openwrt/staging/blogic.git
18 years ago[MIPS] s/__ASSEMBLER__/__ASSEMBLY__/ for clarity sake.
Ralf Baechle [Mon, 25 Sep 2006 14:49:49 +0000 (15:49 +0100)]
[MIPS] s/__ASSEMBLER__/__ASSEMBLY__/ for clarity sake.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Have headers_install install <asm/cachectl.h> and <asm/sysmips.h>.
Ralf Baechle [Sat, 23 Sep 2006 17:08:36 +0000 (18:08 +0100)]
[MIPS] Have headers_install install <asm/cachectl.h> and <asm/sysmips.h>.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] fstatat syscall names
Richard Sandiford [Sun, 17 Sep 2006 19:38:39 +0000 (20:38 +0100)]
[MIPS] fstatat syscall names

MIPS is the only port to call its fstatat()-related syscalls
"__NR_fstatat".  Now I can see why that might be seen as every
other port being wrong, but I think for o32, it is at best confusing.
__NR_fstat provides a plain (32-bit) stat while __NR_fstatat provides a
64-bit stat.  Changing the name to __NR_fstatat64 would make things more
explicit, match x86, and make the glibc port slightly easier.

The current name is more appropriate for n32 and n64, but it would be
appropriate for other 64-bit targets too, and those targets have chosen
to call it __NR_newfstatat instead.  Using the same name for MIPS would
again be more consistent and make the glibc port slightly easier.

I'm not wedded to this idea if the current names are preferred,
but FWIW...

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] The o32 fstatat syscall behaves differently on 32 and 64 bit kernels
Richard Sandiford [Sun, 17 Sep 2006 19:30:46 +0000 (20:30 +0100)]
[MIPS] The o32 fstatat syscall behaves differently on 32 and 64 bit kernels

While working on a glibc patch to support the fstatat() functions[1],
I noticed that the o32 implementation behaves differently on 32-bit and
64-bit kernels; the former provides a stat64 while the latter provides
a plain (o32) stat.  I think the former is what's intended, as there is
no separate fstatat64.  It's also what x86 does.

I think this is just a case of a compat too far.

[1] I've seen Khem's patch, but I don't think it's right.

Signed-off-by: Richard Sandiford <richard@codesourcery.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Remove EV96100 as previously announced.
Ralf Baechle [Wed, 20 Sep 2006 19:56:02 +0000 (20:56 +0100)]
[MIPS] Remove EV96100 as previously announced.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Replace BARRIER with more appropriate hazard barrier.
Ralf Baechle [Fri, 8 Sep 2006 02:16:21 +0000 (04:16 +0200)]
[MIPS] Replace BARRIER with more appropriate hazard barrier.

This is the unchanged part 2 of Chris' hazard cleanup.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Cleanup hazard handling.
Ralf Baechle [Fri, 8 Sep 2006 02:13:49 +0000 (04:13 +0200)]
[MIPS] Cleanup hazard handling.

Mostly based on patch by Chris Dearman and cleanups from Yoichi.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] IP27: Delete useless declaration of allocate_irqno().
Ralf Baechle [Thu, 7 Sep 2006 00:29:57 +0000 (02:29 +0200)]
[MIPS] IP27: Delete useless declaration of allocate_irqno().

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix USER_PTRS_PER_PGD for 64K page size.
Peter Watkins [Wed, 23 Aug 2006 15:15:49 +0000 (11:15 -0400)]
[MIPS] Fix USER_PTRS_PER_PGD for 64K page size.

The code in pgtable-64.h assumes TASK_SIZE is always bigger than a first
level PGDIR_SIZE. This is not the case for 64K pages, where task size is
40 bits (1TB) and a pgd entry can map 42 bits. This leads to
USER_PTRS_PER_PGD being zero for 64K pages.

Signed-off-by: Peter Watkins <treestem@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Add configuration variables for RM9xxx processor
thomas@koeller.dyndns.org [Sun, 27 Aug 2006 11:51:48 +0000 (13:51 +0200)]
[MIPS] Add configuration variables for RM9xxx processor

This patch introduces a number of configuration variables. These allow to
specify presence/absence of integrated peripherals found on the MIPS
RM9xxx processor family, based on the particular processor model used.

Signed-off-by: Thomas Koeller <thomas.koeller@baslerweb.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Move excite_fpga.h to include/asm-mips/mach-excite
thomas@koeller.dyndns.org [Sun, 27 Aug 2006 11:54:31 +0000 (13:54 +0200)]
[MIPS] Move excite_fpga.h to include/asm-mips/mach-excite

excite_fpga.h, like all platform headers, really belongs in the
platform header directory.

Signed-off-by: Thomas Koeller <thomas.koeller@baslerweb.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Suppress compiler warnings
thomas@koeller.dyndns.org [Sun, 27 Aug 2006 11:53:16 +0000 (13:53 +0200)]
[MIPS] Suppress compiler warnings

The excite platform exports hardware resources for device drivers to use.
Any driver wanting to use these resources will look up them by their names.
Since these resources are declared to have static linkage, but are not
used in the source file defining them, the compiler used to emit an
'unused' warning, which this patch suppresses.

Signed-off-by: Thomas Koeller <thomas.koeller@baslerweb.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Reformat missformated SMTC bits.
Ralf Baechle [Fri, 25 Aug 2006 11:34:33 +0000 (12:34 +0100)]
[MIPS] Reformat missformated SMTC bits.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Qemu does not have D-cache aliases
Atsushi Nemoto [Sat, 19 Aug 2006 15:33:38 +0000 (00:33 +0900)]
[MIPS] Qemu does not have D-cache aliases

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Remove F_SETSIG and F_GETSIG in favor of the asm-generic definitions.
Yoichi Yuasa [Wed, 16 Aug 2006 14:10:00 +0000 (23:10 +0900)]
[MIPS] Remove F_SETSIG and F_GETSIG in favor of the asm-generic definitions.

