linux-toradex.git/mm, branch v4.9.76

kaiser: vmstat show NR_KAISERTABLE as nr_overhead

2018-01-05T14:46:33+00:00

The kaiser update made an interesting choice, never to free any shadow
page tables.  Contention on global spinlock was worrying, particularly
with it held across page table scans when freeing.  Something had to be
done: I was going to add refcounting; but simply never to free them is
an appealing choice, minimizing contention without complicating the code
(the more a page table is found already, the less the spinlock is used).

But leaking pages in this way is also a worry: can we get away with it?
At the very least, we need a count to show how bad it actually gets:
in principle, one might end up wasting about 1/256 of memory that way
(1/512 for when direct-mapped pages have to be user-mapped, plus 1/512
for when they are user-mapped from the vmalloc area on another occasion
(but we don't have vmalloc'ed stacks, so only large ldts are vmalloc'ed).

Add per-cpu stat NR_KAISERTABLE: including 256 at startup for the
shared pgd entries, and 1 for each intermediate page table added
thereafter for user-mapping - but leave out the 1 per mm, for its
shadow pgd, because that distracts from the monotonic increase.
Shown in /proc/vmstat as nr_overhead (0 if kaiser not enabled).

In practice, it doesn't look so bad so far: more like 1/12000 after
nine hours of gtests below; and movable pageblock segregation should
tend to cluster the kaiser tables into a subset of the address space
(if not, they will be bad for compaction too).  But production may
tell a different story: keep an eye on this number, and bring back
lighter freeing if it gets out of control (maybe a shrinker).

["nr_overhead" should of course say "nr_kaisertable", if it needs
to stay; but for the moment we are being coy, preferring that when
Joe Blow notices a new line in his /proc/vmstat, he does not get
too curious about what this "kaiser" stuff might be.]

Signed-off-by: Hugh Dickins 
Signed-off-by: Greg Kroah-Hartman

zsmalloc: calling zs_map_object() from irq is a bug

2017-12-14T08:28:23+00:00

[ Upstream commit 1aedcafbf32b3f232c159b14cd0d423fcfe2b861 ]

Use BUG_ON(in_interrupt()) in zs_map_object().  This is not a new
BUG_ON(), it's always been there, but was recently changed to
VM_BUG_ON().  There are several problems there.  First, we use use
per-CPU mappings both in zsmalloc and in zram, and interrupt may easily
corrupt those buffers.  Second, and more importantly, we believe it's
possible to start leaking sensitive information.  Consider the following
case:

-> process P
	swap out
	 zram
	  per-cpu mapping CPU1
	   compress page A
-> IRQ

	swap out
	 zram
	  per-cpu mapping CPU1
	   compress page B
	    write page from per-cpu mapping CPU1 to zsmalloc pool
	iret

-> process P
	    write page from per-cpu mapping CPU1 to zsmalloc pool  [*]
	return

* so we store overwritten data that actually belongs to another
  page (task) and potentially contains sensitive data. And when
  process P will page fault it's going to read (swap in) that
  other task's data.

Link: http://lkml.kernel.org/r/20170929045140.4055-1-sergey.senozhatsky@gmail.com
Signed-off-by: Sergey Senozhatsky 
Acked-by: Minchan Kim 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

thp: fix MADV_DONTNEED vs. numa balancing race

2017-12-14T08:28:16+00:00

commit ced108037c2aa542b3ed8b7afd1576064ad1362a upstream.

In case prot_numa, we are under down_read(mmap_sem).  It's critical to
not clear pmd intermittently to avoid race with MADV_DONTNEED which is
also under down_read(mmap_sem):

	CPU0:				CPU1:
				change_huge_pmd(prot_numa=1)
				 pmdp_huge_get_and_clear_notify()
madvise_dontneed()
 zap_pmd_range()
  pmd_trans_huge(*pmd) == 0 (without ptl)
  // skip the pmd
				 set_pmd_at();
				 // pmd is re-established

The race makes MADV_DONTNEED miss the huge pmd and don't clear it
which may break userspace.

Found by code analysis, never saw triggered.

Link: http://lkml.kernel.org/r/20170302151034.27829-3-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov 
Cc: Andrea Arcangeli 
Cc: Hillf Danton 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
[jwang: adjust context for 4.9 ]
Signed-off-by: Jack Wang 
Signed-off-by: Greg Kroah-Hartman

thp: reduce indentation level in change_huge_pmd()

2017-12-14T08:28:16+00:00

commit 0a85e51d37645e9ce57e5e1a30859e07810ed07c upstream.

Patch series "thp: fix few MADV_DONTNEED races"

For MADV_DONTNEED to work properly with huge pages, it's critical to not
clear pmd intermittently unless you hold down_write(mmap_sem).

Otherwise MADV_DONTNEED can miss the THP which can lead to userspace
breakage.

See example of such race in commit message of patch 2/4.

All these races are found by code inspection.  I haven't seen them
triggered.  I don't think it's worth to apply them to stable@.

This patch (of 4):

Restructure code in preparation for a fix.

Link: http://lkml.kernel.org/r/20170302151034.27829-2-kirill.shutemov@linux.intel.com
Signed-off-by: Kirill A. Shutemov 
Acked-by: Vlastimil Babka 
Cc: Andrea Arcangeli 
Cc: Hillf Danton 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
[jwang: adjust context for 4.9]
Signed-off-by: Jack Wang 
Signed-off-by: Greg Kroah-Hartman

mm: fix remote numa hits statistics

2017-12-09T21:01:51+00:00

[ Upstream commit 2df26639e708a88dcc22171949da638a9998f3bc ]

Jia He has noticed that commit b9f00e147f27 ("mm, page_alloc: reduce
branches in zone_statistics") has an unintentional side effect that
remote node allocation requests are accounted as NUMA_MISS rathat than
NUMA_HIT and NUMA_OTHER if such a request doesn't use __GFP_OTHER_NODE.

There are many of these potentially because the flag is used very rarely
while we have many users of __alloc_pages_node.

Fix this by simply ignoring __GFP_OTHER_NODE (it can be removed in a
follow up patch) and treat all allocations that were satisfied from the
preferred zone's node as NUMA_HITS because this is the same node we
requested the allocation from in most cases.  If this is not the local
node then we just account it as NUMA_OTHER rather than NUMA_LOCAL.

One downsize would be that an allocation request for a node which is
outside of the mempolicy nodemask would be reported as a hit which is a
bit weird but that was the case before b9f00e147f27 already.

Fixes: b9f00e147f27 ("mm, page_alloc: reduce branches in zone_statistics")
Link: http://lkml.kernel.org/r/20170102153057.9451-2-mhocko@kernel.org
Signed-off-by: Michal Hocko 
Reported-by: Jia He 
Reviewed-by: Vlastimil Babka  # with cbmc[1] superpowers
Acked-by: Mel Gorman 
Cc: Johannes Weiner 
Cc: Joonsoo Kim 
Cc: Taku Izumi 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Sasha Levin 
Signed-off-by: Greg Kroah-Hartman

mm, oom_reaper: gather each vma to prevent leaking TLB entry

2017-12-09T21:01:47+00:00

commit 687cb0884a714ff484d038e9190edc874edcf146 upstream.

tlb_gather_mmu(&tlb, mm, 0, -1) means gathering the whole virtual memory
space.  In this case, tlb->fullmm is true.  Some archs like arm64
doesn't flush TLB when tlb->fullmm is true:

  commit 5a7862e83000 ("arm64: tlbflush: avoid flushing when fullmm == 1").

Which causes leaking of tlb entries.

Will clarifies his patch:
 "Basically, we tag each address space with an ASID (PCID on x86) which
  is resident in the TLB. This means we can elide TLB invalidation when
  pulling down a full mm because we won't ever assign that ASID to
  another mm without doing TLB invalidation elsewhere (which actually
  just nukes the whole TLB).

  I think that means that we could potentially not fault on a kernel
  uaccess, because we could hit in the TLB"

There could be a window between complete_signal() sending IPI to other
cores and all threads sharing this mm are really kicked off from cores.
In this window, the oom reaper may calls tlb_flush_mmu_tlbonly() to
flush TLB then frees pages.  However, due to the above problem, the TLB
entries are not really flushed on arm64.  Other threads are possible to
access these pages through TLB entries.  Moreover, a copy_to_user() can
also write to these pages without generating page fault, causes
use-after-free bugs.

This patch gathers each vma instead of gathering full vm space.  In this
case tlb->fullmm is not true.  The behavior of oom reaper become similar
to munmapping before do_exit, which should be safe for all archs.

Link: http://lkml.kernel.org/r/20171107095453.179940-1-wangnan0@huawei.com
Fixes: aac453635549 ("mm, oom: introduce oom reaper")
Signed-off-by: Wang Nan 
Acked-by: Michal Hocko 
Acked-by: David Rientjes 
Cc: Minchan Kim 
Cc: Will Deacon 
Cc: Bob Liu 
Cc: Ingo Molnar 
Cc: Roman Gushchin 
Cc: Konstantin Khlebnikov 
Cc: Andrea Arcangeli 
Cc: 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
[backported to 4.9 stable tree]
Signed-off-by: Michal Hocko 
Signed-off-by: Greg Kroah-Hartman

mm/madvise.c: fix madvise() infinite loop under special circumstances

2017-12-05T10:24:32+00:00

commit 6ea8d958a2c95a1d514015d4e29ba21a8c0a1a91 upstream.

