ppc64: POWER4 lazy icache flushing

arch/ppc64/kernel/head.S

@@ -575,7 +575,8 @@ stab_bolted_user_return:
 	bl	.do_stab_SI
 	b	1f
 
-2:	bl	.do_hash_page_DSI 	/* Try to handle as hpte fault */
+2:	li	r5,0x300
+	bl	.do_hash_page_DSI 	/* Try to handle as hpte fault */
 1:
 	ld      r4,_DAR(r1)
 	ld      r5,_DSISR(r1)
@@ -627,9 +628,8 @@ InstructionAccess_common:
 	bl	.do_stab_SI
 	b	1f
 
-2:	andis.	r0,r23,0x4000		/* no pte found? */
-	beq	1f			/* if so, try to put a PTE */
-	mr	r3,r22			/* into the hash table */
+2:	mr	r3,r22
+	li	r5,0x400
 	bl	.do_hash_page_ISI	/* Try to handle as hpte fault */
 1:
 	mr	r4,r22
@@ -804,6 +804,7 @@ _GLOBAL(do_hash_page_DSI)
 	/*
 	 * r3 contains the faulting address
 	 * r4 contains the required access permissions
+	 * r5 contains the trap number
 	 *
 	 * at return r3 = 0 for success
 	 */
@@ -1119,7 +1120,7 @@ _GLOBAL(save_remaining_regs)
 	rldimi	r22,r20,15,48		/* Insert desired EE value */
 #endif
 
-	mtmsrd  r22
+	mtmsrd  r22,1
 	blr
 
 

arch/ppc64/kernel/htab.c

@@ -195,7 +195,7 @@ static inline unsigned long computeHptePP(unsigned long pte)
  * to be valid via Linux page tables, return 1. If handled return 0
  */
 int __hash_page(unsigned long ea, unsigned long access, unsigned long vsid,
-		pte_t *ptep)
+		pte_t *ptep, unsigned long trap)
 {
 	unsigned long va, vpn;
 	unsigned long newpp, prpn;
@@ -244,6 +244,24 @@ int __hash_page(unsigned long ea, unsigned long access, unsigned long vsid,
 
 	newpp = computeHptePP(pte_val(new_pte));
 
+#define PPC64_HWNOEXEC (1 << 2)
+
+	/* We do lazy icache flushing on POWER4 */
+	if (__is_processor(PV_POWER4) && pfn_valid(pte_pfn(new_pte))) {
+		struct page *page = pte_page(new_pte);
+
+		/* page is dirty */
+		if (!PageReserved(page) &&
+		    !test_bit(PG_arch_1, &page->flags)) {
+			if (trap == 0x400) {
+				__flush_dcache_icache(page_address(page));
+				set_bit(PG_arch_1, &page->flags);
+			} else {
+				newpp |= PPC64_HWNOEXEC;
+			}
+		}
+	}
+
 	/* Check if pte already has an hpte (case 2) */
 	if (pte_val(old_pte) & _PAGE_HASHPTE) {
 		/* There MIGHT be an HPTE for this pte */
@@ -317,7 +335,7 @@ int __hash_page(unsigned long ea, unsigned long access, unsigned long vsid,
 	return 0;
 }
 
-int hash_page(unsigned long ea, unsigned long access)
+int hash_page(unsigned long ea, unsigned long access, unsigned long trap)
 {
 	void *pgdir;
 	unsigned long vsid;
@@ -376,7 +394,7 @@ int hash_page(unsigned long ea, unsigned long access)
 	 */
 	spin_lock(&mm->page_table_lock);
 	ptep = find_linux_pte(pgdir, ea);
-	ret = __hash_page(ea, access, vsid, ptep);
+	ret = __hash_page(ea, access, vsid, ptep, trap);
 	spin_unlock(&mm->page_table_lock);
 
 	return ret;

arch/ppc64/kernel/pSeries_htab.c

@@ -215,7 +215,7 @@ static inline void set_pp_bit(unsigned long pp, HPTE *addr)
 
