1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright 2002 Andi Kleen, SuSE Labs.
   4 * Thanks to Ben LaHaise for precious feedback.
   5 */
   6#include <linux/highmem.h>
   7#include <linux/memblock.h>
   8#include <linux/sched.h>
   9#include <linux/mm.h>
  10#include <linux/interrupt.h>
  11#include <linux/seq_file.h>
  12#include <linux/debugfs.h>
  13#include <linux/pfn.h>
  14#include <linux/percpu.h>
  15#include <linux/gfp.h>
  16#include <linux/pci.h>
  17#include <linux/vmalloc.h>
  18
  19#include <asm/e820/api.h>
  20#include <asm/processor.h>
  21#include <asm/tlbflush.h>
  22#include <asm/sections.h>
  23#include <asm/setup.h>
  24#include <linux/uaccess.h>
  25#include <asm/pgalloc.h>
  26#include <asm/proto.h>
  27#include <asm/pat.h>
  28#include <asm/set_memory.h>
  29
  30#include "mm_internal.h"
  31
  32/*
  33 * The current flushing context - we pass it instead of 5 arguments:
  34 */
  35struct cpa_data {
  36	unsigned long	*vaddr;
  37	pgd_t		*pgd;
  38	pgprot_t	mask_set;
  39	pgprot_t	mask_clr;
  40	unsigned long	numpages;
  41	unsigned long	curpage;
  42	unsigned long	pfn;
  43	unsigned int	flags;
  44	unsigned int	force_split		: 1,
  45			force_static_prot	: 1;
  46	struct page	**pages;
  47};
  48
  49enum cpa_warn {
  50	CPA_CONFLICT,
  51	CPA_PROTECT,
  52	CPA_DETECT,
  53};
  54
  55static const int cpa_warn_level = CPA_PROTECT;
  56
  57/*
  58 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
  59 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
  60 * entries change the page attribute in parallel to some other cpu
  61 * splitting a large page entry along with changing the attribute.
  62 */
  63static DEFINE_SPINLOCK(cpa_lock);
  64
  65#define CPA_FLUSHTLB 1
  66#define CPA_ARRAY 2
  67#define CPA_PAGES_ARRAY 4
  68#define CPA_NO_CHECK_ALIAS 8 /* Do not search for aliases */
  69
  70#ifdef CONFIG_PROC_FS
  71static unsigned long direct_pages_count[PG_LEVEL_NUM];
  72
  73void update_page_count(int level, unsigned long pages)
  74{
  75	/* Protect against CPA */
  76	spin_lock(&pgd_lock);
  77	direct_pages_count[level] += pages;
  78	spin_unlock(&pgd_lock);
  79}
  80
  81static void split_page_count(int level)
  82{
  83	if (direct_pages_count[level] == 0)
  84		return;
  85
  86	direct_pages_count[level]--;
  87	direct_pages_count[level - 1] += PTRS_PER_PTE;
  88}
  89
  90void arch_report_meminfo(struct seq_file *m)
  91{
  92	seq_printf(m, "DirectMap4k:    %8lu kB\n",
  93			direct_pages_count[PG_LEVEL_4K] << 2);
  94#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
  95	seq_printf(m, "DirectMap2M:    %8lu kB\n",
  96			direct_pages_count[PG_LEVEL_2M] << 11);
  97#else
  98	seq_printf(m, "DirectMap4M:    %8lu kB\n",
  99			direct_pages_count[PG_LEVEL_2M] << 12);
 100#endif
 101	if (direct_gbpages)
 102		seq_printf(m, "DirectMap1G:    %8lu kB\n",
 103			direct_pages_count[PG_LEVEL_1G] << 20);
 104}
 105#else
 106static inline void split_page_count(int level) { }
 107#endif
 108
 109#ifdef CONFIG_X86_CPA_STATISTICS
 110
 111static unsigned long cpa_1g_checked;
 112static unsigned long cpa_1g_sameprot;
 113static unsigned long cpa_1g_preserved;
 114static unsigned long cpa_2m_checked;
 115static unsigned long cpa_2m_sameprot;
 116static unsigned long cpa_2m_preserved;
 117static unsigned long cpa_4k_install;
 118
 119static inline void cpa_inc_1g_checked(void)
 120{
 121	cpa_1g_checked++;
 122}
 123
 124static inline void cpa_inc_2m_checked(void)
 125{
 126	cpa_2m_checked++;
 127}
 128
 129static inline void cpa_inc_4k_install(void)
 130{
 131	cpa_4k_install++;
 132}
 133
 134static inline void cpa_inc_lp_sameprot(int level)
 135{
 136	if (level == PG_LEVEL_1G)
 137		cpa_1g_sameprot++;
 138	else
 139		cpa_2m_sameprot++;
 140}
 141
 142static inline void cpa_inc_lp_preserved(int level)
 143{
 144	if (level == PG_LEVEL_1G)
 145		cpa_1g_preserved++;
 146	else
 147		cpa_2m_preserved++;
 148}
 149
 150static int cpastats_show(struct seq_file *m, void *p)
 151{
 152	seq_printf(m, "1G pages checked:     %16lu\n", cpa_1g_checked);
 153	seq_printf(m, "1G pages sameprot:    %16lu\n", cpa_1g_sameprot);
 154	seq_printf(m, "1G pages preserved:   %16lu\n", cpa_1g_preserved);
 155	seq_printf(m, "2M pages checked:     %16lu\n", cpa_2m_checked);
 156	seq_printf(m, "2M pages sameprot:    %16lu\n", cpa_2m_sameprot);
 157	seq_printf(m, "2M pages preserved:   %16lu\n", cpa_2m_preserved);
 158	seq_printf(m, "4K pages set-checked: %16lu\n", cpa_4k_install);
 159	return 0;
 160}
 161
 162static int cpastats_open(struct inode *inode, struct file *file)
 163{
 164	return single_open(file, cpastats_show, NULL);
 165}
 166
 167static const struct file_operations cpastats_fops = {
 168	.open		= cpastats_open,
 169	.read		= seq_read,
 170	.llseek		= seq_lseek,
 171	.release	= single_release,
 172};
 173
 174static int __init cpa_stats_init(void)
 175{
 176	debugfs_create_file("cpa_stats", S_IRUSR, arch_debugfs_dir, NULL,
 177			    &cpastats_fops);
 178	return 0;
 179}
 180late_initcall(cpa_stats_init);
 181#else
 182static inline void cpa_inc_1g_checked(void) { }
 183static inline void cpa_inc_2m_checked(void) { }
 184static inline void cpa_inc_4k_install(void) { }
 185static inline void cpa_inc_lp_sameprot(int level) { }
 186static inline void cpa_inc_lp_preserved(int level) { }
 187#endif
 188
 189
 190static inline int
 191within(unsigned long addr, unsigned long start, unsigned long end)
 192{
 193	return addr >= start && addr < end;
 194}
 195
 196static inline int
 197within_inclusive(unsigned long addr, unsigned long start, unsigned long end)
 198{
 199	return addr >= start && addr <= end;
 200}
 201
 202#ifdef CONFIG_X86_64
 203
 204static inline unsigned long highmap_start_pfn(void)
 205{
 206	return __pa_symbol(_text) >> PAGE_SHIFT;
 207}
 208
 209static inline unsigned long highmap_end_pfn(void)
 210{
 211	/* Do not reference physical address outside the kernel. */
 212	return __pa_symbol(roundup(_brk_end, PMD_SIZE) - 1) >> PAGE_SHIFT;
 213}
 214
 215static bool __cpa_pfn_in_highmap(unsigned long pfn)
 216{
 217	/*
 218	 * Kernel text has an alias mapping at a high address, known
 219	 * here as "highmap".
 220	 */
 221	return within_inclusive(pfn, highmap_start_pfn(), highmap_end_pfn());
 222}
 223
 224#else
 225
 226static bool __cpa_pfn_in_highmap(unsigned long pfn)
 227{
 228	/* There is no highmap on 32-bit */
 229	return false;
 230}
 231
 232#endif
 233
 234/*
 235 * See set_mce_nospec().
 236 *
 237 * Machine check recovery code needs to change cache mode of poisoned pages to
 238 * UC to avoid speculative access logging another error. But passing the
 239 * address of the 1:1 mapping to set_memory_uc() is a fine way to encourage a
 240 * speculative access. So we cheat and flip the top bit of the address. This
 241 * works fine for the code that updates the page tables. But at the end of the
 242 * process we need to flush the TLB and cache and the non-canonical address
 243 * causes a #GP fault when used by the INVLPG and CLFLUSH instructions.
 244 *
 245 * But in the common case we already have a canonical address. This code
 246 * will fix the top bit if needed and is a no-op otherwise.
 247 */
 248static inline unsigned long fix_addr(unsigned long addr)
 249{
 250#ifdef CONFIG_X86_64
 251	return (long)(addr << 1) >> 1;
 252#else
 253	return addr;
 254#endif
 255}
 256
 257static unsigned long __cpa_addr(struct cpa_data *cpa, unsigned long idx)
 258{
 259	if (cpa->flags & CPA_PAGES_ARRAY) {
 260		struct page *page = cpa->pages[idx];
 261
 262		if (unlikely(PageHighMem(page)))
 263			return 0;
 264
 265		return (unsigned long)page_address(page);
 266	}
 267
 268	if (cpa->flags & CPA_ARRAY)
 269		return cpa->vaddr[idx];
 270
 271	return *cpa->vaddr + idx * PAGE_SIZE;
 272}
 273
 274/*
 275 * Flushing functions
 276 */
 277
 278static void clflush_cache_range_opt(void *vaddr, unsigned int size)
 
 
 
 
 
 
 
 
 279{
 280	const unsigned long clflush_size = boot_cpu_data.x86_clflush_size;
 281	void *p = (void *)((unsigned long)vaddr & ~(clflush_size - 1));
 282	void *vend = vaddr + size;
 283
 284	if (p >= vend)
 285		return;
 286
 
