PPC update for the recent changes to the pgd/pmd/pte functions.

This implements ptes-in-highmem for PPC, removes the quicklist
and zero-page stuff.  PTEs in highmem on SMP turned out to need
some significant changes to avoid deadlocks on the hash_table_lock
(now renamed to mmu_hash_lock).  The PMDs now contain the physical
address of the PTE page rather than the virtual address.
Anything that takes the mmu_hash_lock now operates with the DMMU
off to avoid MMU hash-table misses.

arch/ppc/config.in

@@ -258,8 +258,9 @@ endmenu
 mainmenu_option next_comment
 comment 'General setup'
 
-bool 'Prompt for advanced kernel configuration options' CONFIG_ADVANCED_OPTIONS
 bool 'High memory support' CONFIG_HIGHMEM
+dep_bool 'Support for PTEs in high memory' CONFIG_HIGHPTE
+bool 'Prompt for advanced kernel configuration options' CONFIG_ADVANCED_OPTIONS
 if [ "$CONFIG_ADVANCED_OPTIONS" = "y" ]; then
   if [ "$CONFIG_HIGHMEM" = "y" ]; then
     bool "  Set high memory pool address" CONFIG_HIGHMEM_START_BOOL

arch/ppc/kernel/entry.S

@@ -288,15 +288,12 @@ ret_from_syscall_2:
  */	
 _GLOBAL(_switch)
 	stwu	r1,-INT_FRAME_SIZE(r1)
-	stw	r0,GPR0(r1)
-	lwz	r0,0(r1)
-	stw	r0,GPR1(r1)
+	mflr	r0
+	stw	r0,INT_FRAME_SIZE+4(r1)
 	/* r3-r13 are caller saved -- Cort */
-	SAVE_GPR(2, r1)
 	SAVE_8GPRS(14, r1)
 	SAVE_10GPRS(22, r1)
-	mflr	r20		/* Return to switch caller */
-	stw	r20,INT_FRAME_SIZE+4(r1)
+	stw	r0,_NIP(r1)	/* Return to switch caller */
 	mfmsr	r22
 	li	r0,MSR_FP	/* Disable floating-point */
 #ifdef CONFIG_ALTIVEC
@@ -309,17 +306,9 @@ END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
 	andc	r22,r22,r0
 	mtmsr	r22
 	isync
-1:	stw	r20,_NIP(r1)
-	stw	r22,_MSR(r1)
-	stw	r20,_LINK(r1)
+1:	stw	r22,_MSR(r1)
 	mfcr	r20
-	mfctr	r22
-	mfspr	r23,XER
 	stw	r20,_CCR(r1)
-	stw	r22,_CTR(r1)
-	stw	r23,_XER(r1)
-	li	r0,0x0ff0
-	stw	r0,TRAP(r1)
 	stw	r1,KSP(r3)	/* Set old stack pointer */
 
 	tophys(r0,r4)

arch/ppc/kernel/iSeries_head.S

@@ -839,7 +839,7 @@ hash_page:
 
 	/* 
 	 * We hard enable here (but first soft disable) so that the hash_page
-	 * code can spin on the hash_table_lock without problem on a shared
+	 * code can spin on the mmu_hash_lock without problem on a shared
 	 * processor
 	 */
 	li	r0,0

arch/ppc/kernel/iSeries_misc.S

@@ -29,7 +29,7 @@
 	.align	5
 
 #ifdef CONFIG_SMP
-	.comm	hash_table_lock,4
+	.comm	mmu_hash_lock,4
 #endif /* CONFIG_SMP */
 
 _GLOBAL(is_msr_enabled)

arch/ppc/kernel/idle.c

@@ -47,17 +47,10 @@ extern unsigned long yield_count;
 #define run_light_on(x)         do { } while (0)
 #endif /* CONFIG_PPC_ISERIES */
 
-void zero_paged(void);
 void power_save(void);
 
-unsigned long zero_paged_on = 0;
-unsigned long powersave_nap = 0;
-
-unsigned long *zero_cache;    /* head linked list of pre-zero'd pages */
-atomic_t zerototal;      /* # pages zero'd over time */
-atomic_t zeropage_hits;  /* # zero'd pages request that we've done */
-atomic_t zero_sz;	      /* # currently pre-zero'd pages */
-atomic_t zeropage_calls; /* # zero'd pages request that've been made */
+unsigned long zero_paged_on;
+unsigned long powersave_nap;
 
 int idled(void)
 {
@@ -92,7 +85,6 @@ int idled(void)
 		if (need_resched()) {
 			run_light_on(1);
 			schedule();
-			check_pgt_cache();
 		}
 #ifdef CONFIG_PPC_ISERIES
 		else {
@@ -115,141 +107,6 @@ int cpu_idle(void)
 	return 0; 
 }
 
-#if 0
-/*
- * Returns a pre-zero'd page from the list otherwise returns
- * NULL.
- */
-unsigned long get_zero_page_fast(void)
-{
-	unsigned long page = 0;
-
-	atomic_inc(&zero_cache_calls);
-	if ( zero_quicklist )
-	{
-		/* atomically remove this page from the list */
-		register unsigned long tmp;
-		asm (	"101:lwarx  %1,0,%3\n"  /* reserve zero_cache */
-			"    lwz    %0,0(%1)\n" /* get next -- new zero_cache */
-			PPC405_ERR77(0,%3)
-			"    stwcx. %0,0,%3\n"  /* update zero_cache */
-			"    bne-   101b\n"     /* if lost reservation try again */
-			: "=&r" (tmp), "=&r" (page), "+m" (zero_cache)
-			: "r" (&zero_quicklist)
-			: "cc" );
-#ifdef CONFIG_SMP
-		/* if another cpu beat us above this can happen -- Cort */
-		if ( page == 0 ) 
-			return 0;
-#endif /* CONFIG_SMP */		
-		/* we can update zerocount after the fact since it is not
-		 * used for anything but control of a loop which doesn't
-		 * matter since it won't affect anything if it zeros one
-		 * less page -- Cort
-		 */
-		atomic_inc((atomic_t *)&zero_cache_hits);
-		atomic_dec((atomic_t *)&zero_cache_sz);
-		
-		/* zero out the pointer to next in the page */
-		*(unsigned long *)page = 0;
-		return page;
-	}
-	return 0;
-}
-
-/*
- * Experimental stuff to zero out pages in the idle task
- * to speed up get_free_pages(). Zero's out pages until
- * we've reached the limit of zero'd pages.  We handle
- * reschedule()'s in here so when we return we know we've
- * zero'd all we need to for now.
- */
-int zero_cache_water[2] = { 25, 96 }; /* high and low water marks for zero cache */
-void zero_paged(void)
-{
-	unsigned long pageptr = 0;	/* current page being zero'd */
-	unsigned long bytecount = 0;  
-        register unsigned long tmp;
-	pte_t *pte;
-
-	if ( atomic_read(&zero_cache_sz) >= zero_cache_water[0] )
-		return;
-	while ( (atomic_read(&zero_cache_sz) < zero_cache_water[1]) && !need_resched() )
-	{
-		/*
-		 * Mark a page as reserved so we can mess with it
-		 * If we're interrupted we keep this page and our place in it
-		 * since we validly hold it and it's reserved for us.
-		 */
-		pageptr = __get_free_pages(GFP_ATOMIC, 0);
-		if ( !pageptr )
-			return;
-		
-		cond_resched();
-		
-		/*
-		 * Make the page no cache so we don't blow our cache with 0's
-		 */
-		pte = find_pte(&init_mm, pageptr);
-		if ( !pte )
-		{
-			printk("pte NULL in zero_paged()\n");
-			return;
-		}
-		
-		pte_uncache(*pte);
-		flush_tlb_page(find_vma(&init_mm,pageptr),pageptr);
-		/*
-		 * Important here to not take time away from real processes.
-		 */
-		for ( bytecount = 0; bytecount < PAGE_SIZE ; bytecount += 4 )
-		{
-			cond_resched();
-			*(unsigned long *)(bytecount + pageptr) = 0;
-		}
-		
-		/*
-		 * If we finished zero-ing out a page add this page to
-		 * the zero_cache atomically -- we can't use
-		 * down/up since we can't sleep in idle.
-		 * Disabling interrupts is also a bad idea since we would
-		 * steal time away from real processes.
-		 * We can also have several zero_paged's running
-		 * on different processors so we can't interfere with them.
-		 * So we update the list atomically without locking it.
-		 * -- Cort
-		 */
-		
-		/* turn cache on for this page */
-		pte_cache(*pte);
-		flush_tlb_page(find_vma(&init_mm,pageptr),pageptr);
-		/* atomically add this page to the list */
-		asm (	"101:lwarx  %0,0,%2\n"  /* reserve zero_cache */
-			"    stw    %0,0(%3)\n" /* update *pageptr */
-#ifdef CONFIG_SMP
-			"    sync\n"            /* let store settle */
-#endif			
- 			PPC405_ERR77(0,%2)
-			"    stwcx. %3,0,%2\n"  /* update zero_cache in mem */
-			"    bne-   101b\n"     /* if lost reservation try again */
-			: "=&r" (tmp), "+m" (zero_quicklist)
-			: "r" (&zero_quicklist), "r" (pageptr)
-			: "cc" );
-		/*
-		 * This variable is used in the above loop and nowhere
-		 * else so the worst that could happen is we would
-		 * zero out one more or one less page than we want
-		 * per processor on the machine.  This is because
-		 * we could add our page to the list but not have
-		 * zerocount updated yet when another processor
-		 * reads it.  -- Cort
-		 */
-		atomic_inc((atomic_t *)&zero_cache_sz);
-		atomic_inc((atomic_t *)&zero_cache_total);
-	}
-}
-#endif /* 0 */
-
 void power_save(void)
 {
 	unsigned long hid0;

arch/ppc/kernel/misc.S

@@ -369,34 +369,39 @@ _GLOBAL(_tlbia)
 	isync
 #else
 #if defined(CONFIG_SMP)
+	rlwinm	r8,r1,0,0,18
+	lwz	r8,TI_CPU(r8)
+	oris	r8,r8,10
 	mfmsr	r10
 	SYNC
 	rlwinm	r0,r10,0,17,15		/* clear bit 16 (MSR_EE) */
+	rlwinm	r0,r0,0,28,26		/* clear DR */
 	mtmsr	r0
-	SYNC
-	lis	r9,hash_table_lock@h
-	ori	r9,r9,hash_table_lock@l
-	rlwinm	r8,r1,0,0,18
-	lwz	r8,TI_CPU(r8)
-	oris	r8,r8,10
+	SYNC_601
+	isync
+	lis	r9,mmu_hash_lock@h
+	ori	r9,r9,mmu_hash_lock@l
+	tophys(r9,r9)
 10:	lwarx	r7,0,r9
 	cmpi	0,r7,0
 	bne-	10b
-	/* No 405 Erratum 77 fix needed here, because 4xx can't do SMP */
 	stwcx.	r8,0,r9
 	bne-	10b
-#endif /* CONFIG_SMP */
 	sync
 	tlbia
 	sync
-#ifdef CONFIG_SMP
 	TLBSYNC
 	li	r0,0
-	stw	r0,0(r9)		/* clear hash_table_lock */
+	stw	r0,0(r9)		/* clear mmu_hash_lock */
 	mtmsr	r10
-	SYNC
-#endif
-#endif
+	SYNC_601
+	isync
+#else /* CONFIG_SMP */
+	sync
+	tlbia
+	sync
+#endif /* CONFIG_SMP */
+#endif /* CONFIG_4xx */
 	blr	
 