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Move definition of IRIX compat constant into IRIX compat code.
Ralf Baechle [Fri, 4 Aug 2006 00:49:31 +0000 (01:49 +0100)]
[MIPS] Move definition of IRIX compat constant into IRIX compat code.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Use common definitions from asm-generic/signal.h
Yoichi Yuasa [Mon, 31 Jul 2006 14:01:37 +0000 (23:01 +0900)]
[MIPS] Use common definitions from asm-generic/signal.h

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] c-r4k: Convert init functions from inline to __init.
Ralf Baechle [Tue, 1 Aug 2006 22:39:42 +0000 (23:39 +0100)]
[MIPS] c-r4k: Convert init functions from inline to __init.

With more recent compilers inline doesn't necessarily means a function
will always be inlined.  So leave that decission to the compiler and
make the function as __init.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] TLS: set_thread_area returns asmlinkage int not void.
Ralf Baechle [Tue, 19 Sep 2006 16:18:53 +0000 (17:18 +0100)]
[MIPS] TLS: set_thread_area returns  asmlinkage int not void.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] TLS: Delete unused sys32_set_thread_area
Ralf Baechle [Tue, 19 Sep 2006 16:16:56 +0000 (17:16 +0100)]
[MIPS] TLS: Delete unused sys32_set_thread_area

There is no need for a compat version.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Make PROT_WRITE imply PROT_READ.
Ralf Baechle [Sat, 16 Sep 2006 00:29:37 +0000 (01:29 +0100)]
[MIPS] Make PROT_WRITE imply PROT_READ.

18 years ago[MIPS] Atlas: update interrupt handling
Maciej W. Rozycki [Tue, 12 Sep 2006 18:12:18 +0000 (19:12 +0100)]
[MIPS] Atlas: update interrupt handling

The following change updates the Atlas interrupt handling to match that
of Malta.  Tested with a 5Kc and a 34Kf successfully.

Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Atlas: Fix building the RTC driver
Maciej W. Rozycki [Tue, 12 Sep 2006 18:02:44 +0000 (19:02 +0100)]
[MIPS] Atlas: Fix building the RTC driver

Atlas maps its RTC chip in the host mmio space rather than using the
"traditional" location in the PCI/ISA port space.  A change that has
happened to the generic RTC header requires to define ARCH_RTC_LOCATION
now.

Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Patch to arch/mips/mips-boards/generic/time.c
Kevin D. Kissell [Tue, 12 Sep 2006 10:08:08 +0000 (12:08 +0200)]
[MIPS] Patch to arch/mips/mips-boards/generic/time.c

In hooking up the perf counter overflow interrupt to the experimental
deprecated-real-soon-now /proc/perf interface last night, I had to
revisit arch/mips/mips-boards/generic/time.c, and discovered that
when the 2.6.9-based SMTC prototype was merged with the more
recent tree, it was missed that arch/mips/kernel/time.c had changed
so that even in SMP kernels, timer_interrupt() calls
local_timer_interrupt(), so there is no longer a need to invoke it
directly from mips_timer_interrupt() in those cases where
timer_interrupt() has been called.  So I got rid of that, and added the
invocation of perf_irq() if Cause.PCI is set, more-or-less following the
same logic as in the non-SMTC case, with the modifications that (a) a
runtime check for Release 2 isn't done, because it's redundant in SMTC),
and (b) we check for a clock interrupt regardless of the value returned
by the perf counter service - I don't understand why we'd want to control
that with perf_irq(), but maybe one of you knows the story.  I also got
rid of the stupid warning about the unused variable when compiled for
SMTC (another artifact of the merge). The result hasn't been beaten to
death, but boots, seems stable, and supports extended precision event
counting.

Signed-off-by: Kevin D. Kissell <kevink@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Reduce race between cpu_wait() and need_resched() checking
Atsushi Nemoto [Wed, 7 Jun 2006 16:09:01 +0000 (01:09 +0900)]
[MIPS] Reduce race between cpu_wait() and need_resched() checking

If a thread became runnable between need_resched() and the WAIT
instruction, switching to the thread will delay until a next interrupt.
Some CPUs can execute the WAIT instruction with interrupt disabled, so
we can get rid of this race on them (at least UP case).

Original Patch by Atsushi with fixing up for MIPS Technology's cores by
Ralf based on feedback from the RTL designers.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Wire up set_robust_list(2) and get_robust_list(2)
Atsushi Nemoto [Wed, 6 Sep 2006 13:42:02 +0000 (22:42 +0900)]
[MIPS] Wire up set_robust_list(2) and get_robust_list(2)

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix errors detected by "make headers_check"
Atsushi Nemoto [Wed, 6 Sep 2006 16:00:22 +0000 (01:00 +0900)]
[MIPS] Fix errors detected by "make headers_check"

* export asm/sgidefs.h
* include asm/isadep.h only if in kernel
* do not export contents of asm/timex.h and asm/user.h

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Do not lose upper 32-bit on MIPS32 with 64-bit addresses in __pte().
Ralf Baechle [Thu, 31 Aug 2006 18:39:09 +0000 (19:39 +0100)]
[MIPS] Do not lose upper 32-bit on MIPS32 with 64-bit addresses in __pte().

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Replace generic__raw_read_trylock usage
Ralf Baechle [Thu, 31 Aug 2006 13:16:06 +0000 (14:16 +0100)]
[MIPS] Replace generic__raw_read_trylock usage

generic__raw_read_trylock() is a defect generic function actually doing
a __raw_read_lock ...

