gate.S and recent toolchains

gate.S and recent toolchains

[Jes Sorensen]

Hi,

I've been chasing a very strange but since Monday and I am starting
to hit a brick wall.

Basically I am unable to build a kernel from Linus' git tree on one
system since Monday morning, whereas it boots fine if I build it on an
older system. I tried installing an older toolchain on this box but it
still didn't seem to make a difference. The system boots, calls init and
I most often get the message "INIT: 2.85 booting", then a hang. In some
instances I don't even get the INIT message.

What sparked it was Russ Anderson's patch which added a few lines to
include/asm-ia64/asmmacros.h, git id
d2a28ad9fa7bf16761d070d8a3338375e1574b32

I've tracked it down to being the two lines doing a .section/.previous
in the below patch chunk causing it. However it doesn't seem to be the
specific .section name that does it, ie. any random .section/.previous
pair I add to arch/ia64/kernel/gate.S has the same effect.

Now I am not a wiz in linker/gas magic, so I was hoping someone else
had an idea what could be causing this? Anyone seen anything similar?

The specific toolchain I am using is gcc-4.1.0 based, but I tried
3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
another system seems to do fine (admittedly it had binutils 2.14.90.x).

Any suggestions would be greatly appreciated.

Thanks,
Jes



diff --git a/include/asm-ia64/asmmacro.h b/include/asm-ia64/asmmacro.h
index 77af457..d4cec32 100644
--- a/include/asm-ia64/asmmacro.h
+++ b/include/asm-ia64/asmmacro.h
@@ -51,6 +51,17 @@ name:
   [99:]        x

 /*
+ * Tag MCA recoverable instruction ranges.
+ */
+
+       .section "__mca_table", "a"             // declare section &
section attributes
+       .previous
+
+# define MCA_RECOVER_RANGE(y)                  \
+       .xdata4 "__mca_table", y-., 99f-.;      \
+  [99:]
+
+/*
  * Mark instructions that need a load of a virtual address patched to be
  * a load of a physical address.  We use this either in critical
performance
  * path (ivt.S - TLB miss processing) or in places where it might not be

Re: gate.S and recent toolchains

[Russ Anderson]

Jes Sorensen wrote:
> 
> Hi,
> 
> I've been chasing a very strange but since Monday and I am starting
> to hit a brick wall.
> 
> Basically I am unable to build a kernel from Linus' git tree on one
> system since Monday morning, whereas it boots fine if I build it on an
> older system. I tried installing an older toolchain on this box but it
> still didn't seem to make a difference. The system boots, calls init and
> I most often get the message "INIT: 2.85 booting", then a hang. In some
> instances I don't even get the INIT message.

It builds and boots on all the machines I have tried,
including the same build and boot machines Jes uses.
Jes can't get it to work, I can't get it to fail.

> What sparked it was Russ Anderson's patch which added a few lines to
> include/asm-ia64/asmmacros.h, git id
> d2a28ad9fa7bf16761d070d8a3338375e1574b32
> 
> I've tracked it down to being the two lines doing a .section/.previous
> in the below patch chunk causing it. However it doesn't seem to be the
> specific .section name that does it, ie. any random .section/.previous
> pair I add to arch/ia64/kernel/gate.S has the same effect.
> 
> Now I am not a wiz in linker/gas magic, so I was hoping someone else
> had an idea what could be causing this? Anyone seen anything similar?
> 
> The specific toolchain I am using is gcc-4.1.0 based, but I tried
> 3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
> another system seems to do fine (admittedly it had binutils 2.14.90.x).
> 
> Any suggestions would be greatly appreciated.
> 
> Thanks,
> Jes
> 
> 
> 
> diff --git a/include/asm-ia64/asmmacro.h b/include/asm-ia64/asmmacro.h
> index 77af457..d4cec32 100644
> --- a/include/asm-ia64/asmmacro.h
> +++ b/include/asm-ia64/asmmacro.h
> @@ -51,6 +51,17 @@ name:
>    [99:]        x
> 
>  /*
> + * Tag MCA recoverable instruction ranges.
> + */
> +
> +       .section "__mca_table", "a"             // declare section &
> section attributes
> +       .previous
> +
> +# define MCA_RECOVER_RANGE(y)                  \
> +       .xdata4 "__mca_table", y-., 99f-.;      \
> +  [99:]
> +
> +/*
>   * Mark instructions that need a load of a virtual address patched to be
>   * a load of a physical address.  We use this either in critical
> performance
>   * path (ivt.S - TLB miss processing) or in places where it might not be
> 


-- 
Russ Anderson, OS RAS/Partitioning Project Lead  
SGI - Silicon Graphics Inc          rja@sgi.com

Re: gate.S and recent toolchains

[Russ Anderson]

Russ Anderson wrote:
> 
> Hi,
> 
> I've been chasing a very strange but since Monday and I am starting
> to hit a brick wall.
> 
> Basically I am unable to build a kernel from Linus' git tree on one
> system since Monday morning, whereas it boots fine if I build it on an
> older system. I tried installing an older toolchain on this box but it
> still didn't seem to make a difference. The system boots, calls init and
> I most often get the message "INIT: 2.85 booting", then a hang. In some

My kernel says "INIT: version 2.86 booting" and works.

> instances I don't even get the INIT message.
> 
> What sparked it was Russ Anderson's patch which added a few lines to
> include/asm-ia64/asmmacros.h, git id
> d2a28ad9fa7bf16761d070d8a3338375e1574b32
> 
> I've tracked it down to being the two lines doing a .section/.previous
> in the below patch chunk causing it. However it doesn't seem to be the
> specific .section name that does it, ie. any random .section/.previous
> pair I add to arch/ia64/kernel/gate.S has the same effect.
> 
> Now I am not a wiz in linker/gas magic, so I was hoping someone else
> had an idea what could be causing this? Anyone seen anything similar?
> 
> The specific toolchain I am using is gcc-4.1.0 based, but I tried
> 3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
> another system seems to do fine (admittedly it had binutils 2.14.90.x).
> 
> Any suggestions would be greatly appreciated.
> 
> Thanks,
> Jes
> 
> 
> 
> diff --git a/include/asm-ia64/asmmacro.h b/include/asm-ia64/asmmacro.h
> index 77af457..d4cec32 100644
> --- a/include/asm-ia64/asmmacro.h
> +++ b/include/asm-ia64/asmmacro.h
> @@ -51,6 +51,17 @@ name:
>    [99:]        x
> 
>  /*
> + * Tag MCA recoverable instruction ranges.
> + */
> +
> +       .section "__mca_table", "a"             // declare section &
> section attributes
> +       .previous
> +
> +# define MCA_RECOVER_RANGE(y)                  \
> +       .xdata4 "__mca_table", y-., 99f-.;      \
> +  [99:]
> +
> +/*
>   * Mark instructions that need a load of a virtual address patched to be
>   * a load of a physical address.  We use this either in critical
> performance
>   * path (ivt.S - TLB miss processing) or in places where it might not be
> 


-- 
Russ Anderson, OS RAS/Partitioning Project Lead  
SGI - Silicon Graphics Inc          rja@sgi.com

Re: gate.S and recent toolchains

[H. J. Lu]

On Wed, Mar 29, 2006 at 06:07:58PM +0200, Jes Sorensen wrote:
> The specific toolchain I am using is gcc-4.1.0 based, but I tried
> 3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
> another system seems to do fine (admittedly it had binutils 2.14.90.x).
> 

Can you try the current Linux binutils 2.16.91.0.7 at

http://www.kernel.org/pub/linux/devel/binutils/


H.J.

RE: gate.S and recent toolchains

[Chen, Kenneth W]

Russ Anderson wrote on Wednesday, March 29, 2006 8:51 AM
> > Basically I am unable to build a kernel from Linus' git tree on one
> > system since Monday morning, whereas it boots fine if I build it on an
> > older system. I tried installing an older toolchain on this box but it
> > still didn't seem to make a difference. The system boots, calls init and
> > I most often get the message "INIT: 2.85 booting", then a hang. In some
> > instances I don't even get the INIT message.
> 
> It builds and boots on all the machines I have tried,
> including the same build and boot machines Jes uses.
> Jes can't get it to work, I can't get it to fail.

Russ, do you reference __mca_table at run time?  It looks like so.  It's
problematic that __mca_table is in __init section.  Though I doubt that
was the reason Jes ran into.

- Ken

Re: gate.S and recent toolchains

[Russ Anderson]

Chen, Kenneth W wrote:
> 
> Russ, do you reference __mca_table at run time?  It looks like so.  It's
> problematic that __mca_table is in __init section.  Though I doubt that
> was the reason Jes ran into.

It was right next to __ex_table in the original patch.
Yes, it is referenced at run time and should not be in 
the __init section.

-- 
Russ Anderson, OS RAS/Partitioning Project Lead  
SGI - Silicon Graphics Inc          rja@sgi.com

RE: gate.S and recent toolchains

[Chen, Kenneth W]

Russ Anderson wrote on Wednesday, March 29, 2006 9:23 AM
> Chen, Kenneth W wrote:
> > 
> > Russ, do you reference __mca_table at run time?  It looks like so.  It's
> > problematic that __mca_table is in __init section.  Though I doubt that
> > was the reason Jes ran into.
> 
> It was right next to __ex_table in the original patch.
> Yes, it is referenced at run time and should not be in 
> the __init section.

I guess it's a merge error then :-P

- Ken

Re: gate.S and recent toolchains

[Russ Anderson]

Chen, Kenneth W wrote:
> Russ Anderson wrote on Wednesday, March 29, 2006 9:23 AM
> > Chen, Kenneth W wrote:
> > > 
> > > Russ, do you reference __mca_table at run time?  It looks like so.  It's
> > > problematic that __mca_table is in __init section.  Though I doubt that
> > > was the reason Jes ran into.
> > 
> > It was right next to __ex_table in the original patch.
> > Yes, it is referenced at run time and should not be in 
> > the __init section.
> 
> I guess it's a merge error then :-P

This patch moves __mca_table out of the __init section.

Jes, can you give this a try?

-----------------------------------------------------
 arch/ia64/kernel/vmlinux.lds.S |   18 +++++++++---------
 1 files changed, 9 insertions(+), 9 deletions(-)

Index: linux-2.6-linus/arch/ia64/kernel/vmlinux.lds.S
=================================--- linux-2.6-linus.orig/arch/ia64/kernel/vmlinux.lds.S	2006-03-28 09:03:23.000000000 -0600
+++ linux-2.6-linus/arch/ia64/kernel/vmlinux.lds.S	2006-03-29 11:24:56.943922578 -0600
@@ -70,6 +70,15 @@ SECTIONS
 	  __stop___ex_table = .;
 	}
 
+  /* MCA table */
+  . = ALIGN(16);
+  __mca_table : AT(ADDR(__mca_table) - LOAD_OFFSET)
+	{
+	  __start___mca_table = .;
+	  *(__mca_table)
+	  __stop___mca_table = .;
+	}
+
   /* Global data */
   _data = .;
 
@@ -128,15 +137,6 @@ SECTIONS
 	  *(.initcall6.init)
 	  *(.initcall7.init)
 	  __initcall_end = .;
-	}
-
-  /* MCA table */
-  . = ALIGN(16);
-  __mca_table : AT(ADDR(__mca_table) - LOAD_OFFSET)
-	{
-	  __start___mca_table = .;
-	  *(__mca_table)
-	  __stop___mca_table = .;
 	}
 
   .data.patch.vtop : AT(ADDR(.data.patch.vtop) - LOAD_OFFSET)

Re: gate.S and recent toolchains

[Jes Sorensen]

Chen, Kenneth W wrote:
> Russ Anderson wrote on Wednesday, March 29, 2006 8:51 AM
>>> Basically I am unable to build a kernel from Linus' git tree on one
>>> system since Monday morning, whereas it boots fine if I build it on an
>>> older system. I tried installing an older toolchain on this box but it
>>> still didn't seem to make a difference. The system boots, calls init and
>>> I most often get the message "INIT: 2.85 booting", then a hang. In some
>>> instances I don't even get the INIT message.
>> It builds and boots on all the machines I have tried,
>> including the same build and boot machines Jes uses.
>> Jes can't get it to work, I can't get it to fail.
> 
> Russ, do you reference __mca_table at run time?  It looks like so.  It's
> problematic that __mca_table is in __init section.  Though I doubt that
> was the reason Jes ran into.

Hi Ken,

I actually noticed that earlier and tried moving it around, but it made
no difference. I forgot it again in the noise.

Cheers,
Jes

Re: gate.S and recent toolchains

[Jes Sorensen]

H. J. Lu wrote:
> On Wed, Mar 29, 2006 at 06:07:58PM +0200, Jes Sorensen wrote:
>> The specific toolchain I am using is gcc-4.1.0 based, but I tried
>> 3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
>> another system seems to do fine (admittedly it had binutils 2.14.90.x).
>>
> 
> Can you try the current Linux binutils 2.16.91.0.7 at
> 
> http://www.kernel.org/pub/linux/devel/binutils/

Will do first thing in the morning.

Thanks,
Jes

Re: gate.S and recent toolchains

[Jes Sorensen]

Russ Anderson wrote:
> Chen, Kenneth W wrote:
>> Russ Anderson wrote on Wednesday, March 29, 2006 9:23 AM
>>> Chen, Kenneth W wrote:
>>>> Russ, do you reference __mca_table at run time?  It looks like so.  It's
>>>> problematic that __mca_table is in __init section.  Though I doubt that
>>>> was the reason Jes ran into.
>>> It was right next to __ex_table in the original patch.
>>> Yes, it is referenced at run time and should not be in 
>>> the __init section.
>> I guess it's a merge error then :-P
> 
> This patch moves __mca_table out of the __init section.
> 
> Jes, can you give this a try?

I am 99.999% sure it wont make a difference. I even tried reverting your
full patch and just adding the two lines to asmmacros.h and I still hit
the failture.

Cheers,
Jes

RE: gate.S and recent toolchains

[Chen, Kenneth W]

Jes Sorensen wrote on Wednesday, March 29, 2006 10:03 AM
>>>> Basically I am unable to build a kernel from Linus' git tree on one
>>>> system since Monday morning, whereas it boots fine if I build it on an
>>>> older system. I tried installing an older toolchain on this box but it
>>>> still didn't seem to make a difference. The system boots, calls init and
>>>> I most often get the message "INIT: 2.85 booting", then a hang. In some
>>>> instances I don't even get the INIT message.
>>> It builds and boots on all the machines I have tried,
>>> including the same build and boot machines Jes uses.
>>> Jes can't get it to work, I can't get it to fail.
>> 
>> Russ, do you reference __mca_table at run time?  It looks like so.  It's
>> problematic that __mca_table is in __init section.  Though I doubt that
>> was the reason Jes ran into.
>
>I actually noticed that earlier and tried moving it around, but it made
>no difference. I forgot it again in the noise.

That's what I thought as well, very unlikely it is the issue you ran into.

- Ken

Re: gate.S and recent toolchains

[Jack Steiner]

On Wed, Mar 29, 2006 at 10:51:14AM -0600, Russ Anderson wrote:
> Jes Sorensen wrote:
> > 
> > Hi,
> > 
> > I've been chasing a very strange but since Monday and I am starting
> > to hit a brick wall.
> > 
> > Basically I am unable to build a kernel from Linus' git tree on one
> > system since Monday morning, whereas it boots fine if I build it on an
> > older system. I tried installing an older toolchain on this box but it
> > still didn't seem to make a difference. The system boots, calls init and
> > I most often get the message "INIT: 2.85 booting", then a hang. In some
> > instances I don't even get the INIT message.

Have you NMI'ed the system to see where it is hanging.

Also, give it a spin on the simulator. These kinds of bugs are
frequently easier to debug on the simulator. If you
want, send me a pointer to a kernel & I see what happens.


> 
> It builds and boots on all the machines I have tried,
> including the same build and boot machines Jes uses.
> Jes can't get it to work, I can't get it to fail.
> 
> > What sparked it was Russ Anderson's patch which added a few lines to
> > include/asm-ia64/asmmacros.h, git id
> > d2a28ad9fa7bf16761d070d8a3338375e1574b32
> > 
> > I've tracked it down to being the two lines doing a .section/.previous
> > in the below patch chunk causing it. However it doesn't seem to be the
> > specific .section name that does it, ie. any random .section/.previous
> > pair I add to arch/ia64/kernel/gate.S has the same effect.
> > 
> > Now I am not a wiz in linker/gas magic, so I was hoping someone else
> > had an idea what could be causing this? Anyone seen anything similar?
> > 
> > The specific toolchain I am using is gcc-4.1.0 based, but I tried
> > 3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
> > another system seems to do fine (admittedly it had binutils 2.14.90.x).
> > 
> > Any suggestions would be greatly appreciated.
> > 
> > Thanks,
> > Jes
> > 
> > 
> > 
> > diff --git a/include/asm-ia64/asmmacro.h b/include/asm-ia64/asmmacro.h
> > index 77af457..d4cec32 100644
> > --- a/include/asm-ia64/asmmacro.h
> > +++ b/include/asm-ia64/asmmacro.h
> > @@ -51,6 +51,17 @@ name:
> >    [99:]        x
> > 
> >  /*
> > + * Tag MCA recoverable instruction ranges.
> > + */
> > +
> > +       .section "__mca_table", "a"             // declare section &
> > section attributes
> > +       .previous
> > +
> > +# define MCA_RECOVER_RANGE(y)                  \
> > +       .xdata4 "__mca_table", y-., 99f-.;      \
> > +  [99:]
> > +
> > +/*
> >   * Mark instructions that need a load of a virtual address patched to be
> >   * a load of a physical address.  We use this either in critical
> > performance
> >   * path (ivt.S - TLB miss processing) or in places where it might not be
> > 
> 
> 
> -- 
> Russ Anderson, OS RAS/Partitioning Project Lead  
> SGI - Silicon Graphics Inc          rja@sgi.com
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

-- 
Thanks

Jack Steiner (steiner@sgi.com)          651-683-5302
Principal Engineer                      SGI - Silicon Graphics, Inc.

Re: gate.S and recent toolchains

[Jes Sorensen]

>>>>> "HJ" = H J Lu <hjl@lucon.org> writes:

HJ> On Wed, Mar 29, 2006 at 06:07:58PM +0200, Jes Sorensen wrote:
>> The specific toolchain I am using is gcc-4.1.0 based, but I tried
>> 3.3.4 (with binutils 2.15) and it failed as well. However 4.0.2 on
>> another system seems to do fine (admittedly it had binutils
>> 2.14.90.x).
>> 

HJ> Can you try the current Linux binutils 2.16.91.0.7 at

HJ> http://www.kernel.org/pub/linux/devel/binutils/

HJ,

Tried it, no difference :(

Cheers,
Jes