Le 23/04/2019 à 17:17, Laurent Dufour a écrit :
During a different patch review, the check in arch_munmap() was found
spucious due the lake of explanation.

Adding a comment to clarify the test.

Suggested-by: Thomas Gleixner <[email protected]>
Cc: Michael Ellerman <[email protected]>
Signed-off-by: Laurent Dufour <[email protected]>
---
  arch/powerpc/include/asm/mmu_context.h | 11 +++++++++++
  1 file changed, 11 insertions(+)

diff --git a/arch/powerpc/include/asm/mmu_context.h 
b/arch/powerpc/include/asm/mmu_context.h
index 6ee8195a2ffb..638f481b3c9f 100644
--- a/arch/powerpc/include/asm/mmu_context.h
+++ b/arch/powerpc/include/asm/mmu_context.h
@@ -240,6 +240,17 @@ static inline void arch_unmap(struct mm_struct *mm,
                              struct vm_area_struct *vma,
                              unsigned long start, unsigned long end)
  {
+       /*
+        * There are 2 assumptions here:
+        *  1. the VDSO is one page size (guaranteed by vdso_data_store)

Indeed this is not true, only the descriptor is one page size.
This makes that test not handling all the cases, especially if a upper part of the VDSO is unmap (start > mm->context.vdso_base).

I'll sent a new fix asap.

+        *  2. 'start' and 'end' are page aligned (guaranteed by the caller)
+        * The test is wrote in a way to handle unmap operation surrounding the
+        * VDSO area like:
+        *            | VDSO |
+        *    ^ start          ^ end
+        * The test also covers the munmap() operation done to the exact VDSO's
+        * boundaries.
+        */
        if (start <= mm->context.vdso_base && mm->context.vdso_base < end)
                mm->context.vdso_base = 0;
  }


Reply via email to