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] SEAD defconfig build fix
Maciej W. Rozycki [Wed, 30 Aug 2006 13:29:57 +0000 (14:29 +0100)]
[MIPS] SEAD defconfig build fix

Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix for pci config_access on alchemy au1x000
Alexander Bigga [Tue, 29 Aug 2006 14:48:34 +0000 (16:48 +0200)]
[MIPS] Fix for pci config_access on alchemy au1x000

I've encountered a serious problem with PCI config space access on Au1x000
platforms with recent 2.6.x-kernel. With 2.4.31 the same hardware works fine.
So I was looking for the differences:

Symptoms:
- no PCI-device is seen on bootup though two or three cards are present
- lspci output is empty
- OR: lspci shows 20 times the same device
(- OR: in some slot-configurations it worked anyhow)

System(s):
1. platform with Au1500 and three PCI-devices (actually a mycable XXS1500
    with backplane for three PCI-devices)
2. platform with Au1550 and two PCI-devices (custom board)

Debugging:
I digged down to the config_access() of the au1xxx-processors in
arch/mips/pci/ops-au1000.c and switched on DEBUG.

The code of config_access() seems to be almost the same as of the
2.4.x-kernel. But the "pci_cfg_vm->addr" returned by get_vm_area(0x2000, 0)
once on booting is different. That's of course not forbidden. But the
alignment seems to be wrong. In my case, I received:

2.4.31: pci_cfg_vm->addr = c0000000
2.6.18-rc5: pci_cfg_vm->addr = c0101000

To make it short: With 2.6.x it fails on the first config-access with:
"PCI ERR detected: status 83a00356".

Fixup:
My fix is now, to use the VM_IOREMAP-flag in the get_vm_area call. This flag
seems to be introduced in mm/vmalloc.c a long time ago (in 2.6.7-bk13, I
found in gitweb).
Now, the returned address is pci_cfg_vm->addr = c0104000 and everything works
fine.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Make prepare_frametrace() not clobber v0
Atsushi Nemoto [Tue, 29 Aug 2006 03:10:22 +0000 (12:10 +0900)]
[MIPS] Make prepare_frametrace() not clobber v0

Since lmo commit 323a380bf9e1a1679a774a2b053e3c1f2aa3f179 ("Simplify
dump_stack()") made prepare_frametrace() always inlined, using $2 (v0)
in __asm__ is not safe anymore.  We can use $1 (at) instead.  Also we
should use "dla" instead of "la" for 64-bit kernel.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Do not use drop_mmu_context to flusing other task's VIPT I-cache.
Atsushi Nemoto [Fri, 25 Aug 2006 08:55:31 +0000 (17:55 +0900)]
[MIPS] Do not use drop_mmu_context to flusing other task's VIPT I-cache.

c-r4k.c and c-sb1.c use drop_mmu_context() to flush virtually tagged
I-caches, but this does not work for flushing other task's icache.  This
is for example triggered by copy_to_user_page() called from ptrace(2).
Use indexed flush for such cases.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] MT: Fix setting of XTC.
Elizabeth Oldham [Thu, 17 Aug 2006 11:39:21 +0000 (12:39 +0100)]
[MIPS] MT: Fix setting of XTC.

XTC can only be set if VPA is clear, which it may not be. There is
also the possibility of a back to back c0 register access hazard to
take care of.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] SMTC Build fix.
Ralf Baechle [Wed, 16 Aug 2006 13:05:11 +0000 (14:05 +0100)]
[MIPS] SMTC Build fix.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix 32-bit kernel by replacing 64-bit-only code.
Ralf Baechle [Tue, 15 Aug 2006 23:59:40 +0000 (00:59 +0100)]
[MIPS] Fix 32-bit kernel by replacing 64-bit-only code.

dclz() expects its 64-bit argument being passed as a single register
but on 32-bit kernels it'll actually be in a register pair.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] MT: When doing "select SMP" also select SMP's prerequesites or ...
Ralf Baechle [Tue, 15 Aug 2006 18:36:40 +0000 (19:36 +0100)]
[MIPS] MT: When doing "select SMP" also select SMP's prerequesites or ...

... kconfig will do weird stuff.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] eXcite: Don't set SERIAL_RM9000.
Ralf Baechle [Sat, 12 Aug 2006 21:56:03 +0000 (22:56 +0100)]
[MIPS] eXcite: Don't set SERIAL_RM9000.

The driver has not been merged yet so selecting it results in a warning
message.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Retire flush_icache_page from mm use.
Ralf Baechle [Sat, 12 Aug 2006 15:40:08 +0000 (16:40 +0100)]
[MIPS] Retire flush_icache_page from mm use.

On the 34K the redundant cache operations were causing excessive stalls
resulting in realtime code running on the second VPE missing its deadline.
For all other platforms this patch is just a significant performance
improvment as illustrated by below benchmark numbers.

Processor, Processes - times in microseconds - smaller is better
------------------------------------------------------------------------------
Host                 OS  Mhz null null      open slct sig  sig  fork exec sh
                             call  I/O stat clos TCP  inst hndl proc proc proc
--------- ------------- ---- ---- ---- ---- ---- ---- ---- ---- ---- ---- ----
25Kf      2.6.18-rc4     533 0.49 1.16 7.57 33.4 30.5 1.34 12.4 5497 17.K 54.K
25Kf      2.6.18-rc4-p   533 0.49 1.16 6.68 23.0 30.7 1.36 8.55 5030 16.K 48.K
4Kc       2.6.18-rc4      80 4.21 15.0 131. 289. 261. 16.5 258. 18.K 70.K 227K
4Kc       2.6.18-rc4-p    80 4.34 13.1 128. 285. 262. 18.2 258. 12.K 52.K 176K
34Kc      2.6.18-rc4      40 5.01 14.0 61.6 90.0 477. 17.9 94.7 29.K 108K 342K
34Kc      2.6.18-rc4-p    40 4.98 13.9 61.2 89.7 475. 17.6 93.7 8758 44.K 158K
BCM1480   2.6.18-rc4     700 0.28 0.60 3.68 5.92 16.0 0.78 5.08 931. 3163 15.K
BCM1480   2.6.18-rc4-p   700 0.28 0.61 3.65 5.85 16.0 0.79 5.20 395. 1464 8385
TX49-16K  2.6.18-rc3     197 0.73 2.41 19.0 37.8 82.9 2.94 17.5 4438 14.K 56.K
TX49-16K  2.6.18-rc3-p   197 0.73 2.40 19.9 36.3 82.9 2.94 23.4 2577 9103 38.K
TX49-32K  2.6.18-rc3     396 0.36 1.19 6.80 11.8 41.0 1.46 8.17 2738 8465 32.K
TX49-32K  2.6.18-rc3-p   396 0.36 1.19 6.82 10.2 41.0 1.46 8.18 1330 4638 18.K

Original patch by me with enhancements by Atsushi Nemoto.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
18 years ago[MIPS] Avoid double signal restarting.
Ralf Baechle [Tue, 8 Aug 2006 02:47:01 +0000 (03:47 +0100)]
[MIPS] Avoid double signal restarting.

In entry.S resume_userspace ... jal do_notify_resume form a loop through
which the kernel will iterate as long as work is pending.  If we
iterate through this loop more than once with no signal pending for at
least one but the last iteration we will take do the syscall restarting
multiple times resulting in a syscall return prior to the the syscall
instruction in userspace.  This may happen when debugging a multithreaded
program.

Debugging and original fix by Maciej; extended to other ABIs by me.

Signed-off-by: Maciej W. Rozycki <macro@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] MT: Initialise all writable bits in Cause register to zero.
Chris Dearman [Mon, 7 Aug 2006 14:08:01 +0000 (15:08 +0100)]
[MIPS] MT: Initialise all writable bits in Cause register to zero.

Recent 34Ks come out of reset with WP enabled on VPE 1 so we take an
immediate exception when starting the second VPE.

Signed-off-by: Chris Dearman <chris@mips.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Fix EV64120 PCI fixup in Makefile
Yoichi Yuasa [Mon, 31 Jul 2006 14:05:04 +0000 (23:05 +0900)]
[MIPS] Fix EV64120 PCI fixup in Makefile

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Add missing returns in signal code.
Ralf Baechle [Thu, 3 Aug 2006 20:54:13 +0000 (21:54 +0100)]
[MIPS] Add missing returns in signal code.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] IRIX: Crapectopy.
Ralf Baechle [Thu, 3 Aug 2006 20:53:10 +0000 (21:53 +0100)]
[MIPS] IRIX: Crapectopy.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Don't call try_to_freeze in do_signal & co.
Ralf Baechle [Thu, 3 Aug 2006 18:36:37 +0000 (19:36 +0100)]
[MIPS] Don't call try_to_freeze in do_signal & co.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Cleanup leftovers of ARCH_HAS_IRQ_PER_CPU
Ralf Baechle [Thu, 3 Aug 2006 16:34:24 +0000 (17:34 +0100)]
[MIPS] Cleanup leftovers of ARCH_HAS_IRQ_PER_CPU

CONFIG_IRQ_PER_CPU now controls the IRQ_PER_CPU stuff.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Improve unwind_stack()
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:21 +0000 (09:29 +0200)]
[MIPS] Improve unwind_stack()

This patch allows unwind_stack() to return ra for leaf function.
But it tries to detects cases where get_frame_info() wrongly
consider nested function as a leaf one.

It also pass 'unsinged long *sp' instead of 'unsigned long **sp'
as second parameter. The code looks cleaner.

Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Make get_frame_info() more robust
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:20 +0000 (09:29 +0200)]
[MIPS] Make get_frame_info() more robust

Now get_frame_info() wants to detect move sp instruction first. It
assumes that the save ra in the stack instruction can't happen
before allocating frame size space into the stack.

Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Simplify dump_stack()
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:19 +0000 (09:29 +0200)]
[MIPS] Simplify dump_stack()

Make dump_stack() code not depend on CONFIG_KALLSYMS.

It also make prepare_frametrace() always inlined to get
less false entries reported by show_raw_backtrace().

Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Make frame_info_init() more readable.
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:18 +0000 (09:29 +0200)]
[MIPS] Make frame_info_init() more readable.

Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Miscellaneous cleanup in prologue analysis code
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:17 +0000 (09:29 +0200)]
[MIPS] Miscellaneous cleanup in prologue analysis code

We usually use backtrace term for dumping a call tree during
debug. Therefore this patch renames show_frametrace() into
show_backtrace() and show_trace() into show_raw_backtrace().

It also uses the new function print_ip_sym().

Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Remove unused MODULE_RANGE macro.
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:16 +0000 (09:29 +0200)]
[MIPS] Remove unused MODULE_RANGE macro.

Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Make get_frame_info() more readable.
Franck Bui-Huu [Thu, 3 Aug 2006 07:29:15 +0000 (09:29 +0200)]
[MIPS] Make get_frame_info() more readable.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Franck Bui-Huu <vagabon.xyz@gmail.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] c-r4k: Typo fix.
Ralf Baechle [Tue, 1 Aug 2006 22:42:30 +0000 (23:42 +0100)]
[MIPS] c-r4k: Typo fix.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] N32 rt_sigqueueinfo uses O32 padding, not N64
Peter Watkins [Thu, 27 Jul 2006 20:05:49 +0000 (16:05 -0400)]
[MIPS] N32 rt_sigqueueinfo uses O32 padding, not N64

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] dump_stack() based on prologue code analysis
Atsushi Nemoto [Sat, 29 Jul 2006 14:27:20 +0000 (23:27 +0900)]
[MIPS] dump_stack() based on prologue code analysis

Instead of dump all possible address in the stack, unwind the stack frame
based on prologue code analysis, as like as get_wchan() does.  While the
code analysis might fail for some reason, there is a new kernel option
"raw_show_trace" to disable this feature.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] db1x00: Remove unused mirage_ts.c
Yoichi Yuasa [Wed, 26 Jul 2006 14:34:19 +0000 (23:34 +0900)]
[MIPS] db1x00: Remove unused mirage_ts.c

CONFIG_WM97XX_COMODULE doesn't exist.

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Rearrange show_stack, show_trace
Atsushi Nemoto [Tue, 25 Jul 2006 14:51:36 +0000 (23:51 +0900)]
[MIPS] Rearrange show_stack, show_trace

Print call-trace in show_stack() (like on other archs).  Also make
show_trace() static and simplify its argument list.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Updat mpc30x defconfig
Yoichi Yuasa [Tue, 25 Jul 2006 14:26:09 +0000 (23:26 +0900)]
[MIPS] Updat mpc30x defconfig

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Updat workpad defconfig
Yoichi Yuasa [Tue, 25 Jul 2006 14:24:54 +0000 (23:24 +0900)]
[MIPS] Updat workpad defconfig

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Update e55 defconfig
Yoichi Yuasa [Tue, 25 Jul 2006 14:24:47 +0000 (23:24 +0900)]
[MIPS] Update e55 defconfig

Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years ago[MIPS] Honour "panic_on_oops" sysctl.
Maxime Bizon [Thu, 20 Jul 2006 16:52:02 +0000 (18:52 +0200)]
[MIPS] Honour "panic_on_oops" sysctl.

Signed-off-by: Maxime Bizon <mbizon@freebox.fr>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
18 years agoMerge branch 'for-linus' of git://one.firstfloor.org/home/andi/git/linux-2.6
Linus Torvalds [Tue, 26 Sep 2006 20:07:55 +0000 (13:07 -0700)]
Merge branch 'for-linus' of git://one.firstfloor.org/home/andi/git/linux-2.6

* 'for-linus' of git://one.firstfloor.org/home/andi/git/linux-2.6: (225 commits)
  [PATCH] Don't set calgary iommu as default y
  [PATCH] i386/x86-64: New Intel feature flags
  [PATCH] x86: Add a cumulative thermal throttle event counter.
  [PATCH] i386: Make the jiffies compares use the 64bit safe macros.
  [PATCH] x86: Refactor thermal throttle processing
  [PATCH] Add 64bit jiffies compares (for use with get_jiffies_64)
  [PATCH] Fix unwinder warning in traps.c
  [PATCH] x86: Allow disabling early pci scans with pci=noearly or disallowing conf1
  [PATCH] x86: Move direct PCI scanning functions out of line
  [PATCH] i386/x86-64: Make all early PCI scans dependent on CONFIG_PCI
  [PATCH] Don't leak NT bit into next task
  [PATCH] i386/x86-64: Work around gcc bug with noreturn functions in unwinder
  [PATCH] Fix some broken white space in ia32_signal.c
  [PATCH] Initialize argument registers for 32bit signal handlers.
  [PATCH] Remove all traces of signal number conversion
  [PATCH] Don't synchronize time reading on single core AMD systems
  [PATCH] Remove outdated comment in x86-64 mmconfig code
  [PATCH] Use string instructions for Core2 copy/clear
  [PATCH] x86: - restore i8259A eoi status on resume
  [PATCH] i386: Split multi-line printk in oops output.
  ...

18 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/driver-2.6
Linus Torvalds [Tue, 26 Sep 2006 18:49:46 +0000 (11:49 -0700)]
Merge /pub/scm/linux/kernel/git/gregkh/driver-2.6

* master.kernel.org:/pub/scm/linux/kernel/git/gregkh/driver-2.6: (47 commits)
  Driver core: Don't call put methods while holding a spinlock
  Driver core: Remove unneeded routines from driver core
  Driver core: Fix potential deadlock in driver core
  PCI: enable driver multi-threaded probe
  Driver Core: add ability for drivers to do a threaded probe
  sysfs: add proper sysfs_init() prototype
  drivers/base: check errors
  drivers/base: Platform notify needs to occur before drivers attach to the device
  v4l-dev2: handle __must_check
  add CONFIG_ENABLE_MUST_CHECK
  add __must_check to device management code
  Driver core: fixed add_bind_files() definition
  Driver core: fix comments in drivers/base/power/resume.c
  sysfs_remove_bin_file: no return value, dump_stack on error
  kobject: must_check fixes
  Driver core: add ability for devices to create and remove bin files
  Class: add support for class interfaces for devices
  Driver core: create devices/virtual/ tree
  Driver core: add device_rename function
  Driver core: add ability for classes to handle devices properly
  ...

18 years ago[PATCH] s390: fix cmm kernel thread handling
Heiko Carstens [Tue, 26 Sep 2006 06:33:11 +0000 (23:33 -0700)]
[PATCH] s390: fix cmm kernel thread handling

Convert cmm's usage of kernel_thread to kthread_run.  Also create the
cmmthread at module load time, so it is possible to check if creation of
the thread fails.

In addition the cmmthread now gets terminated when the module gets unloaded
instead of leaving a stale kernel thread.  Also check the return values of
other registration functions at module load and handle their return values
appropriately.

Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Make UML use ptrace-abi.h
Jeff Dike [Tue, 26 Sep 2006 06:33:10 +0000 (23:33 -0700)]
[PATCH] Make UML use ptrace-abi.h

Include the host architecture's ptrace-abi.h instead of ptrace.h.

