s390/mm: implement software referenced bits

The last remaining use for the storage key of the s390 architecture
is reference counting. The alternative is to make page table entries
invalid while they are old. On access the fault handler marks the
pte/pmd as young which makes the pte/pmd valid if the access rights
allow read access. The pte/pmd invalidations required for software
managed reference bits cost a bit of performance, on the other hand
the RRBE/RRBM instructions to read and reset the referenced bits are
quite expensive as well.
Reviewed-by: Gerald Schaefer <gerald.schaefer@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/page.h

@@ -140,15 +140,6 @@ static inline int page_reset_referenced(unsigned long addr)
 #define _PAGE_FP_BIT		0x08	/* HW fetch protection bit	*/
 #define _PAGE_ACC_BITS		0xf0	/* HW access control bits	*/
 
-/*
- * Test and clear referenced bit in storage key.
- */
-#define __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
-static inline int page_test_and_clear_young(unsigned long pfn)
-{
-	return page_reset_referenced(pfn << PAGE_SHIFT);
-}
-
 struct page;
 void arch_free_page(struct page *page, int order);
 void arch_alloc_page(struct page *page, int order);

arch/s390/mm/hugetlbpage.c

@@ -10,19 +10,25 @@
 
 static inline pmd_t __pte_to_pmd(pte_t pte)
 {
-	int none, prot;
+	int none, young, prot;
 	pmd_t pmd;
 
 	/*
-	 * Convert encoding	  pte bits	  pmd bits
-	 *			.IR.....wdtp	..R...I.....
-	 * empty		.10.....0000 -> ..0...1.....
-	 * prot-none, clean	.11.....0001 -> ..1...1.....
-	 * prot-none, dirty	.10.....0101 -> ..1...1.....
-	 * read-only, clean	.01.....0001 -> ..1...0.....
-	 * read-only, dirty	.01.....0101 -> ..1...0.....
-	 * read-write, clean	.01.....1001 -> ..0...0.....
-	 * read-write, dirty	.00.....1101 -> ..0...0.....
+	 * Convert encoding		  pte bits	  pmd bits
+	 *				.IR...wrdytp	..R...I...y.
+	 * empty			.10...000000 -> ..0...1...0.
+	 * prot-none, clean, old	.11...000001 -> ..0...1...1.
+	 * prot-none, clean, young	.11...000101 -> ..1...1...1.
+	 * prot-none, dirty, old	.10...001001 -> ..0...1...1.
+	 * prot-none, dirty, young	.10...001101 -> ..1...1...1.
+	 * read-only, clean, old	.11...010001 -> ..1...1...0.
+	 * read-only, clean, young	.01...010101 -> ..1...0...1.
+	 * read-only, dirty, old	.11...011001 -> ..1...1...0.
+	 * read-only, dirty, young	.01...011101 -> ..1...0...1.
+	 * read-write, clean, old	.11...110001 -> ..0...1...0.
+	 * read-write, clean, young	.01...110101 -> ..0...0...1.
+	 * read-write, dirty, old	.10...111001 -> ..0...1...0.
+	 * read-write, dirty, young	.00...111101 -> ..0...0...1.
 	 * Huge ptes are dirty by definition, a clean pte is made dirty
 	 * by the conversion.
 	 */
@@ -31,9 +37,14 @@ static inline pmd_t __pte_to_pmd(pte_t pte)
 		if (pte_val(pte) & _PAGE_INVALID)
 			pmd_val(pmd) |= _SEGMENT_ENTRY_INVALID;
 		none = (pte_val(pte) & _PAGE_PRESENT) &&
-			(pte_val(pte) & _PAGE_INVALID);
-		prot = (pte_val(pte) & _PAGE_PROTECT);
-		if (prot || none)
+			!(pte_val(pte) & _PAGE_READ) &&
+			!(pte_val(pte) & _PAGE_WRITE);
+		prot = (pte_val(pte) & _PAGE_PROTECT) &&
+			!(pte_val(pte) & _PAGE_WRITE);
+		young = pte_val(pte) & _PAGE_YOUNG;
+		if (none || young)
+			pmd_val(pmd) |= _SEGMENT_ENTRY_YOUNG;
+		if (prot || (none && young))
 			pmd_val(pmd) |= _SEGMENT_ENTRY_PROTECT;
 	} else
 		pmd_val(pmd) = _SEGMENT_ENTRY_INVALID;
@@ -46,11 +57,14 @@ static inline pte_t __pmd_to_pte(pmd_t pmd)
 
