[parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[joel.soete]

Hi all,

Thoses problems are not linked but similar by nature:

with evms-1.0.0, I still got thousand of messages:
Apr  3 19:05:48 palinux kernel: evms_lvdisplay(900): unaligned access to
0x0002461e at ip=0x40208b53

and with jfs-1.0.17:
Apr  3 19:11:24 palinux kernel: fsck.jfs(1483): unaligned access to 0xbff00717
at ip=0x0002eb3f

It seems harmless but not sure.

I query the ml db and found similar problems (and its solution :_) )(reiserfs:
http://lists.parisc-linux.org/hypermail/parisc-linux/5862.html), but I do not
very well understand what I have to find the sources (_u32* ?).

Thanks in advance for additional help,
    Joel

-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[Randolph Chung]

> It seems harmless but not sure.

well.... in general this is not a problem, however....

Fleedwood found a bug in the unaligned handler last week. it doesn't
handle the pre/post increment cases of load and store, and there may be
some cases where these messages will indicate the cause of errors.

I'll look into fixing this, but it might take some time.

> I query the ml db and found similar problems (and its solution :_) )(reiserfs:
> http://lists.parisc-linux.org/hypermail/parisc-linux/5862.html), but I do not
> very well understand what I have to find the sources (_u32* ?).

it depends, you need to make sure all the memory accesses happen on
natural boundaries...  (e.g. int on 4 byte boundaries, short on 2 byte
boundaries, etc). From the ip reported by the unaligned message you
should be able to trace back to the asm insn and function (if your
objects are not stripped) that is causing the unaligned access.

randolph
-- 
   @..@                                         http://www.TauSq.org/
  (----)
 ( >__< )
 ^^ ~~ ^^

Re: [parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[joel.soete]

Quoting Randolph Chung <randolph@tausq.org>:

> > It seems harmless but not sure.
> 
> well.... in general this is not a problem, however....
> 
> Fleedwood found a bug in the unaligned handler last week. it doesn't
> handle the pre/post increment cases of load and store, and there may be
> some cases where these messages will indicate the cause of errors.
> 
> I'll look into fixing this, but it might take some time.
> 

Well I will be patient as till now (at least a month) evms seems to works
correctly :-)

> > I query the ml db and found similar problems (and its solution :_)
> )(reiserfs:
> > http://lists.parisc-linux.org/hypermail/parisc-linux/5862.html), but I
> do not
> > very well understand what I have to find the sources (_u32* ?).
> 
> it depends, you need to make sure all the memory accesses happen on
> natural boundaries...  (e.g. int on 4 byte boundaries, short on 2 byte
> boundaries, etc). From the ip reported by the unaligned message you
> should be able to trace back to the asm insn and function (if your
> objects are not stripped) that is causing the unaligned access.

Humm... in evms (not yet investigate jfs) I notice that 4-byte variables are
pass to function parameters actualy 2-byte without casting. Could this be a reason?

> 
> randolph
> -- 
>    @..@                                         http://www.TauSq.org/
>   (----)
>  ( >__< )
>  ^^ ~~ ^^
> 
> 

Thanks for explanation,
    Joel

-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[John David Anglin]

> Quoting Randolph Chung <randolph@tausq.org>:
> 
> > > It seems harmless but not sure.
> > 
> > well.... in general this is not a problem, however....
> > 
> > Fleedwood found a bug in the unaligned handler last week. it doesn't
> > handle the pre/post increment cases of load and store, and there may be
> > some cases where these messages will indicate the cause of errors.
> > 
> > I'll look into fixing this, but it might take some time.
> > 
> 
> Well I will be patient as till now (at least a month) evms seems to works
> correctly :-)

We used to have a huge number of page faults filling log files on gsyprf11.
This was mysteriously fixed a few weeks ago.  Anybody know what fixed the
problem?

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: [parisc-linux] GCC support for dwarf2

[joel.soete]

Hi Dave,

What could we do to help? Do we have to install snapshot gcc? Which programs
would be tested?

Joel

-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] GCC support for dwarf2

[John David Anglin]

Joel,

Thanks for the offer to help.  The gcc 3.1 release criteria are here:
<http://gcc.gnu.org/gcc-3.1/criteria.html>.  There should be info on
how to build these.  There has been discussion of this on the gcc
list.

It would be useful to collect testresults for the listed applications.
It would be really great to compare the status of debian 3.0.4, gcc
cvs 3.1 (prerelease), and gcc cvs 3.2 (experimental).

I suspect that there will be problems with many of the tests.  There
is a integer*1 g77 bug that may affect lapack.  Qt may not build
because of an issue with long calls.  I think that I now understand
how to fix this but it may just go into 3.2.  3.2 has working dwarf2
exceptions and 3.1 doesn't.