There was some cpp mangling of names around the ptrace.h include to avoid
symbol clashes between UML and the host architecture.  Most of these can go
away.  The exception is struct pt_regs, which is convenient to have in
userspace, but must be renamed in order that UML can define its own.

ptrace-x86_64.h needed to have some now-obsolete cpp cruft and a declaration
removed.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Split i386 and x86_64 ptrace.h
Jeff Dike [Tue, 26 Sep 2006 06:33:09 +0000 (23:33 -0700)]
[PATCH] Split i386 and x86_64 ptrace.h

The use of SEGMENT_RPL_MASK in the i386 ptrace.h introduced by
x86-allow-a-kernel-to-not-be-in-ring-0.patch broke the UML build, as UML
includes the underlying architecture's ptrace.h, but has no easy access to the
x86 segment definitions.

Rather than kludging around this, as in the past, this patch splits the
userspace-usable parts, which are the bits that UML needs, of ptrace.h into
ptrace-abi.h, which is included back into ptrace.h.  Thus, there is no net
effect on i386.

As a side-effect, this creates a ptrace header which is close to being usable
in /usr/include.

x86_64 is also treated in this way for consistency.  There was some trailing
whitespace there, which is cleaned up.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Cc: David Woodhouse <dwmw2@infradead.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] UML: tty locking
Alan Cox [Tue, 26 Sep 2006 06:33:08 +0000 (23:33 -0700)]
[PATCH] UML: tty locking

Ensure current->signal->tty doesn't get freed during log_exec().

Signed-off-by: Alan Cox <alan@redhat.com>
Acked-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: stack usage reduction
Jeff Dike [Tue, 26 Sep 2006 06:33:08 +0000 (23:33 -0700)]
[PATCH] uml: stack usage reduction

The KSTK_* macros used an inordinate amount of stack.  In order to overcome
an impedance mismatch between their interface, which just returns a single
register value, and the interface of get_thread_regs, which took a full
pt_regs, the implementation created an on-stack pt_regs, filled it in, and
returned one field.  do_task_stat calls KSTK_* twice, resulting in two
local pt_regs, blowing out the stack.

This patch changes the interface (and name) of get_thread_regs to just
return a single register from a jmp_buf.

The include of archsetjmp.h" in registers.h to get the definition of
jmp_buf exposed a bogus include of <setjmp.h> in start_up.c.  <setjmp.h>
shouldn't be used anywhere any more since UML uses the klibc
setjmp/longjmp.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: clean our set_ether_mac
Paolo 'Blaisorblade' Giarrusso [Tue, 26 Sep 2006 06:33:07 +0000 (23:33 -0700)]
[PATCH] uml: clean our set_ether_mac

Clean set_ether_mac usage.  Maybe could also be removed, but surely it can't
be a global function taking a void* argument.

Signed-off-by: Paolo 'Blaisorblade' Giarrusso <blaisorblade@yahoo.it>
Acked-by: Jeff Dike <jdike@addtoit.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Remove unused variable
Jeff Dike [Tue, 26 Sep 2006 06:33:06 +0000 (23:33 -0700)]
[PATCH] uml: Remove unused variable

timer_irq_inited was useless, so it is removed.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: timer cleanups
Jeff Dike [Tue, 26 Sep 2006 06:33:05 +0000 (23:33 -0700)]
[PATCH] uml: timer cleanups

set_interval returns an error instead of panicing if setitimer fails.  Some of
its callers now check the return.

enable_timer is largely tt-mode-specific, so it is marked as such, and the
only skas-mode caller is made to call set-interval instead.

user_time_init was a no-value-added wrapper around set_interval, so it is
gone.

Since set_interval is now called from kernel code, callers no longer pass
ITIMER_* to it.  Instead, they pass a flag which is converted into ITIMER_REAL
or ITIMER_VIRTUAL.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Move signal handlers to arch code
Jeff Dike [Tue, 26 Sep 2006 06:33:04 +0000 (23:33 -0700)]
[PATCH] uml: Move signal handlers to arch code

Have most signals go through an arch-provided handler which recovers the
sigcontext and then calls a generic handler.  This replaces the
ARCH_GET_SIGCONTEXT macro, which was somewhat fragile.  On x86_64, recovering
%rdx (which holds the sigcontext pointer) must be the first thing that
happens.  sig_handler duly invokes that first, but there is no guarantee that
I can see that instructions won't be reordered such that %rdx is used before
that.  Having the arch provide the handler seems much more robust.

Some signals in some parts of UML require their own handlers - these places
don't call set_handler any more.  They call sigaction or signal themselves.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: SIGIO cleanups
Jeff Dike [Tue, 26 Sep 2006 06:33:04 +0000 (23:33 -0700)]
[PATCH] uml: SIGIO cleanups

- Various cleanups in the sigio code.

- Removed explicit zero-initializations of a few structures.

- Improved some error messages.

- An API change - there was an asymmetry between reactivate_fd calling
  maybe_sigio_broken, which goes through all the machinery of figuring out if
  a file descriptor supports SIGIO and applying the workaround to it if not,
  and deactivate_fd, which just turns off the descriptor.

  This is changed so that only activate_fd calls maybe_sigio_broken, when
  the descriptor is first seen.  reactivate_fd now calls add_sigio_fd, which
  is symmetric with ignore_sigio_fd.

  This removes a recursion which makes a critical section look more critical
  than it really was, obsoleting a big comment to that effect.  This requires
  keeping track of all descriptors which are getting the SIGIO treatment, not
  just the ones being polled at any given moment, so that reactivate_fd,
  through add_sigio_fd, doesn't try to tell the SIGIO thread about descriptors
  it doesn't care about.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Improve SIGBUS diagnostics
Jeff Dike [Tue, 26 Sep 2006 06:33:03 +0000 (23:33 -0700)]
[PATCH] uml: Improve SIGBUS diagnostics

UML can get a SIGBUS anywhere if the tmpfs mount being used for its memory
runs out of space.  This patch adds a printk before the panic to provide a
clue as to what likely went wrong.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Fix handling of failed execs of helpers
Jeff Dike [Tue, 26 Sep 2006 06:33:02 +0000 (23:33 -0700)]
[PATCH] uml: Fix handling of failed execs of helpers

