[linux-2.6-block.git] / arch / powerpc / mm / tlb_64.c

/*
 * This file contains the routines for flushing entries from the
 * TLB and MMU hash table.
 *
 *  Derived from arch/ppc64/mm/init.c:
 *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
 *
 *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
 *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
 *    Copyright (C) 1996 Paul Mackerras
 *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
 *
 *  Derived from "arch/i386/mm/init.c"
 *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
 *
 *  Dave Engebretsen <engebret@us.ibm.com>
 *      Rework for PPC64 port.
 *
 *  This program is free software; you can redistribute it and/or
 *  modify it under the terms of the GNU General Public License
 *  as published by the Free Software Foundation; either version
 *  2 of the License, or (at your option) any later version.
 */

#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/percpu.h>
#include <linux/hardirq.h>
#include <asm/pgalloc.h>
#include <asm/tlbflush.h>
#include <asm/tlb.h>
#include <asm/bug.h>

DEFINE_PER_CPU(struct ppc64_tlb_batch, ppc64_tlb_batch);

/* This is declared as we are using the more or less generic
 * include/asm-powerpc/tlb.h file -- tgall
 */
DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
DEFINE_PER_CPU(struct pte_freelist_batch *, pte_freelist_cur);
unsigned long pte_freelist_forced_free;

struct pte_freelist_batch
{
	struct rcu_head	rcu;
	unsigned int	index;
	pgtable_free_t	tables[0];
};

DEFINE_PER_CPU(struct pte_freelist_batch *, pte_freelist_cur);
unsigned long pte_freelist_forced_free;

#define PTE_FREELIST_SIZE \
	((PAGE_SIZE - sizeof(struct pte_freelist_batch)) \
	  / sizeof(pgtable_free_t))

#ifdef CONFIG_SMP
static void pte_free_smp_sync(void *arg)
{
	/* Do nothing, just ensure we sync with all CPUs */
}
#endif

/* This is only called when we are critically out of memory
 * (and fail to get a page in pte_free_tlb).
 */
static void pgtable_free_now(pgtable_free_t pgf)
{
	pte_freelist_forced_free++;

	smp_call_function(pte_free_smp_sync, NULL, 0, 1);

	pgtable_free(pgf);
}

static void pte_free_rcu_callback(struct rcu_head *head)
{
	struct pte_freelist_batch *batch =
		container_of(head, struct pte_freelist_batch, rcu);
	unsigned int i;

	for (i = 0; i < batch->index; i++)
		pgtable_free(batch->tables[i]);

	free_page((unsigned long)batch);
}

static void pte_free_submit(struct pte_freelist_batch *batch)
{
	INIT_RCU_HEAD(&batch->rcu);
	call_rcu(&batch->rcu, pte_free_rcu_callback);
}

void pgtable_free_tlb(struct mmu_gather *tlb, pgtable_free_t pgf)
{
	/* This is safe since tlb_gather_mmu has disabled preemption */
        cpumask_t local_cpumask = cpumask_of_cpu(smp_processor_id());
	struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);

	if (atomic_read(&tlb->mm->mm_users) < 2 ||
	    cpus_equal(tlb->mm->cpu_vm_mask, local_cpumask)) {
		pgtable_free(pgf);
		return;
	}

	if (*batchp == NULL) {
		*batchp = (struct pte_freelist_batch *)__get_free_page(GFP_ATOMIC);
		if (*batchp == NULL) {
			pgtable_free_now(pgf);
			return;
		}
		(*batchp)->index = 0;
	}
	(*batchp)->tables[(*batchp)->index++] = pgf;
	if ((*batchp)->index == PTE_FREELIST_SIZE) {
		pte_free_submit(*batchp);
		*batchp = NULL;
	}
}

/*
 * A linux PTE was changed and the corresponding hash table entry
 * neesd to be flushed. This function will either perform the flush
 * immediately or will batch it up if the current CPU has an active
 * batch on it.
 *
 * Must be called from within some kind of spinlock/non-preempt region...
 */
void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
		     pte_t *ptep, unsigned long pte, int huge)
{
	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
	unsigned long vsid, vaddr;
	unsigned int psize;
	real_pte_t rpte;
	int i;

	i = batch->index;

	/* We mask the address for the base page size. Huge pages will
	 * have applied their own masking already
	 */
	addr &= PAGE_MASK;

	/* Get page size (maybe move back to caller) */
	if (huge) {
#ifdef CONFIG_HUGETLB_PAGE
		psize = mmu_huge_psize;
#else
		BUG();
		psize = pte_pagesize_index(pte); /* shutup gcc */
#endif
	} else
		psize = pte_pagesize_index(pte);

	/* Build full vaddr */
	if (!is_kernel_addr(addr)) {
		vsid = get_vsid(mm->context.id, addr);
		WARN_ON(vsid == 0);
	} else
		vsid = get_kernel_vsid(addr);
	vaddr = (vsid << 28 ) | (addr & 0x0fffffff);
	rpte = __real_pte(__pte(pte), ptep);

	/*
	 * Check if we have an active batch on this CPU. If not, just
	 * flush now and return. For now, we don global invalidates
	 * in that case, might be worth testing the mm cpu mask though
	 * and decide to use local invalidates instead...
	 */
	if (!batch->active) {
		flush_hash_page(vaddr, rpte, psize, 0);
		return;
	}

	/*
	 * This can happen when we are in the middle of a TLB batch and
	 * we encounter memory pressure (eg copy_page_range when it tries
	 * to allocate a new pte). If we have to reclaim memory and end
	 * up scanning and resetting referenced bits then our batch context
	 * will change mid stream.
	 *
	 * We also need to ensure only one page size is present in a given
	 * batch
	 */
	if (i != 0 && (mm != batch->mm || batch->psize != psize)) {
		__flush_tlb_pending(batch);
		i = 0;
	}
	if (i == 0) {
		batch->mm = mm;
		batch->psize = psize;
	}
	batch->pte[i] = rpte;
	batch->vaddr[i] = vaddr;
	batch->index = ++i;
	if (i >= PPC64_TLB_BATCH_NR)
		__flush_tlb_pending(batch);
}

/*
 * This function is called when terminating an mmu batch or when a batch
 * is full. It will perform the flush of all the entries currently stored
 * in a batch.
 *
 * Must be called from within some kind of spinlock/non-preempt region...
 */
void __flush_tlb_pending(struct ppc64_tlb_batch *batch)
{
	cpumask_t tmp;
	int i, local = 0;

	i = batch->index;
	tmp = cpumask_of_cpu(smp_processor_id());
	if (cpus_equal(batch->mm->cpu_vm_mask, tmp))
		local = 1;
	if (i == 1)
		flush_hash_page(batch->vaddr[0], batch->pte[0],
				batch->psize, local);
	else
		flush_hash_range(i, local);
	batch->index = 0;
}

void pte_free_finish(void)
{
	/* This is safe since tlb_gather_mmu has disabled preemption */
	struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);