> What could we do to help? Do we have to install snapshot gcc? Which programs
> would be tested?

You would need to install the cvs binutils and cvs gcc versions in some
convenient location for the testing.  I use ~/opt/gnu.  Instructions
on cvs access to gcc is on the gcc site <http://gcc.gnu.org/> and
similarly for binutils.  You probably need to define 

LD_RUN_PATH=/home/dave/opt/gnu/lib
LD_LIBRARY_PATH=/home/dave/opt/gnu/lib
LIBRARY_PATH=/home/dave/opt/gnu/lib

in your environment.

Even just building gcc can provide useful feedback.  I know it is very
easy to make mistakes due to the complexity of the package.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: [parisc-linux] GCC support for dwarf2

[joel.soete]

Hi Dave,

Well I finished to co gcc-3.2 and gcc3.1 cvs (that is long with small 33800 baud
modem but that all I can get for cvs)

And I am already back with a problem:

Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:

> You would need to install the cvs binutils and cvs gcc versions in some
> convenient location for the testing. 

Well I just figure out that there is no configure file in cvs of binutils.
So, in the sources directory (/GNU-Dev/parisc-linux/src/binutils) I try autoconf
but it just creates an empty configure?
What do I wrong?


Thanks in advance for additional help,
    Joel


PS: Humm I do not find any glibc CVS so I will test toolchain with glibc 2.2.5.


-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] GCC support for dwarf2

[John David Anglin]

> Well I finished to co gcc-3.2 and gcc3.1 cvs (that is long with small 33800 baud
> modem but that all I can get for cvs)
> 
> And I am already back with a problem:
> 
> Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:
> 
> > You would need to install the cvs binutils and cvs gcc versions in some
> > convenient location for the testing. 
> 
> Well I just figure out that there is no configure file in cvs of binutils.

Remove "configure" and do another checkout.  Use checkout rather than
update since sometimes update has problems in the toplevel directory.

> So, in the sources directory (/GNU-Dev/parisc-linux/src/binutils) I try autoconf
> but it just creates an empty configure?
> What do I wrong?

There are two types of "configure" files.  The toplevel configure script
in binutils (and also gcc) are currently manually maintained.  The
binutils "configure" file isn't autoconf generated.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: [parisc-linux] GCC support for dwarf2

[joel.soete]

Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:

> > Well I finished to co gcc-3.2 and gcc3.1 cvs (that is long with small
> 33800 baud
> > modem but that all I can get for cvs)
> > 
> > And I am already back with a problem:
> > 
> > Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:
> > 
> > > You would need to install the cvs binutils and cvs gcc versions in
> some
> > > convenient location for the testing. 
> > 
> > Well I just figure out that there is no configure file in cvs of
> binutils.
> 
> Remove "configure" and do another checkout.  Use checkout rather than
> update since sometimes update has problems in the toplevel directory.

I just do another co and effectively configure is there now?
(I also notice the problem of update :_( )

> 
> > So, in the sources directory (/GNU-Dev/parisc-linux/src/binutils) I
> try autoconf
> > but it just creates an empty configure?
> > What do I wrong?
> 
> There are two types of "configure" files.  The toplevel configure
> script
> in binutils (and also gcc) are currently manually maintained.  The
> binutils "configure" file isn't autoconf generated.

Ok.

Dave thanks a lot for info,
    Joel

-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] GCC support for dwarf2

[joel.soete]

Hi Dave,

Here am I with something confusing me about glibc building.

In the Carlos O'Donell recipe
(http://www.parisc-linux.org/toolchain/PA-Linux-XC-HOWTO-04.html)

It is set "export PATH=$PATH:$DEST/bin"
So at configure step it was selected gcc-3.0.4 (the last release for hppa-linux box)

Naturally if I put "export PATH=$DEST/bin:$PATH", the configure of glibc select
(for me well) gcc-3.2.

What is the right way? (for me the second)

Thanks for more help,
    Joel

Quoting joel.soete@freebel.net:

> Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:
> 
> > > Well I finished to co gcc-3.2 and gcc3.1 cvs (that is long with
> small
> > 33800 baud
> > > modem but that all I can get for cvs)
> > > 
> > > And I am already back with a problem:
> > > 
> > > Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:
> > > 
> > > > You would need to install the cvs binutils and cvs gcc versions
> in
> > some
> > > > convenient location for the testing. 
> > > 
> > > Well I just figure out that there is no configure file in cvs of
> > binutils.
> > 
> > Remove "configure" and do another checkout.  Use checkout rather than
> > update since sometimes update has problems in the toplevel directory.
> 
> I just do another co and effectively configure is there now?
> (I also notice the problem of update :_( )
> 
> > 
> > > So, in the sources directory (/GNU-Dev/parisc-linux/src/binutils) I
> > try autoconf
> > > but it just creates an empty configure?
> > > What do I wrong?
> > 
> > There are two types of "configure" files.  The toplevel configure
> > script
> > in binutils (and also gcc) are currently manually maintained.  The
> > binutils "configure" file isn't autoconf generated.
> 
> Ok.
> 
> Dave thanks a lot for info,
>     Joel
> 
> -------------------------------------------------
> This mail sent through Tiscali Webmail (http://webmail.tiscali.be)
> 
> 