There were some bugs in handling failures to exec helper programs.  errno was
passed back from the child with the wrong sign.  It was also ignored.  In the
case where it mattered, the errno from the (successful) read in the parent was
used instead.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Whitespace fixes
Jeff Dike [Tue, 26 Sep 2006 06:33:01 +0000 (23:33 -0700)]
[PATCH] uml: Whitespace fixes

arch/um/kernel/tlb.c had some pretty serious whitespace problems.  I also
fixed some returns.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Fix stack alignment
Jeff Dike [Tue, 26 Sep 2006 06:33:01 +0000 (23:33 -0700)]
[PATCH] uml: Fix stack alignment

Stack randomization needs to be conditional on the personality allowing it.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Use ARRAY_SIZE more assiduously
Jeff Dike [Tue, 26 Sep 2006 06:33:00 +0000 (23:33 -0700)]
[PATCH] uml: Use ARRAY_SIZE more assiduously

There were a bunch of missed ARRAY_SIZE opportunities.

Also, some formatting fixes in the affected areas of code.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] uml: Use klibc setjmp/longjmp
Jeff Dike [Tue, 26 Sep 2006 06:32:59 +0000 (23:32 -0700)]
[PATCH] uml: Use klibc setjmp/longjmp

This patch adds an implementation of setjmp and longjmp to UML, allowing
access to the inside of a jmpbuf without needing the access macros formerly
provided by libc.

The implementation is stolen from klibc.  I copy the relevant files into
arch/um.  I have another patch which avoids the copying, but requires klibc be
in the tree.

setjmp and longjmp users required some tweaking.  Includes of <setjmp.h> were
removed and includes of the UML longjmp.h were added where necessary.  There
are also replacements of siglongjmp with UML_LONGJMP which I somehow missed
earlier.

Signed-off-by: Jeff Dike <jdike@addtoit.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PM: Add pm_trace switch
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:58 +0000 (23:32 -0700)]
[PATCH] PM: Add pm_trace switch

Add the pm_trace attribute in /sys/power which has to be explicitly set to
one to really enable the "PM tracing" code compiled in when CONFIG_PM_TRACE
is set (which modifies the machine's CMOS clock in unpredictable ways).

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] i386: Detect clock skew during suspend
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:57 +0000 (23:32 -0700)]
[PATCH] i386: Detect clock skew during suspend

Detect the situations in which the time after a resume from disk would be
earlier than the time before the suspend and prevent them from happening on
i386.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: John Stultz <johnstul@us.ibm.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] suspend: make it possible to disable serial console suspend
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:57 +0000 (23:32 -0700)]
[PATCH] suspend: make it possible to disable serial console suspend

Hack uart_suspend_port() and uart_resume_port() so that serial console
ports are not suspended if CONFIG_DISABLE_CONSOLE_SUSPEND is set.

This makes it possible to debug the suspend and resume routines of all
device drivers as well as the lowest-level swsusp code with the help of the
serial console.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] PM: make it possible to disable console suspending
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:56 +0000 (23:32 -0700)]
[PATCH] PM: make it possible to disable console suspending

Change suspend_console() so that it waits for all consoles to flush the
remaining messages and make it possible to switch the console suspending off
with the help of a Kconfig option.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Cc: Stefan Seyfried <seife@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: Use memory bitmaps during resume
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:55 +0000 (23:32 -0700)]
[PATCH] swsusp: Use memory bitmaps during resume

Make swsusp use memory bitmaps to store its internal information during the
resume phase of the suspend-resume cycle.

If the pfns of saveable pages are saved during the suspend phase instead of
the kernel virtual addresses of these pages, we can use them during the resume
phase directly to set the corresponding bits in a memory bitmap.  Then, this
bitmap is used to mark the page frames corresponding to the pages that were
saveable before the suspend (aka "unsafe" page frames).

Next, we allocate as many page frames as needed to store the entire suspend
image and make sure that there will be some extra free "safe" page frames for
the list of PBEs constructed later.  Subsequently, the image is loaded and, if
possible, the data loaded from it are written into their "original" page
frames (ie.  the ones they had occupied before the suspend).

The image data that cannot be written into their "original" page frames are
loaded into "safe" page frames and their "original" kernel virtual addresses,
as well as the addresses of the "safe" pages containing their copies, are
stored in a list of PBEs.  Finally, the list of PBEs is used to copy the
remaining image data into their "original" page frames (this is done
atomically, by the architecture-dependent parts of swsusp).

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: Introduce memory bitmaps
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:54 +0000 (23:32 -0700)]
[PATCH] swsusp: Introduce memory bitmaps

Introduce the memory bitmap data structure and make swsusp use in the suspend
phase.

The current swsusp's internal data structure is not very efficient from the
memory usage point of view, so it seems reasonable to replace it with a data
structure that will require less memory, such as a pair of bitmaps.

The idea is to use bitmaps that may be allocated as sets of individual pages,
so that we can avoid making allocations of order greater than 0.  For this
reason the memory bitmap structure consists of several linked lists of objects
that contain pointers to memory pages with the actual bitmap data.  Still, for
a typical system all of these lists fit in a single page, so it's reasonable
to introduce an additional mechanism allowing us to allocate all of them
efficiently without sacrificing the generality of the design.  This is done
with the help of the chain_allocator structure and associated functions.

We need to use two memory bitmaps during the suspend phase of the
suspend-resume cycle.  One of them is necessary for marking the saveable
pages, and the second is used to mark the pages in which to store the copies
of them (aka image pages).

