If we enable RADIX but disable HUGETLBFS, the build breaks with:

  arch/powerpc/mm/pgtable-radix.c:557:7: error: implicit declaration of function 'pmd_huge'
  arch/powerpc/mm/pgtable-radix.c:588:7: error: implicit declaration of function 'pud_huge'

Fix it by stubbing those functions when HUGETLBFS=n.

Fixes: 4b5d62ca ("powerpc/mm: add radix__remove_section_mapping()")
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This adds support for using two hypercalls to change the size of the
main hash page table while running as a PAPR guest. For now these
hypercalls are only in experimental qemu versions.

The interface is two part: first H_RESIZE_HPT_PREPARE is used to
allocate and prepare the new hash table. This may be slow, but can be
done asynchronously. Then, H_RESIZE_HPT_COMMIT is used to switch to the
new hash table. This requires that no CPUs be concurrently updating the
HPT, and so must be run under stop_machine().

This also adds a debugfs file which can be used to manually control
HPT resizing or testing purposes.
Signed-off-by: NDavid Gibson <david@gibson.dropbear.id.au>
Reviewed-by: NPaul Mackerras <paulus@samba.org>
[mpe: Rename the debugfs file to "hpt_order"]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Tear down and free the four-level page tables of physical mappings
during memory hotremove.

Borrow the basic structure of remove_pagetable() and friends from the
identically-named x86 functions. Reduce the frequency of tlb flushes and
page_table_lock spinlocks by only doing them in the outermost function.
There was some question as to whether the locking is needed at all.
Leave it for now, but we could consider dropping it.

Memory must be offline to be removed, thus not in use. So there
shouldn't be the sort of concurrent page walking activity here that
might prompt us to use RCU.
Signed-off-by: NReza Arbab <arbab@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Wire up memory hotplug page mapping for radix. Share the mapping
function already used by radix_init_pgtable().
Signed-off-by: NReza Arbab <arbab@linux.vnet.ibm.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

The proto VSID is built using both the MMU context id and effective
segment ID (ESID). We should not have overlapping bits between those.
That could result in us having a VSID collision. With the current code
we missed masking the top bits of the ESID. This implies for kernel
address we ended up using the top 4 bits of the ESID as part of the
proto VSID, which is wrong.

The current code use the top 4 context values (0x7fffc - 0x7ffff) for
the kernel. With those context IDs used for the kernel, we don't run
into VSID collisions because we get the same proto VSID irrespective of
whether we mask the ESID bits or not. eg:

  ea         = 0xf000000000000000
  context    = 0x7ffff

  w/out masking:
  proto_vsid = (0x7ffff << 6 | 0xf000000000000000 >> 40)
	     = (0x1ffffc0 | 0xf00000)
	     =  0x1ffffc0

  with masking:
  proto_vsid = (0x7ffff << 6 | ((0xf000000000000000 >> 40) & 0x3f))
	     = (0x1ffffc0 | (0xf00000 & 0x3f))
	     =  0x1ffffc0 | 0)
	     =  0x1ffffc0

So although there is no bug, the code is still overly subtle, so fix it
to save ourselves pain in future.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

We now support THP with both 64k and 4K page size configuration
for radix. (hash only support THP with 64K page size). Hence we
will have CONFIG_TRANSPARENT_HUGEPAGE enabled for both PPC_64K
and PPC_4K config. Since we only need large pmd page table
with hash configuration (to store the slot information
in the second half of the table) restrict the large pmd page table
to THP and 64K configs.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Reviewed-by: NAnshuman Khandual <khandual@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

We don't do this for other page table entries. So lets keep this simple
and always return false for hugepd check on a 64K page size config.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When we switched to big endian page table, we never updated the hugepd
format such that it can work for both big endian and little endian
config. This patch series update hugepd format such that it is looked at
as __be64 value in big endian page table config.

This patch also switch hugepd_t.pd from signed long to unsigned long.
I did update the FSL hugepd_ok check to check for the top bit instead
of checking > 0.

Fixes: 5dc1ef85 ("powerpc/mm: Use big endian Linux page tables for book3s 64")
Cc: stable@vger.kernel.org # v4.7+
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Memory hotplug is leading to hash page table calls, even on radix:

  arch_add_memory
    create_section_mapping
      htab_bolt_mapping
        BUG_ON(!ppc_md.hpte_insert);

To fix, refactor {create,remove}_section_mapping() into hash__ and
radix__ variants. Leave the radix versions stubbed for now.
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NBalbir Singh <bsingharora@gmail.com>
Signed-off-by: NReza Arbab <arbab@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Add arch specific callback in the generic THP page cache code that will
deposit and withdarw preallocated page table.  Archs like ppc64 use this
preallocated table to store the hash pte slot information.

Testing:
kernel build of the patch series on tmpfs mounted with option huge=always

The related thp stat:
thp_fault_alloc 72939
thp_fault_fallback 60547
thp_collapse_alloc 603
thp_collapse_alloc_failed 0
thp_file_alloc 253763
thp_file_mapped 4251
thp_split_page 51518
thp_split_page_failed 1
thp_deferred_split_page 73566
thp_split_pmd 665
thp_zero_page_alloc 3
thp_zero_page_alloc_failed 0

[akpm@linux-foundation.org: remove unneeded parentheses, per Kirill]
Link: http://lkml.kernel.org/r/20161113150025.17942-2-aneesh.kumar@linux.vnet.ibm.comSigned-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Neuling <mikey@neuling.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Independent of whether the vma is for anonymous memory, some arches like
ppc64 would like to override pmd_move_must_withdraw().

One option is to encapsulate the vma_is_anonymous() check for general
architectures inside pmd_move_must_withdraw() so that is always called
and architectures that need unconditional overriding can override this
function.  ppc64 needs to override the function when the MMU is
configured to use hash PTE's.

[bsingharora@gmail.com: reworked changelog]
Link: http://lkml.kernel.org/r/20161113150025.17942-1-aneesh.kumar@linux.vnet.ibm.comSigned-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Acked-by: Michael Ellerman <mpe@ellerman.id.au> (powerpc)
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Michael Neuling <mikey@neuling.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Balbir Singh <bsingharora@gmail.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Today powerpc64 uses a set of pgtable_caches while powerpc32 uses
standard pages when using 4k pages and a single pgtable_cache
if using other size pages.

In preparation of implementing huge pages on the 8xx, this patch
replaces the specific powerpc32 handling by the 64 bits approach.

This is done by:
* moving 64 bits pgtable_cache_add() and pgtable_cache_init()
in a new file called init-common.c
* modifying pgtable_cache_init() to also handle the case
without PMD
* removing the 32 bits version of pgtable_cache_add() and
pgtable_cache_init()
* copying related header contents from 64 bits into both the
book3s/32 and nohash/32 header files

On the 8xx, the following cache sizes will be used:
* 4k pages mode:
- PGT_CACHE(10) for PGD
- PGT_CACHE(3) for 512k hugepage tables
* 16k pages mode:
- PGT_CACHE(6) for PGD
- PGT_CACHE(7) for 512k hugepage tables
- PGT_CACHE(3) for 8M hugepage tables
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NScott Wood <oss@buserror.net>

This will improve the task exit case, by batching tlb invalidates.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When we are updating a pte, we just need to flush the tlb mapping
that pte. Right now we do a full mm flush because we don't track page
size. Now that we have page size details in pte use that to do the
optimized flush
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

When we are updating a pte, we just need to flush the tlb mapping
that pte. Right now we do a full mm flush because we don't track the page
size. Now that we have page size details in pte use that to do the
optimized flush
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Now that we have page size details encoded in pte using software pte
bits, use that to find the page size needed for tlb flush.

This function should only be used on P9 DD1, so give it a horrible name
to make that clear.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This patch adds a new software defined pte bit. We use the reserved
fields of ISA 3.0 pte definition since we will only be using this on DD1
code paths. We can possibly look at removing this code later.

The software bit will be used to differentiate between 64K/4K and 2M
ptes. This helps in finding the page size mapping by a pte so that we
can do efficient tlb flush.

We don't support 1G hugetlb pages yet. So we add a DEBUG WARN_ON to
catch wrong usage.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

W.r.t hash page table config, we support 16MB and 16GB as the hugepage
size. Update the hstate_get_psize to handle 16M and 16G.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

We will start moving some book3s specific hugetlb functions there.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

MIN_HUGEPTE_SHIFT hasn't been used since commit d1837cba
("powerpc/mm: Cleanup initialization of hugepages on powerpc")
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This changes the way that we support the new ISA v3.00 HPTE format.
Instead of adapting everything that uses HPTE values to handle either
the old format or the new format, depending on which CPU we are on,
we now convert explicitly between old and new formats if necessary
in the low-level routines that actually access HPTEs in memory.
This limits the amount of code that needs to know about the new
format and makes the conversions explicit.  This is OK because the
old format contains all the information that is in the new format.

This also fixes operation under a hypervisor, because the H_ENTER
hypercall (and other hypercalls that deal with HPTEs) will continue
to require the HPTE value to be supplied in the old format.  At
present the kernel will not boot in HPT mode on POWER9 under a
hypervisor.

This fixes and partially reverts commit 50de596d
("powerpc/mm/hash: Add support for Power9 Hash", 2016-04-29).