 
 287	for (; p < vend; p += clflush_size)
 288		clflushopt(p);
 289}
 290
 291/**
 292 * clflush_cache_range - flush a cache range with clflush
 293 * @vaddr:	virtual start address
 294 * @size:	number of bytes to flush
 295 *
 296 * CLFLUSHOPT is an unordered instruction which needs fencing with MFENCE or
 297 * SFENCE to avoid ordering issues.
 298 */
 299void clflush_cache_range(void *vaddr, unsigned int size)
 300{
 301	mb();
 302	clflush_cache_range_opt(vaddr, size);
 303	mb();
 304}
 305EXPORT_SYMBOL_GPL(clflush_cache_range);
 306
 307void arch_invalidate_pmem(void *addr, size_t size)
 308{
 309	clflush_cache_range(addr, size);
 310}
 311EXPORT_SYMBOL_GPL(arch_invalidate_pmem);
 312
 313static void __cpa_flush_all(void *arg)
 314{
 315	unsigned long cache = (unsigned long)arg;
 316
 317	/*
 318	 * Flush all to work around Errata in early athlons regarding
 319	 * large page flushing.
 320	 */
 321	__flush_tlb_all();
 322
 323	if (cache && boot_cpu_data.x86 >= 4)
 324		wbinvd();
 325}
 326
 327static void cpa_flush_all(unsigned long cache)
 328{
 329	BUG_ON(irqs_disabled() && !early_boot_irqs_disabled);
 330
 331	on_each_cpu(__cpa_flush_all, (void *) cache, 1);
 332}
 333
 334void __cpa_flush_tlb(void *data)
 335{
 336	struct cpa_data *cpa = data;
 337	unsigned int i;
 338
 339	for (i = 0; i < cpa->numpages; i++)
 340		__flush_tlb_one_kernel(fix_addr(__cpa_addr(cpa, i)));
 
 341}
 342
 343static void cpa_flush(struct cpa_data *data, int cache)
 344{
 345	struct cpa_data *cpa = data;
 346	unsigned int i;
 347
 348	BUG_ON(irqs_disabled() && !early_boot_irqs_disabled);
 
 349
 350	if (cache && !static_cpu_has(X86_FEATURE_CLFLUSH)) {
 351		cpa_flush_all(cache);
 352		return;
 353	}
 354
 355	if (cpa->numpages <= tlb_single_page_flush_ceiling)
 356		on_each_cpu(__cpa_flush_tlb, cpa, 1);
 357	else
 358		flush_tlb_all();
 359
 360	if (!cache)
 361		return;
 362
 363	mb();
 364	for (i = 0; i < cpa->numpages; i++) {
 365		unsigned long addr = __cpa_addr(cpa, i);
 366		unsigned int level;
 367
 
 
 368		pte_t *pte = lookup_address(addr, &level);
 369
 370		/*
 371		 * Only flush present addresses:
 372		 */
 373		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
 374			clflush_cache_range_opt((void *)fix_addr(addr), PAGE_SIZE);
 375	}
 376	mb();
 377}
 378
 379static bool overlaps(unsigned long r1_start, unsigned long r1_end,
 380		     unsigned long r2_start, unsigned long r2_end)
 381{
 382	return (r1_start <= r2_end && r1_end >= r2_start) ||
 383		(r2_start <= r1_end && r2_end >= r1_start);
 384}
 385
 386#ifdef CONFIG_PCI_BIOS
 387/*
 388 * The BIOS area between 640k and 1Mb needs to be executable for PCI BIOS
 389 * based config access (CONFIG_PCI_GOBIOS) support.
 390 */
 391#define BIOS_PFN	PFN_DOWN(BIOS_BEGIN)
 392#define BIOS_PFN_END	PFN_DOWN(BIOS_END - 1)
 393
 394static pgprotval_t protect_pci_bios(unsigned long spfn, unsigned long epfn)
 395{
 396	if (pcibios_enabled && overlaps(spfn, epfn, BIOS_PFN, BIOS_PFN_END))
 397		return _PAGE_NX;
 398	return 0;
 399}
 400#else
 401static pgprotval_t protect_pci_bios(unsigned long spfn, unsigned long epfn)
 402{
 403	return 0;
 404}
 405#endif
 406
 407/*
 408 * The .rodata section needs to be read-only. Using the pfn catches all
 409 * aliases.  This also includes __ro_after_init, so do not enforce until
 410 * kernel_set_to_readonly is true.
 411 */
 412static pgprotval_t protect_rodata(unsigned long spfn, unsigned long epfn)
 413{
 414	unsigned long epfn_ro, spfn_ro = PFN_DOWN(__pa_symbol(__start_rodata));
 415
 416	/*
 417	 * Note: __end_rodata is at page aligned and not inclusive, so
 418	 * subtract 1 to get the last enforced PFN in the rodata area.
 
 
 
 
 
 419	 */
 420	epfn_ro = PFN_DOWN(__pa_symbol(__end_rodata)) - 1;
 
 
 
 421
 422	if (kernel_set_to_readonly && overlaps(spfn, epfn, spfn_ro, epfn_ro))
 423		return _PAGE_RW;
 424	return 0;
 425}
 426
 427/*
 428 * Protect kernel text against becoming non executable by forbidding
 429 * _PAGE_NX.  This protects only the high kernel mapping (_text -> _etext)
 430 * out of which the kernel actually executes.  Do not protect the low
 431 * mapping.
 432 *
 433 * This does not cover __inittext since that is gone after boot.
 434 */
 435static pgprotval_t protect_kernel_text(unsigned long start, unsigned long end)
 436{
 437	unsigned long t_end = (unsigned long)_etext - 1;
 438	unsigned long t_start = (unsigned long)_text;
 439
 440	if (overlaps(start, end, t_start, t_end))
 441		return _PAGE_NX;
 442	return 0;
 443}
 444
 445#if defined(CONFIG_X86_64)
 446/*
 447 * Once the kernel maps the text as RO (kernel_set_to_readonly is set),
 448 * kernel text mappings for the large page aligned text, rodata sections
 449 * will be always read-only. For the kernel identity mappings covering the
 450 * holes caused by this alignment can be anything that user asks.
 451 *
 452 * This will preserve the large page mappings for kernel text/data at no
 453 * extra cost.
 454 */
 455static pgprotval_t protect_kernel_text_ro(unsigned long start,
 456					  unsigned long end)
 457{
 458	unsigned long t_end = (unsigned long)__end_rodata_hpage_align - 1;
 459	unsigned long t_start = (unsigned long)_text;
 460	unsigned int level;
 461
 462	if (!kernel_set_to_readonly || !overlaps(start, end, t_start, t_end))
 463		return 0;
 464	/*
 465	 * Don't enforce the !RW mapping for the kernel text mapping, if
 466	 * the current mapping is already using small page mapping.  No
 467	 * need to work hard to preserve large page mappings in this case.
 468	 *
 469	 * This also fixes the Linux Xen paravirt guest boot failure caused
 470	 * by unexpected read-only mappings for kernel identity
 471	 * mappings. In this paravirt guest case, the kernel text mapping
 472	 * and the kernel identity mapping share the same page-table pages,
 473	 * so the protections for kernel text and identity mappings have to
 474	 * be the same.
 475	 */
 476	if (lookup_address(start, &level) && (level != PG_LEVEL_4K))
 477		return _PAGE_RW;
 478	return 0;
 479}
 480#else
 481static pgprotval_t protect_kernel_text_ro(unsigned long start,
 482					  unsigned long end)
 483{
 484	return 0;
 485}
 486#endif
 487
 488static inline bool conflicts(pgprot_t prot, pgprotval_t val)
 489{
 490	return (pgprot_val(prot) & ~val) != pgprot_val(prot);
 491}
 492
 493static inline void check_conflict(int warnlvl, pgprot_t prot, pgprotval_t val,
 494				  unsigned long start, unsigned long end,
 495				  unsigned long pfn, const char *txt)
 496{
 497	static const char *lvltxt[] = {
 498		[CPA_CONFLICT]	= "conflict",
 499		[CPA_PROTECT]	= "protect",
 500		[CPA_DETECT]	= "detect",
 501	};
 502
 503	if (warnlvl > cpa_warn_level || !conflicts(prot, val))
 504		return;
 505
 506	pr_warn("CPA %8s %10s: 0x%016lx - 0x%016lx PFN %lx req %016llx prevent %016llx\n",
 507		lvltxt[warnlvl], txt, start, end, pfn, (unsigned long long)pgprot_val(prot),
 508		(unsigned long long)val);
 
 
 
 509}
 510
 511/*
 512 * Certain areas of memory on x86 require very specific protection flags,
 513 * for example the BIOS area or kernel text. Callers don't always get this
 514 * right (again, ioremap() on BIOS memory is not uncommon) so this function
 515 * checks and fixes these known static required protection bits.
 516 */
 517static inline pgprot_t static_protections(pgprot_t prot, unsigned long start,
 518					  unsigned long pfn, unsigned long npg,
 519					  unsigned long lpsize, int warnlvl)
 520{
 521	pgprotval_t forbidden, res;
 522	unsigned long end;
 523
 524	/*
 525	 * There is no point in checking RW/NX conflicts when the requested
 526	 * mapping is setting the page !PRESENT.
 527	 */
 528	if (!(pgprot_val(prot) & _PAGE_PRESENT))
 529		return prot;
 530
 531	/* Operate on the virtual address */
 532	end = start + npg * PAGE_SIZE - 1;
 533
 534	res = protect_kernel_text(start, end);
 535	check_conflict(warnlvl, prot, res, start, end, pfn, "Text NX");
 536	forbidden = res;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 537
 
 538	/*
 539	 * Special case to preserve a large page. If the change spawns the
 540	 * full large page mapping then there is no point to split it
 541	 * up. Happens with ftrace and is going to be removed once ftrace
 542	 * switched to text_poke().
 
