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#include <linux/libnvdimm.h>
 
 
 
 
 
  19
  20#include <asm/e820/api.h>
  21#include <asm/processor.h>
  22#include <asm/tlbflush.h>
  23#include <asm/sections.h>
  24#include <asm/setup.h>
  25#include <linux/uaccess.h>
  26#include <asm/pgalloc.h>
  27#include <asm/proto.h>
  28#include <asm/memtype.h>
  29#include <asm/set_memory.h>
 
  30
  31#include "../mm_internal.h"
  32
  33/*
  34 * The current flushing context - we pass it instead of 5 arguments:
  35 */
  36struct cpa_data {
  37	unsigned long	*vaddr;
  38	pgd_t		*pgd;
  39	pgprot_t	mask_set;
  40	pgprot_t	mask_clr;
  41	unsigned long	numpages;
  42	unsigned long	curpage;
  43	unsigned long	pfn;
  44	unsigned int	flags;
  45	unsigned int	force_split		: 1,
  46			force_static_prot	: 1,
  47			force_flush_all		: 1;
  48	struct page	**pages;
  49};
  50
  51enum cpa_warn {
  52	CPA_CONFLICT,
  53	CPA_PROTECT,
  54	CPA_DETECT,
  55};
  56
  57static const int cpa_warn_level = CPA_PROTECT;
  58
  59/*
  60 * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
  61 * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
  62 * entries change the page attribute in parallel to some other cpu
  63 * splitting a large page entry along with changing the attribute.
  64 */
  65static DEFINE_SPINLOCK(cpa_lock);
  66
  67#define CPA_FLUSHTLB 1
  68#define CPA_ARRAY 2
  69#define CPA_PAGES_ARRAY 4
  70#define CPA_NO_CHECK_ALIAS 8 /* Do not search for aliases */
  71
  72static inline pgprot_t cachemode2pgprot(enum page_cache_mode pcm)
  73{
  74	return __pgprot(cachemode2protval(pcm));
  75}
  76
  77#ifdef CONFIG_PROC_FS
  78static unsigned long direct_pages_count[PG_LEVEL_NUM];
  79
  80void update_page_count(int level, unsigned long pages)
  81{
  82	/* Protect against CPA */
  83	spin_lock(&pgd_lock);
  84	direct_pages_count[level] += pages;
  85	spin_unlock(&pgd_lock);
  86}
  87
  88static void split_page_count(int level)
  89{
  90	if (direct_pages_count[level] == 0)
  91		return;
  92
  93	direct_pages_count[level]--;
 
 
 
 
 