	if (*batchp == NULL)
		return;
	pte_free_submit(*batchp);
	*batchp = NULL;
}
Commit	Line	Data
1da177e4 LT	1	/*
	2	* This file contains the routines for flushing entries from the
	3	* TLB and MMU hash table.
	4	*
	5	* Derived from arch/ppc64/mm/init.c:
	6	* Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
	7	*
	8	* Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
	9	* and Cort Dougan (PReP) (cort@cs.nmt.edu)
	10	* Copyright (C) 1996 Paul Mackerras
	11	* Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
	12	*
	13	* Derived from "arch/i386/mm/init.c"
	14	* Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
	15	*
	16	* Dave Engebretsen <engebret@us.ibm.com>
	17	* Rework for PPC64 port.
	18	*
	19	* This program is free software; you can redistribute it and/or
	20	* modify it under the terms of the GNU General Public License
	21	* as published by the Free Software Foundation; either version
	22	* 2 of the License, or (at your option) any later version.
	23	*/
3c726f8d	24
1da177e4 LT	25	#include <linux/kernel.h>
	26	#include <linux/mm.h>
	27	#include <linux/init.h>
	28	#include <linux/percpu.h>
	29	#include <linux/hardirq.h>
	30	#include <asm/pgalloc.h>
	31	#include <asm/tlbflush.h>
	32	#include <asm/tlb.h>
3c726f8d	33	#include <asm/bug.h>
1da177e4 LT	34
	35	DEFINE_PER_CPU(struct ppc64_tlb_batch, ppc64_tlb_batch);
	36
	37	/* This is declared as we are using the more or less generic
2ef9481e	38	* include/asm-powerpc/tlb.h file -- tgall
1da177e4 LT	39	*/
	40	DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
	41	DEFINE_PER_CPU(struct pte_freelist_batch *, pte_freelist_cur);
	42	unsigned long pte_freelist_forced_free;
	43
e28f7faf DG	44	struct pte_freelist_batch
	45	{
	46	struct rcu_head rcu;
	47	unsigned int index;
	48	pgtable_free_t tables[0];
	49	};
	50
	51	DEFINE_PER_CPU(struct pte_freelist_batch *, pte_freelist_cur);
	52	unsigned long pte_freelist_forced_free;
	53
	54	#define PTE_FREELIST_SIZE \
	55	((PAGE_SIZE - sizeof(struct pte_freelist_batch)) \
	56	/ sizeof(pgtable_free_t))
	57
	58	#ifdef CONFIG_SMP
	59	static void pte_free_smp_sync(void *arg)
	60	{
	61	/* Do nothing, just ensure we sync with all CPUs */
	62	}
	63	#endif
	64
	65	/* This is only called when we are critically out of memory
	66	* (and fail to get a page in pte_free_tlb).
	67	*/
	68	static void pgtable_free_now(pgtable_free_t pgf)
	69	{
	70	pte_freelist_forced_free++;
	71
	72	smp_call_function(pte_free_smp_sync, NULL, 0, 1);
	73
	74	pgtable_free(pgf);
	75	}
	76
	77	static void pte_free_rcu_callback(struct rcu_head *head)
	78	{
	79	struct pte_freelist_batch *batch =
	80	container_of(head, struct pte_freelist_batch, rcu);
	81	unsigned int i;
	82
	83	for (i = 0; i < batch->index; i++)
	84	pgtable_free(batch->tables[i]);
	85
	86	free_page((unsigned long)batch);
	87	}
	88
	89	static void pte_free_submit(struct pte_freelist_batch *batch)
	90	{
	91	INIT_RCU_HEAD(&batch->rcu);
	92	call_rcu(&batch->rcu, pte_free_rcu_callback);
	93	}
	94
	95	void pgtable_free_tlb(struct mmu_gather *tlb, pgtable_free_t pgf)
1da177e4	96	{
01edcd89	97	/* This is safe since tlb_gather_mmu has disabled preemption */
1da177e4 LT	98	cpumask_t local_cpumask = cpumask_of_cpu(smp_processor_id());
	99	struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);
	100
	101	if (atomic_read(&tlb->mm->mm_users) < 2 \|\|
	102	cpus_equal(tlb->mm->cpu_vm_mask, local_cpumask)) {
e28f7faf	103	pgtable_free(pgf);
1da177e4 LT	104	return;
	105	}
	106
	107	if (*batchp == NULL) {
	108	batchp = (struct pte_freelist_batch )__get_free_page(GFP_ATOMIC);
	109	if (*batchp == NULL) {
e28f7faf	110	pgtable_free_now(pgf);
1da177e4 LT	111	return;
	112	}
	113	(*batchp)->index = 0;
	114	}
e28f7faf	115	(batchp)->tables[(batchp)->index++] = pgf;
1da177e4 LT	116	if ((*batchp)->index == PTE_FREELIST_SIZE) {
	117	pte_free_submit(*batchp);
	118	*batchp = NULL;
	119	}
	120	}
	121
	122	/*
a741e679 BH	123	* A linux PTE was changed and the corresponding hash table entry
	124	* neesd to be flushed. This function will either perform the flush
	125	* immediately or will batch it up if the current CPU has an active
	126	* batch on it.
	127	*
	128	* Must be called from within some kind of spinlock/non-preempt region...
1da177e4	129	*/
a741e679 BH	130	void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
a741e679 BH	131	pte_t *ptep, unsigned long pte, int huge)
1da177e4	132	{
1da177e4	133	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);
a741e679	134	unsigned long vsid, vaddr;
bf72aeba	135	unsigned int psize;
a741e679	136	real_pte_t rpte;
61b1a942	137	int i;
1da177e4	138
1da177e4 LT	139	i = batch->index;
1da177e4 LT	140
3c726f8d BH	141	/* We mask the address for the base page size. Huge pages will
	142	* have applied their own masking already
	143	*/
	144	addr &= PAGE_MASK;
	145
	146	/* Get page size (maybe move back to caller) */
	147	if (huge) {
	148	#ifdef CONFIG_HUGETLB_PAGE
	149	psize = mmu_huge_psize;
	150	#else
	151	BUG();
2d69ff32	152	psize = pte_pagesize_index(pte); /* shutup gcc */
3c726f8d	153	#endif
bf72aeba PM	154	} else
bf72aeba PM	155	psize = pte_pagesize_index(pte);
3c726f8d	156
a741e679 BH	157	/* Build full vaddr */
	158	if (!is_kernel_addr(addr)) {
	159	vsid = get_vsid(mm->context.id, addr);
	160	WARN_ON(vsid == 0);
	161	} else
	162	vsid = get_kernel_vsid(addr);
	163	vaddr = (vsid << 28 ) \| (addr & 0x0fffffff);
	164	rpte = __real_pte(__pte(pte), ptep);
	165
	166	/*
	167	* Check if we have an active batch on this CPU. If not, just
	168	* flush now and return. For now, we don global invalidates
	169	* in that case, might be worth testing the mm cpu mask though
	170	* and decide to use local invalidates instead...
	171	*/
	172	if (!batch->active) {
	173	flush_hash_page(vaddr, rpte, psize, 0);
	174	return;
	175	}
	176
1da177e4 LT	177	/*
	178	* This can happen when we are in the middle of a TLB batch and
	179	* we encounter memory pressure (eg copy_page_range when it tries
	180	* to allocate a new pte). If we have to reclaim memory and end
	181	* up scanning and resetting referenced bits then our batch context
	182	* will change mid stream.
3c726f8d BH	183	*
	184	* We also need to ensure only one page size is present in a given
	185	* batch
1da177e4	186	*/
3c726f8d	187	if (i != 0 && (mm != batch->mm \|\| batch->psize != psize)) {
a741e679	188	__flush_tlb_pending(batch);
1da177e4 LT	189	i = 0;
1da177e4 LT	190	}
1da177e4	191	if (i == 0) {
1da177e4	192	batch->mm = mm;
3c726f8d	193	batch->psize = psize;
1da177e4	194	}
a741e679 BH	195	batch->pte[i] = rpte;
a741e679 BH	196	batch->vaddr[i] = vaddr;
1da177e4 LT	197	batch->index = ++i;
1da177e4 LT	198	if (i >= PPC64_TLB_BATCH_NR)
a741e679	199	__flush_tlb_pending(batch);
1da177e4 LT	200	}
1da177e4 LT	201
a741e679 BH	202	/*
	203	* This function is called when terminating an mmu batch or when a batch
	204	* is full. It will perform the flush of all the entries currently stored
	205	* in a batch.
	206	*
	207	* Must be called from within some kind of spinlock/non-preempt region...
	208	*/
1da177e4 LT	209	void __flush_tlb_pending(struct ppc64_tlb_batch *batch)
1da177e4 LT	210	{
1da177e4	211	cpumask_t tmp;
a741e679	212	int i, local = 0;
1da177e4	213
1da177e4	214	i = batch->index;
a741e679	215	tmp = cpumask_of_cpu(smp_processor_id());
1da177e4 LT	216	if (cpus_equal(batch->mm->cpu_vm_mask, tmp))
1da177e4 LT	217	local = 1;
1da177e4	218	if (i == 1)
3c726f8d BH	219	flush_hash_page(batch->vaddr[0], batch->pte[0],
3c726f8d BH	220	batch->psize, local);
1da177e4	221	else
61b1a942	222	flush_hash_range(i, local);
1da177e4	223	batch->index = 0;
1da177e4 LT	224	}
1da177e4 LT	225
1da177e4 LT	226	void pte_free_finish(void)
1da177e4 LT	227	{
01edcd89	228	/* This is safe since tlb_gather_mmu has disabled preemption */
1da177e4 LT	229	struct pte_freelist_batch **batchp = &__get_cpu_var(pte_freelist_cur);
	230
	231	if (*batchp == NULL)
	232	return;
	233	pte_free_submit(*batchp);
	234	*batchp = NULL;
	235	}