Fixes: 50de596d ("powerpc/mm/hash: Add support for Power9 Hash")
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

On some CPUs like the 8xx, _PAGE_RW hence _PAGE_WRITE is defined
as 0 and _PAGE_RO has to be set when a page is not writable

_PAGE_RO is defined by default in pte-common.h, however BOOK3S/64
doesn't include that file so _PAGE_RO has to be defined explicitly
in book3s/64/pgtable.h

Fixes: a7b9f671 ("powerpc32: adds handling of _PAGE_RO")
Signed-off-by: NChristophe Leroy <christophe.leroy@c-s.fr>
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

POWER9 DD1 requires pte to be marked invalid (V=0) before updating
it with the new value. This makes this distinction for the different
revisions.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NMichael Neuling <mikey@neuling.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

POWER9 DD1 uses RTS - 28 for the RTS value but other revisions use
RTS - 31.  This makes this distinction for the different revisions
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Acked-by: NMichael Neuling <mikey@neuling.org>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This replaces a 2-D search through an array with a simple 8-bit table
lookup for determining the actual and/or base page size for a HPT entry.

The encoding in the second doubleword of the HPTE is designed to encode
the actual and base page sizes without using any more bits than would be
needed for a 4k page number, by using between 1 and 8 low-order bits of
the RPN (real page number) field to encode the page sizes.  A single
"large page" bit in the first doubleword indicates that these low-order
bits are to be interpreted like this.

We can determine the page sizes by using the low-order 8 bits of the RPN
to look up a 256-entry table.  For actual page sizes less than 1MB, some
of the upper bits of these 8 bits are going to be real address bits, but
we can cope with that by replicating the entries for those smaller page
sizes.

While we're at it, let's move the hpte_page_size() and hpte_base_page_size()
functions from a KVM-specific header to a header for 64-bit HPT systems,
since this computation doesn't have anything specifically to do with KVM.
Reviewed-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NPaul Mackerras <paulus@ozlabs.org>

We want to initialise register_process_table() before ppc_md is setup,
so that it can be called as part of MMU init (at least on Radix ATM).

That no longer works because probe_machine() requires that ppc_md be
empty before it's called, and we now do probe_machine() much later.

So make register_process_table a global for now. It will probably move
into a mmu_radix_ops struct at some point in the future.

This was broken by me when applying commit 7025776e "powerpc/mm:
Move hash table ops to a separate structure" due to conflicts with other
patches.

Fixes: 7025776e ("powerpc/mm: Move hash table ops to a separate structure")
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This should be same as flush_tlb_page except for hash32. For hash32
I guess the existing code is wrong, because we don't seem to be
flushing tlb for Hash != 0 case at all. Fix this by switching to
calling flush_tlb_page() which does the right thing by flushing
tlb for both hash and nohash case with hash32
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Some archs like ppc64 need to do special things when flushing tlb for
hugepage. Add a new helper to flush hugetlb tlb range. This helps us to
avoid flushing the entire tlb mapping for the pid.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Use the helper instead of open coding the same at multiple place
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Instead of flushing the entire mm, implement a flush_pmd_tlb_range
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Replace opencoding of the same at multiple places with the helper.
No functional change with this patch.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Now that we track page size in mmu_gather, we can use address based
tlbie format when doing a tlb_flush(). We don't do this if we are
invalidating the full address space.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

We plan to use jump label for cpu_has_feature(). In order to implement
this we need to include the linux/jump_label.h in asm/cputable.h.

Unfortunately if we do that it leads to an include loop. The root of the
problem seems to be that reg.h needs cputable.h (for CPU_FTRs), and then
cputable.h via jump_label.h eventually pulls in hw_irq.h which needs
reg.h (for MSR_EE).

So move cpu_has_feature() to a separate file on its own.
Signed-off-by: NKevin Hao <haokexin@gmail.com>
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
[mpe: Rename to cpu_has_feature.h and flesh out change log]
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

This switches early feature checks to use the non static key variant of
the function. In later patches we will be switching cpu_has_feature()
and mmu_has_feature() to use static keys and we can use them only after
static key/jump label is initialized. Any check for feature before jump
label init should be done using this new helper.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently we have radix_enabled() three times, twice in asm/book3s/64/mmu.h
and then a fallback in asm/mmu.h.

Consolidate them in asm/mmu.h. While we're at it convert them to be
static inlines, and change the fallback case to returning a bool, like
mmu_has_feature().
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

MMU feature bits are defined such that we use the lower half to
present MMU family features. Remove the strict split of half and
also move Radix to a mmu family feature. Radix introduce a new MMU
model and strictly speaking it is a new MMU family. This also free
up bits which can be used for individual features later.
Signed-off-by: NAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Like we just did for hash, split the device tree scanning parts out and
call them from mmu_early_init_devtree().
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>

Currently MMU initialisation (early_init_mmu()) consists of a mixture of
scanning the device tree, setting MMU feature bits, and then also doing
actual initialisation of MMU data structures.

We'd like to decouple the setting of the MMU features from the actual
setup. So split out the device tree scanning, and associated code, and
call it from mmu_init_early_devtree().
Signed-off-by: NMichael Ellerman <mpe@ellerman.id.au>