tlb.c 13.2 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
2
3
4
5
6
7
8
9
10
/*
 * TLB support routines.
 *
 * Copyright (C) 1998-2001, 2003 Hewlett-Packard Co
 *	David Mosberger-Tang <davidm@hpl.hp.com>
 *
 * 08/02/00 A. Mallick <asit.k.mallick@intel.com>
 *		Modified RID allocation for SMP
 *          Goutham Rao <goutham.rao@intel.com>
 *              IPI based ptc implementation and A-step IPI implementation.
11
12
 * Rohit Seth <rohit.seth@intel.com>
 * Ken Chen <kenneth.w.chen@intel.com>
13
 * Christophe de Dinechin <ddd@hp.com>: Avoid ptc.e on memory allocation
14
15
16
 * Copyright (C) 2007 Intel Corp
 *	Fenghua Yu <fenghua.yu@intel.com>
 *	Add multiple ptc.g/ptc.ga instruction support in global tlb purge.
Linus Torvalds's avatar
Linus Torvalds committed
17
18
19
20
21
22
23
 */
#include <linux/module.h>
#include <linux/init.h>
#include <linux/kernel.h>
#include <linux/sched.h>
#include <linux/smp.h>
#include <linux/mm.h>
24
#include <linux/bootmem.h>
Linus Torvalds's avatar
Linus Torvalds committed
25
26
27
28
29
30

#include <asm/delay.h>
#include <asm/mmu_context.h>
#include <asm/pgalloc.h>
#include <asm/pal.h>
#include <asm/tlbflush.h>
31
#include <asm/dma.h>
32
#include <asm/processor.h>
33
#include <asm/sal.h>
34
#include <asm/tlb.h>
Linus Torvalds's avatar
Linus Torvalds committed
35
36

static struct {
37
	u64 mask;		/* mask of supported purge page-sizes */
38
	unsigned long max_bits;	/* log2 of largest supported purge page-size */
Linus Torvalds's avatar
Linus Torvalds committed
39
40
41
} purge;

struct ia64_ctx ia64_ctx = {
42
43
44
	.lock =	__SPIN_LOCK_UNLOCKED(ia64_ctx.lock),
	.next =	1,
	.max_ctx = ~0U
Linus Torvalds's avatar
Linus Torvalds committed
45
46
47
};

DEFINE_PER_CPU(u8, ia64_need_tlb_flush);
48
49
50
51
DEFINE_PER_CPU(u8, ia64_tr_num);  /*Number of TR slots in current processor*/
DEFINE_PER_CPU(u8, ia64_tr_used); /*Max Slot number used by kernel*/

struct ia64_tr_entry __per_cpu_idtrs[NR_CPUS][2][IA64_TR_ALLOC_MAX];
Linus Torvalds's avatar
Linus Torvalds committed
52

53
54
55
56
57
58
59
60
61
62
63
64
/*
 * Initializes the ia64_ctx.bitmap array based on max_ctx+1.
 * Called after cpu_init() has setup ia64_ctx.max_ctx based on
 * maximum RID that is supported by boot CPU.
 */
void __init
mmu_context_init (void)
{
	ia64_ctx.bitmap = alloc_bootmem((ia64_ctx.max_ctx+1)>>3);
	ia64_ctx.flushmap = alloc_bootmem((ia64_ctx.max_ctx+1)>>3);
}

Linus Torvalds's avatar
Linus Torvalds committed
65
66
67
68
69
70
/*
 * Acquire the ia64_ctx.lock before calling this function!
 */
void
wrap_mmu_context (struct mm_struct *mm)
{
71
	int i, cpu;
72
	unsigned long flush_bit;
Linus Torvalds's avatar
Linus Torvalds committed
73

74
75
76
	for (i=0; i <= ia64_ctx.max_ctx / BITS_PER_LONG; i++) {
		flush_bit = xchg(&ia64_ctx.flushmap[i], 0);
		ia64_ctx.bitmap[i] ^= flush_bit;
Linus Torvalds's avatar
Linus Torvalds committed
77
	}
78
79
80
81
82
83
84
 
	/* use offset at 300 to skip daemons */
	ia64_ctx.next = find_next_zero_bit(ia64_ctx.bitmap,
				ia64_ctx.max_ctx, 300);
	ia64_ctx.limit = find_next_bit(ia64_ctx.bitmap,
				ia64_ctx.max_ctx, ia64_ctx.next);

