x86: convert CPA users to the new set_page_ API

This patch converts various users of change_page_attr() to the new,
more intent driven set_page_*/set_memory_* API set.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: a new API for drivers/etc to control cache and other page attributes

Right now, if drivers or other code want to change, say, a cache attribute of a
page, the only API they have is change_page_attr(). c-p-a is a really bad API
for this, because it forces the caller to know *ALL* the attributes he wants
for the page, not just the 1 thing he wants to change. So code that wants to
set a page uncachable, needs to be aware of the NX status as well etc etc etc.

This patch introduces a set of new APIs for this, set_pages_<attr> and
set_memory_<attr>, that offer a logical change to the user, and leave all
attributes not implied by the requested logical change alone.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: cpa: move clflush_cache_range()

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: unify ioremap_32 and _64

Unify the now identical ioremap_32.c and ioremap_64.c into the
same ioremap.c file. No code changed.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: unify ioremap

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: use remove_vm_are in ioremap_32 error path

When ioremap_page_range fails, then we can use remove_vm_area instead
of vunmap safely.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: __iomem annotations

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: switch to change_page_attr_addr in ioremap_32.c

Use change_page_attr_addr() instead of change_page_attr(), which
simplifies the code significantly and matches the 64bit
implementation.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: make c_p_a unconditional in ioremap

Make c_p_a unconditional for ioremap and iounmap. This ensures
complete consistency of the flags which are handed to
ioremap_page_range and the real flags in the mappings.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: introduce max_pfn_mapped

64bit uses end_pfn_map and 32bit uses max_low_pfn. There are several
files which have #ifdef'ed defines which map either to end_pfn_map or
max_low_pfn. Replace this by a universal define and clean up all the
other instances.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: cleanup ioremap includes

Get rid of the douplicate define of ISA_START/END_ADDRESS and use the
same headers in 32 and 64 bit code.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: style cleanup of ioremap code

Fix the coding style before going further.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: fix ioremap pgprot inconsistency

The pgprot flags which are handed into ioremap_page_range() are
different to those which are set in change_page_attr(). The
ioremap_page_range flags are executable, while the c_p_a flags are
not.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: fix ioremap pgprot inconsistency

The pgprot flags which are handed into ioremap_page_range() are
different to those which are set in change_page_attr(). The
ioremap_page_range flags are executable, while the c_p_a flags are
not. Also make the mappings global (which is a NOP currently on 32bit,
although CPUs from PPRO+ onwards support it, but that's a separate
fix.)

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

x86: turn the check_exec function into function that

What the check_exec() function really is trying to do is enforce certain
bits in the pgprot that are required by the x86 architecture, but that
callers might not be aware of (such as NX bit exclusion of the BIOS
area for BIOS based PCI access; it's not uncommon to ioremap the BIOS
region for various purposes and normally ioremap() memory has the NX bit
set).

This patch turns the check_exec() function into static_protections()
which also is now used to make sure the kernel text area remains non-NX
and that the .rodata section remains read-only. If the architecture
ends up requiring more such mandatory prot settings for specific areas,
this is now a reasonable place to add these.

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cpa: make self-test depend on DEBUG_KERNEL

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: ioremap_nocache fix

This patch fixes a bug of ioremap_nocache. ioremap_nocache() will call
__ioremap() with flags != 0 to do the real work, which will call
change_page_attr_addr() if phys_addr + size - 1 < (end_pfn_map << PAGE_SHIFT).
But some pages between 0 ~ end_pfn_map << PAGE_SHIFT are not mapped by
identity map, this will make change_page_attr_addr failed.

This patch is based on latest x86 git and has been tested on x86_64 platform.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add PAGE_KERNEL_EXEC_NOCACHE

add PAGE_KERNEL_EXEC_NOCACHE.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix NX bit handling in change_page_attr()

This patch fixes a bug of change_page_attr/change_page_attr_addr on
Intel i386/x86_64 CPUs.  After changing page attribute to be
executable with these functions, the page remains un-executable on
Intel i386/x86_64 CPU.  Because on Intel i386/x86_64 CPU, only if the
"NX" bits of all three level page tables are cleared (PAE is enabled),
the corresponding page is executable (refer to section 4.13.2 of Intel
64 and IA-32 Architectures Software Developer's Manual).  So, the bug
is fixed through clearing the "NX" bit of PMD when splitting the huge
PMD.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: change cpa to pfn based

change CPA to pfn based.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: keep the BIOS area executable

keep the BIOS area executable.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add PG_LEVEL enum

this way PG_LEVEL_1GB will be an easy change.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up lookup_address() declarations

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up arch/x86/mm/pageattr.c

do some leftover cleanups in the now unified arch/x86/mm/pageattr.c
file.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: re-add clflush_cache_range()

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: unify pageattr_32.c and pageattr_64.c

unify the now perfectly identical pageattr_32/64.c files - no code changed.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: prepare for pageattr.c unification

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: backmerge 64-bit details into 32-bit pageattr.c

backmerge 64-bit details into 32-bit pageattr.c.

the pageattr_32.c and pageattr_64.c files are now identical.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: enable DEBUG_PAGEALLOC on 64-bit

enable CONFIG_DEBUG_PAGEALLOC=y on 64-bit kernels too.

preliminary testing shows that it's working fine.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add kernel_map_pages() to 64-bit

needed for DEBUG_PAGEALLOC support and for unification.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: return -EINVAL in __change_page_attr(), instead of 0

careful: might change driver behavior - but this is the right
return value.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up differences between 64-bit and 32-bit

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: 64-bit, add the new split_large_page() function

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: 64-bit pageattr.c, prepare for unification

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: change 64-bit pageattr to use set_pte_atomic()

NOP change - same as set_pte().

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: change 64-bit __change_page_attr() to struct page

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: simplify __change_page_attr()

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: introduce native_set_pte_atomic() on 64-bit too

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up and simplify 64-bit split_large_page()

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: unify header part of pageattr_64.c

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: simplify pageattr_64.c

simplify pageattr_64.c.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: prepare for the unification of the cpa code

prepare for the unification of the cpa code, by unifying the
lookup_address() logic between 32-bit and 64-bit.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: prepare for the unification of the cpa code

prepare for the unification of the cpa code, by unifying the
lookup_address() logic between 32-bit and 64-bit.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cpa self-test, WARN_ON()

add a WARN_ON() to the cpa-self-test failure branch.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: do not PSE on CONFIG_DEBUG_PAGEALLOC=y

get more testing of the c_p_a() code done by not turning off
PSE on DEBUG_PAGEALLOC.

this simplifies the early pagetable setup code, and tests
the largepage-splitup code quite heavily.

In the end, all the largepages will be split up pretty quickly,
so there's no difference to how DEBUG_PAGEALLOC worked before.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cpa: simplify locking

further simplify cpa locking: since the largepage-split is a
slowpath, use the pgd_lock for the whole operation, intead
of the mmap_sem.

This also makes it suitable for DEBUG_PAGEALLOC purposes again.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: simplify cpa largepage split, #3

simplify cpa largepage split: push the reference protection bits
into the largepage-splitting function.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cpa self-test fixes

cpa self-test fixes. change_page_attr_addr() was buggy, it
passed in a virtual address as a physical one.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: further cpa largepage-split cleanups

further cpa largepage-split cleanups: make the splitup isolated
functionality, without leaking details back into __change_page_attr().

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: simplify 32-bit cpa largepage splitting

simplify 32-bit cpa largepage splitting: do a pure split and repeat
the pte lookup to get the new pte modified.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: simplify the 32-bit cpa code

simplify the 32-bit cpa code.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix some bugs about EFI runtime code mapping

This patch fixes some bugs of making EFI runtime code executable.

- Use change_page_attr in i386 too. Because the runtime code may be
  mapped not through ioremap.

- If there is no _PAGE_NX in __supported_pte_mask, the change_page_attr
  is not called.

- Make efi_ioremap map pages as PAGE_KERNEL_EXEC_NOCACHE, because EFI runtime
  code may be mapped through efi_ioremap.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix more non-global TLB flushes

fix more __flush_tlb() instances, out of caution.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix early_ioremap() on 64-bit

Fix early_ioremap() on x86-64

I had ACPI failures on several machines since a few days. Symptom
was NUMA nodes not getting detected or worse cores not getting detected.
They all came from ACPI not being able to read various of its tables. I finally
bisected it down to Jeremy's "put _PAGE_GLOBAL into PAGE_KERNEL" change.
With that the fix was fairly obvious. The problem was that early_ioremap()
didn't use a "_all" flush that would affect the global PTEs too. So
with global bits getting used everywhere now an early_ioremap would
not actually flush a mapping if something else was mapped previously
on that slot (which can happen with early_iounmap inbetween)

This patch changes all flushes in init_64.c to be __flush_tlb_all()
and fixes the problem here.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: remove set_kernel_exec()

The SMP trampoline always runs in real mode, so making it executable
in the page tables doesn't make much sense because it executes
before page tables are set up. That was the only user of
set_kernel_exec(). Remove set_kernel_exec().

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: introduce canon_pgprot()

Introduce canon_pgprot()

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: c_p_a() make it more robust against use of PAT bits

Use the page table level instead of the PSE bit to check if the PTE
is for a 4K page or not. This makes the code more robust when the PAT
bit is changed because the PAT bit on 4K pages is in the same position
as the PSE bit.

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: c_p_a() fix: reorder TLB / cache flushes to follow Intel recommendation

Intel recommends to first flush the TLBs and then the caches
on caching attribute changes. c_p_a() previously did it the
other way round. Reorder that.

The procedure is still not fully compliant to the Intel documentation
because Intel recommends a all CPU synchronization step between
the TLB flushes and the cache flushes.

However on all new Intel CPUs this is now meaningless anyways
because they support Self-Snoop and can skip the cache flush
step anyway.

[ mingo@elte.hu: decoupled from clflush and ported it to x86.git ]

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix c_p_a() boot crash

fix:

> hm, i just found a failing 64-bit .config while testing your CPA
> patchset:
>
>  [    1.916541] CPA mapping 4k 0 large 2048 gb 0 x 0[0-0] miss 0
>  [    1.919874] Unable to handle kernel paging request at 000000000335aea8 RIP:
>  [    1.919874]  [<ffffffff8021d2d3>] change_page_attr+0x3/0x61
>  [    1.919874] PGD 0
>  [    1.919874] Oops: 0000 [1]
>  [    1.919874] CPU 0

This handles addresses which don't have a mem_map entry.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: don't drop NX bit in pte modifier functions on 32-bit

The pte_* modifier functions that cleared bits dropped the NX bit on 32bit
PAE because they only worked in int, but NX is in bit 63. Fix that
by adding appropiate casts so that the arithmetic happens as long long
on PAE kernels.

I decided to just use 64bit arithmetic instead of open coding like
pte_modify() because gcc should generate good enough code for that now.

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add pte_pgprot to 32-bit

64bit already had it.

Needed for later patches.

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: shrink __PAGE_KERNEL/__PAGE_KERNEL_EXEC on non PAE kernels

No need to make it 64bit there.

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cpa: remove unnecessary masking of address

virt_to_page does not care about the bits below the page granuality.
So don't mask them.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cpa: use wbinvd() macro instead of inline assembly in 64bit c_p_a()

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: early_ioremap_init(), enhance warnings

enhance the debug warning in early_ioremap_init().