First, the bitmaps are created and we allocate as many image pages as needed
(the corresponding bits in the second bitmap are set as soon as the pages are
allocated).  Second, the bits corresponding to the saveable pages are set in
the first bitmap and the saveable pages are copied to the image pages.
Finally, the first bitmap is used to save the kernel virtual addresses of the
saveable pages and the second one is used to save the contents of the image
pages.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: Introduce some helpful constants
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:52 +0000 (23:32 -0700)]
[PATCH] swsusp: Introduce some helpful constants

Introduce some constants that hopefully will help improve the readability of
code in kernel/power/snapshot.c.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Change the name of pagedir_nosave
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:52 +0000 (23:32 -0700)]
[PATCH] Change the name of pagedir_nosave

The name of the pagedir_nosave variable does not make sense any more, so it
seems reasonable to change it to something more meaningful.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: clean up suspend header
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:51 +0000 (23:32 -0700)]
[PATCH] swsusp: clean up suspend header

Remove some things that are no longer used or defined elsewhere from suspend.h
and make the inline version of software_suspend() return the right error code.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: Fix alloc_pagedir
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:50 +0000 (23:32 -0700)]
[PATCH] swsusp: Fix alloc_pagedir

Get rid of the FIXME in kernel/power/snapshot.c#alloc_pagedir() and
simplify the functions called by it.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: Reorder memory-allocating functions
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:50 +0000 (23:32 -0700)]
[PATCH] swsusp: Reorder memory-allocating functions

Move some functions in kernel/power/snapshot.c to a better place (in the
same file) and introduce free_image_page() (will be necessary in the
future).

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@suse.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: Fix mark_free_pages
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:49 +0000 (23:32 -0700)]
[PATCH] swsusp: Fix mark_free_pages

Clean up mm/page_alloc.c#mark_free_pages() and make it avoid clearing
PageNosaveFree for PageNosave pages.  This allows us to get rid of an ugly
hack in kernel/power/snapshot.c#copy_data_pages().

Additionally, the page-copying loop in copy_data_pages() is moved to an
inline function.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Disable CPU hotplug during suspend
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:48 +0000 (23:32 -0700)]
[PATCH] Disable CPU hotplug during suspend

The current suspend code has to be run on one CPU, so we use the CPU
hotplug to take the non-boot CPUs offline on SMP machines.  However, we
should also make sure that these CPUs will not be enabled by someone else
after we have disabled them.

The functions disable_nonboot_cpus() and enable_nonboot_cpus() are moved to
kernel/cpu.c, because they now refer to some stuff in there that should
better be static.  Also it's better if disable_nonboot_cpus() returns an
error instead of panicking if something goes wrong, and
enable_nonboot_cpus() has no reason to panic(), because the CPUs may have
been enabled by the userland before it tries to take them online.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] Make swsusp avoid memory holes and reserved memory regions on x86_64
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:46 +0000 (23:32 -0700)]
[PATCH] Make swsusp avoid memory holes and reserved memory regions on x86_64

On x86_64 machines with more than 2 GB of RAM there are large memory gaps
(with no corresponding kernel virtual addresses) and reserved memory
regions between areas of usable physical RAM.  Moreover, if CONFIG_FLATMEM
is set, they appear within the normal zone.  swsusp should not try to save
them, so the corresponding page structs have to be marked as 'nosave'.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Mel Gorman <mel@csn.ul.ie>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: struct snapshot_handle cleanup
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:46 +0000 (23:32 -0700)]
[PATCH] swsusp: struct snapshot_handle cleanup

Add comments describing struct snapshot_handle and its members, change the
confusing name of its member 'page' to 'cur'.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: clean up browsing of pfns
Rafael J. Wysocki [Tue, 26 Sep 2006 06:32:45 +0000 (23:32 -0700)]
[PATCH] swsusp: clean up browsing of pfns

Clean up some loops over pfns for each zone in snapshot.c: reduce the
number of additions to perform, rework detection of saveable pages and make
the code a bit less difficult to understand, hopefully.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Pavel Machek <pavel@ucw.cz>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: read speedup
Andrew Morton [Tue, 26 Sep 2006 06:32:44 +0000 (23:32 -0700)]
[PATCH] swsusp: read speedup

Implement async reads for swsusp resuming.

Crufty old PIII testbox:
15.7 MB/s -> 20.3 MB/s

Sony Vaio:
14.6 MB/s -> 33.3 MB/s

I didn't implement the post-resume bio_set_pages_dirty().  I don't really
understand why resume needs to run set_page_dirty() against these pages.

It might be a worry that this code modifies PG_Uptodate, PG_Error and
PG_Locked against the image pages.  Can this possibly affect the resumed-into
kernel?  Hopefully not, if we're atomically restoring its mem_map?

Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Jens Axboe <axboe@suse.de>
Cc: Laurent Riffard <laurent.riffard@free.fr>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: add read-speed instrumentation
Andrew Morton [Tue, 26 Sep 2006 06:32:43 +0000 (23:32 -0700)]
[PATCH] swsusp: add read-speed instrumentation

Add some instrumentation to the swsusp readin code to show what bandwidth
we're achieving.

Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: write speedup
Andrew Morton [Tue, 26 Sep 2006 06:32:42 +0000 (23:32 -0700)]
[PATCH] swsusp: write speedup

Switch the swsusp writeout code from 4k-at-a-time to 4MB-at-a-time.

Crufty old PIII testbox:
12.9 MB/s -> 20.9 MB/s

Sony Vaio:
14.7 MB/s -> 26.5 MB/s

The implementation is crude.  A better one would use larger BIOs, but wouldn't
gain any performance.

The memcpys will be mostly pipelined with the IO and basically come for free.

The ENOMEM path has not been tested.  It should be.

Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
18 years ago[PATCH] swsusp: add write-speed instrumentation
Andrew Morton [Tue, 26 Sep 2006 06:32:41 +0000 (23:32 -0700)]
[PATCH] swsusp: add write-speed instrumentation

Add some instrumentation to the swsusp writeout code to show what bandwidth
we're achieving.

Cc: Pavel Machek <pavel@ucw.cz>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>