85
86
87
88
89
90
91
92
93
	/*
	 * can't call flush_tlb_all() here because of race condition
	 * with O(1) scheduler [EF]
	 */
	cpu = get_cpu(); /* prevent preemption/migration */
	for_each_online_cpu(i)
		if (i != cpu)
			per_cpu(ia64_need_tlb_flush, i) = 1;
	put_cpu();
Linus Torvalds's avatar
Linus Torvalds committed
94
95
96
	local_flush_tlb_all();
}

97
98
99
100
101
102
/*
 * Implement "spinaphores" ... like counting semaphores, but they
 * spin instead of sleeping.  If there are ever any other users for
 * this primitive it can be moved up to a spinaphore.h header.
 */
struct spinaphore {
103
104
	unsigned long	ticket;
	unsigned long	serve;
105
106
107
108
};

static inline void spinaphore_init(struct spinaphore *ss, int val)
{
109
110
	ss->ticket = 0;
	ss->serve = val;
111
112
113
114
}

static inline void down_spin(struct spinaphore *ss)
{
115
116
117
118
119
120
121
122
123
124
125
126
127
	unsigned long t = ia64_fetchadd(1, &ss->ticket, acq), serve;

	if (time_before(t, ss->serve))
		return;

	ia64_invala();

	for (;;) {
		asm volatile ("ld4.c.nc %0=[%1]" : "=r"(serve) : "r"(&ss->serve) : "memory");
		if (time_before(t, serve))
			return;
		cpu_relax();
	}
128
129
130
131
}

static inline void up_spin(struct spinaphore *ss)
{
132
	ia64_fetchadd(1, &ss->serve, rel);
133
134
135
136
137
138
139
}

static struct spinaphore ptcg_sem;
static u16 nptcg = 1;
static int need_ptcg_sem = 1;
static int toolatetochangeptcgsem = 0;

140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
/*
 * Kernel parameter "nptcg=" overrides max number of concurrent global TLB
 * purges which is reported from either PAL or SAL PALO.
 *
 * We don't have sanity checking for nptcg value. It's the user's responsibility
 * for valid nptcg value on the platform. Otherwise, kernel may hang in some
 * cases.
 */
static int __init
set_nptcg(char *str)
{
	int value = 0;

	get_option(&str, &value);
	setup_ptcg_sem(value, NPTCG_FROM_KERNEL_PARAMETER);

	return 1;
}

__setup("nptcg=", set_nptcg);

161
162
163
164
165
166
167
/*
 * Maximum number of simultaneous ptc.g purges in the system can
 * be defined by PAL_VM_SUMMARY (in which case we should take
 * the smallest value for any cpu in the system) or by the PAL
 * override table (in which case we should ignore the value from
 * PAL_VM_SUMMARY).
 *
168
169
170
171
 * Kernel parameter "nptcg=" overrides maximum number of simultanesous ptc.g
 * purges defined in either PAL_VM_SUMMARY or PAL override table. In this case,
 * we should ignore the value from either PAL_VM_SUMMARY or PAL override table.
 *
172
173
174
175
 * Complicating the logic here is the fact that num_possible_cpus()
 * isn't fully setup until we start bringing cpus online.
 */
void
176
setup_ptcg_sem(int max_purges, int nptcg_from)
177
{
178
179
	static int kp_override;
	static int palo_override;
180
181
182
	static int firstcpu = 1;

	if (toolatetochangeptcgsem) {
183
184
185
186
		if (nptcg_from == NPTCG_FROM_PAL && max_purges == 0)
			BUG_ON(1 < nptcg);
		else
			BUG_ON(max_purges < nptcg);
187
188
189
		return;
	}

190
191
192
193
194
195
196
197
198
199
200
201
	if (nptcg_from == NPTCG_FROM_KERNEL_PARAMETER) {
		kp_override = 1;
		nptcg = max_purges;
		goto resetsema;
	}
	if (kp_override) {
		need_ptcg_sem = num_possible_cpus() > nptcg;
		return;
	}

	if (nptcg_from == NPTCG_FROM_PALO) {
		palo_override = 1;
202
203
204
205
206
207
208
209
210
211
212

