pageattr.c 23.5 KB
Newer Older
1 2
/*
 * Copyright 2002 Andi Kleen, SuSE Labs.
L
Linus Torvalds 已提交
3
 * Thanks to Ben LaHaise for precious feedback.
4
 */
L
Linus Torvalds 已提交
5
#include <linux/highmem.h>
I
Ingo Molnar 已提交
6
#include <linux/bootmem.h>
L
Linus Torvalds 已提交
7
#include <linux/module.h>
8
#include <linux/sched.h>
L
Linus Torvalds 已提交
9
#include <linux/slab.h>
10
#include <linux/mm.h>
11
#include <linux/interrupt.h>
12 13
#include <linux/seq_file.h>
#include <linux/debugfs.h>
14

15
#include <asm/e820.h>
L
Linus Torvalds 已提交
16 17
#include <asm/processor.h>
#include <asm/tlbflush.h>
D
Dave Jones 已提交
18
#include <asm/sections.h>
19 20
#include <asm/uaccess.h>
#include <asm/pgalloc.h>
T
Thomas Gleixner 已提交
21
#include <asm/proto.h>
22
#include <asm/pat.h>
L
Linus Torvalds 已提交
23

I
Ingo Molnar 已提交
24 25 26
/*
 * The current flushing context - we pass it instead of 5 arguments:
 */
T
Thomas Gleixner 已提交
27 28 29 30
struct cpa_data {
	unsigned long	vaddr;
	pgprot_t	mask_set;
	pgprot_t	mask_clr;
31
	int		numpages;
32
	int		flushtlb;
T
Thomas Gleixner 已提交
33
	unsigned long	pfn;
34
	unsigned	force_split : 1;
T
Thomas Gleixner 已提交
35 36
};

T
Thomas Gleixner 已提交
37 38 39 40 41 42 43 44 45 46 47 48 49 50
#ifdef CONFIG_X86_64

static inline unsigned long highmap_start_pfn(void)
{
	return __pa(_text) >> PAGE_SHIFT;
}

static inline unsigned long highmap_end_pfn(void)
{
	return __pa(round_up((unsigned long)_end, PMD_SIZE)) >> PAGE_SHIFT;
}

#endif

51 52 53 54 55 56
#ifdef CONFIG_DEBUG_PAGEALLOC
# define debug_pagealloc 1
#else
# define debug_pagealloc 0
#endif

57 58
static inline int
within(unsigned long addr, unsigned long start, unsigned long end)
I
Ingo Molnar 已提交
59
{
60 61 62
	return addr >= start && addr < end;
}

T
Thomas Gleixner 已提交
63 64 65
/*
 * Flushing functions
 */
66 67 68 69 70 71 72 73 74

/**
 * clflush_cache_range - flush a cache range with clflush
 * @addr:	virtual start address
 * @size:	number of bytes to flush
 *
 * clflush is an unordered instruction which needs fencing with mfence
 * to avoid ordering issues.
 */
I
Ingo Molnar 已提交
75
void clflush_cache_range(void *vaddr, unsigned int size)
T
Thomas Gleixner 已提交
76
{
I
Ingo Molnar 已提交
77
	void *vend = vaddr + size - 1;
T
Thomas Gleixner 已提交
78

79
	mb();
I
Ingo Molnar 已提交
80 81 82 83 84 85 86 87

	for (; vaddr < vend; vaddr += boot_cpu_data.x86_clflush_size)
		clflush(vaddr);
	/*
	 * Flush any possible final partial cacheline:
	 */
	clflush(vend);

88
	mb();
T
Thomas Gleixner 已提交
89 90
}

91
static void __cpa_flush_all(void *arg)
T
Thomas Gleixner 已提交
92
{
93 94
	unsigned long cache = (unsigned long)arg;

T
Thomas Gleixner 已提交
95 96 97 98 99 100
	/*
	 * Flush all to work around Errata in early athlons regarding
	 * large page flushing.
	 */
	__flush_tlb_all();

101
	if (cache && boot_cpu_data.x86_model >= 4)
T
Thomas Gleixner 已提交
102 103 104
		wbinvd();
}

105
static void cpa_flush_all(unsigned long cache)
T
Thomas Gleixner 已提交
106 107 108
{
	BUG_ON(irqs_disabled());

109
	on_each_cpu(__cpa_flush_all, (void *) cache, 1, 1);
T
Thomas Gleixner 已提交
110 111
}

112 113 114 115 116 117 118 119 120 121
static void __cpa_flush_range(void *arg)
{
	/*
	 * We could optimize that further and do individual per page
	 * tlb invalidates for a low number of pages. Caveat: we must
	 * flush the high aliases on 64bit as well.
	 */
	__flush_tlb_all();
}

122
static void cpa_flush_range(unsigned long start, int numpages, int cache)
123
{
I
Ingo Molnar 已提交
124 125 126
	unsigned int i, level;
	unsigned long addr;

