FPU interrupt handler

FPU interrupt handler

[William Lee Irwin III]

        EXPORT(dec_intr_fpu)
dec_intr_fpu:   PANIC("Unimplemented FPU interrupt handler")

I've not encountered this quite yet, but I have an R3K-based
DecStation up and running, and if I understand this properly,
R3K delivers FPU exceptions as interrupts. It looks like
this could take the machine down.


Maciej, I was asked to notify you specifically, but this is
my only known point of contact. If you could look into this,
I would be much obliged.


Thanks,
Bill

Re: FPU interrupt handler

[Maciej W. Rozycki]

On Sun, 25 Nov 2001, William Lee Irwin III wrote:

>         EXPORT(dec_intr_fpu)
> dec_intr_fpu:   PANIC("Unimplemented FPU interrupt handler")
> 
> I've not encountered this quite yet, but I have an R3K-based
> DecStation up and running, and if I understand this properly,
> R3K delivers FPU exceptions as interrupts. It looks like
> this could take the machine down.

 Nope, R3k DECstations used to use the FPU emulator unconditionally. :-(
This would have never got triggered.  An FPU interrupt-to-exception glue
was non-existent.

> Maciej, I was asked to notify you specifically, but this is
> my only known point of contact. If you could look into this,
> I would be much obliged.

 No need to do anything -- I happened to work on the FPU a bit recently
and I sent a patch here on Friday that adds the glue, enables the FPU and
rips the broken code off.  Ralf has already applied the patch. 

 I would *really* appreciate any feedback on DECstation patches I'm
sending here -- for quite some time I have a feeling I'm the last one to
run Linux on a DECstation... :-(

  Maciej

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: FPU interrupt handler

[Houten K.H.C. van (Karel)]

Hi Maciej

You wrote:
> 
>  I would *really* appreciate any feedback on DECstation patches I'm
> sending here -- for quite some time I have a feeling I'm the last one to
> run Linux on a DECstation... :-(
> 

I would love to do more work on my DECstation, but I had some disk
problems recently, and I don't seem te get any newer kernels
then 2.4.0-test9 running after a native compile by the toolchain
provided by H.J. Lu.

Some kernels don't start-up, others hang just before forking init,
and all have problems with my serial console.

When I get a recent kernel running again, I would love to update my
DECStation Linux Website with newer instructions and a new root FS.

Regards,

-- 
Karel van Houten

----------------------------------------------------------
The box said "Requires Windows 95 or better."
I can't understand why it won't work on my Linux computer. 
----------------------------------------------------------

Re: FPU interrupt handler

[Maciej W. Rozycki]

Hi Karel,

> I would love to do more work on my DECstation, but I had some disk
> problems recently, and I don't seem te get any newer kernels
> then 2.4.0-test9 running after a native compile by the toolchain
> provided by H.J. Lu.

 Hmm, for R3k gcc 2.95.3 + binutils 2.11.2 as available at my site seem to
be rock solid.

 For R4k you need binutils 2.11.92, as there is a problem with dla/la
expansion in the Ulf's patch for .mips3+.  That actually can be fixed in
2.11.2 easily but I was going to switch to 2.11.92 anyway, as it has more
MIPS/Linux support integrated and 2.12 is supposedly soon to be released.
Unfortunately 2.11.92 is not as stable as 2.11.2 due to generic ELF code
problems, but I'm trying to track changes and spot a more stable snapshot.

> Some kernels don't start-up, others hang just before forking init,
> and all have problems with my serial console.

 Well, I'm very happy with a /240 running a 2.4.14 snapshot dated
20011123.  For a /260 I need a small, but critical bugfix I'm sending to
the list right now.  I wonder how was it possible for the bug to remain
uncovered for so long as it's absolutely lethal and often triggered (I've
only got my /260 recently and it wasn't even running a few minutes
continuously before the fix). 

 I can't comment other models.

> When I get a recent kernel running again, I would love to update my
> DECStation Linux Website with newer instructions and a new root FS.

 I may upload binaries of my kernels to my site if they are to be useful
-- they are fully monolithic (but with kmod support) due to historical
reasons.  Only IPv6 is modular due to its unstability -- it freezes the
system immediately on my /240 and splashes a bunch of suspicious messages
on my /260 (weird, but no time to debug).  They only support /240 and /260
due to CONFIG_CPU_HAS_WB unset.

  Maciej

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: FPU interrupt handler

[Houten K.H.C. van (Karel)]

Hi Maciej,

>  Hmm, for R3k gcc 2.95.3 + binutils 2.11.2 as available at my site seem to
> be rock solid.
> 
>  For R4k you need binutils 2.11.92, as there is a problem with dla/la
> expansion in the Ulf's patch for .mips3+.  That actually can be fixed in
> 2.11.2 easily but I was going to switch to 2.11.92 anyway, as it has more
> MIPS/Linux support integrated and 2.12 is supposedly soon to be released.
> Unfortunately 2.11.92 is not as stable as 2.11.2 due to generic ELF code
> problems, but I'm trying to track changes and spot a more stable snapshot.

I'm using the RedHat 7.1 packages from oss:
binutils-2.11.92.0.10-1.mips.rpm
gcc-2.96-99.1.mips.rpm


> > Some kernels don't start-up, others hang just before forking init,
> > and all have problems with my serial console.
> 
>  Well, I'm very happy with a /240 running a 2.4.14 snapshot dated
> 20011123.  For a /260 I need a small, but critical bugfix I'm sending to
> the list right now.  I wonder how was it possible for the bug to remain
> uncovered for so long as it's absolutely lethal and often triggered (I've
> only got my /260 recently and it wasn't even running a few minutes
> continuously before the fix). 
My 'main' mips box at home is a /260. I sometimes test things out on
a /240, but I don't have access to other boxes.
 
