gcc 4.1.2 m68k code generator bug fixed (lang/php5 compilation errors)

Hi!

I got a patch from Gunther Nikl, who works on gcc/AmigaOS, which fixes the
jumptable bug in the 68k code generator of gcc 4.1.2. It is responsible for
the compilation errors in pkgsrc/lang/php5 (php 5.2.6).

This patch comes from a backport from mainline (SVN 128601) to gcc 4.1.0,
but I could apply it to 4.1.2 without bigger problems. You will find it
attached.

Also I have put a new gcc binary for NetBSD/68k 4.0 onto the FTP server:
ftp://ftp.NetBSD.org/pub/NetBSD/misc/phx/gcc-4.1.2-update.tar.gz

It is tested, and lang/php5 compiles now.

--

-- 
    _  Frank Wille (frank <at> phoenix.owl.de)
 _ //  http://sun.hasenbraten.de/~frank/
 \X/   Phx  <at>  #AmigaGer

2008-11-06  nobody

	Backport from mainline (SVN 128601):
	2007-09-19  Roman Zippel <zippel <at> linux-m68k.org>

	* config/m68k/m68k.c (output_move_himode): Remove jump table
	recognition.
	config/m68k/m68k.md (lea): Likewise.
	* config/m68k/m68k.c (print_operand_address): Use simple pc

Anyone using current on m68k? (Was: port-atari/39965: "lock error" panic)

From: David Holland <dholland-bugs <at> netbsd.org>

> On Sun, Nov 30, 2008 at 09:12:13PM -0800, David Ross wrote:
> > c)  Same as b) but with BOOTX kernel.  No networking set up.
> > Result: mostly the same as b)...  Different panic but consistent with 
the 
> > "lock error" panic in terms of when it occurred.
> >
> > Screenshots:
> > http://i33.tinypic.com/2enx6zb.jpg
> > http://i37.tinypic.com/okwraw.jpg
> > http://i37.tinypic.com/281vpyg.jpg
> 
> This is reaching biodone() from (what I assume is) an interrupt
> handler while sleeping in biowait() on the same buffer, and choking
> because it already holds the buffer lock for that buffer.
> 
> This should not be possible, so it looks to me as if either the
> interrupt levels or the lock/cv implementation must be broken.

It would be good to confirm if anyone has installed and tested under any 
reasonable load current on another m68k platform - to see if this is a 
general m68k issue or atari specific....

Anyone want to speak up? :)

Re: =?utf-8?q?Anyone_using_current_on_m68k?_(Was:_port-atari/39965:_"lock_erro r"_panic)?=

At 9:57 Uhr +0000 01.12.2008, David Brownlee wrote:
>It would be good to confirm if anyone has installed and tested under any
>reasonable load current on another m68k platform - to see if this is a
>general m68k issue or atari specific....
>
>Anyone want to speak up? :)

I have a netbsd-5 installation here on a Quadra 650 that has been 
pretty stable for a month or so. If that is not enough, I can drop in 
a -current kernel and see how it goes.

	hauke

--

-- 
      The ASCII Ribbon Campaign                    Hauke Fath
()     No HTML/RTF in email            Institut für Nachrichtentechnik
/\     No Word docs in email                     TU Darmstadt
      Respect for open standards              Ruf +49-6151-16-3281

utf-8?q?Anyone_using_current_on_m68k?_(Was:_port-atari/39965:_"lock_erro r"_panic)?=

On Mon, 1 Dec 2008, Hauke Fath wrote:

> At 9:57 Uhr +0000 01.12.2008, David Brownlee wrote:
>> It would be good to confirm if anyone has installed and tested under any
>> reasonable load current on another m68k platform - to see if this is a
>> general m68k issue or atari specific....
>> 
>> Anyone want to speak up? :)
>
> I have a netbsd-5 installation here on a Quadra 650 that has been pretty 
> stable for a month or so. If that is not enough, I can drop in a -current 
> kernel and see how it goes.

 	Thats a SCSI system? Thats probably perfect as a comparison
 	with the TT030, though the Falcon is an IDE system so would
 	be good to find someone with an IDE system to compare.

 	If I recall mac68k has some relatively wacky interrupt setup
 	and handling due to the nubus and the variety of models, but
 	its good that we have at least one working m68k port to
 	provide a reference.

 	If we can work out when atari stopped working (which is sometime
 	post 3.0), maybe some of the changes to mac68k after that point
 	may give a clue...

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

Re: Anyone using current on m68k? (Was: port-atari/39965: "lock error" panic)

abs <at> absd.org wrote:

> > This should not be possible, so it looks to me as if either the
> > interrupt levels or the lock/cv implementation must be broken.
> 
> It would be good to confirm if anyone has installed and tested under any 
> reasonable load current on another m68k platform - to see if this is a 
> general m68k issue or atari specific....

On news68k, upgrading to 5.0_BETA via sysinst
(i.e. fetching and extracting whole binary sets)
worked fine.

- atari has its own pmap.c so some stuff might be out of sync
- <machine/intr.h> implementation could also have problem

etc?

---
Izumi Tsutsui

Re: Anyone using current on m68k?