127
	BUG_ON(irqs_disabled());
I
Ingo Molnar 已提交
128
	WARN_ON(PAGE_ALIGN(start) != start);
129

T
Thomas Gleixner 已提交
130
	on_each_cpu(__cpa_flush_range, NULL, 1, 1);
131

132 133 134
	if (!cache)
		return;

T
Thomas Gleixner 已提交
135 136 137 138 139 140
	/*
	 * We only need to flush on one CPU,
	 * clflush is a MESI-coherent instruction that
	 * will cause all other CPUs to flush the same
	 * cachelines:
	 */
I
Ingo Molnar 已提交
141 142 143 144 145 146
	for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
		pte_t *pte = lookup_address(addr, &level);

		/*
		 * Only flush present addresses:
		 */
147
		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
I
Ingo Molnar 已提交
148 149
			clflush_cache_range((void *) addr, PAGE_SIZE);
	}
150 151
}

152 153 154 155 156 157
/*
 * Certain areas of memory on x86 require very specific protection flags,
 * for example the BIOS area or kernel text. Callers don't always get this
 * right (again, ioremap() on BIOS memory is not uncommon) so this function
 * checks and fixes these known static required protection bits.
 */
T
Thomas Gleixner 已提交
158 159
static inline pgprot_t static_protections(pgprot_t prot, unsigned long address,
				   unsigned long pfn)
160 161 162
{
	pgprot_t forbidden = __pgprot(0);

I
Ingo Molnar 已提交
163
	/*
164 165
	 * The BIOS area between 640k and 1Mb needs to be executable for
	 * PCI BIOS based config access (CONFIG_PCI_GOBIOS) support.
I
Ingo Molnar 已提交
166
	 */
T
Thomas Gleixner 已提交
167
	if (within(pfn, BIOS_BEGIN >> PAGE_SHIFT, BIOS_END >> PAGE_SHIFT))
168 169 170 171
		pgprot_val(forbidden) |= _PAGE_NX;

	/*
	 * The kernel text needs to be executable for obvious reasons
T
Thomas Gleixner 已提交
172 173
	 * Does not cover __inittext since that is gone later on. On
	 * 64bit we do not enforce !NX on the low mapping
174 175 176
	 */
	if (within(address, (unsigned long)_text, (unsigned long)_etext))
		pgprot_val(forbidden) |= _PAGE_NX;
177 178

	/*
T
Thomas Gleixner 已提交
179 180
	 * The .rodata section needs to be read-only. Using the pfn
	 * catches all aliases.
181
	 */
T
Thomas Gleixner 已提交
182 183
	if (within(pfn, __pa((unsigned long)__start_rodata) >> PAGE_SHIFT,
		   __pa((unsigned long)__end_rodata) >> PAGE_SHIFT))
184
		pgprot_val(forbidden) |= _PAGE_RW;
185 186

	prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));
I
Ingo Molnar 已提交
187 188 189 190

	return prot;
}

T
Thomas Gleixner 已提交
191 192 193 194 195 196 197 198
/*
 * Lookup the page table entry for a virtual address. Return a pointer
 * to the entry and the level of the mapping.
 *
 * Note: We return pud and pmd either when the entry is marked large
 * or when the present bit is not set. Otherwise we would return a
 * pointer to a nonexisting mapping.
 */
199
pte_t *lookup_address(unsigned long address, unsigned int *level)
200
{
L
Linus Torvalds 已提交
201 202 203
	pgd_t *pgd = pgd_offset_k(address);
	pud_t *pud;
	pmd_t *pmd;
204

T
Thomas Gleixner 已提交
205 206
	*level = PG_LEVEL_NONE;

L
Linus Torvalds 已提交
207 208
	if (pgd_none(*pgd))
		return NULL;
I
Ingo Molnar 已提交
209

L
Linus Torvalds 已提交
210 211 212
	pud = pud_offset(pgd, address);
	if (pud_none(*pud))
		return NULL;
213 214 215 216 217

	*level = PG_LEVEL_1G;
	if (pud_large(*pud) || !pud_present(*pud))
		return (pte_t *)pud;

L
Linus Torvalds 已提交
218 219 220
	pmd = pmd_offset(pud, address);
	if (pmd_none(*pmd))
		return NULL;
T
Thomas Gleixner 已提交
221 222

	*level = PG_LEVEL_2M;
T
Thomas Gleixner 已提交
223
	if (pmd_large(*pmd) || !pmd_present(*pmd))
L
Linus Torvalds 已提交
224 225
		return (pte_t *)pmd;

T
Thomas Gleixner 已提交
226
	*level = PG_LEVEL_4K;
I
Ingo Molnar 已提交
227

228 229 230
	return pte_offset_kernel(pmd, address);
}

I
Ingo Molnar 已提交
231 232 233
/*
 * Set the new pmd in all the pgds we know about:
 */
I
Ingo Molnar 已提交
234
static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
235 236 237
{
	/* change init_mm */
	set_pte_atomic(kpte, pte);
238
#ifdef CONFIG_X86_32
239
	if (!SHARED_KERNEL_PMD) {
240 241
		struct page *page;

242
		list_for_each_entry(page, &pgd_list, lru) {
243 244 245 246 247 248 249 250 251
			pgd_t *pgd;
			pud_t *pud;
			pmd_t *pmd;

			pgd = (pgd_t *)page_address(page) + pgd_index(address);
			pud = pud_offset(pgd, address);
			pmd = pmd_offset(pud, address);
			set_pte_atomic((pte_t *)pmd, pte);
		}
L
Linus Torvalds 已提交
252
	}
253
#endif
L
Linus Torvalds 已提交
254 255
}

I
Ingo Molnar 已提交
256 257 258
static int
try_preserve_large_page(pte_t *kpte, unsigned long address,
			struct cpa_data *cpa)
259
{
T
Thomas Gleixner 已提交
260
	unsigned long nextpage_addr, numpages, pmask, psize, flags, addr, pfn;
261 262
	pte_t new_pte, old_pte, *tmp;
	pgprot_t old_prot, new_prot;
263
	int i, do_split = 1;
264
	unsigned int level;
265

266 267 268
	if (cpa->force_split)
		return 1;

269 270 271 272 273 274 275 276 277 278 279
	spin_lock_irqsave(&pgd_lock, flags);
	/*
	 * Check for races, another CPU might have split this page
	 * up already:
	 */
	tmp = lookup_address(address, &level);
	if (tmp != kpte)
		goto out_unlock;

	switch (level) {
	case PG_LEVEL_2M:
280 281
		psize = PMD_PAGE_SIZE;
		pmask = PMD_PAGE_MASK;
282
		break;
283
#ifdef CONFIG_X86_64
284
	case PG_LEVEL_1G:
285 286
		psize = PUD_PAGE_SIZE;
		pmask = PUD_PAGE_MASK;
287 288
		break;
#endif
289
	default:
I
Ingo Molnar 已提交
290
		do_split = -EINVAL;
291 292 293 294 295 296 297 298 299
		goto out_unlock;
	}

	/*
	 * Calculate the number of pages, which fit into this large
	 * page starting at address:
	 */
	nextpage_addr = (address + psize) & pmask;
	numpages = (nextpage_addr - address) >> PAGE_SHIFT;
300 301
	if (numpages < cpa->numpages)
		cpa->numpages = numpages;
302 303 304 305 306 307 308 309 310

	/*
	 * We are safe now. Check whether the new pgprot is the same:
	 */
	old_pte = *kpte;
	old_prot = new_prot = pte_pgprot(old_pte);

	pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
	pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
T
Thomas Gleixner 已提交
311 312 313 314 315 316 317 318 319

	/*
	 * old_pte points to the large page base address. So we need
	 * to add the offset of the virtual address:
	 */
	pfn = pte_pfn(old_pte) + ((address & (psize - 1)) >> PAGE_SHIFT);
	cpa->pfn = pfn;

	new_prot = static_protections(new_prot, address, pfn);
320

321 322 323 324 325 326
	/*
	 * We need to check the full range, whether
	 * static_protection() requires a different pgprot for one of
	 * the pages in the range we try to preserve:
	 */
	addr = address + PAGE_SIZE;
T
Thomas Gleixner 已提交
327
	pfn++;
328
	for (i = 1; i < cpa->numpages; i++, addr += PAGE_SIZE, pfn++) {
T
Thomas Gleixner 已提交
329
		pgprot_t chk_prot = static_protections(new_prot, addr, pfn);
330 331 332 333 334

		if (pgprot_val(chk_prot) != pgprot_val(new_prot))
			goto out_unlock;
	}

335 336 337 338 339
	/*
	 * If there are no changes, return. maxpages has been updated
	 * above:
	 */
	if (pgprot_val(new_prot) == pgprot_val(old_prot)) {
I
Ingo Molnar 已提交
340
		do_split = 0;
341 342 343 344 345 346 347 348 349 350 351
		goto out_unlock;
	}

	/*
	 * We need to change the attributes. Check, whether we can
	 * change the large page in one go. We request a split, when
	 * the address is not aligned and the number of pages is
	 * smaller than the number of pages in the large page. Note
	 * that we limited the number of possible pages already to
	 * the number of pages in the large page.
	 */
352
	if (address == (nextpage_addr - psize) && cpa->numpages == numpages) {
353 354 355 356 357 358 359
		/*
		 * The address is aligned and the number of pages
		 * covers the full page.
		 */
		new_pte = pfn_pte(pte_pfn(old_pte), canon_pgprot(new_prot));
		__set_pmd_pte(kpte, address, new_pte);
		cpa->flushtlb = 1;
I
Ingo Molnar 已提交
360
		do_split = 0;
361 362 363 364
	}

out_unlock:
	spin_unlock_irqrestore(&pgd_lock, flags);
I
Ingo Molnar 已提交
365

I
Ingo Molnar 已提交
366
	return do_split;
367 368
}

369 370
static LIST_HEAD(page_pool);
static unsigned long pool_size, pool_pages, pool_low;
371
static unsigned long pool_used, pool_failed;
372

373
static void cpa_fill_pool(struct page **ret)
374 375
{
	gfp_t gfp = GFP_KERNEL;
376 377
	unsigned long flags;
	struct page *p;
378 379

	/*
380 381
	 * Avoid recursion (on debug-pagealloc) and also signal
	 * our priority to get to these pagetables:
382
	 */
383
	if (current->flags & PF_MEMALLOC)
384
		return;
385
	current->flags |= PF_MEMALLOC;
386 387

	/*
388
	 * Allocate atomically from atomic contexts:
389
	 */
390 391
	if (in_atomic() || irqs_disabled() || debug_pagealloc)
		gfp =  GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
392

393
	while (pool_pages < pool_size || (ret && !*ret)) {
394 395 396 397 398
		p = alloc_pages(gfp, 0);
		if (!p) {
			pool_failed++;
			break;
		}
399 400 401 402 403 404 405 406
		/*
		 * If the call site needs a page right now, provide it:
		 */
		if (ret && !*ret) {
			*ret = p;
			continue;
		}
		spin_lock_irqsave(&pgd_lock, flags);
407 408
		list_add(&p->lru, &page_pool);
		pool_pages++;
409
		spin_unlock_irqrestore(&pgd_lock, flags);
410
	}
411 412

	current->flags &= ~PF_MEMALLOC;
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432
}

#define SHIFT_MB		(20 - PAGE_SHIFT)
#define ROUND_MB_GB		((1 << 10) - 1)
#define SHIFT_MB_GB		10
#define POOL_PAGES_PER_GB	16

void __init cpa_init(void)
{
	struct sysinfo si;
	unsigned long gb;

	si_meminfo(&si);
	/*
	 * Calculate the number of pool pages:
	 *
	 * Convert totalram (nr of pages) to MiB and round to the next
	 * GiB. Shift MiB to Gib and multiply the result by
	 * POOL_PAGES_PER_GB:
	 */
433 434 435 436 437 438
	if (debug_pagealloc) {
		gb = ((si.totalram >> SHIFT_MB) + ROUND_MB_GB) >> SHIFT_MB_GB;
		pool_size = POOL_PAGES_PER_GB * gb;
	} else {
		pool_size = 1;
	}
439 440
	pool_low = pool_size;

441
	cpa_fill_pool(NULL);
442 443 444 445 446
	printk(KERN_DEBUG
	       "CPA: page pool initialized %lu of %lu pages preallocated\n",
	       pool_pages, pool_size);
}

447
static int split_large_page(pte_t *kpte, unsigned long address)
448
{
T
Thomas Gleixner 已提交
449
	unsigned long flags, pfn, pfninc = 1;
I
Ingo Molnar 已提交
450
	unsigned int i, level;
451
	pte_t *pbase, *tmp;
I
Ingo Molnar 已提交
452
	pgprot_t ref_prot;
453 454
	struct page *base;

T
Thomas Gleixner 已提交
455 456 457 458 459 460 461 462
	/*
	 * Get a page from the pool. The pool list is protected by the
	 * pgd_lock, which we have to take anyway for the split
	 * operation:
	 */
	spin_lock_irqsave(&pgd_lock, flags);
	if (list_empty(&page_pool)) {
		spin_unlock_irqrestore(&pgd_lock, flags);
463 464 465 466 467 468 469 470 471 472 473 474
		base = NULL;
		cpa_fill_pool(&base);
		if (!base)
			return -ENOMEM;
		spin_lock_irqsave(&pgd_lock, flags);
	} else {
		base = list_first_entry(&page_pool, struct page, lru);
		list_del(&base->lru);
		pool_pages--;

		if (pool_pages < pool_low)
			pool_low = pool_pages;
T
Thomas Gleixner 已提交
475 476
	}

477 478 479 480 481
	/*
	 * Check for races, another CPU might have split this page
	 * up for us already:
	 */
	tmp = lookup_address(address, &level);
I
Ingo Molnar 已提交
482
	if (tmp != kpte)
483 484 485
		goto out_unlock;

	pbase = (pte_t *)page_address(base);
486
	paravirt_alloc_pte(&init_mm, page_to_pfn(base));
T
Thomas Gleixner 已提交
487
	ref_prot = pte_pgprot(pte_clrhuge(*kpte));
488

489 490 491 492 493 494 495
#ifdef CONFIG_X86_64
	if (level == PG_LEVEL_1G) {
		pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
		pgprot_val(ref_prot) |= _PAGE_PSE;
	}
#endif

496 497 498 499
	/*
	 * Get the target pfn from the original entry:
	 */
	pfn = pte_pfn(*kpte);
500
	for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc)
501
		set_pte(&pbase[i], pfn_pte(pfn, ref_prot));
502 503

	/*
T
Thomas Gleixner 已提交
504
	 * Install the new, split up pagetable. Important details here:
505 506 507 508
	 *
	 * On Intel the NX bit of all levels must be cleared to make a
	 * page executable. See section 4.13.2 of Intel 64 and IA-32
	 * Architectures Software Developer's Manual).
T
Thomas Gleixner 已提交
509 510 511
	 *
	 * Mark the entry present. The current mapping might be
	 * set to not present, which we preserved above.
512
	 */
513
	ref_prot = pte_pgprot(pte_mkexec(pte_clrhuge(*kpte)));
T
Thomas Gleixner 已提交
514
	pgprot_val(ref_prot) |= _PAGE_PRESENT;
I
Ingo Molnar 已提交
515
	__set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
516 517 518
	base = NULL;

out_unlock:
T
Thomas Gleixner 已提交
519 520 521 522 523 524 525 526 527
	/*
	 * If we dropped out via the lookup_address check under
	 * pgd_lock then stick the page back into the pool:
	 */
	if (base) {
		list_add(&base->lru, &page_pool);
		pool_pages++;
	} else
		pool_used++;
I
Ingo Molnar 已提交
528
	spin_unlock_irqrestore(&pgd_lock, flags);
529 530 531 532

	return 0;
}

T
Thomas Gleixner 已提交
533
static int __change_page_attr(struct cpa_data *cpa, int primary)
534
{
T
Thomas Gleixner 已提交
535
	unsigned long address = cpa->vaddr;
536 537
	int do_split, err;
	unsigned int level;
T
Thomas Gleixner 已提交
538
	pte_t *kpte, old_pte;
L
Linus Torvalds 已提交
539

540
repeat:
541
	kpte = lookup_address(address, &level);
L
Linus Torvalds 已提交
542
	if (!kpte)
I
Ingo Molnar 已提交
543
		return 0;
T
Thomas Gleixner 已提交
544 545 546 547 548 549 550 551 552

	old_pte = *kpte;
	if (!pte_val(old_pte)) {
		if (!primary)
			return 0;
		printk(KERN_WARNING "CPA: called for zero pte. "
		       "vaddr = %lx cpa->vaddr = %lx\n", address,
		       cpa->vaddr);
		WARN_ON(1);
L
Linus Torvalds 已提交
553
		return -EINVAL;
T
Thomas Gleixner 已提交
554
	}
555

T
Thomas Gleixner 已提交
556
	if (level == PG_LEVEL_4K) {
T
Thomas Gleixner 已提交
557
		pte_t new_pte;
558
		pgprot_t new_prot = pte_pgprot(old_pte);
T
Thomas Gleixner 已提交
559
		unsigned long pfn = pte_pfn(old_pte);
I
Ingo Molnar 已提交
560

T
Thomas Gleixner 已提交
561 562
		pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
		pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
I
Ingo Molnar 已提交
563

T
Thomas Gleixner 已提交
564
		new_prot = static_protections(new_prot, address, pfn);
I
Ingo Molnar 已提交
565

566 567 568 569 570
		/*
		 * We need to keep the pfn from the existing PTE,
		 * after all we're only going to change it's attributes
		 * not the memory it points to
		 */
T
Thomas Gleixner 已提交
571 572
		new_pte = pfn_pte(pfn, canon_pgprot(new_prot));
		cpa->pfn = pfn;
573 574 575 576 577 578 579
		/*
		 * Do we really change anything ?
		 */
		if (pte_val(old_pte) != pte_val(new_pte)) {
			set_pte_atomic(kpte, new_pte);
			cpa->flushtlb = 1;
		}
580
		cpa->numpages = 1;
581
		return 0;
L
Linus Torvalds 已提交
582
	}
583 584 585 586 587

	/*
	 * Check, whether we can keep the large page intact
	 * and just change the pte:
	 */
I
Ingo Molnar 已提交
588
	do_split = try_preserve_large_page(kpte, address, cpa);
589 590
	/*
	 * When the range fits into the existing large page,
591
	 * return. cp->numpages and cpa->tlbflush have been updated in
592 593
	 * try_large_page:
	 */
I
Ingo Molnar 已提交
594 595
	if (do_split <= 0)
		return do_split;
596 597 598 599

	/*
	 * We have to split the large page:
	 */
I
Ingo Molnar 已提交
600 601 602 603 604
	err = split_large_page(kpte, address);
	if (!err) {
		cpa->flushtlb = 1;
		goto repeat;
	}
I
Ingo Molnar 已提交
605

I
Ingo Molnar 已提交
606
	return err;
607
}
L
Linus Torvalds 已提交
608

T
Thomas Gleixner 已提交
609 610 611
static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias);

static int cpa_process_alias(struct cpa_data *cpa)
L
Linus Torvalds 已提交
612
{
T
Thomas Gleixner 已提交
613
	struct cpa_data alias_cpa;
614
	int ret = 0;
615

T
Thomas Gleixner 已提交
616 617
	if (cpa->pfn > max_pfn_mapped)
		return 0;
618

619 620 621 622 623 624
	/*
	 * No need to redo, when the primary call touched the direct
	 * mapping already:
	 */
	if (!within(cpa->vaddr, PAGE_OFFSET,
		    PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))) {
625

626 627 628 629 630
		alias_cpa = *cpa;
		alias_cpa.vaddr = (unsigned long) __va(cpa->pfn << PAGE_SHIFT);

		ret = __change_page_attr_set_clr(&alias_cpa, 0);
	}
631 632

#ifdef CONFIG_X86_64
T
Thomas Gleixner 已提交
633 634
	if (ret)
		return ret;
635 636 637 638 639 640 641
	/*
	 * No need to redo, when the primary call touched the high
	 * mapping already:
	 */
	if (within(cpa->vaddr, (unsigned long) _text, (unsigned long) _end))
		return 0;

A
Arjan van de Ven 已提交
642
	/*
643 644
	 * If the physical address is inside the kernel map, we need
	 * to touch the high mapped kernel as well:
A
Arjan van de Ven 已提交
645
	 */
T
Thomas Gleixner 已提交
646 647
	if (!within(cpa->pfn, highmap_start_pfn(), highmap_end_pfn()))
		return 0;
648

T
Thomas Gleixner 已提交
649 650 651 652 653 654 655 656
	alias_cpa = *cpa;
	alias_cpa.vaddr =
		(cpa->pfn << PAGE_SHIFT) + __START_KERNEL_map - phys_base;

	/*
	 * The high mapping range is imprecise, so ignore the return value.
	 */
	__change_page_attr_set_clr(&alias_cpa, 0);
A
Arjan van de Ven 已提交
657
#endif
T
Thomas Gleixner 已提交
658
	return ret;
L
Linus Torvalds 已提交
659 660
}

T
Thomas Gleixner 已提交
661
static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
662
{
663
	int ret, numpages = cpa->numpages;
664

665 666 667 668 669
	while (numpages) {
		/*
		 * Store the remaining nr of pages for the large page
		 * preservation check.
		 */
670
		cpa->numpages = numpages;
T
Thomas Gleixner 已提交
671 672

		ret = __change_page_attr(cpa, checkalias);
673 674 675
		if (ret)
			return ret;

T
Thomas Gleixner 已提交
676 677 678 679 680 681
		if (checkalias) {
			ret = cpa_process_alias(cpa);
			if (ret)
				return ret;
		}

682 683 684 685 686
		/*
		 * Adjust the number of pages with the result of the
		 * CPA operation. Either a large page has been
		 * preserved or a single page update happened.
		 */
687 688 689
		BUG_ON(cpa->numpages > numpages);
		numpages -= cpa->numpages;
		cpa->vaddr += cpa->numpages * PAGE_SIZE;
690
	}
691 692 693
	return 0;
}

694 695 696 697 698 699
static inline int cache_attr(pgprot_t attr)
{
	return pgprot_val(attr) &
		(_PAGE_PAT | _PAGE_PAT_LARGE | _PAGE_PWT | _PAGE_PCD);
}

700
static int change_page_attr_set_clr(unsigned long addr, int numpages,
701 702
				    pgprot_t mask_set, pgprot_t mask_clr,
				    int force_split)
703
{
T
Thomas Gleixner 已提交
704
	struct cpa_data cpa;
705
	int ret, cache, checkalias;
706 707 708 709 710 711 712

	/*
	 * Check, if we are requested to change a not supported
	 * feature:
	 */
	mask_set = canon_pgprot(mask_set);
	mask_clr = canon_pgprot(mask_clr);
713
	if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
714 715
		return 0;

716 717 718 719 720 721 722 723 724
	/* Ensure we are PAGE_SIZE aligned */
	if (addr & ~PAGE_MASK) {
		addr &= PAGE_MASK;
		/*
		 * People should not be passing in unaligned addresses:
		 */
		WARN_ON_ONCE(1);
	}

T
Thomas Gleixner 已提交
725 726 727 728
	cpa.vaddr = addr;
	cpa.numpages = numpages;
	cpa.mask_set = mask_set;
	cpa.mask_clr = mask_clr;
729
	cpa.flushtlb = 0;
730
	cpa.force_split = force_split;
T
Thomas Gleixner 已提交
731

732 733 734 735
	/* No alias checking for _NX bit modifications */
	checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX;

	ret = __change_page_attr_set_clr(&cpa, checkalias);
736

737 738 739 740
	/*
	 * Check whether we really changed something:
	 */
	if (!cpa.flushtlb)
741
		goto out;
742

743 744 745 746 747 748
	/*
	 * No need to flush, when we did not set any of the caching
	 * attributes:
	 */
	cache = cache_attr(mask_set);

749 750 751
	/*
	 * On success we use clflush, when the CPU supports it to
	 * avoid the wbindv. If the CPU does not support it and in the
752
	 * error case we fall back to cpa_flush_all (which uses
753 754 755
	 * wbindv):
	 */
	if (!ret && cpu_has_clflush)
756
		cpa_flush_range(addr, numpages, cache);
757
	else
758
		cpa_flush_all(cache);
759

760
out:
761 762
	cpa_fill_pool(NULL);

763 764 765
	return ret;
}

766 767
static inline int change_page_attr_set(unsigned long addr, int numpages,
				       pgprot_t mask)
768
{
769
	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0);
770 771
}

772 773
static inline int change_page_attr_clear(unsigned long addr, int numpages,
					 pgprot_t mask)
774
{
775
	return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0);
776 777
}

778
int _set_memory_uc(unsigned long addr, int numpages)
779 780
{
	return change_page_attr_set(addr, numpages,
781
				    __pgprot(_PAGE_CACHE_UC));
782
}
783 784 785 786 787 788 789 790 791

int set_memory_uc(unsigned long addr, int numpages)
{
	if (reserve_memtype(addr, addr + numpages * PAGE_SIZE,
	                    _PAGE_CACHE_UC, NULL))
		return -EINVAL;

	return _set_memory_uc(addr, numpages);
}
792 793
EXPORT_SYMBOL(set_memory_uc);

794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812
int _set_memory_wc(unsigned long addr, int numpages)
{
	return change_page_attr_set(addr, numpages,
				    __pgprot(_PAGE_CACHE_WC));
}

int set_memory_wc(unsigned long addr, int numpages)
{
	if (!pat_wc_enabled)
		return set_memory_uc(addr, numpages);

	if (reserve_memtype(addr, addr + numpages * PAGE_SIZE,
		_PAGE_CACHE_WC, NULL))
		return -EINVAL;

	return _set_memory_wc(addr, numpages);
}
EXPORT_SYMBOL(set_memory_wc);

813
int _set_memory_wb(unsigned long addr, int numpages)
814
{
815
	return change_page_attr_clear(addr, numpages,
816
				      __pgprot(_PAGE_CACHE_MASK));
817
}
818 819 820 821 822 823 824

int set_memory_wb(unsigned long addr, int numpages)
{
	free_memtype(addr, addr + numpages * PAGE_SIZE);

	return _set_memory_wb(addr, numpages);
}
825 826 827 828
EXPORT_SYMBOL(set_memory_wb);

int set_memory_x(unsigned long addr, int numpages)
{
829
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_NX));
830 831 832 833 834
}
EXPORT_SYMBOL(set_memory_x);

int set_memory_nx(unsigned long addr, int numpages)
{
835
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_NX));
836 837 838 839 840
}
EXPORT_SYMBOL(set_memory_nx);

int set_memory_ro(unsigned long addr, int numpages)
{
841
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_RW));
842 843 844 845
}

int set_memory_rw(unsigned long addr, int numpages)
{
846
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_RW));
847
}
I
Ingo Molnar 已提交
848 849 850

int set_memory_np(unsigned long addr, int numpages)
{
851
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_PRESENT));
I
Ingo Molnar 已提交
852
}
853

854 855 856 857 858 859
int set_memory_4k(unsigned long addr, int numpages)
{
	return change_page_attr_set_clr(addr, numpages, __pgprot(0),
					__pgprot(0), 1);
}

860 861 862 863
int set_pages_uc(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

T
Thomas Gleixner 已提交
864
	return set_memory_uc(addr, numpages);
865 866 867 868 869 870 871
}
EXPORT_SYMBOL(set_pages_uc);

int set_pages_wb(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

T
Thomas Gleixner 已提交
872
	return set_memory_wb(addr, numpages);
873 874 875 876 877 878 879
}
EXPORT_SYMBOL(set_pages_wb);

int set_pages_x(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

T
Thomas Gleixner 已提交
880
	return set_memory_x(addr, numpages);
881 882 883 884 885 886 887
}
EXPORT_SYMBOL(set_pages_x);

int set_pages_nx(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

T
Thomas Gleixner 已提交
888
	return set_memory_nx(addr, numpages);
889 890 891 892 893 894 895
}
EXPORT_SYMBOL(set_pages_nx);

int set_pages_ro(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

T
Thomas Gleixner 已提交
896
	return set_memory_ro(addr, numpages);
897 898 899 900 901
}

int set_pages_rw(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);
902

T
Thomas Gleixner 已提交
903
	return set_memory_rw(addr, numpages);
I
Ingo Molnar 已提交
904 905
}