		/* In PALO max_purges == 0 really means it! */
		if (max_purges == 0)
			panic("Whoa! Platform does not support global TLB purges.\n");
		nptcg = max_purges;
		if (nptcg == PALO_MAX_TLB_PURGES) {
			need_ptcg_sem = 0;
			return;
		}
		goto resetsema;
	}
213
	if (palo_override) {
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
		if (nptcg != PALO_MAX_TLB_PURGES)
			need_ptcg_sem = (num_possible_cpus() > nptcg);
		return;
	}

	/* In PAL_VM_SUMMARY max_purges == 0 actually means 1 */
	if (max_purges == 0) max_purges = 1;

	if (firstcpu) {
		nptcg = max_purges;
		firstcpu = 0;
	}
	if (max_purges < nptcg)
		nptcg = max_purges;
	if (nptcg == PAL_MAX_PURGES) {
		need_ptcg_sem = 0;
		return;
	} else
		need_ptcg_sem = (num_possible_cpus() > nptcg);

resetsema:
	spinaphore_init(&ptcg_sem, max_purges);
}

Linus Torvalds's avatar
Linus Torvalds committed
238
void
239
240
ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start,
		       unsigned long end, unsigned long nbits)
Linus Torvalds's avatar
Linus Torvalds committed
241
{
242
243
	struct mm_struct *active_mm = current->active_mm;

244
245
	toolatetochangeptcgsem = 1;

246
247
248
249
250
251
252
253
	if (mm != active_mm) {
		/* Restore region IDs for mm */
		if (mm && active_mm) {
			activate_context(mm);
		} else {
			flush_tlb_all();
			return;
		}
254
255
	}

256
257
258
259
260
261
262
263
264
265
266
267
268
269
	if (need_ptcg_sem)
		down_spin(&ptcg_sem);

	do {
		/*
		 * Flush ALAT entries also.
		 */
		ia64_ptcga(start, (nbits << 2));
		ia64_srlz_i();
		start += (1UL << nbits);
	} while (start < end);

	if (need_ptcg_sem)
		up_spin(&ptcg_sem);
270
271
272
273

        if (mm != active_mm) {
                activate_context(active_mm);
        }
Linus Torvalds's avatar
Linus Torvalds committed
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
}

void
local_flush_tlb_all (void)
{
	unsigned long i, j, flags, count0, count1, stride0, stride1, addr;

	addr    = local_cpu_data->ptce_base;
	count0  = local_cpu_data->ptce_count[0];
	count1  = local_cpu_data->ptce_count[1];
	stride0 = local_cpu_data->ptce_stride[0];
	stride1 = local_cpu_data->ptce_stride[1];

	local_irq_save(flags);
	for (i = 0; i < count0; ++i) {
		for (j = 0; j < count1; ++j) {
			ia64_ptce(addr);
			addr += stride1;
		}
		addr += stride0;
	}
	local_irq_restore(flags);
	ia64_srlz_i();			/* srlz.i implies srlz.d */
}

void
300
301
flush_tlb_range (struct vm_area_struct *vma, unsigned long start,
		 unsigned long end)
Linus Torvalds's avatar
Linus Torvalds committed
302
303
304
305
306
{
	struct mm_struct *mm = vma->vm_mm;
	unsigned long size = end - start;
	unsigned long nbits;

307
#ifndef CONFIG_SMP
Linus Torvalds's avatar
Linus Torvalds committed
308
309
310
311
	if (mm != current->active_mm) {
		mm->context = 0;
		return;
	}
312
#endif
Linus Torvalds's avatar
Linus Torvalds committed
313
314

	nbits = ia64_fls(size + 0xfff);
315
316
	while (unlikely (((1UL << nbits) & purge.mask) == 0) &&
			(nbits < purge.max_bits))
Linus Torvalds's avatar
Linus Torvalds committed
317
318
319
320
321
		++nbits;
	if (nbits > purge.max_bits)
		nbits = purge.max_bits;
	start &= ~((1UL << nbits) - 1);

