* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: David S. Miller @ 2006-05-10 19:40 UTC (permalink / raw)
To: rth; +Cc: linux-arch, linuxppc-dev, paulus, linux-kernel
In-Reply-To: <20060510154702.GA28938@twiddle.net>
From: Richard Henderson <rth@twiddle.net>
Date: Wed, 10 May 2006 08:47:13 -0700
> How do you plan to address the compiler optimizing
...
> Across the schedule, we may have changed cpus, making the cached
> address invalid.
Per-cpu variables need to be accessed only with preemption
disabled. And the preemption enable/disable operations
provide a compiler memory barrier.
#define preempt_disable() \
do { \
inc_preempt_count(); \
barrier(); \
} while (0)
...
#define preempt_enable() \
do { \
preempt_enable_no_resched(); \
barrier(); \
preempt_check_resched(); \
} while (0)
The scheduler itself need to take care to not cause the situation
you mention either.
Therefore this is an issue we had already, not some new thing
introduced by using __thread for per-cpu variables.
^ permalink raw reply
* Re: IMAP_ADDR on PPC 8xx
From: John Rigby @ 2006-05-10 19:58 UTC (permalink / raw)
To: Dan Malek; +Cc: linuxppc-embedded
In-Reply-To: <6CE25E1C-0F7A-4644-926E-746C17A7A851@embeddedalley.com>
Thanks Dan for the extra info.
Just to let everyone know even though this turned out to not
be our problem we here at Freescale are looking at how we
can make sure this does not happen in the future. We want
any uboot shipped with future Freescale boards to have the
correct behavior. We also understand the board documentation
needs to be changed to mention the appropriate value for IMMR
for linux.
John
jrigby@freescale.com
On 5/10/06, Dan Malek <dan@embeddedalley.com> wrote:
>
> On May 10, 2006, at 12:49 PM, Walter L. Wimer III wrote:
>
> > FYI, here's a table from the "MPC885ADS PowerQUICC(tm) Application
> > Development System User's Guide", available on Freescale's website.
> > This table seems to confirm how they've configured their U-Boot -- the
> > IMMR is set to 0x02200000...
>
> Freescale never ported U-Boot to the MPC855ADS platform, and
> I don't believe it came with any software at all. This IMMR value was
> chosen to be backward compatible with some old software tools to
> get a compact address space without the need of using the MMU,
> long before we had Linux running on the platform.
>
> > This may be fine for non-Linux purposes, but it looks like we need to
> > spread some gospel to Freescale regarding the correct IMMR address for
> > U-Boot / Linux....
>
> I think you better be pointing fingers at the clueless person that
> provided the software you have, as it didn't come from Freescale nor
> any of the public U-Boot sources.
>
> Thanks.
>
> -- Dan
>
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded
>
^ permalink raw reply
* Re: IMAP_ADDR on PPC 8xx
From: Walter L. Wimer III @ 2006-05-10 20:19 UTC (permalink / raw)
To: linuxppc-embedded
In-Reply-To: <6CE25E1C-0F7A-4644-926E-746C17A7A851@embeddedalley.com>
On Wed, 2006-05-10 at 13:45 -0400, Dan Malek wrote:
> > This may be fine for non-Linux purposes, but it looks like we need
> > to spread some gospel to Freescale regarding the correct IMMR
> > address for U-Boot / Linux....
>
> I think you better be pointing fingers at the clueless person that
> provided the software you have, as it didn't come from Freescale nor
> any of the public U-Boot sources.
My sincerest apologies to Freescale. Upon further research, it appears
that it was another TimeSys engineer (who has since moved on to another
job) who built and installed U-Boot on our MPC885ADS board. Strangely,
it appears that he started with the community U-Boot 1.1.3 and then
added a patch to change the IMMR value (plus a few other things). I'm
not sure why. He's a very bright guy, so I'm very puzzled....
Anyway, mystery solved.
Again, my sincerest apologies to Freescale.
I built the community U-Boot 1.1.4 and flashed it onto our board. Linux
2.6.16.11 now boots much farther than before (I'm now getting printk()
output). The kernel is hitting a new problem now, but I suspect it may
be related to the TLB code that Marcello has discussed recently. I'm
going to go look at that patch next. :-)
> Thanks.
>
> -- Dan
Thanks again to everyone!!!
Walt Wimer
TimeSys Corporation
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Paul Mackerras @ 2006-05-10 21:05 UTC (permalink / raw)
To: David S. Miller; +Cc: linux-arch, linuxppc-dev, linux-kernel, rth
In-Reply-To: <20060510.124003.04457042.davem@davemloft.net>
David S. Miller writes:
> From: Richard Henderson <rth@twiddle.net>
> Date: Wed, 10 May 2006 08:47:13 -0700
>
> > How do you plan to address the compiler optimizing
> ...