-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] GCC support for dwarf2

[John David Anglin]

> Here am I with something confusing me about glibc building.
> 
> In the Carlos O'Donell recipe
> (http://www.parisc-linux.org/toolchain/PA-Linux-XC-HOWTO-04.html)
> 
> It is set "export PATH=$PATH:$DEST/bin"
> So at configure step it was selected gcc-3.0.4 (the last release for hppa-linux box)
> 
> Naturally if I put "export PATH=$DEST/bin:$PATH", the configure of glibc select
> (for me well) gcc-3.2.
> 
> What is the right way? (for me the second)

Here is the script that I use to build glibc.  This installs to the same
prefix that I use for gcc and binutils.

#!/bin/sh
# /bin/sh must be bash!
export PATH=/usr/local/bin:/usr/bin:/bin:/usr/bin/X11:/usr/games:/home/dave/opt/
gnu/bin
unset LD_RUN_PATH
unset LD_LIBRARY_PATH
export CC=/home/dave/opt/gnu/bin/gcc
export CFLAGS="-O2 -g"
../glibc/configure --build=hppa-linux --host=hppa-linux --target=hppa-linux --pr
efix=/home/dave/opt/gnu --disable-profile --without-cvs --enable-add-ons --enabl
e-hacker-mode &&
make &&
make install

I tried to ensure that I am using the system tools for the build and install
except for the compiler which I specify with CC.  The compiler was built with
--with-gnu-as and --with-gnu-ld, so it knows where to find its tools
independent of the PATH setting.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)

Re: [parisc-linux] GCC support for dwarf2

[joel.soete]

Dave,

It confirms my first idea. (I also just find back:
http://mkhppa1.esiee.fr/archive/script_toolchain).

Thanks a lot,
    Joel

PS: I will send a mail to Carlos to report him this small bug and some update

Quoting John David Anglin <dave@hiauly1.hia.nrc.ca>:

> > Here am I with something confusing me about glibc building.
> > 
> > In the Carlos O'Donell recipe
> > (http://www.parisc-linux.org/toolchain/PA-Linux-XC-HOWTO-04.html)
> > 
> > It is set "export PATH=$PATH:$DEST/bin"
> > So at configure step it was selected gcc-3.0.4 (the last release for
> hppa-linux box)
> > 
> > Naturally if I put "export PATH=$DEST/bin:$PATH", the configure of
> glibc select
> > (for me well) gcc-3.2.
> > 
> > What is the right way? (for me the second)
> 
> Here is the script that I use to build glibc.  This installs to the
> same
> prefix that I use for gcc and binutils.
> 
> #!/bin/sh
> # /bin/sh must be bash!
> export
> PATH=/usr/local/bin:/usr/bin:/bin:/usr/bin/X11:/usr/games:/home/dave/opt/
> gnu/bin
> unset LD_RUN_PATH
> unset LD_LIBRARY_PATH
> export CC=/home/dave/opt/gnu/bin/gcc
> export CFLAGS="-O2 -g"
> ../glibc/configure --build=hppa-linux --host=hppa-linux
> --target=hppa-linux --pr
> efix=/home/dave/opt/gnu --disable-profile --without-cvs --enable-add-ons
> --enabl
> e-hacker-mode &&
> make &&
> make install
> 
> I tried to ensure that I am using the system tools for the build and
> install
> except for the compiler which I specify with CC.  The compiler was built
> with
> --with-gnu-as and --with-gnu-ld, so it knows where to find its tools
> independent of the PATH setting.
> 
> Dave
> -- 
> J. David Anglin                                  dave.anglin@nrc.ca
> National Research Council of Canada              (613) 990-0752 (FAX:
> 952-6605)
> 
> 


-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] Patch to debina glibc-2.2.5 for gcc-3.2 build (was GCC support for dwarf2)

[joel.soete]

Hi Dave,

Could you send me you patch as an attached file or put it somewhere
(I try to grab page but "patch" failed)

Thanks in advance for additional help,
    Joel


-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] GCC support for dwarf2

[Carlos O'Donell Jr.]

> Hi Dave,
> 
> Here am I with something confusing me about glibc building.
> 
> In the Carlos O'Donell recipe
> (http://www.parisc-linux.org/toolchain/PA-Linux-XC-HOWTO-04.html)
> 
> It is set "export PATH=$PATH:$DEST/bin"
> So at configure step it was selected gcc-3.0.4 (the last release for hppa-linux box)
> 
> Naturally if I put "export PATH=$DEST/bin:$PATH", the configure of glibc select
> (for me well) gcc-3.2.
> 
> What is the right way? (for me the second)
> 
> Thanks for more help,
>     Joel
>

The document describes building a cross-compiler for non-hppa->hppa. 
Where gcc would be prefaced with 'hppa-linux' and might be
the only such executable on the system. Obviously if you have multiple
toolchains on your path the order should be '$DEST/bin:$PATH'.

Happily this document also serves as a 'build the toolchain' document :)

I'll update that bit, make some spelling mistakes and email Grant.

Thank Joel!

Cheers,
Carlos.

Re: [parisc-linux] GCC support for dwarf2

[joel.soete]

Quoting "Carlos O'Donell Jr." <carlos@baldric.uwo.ca>:

> > Hi Dave,
> > 
> > Here am I with something confusing me about glibc building.
> > 
> > In the Carlos O'Donell recipe
> > (http://www.parisc-linux.org/toolchain/PA-Linux-XC-HOWTO-04.html)
> > 
> > It is set "export PATH=$PATH:$DEST/bin"
> > So at configure step it was selected gcc-3.0.4 (the last release for
> hppa-linux box)
> > 
> > Naturally if I put "export PATH=$DEST/bin:$PATH", the configure of
> glibc select
> > (for me well) gcc-3.2.
> > 
> > What is the right way? (for me the second)
> > 
> > Thanks for more help,
> >     Joel
> >
> 
> The document describes building a cross-compiler for non-hppa->hppa. 
> Where gcc would be prefaced with 'hppa-linux' and might be
> the only such executable on the system. Obviously if you have multiple
> toolchains on your path the order should be '$DEST/bin:$PATH'.

Ok it is clear.

> 
> Happily this document also serves as a 'build the toolchain' document
> :)
> 
> I'll update that bit, make some spelling mistakes and email Grant.
> 

I just also notice that the glibc release is now 2.2.5 which is extracted as
glibc-2.2.5.orig (so just after "Now onto glibc:" 
    cd glibc-2.2.4
would be:
    cd glibc-2.2.5.orig)

    
Thanks for your attention and your great job,
    Joel

-------------------------------------------------
This mail sent through Tiscali Webmail (http://webmail.tiscali.be)

Re: [parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[Joel Soete]

Hey Randolph,

Randolph Chung wrote:
>>It seems harmless but not sure.
> 
> 
> well.... in general this is not a problem, however....
> 
> Fleedwood found a bug in the unaligned handler last week. it doesn't
> handle the pre/post increment cases of load and store, and there may be
> some cases where these messages will indicate the cause of errors.
> 
> I'll look into fixing this, but it might take some time.

I would just like to know if you already have the opportunity to have a 
look for a fix of this stuff (I know you are very busy this time and do 
not want to stress, just want to got news)

Thanks in advance,
	Joel

Re: [parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[Randolph Chung]

> I would just like to know if you already have the opportunity to have a 
> look for a fix of this stuff (I know you are very busy this time and do 
> not want to stress, just want to got news)

just came back from a trip....

i started fixing this a few weeks back but got sidetracked by the 2.5
merge.... it's definitely on my todo list :-)

fwiw i haven't seen any conclusive proof that this "fix" will actually
fix any problems we've seen....

randolph

Re: [parisc-linux] "unaligned access to ... at ip=..." problem with evms and jfs

[jsoe0708]

>-- Original Message --
>Date: Tue, 6 Aug 2002 23:11:57 -0700
>From: Randolph Chung <tausq@debian.org>
>To: Joel Soete <joel.soete@freebel.net>
>Cc: parisc-linux@lists.parisc-linux.org
>Subject: Re: [parisc-linux] "unaligned access to ... at ip=..." problem
with
>evms and jfs
>Reply-To: Randolph Chung <tausq@debian.org>
>
>
>> I would just like to know if you already have the opportunity to have
a
>
>> look for a fix of this stuff (I know you are very busy this time and do
>
>> not want to stress, just want to got news)
>
>just came back from a trip....
>
>i started fixing this a few weeks back but got sidetracked by the 2.5
>merge.... it's definitely on my todo list :-)
>
Great :))

>fwiw i haven't seen any conclusive proof that this "fix" will actually
>fix any problems we've seen....
>

In fact i am ignoring exactly what is this fix. So I could not presume if
it will or not solve the annoying "unalign access..." messages I encounter
before I test it :)).

Many thanks for help,
    J.