aboutsummaryrefslogtreecommitdiffstats
path: root/arch/tile/mm/homecache.c
blob: 5f7868dcd6d482abc42e00b6349f5234d0c63c3b (plain) (blame)
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
/*
 * Copyright 2010 Tilera Corporation. All Rights Reserved.
 *
 *   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, version 2.
 *
 *   This program is distributed in the hope that it will be useful, but
 *   WITHOUT ANY WARRANTY; without even the implied warranty of
 *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
 *   NON INFRINGEMENT.  See the GNU General Public License for
 *   more details.
 *
 * This code maintains the "home" for each page in the system.
 */

#include <linux/kernel.h>
#include <linux/mm.h>
#include <linux/spinlock.h>
#include <linux/list.h>
#include <linux/bootmem.h>
#include <linux/rmap.h>
#include <linux/pagemap.h>
#include <linux/mutex.h>
#include <linux/interrupt.h>
#include <linux/sysctl.h>
#include <linux/pagevec.h>
#include <linux/ptrace.h>
#include <linux/timex.h>
#include <linux/cache.h>
#include <linux/smp.h>
#include <linux/module.h>
#include <linux/hugetlb.h>

#include <asm/page.h>
#include <asm/sections.h>
#include <asm/tlbflush.h>
#include <asm/pgalloc.h>
#include <asm/homecache.h>

#include <arch/sim.h>

#include "migrate.h"


#if CHIP_HAS_COHERENT_LOCAL_CACHE()

/*
 * The noallocl2 option suppresses all use of the L2 cache to cache
 * locally from a remote home.  There's no point in using it if we
 * don't have coherent local caching, though.
 */
static int __write_once noallocl2;
static int __init set_noallocl2(char *str)
{
	noallocl2 = 1;
	return 0;
}
early_param("noallocl2", set_noallocl2);

#else

#define noallocl2 0

#endif


/*
 * Update the irq_stat for cpus that we are going to interrupt
 * with TLB or cache flushes.  Also handle removing dataplane cpus
 * from the TLB flush set, and setting dataplane_tlb_state instead.
 */
static void hv_flush_update(const struct cpumask *cache_cpumask,
			    struct cpumask *tlb_cpumask,
			    unsigned long tlb_va, unsigned long tlb_length,
			    HV_Remote_ASID *asids, int asidcount)
{
	struct cpumask mask;
	int i, cpu;

	cpumask_clear(&mask);
	if (cache_cpumask)
		cpumask_or(&mask, &mask, cache_cpumask);
	if (tlb_cpumask && tlb_length) {
		cpumask_or(&mask, &mask, tlb_cpumask);
	}

	for (i = 0; i < asidcount; ++i)
		cpumask_set_cpu(asids[i].y * smp_width + asids[i].x, &mask);

	/*
	 * Don't bother to update atomically; losing a count
	 * here is not that critical.
	 */
	for_each_cpu(cpu, &mask)
		++per_cpu(irq_stat, cpu).irq_hv_flush_count;
}

/*
 * This wrapper function around hv_flush_remote() does several things:
 *
 *  - Provides a return value error-checking panic path, since
 *    there's never any good reason for hv_flush_remote() to fail.
 *  - Accepts a 32-bit PFN rather than a 64-bit PA, which generally
 *    is the type that Linux wants to pass around anyway.
 *  - Canonicalizes that lengths of zero make cpumasks NULL.
 *  - Handles deferring TLB flushes for dataplane tiles.
 *  - Tracks remote interrupts in the per-cpu irq_cpustat_t.
 *
 * Note that we have to wait until the cache flush completes before
 * updating the per-cpu last_cache_flush word, since otherwise another
 * concurrent flush can race, conclude the flush has already
 * completed, and start to use the page while it's still dirty
 * remotely (running concurrently with the actual evict, presumably).
 */
void flush_remote(unsigned long cache_pfn, unsigned long cache_control,
		  const struct cpumask *cache_cpumask_orig,
		  HV_VirtAddr tlb_va, unsigned long tlb_length,
		  unsigned long tlb_pgsize,
		  const struct cpumask *tlb_cpumask_orig,
		  HV_Remote_ASID *asids, int asidcount)
{
	int rc;
	struct cpumask cache_cpumask_copy, tlb_cpumask_copy;
	struct cpumask *cache_cpumask, *tlb_cpumask;
	HV_PhysAddr cache_pa;
	char cache_buf[NR_CPUS*5], tlb_buf[NR_CPUS*5];

	mb();   /* provided just to simplify "magic hypervisor" mode */

	/*
	 * Canonicalize and copy the cpumasks.
	 */
	if (cache_cpumask_orig && cache_control) {
		cpumask_copy(&cache_cpumask_copy, cache_cpumask_orig);
		cache_cpumask = &cache_cpumask_copy;
	} else {
		cpumask_clear(&cache_cpumask_copy);
		cache_cpumask = NULL;
	}
	if (cache_cpumask == NULL)
		cache_control = 0;
	if (tlb_cpumask_orig && tlb_length) {
		cpumask_copy(&tlb_cpumask_copy, tlb_cpumask_orig);
		tlb_cpumask = &tlb_cpumask_copy;
	} else {
		cpumask_clear(&tlb_cpumask_copy);
		tlb_cpumask = NULL;
	}

	hv_flush_update(cache_cpumask, tlb_cpumask, tlb_va, tlb_length,
			asids, asidcount);
	cache_pa = (HV_PhysAddr)cache_pfn << PAGE_SHIFT;
	rc = hv_flush_remote(cache_pa, cache_control,
			     cpumask_bits(cache_cpumask),
			     tlb_va, tlb_length, tlb_pgsize,
			     cpumask_bits(tlb_cpumask),
			     asids, asidcount);
	if (rc == 0)
		return;
	cpumask_scnprintf(cache_buf, sizeof(cache_buf), &cache_cpumask_copy);
	cpumask_scnprintf(tlb_buf, sizeof(tlb_buf), &tlb_cpumask_copy);

	pr_err("hv_flush_remote(%#llx, %#lx, %p [%s],"
	       " %#lx, %#lx, %#lx, %p [%s], %p, %d) = %d\n",
	       cache_pa, cache_control, cache_cpumask, cache_buf,
	       (unsigned long)tlb_va, tlb_length, tlb_pgsize,
	       tlb_cpumask, tlb_buf,
	       asids, asidcount, rc);
	panic("Unsafe to continue.");
}

static void homecache_finv_page_va(void* va, int home)
{
	if (home == smp_processor_id()) {
		finv_buffer_local(va, PAGE_SIZE);
	} else if (home == PAGE_HOME_HASH) {
		finv_buffer_remote(va, PAGE_SIZE, 1);
	} else {
		BUG_ON(home < 0 || home >= NR_CPUS);
		finv_buffer_remote(va, PAGE_SIZE, 0);
	}
}

void homecache_finv_map_page(struct page *page, int home)
{
	unsigned long flags;
	unsigned long va;
	pte_t *ptep;
	pte_t pte;

	if (home == PAGE_HOME_UNCACHED)
		return;
	local_irq_save(flags);
#ifdef CONFIG_HIGHMEM
	va = __fix_to_virt(FIX_KMAP_BEGIN + kmap_atomic_idx_push() +
			   (KM_TYPE_NR * smp_processor_id()));
#else
	va = __fix_to_virt(FIX_HOMECACHE_BEGIN + smp_processor_id());
#endif
	ptep = virt_to_pte(NULL, (unsigned long)va);
	pte = pfn_pte(page_to_pfn(page), PAGE_KERNEL);
	__set_pte(ptep, pte_set_home(pte, home));
	homecache_finv_page_va((void *)va, home);
	__pte_clear(ptep);
	hv_flush_page(va, PAGE_SIZE);
#ifdef CONFIG_HIGHMEM
	kmap_atomic_idx_pop();
#endif
	local_irq_restore(flags);
}

static void homecache_finv_page_home(struct page *page, int home)
{
	if (!PageHighMem(page) && home == page_home(page))
		homecache_finv_page_va(page_address(page), home);
	else
		homecache_finv_map_page(page, home);
}

static inline bool incoherent_home(int home)
{
	return home == PAGE_HOME_IMMUTABLE || home == PAGE_HOME_INCOHERENT;
}

static void homecache_finv_page_internal(struct page *page, int force_map)
{
	int home = page_home(page);
	if (home == PAGE_HOME_UNCACHED)
		return;
	if (incoherent_home(home)) {
		int cpu;
		for_each_cpu(cpu, &cpu_cacheable_map)
			homecache_finv_map_page(page, cpu);
	} else if (force_map) {
		/* Force if, e.g., the normal mapping is migrating. */
		homecache_finv_map_page(page, home);
	} else {
		homecache_finv_page_home(page, home);
	}
	sim_validate_lines_evicted(PFN_PHYS(page_to_pfn(page)), PAGE_SIZE);
}

void homecache_finv_page(struct page *page)
{
	homecache_finv_page_internal(page, 0);
}

void homecache_evict(const struct cpumask *mask)
{
	flush_remote(0, HV_FLUSH_EVICT_L2, mask, 0, 0, 0, NULL, NULL, 0);
}

/* Report the home corresponding to a given PTE. */
static int pte_to_home(pte_t pte)
{
	if (hv_pte_get_nc(pte))
		return PAGE_HOME_IMMUTABLE;
	switch (hv_pte_get_mode(pte)) {
	case HV_PTE_MODE_CACHE_TILE_L3:
		return get_remote_cache_cpu(pte);
	case HV_PTE_MODE_CACHE_NO_L3:
		return PAGE_HOME_INCOHERENT;
	case HV_PTE_MODE_UNCACHED:
		return PAGE_HOME_UNCACHED;
#if CHIP_HAS_CBOX_HOME_MAP()
	case HV_PTE_MODE_CACHE_HASH_L3:
		return PAGE_HOME_HASH;
#endif
	}
	panic("Bad PTE %#llx\n", pte.val);
}

/* Update the home of a PTE if necessary (can also be used for a pgprot_t). */
pte_t pte_set_home(pte_t pte, int home)
{
	/* Check for non-linear file mapping "PTEs" and pass them through. */
	if (pte_file(pte))
		return pte;

#if CHIP_HAS_MMIO()
	/* Check for MMIO mappings and pass them through. */
	if (hv_pte_get_mode(pte) == HV_PTE_MODE_MMIO)
		return pte;
#endif


	/*
	 * Only immutable pages get NC mappings.  If we have a
	 * non-coherent PTE, but the underlying page is not
	 * immutable, it's likely the result of a forced
	 * caching setting running up against ptrace setting
	 * the page to be writable underneath.  In this case,
	 * just keep the PTE coherent.
	 */
	if (hv_pte_get_nc(pte) && home != PAGE_HOME_IMMUTABLE) {
		pte = hv_pte_clear_nc(pte);
		pr_err("non-immutable page incoherently referenced: %#llx\n",
		       pte.val);
	}

	switch (home) {

	case PAGE_HOME_UNCACHED:
		pte = hv_pte_set_mode(pte, HV_PTE_MODE_UNCACHED);
		break;

	case PAGE_HOME_INCOHERENT:
		pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_NO_L3);
		break;

	case PAGE_HOME_IMMUTABLE:
		/*
		 * We could home this page anywhere, since it's immutable,
		 * but by default just home it to follow "hash_default".
		 */
		BUG_ON(hv_pte_get_writable(pte));
		if (pte_get_forcecache(pte)) {
			/* Upgrade "force any cpu" to "No L3" for immutable. */
			if (hv_pte_get_mode(pte) == HV_PTE_MODE_CACHE_TILE_L3
			    && pte_get_anyhome(pte)) {
				pte = hv_pte_set_mode(pte,
						      HV_PTE_MODE_CACHE_NO_L3);
			}
		} else
#if CHIP_HAS_CBOX_HOME_MAP()
		if (hash_default)
			pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_HASH_L3);
		else
#endif
			pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_NO_L3);
		pte = hv_pte_set_nc(pte);
		break;

#if CHIP_HAS_CBOX_HOME_MAP()
	case PAGE_HOME_HASH:
		pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_HASH_L3);
		break;
#endif

	default:
		BUG_ON(home < 0 || home >= NR_CPUS ||
		       !cpu_is_valid_lotar(home));
		pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_TILE_L3);
		pte = set_remote_cache_cpu(pte, home);
		break;
	}

#if CHIP_HAS_NC_AND_NOALLOC_BITS()
	if (noallocl2)
		pte = hv_pte_set_no_alloc_l2(pte);

	/* Simplify "no local and no l3" to "uncached" */
	if (hv_pte_get_no_alloc_l2(pte) && hv_pte_get_no_alloc_l1(pte) &&
	    hv_pte_get_mode(pte) == HV_PTE_MODE_CACHE_NO_L3) {
		pte = hv_pte_set_mode(pte, HV_PTE_MODE_UNCACHED);
	}
#endif

	/* Checking this case here gives a better panic than from the hv. */
	BUG_ON(hv_pte_get_mode(pte) == 0);

	return pte;
}
EXPORT_SYMBOL(pte_set_home);

/*
 * The routines in this section are the "static" versions of the normal
 * dynamic homecaching routines; they just set the home cache
 * of a kernel page once, and require a full-chip cache/TLB flush,
 * so they're not suitable for anything but infrequent use.
 */

#if CHIP_HAS_CBOX_HOME_MAP()
static inline int initial_page_home(void) { return PAGE_HOME_HASH; }
#else
static inline int initial_page_home(void) { return 0; }
#endif

int page_home(struct page *page)
{
	if (PageHighMem(page)) {
		return initial_page_home();
	} else {
		unsigned long kva = (unsigned long)page_address(page);
		return pte_to_home(*virt_to_pte(NULL, kva));
	}
}
EXPORT_SYMBOL(page_home);

void homecache_change_page_home(struct page *page, int order, int home)
{
	int i, pages = (1 << order);
	unsigned long kva;

	BUG_ON(PageHighMem(page));
	BUG_ON(page_count(page) > 1);
	BUG_ON(page_mapcount(page) != 0);
	kva = (unsigned long) page_address(page);
	flush_remote(0, HV_FLUSH_EVICT_L2, &cpu_cacheable_map,
		     kva, pages * PAGE_SIZE, PAGE_SIZE, cpu_online_mask,
		     NULL, 0);

	for (i = 0; i < pages; ++i, kva += PAGE_SIZE) {
		pte_t *ptep = virt_to_pte(NULL, kva);
		pte_t pteval = *ptep;
		BUG_ON(!pte_present(pteval) || pte_huge(pteval));
		__set_pte(ptep, pte_set_home(pteval, home));
	}
}

struct page *homecache_alloc_pages(gfp_t gfp_mask,
				   unsigned int order, int home)
{
	struct page *page;
	BUG_ON(gfp_mask & __GFP_HIGHMEM);   /* must be lowmem */
	page = alloc_pages(gfp_mask, order);
	if (page)
		homecache_change_page_home(page, order, home);
	return page;
}
EXPORT_SYMBOL(homecache_alloc_pages);

struct page *homecache_alloc_pages_node(int nid, gfp_t gfp_mask,
					unsigned int order, int home)
{
	struct page *page;
	BUG_ON(gfp_mask & __GFP_HIGHMEM);   /* must be lowmem */
	page = alloc_pages_node(nid, gfp_mask, order);
	if (page)
		homecache_change_page_home(page, order, home);
	return page;
}

void __homecache_free_pages(struct page *page, unsigned int order)
{
	if (put_page_testzero(page)) {
		homecache_change_page_home(page, order, initial_page_home());
		if (order == 0) {
			free_hot_cold_page(page, 0);
		} else {
			init_page_count(page);
			__free_pages(page, order);
		}
	}
}
EXPORT_SYMBOL(__homecache_free_pages);

void homecache_free_pages(unsigned long addr, unsigned int order)
{
	if (addr != 0) {
		VM_BUG_ON(!virt_addr_valid((void *)addr));
		__homecache_free_pages(virt_to_page((void *)addr), order);
	}
}
EXPORT_SYMBOL(homecache_free_pages);