>  I can't comment other models.
> 
> > When I get a recent kernel running again, I would love to update my
> > DECStation Linux Website with newer instructions and a new root FS.
> 
>  I may upload binaries of my kernels to my site if they are to be useful
> -- they are fully monolithic (but with kmod support) due to historical
> reasons.  Only IPv6 is modular due to its unstability -- it freezes the
> system immediately on my /240 and splashes a bunch of suspicious messages
> on my /260 (weird, but no time to debug).  They only support /240 and /260
> due to CONFIG_CPU_HAS_WB unset.

Yes please. I hope to get a new disk this week, so I can build a
stable development server...

Thanks a lot,


-- 
Karel van Houten

----------------------------------------------------------
The box said "Requires Windows 95 or better."
I can't understand why it won't work on my Linux computer. 
----------------------------------------------------------

Re: FPU interrupt handler

[Maciej W. Rozycki]

Hi Karel,

> I'm using the RedHat 7.1 packages from oss:
> binutils-2.11.92.0.10-1.mips.rpm
> gcc-2.96-99.1.mips.rpm

 Can't comment on these.  I feel a bit uneasy about gcc 2.96, possibly due
to the way it was incarnated.  I'll probably switch directly to 3.x when
it's proved stable enough not to distract me from primary tasks.  I don't
know how much binutils 2.11.92.0.10 differ from the mainline.

> >  I may upload binaries of my kernels to my site if they are to be useful
[...]
> Yes please. I hope to get a new disk this week, so I can build a
> stable development server...

 OK -- I should have them by tomorrow.

  Maciej

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: FPU interrupt handler

[Maciej W. Rozycki]

On Mon, 26 Nov 2001, Houten K.H.C. van (Karel) wrote:

> >  I may upload binaries of my kernels to my site if they are to be useful
> > -- they are fully monolithic (but with kmod support) due to historical
> > reasons.  Only IPv6 is modular due to its unstability -- it freezes the
> > system immediately on my /240 and splashes a bunch of suspicious messages
> > on my /260 (weird, but no time to debug).  They only support /240 and /260
> > due to CONFIG_CPU_HAS_WB unset.
> 
> Yes please. I hope to get a new disk this week, so I can build a
> stable development server...

 The kernels are now available at
'ftp://ftp.ds2.pg.gda.pl/pub/macro/linux/'. 

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: FPU interrupt handler

[Ralf Baechle]

On Mon, Nov 26, 2001 at 12:57:23PM +0100, Maciej W. Rozycki wrote:

>  I would *really* appreciate any feedback on DECstation patches I'm
> sending here -- for quite some time I have a feeling I'm the last one to
> run Linux on a DECstation... :-(

Certainly not, on the recent Linux porter meating it was probably the
most used type of non-x86 machine.  So rather blame general passivity
of man kind ...

  Ralf

Re: FPU interrupt handler

[Maciej W. Rozycki]

On Tue, 27 Nov 2001, Ralf Baechle wrote:

> >  I would *really* appreciate any feedback on DECstation patches I'm
> > sending here -- for quite some time I have a feeling I'm the last one to
> > run Linux on a DECstation... :-(
> 
> Certainly not, on the recent Linux porter meating it was probably the
> most used type of non-x86 machine.  So rather blame general passivity
> of man kind ...

 Well, but what can I think when even fatal core errors remain
unnoticed?...  The int-handler.S was one and the file wasn't changed for a
very long time, i.e. since I fixed another fatal error there...  And both
looked like they were present from the day one...

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

Re: FPU interrupt handler

[Ralf Baechle]

On Mon, Nov 26, 2001 at 05:10:53PM +0100, Maciej W. Rozycki wrote:

>  Well, but what can I think when even fatal core errors remain
> unnoticed?...  The int-handler.S was one and the file wasn't changed for a
> very long time, i.e. since I fixed another fatal error there...  And both
> looked like they were present from the day one...

In that case if nobody complains those nobodies all had their fair chance :-)

  Ralf

Re: FPU interrupt handler

[William Lee Irwin III]

On Mon, Nov 26, 2001 at 12:57:23PM +0100, Maciej W. Rozycki wrote:
>  Nope, R3k DECstations used to use the FPU emulator unconditionally. :-(
> This would have never got triggered.  An FPU interrupt-to-exception glue
> was non-existent.

Okay, that is at least some relief.

On Mon, Nov 26, 2001 at 12:57:23PM +0100, Maciej W. Rozycki wrote:
>  No need to do anything -- I happened to work on the FPU a bit recently
> and I sent a patch here on Friday that adds the glue, enables the FPU and
> rips the broken code off.  Ralf has already applied the patch. 

Excellent! I'll apply it as well.

On Mon, Nov 26, 2001 at 12:57:23PM +0100, Maciej W. Rozycki wrote:
>  I would *really* appreciate any feedback on DECstation patches I'm
> sending here -- for quite some time I have a feeling I'm the last one to
> run Linux on a DECstation... :-(

Well, I'm a more than willing tester, and perhaps I can even bring some
code to the table in my spare time at some point for the drivers my
system needs help with.


Cheers,
Bill