322
	preempt_disable();
323
#ifdef CONFIG_SMP
324
	if (mm != current->active_mm || cpumask_weight(mm_cpumask(mm)) != 1) {
325
326
327
328
329
		platform_global_tlb_purge(mm, start, end, nbits);
		preempt_enable();
		return;
	}
#endif
Linus Torvalds's avatar
Linus Torvalds committed
330
331
332
333
	do {
		ia64_ptcl(start, (nbits<<2));
		start += (1UL << nbits);
	} while (start < end);
334
	preempt_enable();
Linus Torvalds's avatar
Linus Torvalds committed
335
336
337
338
339
340
341
	ia64_srlz_i();			/* srlz.i implies srlz.d */
}
EXPORT_SYMBOL(flush_tlb_range);

void __devinit
ia64_tlb_init (void)
{
342
	ia64_ptce_info_t uninitialized_var(ptce_info); /* GCC be quiet */
343
	u64 tr_pgbits;
Linus Torvalds's avatar
Linus Torvalds committed
344
	long status;
345
346
347
	pal_vm_info_1_u_t vm_info_1;
	pal_vm_info_2_u_t vm_info_2;
	int cpu = smp_processor_id();
Linus Torvalds's avatar
Linus Torvalds committed
348
349

	if ((status = ia64_pal_vm_page_size(&tr_pgbits, &purge.mask)) != 0) {
350
		printk(KERN_ERR "PAL_VM_PAGE_SIZE failed with status=%ld; "
Linus Torvalds's avatar
Linus Torvalds committed
351
352
353
354
355
356
357
358
359
360
361
362
		       "defaulting to architected purge page-sizes.\n", status);
		purge.mask = 0x115557000UL;
	}
	purge.max_bits = ia64_fls(purge.mask);

	ia64_get_ptce(&ptce_info);
	local_cpu_data->ptce_base = ptce_info.base;
	local_cpu_data->ptce_count[0] = ptce_info.count[0];
	local_cpu_data->ptce_count[1] = ptce_info.count[1];
	local_cpu_data->ptce_stride[0] = ptce_info.stride[0];
	local_cpu_data->ptce_stride[1] = ptce_info.stride[1];

363
	local_flush_tlb_all();	/* nuke left overs from bootstrapping... */
364
365
366
367
368
369
370
371
372
373
374
375
376
	status = ia64_pal_vm_summary(&vm_info_1, &vm_info_2);

	if (status) {
		printk(KERN_ERR "ia64_pal_vm_summary=%ld\n", status);
		per_cpu(ia64_tr_num, cpu) = 8;
		return;
	}
	per_cpu(ia64_tr_num, cpu) = vm_info_1.pal_vm_info_1_s.max_itr_entry+1;
	if (per_cpu(ia64_tr_num, cpu) >
				(vm_info_1.pal_vm_info_1_s.max_dtr_entry+1))
		per_cpu(ia64_tr_num, cpu) =
				vm_info_1.pal_vm_info_1_s.max_dtr_entry+1;
	if (per_cpu(ia64_tr_num, cpu) > IA64_TR_ALLOC_MAX) {
377
		static int justonce = 1;
378
		per_cpu(ia64_tr_num, cpu) = IA64_TR_ALLOC_MAX;
379
380
381
382
383
		if (justonce) {
			justonce = 0;
			printk(KERN_DEBUG "TR register number exceeds "
			       "IA64_TR_ALLOC_MAX!\n");
		}
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
	}
}

/*
 * is_tr_overlap
 *
 * Check overlap with inserted TRs.
 */
static int is_tr_overlap(struct ia64_tr_entry *p, u64 va, u64 log_size)
{
	u64 tr_log_size;
	u64 tr_end;
	u64 va_rr = ia64_get_rr(va);
	u64 va_rid = RR_TO_RID(va_rr);
	u64 va_end = va + (1<<log_size) - 1;

	if (va_rid != RR_TO_RID(p->rr))
		return 0;
	tr_log_size = (p->itir & 0xff) >> 2;
	tr_end = p->ifa + (1<<tr_log_size) - 1;

	if (va > tr_end || p->ifa > va_end)
		return 0;
	return 1;

}

/*
 * ia64_insert_tr in virtual mode. Allocate a TR slot
 *
 * target_mask : 0x1 : itr, 0x2 : dtr, 0x3 : idtr
 *
 * va 	: virtual address.
 * pte 	: pte entries inserted.
 * log_size: range to be covered.
 *
 * Return value:  <0 :  error No.
 *
 *		  >=0 : slot number allocated for TR.
 * Must be called with preemption disabled.
 */
int ia64_itr_entry(u64 target_mask, u64 va, u64 pte, u64 log_size)
{
	int i, r;
	unsigned long psr;
	struct ia64_tr_entry *p;
	int cpu = smp_processor_id();

