summaryrefslogtreecommitdiff
path: root/arch/powerpc/mm/mmu_context_nohash.c
blob: 22d71a58167f55bb69e7c98819925c61ff565b1d (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
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
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
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
/*
 * This file contains the routines for handling the MMU on those
 * PowerPC implementations where the MMU is not using the hash
 * table, such as 8xx, 4xx, BookE's etc...
 *
 * Copyright 2008 Ben Herrenschmidt <benh@kernel.crashing.org>
 *                IBM Corp.
 *
 *  Derived from previous arch/powerpc/mm/mmu_context.c
 *  and arch/powerpc/include/asm/mmu_context.h
 *
 *  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.
 *
 * TODO:
 *
 *   - The global context lock will not scale very well
 *   - The maps should be dynamically allocated to allow for processors
 *     that support more PID bits at runtime
 *   - Implement flush_tlb_mm() by making the context stale and picking
 *     a new one
 *   - More aggressively clear stale map bits and maybe find some way to
 *     also clear mm->cpu_vm_mask bits when processes are migrated
 */

//#define DEBUG_MAP_CONSISTENCY
//#define DEBUG_CLAMP_LAST_CONTEXT   31
//#define DEBUG_HARDER

/* We don't use DEBUG because it tends to be compiled in always nowadays
 * and this would generate way too much output
 */
#ifdef DEBUG_HARDER
#define pr_hard(args...)	printk(KERN_DEBUG args)
#define pr_hardcont(args...)	printk(KERN_CONT args)
#else
#define pr_hard(args...)	do { } while(0)
#define pr_hardcont(args...)	do { } while(0)
#endif

#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/spinlock.h>
#include <linux/memblock.h>
#include <linux/notifier.h>
#include <linux/cpu.h>
#include <linux/slab.h>

#include <asm/mmu_context.h>
#include <asm/tlbflush.h>

#include "mmu_decl.h"

/*
 * The MPC8xx has only 16 contexts. We rotate through them on each task switch.
 * A better way would be to keep track of tasks that own contexts, and implement
 * an LRU usage. That way very active tasks don't always have to pay the TLB
 * reload overhead. The kernel pages are mapped shared, so the kernel can run on
 * behalf of any task that makes a kernel entry. Shared does not mean they are
 * not protected, just that the ASID comparison is not performed. -- Dan
 *
 * The IBM4xx has 256 contexts, so we can just rotate through these as a way of
 * "switching" contexts. If the TID of the TLB is zero, the PID/TID comparison
 * is disabled, so we can use a TID of zero to represent all kernel pages as
 * shared among all contexts. -- Dan
 *
 * The IBM 47x core supports 16-bit PIDs, thus 65535 contexts. We should
 * normally never have to steal though the facility is present if needed.
 * -- BenH
 */
#define FIRST_CONTEXT 1
#ifdef DEBUG_CLAMP_LAST_CONTEXT
#define LAST_CONTEXT DEBUG_CLAMP_LAST_CONTEXT
#elif defined(CONFIG_PPC_8xx)
#define LAST_CONTEXT 16
#elif defined(CONFIG_PPC_47x)
#define LAST_CONTEXT 65535
#else
#define LAST_CONTEXT 255
#endif

static unsigned int next_context, nr_free_contexts;
static unsigned long *context_map;
#ifdef CONFIG_SMP
static unsigned long *stale_map[NR_CPUS];
#endif
static struct mm_struct **context_mm;
static DEFINE_RAW_SPINLOCK(context_lock);

#define CTX_MAP_SIZE	\
	(sizeof(unsigned long) * (LAST_CONTEXT / BITS_PER_LONG + 1))


/* Steal a context from a task that has one at the moment.
 *
 * This is used when we are running out of available PID numbers
 * on the processors.
 *
 * This isn't an LRU system, it just frees up each context in
 * turn (sort-of pseudo-random replacement :).  This would be the
 * place to implement an LRU scheme if anyone was motivated to do it.
 *  -- paulus
 *
 * For context stealing, we use a slightly different approach for
 * SMP and UP. Basically, the UP one is simpler and doesn't use
 * the stale map as we can just flush the local CPU
 *  -- benh
 */
#ifdef CONFIG_SMP
static unsigned int steal_context_smp(unsigned int id)
{
	struct mm_struct *mm;
	unsigned int cpu, max, i;

	max = LAST_CONTEXT - FIRST_CONTEXT;

