pageattr.c 25.0 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
};

37
#ifdef CONFIG_PROC_FS
38 39
static unsigned long direct_pages_count[PG_LEVEL_NUM];

40
void update_page_count(int level, unsigned long pages)
41 42
{
	unsigned long flags;
43

44 45 46 47
	/* Protect against CPA */
	spin_lock_irqsave(&pgd_lock, flags);
	direct_pages_count[level] += pages;
	spin_unlock_irqrestore(&pgd_lock, flags);
48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64
}

static void split_page_count(int level)
{
	direct_pages_count[level]--;
	direct_pages_count[level - 1] += PTRS_PER_PTE;
}

int arch_report_meminfo(char *page)
{
	int n = sprintf(page, "DirectMap4k:  %8lu\n"
			"DirectMap2M:  %8lu\n",
			direct_pages_count[PG_LEVEL_4K],
			direct_pages_count[PG_LEVEL_2M]);
#ifdef CONFIG_X86_64
	n += sprintf(page + n, "DirectMap1G:  %8lu\n",
		     direct_pages_count[PG_LEVEL_1G]);
65
#endif
66
	return n;
67
}
68 69 70
#else
static inline void split_page_count(int level) { }
#endif
71

T
Thomas Gleixner 已提交
72 73 74 75 76 77 78 79 80 81 82 83 84 85
#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

86 87 88 89 90 91
#ifdef CONFIG_DEBUG_PAGEALLOC
# define debug_pagealloc 1
#else
# define debug_pagealloc 0
#endif

92 93
static inline int
within(unsigned long addr, unsigned long start, unsigned long end)
I
Ingo Molnar 已提交
94
{
95 96 97
	return addr >= start && addr < end;
}

T
Thomas Gleixner 已提交
98 99 100
/*
 * Flushing functions
 */
101 102 103 104 105 106 107 108 109

/**
 * 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
110
void clflush_cache_range(void *vaddr, unsigned int size)
T
Thomas Gleixner 已提交
111
{
I
Ingo Molnar 已提交
112
	void *vend = vaddr + size - 1;
T
Thomas Gleixner 已提交
113

114
	mb();
I
Ingo Molnar 已提交
115 116 117 118 119 120 121 122

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

123
	mb();
T
Thomas Gleixner 已提交
124 125
}

126
static void __cpa_flush_all(void *arg)
T
Thomas Gleixner 已提交
127
{
128 129
	unsigned long cache = (unsigned long)arg;

T
Thomas Gleixner 已提交
130 131 132 133 134 135
	/*
	 * Flush all to work around Errata in early athlons regarding
	 * large page flushing.
	 */
	__flush_tlb_all();

136
	if (cache && boot_cpu_data.x86_model >= 4)
T
Thomas Gleixner 已提交
137 138 139
		wbinvd();
}

140
static void cpa_flush_all(unsigned long cache)
T
Thomas Gleixner 已提交
141 142 143
{
	BUG_ON(irqs_disabled());

144
	on_each_cpu(__cpa_flush_all, (void *) cache, 1);
T
Thomas Gleixner 已提交
145 146
}

147 148 149 150 151 152 153 154 155 156
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();
}

157
static void cpa_flush_range(unsigned long start, int numpages, int cache)
158
{
I
Ingo Molnar 已提交
159 160 161
	unsigned int i, level;
	unsigned long addr;

162
	BUG_ON(irqs_disabled());
I
Ingo Molnar 已提交
163
	WARN_ON(PAGE_ALIGN(start) != start);
164

165
	on_each_cpu(__cpa_flush_range, NULL, 1);
166

167 168 169
	if (!cache)
		return;

T
Thomas Gleixner 已提交
170 171 172 173 174 175
	/*
	 * 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
176 177 178 179 180 181
	for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
		pte_t *pte = lookup_address(addr, &level);

		/*
		 * Only flush present addresses:
		 */
182
		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
I
Ingo Molnar 已提交
183 184
			clflush_cache_range((void *) addr, PAGE_SIZE);
	}
185 186
}