 /*
@@ -415,33 +420,37 @@ _GLOBAL(_tlbie)
 10:
 #else
 #if defined(CONFIG_SMP)
+	rlwinm	r8,r1,0,0,18
+	lwz	r8,TI_CPU(r8)
+	oris	r8,r8,11
 	mfmsr	r10
 	SYNC
 	rlwinm	r0,r10,0,17,15		/* clear bit 16 (MSR_EE) */
+	rlwinm	r0,r0,0,28,26		/* clear DR */
 	mtmsr	r0
-	SYNC
-	lis	r9,hash_table_lock@h
-	ori	r9,r9,hash_table_lock@l
-	rlwinm	r8,r1,0,0,18
-	lwz	r8,TI_CPU(r8)
-	oris	r8,r8,11
+	SYNC_601
+	isync
+	lis	r9,mmu_hash_lock@h
+	ori	r9,r9,mmu_hash_lock@l
+	tophys(r9,r9)
 10:	lwarx	r7,0,r9
 	cmpi	0,r7,0
 	bne-	10b
-	PPC405_ERR77(0,r9)
 	stwcx.	r8,0,r9
 	bne-	10b
 	eieio
-#endif /* CONFIG_SMP */
 	tlbie	r3
 	sync
-#ifdef CONFIG_SMP
 	TLBSYNC
 	li	r0,0
-	stw	r0,0(r9)		/* clear hash_table_lock */
+	stw	r0,0(r9)		/* clear mmu_hash_lock */
 	mtmsr	r10
-	SYNC
-#endif
+	SYNC_601
+	isync
+#else /* CONFIG_SMP */
+	tlbie	r3
+	sync
+#endif /* CONFIG_SMP */
 #endif /* CONFIG_4xx */
 	blr
 
@@ -629,6 +638,40 @@ _GLOBAL(__flush_dcache_icache)
 	sync
 	isync
 	blr
+
+/*
+ * Flush a particular page from the data cache to RAM, identified
+ * by its physical address.  We turn off the MMU so we can just use
+ * the physical address (this may be a highmem page without a kernel
+ * mapping).
+ *
+ *	void __flush_dcache_icache_phys(unsigned long physaddr)
+ */
+_GLOBAL(__flush_dcache_icache_phys)
+	mfspr	r5,PVR
+	rlwinm	r5,r5,16,16,31
+	cmpi	0,r5,1
+	beqlr					/* for 601, do nothing */
+	mfmsr	r10
+	rlwinm	r0,r10,0,28,26			/* clear DR */
+	mtmsr	r0
+	isync
+	rlwinm	r3,r3,0,0,19			/* Get page base address */
+	li	r4,4096/L1_CACHE_LINE_SIZE	/* Number of lines in a page */
+	mtctr	r4
+	mr	r6,r3
+0:	dcbst	0,r3				/* Write line to ram */
+	addi	r3,r3,L1_CACHE_LINE_SIZE
+	bdnz	0b
+	sync
+	mtctr	r4
+1:	icbi	0,r6
+	addi	r6,r6,L1_CACHE_LINE_SIZE
+	bdnz	1b
+	sync
+	mtmsr	r10				/* restore DR */
+	isync
+	blr
 	
 /*
  * Clear a page using the dcbz instruction, which doesn't cause any

arch/ppc/kernel/ppc_ksyms.c

@@ -218,13 +218,13 @@ EXPORT_SYMBOL(__global_sti);
 EXPORT_SYMBOL(__global_save_flags);
 EXPORT_SYMBOL(__global_restore_flags);
 #ifdef SPINLOCK_DEBUG
-EXPORT_SYMBOL(_spin_lock);
-EXPORT_SYMBOL(_spin_unlock);
-EXPORT_SYMBOL(spin_trylock);
-EXPORT_SYMBOL(_read_lock);
-EXPORT_SYMBOL(_read_unlock);
-EXPORT_SYMBOL(_write_lock);
-EXPORT_SYMBOL(_write_unlock);
+EXPORT_SYMBOL(_raw_spin_lock);
+EXPORT_SYMBOL(_raw_spin_unlock);
+EXPORT_SYMBOL(_raw_spin_trylock);
+EXPORT_SYMBOL(_raw_read_lock);
+EXPORT_SYMBOL(_raw_read_unlock);
+EXPORT_SYMBOL(_raw_write_lock);
+EXPORT_SYMBOL(_raw_write_unlock);
 #endif
 EXPORT_SYMBOL(smp_call_function);
 EXPORT_SYMBOL(smp_hw_index);
@@ -361,7 +361,7 @@ EXPORT_SYMBOL(set_context);
 EXPORT_SYMBOL(handle_mm_fault); /* For MOL */
 EXPORT_SYMBOL_NOVERS(disarm_decr);
 #ifdef CONFIG_PPC_STD_MMU
-EXPORT_SYMBOL(flush_hash_page); /* For MOL */
+EXPORT_SYMBOL(flush_hash_pages); /* For MOL */
 extern long *intercept_table;
 EXPORT_SYMBOL(intercept_table);
 #endif

arch/ppc/kernel/smp.c

@@ -291,26 +291,6 @@ void smp_call_function_interrupt(void)
 		atomic_inc(&call_data->finished);
 }
 
-/*
- * Task migration callback.
- */
-void smp_task_migration_interrupt(void *new_task)
-{
-	task_t *p;
-
-	p = new_task;
-	sched_task_migrated(p);
-}
-
-/*
- * This function sends a 'task migration' IPI to another CPU.
- * Must be called from syscall contexts, with interrupts *enabled*.
- */
-void smp_migrate_task(int cpu, task_t *p)
-{
-	__smp_call_function(smp_task_migration_interrupt, p, 0, cpu);
-}
-
 void __init smp_boot_cpus(void)
 {
 	int i, cpu_nr;

arch/ppc/lib/locks.c

@@ -28,7 +28,7 @@
  * since they may inhibit forward progress by other CPUs in getting
  * a lock.
  */
-static unsigned long __spin_trylock(volatile unsigned long *lock)
+unsigned long __spin_trylock(volatile unsigned long *lock)
 {
 	unsigned long ret;
 

arch/ppc/mm/cachemap.c

@@ -50,8 +50,6 @@
 #include <asm/smp.h>
 #include <asm/machdep.h>
 
-extern int get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep);
-
 /* This function will allocate the requested contiguous pages and
  * map them into the kernel's vmalloc() space.  This is done so we
  * get unique mapping for these pages, outside of the kernel's 1:1

arch/ppc/mm/fault.c

@@ -154,6 +154,7 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 
 		/* Since 4xx supports per-page execute permission,
 		 * we lazily flush dcache to icache. */
+		ptep = NULL;
 		if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
 			struct page *page = pte_page(*ptep);
 
@@ -164,9 +165,12 @@ void do_page_fault(struct pt_regs *regs, unsigned long address,
 			}
 			pte_update(ptep, 0, _PAGE_HWEXEC);
 			_tlbie(address);
+			pte_unmap(ptep);
 			up_read(&mm->mmap_sem);
 			return;
 		}
+		if (ptep != NULL)
+			pte_unmap(ptep);
 #endif
 	/* a read */
 	} else {
@@ -289,27 +293,18 @@ pte_t *va_to_pte(unsigned long address)
 	struct mm_struct *mm;
 
 	if (address < TASK_SIZE)
-		mm = current->mm;
-	else
-		mm = &init_mm;
+		return NULL;
 
-	dir = pgd_offset(mm, address & PAGE_MASK);
+	dir = pgd_offset(&init_mm, address);
 	if (dir) {
 		pmd = pmd_offset(dir, address & PAGE_MASK);
 		if (pmd && pmd_present(*pmd)) {
-			pte = pte_offset(pmd, address & PAGE_MASK);
-			if (pte && pte_present(*pte)) {
+			pte = pte_offset_kernel(pmd, address & PAGE_MASK);
+			if (pte && pte_present(*pte))
 				return(pte);
-			}
-		}
-		else {
-			return (0);
 		}
 	}
-	else {
-		return (0);
-	}
-	return (0);
+	return NULL;
 }
 
 unsigned long va_to_phys(unsigned long address)
@@ -334,7 +329,7 @@ print_8xx_pte(struct mm_struct *mm, unsigned long addr)
         if (pgd) {
                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
                 if (pmd && pmd_present(*pmd)) {
-                        pte = pte_offset(pmd, addr & PAGE_MASK);
+                        pte = pte_offset_kernel(pmd, addr & PAGE_MASK);
                         if (pte) {
                                 printk(" (0x%08lx)->(0x%08lx)->0x%08lx\n",
                                         (long)pgd, (long)pte, (long)pte_val(*pte));
@@ -375,9 +370,9 @@ get_8xx_pte(struct mm_struct *mm, unsigned long addr)
         if (pgd) {
                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
                 if (pmd && pmd_present(*pmd)) {
-                        pte = pte_offset(pmd, addr & PAGE_MASK);
+                        pte = pte_offset_kernel(pmd, addr & PAGE_MASK);
                         if (pte) {
-                                        retval = (int)pte_val(*pte);
+				retval = (int)pte_val(*pte);
                         }
                 }
         }

arch/ppc/mm/iSeries_hashtable.c

@@ -39,7 +39,7 @@
 
 int iSeries_hpt_loaded;
 
-static spinlock_t hash_table_lock = SPIN_LOCK_UNLOCKED;
+static spinlock_t mmu_hash_lock = SPIN_LOCK_UNLOCKED;
 
 extern unsigned long htab_reloads;	// Defined in ppc/kernel/ppc_htab.c
 extern unsigned long htab_evicts;
@@ -159,10 +159,10 @@ int iSeries_create_hpte( unsigned long access, unsigned long va )
 	
 	access |= _PAGE_PRESENT;		// _PAGE_PRESENT also needed
 
-	spin_lock( &hash_table_lock );
+	spin_lock( &mmu_hash_lock );
 	// check if pte is in the required state
 	if ( ( access & ~(pte_val(*pt)) ) ) {
-		spin_unlock( &hash_table_lock );
+		spin_unlock( &mmu_hash_lock );
 		return 1;
 	}
 
@@ -177,18 +177,18 @@ int iSeries_create_hpte( unsigned long access, unsigned long va )
 		      va, 
 		      pte_val(*pt));
 
-	spin_unlock( &hash_table_lock );
+	spin_unlock( &mmu_hash_lock );
 	return 0;
 }
 
 void add_hash_page(unsigned context, unsigned long va, pte_t *ptep)
 {
-	spin_lock( &hash_table_lock );
+	spin_lock( &mmu_hash_lock );
 	pte_update(ptep,0,_PAGE_HASHPTE);
 	__create_hpte(CTX_TO_VSID(context, va), 
 		      va, 
 		      pte_val(*ptep));
-	spin_unlock( &hash_table_lock );
+	spin_unlock( &mmu_hash_lock );
 }
 
 int flush_hash_page(unsigned context, unsigned long va, pte_t *ptep)
@@ -208,7 +208,7 @@ int flush_hash_page(unsigned context, unsigned long va, pte_t *ptep)
 	hpte1Ptr = hpte0Ptr + 1;
 	*hpte0Ptr = *hpte1Ptr = 0;
   