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix EISA ioremap

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix early_ioremap()/btmap

fix a long-standing weakness of the early-ioremap allocator: it
uses a single pgd entry for the boot mappings, and was not properly
protecting itself against crossing a 2MB (4MB) boundary.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add early_ioremap() leak detection

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: make early_ioremap_debug early_param

This patch makes "early_ioremap_debug" a early parameter, because
"early_ioreamp/early_iounmap" is only used during early boot stage.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: early_ioremap(), debugging

add early_ioremap() debug printouts via the early_ioremap_debug
boot option.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add debug warnings to early_ioremap()

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: increase the number of boot-mappings

increase max early_ioremap() remapping size from 64K to 256K.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: enhance early_ioremap()

 - allow nesting of up to 4 levels

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: early_ioremap_reset fix

This patch fixes a bug of early_ioremap_reset.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86 32-bit boot: rename bt_ioremap() to early_ioremap()

This patch renames bt_ioremap to early_ioremap, which is used in
x86_64. This makes it easier to merge i386 and x86_64 usage.

[ mingo@elte.hu: fix ]

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: replace boot_ioremap() with enhanced bt_ioremap() - remove boot_ioremap()

This patch replaces boot_ioremap invokation with bt_ioremap and
removes the boot_ioremap implementation.

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

i386 boot: replace boot_ioremap with enhanced bt_ioremap - enhance bt_ioremap

This patch makes it possible for bt_ioremap() to be used before
paging_init(), via providing an early implementation of set_fixmap()
that can be used before paging_init().

This way boot_ioremap() can be replaced by bt_ioremap().

Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: set strong uncacheable where UC is really desired

Also use _PAGE_PWT for all the mappings which need uncache mapping.
Instead of existing PAT2 which is UC- (and can be overwritten by MTRRs),
we now use PAT3 which is strong uncacheable.

This makes it consistent with pgprot_noncached()

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: use __PAGE_KERNEL_EXEC in ioremap_64.c

This patch replaces the manual permission setup for pages in ioremap_64.c with
the pre-defined __PAGE_KERNEL_EXEC value.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: cleanup boot_ioremap_32.c

Coding style cleanup before modifying the file.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up arch/x86/mm/pageattr-test.c

fix 15 checkpatch warnings.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: c_p_a(), add simple self test at boot

Since change_page_attr() is tricky code it is good to have some regression
test code. This patch maps and unmaps some random pages in the direct mapping
at boot and then dumps the state and does some simple sanity checks.

Add it with a CONFIG option.

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: return the page table level in lookup_address()

based on this patch from Andi Kleen:

|  Subject: CPA: Return the page table level in lookup_address()
|  From: Andi Kleen <ak@suse.de>
|
|  Needed for the next change.
|
|  And change all the callers.

and ported it to x86.git.

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add pte accessors for the global bit

Needed for some test code.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up pte_exec

- Rename it to pte_exec() from pte_exec_kernel(). There is nothing
kernel specific in there.
- Move it into the common file because _PAGE_NX is 0 on !PAE and then
pte_exec() will be always evaluate to true.

Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add dump_pagetable helper to X86_32

Similar to x86 64-bit.

Signed-off-by: Harvey Harrison <harvey.harrison@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

c_p_a(): do a simple self test at boot

When CONFIG_DEBUG_RODATA is enabled undo the ro mapping and redo it again.
This gives some simple testing for change_page_attr().

