pageattr.c 20.0 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 <asm/e820.h>
L
Linus Torvalds 已提交
14 15
#include <asm/processor.h>
#include <asm/tlbflush.h>
D
Dave Jones 已提交
16
#include <asm/sections.h>
17 18
#include <asm/uaccess.h>
#include <asm/pgalloc.h>
L
Linus Torvalds 已提交
19

I
Ingo Molnar 已提交
20 21 22
/*
 * The current flushing context - we pass it instead of 5 arguments:
 */
T
Thomas Gleixner 已提交
23 24 25 26
struct cpa_data {
	unsigned long	vaddr;
	pgprot_t	mask_set;
	pgprot_t	mask_clr;
27
	int		numpages;
28
	int		flushtlb;
T
Thomas Gleixner 已提交
29 30
};

31 32
static inline int
within(unsigned long addr, unsigned long start, unsigned long end)
I
Ingo Molnar 已提交
33
{
34 35 36
	return addr >= start && addr < end;
}

T
Thomas Gleixner 已提交
37 38 39
/*
 * Flushing functions
 */
40 41 42 43 44 45 46 47 48

/**
 * 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
49
void clflush_cache_range(void *vaddr, unsigned int size)
T
Thomas Gleixner 已提交
50
{
I
Ingo Molnar 已提交
51
	void *vend = vaddr + size - 1;
T
Thomas Gleixner 已提交
52

53
	mb();
I
Ingo Molnar 已提交
54 55 56 57 58 59 60 61

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

62
	mb();
T
Thomas Gleixner 已提交
63 64
}

65
static void __cpa_flush_all(void *arg)
T
Thomas Gleixner 已提交
66
{
67 68
	unsigned long cache = (unsigned long)arg;

T
Thomas Gleixner 已提交
69 70 71 72 73 74
	/*
	 * Flush all to work around Errata in early athlons regarding
	 * large page flushing.
	 */
	__flush_tlb_all();

75
	if (cache && boot_cpu_data.x86_model >= 4)
T
Thomas Gleixner 已提交
76 77 78
		wbinvd();
}

79
static void cpa_flush_all(unsigned long cache)
T
Thomas Gleixner 已提交
80 81 82
{
	BUG_ON(irqs_disabled());

83
	on_each_cpu(__cpa_flush_all, (void *) cache, 1, 1);
T
Thomas Gleixner 已提交
84 85
}

86 87 88 89 90 91 92 93 94 95
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();
}

96
static void cpa_flush_range(unsigned long start, int numpages, int cache)
97
{
I
Ingo Molnar 已提交
98 99 100
	unsigned int i, level;
	unsigned long addr;

101
	BUG_ON(irqs_disabled());
I
Ingo Molnar 已提交
102
	WARN_ON(PAGE_ALIGN(start) != start);
103

T
Thomas Gleixner 已提交
104
	on_each_cpu(__cpa_flush_range, NULL, 1, 1);
105

106 107 108
	if (!cache)
		return;

T
Thomas Gleixner 已提交
109 110 111 112 113 114
	/*
	 * 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
115 116 117 118 119 120
	for (i = 0, addr = start; i < numpages; i++, addr += PAGE_SIZE) {
		pte_t *pte = lookup_address(addr, &level);

		/*
		 * Only flush present addresses:
		 */
121
		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
I
Ingo Molnar 已提交
122 123
			clflush_cache_range((void *) addr, PAGE_SIZE);
	}
124 125
}

126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141
#define HIGH_MAP_START	__START_KERNEL_map
#define HIGH_MAP_END	(__START_KERNEL_map + KERNEL_TEXT_SIZE)


/*
 * Converts a virtual address to a X86-64 highmap address
 */
static unsigned long virt_to_highmap(void *address)
{
#ifdef CONFIG_X86_64
	return __pa((unsigned long)address) + HIGH_MAP_START - phys_base;
#else
	return (unsigned long)address;
#endif
}

142 143 144 145 146 147 148 149 150 151
/*
 * 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.
 */
static inline pgprot_t static_protections(pgprot_t prot, unsigned long address)
{
	pgprot_t forbidden = __pgprot(0);

I
Ingo Molnar 已提交
152
	/*
153 154
	 * 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
155
	 */
156 157 158 159 160 161 162 163 164
	if (within(__pa(address), BIOS_BEGIN, BIOS_END))
		pgprot_val(forbidden) |= _PAGE_NX;

	/*
	 * The kernel text needs to be executable for obvious reasons
	 * Does not cover __inittext since that is gone later on
	 */
	if (within(address, (unsigned long)_text, (unsigned long)_etext))
		pgprot_val(forbidden) |= _PAGE_NX;
165 166 167 168 169 170
	/*
	 * Do the same for the x86-64 high kernel mapping
	 */
	if (within(address, virt_to_highmap(_text), virt_to_highmap(_etext)))
		pgprot_val(forbidden) |= _PAGE_NX;

171 172 173 174
	/* The .rodata section needs to be read-only */
	if (within(address, (unsigned long)__start_rodata,
				(unsigned long)__end_rodata))
		pgprot_val(forbidden) |= _PAGE_RW;
175 176 177 178 179 180
	/*
	 * Do the same for the x86-64 high kernel mapping
	 */
	if (within(address, virt_to_highmap(__start_rodata),
				virt_to_highmap(__end_rodata)))
		pgprot_val(forbidden) |= _PAGE_RW;
181 182

	prot = __pgprot(pgprot_val(prot) & ~pgprot_val(forbidden));
I
Ingo Molnar 已提交
183 184 185 186

	return prot;
}

T
Thomas Gleixner 已提交
187 188 189 190 191 192 193 194
/*
 * 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.
 */
195
pte_t *lookup_address(unsigned long address, unsigned int *level)
196
{
L
Linus Torvalds 已提交
197 198 199
	pgd_t *pgd = pgd_offset_k(address);
	pud_t *pud;
	pmd_t *pmd;
200

T
Thomas Gleixner 已提交
201 202
	*level = PG_LEVEL_NONE;

L
Linus Torvalds 已提交
203 204
	if (pgd_none(*pgd))
		return NULL;
I
Ingo Molnar 已提交
205

L
Linus Torvalds 已提交
206 207 208
	pud = pud_offset(pgd, address);
	if (pud_none(*pud))
		return NULL;
209 210 211 212 213

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

L
Linus Torvalds 已提交
214 215 216
	pmd = pmd_offset(pud, address);
	if (pmd_none(*pmd))
		return NULL;
T
Thomas Gleixner 已提交
217 218

	*level = PG_LEVEL_2M;
T
Thomas Gleixner 已提交
219
	if (pmd_large(*pmd) || !pmd_present(*pmd))
L
Linus Torvalds 已提交
220 221
		return (pte_t *)pmd;

T
Thomas Gleixner 已提交
222
	*level = PG_LEVEL_4K;
I
Ingo Molnar 已提交
223

224 225 226
	return pte_offset_kernel(pmd, address);
}

I
Ingo Molnar 已提交
227 228 229
/*
 * Set the new pmd in all the pgds we know about:
 */
I
Ingo Molnar 已提交
230
static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
231 232 233
{
	/* change init_mm */
	set_pte_atomic(kpte, pte);
234
#ifdef CONFIG_X86_32
235
	if (!SHARED_KERNEL_PMD) {
236 237
		struct page *page;

238
		list_for_each_entry(page, &pgd_list, lru) {
239 240 241 242 243 244 245 246 247
			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
248
	}
249
#endif
L
Linus Torvalds 已提交
250 251
}

I
Ingo Molnar 已提交
252 253 254
static int
try_preserve_large_page(pte_t *kpte, unsigned long address,
			struct cpa_data *cpa)
255 256 257 258
{
	unsigned long nextpage_addr, numpages, pmask, psize, flags;
	pte_t new_pte, old_pte, *tmp;
	pgprot_t old_prot, new_prot;
259 260
	int do_split = 1;
	unsigned int level;
261 262 263 264 265 266 267 268 269 270 271 272

	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:
273 274
		psize = PMD_PAGE_SIZE;
		pmask = PMD_PAGE_MASK;
275
		break;
276
#ifdef CONFIG_X86_64
277
	case PG_LEVEL_1G:
278 279 280 281
		psize = PMD_PAGE_SIZE;
		pmask = PMD_PAGE_MASK;
		break;
#endif
282
	default:
I
Ingo Molnar 已提交
283
		do_split = -EINVAL;
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
		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;
	if (numpages < cpa->numpages)
		cpa->numpages = numpages;

	/*
	 * 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);
	new_prot = static_protections(new_prot, address);

