[PATCH] mm: page_alloc: increase size of per-cpu-pages

Increase the page allocator's per-cpu magazines from 1/4MB to 1/2MB.

Over 100+ runs for a workload, the difference in mean is about 2%.  The best
results for both are almost same.  Though the max variation in results with
1/2MB is only 2.2%, whereas with 1/4MB it is 12%.

Signed-off-by: Rohit Seth <rohit.seth@intel.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] swaptoken tuning

It turns out that the original swap token implementation, by Song Jiang, only
enforced the swap token while the task holding the token is handling a page
fault.  This patch approximates that, without adding an additional flag to the
mm_struct, by checking whether the mm->mmap_sem is held for reading, like the
page fault code does.

This patch has the effect of automatically, and gradually, disabling the
enforcement of the swap token when there is little or no paging going on, and
"turning up" the intensity of the swap token code the more the task holding
the token is thrashing.

Thanks to Song Jiang for pointing out this aspect of the token based thrashing
control concept.

The new code shows a slight degradation over the old swap token code, but
still a big win over running without the swap token.

2.6.12+ swap token disabled

$ for i in `seq 10` ; do /usr/bin/time ./qsbench -n 30000000 -p 3 ; done
101.74user 23.13system 8:26.91elapsed 24%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (38597major+430315minor)pagefaults 0swaps
101.98user 24.91system 8:03.06elapsed 26%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (33939major+430457minor)pagefaults 0swaps
101.93user 22.12system 7:34.90elapsed 27%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (33166major+421267minor)pagefaults 0swaps
101.82user 22.38system 8:31.40elapsed 24%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (39338major+433262minor)pagefaults 0swaps

2.6.12+ swap token enabled, timeout 300 seconds

$ for i in `seq 4` ; do /usr/bin/time ./qsbench -n 30000000 -p 3 ; done
102.58user 16.08system 3:41.44elapsed 53%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (19707major+285786minor)pagefaults 0swaps
102.07user 19.56system 4:00.64elapsed 50%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (19012major+299259minor)pagefaults 0swaps
102.64user 18.25system 4:07.31elapsed 48%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (21990major+304831minor)pagefaults 0swaps
101.39user 19.41system 5:15.81elapsed 38%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (24850major+323321minor)pagefaults 0swaps

2.6.12+ with new swap token code, timeout 300 seconds

$ for i in `seq 4` ; do /usr/bin/time ./qsbench -n 30000000 -p 3 ; done
101.87user 24.66system 5:53.20elapsed 35%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (26848major+363497minor)pagefaults 0swaps
102.83user 19.95system 4:17.25elapsed 47%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (19946major+305722minor)pagefaults 0swaps
102.09user 19.46system 5:12.57elapsed 38%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (25461major+334994minor)pagefaults 0swaps
101.67user 20.61system 4:52.97elapsed 41%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (22190major+329508minor)pagefaults 0swaps

Signed-off-by: Rik Van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] add sem_is_read/write_locked()

Add sem_is_read/write_locked functions to the read/write semaphores, along the
same lines of the *_is_locked spinlock functions.  The swap token tuning patch
uses sem_is_read_locked; sem_is_write_locked is added for completeness.

Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] fix alpha breakage

barrier.h uses barrier() in non-SMP case.  And doesn't include compiler.h.

Cc: Al Viro <viro@ftp.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] TIMERS: add missing compensation for HZ == 250

Add missing compensation for (HZ == 250) != (1 << SHIFT_HZ) in
second_overflow().

Signed-off-by: YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org>
Cc: "David S. Miller" <davem@davemloft.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

[PATCH] vmalloc_node

This patch adds

vmalloc_node(size, node) -> Allocate necessary memory on the specified node

and

get_vm_area_node(size, flags, node)

and the other functions that it depends on.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Merge master.kernel.org:/home/rmk/linux-2.6-arm

[ARM] 3061/1: cleanup the XIP link address mess

Patch from Nicolas Pitre

Since vmlinux.lds.S is preprocessed, we can use the defines already
present in asm/memory.h (allowed by patch #3060) for the XIP kernel link
address instead of relying on a duplicated Makefile hardcoded value, and
also get rid of its dependency on awk to handle it at the same time.

While at it let's clean XIP stuff even further and make things clearer
in head.S with a nice code reduction.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

[ARM] 3060/1: allow constants found in asm/memory.h to be used in asm code

Patch from Nicolas Pitre

This patch allows for assorted type of cleanups by letting assembly code
use the same set of defines for constant values and avoid duplicated
definitions that might not always be in sync, or that might simply be
confusing due to the different names for the same thing.

