On Thu, 2018-03-15 at 13:09 +0530, Chintan Pandya wrote:

On 3/14/2018 11:31 PM, Toshi Kani wrote:

quoted

Implement pud_free_pmd_page() and pmd_free_pte_page() on x86, which
clear a given pud/pmd entry and free up lower level page table(s).
Address range associated with the pud/pmd entry must have been purged
by INVLPG.

fixes: e61ce6ade404e ("mm: change ioremap to set up huge I/O mappings")
Signed-off-by: Toshi Kani <redacted>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Thomas Gleixner <redacted>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Borislav Petkov <redacted>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: <redacted>
---
  arch/x86/mm/pgtable.c |   28 ++++++++++++++++++++++++++--
  1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/arch/x86/mm/pgtable.c b/arch/x86/mm/pgtable.c
index 1eed7ed518e6..34cda7e0551b 100644
--- a/arch/x86/mm/pgtable.c
+++ b/arch/x86/mm/pgtable.c

@@ -712,7 +712,22 @@ int pmd_clear_huge(pmd_t *pmd)
   */
  int pud_free_pmd_page(pud_t *pud)
  {
-	return pud_none(*pud);
+	pmd_t *pmd;
+	int i;
+
+	if (pud_none(*pud))
+		return 1;
+
+	pmd = (pmd_t *)pud_page_vaddr(*pud);
+
+	for (i = 0; i < PTRS_PER_PMD; i++)
+		if (!pmd_free_pte_page(&pmd[i]))

This is forced action and no optional. Also, pmd_free_pte_page()
doesn't return 0 in any case. So, you may remove _if_ ?

The code needs to be written per the interface definition, not per the
current implementation.

quoted

+			return 0;
+
+	pud_clear(pud);
+	free_page((unsigned long)pmd);
+
+	return 1;
  }
  
  /**

@@ -724,6 +739,15 @@ int pud_free_pmd_page(pud_t *pud)
   */
  int pmd_free_pte_page(pmd_t *pmd)
  {
-	return pmd_none(*pmd);
+	pte_t *pte;
+
+	if (pmd_none(*pmd))

This should also check if pmd is already huge. Same for pud ?

Not necessary. As described in the function header, one of the entry
conditions is that a given pmd range is unmapped.  See
vunmap_pmd_range().

Thanks,
-Toshi