	/* Attempt to free next_context first and then loop until we manage */
	while (max--) {
		/* Pick up the victim mm */
		mm = context_mm[id];

		/* We have a candidate victim, check if it's active, on SMP
		 * we cannot steal active contexts
		 */
		if (mm->context.active) {
			id++;
			if (id > LAST_CONTEXT)
				id = FIRST_CONTEXT;
			continue;
		}
		pr_hardcont(" | steal %d from 0x%p", id, mm);

		/* Mark this mm has having no context anymore */
		mm->context.id = MMU_NO_CONTEXT;

		/* Mark it stale on all CPUs that used this mm. For threaded
		 * implementations, we set it on all threads on each core
		 * represented in the mask. A future implementation will use
		 * a core map instead but this will do for now.
		 */
		for_each_cpu(cpu, mm_cpumask(mm)) {
			for (i = cpu_first_thread_sibling(cpu);
			     i <= cpu_last_thread_sibling(cpu); i++) {
				if (stale_map[i])
					__set_bit(id, stale_map[i]);
			}
			cpu = i - 1;
		}
		return id;
	}

	/* This will happen if you have more CPUs than available contexts,
	 * all we can do here is wait a bit and try again
	 */
	raw_spin_unlock(&context_lock);
	cpu_relax();
	raw_spin_lock(&context_lock);

	/* This will cause the caller to try again */
	return MMU_NO_CONTEXT;
}
#endif  /* CONFIG_SMP */

static unsigned int steal_all_contexts(void)
{
	struct mm_struct *mm;
#ifdef CONFIG_SMP
	int cpu = smp_processor_id();
#endif
	unsigned int id;

	for (id = FIRST_CONTEXT; id <= LAST_CONTEXT; id++) {
		/* Pick up the victim mm */
		mm = context_mm[id];

		pr_hardcont(" | steal %d from 0x%p", id, mm);

		/* Mark this mm as having no context anymore */
		mm->context.id = MMU_NO_CONTEXT;
		if (id != FIRST_CONTEXT) {
			context_mm[id] = NULL;
			__clear_bit(id, context_map);
#ifdef DEBUG_MAP_CONSISTENCY
			mm->context.active = 0;
#endif
		}
#ifdef CONFIG_SMP
		__clear_bit(id, stale_map[cpu]);
#endif
	}

	/* Flush the TLB for all contexts (not to be used on SMP) */
	_tlbil_all();

	nr_free_contexts = LAST_CONTEXT - FIRST_CONTEXT;

	return FIRST_CONTEXT;
}

/* Note that this will also be called on SMP if all other CPUs are
 * offlined, which means that it may be called for cpu != 0. For
 * this to work, we somewhat assume that CPUs that are onlined
 * come up with a fully clean TLB (or are cleaned when offlined)
 */
static unsigned int steal_context_up(unsigned int id)
{
	struct mm_struct *mm;
#ifdef CONFIG_SMP
	int cpu = smp_processor_id();
#endif

	/* Pick up the victim mm */
	mm = context_mm[id];

	pr_hardcont(" | steal %d from 0x%p", id, mm);

	/* Flush the TLB for that context */
	local_flush_tlb_mm(mm);

	/* Mark this mm has having no context anymore */
	mm->context.id = MMU_NO_CONTEXT;

	/* XXX This clear should ultimately be part of local_flush_tlb_mm */
#ifdef CONFIG_SMP
	__clear_bit(id, stale_map[cpu]);
#endif

	return id;
}

#ifdef DEBUG_MAP_CONSISTENCY
static void context_check_map(void)
{
	unsigned int id, nrf, nact;

