My tiger-4 machine has 16GB memory. Kernel 2.6.8 fails to boot on it
when command line parameter mem=8G, and it also fails when mem being
set to other value, such as 7G, 10G.

Basically, in function efi_memmap_walk, md->num_pages might be decreased
if mem_limit is set, and then at the next time when efi_memmap_walk is
called, trim_top might trim the md again because of IA64_GRANULE_SIZE
alignment, then another md which is beyond mem_limit at the beginning
will be chosen, and its physical page number is larger than max_pfn. Then,
a BUG check is triggered.
Signed-off-by: Zhang Yanmin <yanmin.zhang@intel.com>
Signed-off-by: Yao Jun      <junx.yao@intel.com>
Signed-off-by: Tony Luck    <tony.luck@intel.com>

into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.8.1

Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Update comments in function ia64_switch_mode_virt() to reflect actual
implementation from recent region 5 init_task bug fix.
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Small patch to fix the includes for the cyclone timer.  We just happen to be
getting these if CONFIG_SMP is on, but if it's turned off the build breaks.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Current SGI Altix systems have a NUMAlink domain size of 1024
compute nodes and are fully cache coherent up to 256 compute
nodes (compute nodes are even-numbered).  Systems larger than
256 nodes are partitioned into multiple cache coherent systems.
This patch exports a partition's coherence id to users via the
/proc/sgi_sn/coherence_id file.
Signed-off-by: Dean Roe <roe@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

This patch turns the per-CPU frequency printk into a KERN_DEBUG instead of a
KERN_INFO so it'll show up in the system log but won't be printed at boot,
since it's a big pain on system with a lot of CPUs.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

It's pretty unlikely these page allocations would fail, but we should
still check them.
Signed-off-by: Greg Edwards <edwardsg@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

A recent patch caused a warning about not using expressions as lvalues
to crop up in bte.c (or maybe it's just that I'm using gcc-3.4.1 now).
This patch fixes it by creating a temporary to store the register whose
address we want to get and stuff into the per-bte info structure.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

In working on the patch to export PCI ROM space via sysfs, I found that the
sn2 PCI code doesn't assign a parent resource to any of the PCI device
resources as it builds them.  This provides a simple fix for that problem.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: Andreas Schwab <schwab@suse.de>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

The biggest change is better overflow handling, partly ported from the AMD64 
IOMMU  code.  This is important for some 3d drivers who make use of this and can 
map a lot of memory.

First it increases the default memory usage of swiotlb to 64MB.
This was the value used in production systems in 2.4 (e.g. in SLES8) and the default
was far too small.  Note this is only allocated when the swiotlb is needed.

pci_map_single will not panic anymore on an overflow. Instead it returns
a pointer to a scratch area and allows the caller to check using 
dma_mapping_error(). It still panics when the overflow buffer is exceeded.

dma_mapping_error support is implemented.

pci_map_sg will never panic now, but just return 0. Near all callers 
of it seem to have proper error checking for this (IDE will even handle
it without erroring out) 

The handling of passing a NULL device is improved, previously it would 
never map in this case. i386 and other architectures assume NULL means a 4GB
limit instead.   This code does this now too.

I added support for swiotlb=force for easier testing.

Patch supplied by Andi Kleen & Suresh Siddha
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Fix stupid thinkos in the fcntl f_op removal code.

Update lasi 82596 driver:
 - Fix boot messages (Helge Deller)
 - Whitespace and __FILE__ usage (Joel Soete)
 - 64-bit compile warning fixes (James Bottomley)

This trivial fix allows mounting MO-drives readwrite.

Acked by Jens.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

 - __PAGE_OFFSET is 0x10000000 (Randolph Chung)
 - PA8800 support (Grant Grundler)
 - debuglocks (Thibaut Varene)
 - PDC chassis disabling (Thibaut Varene)
 - Distinguish between Dinos in request_irq (Thibaut Varene)
 - Document interrupt registers (Randolph Chung)
 - Revamp CONFIG_DISCONTIGMEM support (Randolph Chung)
 - Remove STI console warning and special casing (Randolph Chung)
 - n4000 defconfig (Randolph Chung)
 - iosapic fixes (Bjorn Helgaas)
 - Fix a bug in entry.S where pa_dbit_lock was being trashed (Randolph Chung)
 - SMP support (Randolph Chung, Grant Grundler, James Bottomley)
 - Clear the pte in the fault handler (Joel Soete)
 - Change _exit prototype (Carlos O'Donell)
 - Better unwinding support (Randolph Chung)
 - GCC 3.4 fixes (Carlos O'Donell, Randolph Chung)

The newly introduced ->fcntl file_operation is badly thought out,
not to mention undocumented.  This patch replaces it with two better
defined operations -- check_flags and dir_notify.  Any other fcntl()s
that filesystems are interested in can have their own properly typed
f_op method when they need it.
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

Patch from Marc Singer

The work already done to rectify the timer code initialization missed
a necessary change.  This patch adds a timer initialization function
to the LPD7A400 machine type.  It also makes a small correction to the
kernel configuration that is out of step with the machine specific
Kconfig.

Patch from Marc Singer

This patch aligns the lh7a40x port with the updated structure of IDE
drivers.  It also removes an unused FIQ handler.  This patch is made
in order to get the lh7a40x port to build.

Patch from Catalin Marinas

This patch replaces the -mshort-load-bytes option with $(call
check_gcc,-malignment-traps,-mshort-load-bytes) in the CFLAGS
setting in arch/arm/Makefile.

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

into kernel.bkbits.net:/home/davem/sparc-2.6

In the *ppos cleanups, proc_dol2crvec was updated, but the prototype
found at the top of kernel/sysctl.h was not, generating warning.  This
corrects the prototype to match the code.

(I'm gonna take a stab at moving these into arch/ppc shortly)
Signed-off-by: Tom Rini <trini@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

We allow more commands if the disk was opened read-write.

They'll need it for permission checking.

Signed-off-by: David S. Miller <davem@redhat.com>

Normal users shouldn't have access to the raw device anyway
unless they are in the trusted "disk" group, but let's require
RAWIO capabilities. That's what the original SCSI interfaces
did anyhoo.

We probably _should_ just require write access, but that will
need more of a code change to pass down the file descriptor.

We must setup the queue before calling elevator_init(), otherwise the
elevator cannot use the queue members. This really hurts cfq since it
checks ->nr_requests, which is zero at that time.
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

CONFIG_SERIAL_8250_MANY_PORTS should not be set for these boards, as
they only have 2 serial ports.
Signed-off-by: Tom Rini <trini@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

With 'len' being unsigned now, we must make sure it never tries
to go negative...

Thanks: griffon26@kfk4ever.com (Maurice van der Pot)
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Eliminates costly atomic bitops in do_sparc64_fault.
Signed-off-by: David S. Miller <davem@redhat.com>