> > Across the schedule, we may have changed cpus, making the cached
> > address invalid.
>
> Per-cpu variables need to be accessed only with preemption
> disabled. And the preemption enable/disable operations
> provide a compiler memory barrier.
No, Richard has a point, it's not the value that is the concern, it's
the address, which gcc could assume is still valid after a barrier.
Drat.
Paul.
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Christoph Cebulla @ 2006-05-10 20:54 UTC (permalink / raw)
To: Wolfgang Pfeiffer; +Cc: linuxppc-dev list, Johannes Berg
In-Reply-To: <20060510172443.GF3878@localhost>
Hi,
> > Did it ever work?
Kind of... with a patch[1] originally by Ben Collins and adapted by
Gaudenz Steinlin it at least lets you play your favorite music. The
kernel I use is a 2.6.17-rc3-git5.
> Not with Debian, but with this Ubuntu live CD I have sound:
> Ubuntu 6.06 "Dapper Drake" Beta 2 powerpc (20060427)
>
> The kernel on their CD is a 2.6.15-21-powerpc
This patch applied cleanly on every kernel I tried since 2.6.15...
> 5.8 is, IINM, the last PPC 15" Powerbook that Apple made
>
> > Also, try snd-aoa.
Best regards,
Christoph
[1] http://lists.debian.org/debian-powerpc/2006/03/msg00226.html
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Wolfgang Pfeiffer @ 2006-05-10 21:30 UTC (permalink / raw)
To: Johannes Berg; +Cc: linuxppc-dev list
In-Reply-To: <2829.131.234.104.112.1147277053.squirrel@secure.sipsolutions.net>
On Wed, May 10, 2006 at 06:04:13PM +0200, Johannes Berg wrote:
> Also, try snd-aoa.
Impossible here, it does not compile here. Neither with gcc 4.0 or 4.1
I took your instructions from
http://lists.debian.org/debian-powerpc/2006/03/msg00470.html
And if I understand them correctly it's not necessary to descend to
some directory level of the kernel source for my current kernel. I
slightly modified your instructions (being addicted to git, Sorry
about that ... :) .. :
Something like:
mkdir snd-aoa
cd snd-aoa/
git clone http://johannes.sipsolutions.net/snd-aoa.git/
cd snd-aoa/
make
The above did not work: compile problems:
-----------------
include/sound/pcm.h:742: error: dereferencing pointer to incomplete type
include/sound/pcm.h: In function 'snd_pcm_capture_empty':
include/sound/pcm.h:755: error: dereferencing pointer to incomplete type
include/sound/pcm.h: In function 'snd_pcm_trigger_done':
include/sound/pcm.h:762: error: dereferencing pointer to incomplete type
include/sound/pcm.h: At top level:
include/sound/pcm.h:844: error: syntax error before 'u_int32_t'
include/sound/pcm.h:844: warning: function declaration isn't a prototype
include/sound/pcm.h:846: error: syntax error before 'u_int64_t'
include/sound/pcm.h:846: warning: function declaration isn't a prototype
include/sound/pcm.h:901: error: syntax error before 'snd_pcm_format_size'
include/sound/pcm.h:901: error: syntax error before 'size_t'
include/sound/pcm.h:901: warning: type defaults to 'int' in declaration of 'snd_pcm_format_size'
include/sound/pcm.h:901: warning: function declaration isn't a prototype
include/sound/pcm.h:901: warning: data definition has no type or storage class
include/sound/pcm.h: In function 'snd_pcm_set_runtime_buffer':
include/sound/pcm.h:936: error: dereferencing pointer to incomplete type
include/sound/pcm.h:938: error: dereferencing pointer to incomplete type
include/sound/pcm.h:939: error: dereferencing pointer to incomplete type
include/sound/pcm.h:939: error: dereferencing pointer to incomplete type
include/sound/pcm.h:940: error: dereferencing pointer to incomplete type
include/sound/pcm.h:940: error: dereferencing pointer to incomplete type
include/sound/pcm.h:941: error: dereferencing pointer to incomplete type
include/sound/pcm.h:941: error: dereferencing pointer to incomplete type
include/sound/pcm.h:943: error: dereferencing pointer to incomplete type
include/sound/pcm.h:944: error: dereferencing pointer to incomplete type
include/sound/pcm.h:945: error: dereferencing pointer to incomplete type
include/sound/pcm.h:946: error: dereferencing pointer to incomplete type
include/sound/pcm.h: At top level:
include/sound/pcm.h:966: error: syntax error before 'size_t'
include/sound/pcm.h:966: warning: function declaration isn't a prototype
include/sound/pcm.h:969: error: syntax error before 'size_t'
include/sound/pcm.h:969: warning: function declaration isn't a prototype
include/sound/pcm.h:970: error: syntax error before 'size_t'
include/sound/pcm.h:970: warning: function declaration isn't a prototype
include/sound/pcm.h: In function 'snd_pcm_mmap_data_open':
include/sound/pcm.h:981: error: dereferencing pointer to incomplete type
include/sound/pcm.h:982: error: dereferencing pointer to incomplete type
include/sound/pcm.h: In function 'snd_pcm_mmap_data_close':
include/sound/pcm.h:987: error: dereferencing pointer to incomplete type
include/sound/pcm.h:988: error: dereferencing pointer to incomplete type
include/sound/pcm.h: At top level:
include/sound/pcm.h:1000: error: syntax error before 'size_t'
include/sound/pcm.h:1001: warning: function declaration isn't a prototype
include/sound/pcm.h: In function 'snd_pcm_limit_isa_dma_size':
include/sound/pcm.h:1002: error: 'max' undeclared (first use in this function)
include/sound/pcm.h:1002: error: 'dma' undeclared (first use in this function)
In file included from /home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/aoa.h:21,
from /home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:14:
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h: At top level:
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:37: error: syntax error before 'u64'
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:37: warning: no semicolon at end of struct or union
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:40: error: syntax error before 'transfer_in'
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:44: error: syntax error before '}' token
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:142: error: field 'ofdev' has incomplete type
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:169: error: syntax error before 'u32'
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:169: warning: no semicolon at end of struct or union
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:193: error: field 'driver' has incomplete type
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c: In function 'attach_codec_to_fabric':
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:33: error: 'ENOENT' undeclared (first use in this function)
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c: In function 'aoa_fabric_register':
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:85: error: 'EALREADY' undeclared (first use in this function)
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:89: error: 'EEXIST' undeclared (first use in this function)
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:91: error: 'EINVAL' undeclared (first use in this function)
make[3]: *** [/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.o] Error 1
make[2]: *** [/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa] Error 2
make[1]: *** [_module_/home/shorty/kernel-factory/git/snd-aoa/snd-aoa] Error 2
make[1]: Leaving directory `/home/shorty/kernel-factory/git/linux-2.6'
make: *** [modules] Error 2
----------------
FYI:
------------------------
$ gcc --version
gcc (GCC) 4.0.4 20060422 (prerelease) (Debian 4.0.3-2)
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
-------------------------------------------
Same being true for this:
$ pwd
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa
$ make clean
$ MAKEFLAGS="CC=gcc-4.1" make
The end from the latter command:
--------------
include/sound/pcm.h:669: error: 'struct snd_pcm_runtime' has no member named 'status'
include/sound/pcm.h:669: error: 'struct snd_pcm_runtime' has no member named 'control'
include/sound/pcm.h:671: error: 'struct snd_pcm_runtime' has no member named 'boundary'
include/sound/pcm.h: In function 'snd_pcm_playback_ready':
include/sound/pcm.h:695: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h:696: error: 'struct snd_pcm_runtime' has no member named 'control'
include/sound/pcm.h: In function 'snd_pcm_capture_ready':
include/sound/pcm.h:709: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h:710: error: 'struct snd_pcm_runtime' has no member named 'control'
include/sound/pcm.h: In function 'snd_pcm_playback_data':
include/sound/pcm.h:724: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h:726: error: 'struct snd_pcm_runtime' has no member named 'stop_threshold'
include/sound/pcm.h:726: error: 'struct snd_pcm_runtime' has no member named 'boundary'
include/sound/pcm.h: In function 'snd_pcm_playback_empty':
include/sound/pcm.h:741: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h: In function 'snd_pcm_capture_empty':
include/sound/pcm.h:755: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h: In function 'snd_pcm_trigger_done':
include/sound/pcm.h:762: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h: At top level:
include/sound/pcm.h:844: error: expected declaration specifiers or '...' before 'u_int32_t'
include/sound/pcm.h:846: error: expected declaration specifiers or '...' before 'u_int64_t'
include/sound/pcm.h:901: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'snd_pcm_format_size'
include/sound/pcm.h: In function 'snd_pcm_set_runtime_buffer':
include/sound/pcm.h:936: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h:938: error: 'struct snd_pcm_runtime' has no member named 'dma_buffer_p'