187 188 189 190 191 192
/*
 * 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
193 194
static inline pgprot_t static_protections(pgprot_t prot, unsigned long address,
				   unsigned long pfn)
195 196 197
{
	pgprot_t forbidden = __pgprot(0);

I
Ingo Molnar 已提交
198
	/*
199 200
	 * 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
201
	 */
T
Thomas Gleixner 已提交
202
	if (within(pfn, BIOS_BEGIN >> PAGE_SHIFT, BIOS_END >> PAGE_SHIFT))
203 204 205 206
		pgprot_val(forbidden) |= _PAGE_NX;

	/*
	 * The kernel text needs to be executable for obvious reasons
T
Thomas Gleixner 已提交
207 208
	 * Does not cover __inittext since that is gone later on. On
	 * 64bit we do not enforce !NX on the low mapping
209 210 211
	 */
	if (within(address, (unsigned long)_text, (unsigned long)_etext))
		pgprot_val(forbidden) |= _PAGE_NX;
212 213

	/*
T
Thomas Gleixner 已提交
214 215
	 * The .rodata section needs to be read-only. Using the pfn
	 * catches all aliases.
216
	 */
T
Thomas Gleixner 已提交
217 218
	if (within(pfn, __pa((unsigned long)__start_rodata) >> PAGE_SHIFT,
		   __pa((unsigned long)__end_rodata) >> PAGE_SHIFT))
219
		pgprot_val(forbidden) |= _PAGE_RW;
220 221

	prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));
I
Ingo Molnar 已提交
222 223 224 225

	return prot;
}

T
Thomas Gleixner 已提交
226 227 228 229 230 231 232 233
/*
 * 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.
 */
234
pte_t *lookup_address(unsigned long address, unsigned int *level)
235
{
L
Linus Torvalds 已提交
236 237 238
	pgd_t *pgd = pgd_offset_k(address);
	pud_t *pud;
	pmd_t *pmd;
239

T
Thomas Gleixner 已提交
240 241
	*level = PG_LEVEL_NONE;

L
Linus Torvalds 已提交
242 243
	if (pgd_none(*pgd))
		return NULL;
I
Ingo Molnar 已提交
244

L
Linus Torvalds 已提交
245 246 247
	pud = pud_offset(pgd, address);
	if (pud_none(*pud))
		return NULL;
248 249 250 251 252

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

L
Linus Torvalds 已提交
253 254 255
	pmd = pmd_offset(pud, address);
	if (pmd_none(*pmd))
		return NULL;
T
Thomas Gleixner 已提交
256 257

	*level = PG_LEVEL_2M;
T
Thomas Gleixner 已提交
258
	if (pmd_large(*pmd) || !pmd_present(*pmd))
L
Linus Torvalds 已提交
259 260
		return (pte_t *)pmd;

T
Thomas Gleixner 已提交
261
	*level = PG_LEVEL_4K;
I
Ingo Molnar 已提交
262

263 264
	return pte_offset_kernel(pmd, address);
}
265
EXPORT_SYMBOL_GPL(lookup_address);
266

I
Ingo Molnar 已提交
267 268 269
/*
 * Set the new pmd in all the pgds we know about:
 */
I
Ingo Molnar 已提交
270
static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
271 272 273
{
	/* change init_mm */
	set_pte_atomic(kpte, pte);
274
#ifdef CONFIG_X86_32
275
	if (!SHARED_KERNEL_PMD) {
276 277
		struct page *page;

278
		list_for_each_entry(page, &pgd_list, lru) {
279 280 281 282 283 284 285 286 287
			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
288
	}
289
#endif
L
Linus Torvalds 已提交
290 291
}

I
Ingo Molnar 已提交
292 293 294
static int
try_preserve_large_page(pte_t *kpte, unsigned long address,
			struct cpa_data *cpa)
295
{
T
Thomas Gleixner 已提交
296
	unsigned long nextpage_addr, numpages, pmask, psize, flags, addr, pfn;
297 298
	pte_t new_pte, old_pte, *tmp;
	pgprot_t old_prot, new_prot;
299
	int i, do_split = 1;
300
	unsigned int level;
301