-	spin_lock( &hash_table_lock );
+	spin_lock( &mmu_hash_lock );
 	rtnIndex = HvCallHpt_findValid( &hpte, vpn );
   
 	if ( hpte.v ) {
@@ -217,7 +217,7 @@ int flush_hash_page(unsigned context, unsigned long va, pte_t *ptep)
 		rc = 0;
 	} else
 		rc = 1;
-	spin_unlock( &hash_table_lock );
+	spin_unlock( &mmu_hash_lock );
 	return rc;
 }
 

arch/ppc/mm/iSeries_mmu.c

@@ -175,12 +175,13 @@ void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 	pte_t *ptep;
 	static int nopreload;
 
-	if (nopreload)
+	if (nopreload || address >= TASK_SIZE)
 		return;
-	mm = (address < TASK_SIZE)? vma->vm_mm: &init_mm;
+	mm = vma->vm_mm;
 	pmd = pmd_offset(pgd_offset(mm, address), address);
 	if (!pmd_none(*pmd)) {
-		ptep = pte_offset(pmd, address);
+		ptep = pte_offset_map(pmd, address);
 		add_hash_page(mm->context, address, ptep);
+		pte_unmap(ptep);
 	}
 }

arch/ppc/mm/init.c

@@ -113,24 +113,6 @@ unsigned long __max_memory;
 /* max amount of low RAM to map in */
 unsigned long __max_low_memory = MAX_LOW_MEM;
 
-int do_check_pgt_cache(int low, int high)
-{
-	int freed = 0;
-	if (pgtable_cache_size > high) {
-		do {
-                        if (pgd_quicklist) {
-				free_pgd_slow(get_pgd_fast());
-				freed++;
-			}
-			if (pte_quicklist) {
-				pte_free_slow(pte_alloc_one_fast(NULL, 0));
-				freed++;
-			}
-		} while (pgtable_cache_size > low);
-	}
-	return freed;
-}
-
 void show_mem(void)
 {
 	int i,free = 0,total = 0,reserved = 0;
@@ -160,7 +142,6 @@ void show_mem(void)
 	printk("%d reserved pages\n",reserved);
 	printk("%d pages shared\n",shared);
 	printk("%d pages swap cached\n",cached);
-	printk("%d pages in page table cache\n",(int)pgtable_cache_size);
 	show_buffers();
 }
 
@@ -396,9 +377,11 @@ void __init paging_init(void)
 
 #ifdef CONFIG_HIGHMEM
 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
-	pkmap_page_table = pte_offset(pmd_offset(pgd_offset_k(PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
+	pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
+			(PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
 	map_page(KMAP_FIX_BEGIN, 0, 0);	/* XXX gross */
-	kmap_pte = pte_offset(pmd_offset(pgd_offset_k(KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
+	kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
+			(KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
 	kmap_prot = PAGE_KERNEL;
 #endif /* CONFIG_HIGHMEM */
 
@@ -588,10 +571,12 @@ void flush_dcache_page(struct page *page)
 
 void flush_icache_page(struct vm_area_struct *vma, struct page *page)
 {
+	unsigned long phys;
+
 	if (page->mapping && !PageReserved(page)
 	    && !test_bit(PG_arch_1, &page->flags)) {
-		__flush_dcache_icache(kmap(page));
-		kunmap(page);
+		phys = ((page - mem_map) << PAGE_SHIFT) + PPC_MEMSTART;
+		__flush_dcache_icache_phys(phys);
 		set_bit(PG_arch_1, &page->flags);
 	}
 }

arch/ppc/mm/mmu_decl.h

@@ -61,11 +61,12 @@ extern void MMU_init_hw(void);
  * which includes all new 82xx processors.  We need tlbie/tlbsync here
  * in that case (I think). -- Dan.
  */
-static inline void flush_HPTE(unsigned context, unsigned long va, pte_t *pg)
+static inline void flush_HPTE(unsigned context, unsigned long va,
+			      unsigned long pdval)
 {
 	if ((Hash != 0) &&
 	    (cur_cpu_spec[0]->cpu_features & CPU_FTR_HPTE_TABLE))
-		flush_hash_page(0, va, pg);
+		flush_hash_pages(0, va, pdval, 1);
 	else
 		_tlbie(va);
 }

arch/ppc/mm/pgtable.c

@@ -39,10 +39,6 @@ unsigned long ioremap_base;
 unsigned long ioremap_bot;
 int io_bat_index;
 
-#ifndef CONFIG_SMP
-struct pgtable_cache_struct quicklists;
-#endif
-
 #if defined(CONFIG_6xx) || defined(CONFIG_POWER3)
 #define HAVE_BATS	1
 #endif
@@ -173,12 +169,12 @@ map_page(unsigned long va, unsigned long pa, int flags)
 	/* Use upper 10 bits of VA to index the first level map */
 	pd = pmd_offset(pgd_offset_k(va), va);
 	/* Use middle 10 bits of VA to index the second-level map */
-	pg = pte_alloc(&init_mm, pd, va);
+	pg = pte_alloc_kernel(&init_mm, pd, va);
 	if (pg != 0) {
 		err = 0;
 		set_pte(pg, mk_pte_phys(pa & PAGE_MASK, __pgprot(flags)));
 		if (mem_init_done)
-			flush_HPTE(0, va, pg);
+			flush_HPTE(0, va, pmd_val(*pd));
 	}
 	spin_unlock(&init_mm.page_table_lock);
 	return err;
@@ -272,10 +268,11 @@ get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep)
         if (pgd) {
                 pmd = pmd_offset(pgd, addr & PAGE_MASK);
                 if (pmd_present(*pmd)) {
-                        pte = pte_offset(pmd, addr & PAGE_MASK);
+                        pte = pte_offset_map(pmd, addr & PAGE_MASK);
                         if (pte) {
 				retval = 1;
 				*ptep = pte;
+				/* XXX caller needs to do pte_unmap, yuck */
                         }
                 }
         }
@@ -312,8 +309,10 @@ unsigned long iopa(unsigned long addr)
 		mm = &init_mm;
 	
 	pa = 0;
-	if (get_pteptr(mm, addr, &pte))
+	if (get_pteptr(mm, addr, &pte)) {
 		pa = (pte_val(*pte) & PAGE_MASK) | (addr & ~PAGE_MASK);
+		pte_unmap(pte);
+	}
 
 	return(pa);
 }

arch/ppc/mm/ppc_mmu.c

@@ -33,6 +33,7 @@
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include <linux/init.h>
+#include <linux/highmem.h>
 
