s390/mm: fix refcount usage for 4K pgste

s390 no longer uses the _mapcount field in struct page to identify
the page table format being used. While the code was diligent in handling
the different mappings, it neglected to turn "off" the map bits when
alloc_pgste was being used. This resulted in bits remaining "on" in the
_refcount field, and thus an artifically huge "in use" count that prevents
the pages from actually being released by __free_page.

There's opportunity for improvement in the "1 vs 3" vs "1U vs 3U" vs
"0x1 vs 0x11" etc. variations for all these calls, I am just keeping
things simple compared to neighboring code.

Fixes: 620b4e903179 ("s390: use _refcount for pgtables")
Reported-by: Halil Pasic <pasic@linux.ibm.com>
Bisected-by: Vasily Gorbik <gor@linux.ibm.com>
Signed-off-by: Eric Farman <farman@linux.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

diff --git a/arch/s390/mm/pgalloc.c b/arch/s390/mm/pgalloc.c
index 84bd6329a88dd3ace39e612197dccec0a48dc4fc..e3bd5627afef3452c50325e955611a5814341b6f 100644 (file)
--- a/arch/s390/mm/pgalloc.c
+++ b/arch/s390/mm/pgalloc.c
@@ -252,6 +252,8 @@ void page_table_free(struct mm_struct *mm, unsigned long *table)
                spin_unlock_bh(&mm->context.lock);
                if (mask != 0)
                        return;
+       } else {
+               atomic_xor_bits(&page->_refcount, 3U << 24);
        }
 
        pgtable_page_dtor(page);
@@ -304,6 +306,8 @@ static void __tlb_remove_table(void *_table)
                        break;
                /* fallthrough */
        case 3:         /* 4K page table with pgstes */
+               if (mask & 3)
+                       atomic_xor_bits(&page->_refcount, 3 << 24);
                pgtable_page_dtor(page);
                __free_page(page);
                break;