302 303 304
	if (cpa->force_split)
		return 1;

305 306 307 308 309 310 311 312 313 314 315
	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:
316 317
		psize = PMD_PAGE_SIZE;
		pmask = PMD_PAGE_MASK;
318
		break;
319
#ifdef CONFIG_X86_64
320
	case PG_LEVEL_1G:
321 322
		psize = PUD_PAGE_SIZE;
		pmask = PUD_PAGE_MASK;
323 324
		break;
#endif
325
	default:
I
Ingo Molnar 已提交
326
		do_split = -EINVAL;
327 328 329 330 331 332 333 334 335
		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;
336 337
	if (numpages < cpa->numpages)
		cpa->numpages = numpages;
338 339 340 341 342 343 344 345 346

	/*
	 * 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
347 348 349 350 351 352 353 354 355

	/*
	 * 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);
356

357 358 359 360 361 362
	/*
	 * 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
363
	pfn++;
364
	for (i = 1; i < cpa->numpages; i++, addr += PAGE_SIZE, pfn++) {
T
Thomas Gleixner 已提交
365
		pgprot_t chk_prot = static_protections(new_prot, addr, pfn);
366 367 368 369 370

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

371 372 373 374 375
	/*
	 * 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
376
		do_split = 0;
377 378 379 380 381 382 383 384 385 386 387
		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.
	 */
388
	if (address == (nextpage_addr - psize) && cpa->numpages == numpages) {
389 390 391 392 393 394 395
		/*
		 * 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
396
		do_split = 0;
397 398 399 400
	}

out_unlock:
	spin_unlock_irqrestore(&pgd_lock, flags);
I
Ingo Molnar 已提交
401

I
Ingo Molnar 已提交
402
	return do_split;
403 404
}

405 406
static LIST_HEAD(page_pool);
static unsigned long pool_size, pool_pages, pool_low;
407
static unsigned long pool_used, pool_failed;
408

409
static void cpa_fill_pool(struct page **ret)
410 411
{
	gfp_t gfp = GFP_KERNEL;
412 413
	unsigned long flags;
	struct page *p;
414 415

	/*
416 417
	 * Avoid recursion (on debug-pagealloc) and also signal
	 * our priority to get to these pagetables:
418
	 */
419
	if (current->flags & PF_MEMALLOC)
420
		return;
421
	current->flags |= PF_MEMALLOC;
422 423

	/*
424
	 * Allocate atomically from atomic contexts:
425
	 */
426 427
	if (in_atomic() || irqs_disabled() || debug_pagealloc)
		gfp =  GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
428

429
	while (pool_pages < pool_size || (ret && !*ret)) {
430 431 432 433 434
		p = alloc_pages(gfp, 0);
		if (!p) {
			pool_failed++;
			break;
		}
435 436 437 438 439 440 441 442
		/*
		 * If the call site needs a page right now, provide it:
		 */
		if (ret && !*ret) {
			*ret = p;
			continue;
		}
		spin_lock_irqsave(&pgd_lock, flags);
443 444
		list_add(&p->lru, &page_pool);
		pool_pages++;
445
		spin_unlock_irqrestore(&pgd_lock, flags);
446
	}
447 448

	current->flags &= ~PF_MEMALLOC;
449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468
}

#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:
	 */
469 470 471 472 473 474
	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;
	}
475 476
	pool_low = pool_size;

477
	cpa_fill_pool(NULL);
478 479 480 481 482
	printk(KERN_DEBUG
	       "CPA: page pool initialized %lu of %lu pages preallocated\n",
	       pool_pages, pool_size);
}

483
static int split_large_page(pte_t *kpte, unsigned long address)
484
{
T
Thomas Gleixner 已提交
485
	unsigned long flags, pfn, pfninc = 1;
I
Ingo Molnar 已提交
486
	unsigned int i, level;
487
	pte_t *pbase, *tmp;
I
Ingo Molnar 已提交
488
	pgprot_t ref_prot;
489 490
	struct page *base;

