1.6.1 lack of stability on cats

Hi,

I've just been running 1.6.1 RC1 on my cats box and am concerned about
the stability, I've had a coupld of core dumps, and even had it
disconnect me part way through an xsrc build.  This didn't happen with
1.6, I'm wondering what major changes have occured on the arm port to
cause this.  Anyone any thoughts on this?

from the xsrc build I got 3 .core dumps:
./programs/sh.core
./lib/Xaw6/cpp0.core
./exports/include/X11/Xaw/sh.core

which is far too many for me to think the system is stable, none of the
core dumps can be back traced either...

I know there was talk it might be the optimisation flags to the kernel,
cats kernels are compiled using:
COPTS="-O2 -march=armv4 -mtune=strongarm"

Is this a likely cause, and if so, should we be considering disabling
the optimisations for 1.6.1?

Does anyone else running an arm box with 1.6.1 have stability issues?

Chris

Re: 1.6.1 lack of stability on cats

Hi,

On Mon, Feb 03, 2003 at 10:39:21AM +0000, Chris Gilbert wrote:

> COPTS="-O2 -march=armv4 -mtune=strongarm"
> 
> Is this a likely cause, and if so, should we be considering disabling
> the optimisations for 1.6.1?
> 
> Does anyone else running an arm box with 1.6.1 have stability issues?

I have a PR open about this. It's not clear whether the compiler is the
culprit or just an innocent messenger, but I made my Shark much more stable
by running an kernel without the -march= and mtune= parts.

Regards,
	Ignatios

Re: 1.6.1 lack of stability on cats

On Mon, 3 Feb 2003, Chris Gilbert wrote:

> I've just been running 1.6.1 RC1 on my cats box and am concerned about
> the stability, I've had a coupld of core dumps, and even had it
> disconnect me part way through an xsrc build.  This didn't happen with
> 1.6, I'm wondering what major changes have occured on the arm port to
> cause this.  Anyone any thoughts on this?

My immediate thought (and something that's been sitting at the back of my
mind and completely failing to get any useful attention for months) is
that we might have some but not all of the copyin/out updates in the 1.6
branch, and might hence have a copyin/out that handles misaligned buffers
wrongly.

Unfortunately, I haven't unpacked any of my NetBSD/arm boxes since I moved
house in November (well, apart from the NC, but it's not really a
development box), and I've been busy making PuTTY go on Mac OS, so I've
been kind of neglecting NetBSD.

--

-- 
Ben Harris                                                   <bjh21 <at> netbsd.org>
Portmaster, NetBSD/acorn26           <URL:http://www.netbsd.org/Ports/acorn26/>

Re: 1.6.1 lack of stability on cats

> On Mon, 3 Feb 2003, Chris Gilbert wrote:
> 
> > I've just been running 1.6.1 RC1 on my cats box and am concerned about
> > the stability, I've had a coupld of core dumps, and even had it
> > disconnect me part way through an xsrc build.  This didn't happen with
> > 1.6, I'm wondering what major changes have occured on the arm port to
> > cause this.  Anyone any thoughts on this?
> 
> My immediate thought (and something that's been sitting at the back of my
> mind and completely failing to get any useful attention for months) is
> that we might have some but not all of the copyin/out updates in the 1.6
> branch, and might hence have a copyin/out that handles misaligned buffers
> wrongly.
> 
> Unfortunately, I haven't unpacked any of my NetBSD/arm boxes since I moved
> house in November (well, apart from the NC, but it's not really a
> development box), and I've been busy making PuTTY go on Mac OS, so I've
> been kind of neglecting NetBSD.

Sounds very likely,  bcopyinout.S on the branch is effectively 1.1 from 
the trunk (though CVS thinks it's parented off 1.9 for some strange 
reason).

R.

Re: 1.6.1 lack of stability on cats

Richard Earnshaw said:
> bjh21 said:
>> My immediate thought (and something that's been sitting at the back of
>> my mind and completely failing to get any useful attention for months)
>> is that we might have some but not all of the copyin/out updates in
>> the 1.6 branch, and might hence have a copyin/out that handles
>> misaligned buffers wrongly.
>>
> Sounds very likely,  bcopyinout.S on the branch is effectively 1.1 from
> the trunk (though CVS thinks it's parented off 1.9 for some strange
> reason).

hmm, possibly, I'll see when I get home if the xsrc build has got any
further with a non arch/tune kernel.  Certainly looking at the diffs it's
rather worrying.

How standalone is bcopyinout.S?  IE could it be updated on it's own
(noting that it's now full of mulitprocessor stuff.

Chris

Re: 1.6.1 lack of stability on cats

On Mon, 3 Feb 2003 11:46:50 +0100
Ignatios Souvatzis <ignatios <at> theory.cs.uni-bonn.de> wrote:

> Hi,
> 
> On Mon, Feb 03, 2003 at 10:39:21AM +0000, Chris Gilbert wrote:
> 
> > COPTS="-O2 -march=armv4 -mtune=strongarm"
> > 
> > Is this a likely cause, and if so, should we be considering
> > disabling the optimisations for 1.6.1?
> > 
> > Does anyone else running an arm box with 1.6.1 have stability
> > issues?
> 
> I have a PR open about this. It's not clear whether the compiler is
> the culprit or just an innocent messenger, but I made my Shark much
> more stable by running an kernel without the -march= and mtune= parts.