include/sound/pcm.h:939: error: 'struct snd_pcm_runtime' has no member named 'dma_area'
include/sound/pcm.h:940: error: 'struct snd_pcm_runtime' has no member named 'dma_addr'
include/sound/pcm.h:940: error: 'struct snd_dma_buffer' has no member named 'addr'
include/sound/pcm.h:941: error: 'struct snd_pcm_runtime' has no member named 'dma_bytes'
include/sound/pcm.h:941: error: 'struct snd_dma_buffer' has no member named 'bytes'
include/sound/pcm.h:943: error: 'struct snd_pcm_runtime' has no member named 'dma_buffer_p'
include/sound/pcm.h:944: error: 'struct snd_pcm_runtime' has no member named 'dma_area'
include/sound/pcm.h:945: error: 'struct snd_pcm_runtime' has no member named 'dma_addr'
include/sound/pcm.h:946: error: 'struct snd_pcm_runtime' has no member named 'dma_bytes'
include/sound/pcm.h: At top level:
include/sound/pcm.h:966: error: expected declaration specifiers or '...' before 'size_t'
include/sound/pcm.h:966: error: expected declaration specifiers or '...' before 'size_t'
include/sound/pcm.h:969: error: expected declaration specifiers or '...' before 'size_t'
include/sound/pcm.h:969: error: expected declaration specifiers or '...' before 'size_t'
include/sound/pcm.h:970: error: expected declaration specifiers or '...' before 'size_t'
include/sound/pcm.h: In function 'snd_pcm_mmap_data_open':
include/sound/pcm.h:981: error: 'struct vm_area_struct' has no member named 'vm_private_data'
include/sound/pcm.h:982: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h: In function 'snd_pcm_mmap_data_close':
include/sound/pcm.h:987: error: 'struct vm_area_struct' has no member named 'vm_private_data'
include/sound/pcm.h:988: error: 'struct snd_pcm_substream' has no member named 'runtime'
include/sound/pcm.h: At top level:
include/sound/pcm.h:1000: error: expected declaration specifiers or '...' before 'size_t'
include/sound/pcm.h: In function 'snd_pcm_limit_isa_dma_size':
include/sound/pcm.h:1002: error: 'max' undeclared (first use in this function)
In file included from /home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/aoa.h:21,
from /home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:14:
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h: At top level:
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:37: error: expected specifier-qualifier-list before 'u64'
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:142: error: field 'ofdev' has incomplete type
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:169: error: expected specifier-qualifier-list before 'u32'
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/../soundbus/soundbus.h:193: error: field 'driver' has incomplete type
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c: In function 'attach_codec_to_fabric':
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:33: error: 'ENOENT' undeclared (first use in this function)
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c: In function 'aoa_fabric_register':
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:85: error: 'EALREADY' undeclared (first use in this function)
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:89: error: 'EEXIST' undeclared (first use in this function)
/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.c:91: error: 'EINVAL' undeclared (first use in this function)
make[3]: *** [/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa/snd-aoa-core.o] Error 1
make[2]: *** [/home/shorty/kernel-factory/git/snd-aoa/snd-aoa/aoa] Error 2
make[1]: *** [_module_/home/shorty/kernel-factory/git/snd-aoa/snd-aoa] Error 2
make[1]: Leaving directory `/home/shorty/kernel-factory/git/linux-2.6'
make: *** [modules] Error 2
-------------
I should compile with gcc 4.1, IINM:
$ cat /proc/version
Linux version 2.6.17-rc3-g5528e568-dirty (root@debby1-6) (gcc version
4.1.1 20060428 (prerelease) (Debian 4.1.0-2)) #1 Sun May 7 23:51:15
CEST 2006
Does it help?
Regards
Wolfgang
--
Wolfgang Pfeiffer: /ICQ: 286585973/ + + + /AIM: crashinglinux/
http://profiles.yahoo.com/wolfgangpfeiffer
Key ID: E3037113
http://keyserver.mine.nu/pks/lookup?search=0xE3037113&fingerprint=on
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: David S. Miller @ 2006-05-10 22:25 UTC (permalink / raw)
To: paulus; +Cc: linux-arch, linuxppc-dev, linux-kernel, rth
In-Reply-To: <17506.21908.857189.645889@cargo.ozlabs.ibm.com>
From: Paul Mackerras <paulus@samba.org>
Date: Thu, 11 May 2006 07:05:24 +1000
> No, Richard has a point, it's not the value that is the concern, it's
> the address, which gcc could assume is still valid after a barrier.
> Drat.
Oh right, and that's currently part of why we obfuscate the
address computation with the RELOC_HIDE() buisness.
Once we expose what's really going on with something like
__thread, gcc can now be "smart" about it.
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Paul Mackerras @ 2006-05-10 23:05 UTC (permalink / raw)
To: Richard Henderson; +Cc: linux-arch, linuxppc-dev, linux-kernel, t
In-Reply-To: <20060510154702.GA28938@twiddle.net>
Richard Henderson writes:
> How do you plan to address the compiler optimizing
>
> __thread int foo;
> {
> use(foo);
> schedule();
> use(foo);
> }
>
> into
>
> {
> int *tmp = &foo; // tls arithmetic here
> use(*tmp);
> schedule();
> use(*tmp);
> }
Hmmm... Would it be sufficient to use a RELOC_HIDE in __get_cpu_var,
like this?
#define __get_cpu_var(x) (*(RELOC_HIDE(&per_cpu__##x, 0)))
Paul.
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Wolfgang Pfeiffer @ 2006-05-10 23:08 UTC (permalink / raw)
To: Johannes Berg; +Cc: linuxppc-dev list
In-Reply-To: <20060510213028.GG3878@localhost>
On Wed, May 10, 2006 at 11:30:28PM +0200, Wolfgang Pfeiffer wrote:
> On Wed, May 10, 2006 at 06:04:13PM +0200, Johannes Berg wrote:
>
> > Also, try snd-aoa.
>
> Impossible here, it does not compile here. Neither with gcc 4.0 or 4.1
>
Not being sure tho' whether the fact my source dir for the running
kernel is cleaned up, IINM, (like "fakeroot make-kpkg clean") is the
culprit for the failed compile ... I'll have a look into that at
next daylight ... :)
Until then
Wolfgang
--
Wolfgang Pfeiffer: /ICQ: 286585973/ + + + /AIM: crashinglinux/
http://profiles.yahoo.com/wolfgangpfeiffer
Key ID: E3037113
http://keyserver.mine.nu/pks/lookup?search=0xE3037113&fingerprint=on
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Segher Boessenkool @ 2006-05-10 23:17 UTC (permalink / raw)
To: Paul Mackerras
Cc: linux-arch, linuxppc-dev, rth, David S. Miller, linux-kernel
In-Reply-To: <17506.21908.857189.645889@cargo.ozlabs.ibm.com>
>>> How do you plan to address the compiler optimizing
>> ...
>>> Across the schedule, we may have changed cpus, making the cached
>>> address invalid.
>>
>> Per-cpu variables need to be accessed only with preemption
>> disabled. And the preemption enable/disable operations
>> provide a compiler memory barrier.
>
> No, Richard has a point, it's not the value that is the concern, it's
> the address, which gcc could assume is still valid after a barrier.
> Drat.
Would an asm clobber of GPR13 in the schedule routines (or a wrapper
for them, or whatever) work?
Segher
^ permalink raw reply
* [RFC-PATCH] Prevent tasks from sleeping in die()
From: David Wilder @ 2006-05-11 0:19 UTC (permalink / raw)
To: linuxppc-dev
I am seeing an issue in die() when voluntary preemption is enabled.
die()
->>show_regs()->>show_instructions()->>__get_user_nocheck()->>might_sleep()
If multiple CPUs call die() and one should sleep other CPUs may block on
the die_lock held by the sleeping CPU. This problem is seen when a
soft-reset is issued as all CPUs call die() at roughly the same time.
Is this the correct way to fix this problem?
Signed-of-by: <wilder@us.ibm.com>
diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c
index 064a525..dc45bcd 100644
--- a/arch/powerpc/kernel/traps.c
+++ b/arch/powerpc/kernel/traps.c
@@ -101,6 +101,15 @@ int die(const char *str, struct pt_regs
if (debugger(regs))
return 1;
+ /* If voluntary preemption is on we can sleep
+ * in show_regs(). This is bad as we hold the
+ * die_lock. Ether the task is exiting or the system
+ * is crashing so there is no need to restore the
+ * NEED_RESCHED flag.
+ */
+ clear_need_resched();
+
+
console_verbose();
spin_lock_irq(&die_lock);
bust_spinlocks(1);
--
David Wilder
IBM Linux Technology Center
Beaverton, Oregon, USA
dwilder@us.ibm.com
(503)578-3789
^ permalink raw reply related
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Paul Mackerras @ 2006-05-10 23:44 UTC (permalink / raw)
To: Richard Henderson, t, linux-kernel, linux-arch, linuxppc-dev,
amodra
In-Reply-To: <17506.29128.591758.502430@cargo.ozlabs.ibm.com>
I wrote:
> Hmmm... Would it be sufficient to use a RELOC_HIDE in __get_cpu_var,
> like this?