T
Thomas Gleixner 已提交
491 492 493 494 495 496 497 498
	/*
	 * 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);
499 500 501 502 503 504 505 506 507 508 509 510
		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
511 512
	}

513 514 515 516 517
	/*
	 * 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
518
	if (tmp != kpte)
519 520 521
		goto out_unlock;

	pbase = (pte_t *)page_address(base);
522
	paravirt_alloc_pte(&init_mm, page_to_pfn(base));
T
Thomas Gleixner 已提交
523
	ref_prot = pte_pgprot(pte_clrhuge(*kpte));
524

525 526 527 528 529 530 531
#ifdef CONFIG_X86_64
	if (level == PG_LEVEL_1G) {
		pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
		pgprot_val(ref_prot) |= _PAGE_PSE;
	}
#endif

532 533 534 535
	/*
	 * Get the target pfn from the original entry:
	 */
	pfn = pte_pfn(*kpte);
536
	for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc)
537
		set_pte(&pbase[i], pfn_pte(pfn, ref_prot));
538

539
	if (address >= (unsigned long)__va(0) &&
540 541 542 543 544
		address < (unsigned long)__va(max_low_pfn_mapped << PAGE_SHIFT))
		split_page_count(level);

#ifdef CONFIG_X86_64
	if (address >= (unsigned long)__va(1UL<<32) &&
545 546
		address < (unsigned long)__va(max_pfn_mapped << PAGE_SHIFT))
		split_page_count(level);
547
#endif
548

549
	/*
T
Thomas Gleixner 已提交
550
	 * Install the new, split up pagetable. Important details here:
551 552 553 554
	 *
	 * 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
555 556 557
	 *
	 * Mark the entry present. The current mapping might be
	 * set to not present, which we preserved above.
558
	 */
559
	ref_prot = pte_pgprot(pte_mkexec(pte_clrhuge(*kpte)));
T
Thomas Gleixner 已提交
560
	pgprot_val(ref_prot) |= _PAGE_PRESENT;
I
Ingo Molnar 已提交
561
	__set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
562 563 564
	base = NULL;

out_unlock:
T
Thomas Gleixner 已提交
565 566 567 568 569 570 571 572 573
	/*
	 * 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
574
	spin_unlock_irqrestore(&pgd_lock, flags);
575 576 577 578

	return 0;
}

T
Thomas Gleixner 已提交
579
static int __change_page_attr(struct cpa_data *cpa, int primary)
580
{
T
Thomas Gleixner 已提交
581
	unsigned long address = cpa->vaddr;
582 583
	int do_split, err;
	unsigned int level;
T
Thomas Gleixner 已提交
584
	pte_t *kpte, old_pte;
L
Linus Torvalds 已提交
585

586
repeat:
587
	kpte = lookup_address(address, &level);
L
Linus Torvalds 已提交
588
	if (!kpte)
I
Ingo Molnar 已提交
589
		return 0;
T
Thomas Gleixner 已提交
590 591 592 593 594 595 596 597 598

	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
599
		return -EINVAL;
T
Thomas Gleixner 已提交
600
	}
601

T
Thomas Gleixner 已提交
602
	if (level == PG_LEVEL_4K) {
T
Thomas Gleixner 已提交
603
		pte_t new_pte;
604
		pgprot_t new_prot = pte_pgprot(old_pte);
T
Thomas Gleixner 已提交
605
		unsigned long pfn = pte_pfn(old_pte);
I
Ingo Molnar 已提交
606

T
Thomas Gleixner 已提交
607 608
		pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
		pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
I
Ingo Molnar 已提交
609

T
Thomas Gleixner 已提交
610
		new_prot = static_protections(new_prot, address, pfn);
I
Ingo Molnar 已提交
611