 
 
 543	 */
 544	if (lpsize != (npg * PAGE_SIZE) || (start & (lpsize - 1))) {
 545		res = protect_kernel_text_ro(start, end);
 546		check_conflict(warnlvl, prot, res, start, end, pfn, "Text RO");
 547		forbidden |= res;
 548	}
 549
 550	/* Check the PFN directly */
 551	res = protect_pci_bios(pfn, pfn + npg - 1);
 552	check_conflict(warnlvl, prot, res, start, end, pfn, "PCIBIOS NX");
 553	forbidden |= res;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 554
 555	res = protect_rodata(pfn, pfn + npg - 1);
 556	check_conflict(warnlvl, prot, res, start, end, pfn, "Rodata RO");
 557	forbidden |= res;
 558
 559	return __pgprot(pgprot_val(prot) & ~forbidden);
 560}
 561
 562/*
 563 * Lookup the page table entry for a virtual address in a specific pgd.
 564 * Return a pointer to the entry and the level of the mapping.
 565 */
 566pte_t *lookup_address_in_pgd(pgd_t *pgd, unsigned long address,
 567			     unsigned int *level)
 568{
 569	p4d_t *p4d;
 570	pud_t *pud;
 571	pmd_t *pmd;
 572
 573	*level = PG_LEVEL_NONE;
 574
 575	if (pgd_none(*pgd))
 576		return NULL;
 577
 578	p4d = p4d_offset(pgd, address);
 579	if (p4d_none(*p4d))
 580		return NULL;
 581
 582	*level = PG_LEVEL_512G;
 583	if (p4d_large(*p4d) || !p4d_present(*p4d))
 584		return (pte_t *)p4d;
 585
 586	pud = pud_offset(p4d, address);
 587	if (pud_none(*pud))
 588		return NULL;
 589
 590	*level = PG_LEVEL_1G;
 591	if (pud_large(*pud) || !pud_present(*pud))
 592		return (pte_t *)pud;
 593
 594	pmd = pmd_offset(pud, address);
 595	if (pmd_none(*pmd))
 596		return NULL;
 597
 598	*level = PG_LEVEL_2M;
 599	if (pmd_large(*pmd) || !pmd_present(*pmd))
 600		return (pte_t *)pmd;
 601
 602	*level = PG_LEVEL_4K;
 603
 604	return pte_offset_kernel(pmd, address);
 605}
 606
 607/*
 608 * Lookup the page table entry for a virtual address. Return a pointer
 609 * to the entry and the level of the mapping.
 610 *
 611 * Note: We return pud and pmd either when the entry is marked large
 612 * or when the present bit is not set. Otherwise we would return a
 613 * pointer to a nonexisting mapping.
 614 */
 615pte_t *lookup_address(unsigned long address, unsigned int *level)
 616{
 617	return lookup_address_in_pgd(pgd_offset_k(address), address, level);
 618}
 619EXPORT_SYMBOL_GPL(lookup_address);
 620
 621static pte_t *_lookup_address_cpa(struct cpa_data *cpa, unsigned long address,
 622				  unsigned int *level)
 623{
 624	if (cpa->pgd)
 625		return lookup_address_in_pgd(cpa->pgd + pgd_index(address),
 626					       address, level);
 627
 628	return lookup_address(address, level);
 629}
 630
 631/*
 632 * Lookup the PMD entry for a virtual address. Return a pointer to the entry
 633 * or NULL if not present.
 634 */
 635pmd_t *lookup_pmd_address(unsigned long address)
 636{
 637	pgd_t *pgd;
 638	p4d_t *p4d;
 639	pud_t *pud;
 640
 641	pgd = pgd_offset_k(address);
 642	if (pgd_none(*pgd))
 643		return NULL;
 644
 645	p4d = p4d_offset(pgd, address);
 646	if (p4d_none(*p4d) || p4d_large(*p4d) || !p4d_present(*p4d))
 647		return NULL;
 648
 649	pud = pud_offset(p4d, address);
 650	if (pud_none(*pud) || pud_large(*pud) || !pud_present(*pud))
 651		return NULL;
 652
 653	return pmd_offset(pud, address);
 654}
 655
 656/*
 657 * This is necessary because __pa() does not work on some
 658 * kinds of memory, like vmalloc() or the alloc_remap()
 659 * areas on 32-bit NUMA systems.  The percpu areas can
 660 * end up in this kind of memory, for instance.
 661 *
 662 * This could be optimized, but it is only intended to be
 663 * used at inititalization time, and keeping it
 664 * unoptimized should increase the testing coverage for
 665 * the more obscure platforms.
 666 */
 667phys_addr_t slow_virt_to_phys(void *__virt_addr)
 668{
 669	unsigned long virt_addr = (unsigned long)__virt_addr;
 670	phys_addr_t phys_addr;
 671	unsigned long offset;
 672	enum pg_level level;
 673	pte_t *pte;
 674
 675	pte = lookup_address(virt_addr, &level);
 676	BUG_ON(!pte);
 677
 678	/*
 679	 * pXX_pfn() returns unsigned long, which must be cast to phys_addr_t
 680	 * before being left-shifted PAGE_SHIFT bits -- this trick is to
 681	 * make 32-PAE kernel work correctly.
 682	 */
 683	switch (level) {
 684	case PG_LEVEL_1G:
 685		phys_addr = (phys_addr_t)pud_pfn(*(pud_t *)pte) << PAGE_SHIFT;
 686		offset = virt_addr & ~PUD_PAGE_MASK;
 687		break;
 688	case PG_LEVEL_2M:
 689		phys_addr = (phys_addr_t)pmd_pfn(*(pmd_t *)pte) << PAGE_SHIFT;
 690		offset = virt_addr & ~PMD_PAGE_MASK;
 691		break;
 692	default:
 693		phys_addr = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
 694		offset = virt_addr & ~PAGE_MASK;
 695	}
 696
 697	return (phys_addr_t)(phys_addr | offset);
 698}
 699EXPORT_SYMBOL_GPL(slow_virt_to_phys);
 700
 701/*
 702 * Set the new pmd in all the pgds we know about:
 703 */
 704static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
 705{
 706	/* change init_mm */
 707	set_pte_atomic(kpte, pte);
 708#ifdef CONFIG_X86_32
 709	if (!SHARED_KERNEL_PMD) {
 710		struct page *page;
 711
 712		list_for_each_entry(page, &pgd_list, lru) {
 713			pgd_t *pgd;
 714			p4d_t *p4d;
 715			pud_t *pud;
 716			pmd_t *pmd;
 717
 718			pgd = (pgd_t *)page_address(page) + pgd_index(address);
 719			p4d = p4d_offset(pgd, address);
 720			pud = pud_offset(p4d, address);
 721			pmd = pmd_offset(pud, address);
 722			set_pte_atomic((pte_t *)pmd, pte);
 723		}
 724	}
 725#endif
 726}
 727
 728static pgprot_t pgprot_clear_protnone_bits(pgprot_t prot)
 729{
 730	/*
 731	 * _PAGE_GLOBAL means "global page" for present PTEs.
 732	 * But, it is also used to indicate _PAGE_PROTNONE
 733	 * for non-present PTEs.
 734	 *
 735	 * This ensures that a _PAGE_GLOBAL PTE going from
 736	 * present to non-present is not confused as
 737	 * _PAGE_PROTNONE.
 738	 */
 739	if (!(pgprot_val(prot) & _PAGE_PRESENT))
 740		pgprot_val(prot) &= ~_PAGE_GLOBAL;
 741
 742	return prot;
 743}
 744
 745static int __should_split_large_page(pte_t *kpte, unsigned long address,
 746				     struct cpa_data *cpa)
 
 747{
 748	unsigned long numpages, pmask, psize, lpaddr, pfn, old_pfn;
 749	pgprot_t old_prot, new_prot, req_prot, chk_prot;
 750	pte_t new_pte, *tmp;
 
 751	enum pg_level level;
 752
 
 
 
 
 753	/*
 754	 * Check for races, another CPU might have split this page
 755	 * up already:
 756	 */
 757	tmp = _lookup_address_cpa(cpa, address, &level);
 758	if (tmp != kpte)
 759		return 1;
 760
 761	switch (level) {
 762	case PG_LEVEL_2M:
 763		old_prot = pmd_pgprot(*(pmd_t *)kpte);
 764		old_pfn = pmd_pfn(*(pmd_t *)kpte);
 765		cpa_inc_2m_checked();
 766		break;
 767	case PG_LEVEL_1G:
 768		old_prot = pud_pgprot(*(pud_t *)kpte);
 769		old_pfn = pud_pfn(*(pud_t *)kpte);
 770		cpa_inc_1g_checked();
 771		break;
 772	default:
 773		return -EINVAL;
 
 774	}
 775
 776	psize = page_level_size(level);
 777	pmask = page_level_mask(level);
 778
 779	/*
 780	 * Calculate the number of pages, which fit into this large
 781	 * page starting at address:
 782	 */
 783	lpaddr = (address + psize) & pmask;
 784	numpages = (lpaddr - address) >> PAGE_SHIFT;
 785	if (numpages < cpa->numpages)
 786		cpa->numpages = numpages;
 787
 788	/*
 789	 * We are safe now. Check whether the new pgprot is the same:
 790	 * Convert protection attributes to 4k-format, as cpa->mask* are set
 791	 * up accordingly.
 792	 */
 793
 794	/* Clear PSE (aka _PAGE_PAT) and move PAT bit to correct position */
 795	req_prot = pgprot_large_2_4k(old_prot);
 796
 797	pgprot_val(req_prot) &= ~pgprot_val(cpa->mask_clr);
 798	pgprot_val(req_prot) |= pgprot_val(cpa->mask_set);
 799
 800	/*
 801	 * req_prot is in format of 4k pages. It must be converted to large
 802	 * page format: the caching mode includes the PAT bit located at
 803	 * different bit positions in the two formats.
 804	 */
 805	req_prot = pgprot_4k_2_large(req_prot);
 806	req_prot = pgprot_clear_protnone_bits(req_prot);
 807	if (pgprot_val(req_prot) & _PAGE_PRESENT)
 808		pgprot_val(req_prot) |= _PAGE_PSE;
 809
 810	/*
 811	 * old_pfn points to the large page base pfn. So we need to add the
 812	 * offset of the virtual address:
 813	 */
 814	pfn = old_pfn + ((address & (psize - 1)) >> PAGE_SHIFT);
 815	cpa->pfn = pfn;
 816
 817	/*
 818	 * Calculate the large page base address and the number of 4K pages
 819	 * in the large page
 820	 */
 821	lpaddr = address & pmask;
 822	numpages = psize >> PAGE_SHIFT;
 823
 824	/*
 825	 * Sanity check that the existing mapping is correct versus the static
 826	 * protections. static_protections() guards against !PRESENT, so no
 827	 * extra conditional required here.
 828	 */
 829	chk_prot = static_protections(old_prot, lpaddr, old_pfn, numpages,
 830				      psize, CPA_CONFLICT);
 
 
 831
 832	if (WARN_ON_ONCE(pgprot_val(chk_prot) != pgprot_val(old_prot))) {
 833		/*
 834		 * Split the large page and tell the split code to
 835		 * enforce static protections.
 836		 */
 837		cpa->force_static_prot = 1;
 838		return 1;
 839	}
 840
 841	/*
 842	 * Optimization: If the requested pgprot is the same as the current
 843	 * pgprot, then the large page can be preserved and no updates are
 844	 * required independent of alignment and length of the requested
 845	 * range. The above already established that the current pgprot is
 846	 * correct, which in consequence makes the requested pgprot correct
 847	 * as well if it is the same. The static protection scan below will
 848	 * not come to a different conclusion.
 849	 */
 850	if (pgprot_val(req_prot) == pgprot_val(old_prot)) {
 851		cpa_inc_lp_sameprot(level);
 852		return 0;
 853	}
 854
 855	/*
 856	 * If the requested range does not cover the full page, split it up
 857	 */
 858	if (address != lpaddr || cpa->numpages != numpages)
 859		return 1;
 860
 861	/*
 862	 * Check whether the requested pgprot is conflicting with a static
 863	 * protection requirement in the large page.
 864	 */
 865	new_prot = static_protections(req_prot, lpaddr, old_pfn, numpages,
 866				      psize, CPA_DETECT);
 867
 868	/*
 869	 * If there is a conflict, split the large page.
 870	 *
 871	 * There used to be a 4k wise evaluation trying really hard to
 872	 * preserve the large pages, but experimentation has shown, that this
 873	 * does not help at all. There might be corner cases which would
 874	 * preserve one large page occasionally, but it's really not worth the
 875	 * extra code and cycles for the common case.
 876	 */
 877	if (pgprot_val(req_prot) != pgprot_val(new_prot))
 878		return 1;
 879
 880	/* All checks passed. Update the large page mapping. */
 881	new_pte = pfn_pte(old_pfn, new_prot);
 882	__set_pmd_pte(kpte, address, new_pte);
 883	cpa->flags |= CPA_FLUSHTLB;
 884	cpa_inc_lp_preserved(level);
 885	return 0;
 886}
 887
 888static int should_split_large_page(pte_t *kpte, unsigned long address,
 889				   struct cpa_data *cpa)
 890{
 891	int do_split;
 892
 893	if (cpa->force_split)
 894		return 1;
 895
 896	spin_lock(&pgd_lock);
 897	do_split = __should_split_large_page(kpte, address, cpa);
 898	spin_unlock(&pgd_lock);
 899
 900	return do_split;
 901}
 902
 903static void split_set_pte(struct cpa_data *cpa, pte_t *pte, unsigned long pfn,
 904			  pgprot_t ref_prot, unsigned long address,
 905			  unsigned long size)
 906{
 907	unsigned int npg = PFN_DOWN(size);
 908	pgprot_t prot;
 909
 910	/*
 911	 * If should_split_large_page() discovered an inconsistent mapping,
 912	 * remove the invalid protection in the split mapping.
 913	 */
 914	if (!cpa->force_static_prot)
 915		goto set;
 916
 917	/* Hand in lpsize = 0 to enforce the protection mechanism */
 918	prot = static_protections(ref_prot, address, pfn, npg, 0, CPA_PROTECT);
 919
 920	if (pgprot_val(prot) == pgprot_val(ref_prot))
 921		goto set;
 922
 923	/*
 924	 * If this is splitting a PMD, fix it up. PUD splits cannot be
 925	 * fixed trivially as that would require to rescan the newly
 926	 * installed PMD mappings after returning from split_large_page()
 927	 * so an eventual further split can allocate the necessary PTE
 928	 * pages. Warn for now and revisit it in case this actually
 929	 * happens.
 930	 */
 931	if (size == PAGE_SIZE)
 932		ref_prot = prot;
 933	else
 934		pr_warn_once("CPA: Cannot fixup static protections for PUD split\n");
 935set:
 936	set_pte(pte, pfn_pte(pfn, ref_prot));
 937}
 938
 939static int
 940__split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
 941		   struct page *base)
 942{
 943	unsigned long lpaddr, lpinc, ref_pfn, pfn, pfninc = 1;
 944	pte_t *pbase = (pte_t *)page_address(base);
 
 945	unsigned int i, level;
 946	pgprot_t ref_prot;
 947	pte_t *tmp;
 
 948
 949	spin_lock(&pgd_lock);
 950	/*
 951	 * Check for races, another CPU might have split this page
 952	 * up for us already:
 953	 */
 954	tmp = _lookup_address_cpa(cpa, address, &level);
 955	if (tmp != kpte) {
 956		spin_unlock(&pgd_lock);
 957		return 1;
 958	}
 959
 960	paravirt_alloc_pte(&init_mm, page_to_pfn(base));
 961
 962	switch (level) {
 963	case PG_LEVEL_2M:
 964		ref_prot = pmd_pgprot(*(pmd_t *)kpte);
 965		/*
 966		 * Clear PSE (aka _PAGE_PAT) and move
 967		 * PAT bit to correct position.
 968		 */
 969		ref_prot = pgprot_large_2_4k(ref_prot);
 
 970		ref_pfn = pmd_pfn(*(pmd_t *)kpte);
 971		lpaddr = address & PMD_MASK;
 972		lpinc = PAGE_SIZE;
 973		break;
 974
 975	case PG_LEVEL_1G:
 976		ref_prot = pud_pgprot(*(pud_t *)kpte);
 977		ref_pfn = pud_pfn(*(pud_t *)kpte);
 978		pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
 979		lpaddr = address & PUD_MASK;
 980		lpinc = PMD_SIZE;
 981		/*
 982		 * Clear the PSE flags if the PRESENT flag is not set
 983		 * otherwise pmd_present/pmd_huge will return true
 984		 * even on a non present pmd.
 985		 */
 986		if (!(pgprot_val(ref_prot) & _PAGE_PRESENT))
 987			pgprot_val(ref_prot) &= ~_PAGE_PSE;
 988		break;
 989
 990	default:
 991		spin_unlock(&pgd_lock);
 992		return 1;
 993	}
 994
 995	ref_prot = pgprot_clear_protnone_bits(ref_prot);
 996
 997	/*
 998	 * Get the target pfn from the original entry:
 999	 */
1000	pfn = ref_pfn;
1001	for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc, lpaddr += lpinc)
1002		split_set_pte(cpa, pbase + i, pfn, ref_prot, lpaddr, lpinc);
1003
1004	if (virt_addr_valid(address)) {
1005		unsigned long pfn = PFN_DOWN(__pa(address));
1006
1007		if (pfn_range_is_mapped(pfn, pfn + 1))
1008			split_page_count(level);
1009	}
1010
1011	/*
1012	 * Install the new, split up pagetable.
1013	 *
1014	 * We use the standard kernel pagetable protections for the new
1015	 * pagetable protections, the actual ptes set above control the
1016	 * primary protection behavior:
1017	 */
1018	__set_pmd_pte(kpte, address, mk_pte(base, __pgprot(_KERNPG_TABLE)));
1019
1020	/*
1021	 * Do a global flush tlb after splitting the large page
1022	 * and before we do the actual change page attribute in the PTE.
1023	 *
1024	 * Without this, we violate the TLB application note, that says:
1025	 * "The TLBs may contain both ordinary and large-page
1026	 *  translations for a 4-KByte range of linear addresses. This
1027	 *  may occur if software modifies the paging structures so that
1028	 *  the page size used for the address range changes. If the two
1029	 *  translations differ with respect to page frame or attributes
1030	 *  (e.g., permissions), processor behavior is undefined and may
1031	 *  be implementation-specific."
1032	 *
1033	 * We do this global tlb flush inside the cpa_lock, so that we
1034	 * don't allow any other cpu, with stale tlb entries change the
1035	 * page attribute in parallel, that also falls into the
1036	 * just split large page entry.
1037	 */
1038	flush_tlb_all();
1039	spin_unlock(&pgd_lock);
1040
1041	return 0;
1042}
1043
1044static int split_large_page(struct cpa_data *cpa, pte_t *kpte,
1045			    unsigned long address)
1046{
1047	struct page *base;
1048
1049	if (!debug_pagealloc_enabled())
1050		spin_unlock(&cpa_lock);
1051	base = alloc_pages(GFP_KERNEL, 0);
1052	if (!debug_pagealloc_enabled())
1053		spin_lock(&cpa_lock);
1054	if (!base)
1055		return -ENOMEM;
1056
1057	if (__split_large_page(cpa, kpte, address, base))
1058		__free_page(base);
1059
1060	return 0;
1061}
1062
1063static bool try_to_free_pte_page(pte_t *pte)
1064{
1065	int i;
1066
1067	for (i = 0; i < PTRS_PER_PTE; i++)
1068		if (!pte_none(pte[i]))
1069			return false;
1070
1071	free_page((unsigned long)pte);
1072	return true;
1073}
1074
1075static bool try_to_free_pmd_page(pmd_t *pmd)
1076{
1077	int i;
1078
1079	for (i = 0; i < PTRS_PER_PMD; i++)
1080		if (!pmd_none(pmd[i]))
1081			return false;
1082
1083	free_page((unsigned long)pmd);
1084	return true;
1085}
1086
1087static bool unmap_pte_range(pmd_t *pmd, unsigned long start, unsigned long end)
1088{
1089	pte_t *pte = pte_offset_kernel(pmd, start);
1090
1091	while (start < end) {
1092		set_pte(pte, __pte(0));
1093
1094		start += PAGE_SIZE;
1095		pte++;
1096	}
1097
1098	if (try_to_free_pte_page((pte_t *)pmd_page_vaddr(*pmd))) {
1099		pmd_clear(pmd);
1100		return true;
1101	}
1102	return false;
1103}
1104
1105static void __unmap_pmd_range(pud_t *pud, pmd_t *pmd,
1106			      unsigned long start, unsigned long end)
1107{
1108	if (unmap_pte_range(pmd, start, end))
1109		if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud)))
1110			pud_clear(pud);
1111}
1112
1113static void unmap_pmd_range(pud_t *pud, unsigned long start, unsigned long end)
1114{
1115	pmd_t *pmd = pmd_offset(pud, start);
1116
1117	/*
1118	 * Not on a 2MB page boundary?
1119	 */
1120	if (start & (PMD_SIZE - 1)) {
1121		unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
1122		unsigned long pre_end = min_t(unsigned long, end, next_page);
1123
1124		__unmap_pmd_range(pud, pmd, start, pre_end);
1125
1126		start = pre_end;
1127		pmd++;
1128	}
1129
1130	/*
1131	 * Try to unmap in 2M chunks.
1132	 */
1133	while (end - start >= PMD_SIZE) {
1134		if (pmd_large(*pmd))
1135			pmd_clear(pmd);
1136		else
1137			__unmap_pmd_range(pud, pmd, start, start + PMD_SIZE);
1138
1139		start += PMD_SIZE;
1140		pmd++;
1141	}
1142
1143	/*
1144	 * 4K leftovers?
1145	 */
1146	if (start < end)
1147		return __unmap_pmd_range(pud, pmd, start, end);
1148
1149	/*
1150	 * Try again to free the PMD page if haven't succeeded above.
1151	 */
1152	if (!pud_none(*pud))
1153		if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud)))
1154			pud_clear(pud);
1155}
1156
1157static void unmap_pud_range(p4d_t *p4d, unsigned long start, unsigned long end)
1158{
1159	pud_t *pud = pud_offset(p4d, start);
1160
1161	/*
1162	 * Not on a GB page boundary?
1163	 */
1164	if (start & (PUD_SIZE - 1)) {
1165		unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1166		unsigned long pre_end	= min_t(unsigned long, end, next_page);
1167
1168		unmap_pmd_range(pud, start, pre_end);
1169
1170		start = pre_end;
1171		pud++;
1172	}
1173
1174	/*
1175	 * Try to unmap in 1G chunks?
1176	 */
1177	while (end - start >= PUD_SIZE) {
1178
1179		if (pud_large(*pud))
1180			pud_clear(pud);
1181		else
1182			unmap_pmd_range(pud, start, start + PUD_SIZE);
1183
1184		start += PUD_SIZE;
1185		pud++;
1186	}
1187
1188	/*
1189	 * 2M leftovers?
1190	 */
1191	if (start < end)
1192		unmap_pmd_range(pud, start, end);
1193
1194	/*
1195	 * No need to try to free the PUD page because we'll free it in
1196	 * populate_pgd's error path
1197	 */
1198}
1199
1200static int alloc_pte_page(pmd_t *pmd)
1201{
1202	pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
1203	if (!pte)
1204		return -1;
1205
1206	set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
1207	return 0;
1208}
1209
1210static int alloc_pmd_page(pud_t *pud)
1211{
1212	pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
1213	if (!pmd)
1214		return -1;
1215
1216	set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
1217	return 0;
1218}
1219
1220static void populate_pte(struct cpa_data *cpa,
1221			 unsigned long start, unsigned long end,
1222			 unsigned num_pages, pmd_t *pmd, pgprot_t pgprot)
1223{
1224	pte_t *pte;
1225
1226	pte = pte_offset_kernel(pmd, start);
1227
1228	pgprot = pgprot_clear_protnone_bits(pgprot);
1229
1230	while (num_pages-- && start < end) {
1231		set_pte(pte, pfn_pte(cpa->pfn, pgprot));
1232
1233		start	 += PAGE_SIZE;
1234		cpa->pfn++;
1235		pte++;
1236	}
1237}
1238
1239static long populate_pmd(struct cpa_data *cpa,
1240			 unsigned long start, unsigned long end,
1241			 unsigned num_pages, pud_t *pud, pgprot_t pgprot)
1242{
1243	long cur_pages = 0;
1244	pmd_t *pmd;
1245	pgprot_t pmd_pgprot;
1246
1247	/*
1248	 * Not on a 2M boundary?
1249	 */
1250	if (start & (PMD_SIZE - 1)) {
1251		unsigned long pre_end = start + (num_pages << PAGE_SHIFT);
1252		unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
1253
1254		pre_end   = min_t(unsigned long, pre_end, next_page);
1255		cur_pages = (pre_end - start) >> PAGE_SHIFT;
1256		cur_pages = min_t(unsigned int, num_pages, cur_pages);
1257
1258		/*
1259		 * Need a PTE page?
1260		 */
1261		pmd = pmd_offset(pud, start);
1262		if (pmd_none(*pmd))
1263			if (alloc_pte_page(pmd))
1264				return -1;
1265
1266		populate_pte(cpa, start, pre_end, cur_pages, pmd, pgprot);
1267
1268		start = pre_end;
1269	}
1270
1271	/*
1272	 * We mapped them all?
1273	 */
1274	if (num_pages == cur_pages)
1275		return cur_pages;
1276
1277	pmd_pgprot = pgprot_4k_2_large(pgprot);
1278
1279	while (end - start >= PMD_SIZE) {
1280
1281		/*
1282		 * We cannot use a 1G page so allocate a PMD page if needed.
1283		 */
1284		if (pud_none(*pud))
1285			if (alloc_pmd_page(pud))
1286				return -1;
1287
1288		pmd = pmd_offset(pud, start);
1289
1290		set_pmd(pmd, pmd_mkhuge(pfn_pmd(cpa->pfn,
1291					canon_pgprot(pmd_pgprot))));
1292
1293		start	  += PMD_SIZE;
1294		cpa->pfn  += PMD_SIZE >> PAGE_SHIFT;
1295		cur_pages += PMD_SIZE >> PAGE_SHIFT;
1296	}
1297
1298	/*
1299	 * Map trailing 4K pages.
1300	 */
1301	if (start < end) {
1302		pmd = pmd_offset(pud, start);
1303		if (pmd_none(*pmd))
1304			if (alloc_pte_page(pmd))
1305				return -1;
1306
1307		populate_pte(cpa, start, end, num_pages - cur_pages,
1308			     pmd, pgprot);
1309	}
1310	return num_pages;
1311}
1312
1313static int populate_pud(struct cpa_data *cpa, unsigned long start, p4d_t *p4d,
1314			pgprot_t pgprot)
1315{
1316	pud_t *pud;
1317	unsigned long end;
1318	long cur_pages = 0;
1319	pgprot_t pud_pgprot;
1320
1321	end = start + (cpa->numpages << PAGE_SHIFT);
1322
1323	/*
1324	 * Not on a Gb page boundary? => map everything up to it with
1325	 * smaller pages.
1326	 */
1327	if (start & (PUD_SIZE - 1)) {
1328		unsigned long pre_end;
1329		unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1330
1331		pre_end   = min_t(unsigned long, end, next_page);
1332		cur_pages = (pre_end - start) >> PAGE_SHIFT;
1333		cur_pages = min_t(int, (int)cpa->numpages, cur_pages);
1334
1335		pud = pud_offset(p4d, start);
1336
1337		/*
1338		 * Need a PMD page?
1339		 */
1340		if (pud_none(*pud))
1341			if (alloc_pmd_page(pud))
1342				return -1;
1343
1344		cur_pages = populate_pmd(cpa, start, pre_end, cur_pages,
1345					 pud, pgprot);
1346		if (cur_pages < 0)
1347			return cur_pages;
1348
1349		start = pre_end;
1350	}
1351
1352	/* We mapped them all? */
1353	if (cpa->numpages == cur_pages)
1354		return cur_pages;
1355
1356	pud = pud_offset(p4d, start);
1357	pud_pgprot = pgprot_4k_2_large(pgprot);
1358
1359	/*
1360	 * Map everything starting from the Gb boundary, possibly with 1G pages
1361	 */
1362	while (boot_cpu_has(X86_FEATURE_GBPAGES) && end - start >= PUD_SIZE) {
1363		set_pud(pud, pud_mkhuge(pfn_pud(cpa->pfn,
1364				   canon_pgprot(pud_pgprot))));
1365
1366		start	  += PUD_SIZE;
1367		cpa->pfn  += PUD_SIZE >> PAGE_SHIFT;
1368		cur_pages += PUD_SIZE >> PAGE_SHIFT;
1369		pud++;
1370	}
1371
1372	/* Map trailing leftover */
1373	if (start < end) {
1374		long tmp;
1375
1376		pud = pud_offset(p4d, start);
1377		if (pud_none(*pud))
1378			if (alloc_pmd_page(pud))
1379				return -1;
1380
1381		tmp = populate_pmd(cpa, start, end, cpa->numpages - cur_pages,
1382				   pud, pgprot);
1383		if (tmp < 0)
1384			return cur_pages;
1385
1386		cur_pages += tmp;
1387	}
1388	return cur_pages;
1389}
1390
1391/*
1392 * Restrictions for kernel page table do not necessarily apply when mapping in
1393 * an alternate PGD.
1394 */
1395static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
1396{
1397	pgprot_t pgprot = __pgprot(_KERNPG_TABLE);
1398	pud_t *pud = NULL;	/* shut up gcc */
1399	p4d_t *p4d;
1400	pgd_t *pgd_entry;
1401	long ret;
1402
1403	pgd_entry = cpa->pgd + pgd_index(addr);
1404
1405	if (pgd_none(*pgd_entry)) {
1406		p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
1407		if (!p4d)
1408			return -1;
1409
1410		set_pgd(pgd_entry, __pgd(__pa(p4d) | _KERNPG_TABLE));
1411	}
1412
1413	/*
1414	 * Allocate a PUD page and hand it down for mapping.
1415	 */
1416	p4d = p4d_offset(pgd_entry, addr);
1417	if (p4d_none(*p4d)) {
1418		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
1419		if (!pud)
1420			return -1;
1421
1422		set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
1423	}
1424
1425	pgprot_val(pgprot) &= ~pgprot_val(cpa->mask_clr);
1426	pgprot_val(pgprot) |=  pgprot_val(cpa->mask_set);
1427
1428	ret = populate_pud(cpa, addr, p4d, pgprot);
1429	if (ret < 0) {
1430		/*
1431		 * Leave the PUD page in place in case some other CPU or thread
1432		 * already found it, but remove any useless entries we just
1433		 * added to it.
1434		 */
1435		unmap_pud_range(p4d, addr,
1436				addr + (cpa->numpages << PAGE_SHIFT));
1437		return ret;
1438	}
1439
1440	cpa->numpages = ret;
1441	return 0;
1442}
1443
1444static int __cpa_process_fault(struct cpa_data *cpa, unsigned long vaddr,
1445			       int primary)
1446{
1447	if (cpa->pgd) {
1448		/*
1449		 * Right now, we only execute this code path when mapping
1450		 * the EFI virtual memory map regions, no other users
1451		 * provide a ->pgd value. This may change in the future.
1452		 */
1453		return populate_pgd(cpa, vaddr);
1454	}
1455
1456	/*
1457	 * Ignore all non primary paths.
1458	 */
1459	if (!primary) {
1460		cpa->numpages = 1;
1461		return 0;
1462	}
1463
1464	/*
1465	 * Ignore the NULL PTE for kernel identity mapping, as it is expected
1466	 * to have holes.
1467	 * Also set numpages to '1' indicating that we processed cpa req for
1468	 * one virtual address page and its pfn. TBD: numpages can be set based
1469	 * on the initial value and the level returned by lookup_address().
1470	 */
1471	if (within(vaddr, PAGE_OFFSET,
1472		   PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))) {
1473		cpa->numpages = 1;
1474		cpa->pfn = __pa(vaddr) >> PAGE_SHIFT;
1475		return 0;
1476
1477	} else if (__cpa_pfn_in_highmap(cpa->pfn)) {
1478		/* Faults in the highmap are OK, so do not warn: */
1479		return -EFAULT;
1480	} else {
1481		WARN(1, KERN_WARNING "CPA: called for zero pte. "
1482			"vaddr = %lx cpa->vaddr = %lx\n", vaddr,
1483			*cpa->vaddr);
1484
1485		return -EFAULT;
1486	}
1487}
1488
1489static int __change_page_attr(struct cpa_data *cpa, int primary)
1490{
1491	unsigned long address;
1492	int do_split, err;
1493	unsigned int level;
1494	pte_t *kpte, old_pte;
1495
1496	address = __cpa_addr(cpa, cpa->curpage);
 
 
 
 
 
 
 
 
1497repeat:
1498	kpte = _lookup_address_cpa(cpa, address, &level);
1499	if (!kpte)
1500		return __cpa_process_fault(cpa, address, primary);
1501
1502	old_pte = *kpte;
1503	if (pte_none(old_pte))
1504		return __cpa_process_fault(cpa, address, primary);
1505
1506	if (level == PG_LEVEL_4K) {
1507		pte_t new_pte;
1508		pgprot_t new_prot = pte_pgprot(old_pte);
1509		unsigned long pfn = pte_pfn(old_pte);
1510
1511		pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
1512		pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
1513
1514		cpa_inc_4k_install();
1515		/* Hand in lpsize = 0 to enforce the protection mechanism */
1516		new_prot = static_protections(new_prot, address, pfn, 1, 0,
1517					      CPA_PROTECT);
1518
1519		new_prot = pgprot_clear_protnone_bits(new_prot);
1520
1521		/*
1522		 * We need to keep the pfn from the existing PTE,
1523		 * after all we're only going to change it's attributes
1524		 * not the memory it points to
1525		 */
1526		new_pte = pfn_pte(pfn, new_prot);
1527		cpa->pfn = pfn;
1528		/*
1529		 * Do we really change anything ?
1530		 */
1531		if (pte_val(old_pte) != pte_val(new_pte)) {
1532			set_pte_atomic(kpte, new_pte);
1533			cpa->flags |= CPA_FLUSHTLB;
1534		}
1535		cpa->numpages = 1;
1536		return 0;
1537	}
1538
1539	/*
1540	 * Check, whether we can keep the large page intact
1541	 * and just change the pte:
1542	 */
1543	do_split = should_split_large_page(kpte, address, cpa);
1544	/*
1545	 * When the range fits into the existing large page,
1546	 * return. cp->numpages and cpa->tlbflush have been updated in
1547	 * try_large_page:
1548	 */
1549	if (do_split <= 0)
1550		return do_split;
1551
1552	/*
1553	 * We have to split the large page:
1554	 */
1555	err = split_large_page(cpa, kpte, address);
1556	if (!err)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1557		goto repeat;
 
1558
1559	return err;
1560}
1561
1562static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias);
1563
1564static int cpa_process_alias(struct cpa_data *cpa)
1565{
1566	struct cpa_data alias_cpa;
1567	unsigned long laddr = (unsigned long)__va(cpa->pfn << PAGE_SHIFT);
1568	unsigned long vaddr;
1569	int ret;
1570
1571	if (!pfn_range_is_mapped(cpa->pfn, cpa->pfn + 1))
1572		return 0;
1573
1574	/*
1575	 * No need to redo, when the primary call touched the direct
1576	 * mapping already:
1577	 */
1578	vaddr = __cpa_addr(cpa, cpa->curpage);
 
 
 
 
 
 
 
 
 
1579	if (!(within(vaddr, PAGE_OFFSET,
1580		    PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT)))) {
1581
1582		alias_cpa = *cpa;
1583		alias_cpa.vaddr = &laddr;
1584		alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1585		alias_cpa.curpage = 0;
1586
1587		ret = __change_page_attr_set_clr(&alias_cpa, 0);
1588		if (ret)
1589			return ret;
1590	}
1591
1592#ifdef CONFIG_X86_64
1593	/*
1594	 * If the primary call didn't touch the high mapping already
1595	 * and the physical address is inside the kernel map, we need
1596	 * to touch the high mapped kernel as well:
1597	 */
1598	if (!within(vaddr, (unsigned long)_text, _brk_end) &&
1599	    __cpa_pfn_in_highmap(cpa->pfn)) {
1600		unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) +
1601					       __START_KERNEL_map - phys_base;
1602		alias_cpa = *cpa;
1603		alias_cpa.vaddr = &temp_cpa_vaddr;
1604		alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1605		alias_cpa.curpage = 0;
1606
1607		/*
1608		 * The high mapping range is imprecise, so ignore the
1609		 * return value.
1610		 */
1611		__change_page_attr_set_clr(&alias_cpa, 0);
1612	}
1613#endif
1614
1615	return 0;
1616}
1617
1618static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
1619{
1620	unsigned long numpages = cpa->numpages;
1621	unsigned long rempages = numpages;
1622	int ret = 0;
1623
1624	while (rempages) {
1625		/*
1626		 * Store the remaining nr of pages for the large page
1627		 * preservation check.
1628		 */
1629		cpa->numpages = rempages;
1630		/* for array changes, we can't use large page */
1631		if (cpa->flags & (CPA_ARRAY | CPA_PAGES_ARRAY))
1632			cpa->numpages = 1;
1633
1634		if (!debug_pagealloc_enabled())
1635			spin_lock(&cpa_lock);
1636		ret = __change_page_attr(cpa, checkalias);
1637		if (!debug_pagealloc_enabled())
1638			spin_unlock(&cpa_lock);
1639		if (ret)
1640			goto out;
1641
1642		if (checkalias) {
1643			ret = cpa_process_alias(cpa);
1644			if (ret)
1645				goto out;
1646		}
1647
1648		/*
1649		 * Adjust the number of pages with the result of the
1650		 * CPA operation. Either a large page has been
1651		 * preserved or a single page update happened.
1652		 */
1653		BUG_ON(cpa->numpages > rempages || !cpa->numpages);
1654		rempages -= cpa->numpages;
1655		cpa->curpage += cpa->numpages;
1656	}
 
 
1657
1658out:
1659	/* Restore the original numpages */
1660	cpa->numpages = numpages;
1661	return ret;
1662}
1663
1664static int change_page_attr_set_clr(unsigned long *addr, int numpages,
1665				    pgprot_t mask_set, pgprot_t mask_clr,
1666				    int force_split, int in_flag,
1667				    struct page **pages)
1668{
1669	struct cpa_data cpa;
1670	int ret, cache, checkalias;
 
1671
1672	memset(&cpa, 0, sizeof(cpa));
1673
1674	/*
1675	 * Check, if we are requested to set a not supported
1676	 * feature.  Clearing non-supported features is OK.
1677	 */
1678	mask_set = canon_pgprot(mask_set);
1679
1680	if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
1681		return 0;
1682
1683	/* Ensure we are PAGE_SIZE aligned */
1684	if (in_flag & CPA_ARRAY) {
1685		int i;
1686		for (i = 0; i < numpages; i++) {
1687			if (addr[i] & ~PAGE_MASK) {
1688				addr[i] &= PAGE_MASK;
1689				WARN_ON_ONCE(1);
1690			}
1691		}
1692	} else if (!(in_flag & CPA_PAGES_ARRAY)) {
1693		/*
1694		 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
1695		 * No need to check in that case
1696		 */
1697		if (*addr & ~PAGE_MASK) {
1698			*addr &= PAGE_MASK;
1699			/*
1700			 * People should not be passing in unaligned addresses:
1701			 */
1702			WARN_ON_ONCE(1);
1703		}
 
 
 
 
 
1704	}
1705
1706	/* Must avoid aliasing mappings in the highmem code */
1707	kmap_flush_unused();
1708
1709	vm_unmap_aliases();
1710
1711	cpa.vaddr = addr;
1712	cpa.pages = pages;
1713	cpa.numpages = numpages;
1714	cpa.mask_set = mask_set;
1715	cpa.mask_clr = mask_clr;
1716	cpa.flags = 0;
1717	cpa.curpage = 0;
1718	cpa.force_split = force_split;
1719
1720	if (in_flag & (CPA_ARRAY | CPA_PAGES_ARRAY))
1721		cpa.flags |= in_flag;
1722
1723	/* No alias checking for _NX bit modifications */
1724	checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX;
1725	/* Has caller explicitly disabled alias checking? */
1726	if (in_flag & CPA_NO_CHECK_ALIAS)
1727		checkalias = 0;
1728
1729	ret = __change_page_attr_set_clr(&cpa, checkalias);
1730
1731	/*
1732	 * Check whether we really changed something:
1733	 */
1734	if (!(cpa.flags & CPA_FLUSHTLB))
1735		goto out;
1736
1737	/*
1738	 * No need to flush, when we did not set any of the caching
1739	 * attributes:
1740	 */
1741	cache = !!pgprot2cachemode(mask_set);
1742
1743	/*
1744	 * On error; flush everything to be sure.
1745	 */
1746	if (ret) {
 
 
 
 
 
 
 
 
 
1747		cpa_flush_all(cache);
1748		goto out;
1749	}
1750
1751	cpa_flush(&cpa, cache);
1752out:
1753	return ret;
1754}
1755
1756static inline int change_page_attr_set(unsigned long *addr, int numpages,
1757				       pgprot_t mask, int array)
1758{
1759	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0,
1760		(array ? CPA_ARRAY : 0), NULL);
1761}
1762
1763static inline int change_page_attr_clear(unsigned long *addr, int numpages,
1764					 pgprot_t mask, int array)
1765{
1766	return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0,
1767		(array ? CPA_ARRAY : 0), NULL);
1768}
1769
1770static inline int cpa_set_pages_array(struct page **pages, int numpages,
1771				       pgprot_t mask)
1772{
1773	return change_page_attr_set_clr(NULL, numpages, mask, __pgprot(0), 0,
1774		CPA_PAGES_ARRAY, pages);
1775}
1776
1777static inline int cpa_clear_pages_array(struct page **pages, int numpages,
1778					 pgprot_t mask)
1779{
1780	return change_page_attr_set_clr(NULL, numpages, __pgprot(0), mask, 0,
1781		CPA_PAGES_ARRAY, pages);
1782}
1783
1784int _set_memory_uc(unsigned long addr, int numpages)
1785{
1786	/*
1787	 * for now UC MINUS. see comments in ioremap_nocache()
1788	 * If you really need strong UC use ioremap_uc(), but note
1789	 * that you cannot override IO areas with set_memory_*() as
1790	 * these helpers cannot work with IO memory.
1791	 */
1792	return change_page_attr_set(&addr, numpages,
1793				    cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1794				    0);
1795}
1796
1797int set_memory_uc(unsigned long addr, int numpages)
1798{
1799	int ret;
1800
1801	/*
1802	 * for now UC MINUS. see comments in ioremap_nocache()
1803	 */
1804	ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1805			      _PAGE_CACHE_MODE_UC_MINUS, NULL);
1806	if (ret)
1807		goto out_err;
1808
1809	ret = _set_memory_uc(addr, numpages);
1810	if (ret)
1811		goto out_free;
1812
1813	return 0;
1814
1815out_free:
1816	free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1817out_err:
1818	return ret;
1819}
1820EXPORT_SYMBOL(set_memory_uc);
1821
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1822int _set_memory_wc(unsigned long addr, int numpages)
1823{
1824	int ret;
 
1825
1826	ret = change_page_attr_set(&addr, numpages,
1827				   cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1828				   0);
1829	if (!ret) {
1830		ret = change_page_attr_set_clr(&addr, numpages,
1831					       cachemode2pgprot(_PAGE_CACHE_MODE_WC),
 
1832					       __pgprot(_PAGE_CACHE_MASK),
1833					       0, 0, NULL);
1834	}
1835	return ret;
1836}
1837
1838int set_memory_wc(unsigned long addr, int numpages)
1839{
1840	int ret;
1841
1842	ret = reserve_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1843		_PAGE_CACHE_MODE_WC, NULL);
1844	if (ret)
1845		return ret;
1846
1847	ret = _set_memory_wc(addr, numpages);
1848	if (ret)
1849		free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1850
1851	return ret;
1852}
1853EXPORT_SYMBOL(set_memory_wc);
1854
1855int _set_memory_wt(unsigned long addr, int numpages)
1856{
1857	return change_page_attr_set(&addr, numpages,
1858				    cachemode2pgprot(_PAGE_CACHE_MODE_WT), 0);
1859}
1860
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1861int _set_memory_wb(unsigned long addr, int numpages)
1862{
1863	/* WB cache mode is hard wired to all cache attribute bits being 0 */
1864	return change_page_attr_clear(&addr, numpages,
1865				      __pgprot(_PAGE_CACHE_MASK), 0);
1866}
1867
1868int set_memory_wb(unsigned long addr, int numpages)
1869{
1870	int ret;
1871
1872	ret = _set_memory_wb(addr, numpages);
1873	if (ret)
1874		return ret;
1875
1876	free_memtype(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1877	return 0;
1878}
1879EXPORT_SYMBOL(set_memory_wb);
1880
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1881int set_memory_x(unsigned long addr, int numpages)
1882{
1883	if (!(__supported_pte_mask & _PAGE_NX))
1884		return 0;
1885
1886	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0);
1887}
 
1888
1889int set_memory_nx(unsigned long addr, int numpages)
1890{
1891	if (!(__supported_pte_mask & _PAGE_NX))
1892		return 0;
1893
1894	return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0);
1895}
 
1896
1897int set_memory_ro(unsigned long addr, int numpages)
1898{
1899	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0);
1900}
1901
1902int set_memory_rw(unsigned long addr, int numpages)
1903{
1904	return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
1905}
1906
1907int set_memory_np(unsigned long addr, int numpages)
1908{
1909	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
1910}
1911
1912int set_memory_np_noalias(unsigned long addr, int numpages)
1913{
1914	int cpa_flags = CPA_NO_CHECK_ALIAS;
1915
1916	return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
1917					__pgprot(_PAGE_PRESENT), 0,
1918					cpa_flags, NULL);
1919}
1920
1921int set_memory_4k(unsigned long addr, int numpages)
1922{
1923	return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
1924					__pgprot(0), 1, 0, NULL);
1925}
1926
1927int set_memory_nonglobal(unsigned long addr, int numpages)
1928{
1929	return change_page_attr_clear(&addr, numpages,
1930				      __pgprot(_PAGE_GLOBAL), 0);
1931}
1932
1933int set_memory_global(unsigned long addr, int numpages)
1934{
1935	return change_page_attr_set(&addr, numpages,
1936				    __pgprot(_PAGE_GLOBAL), 0);
1937}
1938
1939static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc)
1940{
1941	struct cpa_data cpa;
 
1942	int ret;
1943
1944	/* Nothing to do if memory encryption is not active */
1945	if (!mem_encrypt_active())
1946		return 0;
1947
1948	/* Should not be working on unaligned addresses */
1949	if (WARN_ONCE(addr & ~PAGE_MASK, "misaligned address: %#lx\n", addr))
1950		addr &= PAGE_MASK;
1951
 
 
1952	memset(&cpa, 0, sizeof(cpa));
1953	cpa.vaddr = &addr;
1954	cpa.numpages = numpages;
1955	cpa.mask_set = enc ? __pgprot(_PAGE_ENC) : __pgprot(0);
1956	cpa.mask_clr = enc ? __pgprot(0) : __pgprot(_PAGE_ENC);
1957	cpa.pgd = init_mm.pgd;
1958
1959	/* Must avoid aliasing mappings in the highmem code */
1960	kmap_flush_unused();
1961	vm_unmap_aliases();
1962
1963	/*
1964	 * Before changing the encryption attribute, we need to flush caches.
1965	 */
1966	cpa_flush(&cpa, 1);
 
 
 
1967
1968	ret = __change_page_attr_set_clr(&cpa, 1);
1969
1970	/*
1971	 * After changing the encryption attribute, we need to flush TLBs again
1972	 * in case any speculative TLB caching occurred (but no need to flush
1973	 * caches again).  We could just use cpa_flush_all(), but in case TLB
1974	 * flushing gets optimized in the cpa_flush() path use the same logic
1975	 * as above.
1976	 */
1977	cpa_flush(&cpa, 0);
 
 
 