>
> #define __get_cpu_var(x) (*(RELOC_HIDE(&per_cpu__##x, 0)))
But that won't work because the compiler can still cache &per_cpu__x.
I guess I have to do this:
#define __get_cpu_var(var, cpu) \
(*(__typeof__(&per_cpu__##var))({ \
void *__ptr; \
asm("addi %0,13,per_cpu__"#var"@tprel" \
: "=r" (__ptr)); \
__ptr; \
}))
That means we lose the possible optimization of combining the addi
into a following load or store. Bah. However, I guess it's still
better than what we do at the moment.
Paul.
^ permalink raw reply
* Re: [RFC-PATCH] Prevent tasks from sleeping in die()
From: Paul Mackerras @ 2006-05-10 23:45 UTC (permalink / raw)
To: David Wilder; +Cc: linuxppc-dev
In-Reply-To: <44628316.80600@us.ibm.com>
David Wilder writes:
> I am seeing an issue in die() when voluntary preemption is enabled.
> die()
> ->>show_regs()->>show_instructions()->>__get_user_nocheck()->>might_sleep()
Are you using Linus' current git tree? It has a patch that fixes this
problem by making __get_user_nocheck do the might_sleep only if the
address you give it is a user address.
Paul.
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Benjamin Herrenschmidt @ 2006-05-10 23:47 UTC (permalink / raw)
To: Johannes Berg; +Cc: linuxppc-dev list
In-Reply-To: <2829.131.234.104.112.1147277053.squirrel@secure.sipsolutions.net>
On Wed, 2006-05-10 at 18:04 +0200, Johannes Berg wrote:
> > As to the reason why I came here:
> > As I can't be sure whether the issue announced in the subject line is
> > really a kernel issue
>
> Did it ever work? I don't quite know what machine a 5,8 is since I have a
> 5,6 only. Also, try snd-aoa.
Same as mine, it works fine with snd-aoa
Ben.
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Benjamin Herrenschmidt @ 2006-05-10 23:48 UTC (permalink / raw)
To: Wolfgang Pfeiffer; +Cc: linuxppc-dev list, Johannes Berg
In-Reply-To: <20060510172443.GF3878@localhost>
On Wed, 2006-05-10 at 19:24 +0200, Wolfgang Pfeiffer wrote:
> On Wed, May 10, 2006 at 06:04:13PM +0200, Johannes Berg wrote:
> > > As to the reason why I came here:
> > > As I can't be sure whether the issue announced in the subject line is
> > > really a kernel issue
> >
> > Did it ever work?
>
> Not with Debian, but with this Ubuntu live CD I have sound:
> Ubuntu 6.06 "Dapper Drake" Beta 2 powerpc (20060427)
>
> The kernel on their CD is a 2.6.15-21-powerpc
Dapper kernel has the hack to consider it as a toonie. No volume control
nor anything.
> > I don't quite know what machine a 5,8 is since I have a
> > 5,6 only.
>
>
> 5.8 is, IINM, the last PPC 15" Powerbook that Apple made
>
> > Also, try snd-aoa.
>
> With the git kernel mentioned in my previous mail there's no snd-aoa
> available, IINM. I'll try figure out how to get the source and compile
> another kernel with this driver ...
It's not merged upstream yet.
Ben.
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Benjamin Herrenschmidt @ 2006-05-10 23:49 UTC (permalink / raw)
To: Wolfgang Pfeiffer; +Cc: linuxppc-dev list, Johannes Berg
In-Reply-To: <20060510230811.GH3878@localhost>
On Thu, 2006-05-11 at 01:08 +0200, Wolfgang Pfeiffer wrote:
> On Wed, May 10, 2006 at 11:30:28PM +0200, Wolfgang Pfeiffer wrote:
> > On Wed, May 10, 2006 at 06:04:13PM +0200, Johannes Berg wrote:
> >
> > > Also, try snd-aoa.
> >
> > Impossible here, it does not compile here. Neither with gcc 4.0 or 4.1
> >
>
> Not being sure tho' whether the fact my source dir for the running
> kernel is cleaned up, IINM, (like "fakeroot make-kpkg clean") is the
> culprit for the failed compile ... I'll have a look into that at
> next daylight ... :)
You can't build modules with "cleaned" kernel headers.
Ben.
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: David S. Miller @ 2006-05-11 0:11 UTC (permalink / raw)
To: paulus; +Cc: linux-arch, linux-kernel, linuxppc-dev, amodra, rth, t
In-Reply-To: <17506.31456.68099.57515@cargo.ozlabs.ibm.com>
From: Paul Mackerras <paulus@samba.org>
Date: Thu, 11 May 2006 09:44:32 +1000
> That means we lose the possible optimization of combining the addi
> into a following load or store. Bah. However, I guess it's still
> better than what we do at the moment.