612 613 614 615 616
		/*
		 * 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
617 618
		new_pte = pfn_pte(pfn, canon_pgprot(new_prot));
		cpa->pfn = pfn;
619 620 621 622 623 624 625
		/*
		 * Do we really change anything ?
		 */
		if (pte_val(old_pte) != pte_val(new_pte)) {
			set_pte_atomic(kpte, new_pte);
			cpa->flushtlb = 1;
		}
626
		cpa->numpages = 1;
627
		return 0;
L
Linus Torvalds 已提交
628
	}
629 630 631 632 633

	/*
	 * Check, whether we can keep the large page intact
	 * and just change the pte:
	 */
I
Ingo Molnar 已提交
634
	do_split = try_preserve_large_page(kpte, address, cpa);
635 636
	/*
	 * When the range fits into the existing large page,
637
	 * return. cp->numpages and cpa->tlbflush have been updated in
638 639
	 * try_large_page:
	 */
I
Ingo Molnar 已提交
640 641
	if (do_split <= 0)
		return do_split;
642 643 644 645

	/*
	 * We have to split the large page:
	 */
I
Ingo Molnar 已提交
646 647 648 649 650
	err = split_large_page(kpte, address);
	if (!err) {
		cpa->flushtlb = 1;
		goto repeat;
	}
I
Ingo Molnar 已提交
651

I
Ingo Molnar 已提交
652
	return err;
653
}
L
Linus Torvalds 已提交
654

T
Thomas Gleixner 已提交
655 656 657
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
658
{
T
Thomas Gleixner 已提交
659
	struct cpa_data alias_cpa;
660
	int ret = 0;
661

Y
Yinghai Lu 已提交
662
	if (cpa->pfn >= max_pfn_mapped)
T
Thomas Gleixner 已提交
663
		return 0;
664

665
#ifdef CONFIG_X86_64
Y
Yinghai Lu 已提交
666
	if (cpa->pfn >= max_low_pfn_mapped && cpa->pfn < (1UL<<(32-PAGE_SHIFT)))
667 668
		return 0;
#endif
669 670 671 672
	/*
	 * No need to redo, when the primary call touched the direct
	 * mapping already:
	 */
673 674 675 676 677 678 679
	if (!(within(cpa->vaddr, PAGE_OFFSET,
		    PAGE_OFFSET + (max_low_pfn_mapped << PAGE_SHIFT))
#ifdef CONFIG_X86_64
		|| within(cpa->vaddr, PAGE_OFFSET + (1UL<<32),
		    PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))
#endif
	)) {
680

681 682 683 684 685
		alias_cpa = *cpa;
		alias_cpa.vaddr = (unsigned long) __va(cpa->pfn << PAGE_SHIFT);

		ret = __change_page_attr_set_clr(&alias_cpa, 0);
	}
686 687

#ifdef CONFIG_X86_64
T
Thomas Gleixner 已提交
688 689
	if (ret)
		return ret;
690 691 692 693 694 695 696
	/*
	 * 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
697
	/*
698 699
	 * 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
700
	 */
T
Thomas Gleixner 已提交
701 702
	if (!within(cpa->pfn, highmap_start_pfn(), highmap_end_pfn()))
		return 0;
703

T
Thomas Gleixner 已提交
704 705 706 707 708 709 710 711
	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
712
#endif
T
Thomas Gleixner 已提交
713
	return ret;
L
Linus Torvalds 已提交
714 715
}

T
Thomas Gleixner 已提交
716
static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
717
{
718
	int ret, numpages = cpa->numpages;
719

720 721 722 723 724
	while (numpages) {
		/*
		 * Store the remaining nr of pages for the large page
		 * preservation check.
		 */
725
		cpa->numpages = numpages;
T
Thomas Gleixner 已提交
726 727

		ret = __change_page_attr(cpa, checkalias);
728 729 730
		if (ret)
			return ret;

T
Thomas Gleixner 已提交
731 732 733 734 735 736
		if (checkalias) {
			ret = cpa_process_alias(cpa);
			if (ret)
				return ret;
		}

737 738 739 740 741
		/*
		 * 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.
		 */
742 743 744
		BUG_ON(cpa->numpages > numpages);
		numpages -= cpa->numpages;
		cpa->vaddr += cpa->numpages * PAGE_SIZE;
745
	}
746 747 748
	return 0;
}

749 750 751 752 753 754
static inline int cache_attr(pgprot_t attr)
{
	return pgprot_val(attr) &
		(_PAGE_PAT | _PAGE_PAT_LARGE | _PAGE_PWT | _PAGE_PCD);
}

755
static int change_page_attr_set_clr(unsigned long addr, int numpages,
756 757
				    pgprot_t mask_set, pgprot_t mask_clr,
				    int force_split)
758
{
T
Thomas Gleixner 已提交
759
	struct cpa_data cpa;
760
	int ret, cache, checkalias;
761 762 763 764 765 766 767

	/*
	 * Check, if we are requested to change a not supported
	 * feature:
	 */
	mask_set = canon_pgprot(mask_set);
	mask_clr = canon_pgprot(mask_clr);
768
	if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
769 770
		return 0;

771 772 773 774 775 776 777 778 779
	/* 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
780 781 782 783
	cpa.vaddr = addr;
	cpa.numpages = numpages;
	cpa.mask_set = mask_set;
	cpa.mask_clr = mask_clr;
784
	cpa.flushtlb = 0;
785
	cpa.force_split = force_split;
T
Thomas Gleixner 已提交
786

787 788 789 790
	/* 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);
791

792 793 794 795
	/*
	 * Check whether we really changed something:
	 */
	if (!cpa.flushtlb)
796
		goto out;
797

798 799 800 801 802 803
	/*
	 * No need to flush, when we did not set any of the caching
	 * attributes:
	 */
	cache = cache_attr(mask_set);

804 805 806
	/*
	 * On success we use clflush, when the CPU supports it to
	 * avoid the wbindv. If the CPU does not support it and in the
807
	 * error case we fall back to cpa_flush_all (which uses
808 809 810
	 * wbindv):
	 */
	if (!ret && cpu_has_clflush)
811
		cpa_flush_range(addr, numpages, cache);
812
	else
813
		cpa_flush_all(cache);
814

815
out:
816 817
	cpa_fill_pool(NULL);

818 819 820
	return ret;
}

821 822
static inline int change_page_attr_set(unsigned long addr, int numpages,
				       pgprot_t mask)
823
{
824
	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0);
825 826
}

827 828
static inline int change_page_attr_clear(unsigned long addr, int numpages,
					 pgprot_t mask)
829
{
830
	return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0);
831 832
}

833
int _set_memory_uc(unsigned long addr, int numpages)
834
{
835 836 837
	/*
	 * for now UC MINUS. see comments in ioremap_nocache()
	 */
838
	return change_page_attr_set(addr, numpages,
839
				    __pgprot(_PAGE_CACHE_UC_MINUS));
840
}
841 842 843

int set_memory_uc(unsigned long addr, int numpages)
{
844 845 846
	/*
	 * for now UC MINUS. see comments in ioremap_nocache()
	 */
847
	if (reserve_memtype(addr, addr + numpages * PAGE_SIZE,
848
			    _PAGE_CACHE_UC_MINUS, NULL))
849 850 851 852
		return -EINVAL;

	return _set_memory_uc(addr, numpages);
}
853 854
EXPORT_SYMBOL(set_memory_uc);

855 856 857 858 859 860 861 862
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)
{
863
	if (!pat_enabled)
864 865 866 867 868 869 870 871 872 873
		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);

874
int _set_memory_wb(unsigned long addr, int numpages)
875
{
876
	return change_page_attr_clear(addr, numpages,
877
				      __pgprot(_PAGE_CACHE_MASK));
878
}
879 880 881 882 883 884 885