 
  94	direct_pages_count[level - 1] += PTRS_PER_PTE;
  95}
  96
  97void arch_report_meminfo(struct seq_file *m)
  98{
  99	seq_printf(m, "DirectMap4k:    %8lu kB\n",
 100			direct_pages_count[PG_LEVEL_4K] << 2);
 101#if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
 102	seq_printf(m, "DirectMap2M:    %8lu kB\n",
 103			direct_pages_count[PG_LEVEL_2M] << 11);
 104#else
 105	seq_printf(m, "DirectMap4M:    %8lu kB\n",
 106			direct_pages_count[PG_LEVEL_2M] << 12);
 107#endif
 108	if (direct_gbpages)
 109		seq_printf(m, "DirectMap1G:    %8lu kB\n",
 110			direct_pages_count[PG_LEVEL_1G] << 20);
 111}
 112#else
 113static inline void split_page_count(int level) { }
 114#endif
 115
 116#ifdef CONFIG_X86_CPA_STATISTICS
 117
 118static unsigned long cpa_1g_checked;
 119static unsigned long cpa_1g_sameprot;
 120static unsigned long cpa_1g_preserved;
 121static unsigned long cpa_2m_checked;
 122static unsigned long cpa_2m_sameprot;
 123static unsigned long cpa_2m_preserved;
 124static unsigned long cpa_4k_install;
 125
 126static inline void cpa_inc_1g_checked(void)
 127{
 128	cpa_1g_checked++;
 129}
 130
 131static inline void cpa_inc_2m_checked(void)
 132{
 133	cpa_2m_checked++;
 134}
 135
 136static inline void cpa_inc_4k_install(void)
 137{
 138	data_race(cpa_4k_install++);
 139}
 140
 141static inline void cpa_inc_lp_sameprot(int level)
 142{
 143	if (level == PG_LEVEL_1G)
 144		cpa_1g_sameprot++;
 145	else
 146		cpa_2m_sameprot++;
 147}
 148
 149static inline void cpa_inc_lp_preserved(int level)
 150{
 151	if (level == PG_LEVEL_1G)
 152		cpa_1g_preserved++;
 153	else
 154		cpa_2m_preserved++;
 155}
 156
 157static int cpastats_show(struct seq_file *m, void *p)
 158{
 159	seq_printf(m, "1G pages checked:     %16lu\n", cpa_1g_checked);
 160	seq_printf(m, "1G pages sameprot:    %16lu\n", cpa_1g_sameprot);
 161	seq_printf(m, "1G pages preserved:   %16lu\n", cpa_1g_preserved);
 162	seq_printf(m, "2M pages checked:     %16lu\n", cpa_2m_checked);
 163	seq_printf(m, "2M pages sameprot:    %16lu\n", cpa_2m_sameprot);
 164	seq_printf(m, "2M pages preserved:   %16lu\n", cpa_2m_preserved);
 165	seq_printf(m, "4K pages set-checked: %16lu\n", cpa_4k_install);
 166	return 0;
 167}
 168
 169static int cpastats_open(struct inode *inode, struct file *file)
 170{
 171	return single_open(file, cpastats_show, NULL);
 172}
 173
 174static const struct file_operations cpastats_fops = {
 175	.open		= cpastats_open,
 176	.read		= seq_read,
 177	.llseek		= seq_lseek,
 178	.release	= single_release,
 179};
 180
 181static int __init cpa_stats_init(void)
 182{
 183	debugfs_create_file("cpa_stats", S_IRUSR, arch_debugfs_dir, NULL,
 184			    &cpastats_fops);
 185	return 0;
 186}
 187late_initcall(cpa_stats_init);
 188#else
 189static inline void cpa_inc_1g_checked(void) { }
 190static inline void cpa_inc_2m_checked(void) { }
 191static inline void cpa_inc_4k_install(void) { }
 192static inline void cpa_inc_lp_sameprot(int level) { }
 193static inline void cpa_inc_lp_preserved(int level) { }
 194#endif
 195
 196
 197static inline int
 198within(unsigned long addr, unsigned long start, unsigned long end)
 199{
 200	return addr >= start && addr < end;
 201}
 202
 203static inline int
 204within_inclusive(unsigned long addr, unsigned long start, unsigned long end)
 205{
 206	return addr >= start && addr <= end;
 207}
 208
 209#ifdef CONFIG_X86_64
 210
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 211static inline unsigned long highmap_start_pfn(void)
 212{
 213	return __pa_symbol(_text) >> PAGE_SHIFT;
 214}
 215
 216static inline unsigned long highmap_end_pfn(void)
 217{
 218	/* Do not reference physical address outside the kernel. */
 219	return __pa_symbol(roundup(_brk_end, PMD_SIZE) - 1) >> PAGE_SHIFT;
 220}
 221
 222static bool __cpa_pfn_in_highmap(unsigned long pfn)
 223{
 224	/*
 225	 * Kernel text has an alias mapping at a high address, known
 226	 * here as "highmap".
 227	 */
 228	return within_inclusive(pfn, highmap_start_pfn(), highmap_end_pfn());
 229}
 230
 231#else
 232
 233static bool __cpa_pfn_in_highmap(unsigned long pfn)
 234{
 235	/* There is no highmap on 32-bit */
 236	return false;
 237}
 238