If you have to hide the operation so deeply like this, maybe you can
do something similar to sparc64, by explicitly doing the per-cpu fixed
register and offsets, and still get the single instruction relocs that
powerpc can do for up to 64K by doing something like:
&per_cpu_blah - &per_cpu_base
to calculate the offset.
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Richard Henderson @ 2006-05-11 0:22 UTC (permalink / raw)
To: Segher Boessenkool
Cc: linux-arch, linuxppc-dev, Paul Mackerras, David S. Miller,
linux-kernel
In-Reply-To: <49BB818F-DF88-43A3-8B6A-7F9F5C7A2C3C@kernel.crashing.org>
On Thu, May 11, 2006 at 01:17:50AM +0200, Segher Boessenkool wrote:
> Would an asm clobber of GPR13 in the schedule routines (or a wrapper
> for them, or whatever) work?
No. The address is cse'd symbolically long before the r13
reference is exposed.
r~
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Alan Modra @ 2006-05-11 1:04 UTC (permalink / raw)
To: Paul Mackerras
Cc: linux-arch, linuxppc-dev, rth, David S. Miller, linux-kernel
In-Reply-To: <17506.21908.857189.645889@cargo.ozlabs.ibm.com>
On Thu, May 11, 2006 at 07:05:24AM +1000, Paul Mackerras wrote:
> No, Richard has a point, it's not the value that is the concern, it's
> the address, which gcc could assume is still valid after a barrier.
> Drat.
That may never happen, at least with a compiler that knows how to
optimise away the addi. You're using -mtls-size=16 so all your accesses
should look like
lwz rn,per_cpu_var@tprel(13)
gcc shouldn't think there is any reason to cache the address.
--
Alan Modra
IBM OzLabs - Linux Technology Centre
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Paul Mackerras @ 2006-05-11 1:21 UTC (permalink / raw)
To: Alan Modra; +Cc: linux-arch, linuxppc-dev, rth, David S. Miller, linux-kernel
In-Reply-To: <20060511010438.GE24458@bubble.grove.modra.org>
Alan Modra writes:
> gcc shouldn't think there is any reason to cache the address.
Can I rely on that being true in the future?
Paul.
^ permalink raw reply
* Re: [RFC/PATCH] Make powerpc64 use __thread for per-cpu variables
From: Alan Modra @ 2006-05-11 2:01 UTC (permalink / raw)
To: Paul Mackerras
Cc: linux-arch, linuxppc-dev, rth, David S. Miller, linux-kernel
In-Reply-To: <17506.37259.755099.974824@cargo.ozlabs.ibm.com>
On Thu, May 11, 2006 at 11:21:15AM +1000, Paul Mackerras wrote:
> Alan Modra writes:
>
> > gcc shouldn't think there is any reason to cache the address.
>
> Can I rely on that being true in the future?
It isn't true in the *present*, except with a compiler on my home
machine. :-)
__thread int i1;
void
f3 (void)
{
int x = i1;
__asm__ __volatile__ ("#dragons be here. %0" : "+r" (x));
i1 = x;
}
current mainline with -O2 -S -mtls-size=16
f3:
addi 9,2,i1@tprel
lwz 0,0(9)
#APP
#dragons be here. 0
#NO_APP
stw 0,0(9)
blr
Same thing with my modified compiler.
f3:
lwz 0,i1@tprel(2)
#APP
#dragons be here. 0
#NO_APP
stw 0,i1@tprel(2)
blr
--
Alan Modra
IBM OzLabs - Linux Technology Centre
^ permalink raw reply
* mpc8270 : i2c support
From: Heiko Schocher @ 2006-05-11 5:12 UTC (permalink / raw)
To: linuxppc-embedded
Hello,
on Thu May 4 06:11:17 2006 i wrote:
> Hmmm.... I think there are differences in the memory map between
> MPC826x and MPC827x ... can you try following Hack in
> include/asm-ppc/cpm_8260.h?
>
> -#define PROFF_I2C ((16 * 1024) - 64)
> +#define PROFF_I2C ((8 * 1024) - 64)
>
> [If it solves the problem, we must do this on a better way ;-)]
I don t know, if this was necessary to solve the problem, but i think
the follwing patch is useful, but I have no Hardware to try it out.
Any comments?
