Following the CPUID extensions and APIC-ID setup presented by Richard
Brunner at Kernel Summit.

Without this the kernel's node discovery would assign the wrong nodes on
dual core systems and cause performance degradation.

Only handles power of two cored CPUs, others will be ignored.

/proc/cpuinfo has a new field cpu cores: <number of CPU cores> and also
show the other core as HT sibling (this is needed to make old CPU based
license managers to count one CPU package as only one physical CPU)

FIXME this currently misses the disable code for SMT idle.  That's because
2.6.9rc1-bk6 removed the arch hooks need for that.  Needs to be readded
later.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This patch contains - Removes __initdata from lmb definition (struct lmb
lmb;) and modified the existing page_is_ram function.

Also changed the current argument from physical address to pfn to make it
compatible across architectures.  Please review them and send me your
comments/suggestions.  If you are OK with any one patch, please include it
in the mainline kernel.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Wrap a lot more of processor.h in #ifdef __KERNEL__ / __ASSEMBLY__.  Remove
now unised EXC_FRAME_SIZE.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Instead of statically allocating the NUMA node structures, do it at runtime
with node local memory where possible.  With NR_CPUS=128 we had over 800kB
of memory allocated before this patch, now we allocate only as many nodes
as the machine requires.

We have to do some fancy footwork in do_init_bootmem since we use both the
LMB allocator and the bootmem allocator at the same time.  The problem has
always been there although I only recently found a test case that hit it.
Wrap that logic in careful_allocation which uses the LMB allocator then
falls back to the bootmem allocator if required.

Now alloc_bootmem on a node with no memory doesnt panic, we dont need the
hack in paging_init so remove it.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

On some machines we would print hundreds of lines of NUMA debug output. 
The following patch cleans it up so we only print a summary of cpus and
memory vs nodes:

Node 0 CPUs: 0-1
Node 1 CPUs: 16-17
Node 2 CPUs: 32-33
Node 3 CPUs: 48-49
Node 0 Memory: 0x0-0x400000000
Node 1 Memory: 0x400000000-0x800000000

I lifted the code to do this out of xmon.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Update iSeries_defconfig.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Update pSeries_defconfig.  This config should better cover POWER5 systems
with SPLPAR spinlocks, IPR SCSI, NR_CPUS = 128, NUMA, HVCS, and irq stacks
enabled, VSCSI and VETH as modules.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Now we set up cpu_possible_map early we can dynamically allocate the
emergency stacks.  Previously we would allocate NR_CPUS * PAGE_SIZE,
potentially wasting quite a lot of memory.

Clean a comment up in irqstack_early_init and use unsigned int for cpu ids
while in the area.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This patch uses the firmware-defined error log buffer length for calls to
the firmware routine 'check-exception'.  It also simplifies code in rtasd.c
that is attempting to obtain the error log length.
Signed-off-by: Linas Vepstas <linas@linas.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Implement the HCALLs to do more than one TCE setup or invalidation at a
time on pSeries LPAR.  Previous implementation did one hypervisor call per
setup or teardown, resulting in significant overhead.

A simple test of "time dd if=/dev/sda of=/dev/null bs=128k" shows the
amount of system time go down by about 5% by using the multi-tce calls.
Signed-off-by: Olof Johansson <olof@austin.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

To be able to use the cur_cpu_spec->firmware_features values in early
setup, the function initializing them has to be moved up to before the
early init calls.
Signed-off-by: Olof Johansson <olof@austin.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

ppc64 has constraints on the mixing of small and large pages, such that
they cannot be mixed in the same 256Mb segment.  This patch provides an
architecture specific arch_get_unmapped_area_topdown() which implements
these contraints.
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Recent patches introduced a top down user process address space allocation
policy; further patches enable this for ppc64.  Although these work
correctly for normal maps, the topdown algorithm does not take into account
stringent mixing constraints for small and large pages on this
architecture.  These patches introduce a ppc64 specific
arch_get_unused_area_topdown() variant.  The first introduces
infrastructure to allow replacement of the generic
arch_get_unused_area_topdown() and the second utilises this infrastructure.

In this patch I have followed the pattern set by the arch_get_unused_area()
using HAVE_ARCH_UNMAPPED_AREA_TOPDOWN to be consistent.  However, it would
also be possible to simply have a ppc64_get_unused_area_topdown() in the
arch/ppc64/mm/mmap.c or to use weak bindings.

This patch:

Allow an architecture to override the default definition of
arch_get_unmapped_area_topdown().
Signed-off-by: Andy Whitcroft <apw@shadowen.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

We should print a stack backtrace when we get EEH errors, it makes
debugging the cause of the fail easier.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This patch adds explicit checking for EEH slot isolation events into the
PCI config space read path.  The change itself would have been minor,
except that pci config reads don't have a pointer to a struct pci_dev.
Thus, I had to restructure the eeh code to accomodate this, which seems to
be a good thing anyway, making it a tad cleaner.
Signed-off-by: Linas Vepstas <linas@linas.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

DEBUG_SPINLOCK_SLEEP is useful on ppc64, so allow it to be selected.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

This has been given basic testing on Power4 pSeries and iSeries machines.

