This patch cleans up the per-CPU MCA mess with the following changes
(and yields a UP kernel that actually boots again):

 - In percpu.h, make per_cpu_init() a function-call even for the
   UP case.
 - In contig.c, enable per_cpu_init() even for UP since we need to
   allocate the per-CPU MCA data in that case as well.
 - Move the MCA-related stuff out of the cpuinfo structure into
   per-CPU variables defined by mca.c.
 - Rename IA64_KR_PA_CPU_INFO to IA64_KR_PER_CPU_DATA, since it really
   is a per-CPU pointer now.
 - In mca.h, move IA64_MCA_STACK_SIZE early enough so it gets defined
   for assembly-code, too.  Tidy up struct ia64_mca_struct.  Add declaration
   of ia64_mca_cpu_init().
 - In mca_asm.[hS], replace various GET_*() macros with a single
   GET_PERCPU_ADDR() which loads the physical address of an
   arbitrary per-CPU variable.  Remove all dependencies on the
   layout of the cpuinfo structure.  Replace hardcoded stack-size
   with IA64_MCA_STACK_SIZE constant.  Replace hardcoded references
   to ar.k3 with IA64_KR(PER_CPU_DATA).
 - In setup.c:cpu_init(), initialize ar.k3 to be the physical equivalent
   of the per-CPU data pointer.
 - Nuke silly ia64_mca_cpu_t typedef and just use struct ia64_mca_cpu instead.
 - Move __per_cpu_mca[] from setup.c to mca.c.
 - Rename set_mca_pointer() to ia64_mca_cpu_init() and sanitize it.
 - Rename efi.c:pal_code_memdesc() to efi_get_pal_addr() and make it
   return the PAL address, rather than a memory-descriptor.
 - Make efi_map_pal_code() use efi_get_pal_addr().
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

A few reports of illegal instruction panics while trying to boot
were tracked to this.  Fix by David Mosberger.
Signed-off-by: Tony Luck <tony.luck@intel.com>

Spotted by Andreas Schwab, fix from Matthew Wilcox
and David Mosberger.
Signed-off-by: Tony Luck <tony.luck@intel.com>

I noticed that the PTRACE_POKEUSR code incorrectly clears bits 56-58
of _all_ debug registers.  The intention was to only clear it for
odd-numbered registers, to ensure that user-level can only set
user-level data/instruction-breakpoints.  Patch below fixes this problem.
The patch also replaces explicit clearing of the single-step and
taken-branch PSR bits with a call to ptrace_disable() for PTRACE_KILL.
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

This patch replaces the idiom:

	func (args..., long stack) {
		struct pt_regs *regs = (struct pt_regs *) &stack;

with the more commonly used:

	func (args..., struct pt_regs regs) {

The latter didn't used to work with the very earliest kernels and
compilers (anybody remember egcs?) but gcc-3.3 and probably even
gcc-2.96 don't have a problem with it anymore.

The change also makes sparse happier, since it doesn't like it when
you access memory past the end of the declared size of that variable.
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

David thinks this might make Jesse and Willy happy (or
at least happier).  If they can cope with line breaks
before a binary operator, rather than after, then maybe
it will :-)
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

This version doesn't cost us any extra cycles.
Signed-off-by: Ken Chen <kenneth.w.chen@intel.com>
Signed-off-by: Rohit Seth <rohit.seth@intel.com>
Acked-by: David Mosberger <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Anton Blanchard wrote:
>Your recent patch looks to break module kallsyms lookups....
>It looks like if CONFIG_KALLSYMS_ALL is set then we never look up module
>addresses.

Separate lookups for kernel and modules when CONFIG_KALLSYMS_ALL=y.
Signed-off-by: Keith Owens <kaos@ocs.com.au>
Acked-by: Chris Wedgwood <cw@f00f.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Fix warning:

In file included from include/asm/numa.h:5,
                 from arch/x86_64/kernel/setup64.c:27:
include/asm/numnodes.h:6:1: warning: "NODES_SHIFT" redefined
In file included from include/linux/mmzone.h:13,
                 from include/linux/gfp.h:4,
                 from include/linux/slab.h:15,
                 from include/linux/percpu.h:4,
                 from include/linux/sched.h:33,
                 from arch/x86_64/kernel/setup64.c:11:
include/linux/numa.h:11:1: warning: this is the location of the previous definition

in UP builds.
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

From Terence Ripperda <tripperda@nvidia.com>

When doing iounmap don't try to change_page_attr back the guard
page that ioremap added.

Since the last round of change_page_attr changes this would
trigger an BUG because the reference count on the changed pages
wouldn't match up.

The problem would be only visible on machines with >3GB of memory,
because only then the PCI memory hole is below end_pfn and
change_page_attr is used.

Fixed for both i386 and x86-64.

This was actually discovered&fixed by Andrea earlier, but I goofed up
while doing the last ioremap fixes merge and this change got lost.
Poor Terence had to debug it again. Sorry about that.

cc: andrea@suse.de
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Undo bogus change that was introduced with kprobes.  It's not
really needed and it breaks some user applications because
it changes the signal for int 3 from SIGTRAP to SIGSEGV.

Cc: <prasanna@in.ibm.com>
Signed-off-by: Andi Kleen <ak@suse.de>
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

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

- If dst/src are equal, memcpy can be used.
- Eliminate register writes which were unused
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Tony Luck <tony.luck@intel.com>

Patch from Christoph Hellwig to:
- irq_desc and irq_to_vector machvecs.  SN2 has it's own versions,
  but they're the same as the generic ones
- kill do do_IRQ and use __do_IRQ directly everywhere
- kill dead X86 ifdefs
- move some variable declarations around in irq.c to recuce # of ifdefs
Signed-off-by: Tony Luck <tony.luck@intel.com>

When I converted the sn2 code over to the new DMA API, I left the old routines
in place and added wrappers to call them from the generic DMA API functions.
This added an unnecessary level of obfuscation since the generic ia64 code
calls those functions when any of the old style PCI DMA API functions are
called.  This patch rectifies the problem making the code much easier to
understand and hopefully a little more efficient (though I'm sure gcc was
already inlining things pretty well, there were a bunch of unnecessary checks
that I took this opportunity to remove).  It also shrinks the size of the sn2
pci_dma.c quite a bit.
 
  pci_dma.c |  480 +++++++++++++++++++-----------------------------------------
  1 files changed, 151 insertions(+), 329 deletions(-)
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

sn2 does early initialization of the SAL so it can use it for early console
support.  Unfortunately, the loop to find the SAL entry point was buggy so
when we tried out new EFI and SAL system table layouts, the loop didn't
terminate.  Here's the fix (doh!, use two different loop counters instead of
one and just return if we find the SAL entry point).
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Ok, here you go Tony.  This one fixes the loop and also fixes drm_vm.c.  All
of the bits aside from the efi.h bit are ia64 specific (either under
arch/ia64 or __ia64__), so your tree is probably the right place for all of
it.
 
This patch adds efi_range_is_wc() to efi.h.  It's used to determine whether an
address range can be mapped with the write coalescing attribute.  It also
fixes up some ia64 specific callers to use the new routine instead of
unconditionally calling pgprot_writecombined, which can be dangerous if used
on ranges that don't support it.
Signed-off-by: Jesse Barnes <jbarnes@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

The following patch fixes the ia64_pal_prefetch_visibility function to
take a transaction type argument for either virtual or physical memory
as specified in the System Architechture Manual page 2:358.
Signed-Off-By: Jes Sorensen <jes@trained-monkey.org>
Signed-Off-By: Tony Luck <tony.luck@intel.com>

Problem:
	There exists a case where we stop monitoring, i.e. clear
	psr.pp/dcr.pp, via IPI. This is when the stop is triggered
	by a close(), either explicit in the application or implicit
	via exit_files(). The IPI is necessary because at the time the
	thread (controlling the context) issues a close() it may not run
	on the CPU the context is bound to. Yet the call must succeed,
	hence we need to propagate the call to the right CPU.
	But what is the problem then?
	Under IPI, we invoke a perfmon routine which clear the kernel
	(live) kernel psr.pp bit and also dcr.pp. Then we return from
	the function and execute the kernel exit path which restores
	the interrupted state. Unfortunately, this restores the kernel
	psr from ipsr which now contains a stale value. Therefore
	monitoring in the kernel will be active even though we stopped it.
	You cannot modify the "global" psr in an interrupt routine because
	it will be systematically restored on the way back.
	
Solution:
	We need to patch ipsr.pp in the kernel exit path to reflect the
	kernel value of the kernel psr.pp bit. This must be done only when
	returning to kernel. 

	The proposed patch does patch ipsr.pp such that it is identical
	to psr.pp. The patch is subtle because the exit path does not have
	a lot of free registers and also because we need to schedule for
	a psr read. I had to shuffle  things around a little bit.

	The patch is important because there will be another situation where
	this problem can occur once we incorporate the support for event set
	and multiplexing. In this configuration, you may be in the middle of
	the idle loop and on a timer interrupt, you may stop monitoring.
	Slightly different condition, yet same problem with ipsr.pp vs. psr.pp.


Changelog:
	- update kernel exit path when returning to kernel to copy
	psr.pp to ipsr.pp. This is necesary to ensure that if psr.pp
	was modified during the kernel entry, the change is propagated
	the the psr.pp of the of the interrupted thread. Psr.pp can
	be modified as a consequence of an IPI under certain conditions,
	such as when a system-wide context is closed from a remote CPU.

Special thanks to David for reworking the patch to fit
into the enhanced exit path.
signed-off-by: stephane eranian <eranian@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Patch from yanmin.zhang@intel.com to fix up some corner cases
in ptrace.  Many thanks to davidm for reviewing and improving.
Signed-off-by: Tony Luck <tony.luck@intel.com>

Label is unused, and so the compiler generates a warning.
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Use srlz.d instead of srlz.i.  Safe because we don't care whether
the VHPT walker sees the clearing of PSR.ic (if it does, that's fine
if it doesn't, it's OK too since the kernel-text is pinned anyhow).
Good for another 11+ cycles in (normal) getpid().
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Optimize ia64_leave_syscall() a bit better for McKinley-type cores.
The patch looks big, but that's mostly due to renaming r16/r17 to r2/r3.
Good for a 13 cycle improvement.
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Trivial patch: align rse_clear_invalid to 32-byte boundary.  Good for
a 9 cycle speed up.
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Kernel-threads had both pUStk and pKStk set to FALSE, which was
unintentional.  I don't think the bug has shown any ill effects, but
it's clearly wrong and could come around to bite us later, so let's
fix it now.  Depends on the previous patch to clean up C usage of the
global/root-function predicates.
Signed-off-by: Tony Luck <tony.luck@intel.com>

 
The patch below is purely a cleanup but it's a prerequisite for the
next bug fix patch.
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Use msleep() instead of schedule_timeout() to guarantee the task
delays as expected.
Signed-off-by: Nishanth Aravamudan <nacc@us.ibm.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

The fixed sized array of pages that are isolated because of 2xECC
memory errors can run out.  Increasing the size of the array is a
band-aid measure, the real fix will require changes to generic code
to add some bits to page_flags so that pages with errors can be
marked so as to prevent them ever being examined.
Signed-off-by: Russ Anderson <rja@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

Signed-off-by: Keith Owens <kaos@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

When compiling for non-NUMA variable "node" ended up not being used,
which solicits a warning from GCC.  Fix is to evluate cpu_to_node(i)
in place.  This has the effect of doing cpu_to_node(i) twice on NUMA,
but this is init code, so performance is not an issue (and even if
it were, you can just declare cpu_to_node() as being a pure function,
so the compiler can eliminate the second call).
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

I was getting a lot of spurious warnings with "make C=1", due to a
missing "__user" attribute.  I'm not sure whether this got lost or
whether earlier versions of sparse failed to warn about it, but it's
clearly needed (and the i386 version of uaccess.h does the same).
Signed-off-by: David Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

It seems that in ia64_elf32_init, instead of calling return, if we insert
an overlapping vma, we should instead BUG(). We should never get into this
code path, because the vma's are set above PAGE_OFFSET, and thus a
Xmalicious user can not trigger this code path. This change is being
suggested mainly for clarity. Thanks to Stephen Tweedie for pointing out
that returning early in ia64_elf32_init could have unpredictable results.
Signed-off-by: Tony Luck <tony.luck@intel.com>