	/*
	 * 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
311
		do_split = 0;
312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
		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.
	 */
	if (address == (nextpage_addr - psize) && cpa->numpages == numpages) {
		/*
		 * 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
331
		do_split = 0;
332 333 334 335
	}

out_unlock:
	spin_unlock_irqrestore(&pgd_lock, flags);
I
Ingo Molnar 已提交
336

I
Ingo Molnar 已提交
337
	return do_split;
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
static LIST_HEAD(page_pool);
static unsigned long pool_size, pool_pages, pool_low;
static unsigned long pool_used, pool_failed, pool_refill;

static void cpa_fill_pool(void)
{
	struct page *p;
	gfp_t gfp = GFP_KERNEL;

	/* Do not allocate from interrupt context */
	if (in_irq() || irqs_disabled())
		return;
	/*
	 * Check unlocked. I does not matter when we have one more
	 * page in the pool. The bit lock avoids recursive pool
	 * allocations:
	 */
	if (pool_pages >= pool_size || test_and_set_bit_lock(0, &pool_refill))
		return;

#ifdef CONFIG_DEBUG_PAGEALLOC
	/*
	 * We could do:
	 * gfp = in_atomic() ? GFP_ATOMIC : GFP_KERNEL;
	 * but this fails on !PREEMPT kernels
	 */
	gfp =  GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
#endif

	while (pool_pages < pool_size) {
		p = alloc_pages(gfp, 0);
		if (!p) {
			pool_failed++;
			break;
		}
		spin_lock_irq(&pgd_lock);
		list_add(&p->lru, &page_pool);
		pool_pages++;
		spin_unlock_irq(&pgd_lock);
	}
	clear_bit_unlock(0, &pool_refill);
}

#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:
	 */
	gb = ((si.totalram >> SHIFT_MB) + ROUND_MB_GB) >> SHIFT_MB_GB;
	pool_size = POOL_PAGES_PER_GB * gb;
	pool_low = pool_size;

	cpa_fill_pool();
	printk(KERN_DEBUG
	       "CPA: page pool initialized %lu of %lu pages preallocated\n",
	       pool_pages, pool_size);
}

411
static int split_large_page(pte_t *kpte, unsigned long address)
412
{
T
Thomas Gleixner 已提交
413
	unsigned long flags, pfn, pfninc = 1;
I
Ingo Molnar 已提交
414 415
	gfp_t gfp_flags = GFP_KERNEL;
	unsigned int i, level;
416
	pte_t *pbase, *tmp;
I
Ingo Molnar 已提交
417
	pgprot_t ref_prot;
418 419
	struct page *base;

420
#ifdef CONFIG_DEBUG_PAGEALLOC
I
Ingo Molnar 已提交
421
	gfp_flags = GFP_ATOMIC | __GFP_NOWARN;
422 423
#endif
	base = alloc_pages(gfp_flags, 0);
424 425 426
	if (!base)
		return -ENOMEM;

I
Ingo Molnar 已提交
427
	spin_lock_irqsave(&pgd_lock, flags);
428 429 430 431 432
	/*
	 * 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
433
	if (tmp != kpte)
434 435 436
		goto out_unlock;

	pbase = (pte_t *)page_address(base);
437
#ifdef CONFIG_X86_32
438
	paravirt_alloc_pt(&init_mm, page_to_pfn(base));
439
#endif
T
Thomas Gleixner 已提交
440
	ref_prot = pte_pgprot(pte_clrhuge(*kpte));
441

442 443 444 445 446 447 448
#ifdef CONFIG_X86_64
	if (level == PG_LEVEL_1G) {
		pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
		pgprot_val(ref_prot) |= _PAGE_PSE;
	}
#endif

449 450 451 452
	/*
	 * Get the target pfn from the original entry:
	 */
	pfn = pte_pfn(*kpte);
453
	for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc)
454
		set_pte(&pbase[i], pfn_pte(pfn, ref_prot));
455 456