Best regards
Heiko
diff --git a/include/asm-ppc/cpm2.h b/include/asm-ppc/cpm2.h
index 6197986..738259c 100644
--- a/include/asm-ppc/cpm2.h
+++ b/include/asm-ppc/cpm2.h
@@ -181,7 +181,11 @@ #define PROFF_IDMA4_BASE ((uint)0x8afe)
*/
#define PROFF_SMC1 (0)
#define PROFF_SMC2 (64)
+#if defined(CONFIG_8272) || defined(CONFIG_MPC8555)
+#define PROFF_I2C ((8 * 1024) - 64)
+#else
#define PROFF_I2C ((16 * 1024) - 64)
+#endif
/* Define enough so I can at least use the serial port as a UART.
*/
^ permalink raw reply related
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Johannes Berg @ 2006-05-11 11:58 UTC (permalink / raw)
To: Benjamin Herrenschmidt; +Cc: linuxppc-dev list
In-Reply-To: <1147304824.32448.82.camel@localhost.localdomain>
[-- Attachment #1: Type: text/plain, Size: 633 bytes --]
On Thu, 2006-05-11 at 09:47 +1000, Benjamin Herrenschmidt wrote:
> Same as mine, it works fine with snd-aoa
I should build a database with a machine string -> user* mapping... This
means that even my brother has this one and it works great with snd-aoa
as far as I know :)
Though, since the alsa userspace libs absolutely *SUCK* no program I
know of can properly represent the mixer controls except maybe amixer...
I'd hate to work around this in the kernel, but the alsa libs are such
complex beasts that I have a feeling they won't be fixed.
Anyone up to inventing a new mixer API that uses sysfs? ;)
johannes
[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 793 bytes --]
^ permalink raw reply
* Re: Alubook 5,8: No sound with 2.6.17-rc3-g5528e568-dirty
From: Benjamin Herrenschmidt @ 2006-05-11 12:04 UTC (permalink / raw)
To: Johannes Berg; +Cc: linuxppc-dev list
In-Reply-To: <1147348702.9630.7.camel@johannes>
On Thu, 2006-05-11 at 13:58 +0200, Johannes Berg wrote:
> On Thu, 2006-05-11 at 09:47 +1000, Benjamin Herrenschmidt wrote:
>
> > Same as mine, it works fine with snd-aoa
>
> I should build a database with a machine string -> user* mapping... This
> means that even my brother has this one and it works great with snd-aoa
> as far as I know :)
>
> Though, since the alsa userspace libs absolutely *SUCK* no program I
> know of can properly represent the mixer controls except maybe amixer...
> I'd hate to work around this in the kernel, but the alsa libs are such
> complex beasts that I have a feeling they won't be fixed.
kmix looks approx. ok
> Anyone up to inventing a new mixer API that uses sysfs? ;)
>
> johannes
^ permalink raw reply
* Help: Linux porting to custom target hw
From: Thiago Galesi @ 2006-05-11 12:02 UTC (permalink / raw)
To: Jayanta Das; +Cc: linuxppc-embedded
In-Reply-To: <8584FDC94AFF7640B17B8A89B23B19B331C626@sbsserver.AlphionCorp.local>
Ok
What is the specific problem you are having?? Based on what you said
all went well, didn't it?
Basically, I think the main thing you have to configure there is flash
location (in the kernel), to be able to use it via MTD.
Booting from NFS or Flash is pretty much effortless, configurable via cmdli=
ne.
Be more specific, so we can help you more...
Thiago
On 5/10/06, Jayanta Das <JDas@alphion.com> wrote:
>
>
> Hello:
>
> Can someone point me in the right direction for some good documentation o=
n
> the above topic. Following is what I so far have done and what I need to =
do.
>
> 1. Set up host environment based on Fedora Core 4
> 2. Downloaded 'ppc4xx' tool chain, ELDK and kernel
> 3. Built U-Boot and uImage
> 4. Flashed U-Boot on the 2nd flash on the Ocotea board (AMCC440GX PowerPC=
)
> 5. Changed the environment variable for NFS mounted root fs and other MAC
> and IP addresses as needed
> 6. TFTP uImage at 400000
> 7. bootm and kernel boots all right with root fs mounted on the host
>
> I am expecting my hardware based on 440GX end of the month. I told the HW
> engineer to follow the peripheral i/f as much possible close to the ref.
> design. We are putting 32MB of Flash and 256MB of RAM.
>
> I need some guidance so that I can port U-Boot and the kernel (minimal
> changes) so that I can bring up the board initially with root fs NFS mint=
ed
> and later on the RAMDISK.
>
> Appreciate the help.
>
> Regards.
> _______________________________________________
> Linuxppc-embedded mailing list
> Linuxppc-embedded@ozlabs.org
> https://ozlabs.org/mailman/listinfo/linuxppc-embedded
>
>
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox