On Tue, Jun 27, 2017 at 08:54:07PM +0530, Aneesh Kumar K.V wrote:

On Tuesday 27 June 2017 03:41 PM, Ram Pai wrote:

quoted

Pass the correct protection key value to the hash functions on
page fault.

Signed-off-by: Ram Pai <redacted>
---
 arch/powerpc/include/asm/pkeys.h | 11 +++++++++++
 arch/powerpc/mm/hash_utils_64.c  |  4 ++++
 arch/powerpc/mm/mem.c            |  6 ++++++
 3 files changed, 21 insertions(+)

diff --git a/arch/powerpc/include/asm/pkeys.h b/arch/powerpc/include/asm/pkeys.h
index ef1c601..1370b3f 100644
--- a/arch/powerpc/include/asm/pkeys.h
+++ b/arch/powerpc/include/asm/pkeys.h

@@ -74,6 +74,17 @@ static inline bool mm_pkey_is_allocated(struct mm_struct *mm, int pkey)
 }

 /*
+ * return the protection key of the vma corresponding to the
+ * given effective address @ea.
+ */
+static inline int mm_pkey(struct mm_struct *mm, unsigned long ea)
+{
+	struct vm_area_struct *vma = find_vma(mm, ea);
+	int pkey = vma ? vma_pkey(vma) : 0;
+	return pkey;
+}
+
+/*

That is not going to work in hash fault path right ? We can't do a
find_vma there without holding the mmap_sem

There is a fundamental problem with this new design. Looks like we can't
hold a lock in that path, without badly hurting the performance.

I am moving back to the old design. Cant by-pass the pte. The
keys will be programmed into the pte which will than be used
to program the hpte.

RP