	/*
T
Thomas Gleixner 已提交
457
	 * Install the new, split up pagetable. Important details here:
458 459 460 461
	 *
	 * 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
462 463 464
	 *
	 * Mark the entry present. The current mapping might be
	 * set to not present, which we preserved above.
465
	 */
466
	ref_prot = pte_pgprot(pte_mkexec(pte_clrhuge(*kpte)));
T
Thomas Gleixner 已提交
467
	pgprot_val(ref_prot) |= _PAGE_PRESENT;
I
Ingo Molnar 已提交
468
	__set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
469 470 471
	base = NULL;

out_unlock:
I
Ingo Molnar 已提交
472
	spin_unlock_irqrestore(&pgd_lock, flags);
473 474 475 476 477 478 479

	if (base)
		__free_pages(base, 0);

	return 0;
}

T
Thomas Gleixner 已提交
480
static int __change_page_attr(unsigned long address, struct cpa_data *cpa)
481
{
482 483
	int do_split, err;
	unsigned int level;
L
Linus Torvalds 已提交
484
	struct page *kpte_page;
485
	pte_t *kpte;
L
Linus Torvalds 已提交
486

487
repeat:
488
	kpte = lookup_address(address, &level);
L
Linus Torvalds 已提交
489 490
	if (!kpte)
		return -EINVAL;
491

L
Linus Torvalds 已提交
492
	kpte_page = virt_to_page(kpte);
493 494 495
	BUG_ON(PageLRU(kpte_page));
	BUG_ON(PageCompound(kpte_page));

T
Thomas Gleixner 已提交
496
	if (level == PG_LEVEL_4K) {
I
Ingo Molnar 已提交
497
		pte_t new_pte, old_pte = *kpte;
498 499 500
		pgprot_t new_prot = pte_pgprot(old_pte);

		if(!pte_val(old_pte)) {
T
Thomas Gleixner 已提交
501 502 503 504
			printk(KERN_WARNING "CPA: called for zero pte. "
			       "vaddr = %lx cpa->vaddr = %lx\n", address,
				cpa->vaddr);
			WARN_ON(1);
505 506
			return -EINVAL;
		}
I
Ingo Molnar 已提交
507

T
Thomas Gleixner 已提交
508 509
		pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
		pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
I
Ingo Molnar 已提交
510 511 512

		new_prot = static_protections(new_prot, address);

513 514 515 516 517 518
		/*
		 * 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
		 */
		new_pte = pfn_pte(pte_pfn(old_pte), canon_pgprot(new_prot));
519 520 521 522 523 524 525 526

		/*
		 * Do we really change anything ?
		 */
		if (pte_val(old_pte) != pte_val(new_pte)) {
			set_pte_atomic(kpte, new_pte);
			cpa->flushtlb = 1;
		}
527 528
		cpa->numpages = 1;
		return 0;
L
Linus Torvalds 已提交
529
	}
530 531 532 533 534

	/*
	 * Check, whether we can keep the large page intact
	 * and just change the pte:
	 */
I
Ingo Molnar 已提交
535
	do_split = try_preserve_large_page(kpte, address, cpa);
536 537 538 539 540
	/*
	 * When the range fits into the existing large page,
	 * return. cp->numpages and cpa->tlbflush have been updated in
	 * try_large_page:
	 */
I
Ingo Molnar 已提交
541 542
	if (do_split <= 0)
		return do_split;
543 544 545 546

	/*
	 * We have to split the large page:
	 */
I
Ingo Molnar 已提交
547 548 549 550 551
	err = split_large_page(kpte, address);
	if (!err) {
		cpa->flushtlb = 1;
		goto repeat;
	}
I
Ingo Molnar 已提交
552

I
Ingo Molnar 已提交
553
	return err;
554
}
L
Linus Torvalds 已提交
555

556 557 558 559
/**
 * change_page_attr_addr - Change page table attributes in linear mapping
 * @address: Virtual address in linear mapping.
 * @prot:    New page table attribute (PAGE_*)
L
Linus Torvalds 已提交
560
 *
561 562 563
 * Change page attributes of a page in the direct mapping. This is a variant
 * of change_page_attr() that also works on memory holes that do not have
 * mem_map entry (pfn_valid() is false).
564
 *
565
 * See change_page_attr() documentation for more details.
566 567
 *
 * Modules and drivers should use the set_memory_* APIs instead.
L
Linus Torvalds 已提交
568
 */
T
Thomas Gleixner 已提交
569
static int change_page_attr_addr(struct cpa_data *cpa)
L
Linus Torvalds 已提交
570
{
571
	int err;
T
Thomas Gleixner 已提交
572
	unsigned long address = cpa->vaddr;
573 574

#ifdef CONFIG_X86_64
575 576
	unsigned long phys_addr = __pa(address);

577 578 579 580 581 582 583
	/*
	 * If we are inside the high mapped kernel range, then we
	 * fixup the low mapping first. __va() returns the virtual
	 * address in the linear mapping:
	 */
	if (within(address, HIGH_MAP_START, HIGH_MAP_END))
		address = (unsigned long) __va(phys_addr);
584 585
#endif

T
Thomas Gleixner 已提交
586
	err = __change_page_attr(address, cpa);
587 588
	if (err)
		return err;
589 590

#ifdef CONFIG_X86_64
A
Arjan van de Ven 已提交
591
	/*
592 593
	 * 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
594
	 */
595 596 597 598
	if (within(phys_addr, 0, KERNEL_TEXT_SIZE)) {
		/*
		 * Calc the high mapping address. See __phys_addr()
		 * for the non obvious details.
599 600 601
		 *
		 * Note that NX and other required permissions are
		 * checked in static_protections().
602 603 604
		 */
		address = phys_addr + HIGH_MAP_START - phys_base;

I
Ingo Molnar 已提交
605
		/*
606 607 608
		 * Our high aliases are imprecise, because we check
		 * everything between 0 and KERNEL_TEXT_SIZE, so do
		 * not propagate lookup failures back to users:
I
Ingo Molnar 已提交
609
		 */
T
Thomas Gleixner 已提交
610
		__change_page_attr(address, cpa);
611
	}
A
Arjan van de Ven 已提交
612
#endif
L
Linus Torvalds 已提交
613 614 615
	return err;
}

T
Thomas Gleixner 已提交
616
static int __change_page_attr_set_clr(struct cpa_data *cpa)
617
{
618
	int ret, numpages = cpa->numpages;
619

620 621 622 623 624 625
	while (numpages) {
		/*
		 * Store the remaining nr of pages for the large page
		 * preservation check.
		 */
		cpa->numpages = numpages;
T
Thomas Gleixner 已提交
626
		ret = change_page_attr_addr(cpa);
627 628 629
		if (ret)
			return ret;

630 631 632 633 634 635 636 637 638
		/*
		 * 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.
		 */
		BUG_ON(cpa->numpages > numpages);
		numpages -= cpa->numpages;
		cpa->vaddr += cpa->numpages * PAGE_SIZE;
	}
639 640 641
	return 0;
}

642 643 644 645 646 647
static inline int cache_attr(pgprot_t attr)
{
	return pgprot_val(attr) &
		(_PAGE_PAT | _PAGE_PAT_LARGE | _PAGE_PWT | _PAGE_PCD);
}

648 649 650
static int change_page_attr_set_clr(unsigned long addr, int numpages,
				    pgprot_t mask_set, pgprot_t mask_clr)
{
T
Thomas Gleixner 已提交
651
	struct cpa_data cpa;
652
	int ret, cache;
653 654 655 656 657 658 659 660 661 662

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

T
Thomas Gleixner 已提交
663 664 665 666
	cpa.vaddr = addr;
	cpa.numpages = numpages;
	cpa.mask_set = mask_set;
	cpa.mask_clr = mask_clr;
667
	cpa.flushtlb = 0;
T
Thomas Gleixner 已提交
668 669

	ret = __change_page_attr_set_clr(&cpa);
670

671 672 673 674
	/*
	 * Check whether we really changed something:
	 */
	if (!cpa.flushtlb)
675
		goto out;
676

677 678 679 680 681 682
	/*
	 * No need to flush, when we did not set any of the caching
	 * attributes:
	 */
	cache = cache_attr(mask_set);

683 684 685
	/*
	 * On success we use clflush, when the CPU supports it to
	 * avoid the wbindv. If the CPU does not support it and in the
686
	 * error case we fall back to cpa_flush_all (which uses
687 688 689
	 * wbindv):
	 */
	if (!ret && cpu_has_clflush)
690
		cpa_flush_range(addr, numpages, cache);
691
	else
692
		cpa_flush_all(cache);
693

694 695
out:
	cpa_fill_pool();
696 697 698
	return ret;
}

699 700
static inline int change_page_attr_set(unsigned long addr, int numpages,
				       pgprot_t mask)
701
{
702
	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0));
703 704
}

705 706
static inline int change_page_attr_clear(unsigned long addr, int numpages,
					 pgprot_t mask)
707
{
H
Huang, Ying 已提交
708
	return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask);
709 710 711 712 713 714
}

int set_memory_uc(unsigned long addr, int numpages)
{
	return change_page_attr_set(addr, numpages,
				    __pgprot(_PAGE_PCD | _PAGE_PWT));
715 716 717 718 719
}
EXPORT_SYMBOL(set_memory_uc);

int set_memory_wb(unsigned long addr, int numpages)
{
720 721
	return change_page_attr_clear(addr, numpages,
				      __pgprot(_PAGE_PCD | _PAGE_PWT));
722 723 724 725 726
}
EXPORT_SYMBOL(set_memory_wb);

int set_memory_x(unsigned long addr, int numpages)
{
727
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_NX));
728 729 730 731 732
}
EXPORT_SYMBOL(set_memory_x);

int set_memory_nx(unsigned long addr, int numpages)
{
733
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_NX));
734 735 736 737 738
}
EXPORT_SYMBOL(set_memory_nx);

int set_memory_ro(unsigned long addr, int numpages)
{
739
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_RW));
740 741 742 743
}

int set_memory_rw(unsigned long addr, int numpages)
{
744
	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_RW));
745
}
I
Ingo Molnar 已提交
746 747 748

int set_memory_np(unsigned long addr, int numpages)
{
749
	return change_page_attr_clear(addr, numpages, __pgprot(_PAGE_PRESENT));
I
Ingo Molnar 已提交
750
}
751 752 753 754 755

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

T
Thomas Gleixner 已提交
756
	return set_memory_uc(addr, numpages);
757 758 759 760 761 762 763
}
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
764
	return set_memory_wb(addr, numpages);
765 766 767 768 769 770 771
}
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
772
	return set_memory_x(addr, numpages);
773 774 775 776 777 778 779
}
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
780
	return set_memory_nx(addr, numpages);
781 782 783 784 785 786 787
}
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
788
	return set_memory_ro(addr, numpages);
789 790 791 792 793
}

int set_pages_rw(struct page *page, int numpages)
{
	unsigned long addr = (unsigned long)page_address(page);
794

T
Thomas Gleixner 已提交
795
	return set_memory_rw(addr, numpages);
I
Ingo Molnar 已提交
796 797
}

L
Linus Torvalds 已提交
798
#ifdef CONFIG_DEBUG_PAGEALLOC
I
Ingo Molnar 已提交
799 800 801

static int __set_pages_p(struct page *page, int numpages)
{
T
Thomas Gleixner 已提交
802 803 804 805
	struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page),
				.numpages = numpages,
				.mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
				.mask_clr = __pgprot(0)};
806

T
Thomas Gleixner 已提交
807
	return __change_page_attr_set_clr(&cpa);
I
Ingo Molnar 已提交
808 809 810 811
}

static int __set_pages_np(struct page *page, int numpages)
{
T
Thomas Gleixner 已提交
812 813 814 815
	struct cpa_data cpa = { .vaddr = (unsigned long) page_address(page),
				.numpages = numpages,
				.mask_set = __pgprot(0),
				.mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW)};
816

T
Thomas Gleixner 已提交
817
	return __change_page_attr_set_clr(&cpa);
I
Ingo Molnar 已提交
818 819
}

L
Linus Torvalds 已提交
820 821 822 823
void kernel_map_pages(struct page *page, int numpages, int enable)
{
	if (PageHighMem(page))
		return;
824
	if (!enable) {
825 826
		debug_check_no_locks_freed(page_address(page),
					   numpages * PAGE_SIZE);
827
	}
828

829 830 831 832 833 834
	/*
	 * If page allocator is not up yet then do not call c_p_a():
	 */
	if (!debug_pagealloc_enabled)
		return;

835
	/*
836 837
	 * The return value is ignored - the calls cannot fail,
	 * large pages are disabled at boot time:
L
Linus Torvalds 已提交
838
	 */
I
Ingo Molnar 已提交
839 840 841 842
	if (enable)
		__set_pages_p(page, numpages);
	else
		__set_pages_np(page, numpages);
843 844

	/*
845 846
	 * We should perform an IPI and flush all tlbs,
	 * but that can deadlock->flush only current cpu:
L
Linus Torvalds 已提交
847 848
	 */
	__flush_tlb_all();
849 850 851 852 853 854

	/*
	 * Try to refill the page pool here. We can do this only after
	 * the tlb flush.
	 */
	cpa_fill_pool();
L
Linus Torvalds 已提交
855 856
}
#endif
857 858 859 860 861 862 863 864

/*
 * 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