L
Linus Torvalds 已提交
906
#ifdef CONFIG_DEBUG_PAGEALLOC
I
Ingo Molnar 已提交
907 908 909

static int __set_pages_p(struct page *page, int numpages)
{
T
Thomas Gleixner 已提交
910 911 912 913
	struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page),
				.numpages = numpages,
				.mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
				.mask_clr = __pgprot(0)};
914

T
Thomas Gleixner 已提交
915
	return __change_page_attr_set_clr(&cpa, 1);
I
Ingo Molnar 已提交
916 917 918 919
}

static int __set_pages_np(struct page *page, int numpages)
{
T
Thomas Gleixner 已提交
920 921 922 923
	struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page),
				.numpages = numpages,
				.mask_set = __pgprot(0),
				.mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW)};
924

T
Thomas Gleixner 已提交
925
	return __change_page_attr_set_clr(&cpa, 1);
I
Ingo Molnar 已提交
926 927
}

L
Linus Torvalds 已提交
928 929 930 931
void kernel_map_pages(struct page *page, int numpages, int enable)
{
	if (PageHighMem(page))
		return;
932
	if (!enable) {
933 934
		debug_check_no_locks_freed(page_address(page),
					   numpages * PAGE_SIZE);
935
	}
936

937 938 939 940 941 942
	/*
	 * If page allocator is not up yet then do not call c_p_a():
	 */
	if (!debug_pagealloc_enabled)
		return;

943
	/*
I
Ingo Molnar 已提交
944 945 946 947 948 949
	 * The return value is ignored as the calls cannot fail.
	 * Large pages are kept enabled at boot time, and are
	 * split up quickly with DEBUG_PAGEALLOC. If a splitup
	 * fails here (due to temporary memory shortage) no damage
	 * is done because we just keep the largepage intact up
	 * to the next attempt when it will likely be split up:
L
Linus Torvalds 已提交
950
	 */
I
Ingo Molnar 已提交
951 952 953 954
	if (enable)
		__set_pages_p(page, numpages);
	else
		__set_pages_np(page, numpages);
955 956

	/*
957 958
	 * We should perform an IPI and flush all tlbs,
	 * but that can deadlock->flush only current cpu:
L
Linus Torvalds 已提交
959 960
	 */
	__flush_tlb_all();
961 962 963 964 965

	/*
	 * Try to refill the page pool here. We can do this only after
	 * the tlb flush.
	 */
966
	cpa_fill_pool(NULL);
L
Linus Torvalds 已提交
967
}
968

969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993
#ifdef CONFIG_DEBUG_FS
static int dpa_show(struct seq_file *m, void *v)
{
	seq_puts(m, "DEBUG_PAGEALLOC\n");
	seq_printf(m, "pool_size     : %lu\n", pool_size);
	seq_printf(m, "pool_pages    : %lu\n", pool_pages);
	seq_printf(m, "pool_low      : %lu\n", pool_low);
	seq_printf(m, "pool_used     : %lu\n", pool_used);
	seq_printf(m, "pool_failed   : %lu\n", pool_failed);

	return 0;
}

static int dpa_open(struct inode *inode, struct file *filp)
{
	return single_open(filp, dpa_show, NULL);
}

static const struct file_operations dpa_fops = {
	.open		= dpa_open,
	.read		= seq_read,
	.llseek		= seq_lseek,
	.release	= single_release,
};

I
Ingo Molnar 已提交
994
static int __init debug_pagealloc_proc_init(void)
995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007
{
	struct dentry *de;

	de = debugfs_create_file("debug_pagealloc", 0600, NULL, NULL,
				 &dpa_fops);
	if (!de)
		return -ENOMEM;

	return 0;
}
__initcall(debug_pagealloc_proc_init);
#endif

1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024
#ifdef CONFIG_HIBERNATION

bool kernel_page_present(struct page *page)
{
	unsigned int level;
	pte_t *pte;

	if (PageHighMem(page))
		return false;

	pte = lookup_address((unsigned long)page_address(page), &level);
	return (pte_val(*pte) & _PAGE_PRESENT);
}

#endif /* CONFIG_HIBERNATION */

#endif /* CONFIG_DEBUG_PAGEALLOC */
1025 1026 1027 1028 1029 1030 1031 1032

/*
 * The testcases use internal knowledge of the implementation that shouldn't
 * be exposed to the rest of the kernel. Include these directly here.
 */
#ifdef CONFIG_CPA_DEBUG
#include "pageattr-test.c"
#endif