 #include <asm/prom.h>
 #include <asm/mmu.h>
@@ -289,7 +290,6 @@ void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 {
 	struct mm_struct *mm;
 	pmd_t *pmd;
-	pte_t *ptep;
 	static int nopreload;
 
 	if (Hash == 0 || nopreload)
@@ -299,8 +299,6 @@ void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
 		return;
 	mm = (address < TASK_SIZE)? vma->vm_mm: &init_mm;
 	pmd = pmd_offset(pgd_offset(mm, address), address);
-	if (!pmd_none(*pmd)) {
-		ptep = pte_offset(pmd, address);
-		add_hash_page(mm->context, address, ptep);
-	}
+	if (!pmd_none(*pmd))
+		add_hash_page(mm->context, address, pmd_val(*pmd));
 }

arch/ppc/mm/tlb.c

@@ -30,6 +30,7 @@
 #include <linux/kernel.h>
 #include <linux/mm.h>
 #include <linux/init.h>
+#include <linux/highmem.h>
 
 #include "mmu_decl.h"
 
@@ -104,7 +105,6 @@ local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
 {
 	struct mm_struct *mm;
 	pmd_t *pmd;
-	pte_t *pte;
 
 	if (Hash == 0) {
 		_tlbie(vmaddr);
@@ -112,11 +112,8 @@ local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
 	}
 	mm = (vmaddr < TASK_SIZE)? vma->vm_mm: &init_mm;
 	pmd = pmd_offset(pgd_offset(mm, vmaddr), vmaddr);
-	if (!pmd_none(*pmd)) {
-		pte = pte_offset(pmd, vmaddr);
-		if (pte_val(*pte) & _PAGE_HASHPTE)
-			flush_hash_page(mm->context, vmaddr, pte);
-	}
+	if (!pmd_none(*pmd))
+		flush_hash_pages(mm->context, vmaddr, pmd_val(*pmd), 1);
 #ifdef CONFIG_SMP
 	smp_send_tlb_invalidate(0);
 #endif	
@@ -133,8 +130,8 @@ local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned
 {
 	struct mm_struct *mm = vma->vm_mm;
 	pmd_t *pmd;
-	pte_t *pte;
 	unsigned long pmd_end;
+	int count;
 	unsigned int ctx = mm->context;
 
 	if (Hash == 0) {
@@ -144,24 +141,21 @@ local_flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned
 	start &= PAGE_MASK;
 	if (start >= end)
 		return;
+	end = (end - 1) | ~PAGE_MASK;
 	pmd = pmd_offset(pgd_offset(mm, start), start);
-	do {
-		pmd_end = (start + PGDIR_SIZE) & PGDIR_MASK;
+	for (;;) {
+		pmd_end = ((start + PGDIR_SIZE) & PGDIR_MASK) - 1;
+		if (pmd_end > end)
+			pmd_end = end;
 		if (!pmd_none(*pmd)) {
-			if (!pmd_end || pmd_end > end)
-				pmd_end = end;
-			pte = pte_offset(pmd, start);
-			do {
-				if ((pte_val(*pte) & _PAGE_HASHPTE) != 0)
-					flush_hash_page(ctx, start, pte);
-				start += PAGE_SIZE;
-				++pte;
-			} while (start && start < pmd_end);
-		} else {
-			start = pmd_end;
+			count = ((pmd_end - start) >> PAGE_SHIFT) + 1;
+			flush_hash_pages(ctx, start, pmd_val(*pmd), count);
 		}
+		if (pmd_end == end)
+			break;
+		start = pmd_end + 1;
 		++pmd;
-	} while (start && start < end);
+	}
 
 #ifdef CONFIG_SMP
 	smp_send_tlb_invalidate(0);

include/asm-ppc/highmem.h

@@ -106,7 +106,7 @@ static inline void *kmap_atomic(struct page *page, enum km_type type)
 static inline void kunmap_atomic(void *kvaddr, enum km_type type)
 {
 #if HIGHMEM_DEBUG
-	unsigned long vaddr = (unsigned long) kvaddr;
+	unsigned long vaddr = (unsigned long) kvaddr & PAGE_MASK;
 	unsigned int idx = type + KM_TYPE_NR*smp_processor_id();
 
 	if (vaddr < KMAP_FIX_BEGIN) // FIXME

include/asm-ppc/kmap_types.h

 /*
- * BK Id: SCCS/s.kmap_types.h 1.9 08/29/01 14:03:05 paulus
+ * BK Id: %F% %I% %G% %U% %#%
  */
 #ifdef __KERNEL__
 #ifndef _ASM_KMAP_TYPES_H
@@ -12,6 +12,8 @@ enum km_type {
 	KM_USER0,
 	KM_USER1,
 	KM_BIO_IRQ,
+	KM_PTE0,
+	KM_PTE1,
 	KM_TYPE_NR
 };
 

include/asm-ppc/pgalloc.h

 /*
- * BK Id: SCCS/s.pgalloc.h 1.9 05/17/01 18:14:25 cort
+ * BK Id: %F% %I% %G% %U% %#%
  */
 #ifdef __KERNEL__
 #ifndef _PPC_PGALLOC_H
@@ -7,55 +7,12 @@
 
 #include <linux/config.h>
 #include <linux/threads.h>
+#include <linux/highmem.h>
 #include <asm/processor.h>
 
-/*
- * This is handled very differently on the PPC since out page tables
- * are all 0's and I want to be able to use these zero'd pages elsewhere
- * as well - it gives us quite a speedup.
- *
- * Note that the SMP/UP versions are the same but we don't need a
- * per cpu list of zero pages because we do the zero-ing with the cache
- * off and the access routines are lock-free but the pgt cache stuff
- * is per-cpu since it isn't done with any lock-free access routines
- * (although I think we need arch-specific routines so I can do lock-free).
- *
- * I need to generalize this so we can use it for other arch's as well.
- * -- Cort
- */
-#ifdef CONFIG_SMP
-#define quicklists	cpu_data[smp_processor_id()]
-#else
-extern struct pgtable_cache_struct {
-	unsigned long *pgd_cache;
-	unsigned long *pte_cache;
-	unsigned long pgtable_cache_sz;
-} quicklists;
-#endif
-
-#define pgd_quicklist 		(quicklists.pgd_cache)
-#define pmd_quicklist 		((unsigned long *)0)
-#define pte_quicklist 		(quicklists.pte_cache)
-#define pgtable_cache_size 	(quicklists.pgtable_cache_sz)
-
-extern unsigned long *zero_cache;    /* head linked list of pre-zero'd pages */
-extern atomic_t zero_sz;	     /* # currently pre-zero'd pages */
-extern atomic_t zeropage_hits;	     /* # zero'd pages request that we've done */
-extern atomic_t zeropage_calls;      /* # zero'd pages request that've been made */
-extern atomic_t zerototal;	     /* # pages zero'd over time */
-
-#define zero_quicklist     	(zero_cache)
-#define zero_cache_sz  	 	(zero_sz)
-#define zero_cache_calls 	(zeropage_calls)
-#define zero_cache_hits  	(zeropage_hits)
-#define zero_cache_total 	(zerototal)
-
-/* return a pre-zero'd page from the list, return NULL if none available -- Cort */
-extern unsigned long get_zero_page_fast(void);
-
 extern void __bad_pte(pmd_t *pmd);
 
-extern __inline__ pgd_t *get_pgd_slow(void)
+static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 {
 	pgd_t *ret;
 
@@ -64,85 +21,75 @@ extern __inline__ pgd_t *get_pgd_slow(void)
 	return ret;
 }
 
-extern __inline__ pgd_t *get_pgd_fast(void)
-{
-        unsigned long *ret;
-
-        if ((ret = pgd_quicklist) != NULL) {
-                pgd_quicklist = (unsigned long *)(*ret);
-                ret[0] = 0;
-                pgtable_cache_size--;
-        } else
-                ret = (unsigned long *)get_pgd_slow();
-        return (pgd_t *)ret;
-}
-
-extern __inline__ void free_pgd_fast(pgd_t *pgd)
-{
-        *(unsigned long **)pgd = pgd_quicklist;
-        pgd_quicklist = (unsigned long *) pgd;
-        pgtable_cache_size++;
-}
-
-extern __inline__ void free_pgd_slow(pgd_t *pgd)
+extern __inline__ void pgd_free(pgd_t *pgd)
 {
 	free_page((unsigned long)pgd);
 }
 
-#define pgd_free(pgd)		free_pgd_fast(pgd)
-#define pgd_alloc(mm)		get_pgd_fast()
-
 /*
  * We don't have any real pmd's, and this code never triggers because
  * the pgd will always be present..
  */
-#define pmd_alloc_one_fast(mm, address) ({ BUG(); ((pmd_t *)1); })
 #define pmd_alloc_one(mm,address)       ({ BUG(); ((pmd_t *)2); })
 #define pmd_free(x)                     do { } while (0)
 #define pgd_populate(mm, pmd, pte)      BUG()
 
-static inline pte_t *pte_alloc_one(struct mm_struct *mm, unsigned long address)
+static inline pte_t *
+pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
 {
 	pte_t *pte;
 	extern int mem_init_done;
 	extern void *early_get_page(void);
-
-	if (mem_init_done)
-		pte = (pte_t *) __get_free_page(GFP_KERNEL);
-	else
+	int timeout = 0;
+
+	if (mem_init_done) {
+		while ((pte = (pte_t *) __get_free_page(GFP_KERNEL)) == NULL
+		       && ++timeout < 10) {
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			schedule_timeout(HZ);
+		}
+	} else
 		pte = (pte_t *) early_get_page();
 	if (pte != NULL)
 		clear_page(pte);
 	return pte;
 }
 
-static inline pte_t *pte_alloc_one_fast(struct mm_struct *mm, unsigned long address)
+static inline struct page *
+pte_alloc_one(struct mm_struct *mm, unsigned long address)
 {
-        unsigned long *ret;
+	struct page *pte;
+	int timeout = 0;
+#ifdef CONFIG_HIGHPTE
+	int flags = GFP_KERNEL | __GFP_HIGHMEM;
+#else
+	int flags = GFP_KERNEL;
+#endif
 
-        if ((ret = pte_quicklist) != NULL) {
-                pte_quicklist = (unsigned long *)(*ret);
-                ret[0] = 0;
-                pgtable_cache_size--;
+	while ((pte = alloc_pages(flags, 0)) == NULL) {
+		if (++timeout >= 10)
+			return NULL;
+		set_current_state(TASK_UNINTERRUPTIBLE);
+		schedule_timeout(HZ);
 	}
-        return (pte_t *)ret;
+	clear_highpage(pte);
+	return pte;
 }
 
-extern __inline__ void pte_free_fast(pte_t *pte)
+static inline void pte_free_kernel(pte_t *pte)
 {
-        *(unsigned long **)pte = pte_quicklist;
-        pte_quicklist = (unsigned long *) pte;
-        pgtable_cache_size++;
+	free_page((unsigned long)pte);
 }
 
-extern __inline__ void pte_free_slow(pte_t *pte)
+static inline void pte_free(struct page *pte)
 {
-	free_page((unsigned long)pte);
+	__free_page(pte);
 }
 
-#define pte_free(pte)    pte_free_slow(pte)
-
-#define pmd_populate(mm, pmd, pte)	(pmd_val(*(pmd)) = (unsigned long) (pte))
+#define pmd_populate_kernel(mm, pmd, pte)	\
+		(pmd_val(*(pmd)) = __pa(pte))
+#define pmd_populate(mm, pmd, pte)	\
+		(pmd_val(*(pmd)) = ((pte) - mem_map) << PAGE_SHIFT)
 
 extern int do_check_pgt_cache(int, int);
 

include/asm-ppc/pgtable.h

@@ -13,6 +13,7 @@
 #include <asm/processor.h>		/* For TASK_SIZE */
 #include <asm/mmu.h>
 #include <asm/page.h>
+#include <asm/kmap_types.h>
 
 extern void _tlbie(unsigned long address);
 extern void _tlbia(void);
@@ -98,6 +99,7 @@ extern void flush_icache_user_range(struct vm_area_struct *vma,
 		struct page *page, unsigned long addr, int len);
 extern void flush_icache_range(unsigned long, unsigned long);
 extern void __flush_dcache_icache(void *page_va);
+extern void __flush_dcache_icache_phys(unsigned long physaddr);
 extern void flush_dcache_page(struct page *page);
 extern void flush_icache_page(struct vm_area_struct *vma, struct page *page);
 
@@ -274,7 +276,6 @@ extern unsigned long ioremap_bot, ioremap_base;
 #define _PAGE_HWWRITE	0x0100	/* h/w write enable: never set in Linux PTE */
 #define _PAGE_USER	0x0800	/* One of the PP bits, the other is USER&~RW */
 
-#define _PMD_PRESENT	0x0001
 #define _PMD_PAGE_MASK	0x000c
 #define _PMD_PAGE_8M	0x000c
 
@@ -385,8 +386,8 @@ extern unsigned long empty_zero_page[1024];
 #define pte_clear(ptep)		do { set_pte((ptep), __pte(0)); } while (0)
 
 #define pmd_none(pmd)		(!pmd_val(pmd))
-#define	pmd_bad(pmd)		((pmd_val(pmd) & ~PAGE_MASK) != 0)
-#define	pmd_present(pmd)	((pmd_val(pmd) & PAGE_MASK) != 0)
+#define	pmd_bad(pmd)		(0)
+#define	pmd_present(pmd)	(pmd_val(pmd) != 0)
 #define	pmd_clear(pmdp)		do { pmd_val(*(pmdp)) = 0; } while (0)
 
 #define pte_page(x)		(mem_map+(unsigned long)((pte_val(x)-PPC_MEMSTART) >> PAGE_SHIFT))
@@ -530,7 +531,10 @@ static inline void ptep_mkdirty(pte_t *ptep)
 
 #define pte_same(A,B)	(((pte_val(A) ^ pte_val(B)) & ~_PAGE_HASHPTE) == 0)
 
-#define pmd_page(pmd)	(pmd_val(pmd) & PAGE_MASK)
+#define pmd_page_kernel(pmd)	\
+	((unsigned long) __va(pmd_val(pmd) & PAGE_MASK))
+#define pmd_page(pmd)		\
+	(mem_map + (pmd_val(pmd) >> PAGE_SHIFT))
 
 /* to find an entry in a kernel page-table-directory */
 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
@@ -545,11 +549,18 @@ static inline pmd_t * pmd_offset(pgd_t * dir, unsigned long address)
 	return (pmd_t *) dir;
 }
 
-/* Find an entry in the third-level page table.. */ 
-static inline pte_t * pte_offset(pmd_t * dir, unsigned long address)
-{
-	return (pte_t *) pmd_page(*dir) + ((address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1));
-}
+/* Find an entry in the third-level page table.. */
+#define __pte_offset(address)		\
+	((address >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
+#define pte_offset_kernel(dir, addr)	\
+	((pte_t *) pmd_page_kernel(*(dir)) + __pte_offset(addr))
+#define pte_offset_map(dir, addr)		\
+	((pte_t *) kmap_atomic(pmd_page(*(dir)), KM_PTE0) + __pte_offset(addr))
+#define pte_offset_map_nested(dir, addr)	\
+	((pte_t *) kmap_atomic(pmd_page(*(dir)), KM_PTE1) + __pte_offset(addr))
+
+#define pte_unmap(pte)		kunmap_atomic(pte, KM_PTE0)
+#define pte_unmap_nested(pte)	kunmap_atomic(pte, KM_PTE1)
 
 extern pgd_t swapper_pg_dir[1024];
 extern void paging_init(void);
@@ -558,10 +569,12 @@ extern void paging_init(void);
  * When flushing the tlb entry for a page, we also need to flush the hash
  * table entry.  flush_hash_page is assembler (for speed) in hashtable.S.
  */
-extern int flush_hash_page(unsigned context, unsigned long va, pte_t *ptep);
+extern int flush_hash_pages(unsigned context, unsigned long va,
+			    unsigned long pmdval, int count);
 
 /* Add an HPTE to the hash table */
-extern void add_hash_page(unsigned context, unsigned long va, pte_t *ptep);
+extern void add_hash_page(unsigned context, unsigned long va,
+			  unsigned long pmdval);
 
 /*
  * Encode and decode a swap entry.