	r = -EINVAL;
	/*Check overlap with existing TR entries*/
	if (target_mask & 0x1) {
		p = &__per_cpu_idtrs[cpu][0][0];
		for (i = IA64_TR_ALLOC_BASE; i <= per_cpu(ia64_tr_used, cpu);
								i++, p++) {
			if (p->pte & 0x1)
				if (is_tr_overlap(p, va, log_size)) {
					printk(KERN_DEBUG "Overlapped Entry"
						"Inserted for TR Reigster!!\n");
					goto out;
			}
		}
	}
	if (target_mask & 0x2) {
		p = &__per_cpu_idtrs[cpu][1][0];
		for (i = IA64_TR_ALLOC_BASE; i <= per_cpu(ia64_tr_used, cpu);
								i++, p++) {
			if (p->pte & 0x1)
				if (is_tr_overlap(p, va, log_size)) {
					printk(KERN_DEBUG "Overlapped Entry"
						"Inserted for TR Reigster!!\n");
					goto out;
				}
		}
	}

	for (i = IA64_TR_ALLOC_BASE; i < per_cpu(ia64_tr_num, cpu); i++) {
		switch (target_mask & 0x3) {
		case 1:
			if (!(__per_cpu_idtrs[cpu][0][i].pte & 0x1))
				goto found;
			continue;
		case 2:
			if (!(__per_cpu_idtrs[cpu][1][i].pte & 0x1))
				goto found;
			continue;
		case 3:
			if (!(__per_cpu_idtrs[cpu][0][i].pte & 0x1) &&
				!(__per_cpu_idtrs[cpu][1][i].pte & 0x1))
				goto found;
			continue;
		default:
			r = -EINVAL;
			goto out;
		}
	}
found:
	if (i >= per_cpu(ia64_tr_num, cpu))
		return -EBUSY;

	/*Record tr info for mca hander use!*/
	if (i > per_cpu(ia64_tr_used, cpu))
		per_cpu(ia64_tr_used, cpu) = i;

	psr = ia64_clear_ic();
	if (target_mask & 0x1) {
		ia64_itr(0x1, i, va, pte, log_size);
		ia64_srlz_i();
		p = &__per_cpu_idtrs[cpu][0][i];
		p->ifa = va;
		p->pte = pte;
		p->itir = log_size << 2;
		p->rr = ia64_get_rr(va);
	}
	if (target_mask & 0x2) {
		ia64_itr(0x2, i, va, pte, log_size);
		ia64_srlz_i();
		p = &__per_cpu_idtrs[cpu][1][i];
		p->ifa = va;
		p->pte = pte;
		p->itir = log_size << 2;
		p->rr = ia64_get_rr(va);
	}
	ia64_set_psr(psr);
	r = i;
out:
	return r;
}
EXPORT_SYMBOL_GPL(ia64_itr_entry);

/*
 * ia64_purge_tr
 *
 * target_mask: 0x1: purge itr, 0x2 : purge dtr, 0x3 purge idtr.
 * slot: slot number to be freed.
 *
 * Must be called with preemption disabled.
 */
void ia64_ptr_entry(u64 target_mask, int slot)
{
	int cpu = smp_processor_id();
	int i;
	struct ia64_tr_entry *p;

	if (slot < IA64_TR_ALLOC_BASE || slot >= per_cpu(ia64_tr_num, cpu))
		return;

	if (target_mask & 0x1) {
		p = &__per_cpu_idtrs[cpu][0][slot];
		if ((p->pte&0x1) && is_tr_overlap(p, p->ifa, p->itir>>2)) {
			p->pte = 0;
			ia64_ptr(0x1, p->ifa, p->itir>>2);
			ia64_srlz_i();
		}
	}

	if (target_mask & 0x2) {
		p = &__per_cpu_idtrs[cpu][1][slot];
		if ((p->pte & 0x1) && is_tr_overlap(p, p->ifa, p->itir>>2)) {
			p->pte = 0;
			ia64_ptr(0x2, p->ifa, p->itir>>2);
			ia64_srlz_i();
		}
	}

	for (i = per_cpu(ia64_tr_used, cpu); i >= IA64_TR_ALLOC_BASE; i--) {
		if ((__per_cpu_idtrs[cpu][0][i].pte & 0x1) ||
				(__per_cpu_idtrs[cpu][1][i].pte & 0x1))
			break;
	}
	per_cpu(ia64_tr_used, cpu) = i;
Linus Torvalds's avatar
Linus Torvalds committed
554
}
555
EXPORT_SYMBOL_GPL(ia64_ptr_entry);