Signed-off-by: Andi Kleen <ak@suse.de>
Acked-by: Jan Beulich <jbeulich@novell.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up arch/x86/mm/pageattr_64.c

clean up arch/x86/mm/pageattr_64.c.

no code changed:

   text    data     bss     dec     hex filename
   1751      16       0    1767     6e7 pageattr_64.o.before
   1751      16       0    1767     6e7 pageattr_64.o.after

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: clean up arch/x86/mm/pageattr_32.c

clean up arch/x86/mm/pageattr_32.c.

no code changed:

   text    data     bss     dec     hex filename
   1255      40       0    1295     50f pageattr_32.o.before
   1255      40       0    1295     50f pageattr_32.o.after

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: change ioremap() to default to uncached

Prepare ioremap() to default to uncached. This will be the
safest - but first we have to fix CPA.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: fix recursion in arch/x86/kernel/cpu/mcheck/mce_amd_64.c

remove the recursion from this function.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: allocate and initialize unshared pmds

If SHARED_KERNEL_PMD is false, then we need to allocate and initialize
the kernel pmd.  We can easily piggy-back this onto the existing pmd
prepopulation code.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: preallocate pmds at pgd creation time

In PAE mode, an update to the pgd requires a cr3 reload to make sure
the processor notices the changes.  Since this also has the
side-effect of flushing the tlb, its an expensive operation which we
want to avoid where possible.

This patch mitigates the cost of installing the initial set of pmds on
process creation by preallocating them when the pgd is allocated.
This avoids up to three tlb flushes during exec, as it creates the new
process address space while the pagetable is in active use.

The pmds will be freed as part of the normal pagetable teardown in
free_pgtables, which is called in munmap and process exit.  However,
free_pgtables will only free parts of the pagetable which actually
contain mappings, so stray pmds may still be attached to the pgd at
pgd_free time.  We must mop them up to prevent a memory leak.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Andi Kleen <ak@suse.de>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: William Irwin <wli@holomorphy.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

xen: deal with pmd being allocated/freed

Deal properly with pmd-level pages being allocated and freed
dynamically.  We can handle them more or less the same as pte pages.

Also, deal with early_ioremap pagetable manipulations.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: demacro asm-x86/pgalloc_32.h

Convert macros into inline functions, for better type-checking.

This patch required a little bit of fiddling with headers in order to
make __(pte|pmd)_free_tlb inline rather than macros.
asm-generic/tlb.h includes asm/pgalloc.h, though it doesn't directly
use any pgalloc definitions.  I removed this include to avoid an
include cycle, but it may cause secondary compile failures by things
depending on the indirect inclusion; arch/x86/mm/hugetlbpage.c was one
such place; there may be others.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: add mm parameter to paravirt_alloc_pd

Add mm to paravirt_alloc_pd, partly to make it consistent with
paravirt_alloc_pt, and because later changes will make use of it.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

xen: fix mismerge in masking pte flags

Looks like a mismerge/misapply dropped one of the cases of pte flag
masking for Xen.  Also, only mask the flags for present ptes.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: change aper valid checking sequence

old sequence:
  size ==> >4G  ==> point to RAM

changed to:
  >4G ==> point to RAM ==> size

some bios even leave aper to unclear, so check size at last.

To avoid reporting:

  Node 0: Aperture @ 4a42000000 size 32 MB
  Aperture too small (32 MB)

with this change we will get:

  Node 0: Aperture @ 4a42000000 size 32 MB
  Aperture beyond 4G. Ignoring.

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: force enable HPET on (some?) ICH9 boards

Some consumer ICH9 boards (such as the Abit IP35 Pro) do not provide a BIOS
option for enabling the HPET. The same ICH workaround used for 6,7,8 can be
applied to 9. Here I enable the only PCI id that was visible on my system.

I have confirmed the HPETs work both from userspace and as a clocksource for
the running kernel (2.6.24 here) after applying this patch.

 Force enabled HPET at base address 0xfed00000
 hpet clockevent registered
 hpet0: at MMIO 0xfed00000, IRQs 2, 8, 0, 0
 hpet0: 4 64-bit timers, 14318180 Hz

Signed-off-by: Alistair John Strachan <alistair@devzero.co.uk>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

x86: not set boot cpu in cpu_online_map in smp_prepare_boot_cpu()

in init/main.c boot_cpu_init() does that before.

Signed-off-by: Yinghai Lu <yinghai.lu@sun.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>