 239#endif
 240
 241/*
 242 * See set_mce_nospec().
 243 *
 244 * Machine check recovery code needs to change cache mode of poisoned pages to
 245 * UC to avoid speculative access logging another error. But passing the
 246 * address of the 1:1 mapping to set_memory_uc() is a fine way to encourage a
 247 * speculative access. So we cheat and flip the top bit of the address. This
 248 * works fine for the code that updates the page tables. But at the end of the
 249 * process we need to flush the TLB and cache and the non-canonical address
 250 * causes a #GP fault when used by the INVLPG and CLFLUSH instructions.
 251 *
 252 * But in the common case we already have a canonical address. This code
 253 * will fix the top bit if needed and is a no-op otherwise.
 254 */
 255static inline unsigned long fix_addr(unsigned long addr)
 256{
 257#ifdef CONFIG_X86_64
 258	return (long)(addr << 1) >> 1;
 259#else
 260	return addr;
 261#endif
 262}
 263
 264static unsigned long __cpa_addr(struct cpa_data *cpa, unsigned long idx)
 265{
 266	if (cpa->flags & CPA_PAGES_ARRAY) {
 267		struct page *page = cpa->pages[idx];
 268
 269		if (unlikely(PageHighMem(page)))
 270			return 0;
 271
 272		return (unsigned long)page_address(page);
 273	}
 274
 275	if (cpa->flags & CPA_ARRAY)
 276		return cpa->vaddr[idx];
 277
 278	return *cpa->vaddr + idx * PAGE_SIZE;
 279}
 280
 281/*
 282 * Flushing functions
 283 */
 284
 285static void clflush_cache_range_opt(void *vaddr, unsigned int size)
 286{
 287	const unsigned long clflush_size = boot_cpu_data.x86_clflush_size;
 288	void *p = (void *)((unsigned long)vaddr & ~(clflush_size - 1));
 289	void *vend = vaddr + size;
 290
 291	if (p >= vend)
 292		return;
 293
 294	for (; p < vend; p += clflush_size)
 295		clflushopt(p);
 296}
 297
 298/**
 299 * clflush_cache_range - flush a cache range with clflush
 300 * @vaddr:	virtual start address
 301 * @size:	number of bytes to flush
 302 *
 303 * CLFLUSHOPT is an unordered instruction which needs fencing with MFENCE or
 304 * SFENCE to avoid ordering issues.
 305 */
 306void clflush_cache_range(void *vaddr, unsigned int size)
 307{
 308	mb();
 309	clflush_cache_range_opt(vaddr, size);
 310	mb();
 311}
 312EXPORT_SYMBOL_GPL(clflush_cache_range);
 313
 314#ifdef CONFIG_ARCH_HAS_PMEM_API
 315void arch_invalidate_pmem(void *addr, size_t size)
 316{
 317	clflush_cache_range(addr, size);
 318}
 319EXPORT_SYMBOL_GPL(arch_invalidate_pmem);
 320#endif
 321
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 322static void __cpa_flush_all(void *arg)
 323{
 324	unsigned long cache = (unsigned long)arg;
 325
 326	/*
 327	 * Flush all to work around Errata in early athlons regarding
 328	 * large page flushing.
 329	 */
 330	__flush_tlb_all();
 331
 332	if (cache && boot_cpu_data.x86 >= 4)
 333		wbinvd();
 334}
 335
 336static void cpa_flush_all(unsigned long cache)
 337{
 338	BUG_ON(irqs_disabled() && !early_boot_irqs_disabled);
 339
 340	on_each_cpu(__cpa_flush_all, (void *) cache, 1);
 341}
 342
 343static void __cpa_flush_tlb(void *data)
 344{
 345	struct cpa_data *cpa = data;
 346	unsigned int i;
 347
 348	for (i = 0; i < cpa->numpages; i++)
 349		flush_tlb_one_kernel(fix_addr(__cpa_addr(cpa, i)));
 350}
 351
 352static void cpa_flush(struct cpa_data *data, int cache)
 353{
 354	struct cpa_data *cpa = data;
 355	unsigned int i;
 356
 357	BUG_ON(irqs_disabled() && !early_boot_irqs_disabled);
 358
 359	if (cache && !static_cpu_has(X86_FEATURE_CLFLUSH)) {
 360		cpa_flush_all(cache);
 361		return;
 362	}
 363
 364	if (cpa->force_flush_all || cpa->numpages > tlb_single_page_flush_ceiling)
 365		flush_tlb_all();
 366	else
 367		on_each_cpu(__cpa_flush_tlb, cpa, 1);
 368
 369	if (!cache)
 370		return;
 371
 372	mb();
 373	for (i = 0; i < cpa->numpages; i++) {
 374		unsigned long addr = __cpa_addr(cpa, i);
 375		unsigned int level;
 376
 377		pte_t *pte = lookup_address(addr, &level);
 378
 379		/*
 380		 * Only flush present addresses:
 381		 */
 382		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
 383			clflush_cache_range_opt((void *)fix_addr(addr), PAGE_SIZE);
 384	}
 385	mb();
 386}
 387
 388static bool overlaps(unsigned long r1_start, unsigned long r1_end,