	nrf = nact = 0;
	for (id = FIRST_CONTEXT; id <= LAST_CONTEXT; id++) {
		int used = test_bit(id, context_map);
		if (!used)
			nrf++;
		if (used != (context_mm[id] != NULL))
			pr_err("MMU: Context %d is %s and MM is %p !\n",
			       id, used ? "used" : "free", context_mm[id]);
		if (context_mm[id] != NULL)
			nact += context_mm[id]->context.active;
	}
	if (nrf != nr_free_contexts) {
		pr_err("MMU: Free context count out of sync ! (%d vs %d)\n",
		       nr_free_contexts, nrf);
		nr_free_contexts = nrf;
	}
	if (nact > num_online_cpus())
		pr_err("MMU: More active contexts than CPUs ! (%d vs %d)\n",
		       nact, num_online_cpus());
	if (FIRST_CONTEXT > 0 && !test_bit(0, context_map))
		pr_err("MMU: Context 0 has been freed !!!\n");
}
#else
static void context_check_map(void) { }
#endif

void switch_mmu_context(struct mm_struct *prev, struct mm_struct *next,
			struct task_struct *tsk)
{
	unsigned int id;
#ifdef CONFIG_SMP
	unsigned int i, cpu = smp_processor_id();
#endif
	unsigned long *map;

	/* No lockless fast path .. yet */
	raw_spin_lock(&context_lock);

	pr_hard("[%d] activating context for mm @%p, active=%d, id=%d",
		cpu, next, next->context.active, next->context.id);

#ifdef CONFIG_SMP
	/* Mark us active and the previous one not anymore */
	next->context.active++;
	if (prev) {
		pr_hardcont(" (old=0x%p a=%d)", prev, prev->context.active);
		WARN_ON(prev->context.active < 1);
		prev->context.active--;
	}

 again:
#endif /* CONFIG_SMP */

	/* If we already have a valid assigned context, skip all that */
	id = next->context.id;
	if (likely(id != MMU_NO_CONTEXT)) {
#ifdef DEBUG_MAP_CONSISTENCY
		if (context_mm[id] != next)
			pr_err("MMU: mm 0x%p has id %d but context_mm[%d] says 0x%p\n",
			       next, id, id, context_mm[id]);
#endif
		goto ctxt_ok;
	}

	/* We really don't have a context, let's try to acquire one */
	id = next_context;
	if (id > LAST_CONTEXT)
		id = FIRST_CONTEXT;
	map = context_map;

	/* No more free contexts, let's try to steal one */
	if (nr_free_contexts == 0) {
#ifdef CONFIG_SMP
		if (num_online_cpus() > 1) {
			id = steal_context_smp(id);
			if (id == MMU_NO_CONTEXT)
				goto again;
			goto stolen;
		}
#endif /* CONFIG_SMP */
		if (IS_ENABLED(CONFIG_PPC_8xx))
			id = steal_all_contexts();
		else
			id = steal_context_up(id);
		goto stolen;
	}
	nr_free_contexts--;

	/* We know there's at least one free context, try to find it */
	while (__test_and_set_bit(id, map)) {
		id = find_next_zero_bit(map, LAST_CONTEXT+1, id);
		if (id > LAST_CONTEXT)
			id = FIRST_CONTEXT;
	}
 stolen:
	next_context = id + 1;
	context_mm[id] = next;
	next->context.id = id;
	pr_hardcont(" | new id=%d,nrf=%d", id, nr_free_contexts);

	context_check_map();
 ctxt_ok:

	/* If that context got marked stale on this CPU, then flush the
	 * local TLB for it and unmark it before we use it
	 */
#ifdef CONFIG_SMP
	if (test_bit(id, stale_map[cpu])) {
		pr_hardcont(" | stale flush %d [%d..%d]",
			    id, cpu_first_thread_sibling(cpu),
			    cpu_last_thread_sibling(cpu));

		local_flush_tlb_mm(next);

		/* XXX This clear should ultimately be part of local_flush_tlb_mm */
		for (i = cpu_first_thread_sibling(cpu);
		     i <= cpu_last_thread_sibling(cpu); i++) {
			if (stale_map[i])
				__clear_bit(id, stale_map[i]);
		}
	}
#endif

	/* Flick the MMU and release lock */
	pr_hardcont(" -> %d\n", id);
	set_context(id, next->pgd);
	raw_spin_unlock(&context_lock);
}

/*
 * Set up the context for a new address space.
 */
int init_new_context(struct task_struct *t, struct mm_struct *mm)
{
	pr_hard("initing context for mm @%p\n", mm);