1978
1979	return ret;
1980}
1981
1982int set_memory_encrypted(unsigned long addr, int numpages)
1983{
1984	return __set_memory_enc_dec(addr, numpages, true);
1985}
1986EXPORT_SYMBOL_GPL(set_memory_encrypted);
1987
1988int set_memory_decrypted(unsigned long addr, int numpages)
1989{
1990	return __set_memory_enc_dec(addr, numpages, false);
1991}
1992EXPORT_SYMBOL_GPL(set_memory_decrypted);
1993
1994int set_pages_uc(struct page *page, int numpages)
1995{
1996	unsigned long addr = (unsigned long)page_address(page);
1997
1998	return set_memory_uc(addr, numpages);
1999}
2000EXPORT_SYMBOL(set_pages_uc);
2001
2002static int _set_pages_array(struct page **pages, int numpages,
2003		enum page_cache_mode new_type)
2004{
2005	unsigned long start;
2006	unsigned long end;
2007	enum page_cache_mode set_type;
2008	int i;
2009	int free_idx;
2010	int ret;
2011
2012	for (i = 0; i < numpages; i++) {
2013		if (PageHighMem(pages[i]))
2014			continue;
2015		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2016		end = start + PAGE_SIZE;
2017		if (reserve_memtype(start, end, new_type, NULL))
2018			goto err_out;
2019	}
2020
2021	/* If WC, set to UC- first and then WC */
2022	set_type = (new_type == _PAGE_CACHE_MODE_WC) ?
2023				_PAGE_CACHE_MODE_UC_MINUS : new_type;
2024
2025	ret = cpa_set_pages_array(pages, numpages,
2026				  cachemode2pgprot(set_type));
2027	if (!ret && new_type == _PAGE_CACHE_MODE_WC)
2028		ret = change_page_attr_set_clr(NULL, numpages,
2029					       cachemode2pgprot(
2030						_PAGE_CACHE_MODE_WC),
2031					       __pgprot(_PAGE_CACHE_MASK),
2032					       0, CPA_PAGES_ARRAY, pages);
2033	if (ret)
2034		goto err_out;
2035	return 0; /* Success */
2036err_out:
2037	free_idx = i;
2038	for (i = 0; i < free_idx; i++) {
2039		if (PageHighMem(pages[i]))
2040			continue;
2041		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2042		end = start + PAGE_SIZE;
2043		free_memtype(start, end);
2044	}
2045	return -EINVAL;
2046}
2047
2048int set_pages_array_uc(struct page **pages, int numpages)
2049{
2050	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_UC_MINUS);
2051}
2052EXPORT_SYMBOL(set_pages_array_uc);
2053
2054int set_pages_array_wc(struct page **pages, int numpages)
2055{
2056	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WC);
2057}
2058EXPORT_SYMBOL(set_pages_array_wc);
2059
2060int set_pages_array_wt(struct page **pages, int numpages)
2061{
2062	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WT);
2063}
2064EXPORT_SYMBOL_GPL(set_pages_array_wt);
2065
2066int set_pages_wb(struct page *page, int numpages)
2067{
2068	unsigned long addr = (unsigned long)page_address(page);
2069
2070	return set_memory_wb(addr, numpages);
2071}
2072EXPORT_SYMBOL(set_pages_wb);
2073
2074int set_pages_array_wb(struct page **pages, int numpages)
2075{
2076	int retval;
2077	unsigned long start;
2078	unsigned long end;
2079	int i;
2080
2081	/* WB cache mode is hard wired to all cache attribute bits being 0 */
2082	retval = cpa_clear_pages_array(pages, numpages,
2083			__pgprot(_PAGE_CACHE_MASK));
2084	if (retval)
2085		return retval;
2086
2087	for (i = 0; i < numpages; i++) {
2088		if (PageHighMem(pages[i]))
2089			continue;
2090		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2091		end = start + PAGE_SIZE;
2092		free_memtype(start, end);
2093	}
2094
2095	return 0;
2096}
2097EXPORT_SYMBOL(set_pages_array_wb);
2098
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2099int set_pages_ro(struct page *page, int numpages)
2100{
2101	unsigned long addr = (unsigned long)page_address(page);
2102
2103	return set_memory_ro(addr, numpages);
2104}
2105
2106int set_pages_rw(struct page *page, int numpages)
2107{
2108	unsigned long addr = (unsigned long)page_address(page);
2109
2110	return set_memory_rw(addr, numpages);
2111}
2112
 
 
2113static int __set_pages_p(struct page *page, int numpages)
2114{
2115	unsigned long tempaddr = (unsigned long) page_address(page);
2116	struct cpa_data cpa = { .vaddr = &tempaddr,
2117				.pgd = NULL,
2118				.numpages = numpages,
2119				.mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
2120				.mask_clr = __pgprot(0),
2121				.flags = 0};
2122
2123	/*
2124	 * No alias checking needed for setting present flag. otherwise,
2125	 * we may need to break large pages for 64-bit kernel text
2126	 * mappings (this adds to complexity if we want to do this from
2127	 * atomic context especially). Let's keep it simple!
2128	 */
2129	return __change_page_attr_set_clr(&cpa, 0);
2130}
2131
2132static int __set_pages_np(struct page *page, int numpages)
2133{
2134	unsigned long tempaddr = (unsigned long) page_address(page);
2135	struct cpa_data cpa = { .vaddr = &tempaddr,
2136				.pgd = NULL,
2137				.numpages = numpages,
2138				.mask_set = __pgprot(0),
2139				.mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW),
2140				.flags = 0};
2141
2142	/*
2143	 * No alias checking needed for setting not present flag. otherwise,
2144	 * we may need to break large pages for 64-bit kernel text
2145	 * mappings (this adds to complexity if we want to do this from
2146	 * atomic context especially). Let's keep it simple!
2147	 */
2148	return __change_page_attr_set_clr(&cpa, 0);
2149}
2150
2151int set_direct_map_invalid_noflush(struct page *page)
2152{
2153	return __set_pages_np(page, 1);
2154}
2155
2156int set_direct_map_default_noflush(struct page *page)
2157{
2158	return __set_pages_p(page, 1);
2159}
2160
2161void __kernel_map_pages(struct page *page, int numpages, int enable)
2162{
2163	if (PageHighMem(page))
2164		return;
2165	if (!enable) {
2166		debug_check_no_locks_freed(page_address(page),
2167					   numpages * PAGE_SIZE);
2168	}
2169
2170	/*
2171	 * The return value is ignored as the calls cannot fail.
2172	 * Large pages for identity mappings are not used at boot time
2173	 * and hence no memory allocations during large page split.
2174	 */
2175	if (enable)
2176		__set_pages_p(page, numpages);
2177	else
2178		__set_pages_np(page, numpages);
2179
2180	/*
2181	 * We should perform an IPI and flush all tlbs,
2182	 * but that can deadlock->flush only current cpu.
2183	 * Preemption needs to be disabled around __flush_tlb_all() due to
2184	 * CR3 reload in __native_flush_tlb().
2185	 */
2186	preempt_disable();
2187	__flush_tlb_all();
2188	preempt_enable();
2189
2190	arch_flush_lazy_mmu_mode();
2191}
2192
2193#ifdef CONFIG_HIBERNATION
 
2194bool kernel_page_present(struct page *page)
2195{
2196	unsigned int level;
2197	pte_t *pte;
2198
2199	if (PageHighMem(page))
2200		return false;
2201
2202	pte = lookup_address((unsigned long)page_address(page), &level);
2203	return (pte_val(*pte) & _PAGE_PRESENT);
2204}
 
2205#endif /* CONFIG_HIBERNATION */
2206
2207int __init kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address,
2208				   unsigned numpages, unsigned long page_flags)
 
 
2209{
2210	int retval = -EINVAL;
2211
2212	struct cpa_data cpa = {
2213		.vaddr = &address,
2214		.pfn = pfn,
2215		.pgd = pgd,
2216		.numpages = numpages,
2217		.mask_set = __pgprot(0),
2218		.mask_clr = __pgprot(0),
2219		.flags = 0,
2220	};
2221
2222	WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2223
2224	if (!(__supported_pte_mask & _PAGE_NX))
2225		goto out;
2226
2227	if (!(page_flags & _PAGE_NX))
2228		cpa.mask_clr = __pgprot(_PAGE_NX);
2229
2230	if (!(page_flags & _PAGE_RW))
2231		cpa.mask_clr = __pgprot(_PAGE_RW);
2232
2233	if (!(page_flags & _PAGE_ENC))
2234		cpa.mask_clr = pgprot_encrypted(cpa.mask_clr);
2235
2236	cpa.mask_set = __pgprot(_PAGE_PRESENT | page_flags);
2237
2238	retval = __change_page_attr_set_clr(&cpa, 0);
2239	__flush_tlb_all();
2240
2241out:
2242	return retval;
2243}
2244
2245/*
2246 * __flush_tlb_all() flushes mappings only on current CPU and hence this
2247 * function shouldn't be used in an SMP environment. Presently, it's used only
2248 * during boot (way before smp_init()) by EFI subsystem and hence is ok.
2249 */
2250int __init kernel_unmap_pages_in_pgd(pgd_t *pgd, unsigned long address,
2251				     unsigned long numpages)
2252{
2253	int retval;
2254
2255	/*
2256	 * The typical sequence for unmapping is to find a pte through
2257	 * lookup_address_in_pgd() (ideally, it should never return NULL because
2258	 * the address is already mapped) and change it's protections. As pfn is
2259	 * the *target* of a mapping, it's not useful while unmapping.
2260	 */
2261	struct cpa_data cpa = {
2262		.vaddr		= &address,
2263		.pfn		= 0,
2264		.pgd		= pgd,
2265		.numpages	= numpages,
2266		.mask_set	= __pgprot(0),
2267		.mask_clr	= __pgprot(_PAGE_PRESENT | _PAGE_RW),
2268		.flags		= 0,
2269	};
2270
2271	WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2272
2273	retval = __change_page_attr_set_clr(&cpa, 0);
2274	__flush_tlb_all();
2275
2276	return retval;
2277}
2278
2279/*
2280 * The testcases use internal knowledge of the implementation that shouldn't
2281 * be exposed to the rest of the kernel. Include these directly here.
2282 */
2283#ifdef CONFIG_CPA_DEBUG
2284#include "pageattr-test.c"
2285#endif