 	/*
 	 * Convert encoding	  pmd bits	  pte bits
-	 *			..R...I.....	.IR.....wdtp
-	 * empty		..0...1..... -> .10.....0000
-	 * prot-none, young	..1...1..... -> .10.....0101
-	 * read-only, young	..1...0..... -> .01.....0101
-	 * read-write, young	..0...0..... -> .00.....1101
+	 *			..R...I...y.	.IR...wrdytp
+	 * empty		..0...1...0. -> .10...000000
+	 * prot-none, old	..0...1...1. -> .10...001001
+	 * prot-none, young	..1...1...1. -> .10...001101
+	 * read-only, old	..1...1...0. -> .11...011001
+	 * read-only, young	..1...0...1. -> .01...011101
+	 * read-write, old	..0...1...0. -> .10...111001
+	 * read-write, young	..0...0...1. -> .00...111101
 	 * Huge ptes are dirty by definition
 	 */
 	if (pmd_present(pmd)) {
@@ -58,11 +72,17 @@ static inline pte_t __pmd_to_pte(pmd_t pmd)
 			(pmd_val(pmd) & PAGE_MASK);
 		if (pmd_val(pmd) & _SEGMENT_ENTRY_INVALID)
 			pte_val(pte) |= _PAGE_INVALID;
-		else {
+		if (pmd_prot_none(pmd)) {
+			if (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT)
+				pte_val(pte) |= _PAGE_YOUNG;
+		} else {
+			pte_val(pte) |= _PAGE_READ;
 			if (pmd_val(pmd) & _SEGMENT_ENTRY_PROTECT)
 				pte_val(pte) |= _PAGE_PROTECT;
 			else
 				pte_val(pte) |= _PAGE_WRITE;
+			if (pmd_val(pmd) & _SEGMENT_ENTRY_YOUNG)
+				pte_val(pte) |= _PAGE_YOUNG;
 		}
 	} else
 		pte_val(pte) = _PAGE_INVALID;

arch/s390/mm/pgtable.c

@@ -754,7 +754,8 @@ static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm,
 	atomic_set(&page->_mapcount, 3);
 	table = (unsigned long *) page_to_phys(page);
 	clear_table(table, _PAGE_INVALID, PAGE_SIZE/2);
-	clear_table(table + PTRS_PER_PTE, 0, PAGE_SIZE/2);
+	clear_table(table + PTRS_PER_PTE, PGSTE_HR_BIT | PGSTE_HC_BIT,
+		    PAGE_SIZE/2);
 	return table;
 }
 
@@ -792,26 +793,21 @@ int set_guest_storage_key(struct mm_struct *mm, unsigned long addr,
 	pgste_val(new) |= (key & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48;
 	pgste_val(new) |= (key & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56;
 	if (!(pte_val(*ptep) & _PAGE_INVALID)) {
-		unsigned long address, bits;
-		unsigned char skey;
+		unsigned long address, bits, skey;
 
 		address = pte_val(*ptep) & PAGE_MASK;
-		skey = page_get_storage_key(address);
+		skey = (unsigned long) page_get_storage_key(address);
 		bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED);
+		skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT);
 		/* Set storage key ACC and FP */
-		page_set_storage_key(address,
-				(key & (_PAGE_ACC_BITS | _PAGE_FP_BIT)),
-				!nq);
-
+		page_set_storage_key(address, skey, !nq);
 		/* Merge host changed & referenced into pgste  */
 		pgste_val(new) |= bits << 52;
-		/* Transfer skey changed & referenced bit to kvm user bits */
-		pgste_val(new) |= bits << 45;	/* PGSTE_UR_BIT & PGSTE_UC_BIT */
 	}
 	/* changing the guest storage key is considered a change of the page */
 	if ((pgste_val(new) ^ pgste_val(old)) &
 	    (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT))
-		pgste_val(new) |= PGSTE_UC_BIT;
+		pgste_val(new) |= PGSTE_HC_BIT;
 
 	pgste_set_unlock(ptep, new);
 	pte_unmap_unlock(*ptep, ptl);

arch/s390/mm/vmem.c

@@ -118,6 +118,7 @@ static int vmem_add_mem(unsigned long start, unsigned long size, int ro)
 		    !(address & ~PMD_MASK) && (address + PMD_SIZE <= end)) {
 			pmd_val(*pm_dir) = __pa(address) |
 				_SEGMENT_ENTRY | _SEGMENT_ENTRY_LARGE |
+				_SEGMENT_ENTRY_YOUNG |
 				(ro ? _SEGMENT_ENTRY_PROTECT : 0);
 			address += PMD_SIZE;
 			continue;

include/asm-generic/pgtable.h

@@ -208,10 +208,6 @@ static inline int pmd_same(pmd_t pmd_a, pmd_t pmd_b)
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 #endif
 
-#ifndef __HAVE_ARCH_PAGE_TEST_AND_CLEAR_YOUNG
-#define page_test_and_clear_young(pfn) (0)
-#endif
-
 #ifndef __HAVE_ARCH_PGD_OFFSET_GATE
 #define pgd_offset_gate(mm, addr)	pgd_offset(mm, addr)
 #endif

mm/rmap.c

@@ -873,9 +873,6 @@ int page_referenced(struct page *page,
 								vm_flags);
 		if (we_locked)
 			unlock_page(page);
-
-		if (page_test_and_clear_young(page_to_pfn(page)))
-			referenced++;
 	}
 out:
 	return referenced;