	/*
	 * We have MMU_NO_CONTEXT set to be ~0. Hence check
	 * explicitly against context.id == 0. This ensures that we properly
	 * initialize context slice details for newly allocated mm's (which will
	 * have id == 0) and don't alter context slice inherited via fork (which
	 * will have id != 0).
	 */
	if (mm->context.id == 0)
		slice_init_new_context_exec(mm);
	mm->context.id = MMU_NO_CONTEXT;
	mm->context.active = 0;
	pte_frag_set(&mm->context, NULL);
	return 0;
}

/*
 * We're finished using the context for an address space.
 */
void destroy_context(struct mm_struct *mm)
{
	unsigned long flags;
	unsigned int id;

	if (mm->context.id == MMU_NO_CONTEXT)
		return;

	WARN_ON(mm->context.active != 0);

	raw_spin_lock_irqsave(&context_lock, flags);
	id = mm->context.id;
	if (id != MMU_NO_CONTEXT) {
		__clear_bit(id, context_map);
		mm->context.id = MMU_NO_CONTEXT;
#ifdef DEBUG_MAP_CONSISTENCY
		mm->context.active = 0;
#endif
		context_mm[id] = NULL;
		nr_free_contexts++;
	}
	raw_spin_unlock_irqrestore(&context_lock, flags);
}

#ifdef CONFIG_SMP
static int mmu_ctx_cpu_prepare(unsigned int cpu)
{
	/* We don't touch CPU 0 map, it's allocated at aboot and kept
	 * around forever
	 */
	if (cpu == boot_cpuid)
		return 0;

	pr_devel("MMU: Allocating stale context map for CPU %d\n", cpu);
	stale_map[cpu] = kzalloc(CTX_MAP_SIZE, GFP_KERNEL);
	return 0;
}

static int mmu_ctx_cpu_dead(unsigned int cpu)
{
#ifdef CONFIG_HOTPLUG_CPU
	if (cpu == boot_cpuid)
		return 0;

	pr_devel("MMU: Freeing stale context map for CPU %d\n", cpu);
	kfree(stale_map[cpu]);
	stale_map[cpu] = NULL;

	/* We also clear the cpu_vm_mask bits of CPUs going away */
	clear_tasks_mm_cpumask(cpu);
#endif
	return 0;
}

#endif /* CONFIG_SMP */

/*
 * Initialize the context management stuff.
 */
void __init mmu_context_init(void)
{
	/* Mark init_mm as being active on all possible CPUs since
	 * we'll get called with prev == init_mm the first time
	 * we schedule on a given CPU
	 */
	init_mm.context.active = NR_CPUS;

	/*
	 * Allocate the maps used by context management
	 */
	context_map = memblock_alloc(CTX_MAP_SIZE, SMP_CACHE_BYTES);
	context_mm = memblock_alloc(sizeof(void *) * (LAST_CONTEXT + 1),
				    SMP_CACHE_BYTES);
#ifdef CONFIG_SMP
	stale_map[boot_cpuid] = memblock_alloc(CTX_MAP_SIZE, SMP_CACHE_BYTES);

	cpuhp_setup_state_nocalls(CPUHP_POWERPC_MMU_CTX_PREPARE,
				  "powerpc/mmu/ctx:prepare",
				  mmu_ctx_cpu_prepare, mmu_ctx_cpu_dead);
#endif

	printk(KERN_INFO
	       "MMU: Allocated %zu bytes of context maps for %d contexts\n",
	       2 * CTX_MAP_SIZE + (sizeof(void *) * (LAST_CONTEXT + 1)),
	       LAST_CONTEXT - FIRST_CONTEXT + 1);

	/*
	 * Some processors have too few contexts to reserve one for
	 * init_mm, and require using context 0 for a normal task.
	 * Other processors reserve the use of context zero for the kernel.
	 * This code assumes FIRST_CONTEXT < 32.
	 */
	context_map[0] = (1 << FIRST_CONTEXT) - 1;
	next_context = FIRST_CONTEXT;
	nr_free_contexts = LAST_CONTEXT - FIRST_CONTEXT + 1;
}