 	__asm__ __volatile__(
 	"1:	ldarx	%0,0,%3\n\
-		rldimi	%0,%2,0,62\n\
+		rldimi	%0,%2,0,61\n\
 		stdcx.	%0,0,%3\n\
 		bne	1b"
 	: "=&r" (old), "=m" (*p)
@@ -266,8 +266,6 @@ static long pSeries_hpte_updatepp(unsigned long slot, unsigned long newpp,
 	unsigned long vpn, avpn;
 	unsigned long flags;
 
-	udbg_printf("updatepp\n");
-
 	if (large)
 		vpn = va >> LARGE_PAGE_SHIFT;
 	else

arch/ppc64/kernel/pSeries_lpar.c

@@ -647,11 +647,9 @@ static long pSeries_lpar_hpte_updatepp(unsigned long slot, unsigned long newpp,
 {
 	unsigned long lpar_rc;
 	unsigned long flags;
-	flags = (newpp & 3) | H_AVPN;
+	flags = (newpp & 7) | H_AVPN;
 	unsigned long vpn = va >> PAGE_SHIFT;
 
-	udbg_printf("updatepp\n");
-
 	lpar_rc = plpar_pte_protect(flags, slot, (vpn >> 4) & ~0x7fUL);
 
 	if (lpar_rc == H_Not_Found) {

arch/ppc64/mm/init.c

@@ -113,13 +113,10 @@ void show_mem(void)
 			reserved++;
 		else if (PageSwapCache(mem_map+i))
 			cached++;
-		else if (!atomic_read(&mem_map[i].count))
-			free++;
-		else
-			shared += atomic_read(&mem_map[i].count) - 1;
+		else if (page_count(mem_map+i))
+			shared += page_count(mem_map+i) - 1;
 	}
 	printk("%d pages of RAM\n",total);
-	printk("%d free pages\n",free);
 	printk("%d reserved pages\n",reserved);
 	printk("%d pages shared\n",shared);
 	printk("%d pages swap cached\n",cached);
@@ -575,6 +572,12 @@ void flush_dcache_page(struct page *page)
 
 void flush_icache_page(struct vm_area_struct *vma, struct page *page)
 {
+	if (__is_processor(PV_POWER4))
+		return;
+
+	if ((vma->vm_flags & VM_EXEC) == 0)
+		return;
+
 	if (page->mapping && !PageReserved(page)
 	    && !test_bit(PG_arch_1, &page->flags)) {
 		__flush_dcache_icache(page_address(page));
@@ -585,13 +588,32 @@ void flush_icache_page(struct vm_area_struct *vma, struct page *page)
 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
 {
 	clear_page(page);
+
+	/* XXX we shouldnt have to do this, but glibc requires it */
+	if (__is_processor(PV_POWER4))
+		clear_bit(PG_arch_1, &pg->flags);
+	else
+		__flush_dcache_icache(page);
 }
 
 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 		    struct page *pg)
 {
 	copy_page(vto, vfrom);
-	__flush_dcache_icache(vto);
+
+	/*
+	 * Unfortunately we havent always marked our GOT and PLT sections
+	 * as executable, so we need to flush all file regions - Anton
+	 */
+#if 0
+	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
+		return;
+#endif
+
+	if (__is_processor(PV_POWER4))
+		clear_bit(PG_arch_1, &pg->flags);
+	else
+		__flush_dcache_icache(vto);
 }
 
 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
@@ -605,7 +627,7 @@ void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
 
 extern pte_t *find_linux_pte(pgd_t *pgdir, unsigned long ea);
 int __hash_page(unsigned long ea, unsigned long access, unsigned long vsid,
-		pte_t *ptep);
+		pte_t *ptep, unsigned long trap);
 
 /*
  * This is called at the end of handling a user page fault, when the
@@ -633,5 +655,6 @@ void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
 	ptep = find_linux_pte(pgdir, ea);
 	vsid = get_vsid(vma->vm_mm->context, ea);
 
-	__hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep);
+	__hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
+		    0x300);
 }