MADVISE_WILLNEED has always been a noop for DAX (formerly XIP) mappings.
Unfortunately madvise_willneed() doesn't communicate this information
properly to the generic madvise syscall implementation.  The calling
convention is quite subtle there.  madvise_vma() is supposed to either
return an error or update &prev otherwise the main loop will never
advance to the next vma and it will keep looping for ever without a way
to get out of the kernel.

It seems this has been broken since introduction.  Nobody has noticed
because nobody seems to be using MADVISE_WILLNEED on these DAX mappings.

[mhocko@suse.com: rewrite changelog]
Link: http://lkml.kernel.org/r/20171127115318.911-1-guoxuenan@huawei.com
Fixes: fe77ba6f4f97 ("[PATCH] xip: madvice/fadvice: execute in place")
Signed-off-by: chenjie 
Signed-off-by: guoxuenan 
Acked-by: Michal Hocko 
Cc: Minchan Kim 
Cc: zhangyi (F) 
Cc: Miao Xie 
Cc: Mike Rapoport 
Cc: Shaohua Li 
Cc: Andrea Arcangeli 
Cc: Mel Gorman 
Cc: Kirill A. Shutemov 
Cc: David Rientjes 
Cc: Anshuman Khandual 
Cc: Rik van Riel 
Cc: Carsten Otte 
Cc: Dan Williams 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm, hugetlbfs: introduce ->split() to vm_operations_struct

2017-12-05T10:24:32+00:00

commit 31383c6865a578834dd953d9dbc88e6b19fe3997 upstream.

Patch series "device-dax: fix unaligned munmap handling"

When device-dax is operating in huge-page mode we want it to behave like
hugetlbfs and fail attempts to split vmas into unaligned ranges.  It
would be messy to teach the munmap path about device-dax alignment
constraints in the same (hstate) way that hugetlbfs communicates this
constraint.  Instead, these patches introduce a new ->split() vm
operation.

This patch (of 2):

The device-dax interface has similar constraints as hugetlbfs in that it
requires the munmap path to unmap in huge page aligned units.  Rather
than add more custom vma handling code in __split_vma() introduce a new
vm operation to perform this vma specific check.

Link: http://lkml.kernel.org/r/151130418135.4029.6783191281930729710.stgit@dwillia2-desk3.amr.corp.intel.com
Fixes: dee410792419 ("/dev/dax, core: file operations and dax-mmap")
Signed-off-by: Dan Williams 
Cc: Jeff Moyer 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm/cma: fix alloc_contig_range ret code/potential leak

2017-12-05T10:24:31+00:00

commit 63cd448908b5eb51d84c52f02b31b9b4ccd1cb5a upstream.

If the call __alloc_contig_migrate_range() in alloc_contig_range returns
-EBUSY, processing continues so that test_pages_isolated() is called
where there is a tracepoint to identify the busy pages.  However, it is
possible for busy pages to become available between the calls to these
two routines.  In this case, the range of pages may be allocated.
Unfortunately, the original return code (ret == -EBUSY) is still set and
returned to the caller.  Therefore, the caller believes the pages were
not allocated and they are leaked.

Update the comment to indicate that allocation is still possible even if
__alloc_contig_migrate_range returns -EBUSY.  Also, clear return code in
this case so that it is not accidentally used or returned to caller.

Link: http://lkml.kernel.org/r/20171122185214.25285-1-mike.kravetz@oracle.com
Fixes: 8ef5849fa8a2 ("mm/cma: always check which page caused allocation failure")
Signed-off-by: Mike Kravetz 
Acked-by: Vlastimil Babka 
Acked-by: Michal Hocko 
Acked-by: Johannes Weiner 
Acked-by: Joonsoo Kim 
Cc: Michal Nazarewicz 
Cc: Laura Abbott 
Cc: Michal Hocko 
Cc: Mel Gorman 
Signed-off-by: Andrew Morton 
Signed-off-by: Linus Torvalds 
Signed-off-by: Greg Kroah-Hartman

mm, thp: Do not make page table dirty unconditionally in touch_p[mu]d()

2017-12-05T10:24:31+00:00

commit a8f97366452ed491d13cf1e44241bc0b5740b1f0 upstream.

Currently, we unconditionally make page table dirty in touch_pmd().
It may result in false-positive can_follow_write_pmd().

We may avoid the situation, if we would only make the page table entry
dirty if caller asks for write access -- FOLL_WRITE.

The patch also changes touch_pud() in the same way.

Signed-off-by: Kirill A. Shutemov 
Cc: Michal Hocko 
Cc: Hugh Dickins 
Signed-off-by: Linus Torvalds 
[Salvatore Bonaccorso: backport for 4.9:
 - Adjust context
 - Drop specific part for PUD-sized transparent hugepages. Support
   for PUD-sized transparent hugepages was added in v4.11-rc1
]
Signed-off-by: Ben Hutchings 
Signed-off-by: Greg Kroah-Hartman