At present, the SLB miss handler has to check the SLB slot it is about to
use to ensure that it does not contain the SLBE for the current kernel
stack - throwing out the SLBE for the kernel stack can trigger the
"megabug": we have no SLBE for the stack, but we don't fault immediately
because we have an ERAT entry for it.  The ERAT entry is then lost due to a
tlbie on another CPU during the unrecoverable section of the exception
return path.

This patch implements a different approach - with this patch SLB slot 2
always (well, nearly always) contains an SLBE for the stack.  This slot is
never cast out by the normal SLB miss path.  On context switch, an SLBE for
the new stack is pinned into this slot, unless the new stack is the the
bolted segment.

For iSeries we need a special workaround because there is no way of
ensuring the stack SLBE is preserved an a shared processor switch.  So, we
still need to handle taking an SLB miss on the stack, in which case we must
make sure it is loaded into slot 2, rather than using the normal
round-robin.

This approach shaves a few ns off the slb miss time (on pSeries), but more
importantly makes it easier to experiment with different SLB castout
aporoaches without worrying about reinstating the megabug.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

 ... but a better solution should probably be found, that's saved_auxv
in struct mm_struct is ... hrm... disgusting at least ;)

So the problem is that we currently do

	unsigned long saved_auxv[40];

That is we assume that all archs will have an AUXV entry of no more than
2 unsigned longs and we'll have no more than 20 entries in the table.

Unfortunately, this is wrong.

PPC/PPC64 adds 5 platform entries. then we add 14 in the common code,
and then we have

	if (k_platform) {
		NEW_AUX_ENT(AT_PLATFORM, (elf_addr_t)(unsigned long)u_platform);
	}
	if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
		NEW_AUX_ENT(AT_EXECFD, (elf_addr_t) bprm->interp_data);
	}

which may add a 20th entry... which means we are already out of room for
the terminating AT_NULL entry... bad bad bad.

Now I'm adding AT_SYSINFO_EHDR to ppc/ppc64 and it breaks all the time.

The quick fix is to bump the number of entries in struct mm_struct to 42,
though that's worth a big BOLD comment in binfmt_elf in case anybody adds
a new entry in there. (Patch below). I wonder if we shouldn't add a
ARCH_MAX_AUXV_ENTRIES macro to the asm-arch/elf.h though to get the max
amount of entry the arch may add to the normal count, and have the mm
struct array use that plus the number of "normal" entries...

This patch increase the max amount of auxv entries in struct mm_struct to
42 so that ppc & ppc64 don't get a chance to blow the array up.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

drivers/scsi/ipr.c: In function `ipr_worker_thread':
drivers/scsi/ipr.c:1825: wrong type argument to unary exclamation mark
drivers/scsi/ipr.c: In function `ipr_read_dump':
drivers/scsi/ipr.c:2426: wrong type argument to unary exclamation mark

Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Brian King <brking@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

I've been hunting down a bug affecting IBM x440/x445 systems where the
floppy driver would get spurious interrupts and would not initialize
properly.

After digging James Cleverdon and Bill Irwin pointed out that target_cpus()
is routing the interrupts to the clustered apic broadcast mask using
destFixed deliver mode.  This was causing multiple interrupts to show up,
breaking the floppy init code.

This fix simply changes the delivery mode to dest_LowestPrio and initially
routes interrupts to the first cpu to resolve this issue.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

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

An attempt to fix the random poking in legacy io space...

ask for permissions before poking at io ports.  this affects
detect_and_report_winbond() and detect_and_report_smsc().

Its unclear what port number is returned in show_parconfig_winbond(), so
just request the region as well and free the probing region.
Signed-off-by: Olaf Hering <olh@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Page allocation failures are expected when add_to_swap() tries to allocate
radix-tree nodes without PF_MEMALLOC.  Kill the __alloc_pages() warnings.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

sound/core/pcm_native.c was assuming that all architectures implement
io_remap_page_range() in the same way.  They don't, so the sparc64 build
broke.

mmap callback is added to PCM ops.  The driver can implement its own mmap
method.

Also, mmapping of the status and control records is available only on known
coherent architectures now.
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Fix the build with CONFIG_SHMEM=n
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

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

into ppc970.osdl.org:/home/torvalds/v2.6/linux

into ppc970.osdl.org:/home/torvalds/v2.6/linux

Dmitry Torokhov triggered a problem in the new pidhash macros: These
do_each_task_pid()/while_each_task_pid() do loop forever if you use a
'continue' inside them. 

The end of the loop has to be inside the "while()" statement, so as to
not make "continue" jump over it.

into ppc970.osdl.org:/home/torvalds/v2.6/linux

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9

Patch developed by Ian Wienand under advice from Bill Irwin.
Signed-off-by: William Irwin <wli@holomorphy.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

There is a possible deadlock condition with conntrack/nat-helpers:

CPU1:
conntrack-helper:help: lock(private_lock)
ip_conntrack_expect_related: write_lock(ip_conntrack_lock)

CPU2:
nat-core:do_bindings: read_lock(ip_conntrack_lock)
nat-helper:help: lock(private_lock)

The lock in the nat-helper is unneccessary because the expectation
is never changed and is protected by ip_conntrack_lock.
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: Harald Welte <laforge@netfilter.org>
Signed-off-by: David S. Miller <davem@davemloft.net>