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

	return _set_memory_wb(addr, numpages);
}
886 887 888 889
EXPORT_SYMBOL(set_memory_wb);

int set_memory_x(unsigned long addr, int numpages)
{
890
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_NX));
891 892 893 894 895
}
EXPORT_SYMBOL(set_memory_x);

int set_memory_nx(unsigned long addr, int numpages)
{
896
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_NX));
897 898 899 900 901
}
EXPORT_SYMBOL(set_memory_nx);

int set_memory_ro(unsigned long addr, int numpages)
{
902
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_RW));
903 904 905 906
}

int set_memory_rw(unsigned long addr, int numpages)
{
907
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_RW));
908
}
I
Ingo Molnar 已提交
909 910 911

int set_memory_np(unsigned long addr, int numpages)
{
912
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_PRESENT));
I
Ingo Molnar 已提交
913
}
914

915 916 917 918 919 920
int set_memory_4k(unsigned long addr, int numpages)
{
	return change_page_attr_set_clr(addr, numpages, __pgprot(0),
					__pgprot(0), 1);
}

921 922 923 924
int set_pages_uc(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);

T
Thomas Gleixner 已提交
925
	return set_memory_uc(addr, numpages);
926 927 928 929 930 931 932
}
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
933
	return set_memory_wb(addr, numpages);
934 935 936 937 938 939 940
}
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
941
	return set_memory_x(addr, numpages);
942 943 944 945 946 947 948
}
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
949
	return set_memory_nx(addr, numpages);
950 951 952 953 954 955 956
}
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
957
	return set_memory_ro(addr, numpages);
958 959 960 961 962
}

int set_pages_rw(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);
963

T
Thomas Gleixner 已提交
964
	return set_memory_rw(addr, numpages);
I
Ingo Molnar 已提交
965 966
}

L
Linus Torvalds 已提交
967
#ifdef CONFIG_DEBUG_PAGEALLOC
I
Ingo Molnar 已提交
968 969 970

static int __set_pages_p(struct page *page, int numpages)
{
T
Thomas Gleixner 已提交
971 972 973 974
	struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page),
				.numpages = numpages,
				.mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
				.mask_clr = __pgprot(0)};
975

T
Thomas Gleixner 已提交
976
	return __change_page_attr_set_clr(&cpa, 1);
I
Ingo Molnar 已提交
977 978 979 980
}

static int __set_pages_np(struct page *page, int numpages)
{
T
Thomas Gleixner 已提交
981 982 983 984
	struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page),
				.numpages = numpages,
				.mask_set = __pgprot(0),
				.mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW)};
985

T
Thomas Gleixner 已提交
986
	return __change_page_attr_set_clr(&cpa, 1);
I
Ingo Molnar 已提交
987 988
}

L
Linus Torvalds 已提交
989 990 991 992
void kernel_map_pages(struct page *page, int numpages, int enable)
{
	if (PageHighMem(page))
		return;
993
	if (!enable) {
994 995
		debug_check_no_locks_freed(page_address(page),
					   numpages * PAGE_SIZE);
996
	}
997

998 999 1000 1001 1002 1003
	/*
	 * If page allocator is not up yet then do not call c_p_a():
	 */
	if (!debug_pagealloc_enabled)
		return;

1004
	/*
I
Ingo Molnar 已提交
1005 1006 1007 1008 1009 1010
	 * 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
1011
	 */
I
Ingo Molnar 已提交
1012 1013 1014 1015
	if (enable)
		__set_pages_p(page, numpages);
	else
		__set_pages_np(page, numpages);
1016 1017

	/*
1018 1019
	 * We should perform an IPI and flush all tlbs,
	 * but that can deadlock->flush only current cpu:
L
Linus Torvalds 已提交
1020 1021
	 */
	__flush_tlb_all();
1022 1023 1024 1025 1026

	/*
	 * Try to refill the page pool here. We can do this only after
	 * the tlb flush.
	 */
1027
	cpa_fill_pool(NULL);
L
Linus Torvalds 已提交
1028
}
1029

1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
#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
1055
static int __init debug_pagealloc_proc_init(void)
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068
{
	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

1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085
#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 */
1086 1087 1088 1089 1090 1091 1092 1093

/*
 * 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