In the random sample of does xsrc build now, the answer is yes it did
without the arch and tune flags, so for as to why I've no idea *sigh*

Anyway at least I got x server for 1.6.1 cats built now 8)

Cheers,
Chris

Re: bug fix for cats core dumping

On Sun, 17 Nov 2002 23:57:08 +0000
Richard Earnshaw <rearnsha <at> buzzard.freeserve.co.uk> wrote:

> > Hi,
> > 
> > Not sure if other arm archs are suffering issues dumping kernel
> > cores, but attached it a patch that fixed it for my cats box (well
> > it doesn't hang any more when doing a dump (or it didn't for the one
> > go I tried)
> > 
> > however I'm wondering if someone can check that it's sane, before I
> > submit it.
> 
> What about cacheing?  can we be sure that the cache has been cleaned 
> before we potentially doubly map each page?  Otherwise we won't really
> be dumping what the processor sees.

Meant to sort this out sooner, but time etc...

How about the attached?  I believe it does the right thing now, re
flushing cache to memory.

Cheers,
Chris

Index: stubs.c
===================================================================
RCS file: /cvsroot/src/sys/arch/arm/arm32/stubs.c,v
retrieving revision 1.11

Re: 1.6.1 lack of stability on cats

> On Mon, 3 Feb 2003 11:46:50 +0100
> Ignatios Souvatzis <ignatios <at> theory.cs.uni-bonn.de> wrote:
> 
> > Hi,
> > 
> > On Mon, Feb 03, 2003 at 10:39:21AM +0000, Chris Gilbert wrote:
> > 
> > > COPTS="-O2 -march=armv4 -mtune=strongarm"
> > > 
> > > Is this a likely cause, and if so, should we be considering
> > > disabling the optimisations for 1.6.1?
> > > 
> > > Does anyone else running an arm box with 1.6.1 have stability
> > > issues?
> > 
> > I have a PR open about this. It's not clear whether the compiler is
> > the culprit or just an innocent messenger, but I made my Shark much
> > more stable by running an kernel without the -march= and mtune= parts.
> 
> In the random sample of does xsrc build now, the answer is yes it did
> without the arch and tune flags, so for as to why I've no idea *sigh*
> 
> Anyway at least I got x server for 1.6.1 cats built now 8)

I've just built a 1.6.1_RC1 kernel with strongarm optimizations enabled, 
but with bcopyinout.S pulled up to revision 1.6 and the kernel seems fine 
(sufficient to completely bootstrap and regtest gcc without any untoward 
problems).

R.

Re: 1.6.1 lack of stability on cats

On Sun, 09 Feb 2003 18:33:58 +0000
Richard Earnshaw <rearnsha <at> buzzard.freeserve.co.uk> wrote:

> > On Mon, 3 Feb 2003 11:46:50 +0100
> > Ignatios Souvatzis <ignatios <at> theory.cs.uni-bonn.de> wrote:
> > 
> > > Hi,
> > > 
> > > On Mon, Feb 03, 2003 at 10:39:21AM +0000, Chris Gilbert wrote:
> > > 
> > > > COPTS="-O2 -march=armv4 -mtune=strongarm"
> > > > 
> > > > Is this a likely cause, and if so, should we be considering
> > > > disabling the optimisations for 1.6.1?
> > > > 
> > > > Does anyone else running an arm box with 1.6.1 have stability
> > > > issues?
> > > 
> > > I have a PR open about this. It's not clear whether the compiler
> > > is the culprit or just an innocent messenger, but I made my Shark
> > > much more stable by running an kernel without the -march= and
> > > mtune= parts.
> > 
> > In the random sample of does xsrc build now, the answer is yes it
> > did without the arch and tune flags, so for as to why I've no idea
> > *sigh*
> > 
> > Anyway at least I got x server for 1.6.1 cats built now 8)
> 
> I've just built a 1.6.1_RC1 kernel with strongarm optimizations

Re: 1.6.1 lack of stability on cats

Hi,

> On Sun, 09 Feb 2003 18:33:58 +0000
> Richard Earnshaw <rearnsha <at> buzzard.freeserve.co.uk> wrote:
> > I've just built a 1.6.1_RC1 kernel with strongarm optimizations
> > enabled, but with bcopyinout.S pulled up to revision 1.6 and the
> > kernel seems fine (sufficient to completely bootstrap and regtest gcc
> > without any untoward problems).
> 

but on Sun, Feb 09, 2003 at 06:43:25PM +0000, Chris Gilbert wrote:
> Hmmm, just wondering why, certainly it sounds like we should get it
> pulled up ASAP.  I'll put a pullup request in.  

... which requests the latest version of that file, which is 1.9.
However, Richard tested version 1.6 in the netbsd-1-6 context. 

Shouldn't the pullup request be for 1.6?

Regards,
	Ignatios
--

-- 
seal your e-mail: http://www.gnupg.org/