Signed-off-by: Nicolas Pitre <nico@cam.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

Merge branch 'upstream' of git://ftp.linux-mips.org/pub/scm/upstream-linus

Update MIPS defconfig files.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

prom_free_prom_memory() returns unsigned long

    Some boards declare prom_free_prom_memory as a void function but the
    caller free_initmem() expects a return value.

    Fix those up and return 0 instead, just like everyone else does.

Signed-off-by: Arthur Othieno <a.othieno@bluewin.ch>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Get rid of SINGLE_ONLY_FPU.  Linux does not support half FPU other than
by emulation of a full FPU.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix all the get_user / put_user related sparse warnings.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Delete unused ieee754_cname[] and declaration.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Include for prototypes.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Protect against multiple inclusion.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Remove useless casts of kmalloc return values.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Hack to resolve longstanding prefetch issue

Prefetching may be fatal on some systems if we're prefetching beyond the
end of memory on some systems.  It's also a seriously bad idea on non
dma-coherent systems.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

More foolproofing of the CPU configuration.

Limit the number of cpu type options in the cpu menu to just those
types that are actually available for the select platform.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

pci-expmem-hack

CFE 1.2.5 and earlier fails to turn on the ExpMemEn bit in the
PCIFeatureControl register, which means that DMA does not work
beyond physical address 01_0000_0000, ergo to DRAM beyond 1GB.

With ExpMemEn turned on, 01_0000_0000-0f_ffff_ffff is mapped,
so DMA works for up to 61 GB of DRAM.

Will be fixed in CFE 1.2.6 (yet to be released).

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

BCM1480 HT support

PCI support code for PLX 7250 PCI-X tunnel on BCM91480B BigSur board.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Support for the BCM1480 on-chip PCI-X bridge.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

SB1 cache exception handling.

Expand SB1 cache error handling by adding SB1_CEX_ALWAYS_FATAL and
SB1_CEX_STALL, allowing configurable behavior on cache errors.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Support for BigSur board.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Add support for BCM1480 family of chips.

 - Kconfig and Makefile changes
 - arch/mips/sibyte/bcm1480/
 - changes to sibyte common code to support 1480

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Add support for SB1A CPU.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Sibyte header cleanup

Update sibyte headers to match Broadcom internal copies:
 - comment cleanup and updates
 - fix LittleSur part number to match the board silkscreen

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

BCM1480 headers

Add header files for BCM1480/1280/1455/1255 family of chips, and
update sb1250 headers which are shared by BCM1480 family.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
diff --git a/include/asm-mips/sibyte/bcm1480_int.h b/include/asm-mips/sibyte/bcm1480_int.h
new file mode 100644

Sibyte fixes

Fix typo in cpu_probe_sibyte.

Signed-Off-By: Andy Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Make UL what should be UL.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix zero length sys_cacheflush

Cacheflush(0, 0, 0) was crashing the system.  This is because
flush_icache_range(start, end) tries to flushing whole address space
(0 - ~0UL) if both start and end are zero.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Get 64-bit right in the kgdb stub.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Sys_lookup_dcookie arguments occupy 4 argument slots.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

FPU emulator garbage collection.

First argument of fpu_emulator_cop1Handler() was unused.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Don't print file name and line in die and die_if_kernel.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Rename page argument of flush_cache_page to something more descriptive.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Define EOWNERDEAD and ENOTRECOVERABLE.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Sliceup Kconfig; it's grown too large.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

More configcheck fixes.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

2.6.14-rc1 updates for MIPS compat types.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Complete the fcntl.h cleanup.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Cleanup Sibyte Kconfig a bit further.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Date:   Fri Jan 14 03:03:23 2005 +0000

Locking cleanups.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix weirdness in <asm/bug.h>

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix wrong comment.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fixup a few lose ends in explicit support for MIPS R1/R2.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Document the meaning of the CPU_MIPS32, CPU_MIPS64, CPU_MIPSR1 and
CPU_MIPSR2.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Protect manipulation of c0_status against preemption and multithreading.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Detect 4KSD and treat it like 4KSc.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

We're no longer hosted on oss for ages ...

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Convert the remaining SPIN_LOCK_UNLOCKED instances to DEFINE_SPINLOCK.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Define and initialize kdb_lock using DEFINE_SPINLOCK.
Convert kgdb_cpulock into a raw_spinlock_t.

SPIN_LOCK_UNLOCKED is deprecated and it's replacement DEFINE_SPINLOCK is
not suitable for arrays of spinlocks.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Make kgdb_wait static.

Nothing outside gdb-stub.c uses kgdb_wait, so change it's definition to
static.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Don't copy SB1 cache error handler to uncached memory.

This may have made sense on a paranoid day with pass 1 BCM1250 processors
that were throwing cache error exception left and right for no good
reason.  On modern silicion that hardly makes sense and the code had
gotten just an obscurity ...

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Provide 64-bit address space definitions for the Sibyte SB1 CPU core.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

No need to explicitly call __read_64bit_c0_split; __read_64bit_c0_register
will do that itself iff needed.  Fix format string.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix stale comment in c-sb1.c.

Signed-Off-By: Andrew Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Cleanup the mess in cpu_cache_init.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Use cpumask_t rather than hand-rolled bitmask code in sb1250_set_affinity.

Signed-Off-By: Andrew Isaacson <adi@broadcom.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Use R4000 TLB routines for SB1 also.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix build error caused by missmatching duplicate declaration.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Don't call memset to clean irq_desc; these data fields have already
previously been initialized statically in kernel/irq/handle.c.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix ARCH_KMALLOC_MINALIGN values on MIPS

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Support for MIPSsim, the cycle accurate MIPS simulator.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Switch Sibyte profiling driver to ->compat_ioctl

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Revise MIPS 64-bit ptrace interface

Change the N32 debugging ABI to something more sane, and add support
for o32 and n32 debuggers to trace n64 programs.

Signed-off-by: Daniel Jacobowitz <dan@codesourcery.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Sync c-tx39.c with c-r4k.c.

tx39_flush_cache_range() does nothing if !cpu_has_dc_aliases.  It should
flush d-cache and invalidate i-cache since the TX39(H2) has separate I/D
cache.

Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Glue again after removal of BUILD_BUG().

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Let it compile again after i2c algo id removal. Sort headers.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

SMP on Malta needs to define ARCH_HAS_IRQ_PER_CPU since 2.6.14-rc1.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
diff --git a/include/asm-mips/mach-mips/irq.h b/include/asm-mips/mach-mips/irq.h
new file mode 100644

smc91x platform support; requires patch to smc91x.h which was sent
upstream.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Au1[12]00 mmc driver. Only tested on the Au1200 at this point though
it should work on the Au1100 as well. Updated defconfig to include driver.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Add SOCK_DCCP definition for MIPS also.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

The type of sum in csum_tcpudp_nofold is "unsigned int", so when we assign
to it in an asm() block, and we're running on a system with 64-bit
registers, it is vitally important that we sign extend it correctly before
returning to C.  Otherwise the stray high bits will be preserved into
csum_fold, and on the SB-1 processor, 32-bit arithmetic on a non
sign-extended register will yield surprising results.

This caused incorrect checksums in some UDP packets for NFS root.  The
problem was mild when using a 10.0.1.x IP address, but severe when
using 192.168.1.x.

Signed-off-by: Daniel Jacobowitz <dan@codesourcery.com>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Kernel gpio/2 routines that will be used by some drivers.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix a cache snooping issue.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Updated pcmcia driver with pb1200 and db1200 support.
Updated db1200_defconfig so pcmcia is enabled by default.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

- moved platform structure to platform.c
- fixed an iounmap warning
- export fixup_xx, needed by the module

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Fix excessive signal latencies.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Reorder & reformat a bit.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Use ARCH_MAY_HAVE_PC_FDC where needed.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

GCC bulletproofing: __mips64 is only defined when compiling for 64-bit
processors.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

The values for SO_SNDBUFFORCE / SO_RCVBUFFORCE were already taken ...

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Futexes for MIPS, for the time being only the R10000_LLSC_WAR version.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Define pcibus_to_node() for IP27.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Cleaned up AMD Au1200 IDE driver:
- converted to platform bus
- removed pci dependencies
- removed virt_to_phys/phys_to_virt calls

System now can root off of a disk.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
diff --git a/Documentation/mips/AU1xxx_IDE.README b/Documentation/mips/AU1xxx_IDE.README
new file mode 100644

Misc au1200 updates.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Delete the SABLE_RTL case.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

When no yamon command line is passed to the kernel, preserve the default
compiled in command line.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Provide MODULE_ARCH_VERMAGIC for MIPS.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Add/Fix missing bit of R4600 hit cacheop workaround.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Minor code cleanup.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

R4600 v2.0 needs a nop before tlbp.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Don't set up a sg dma address if we have no page address for some reason.
Code cleanup.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Move MIPS Technologies processor IDs to where they belong.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

More .set push/pop.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Move genrtc.c's functions into <asm/rtc.h>

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Let r4600 PRID detection match only legacy CPUs, cleanups.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

Define some more common ip22 CPU features.

Signed-off-by: Thiemo Seufer <ths@networkno.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>