Commit 954f8571 authored by Andrea Arcangeli's avatar Andrea Arcangeli Committed by Ingo Molnar
Browse files

Revert "x86, mm: Make spurious_fault check explicitly check explicitly check the PRESENT bit"

I got a report for a minor regression introduced by commit
027ef6c8 ("mm: thp: fix pmd_present for split_huge_page and
PROT_NONE with THP").

So the problem is, pageattr creates kernel pagetables (pte and
pmds) that breaks pte_present/pmd_present and the patch above
exposed this invariant breakage for pmd_present.

The same problem already existed for the pte and pte_present and
it was fixed by commit 660a293e

 ("x86, mm: Make
spurious_fault check explicitly check the PRESENT bit") (if it
wasn't for that commit, it wouldn't even be a regression).  That
fix avoids the pagefault to use pte_present.  I could follow
through by stopping using pmd_present/pmd_huge too.

However I think it's more robust to fix pageattr and to clear
the PSE/GLOBAL bitflags too in addition to the present bitflag.
So the kernel page fault can keep using the regular
pte_present/pmd_present/pmd_huge.

The confusion arises because _PAGE_GLOBAL and _PAGE_PROTNONE are
sharing the same bit, and in the pmd case we pretend _PAGE_PSE
to be set only in present pmds (to facilitate split_huge_page
final tlb flush).
Signed-off-by: default avatarAndrea Arcangeli <aarcange@redhat.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Shaohua Li <shaohua.li@intel.com>
Cc: "H. Peter Anvin" <hpa@linux.intel.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: default avatarIngo Molnar <mingo@kernel.org>
parent 942670d0
......@@ -939,14 +939,8 @@ spurious_fault(unsigned long error_code, unsigned long address)
if (pmd_large(*pmd))
return spurious_fault_check(error_code, (pte_t *) pmd);
/*
* Note: don't use pte_present() here, since it returns true
* if the _PAGE_PROTNONE bit is set. However, this aliases the
* _PAGE_GLOBAL bit, which for kernel pages give false positives
* when CONFIG_DEBUG_PAGEALLOC is used.
*/
pte = pte_offset_kernel(pmd, address);
if (!(pte_flags(*pte) & _PAGE_PRESENT))
if (!pte_present(*pte))
return 0;
ret = spurious_fault_check(error_code, pte);
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment