem.
>
> Damn... does this mean that the 7450 machines are lemons (wouldn't
>surprise me if the answer was a resounding 'yes')?

I can't say for sure. It seems we are hitting a CPU bug, but after
studying the errata and trying all sort of ways to "catch" the dead
CPU when the lockup happen, I had to give up.
It seems even tweaking the CPU reset line won't get it back from the
deadlock situation (I'm tweaking it using the firewire controller
tapping Apple's KeyLargo ASIC GPIO registers, this works fine when
the CPU isn't locked up in this state).

> herrenschmidt.wcg.mcgill.ca

hehe, nice ;) though I wouldn't like having to type that name every
day to telnet to the box ;)

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Christopher Murtagh]

On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
>> herrenschmidt.wcg.mcgill.ca
>
>hehe, nice ;) though I wouldn't like having to type that name every
>day to telnet to the box ;)

 Lets just say, I've learned to type your last name really well. :-)

Cheers,

Chris


--

Christopher Murtagh
Webmaster / Sysadmin
Web Communications Group
McGill University
Montreal, Quebec
Canada

Tel.: (514) 398-3122
Fax:  (514) 398-2017


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Benjamin Herrenschmidt]

>Apple did hold off on dual 7450 until rev 2.1 came out and fixed some
>of the nastier SMP errata, so presumably they at least thought there
>were no SMP showstoppers left in rev 2.1.  As far as I can tell they
>are pretty stable under OS X.

There is an SMP showstopper with HW hashtable walk, but Apple
claims to have a HW workaround.

>Linux is probably just missing some deep magic needed for working
>around one or more of the errata that do still apply to rev 2.1.  The
>trouble is figuring out exactly what's wrong... it can be VERY
>difficult to understand all the ramifications of a CPU bug
>(particularly a SMP related bug) until you use a bus analyzer to
>snapshot bus traffic during a crash, or probe CPU registers via JTAG
>after it has crashed, etc.

Especially since you don't have the connector on the motherboard
to put such a tool, and since I don't even have physical access
to any of these machines.

Ben.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Timothy A. Seufert]

At 1:09 PM -0400 6/6/02, Christopher Murtagh wrote:
>On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
>>Xserve uses 7455 (like newer dual G4s), which doesn't seem to exhibit
>>the problem.
>
>  Damn... does this mean that the 7450 machines are lemons (wouldn't
>surprise me if the answer was a resounding 'yes')?

Apple did hold off on dual 7450 until rev 2.1 came out and fixed some
of the nastier SMP errata, so presumably they at least thought there
were no SMP showstoppers left in rev 2.1.  As far as I can tell they
are pretty stable under OS X.

Linux is probably just missing some deep magic needed for working
around one or more of the errata that do still apply to rev 2.1.  The
trouble is figuring out exactly what's wrong... it can be VERY
difficult to understand all the ramifications of a CPU bug
(particularly a SMP related bug) until you use a bus analyzer to
snapshot bus traffic during a crash, or probe CPU registers via JTAG
after it has crashed, etc.
--
Tim Seufert

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Geert Uytterhoeven]

On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
> > herrenschmidt.wcg.mcgill.ca
>
> hehe, nice ;) though I wouldn't like having to type that name every
> day to telnet to the box ;)

Perhaps they can add an alias benh.wcg.mcgill.ca?

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Christopher Murtagh]

On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
>Xserve uses 7455 (like newer dual G4s), which doesn't seem to exhibit
>the problem.

 Does this mean that the Dual 1Ghz machines (Quicksilver) are running fine
with HIGHMEM and SMP? Replacing the Dual 800 with one might by my only
option, I plan on adding some services to our central machine and I'm
worried about CPU resources. Does anyone here have one of these running?

Cheers,

Chris

--

Christopher Murtagh
Webmaster / Sysadmin
Web Communications Group
McGill University
Montreal, Quebec
Canada

Tel.: (514) 398-3122
Fax:  (514) 398-2017


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: SMP/HIGHMEM/7450 status ?

[Dan Burcaw]

Chris,

I have a dual 1GHz as my primary build box w/ HIGHMEM
(1.5GB installed).. and I have had no stability problems w/
the latest 2.4.19...

-Dan

On Sat, 2002-06-08 at 17:10, Christopher Murtagh wrote:
>
> On Thu, 6 Jun 2002 benh@kernel.crashing.org wrote:
> >Xserve uses 7455 (like newer dual G4s), which doesn't seem to exhibit
> >the problem.
>
>  Does this mean that the Dual 1Ghz machines (Quicksilver) are running fine
> with HIGHMEM and SMP? Replacing the Dual 800 with one might by my only
> option, I plan on adding some services to our central machine and I'm
> worried about CPU resources. Does anyone here have one of these running?
>
> Cheers,
>
> Chris
>
> --
>
> Christopher Murtagh
> Webmaster / Sysadmin
> Web Communications Group
> McGill University
> Montreal, Quebec
> Canada
>
> Tel.: (514) 398-3122
> Fax:  (514) 398-2017
>
>
>


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/