At 13:05 Uhr +0000 01.12.2008, David Brownlee wrote:
>>  I have a netbsd-5 installation here on a Quadra 650 that has been
>>pretty stable for a month or so. If that is not enough, I can drop
>>in a -current kernel and see how it goes.
>
>	Thats a SCSI system? Thats probably perfect as a comparison
>	with the TT030, though the Falcon is an IDE system so would
>	be good to find someone with an IDE system to compare.

Well, I never got my hands on a TT, but - for one the Quadras use the 
53c90 esp(4) driver, not the 53c80 ncrscsi(4), and for another they 
are not DMA capable. If there's anything worse than their interrup 
scheme, it's their pseudo DMA...

I remember 
<http://www.netbsd.org/cgi-bin/query-pr-single.pl?number=38758>, 
though, which may or may not be relevant here - somebody adding MD 
locking code in a fell swoop without the hardware to actually test on.

>	If we can work out when atari stopped working (which is sometime
>	post 3.0), maybe some of the changes to mac68k after that point
>	may give a clue...

netbsd-3? That's been a while... My Quadra booted a 5.99.4 kernel 
nicely in the meantime, btw.

	hauke

--

-- 
      The ASCII Ribbon Campaign                    Hauke Fath

Re: Anyone using current on m68k? (Was: port-atari/39965: "lock error" panic)

On Mon, 1 Dec 2008, Izumi Tsutsui wrote:

> abs <at> absd.org wrote:
>
>>> This should not be possible, so it looks to me as if either the
>>> interrupt levels or the lock/cv implementation must be broken.
>>
>> It would be good to confirm if anyone has installed and tested under any
>> reasonable load current on another m68k platform - to see if this is a
>> general m68k issue or atari specific....
>
> On news68k, upgrading to 5.0_BETA via sysinst
> (i.e. fetching and extracting whole binary sets)
> worked fine.
>
> - atari has its own pmap.c so some stuff might be out of sync
> - <machine/intr.h> implementation could also have problem

 	Excellent point - syncing sys/arch/atari/atari/pmap.c with
 	sys/arch/m68k/m68k/pmap_motorola.c with a view towards switching
 	atari across to the shared pmap would certainly be a good place
 	to start, and worthwhile even if it doesn't fix this particular
 	issue.

--

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --

Re: Anyone using current on m68k? (Was: port-atari/39965: "lock error" panic)

On Mon, 1 Dec 2008, David Brownlee wrote:

> It would be good to confirm if anyone has installed and tested under any
> reasonable load current on another m68k platform - to see if this is a
> general m68k issue or atari specific....
>
> Anyone want to speak up? :)

   I am running a netbsd-5 branch on my amiga (and am in the process of 
slowly doing a native build).  The next time the build stops, I can boot
a -current kernel and see if that works.

--
Michael L. Hitch			mhitch <at> montana.edu
Computer Consultant
Information Technology Center
Montana State University	Bozeman, MT	USA

Anyone using current on m68k? (Was: port-atari/39965: "lock_error" panic)

On Mon, 1 Dec 2008, Michael L. Hitch wrote:

> On Mon, 1 Dec 2008, David Brownlee wrote:
>
>> It would be good to confirm if anyone has installed and tested under any
>> reasonable load current on another m68k platform - to see if this is a
>> general m68k issue or atari specific....
>> 
>> Anyone want to speak up? :)
>
>  I am running a netbsd-5 branch on my amiga (and am in the process of slowly 
> doing a native build).  The next time the build stops, I can boot
> a -current kernel and see if that works.

 	NetBSD-5 is probably close enough for this test - the issue
 	has been showing up on NetBSD/atari current from before the
 	netbsd-5 branch.
 	Thats another good data point though - amiga and atari appear
 	to be among the few m68k ports which don't use the common
 	sys/arch/m68k/m68k/pmap_motorola.c and their pmap.c files
 	are *very* similar, apart from a various cleanups and fixes
 	to the amiga pmap.

 	So... quickest shortterm fix might be to sync the atari pmap
 	with the amiga one, though a better solution (tho' more effort)
 	would be to switch atari to sys/arch/m68k/m68k/pmap_motorola.c...

 	Thanks

--

-- 

Re: Anyone using current on m68k? (Was: port-atari/39965: "lock error" panic)

On Mon, 1 Dec 2008, David Brownlee wrote:

>> - atari has its own pmap.c so some stuff might be out of sync
>> - <machine/intr.h> implementation could also have problem
>
> 	Excellent point - syncing sys/arch/atari/atari/pmap.c with
> 	sys/arch/m68k/m68k/pmap_motorola.c with a view towards switching
> 	atari across to the shared pmap would certainly be a good place
> 	to start, and worthwhile even if it doesn't fix this particular
> 	issue.

   The amiga has it's own pmap as well, although I've got it to the point
where the amiga can use either its local pmap or pmap_motorola.  Hmm, I
think I'm currently running with pmap_motorola on my 5.0_BETA kernel.
I'll need to build a kernel using the old pmap as well.

--

-- 
Michael L. Hitch			mhitch <at> montana.edu
Operations Consulting, Information Technology Center
Montana State University	Bozeman, MT	USA