 389		     unsigned long r2_start, unsigned long r2_end)
 390{
 391	return (r1_start <= r2_end && r1_end >= r2_start) ||
 392		(r2_start <= r1_end && r2_end >= r1_start);
 393}
 394
 395#ifdef CONFIG_PCI_BIOS
 396/*
 397 * The BIOS area between 640k and 1Mb needs to be executable for PCI BIOS
 398 * based config access (CONFIG_PCI_GOBIOS) support.
 399 */
 400#define BIOS_PFN	PFN_DOWN(BIOS_BEGIN)
 401#define BIOS_PFN_END	PFN_DOWN(BIOS_END - 1)
 402
 403static pgprotval_t protect_pci_bios(unsigned long spfn, unsigned long epfn)
 404{
 405	if (pcibios_enabled && overlaps(spfn, epfn, BIOS_PFN, BIOS_PFN_END))
 406		return _PAGE_NX;
 407	return 0;
 408}
 409#else
 410static pgprotval_t protect_pci_bios(unsigned long spfn, unsigned long epfn)
 411{
 412	return 0;
 413}
 414#endif
 415
 416/*
 417 * The .rodata section needs to be read-only. Using the pfn catches all
 418 * aliases.  This also includes __ro_after_init, so do not enforce until
 419 * kernel_set_to_readonly is true.
 420 */
 421static pgprotval_t protect_rodata(unsigned long spfn, unsigned long epfn)
 422{
 423	unsigned long epfn_ro, spfn_ro = PFN_DOWN(__pa_symbol(__start_rodata));
 424
 425	/*
 426	 * Note: __end_rodata is at page aligned and not inclusive, so
 427	 * subtract 1 to get the last enforced PFN in the rodata area.
 428	 */
 429	epfn_ro = PFN_DOWN(__pa_symbol(__end_rodata)) - 1;
 430
 431	if (kernel_set_to_readonly && overlaps(spfn, epfn, spfn_ro, epfn_ro))
 432		return _PAGE_RW;
 433	return 0;
 434}
 435
 436/*
 437 * Protect kernel text against becoming non executable by forbidding
 438 * _PAGE_NX.  This protects only the high kernel mapping (_text -> _etext)
 439 * out of which the kernel actually executes.  Do not protect the low
 440 * mapping.
 441 *
 442 * This does not cover __inittext since that is gone after boot.
 443 */
 444static pgprotval_t protect_kernel_text(unsigned long start, unsigned long end)
 445{
 446	unsigned long t_end = (unsigned long)_etext - 1;
 447	unsigned long t_start = (unsigned long)_text;
 448
 449	if (overlaps(start, end, t_start, t_end))
 450		return _PAGE_NX;
 451	return 0;
 452}
 453
 454#if defined(CONFIG_X86_64)
 455/*
 456 * Once the kernel maps the text as RO (kernel_set_to_readonly is set),
 457 * kernel text mappings for the large page aligned text, rodata sections
 458 * will be always read-only. For the kernel identity mappings covering the
 459 * holes caused by this alignment can be anything that user asks.
 460 *
 461 * This will preserve the large page mappings for kernel text/data at no
 462 * extra cost.
 463 */
 464static pgprotval_t protect_kernel_text_ro(unsigned long start,
 465					  unsigned long end)
 466{
 467	unsigned long t_end = (unsigned long)__end_rodata_hpage_align - 1;
 468	unsigned long t_start = (unsigned long)_text;
 469	unsigned int level;
 470
 471	if (!kernel_set_to_readonly || !overlaps(start, end, t_start, t_end))
 472		return 0;
 473	/*
 474	 * Don't enforce the !RW mapping for the kernel text mapping, if
 475	 * the current mapping is already using small page mapping.  No
 476	 * need to work hard to preserve large page mappings in this case.
 477	 *
 478	 * This also fixes the Linux Xen paravirt guest boot failure caused
 479	 * by unexpected read-only mappings for kernel identity
 480	 * mappings. In this paravirt guest case, the kernel text mapping
 481	 * and the kernel identity mapping share the same page-table pages,
 482	 * so the protections for kernel text and identity mappings have to
 483	 * be the same.
 484	 */
 485	if (lookup_address(start, &level) && (level != PG_LEVEL_4K))
 486		return _PAGE_RW;
 487	return 0;
 488}
 489#else
 490static pgprotval_t protect_kernel_text_ro(unsigned long start,
 491					  unsigned long end)
 492{
 493	return 0;
 494}
 495#endif
 496
 497static inline bool conflicts(pgprot_t prot, pgprotval_t val)
 498{
 499	return (pgprot_val(prot) & ~val) != pgprot_val(prot);
 500}
 501
 502static inline void check_conflict(int warnlvl, pgprot_t prot, pgprotval_t val,
 503				  unsigned long start, unsigned long end,
 504				  unsigned long pfn, const char *txt)
 505{
 506	static const char *lvltxt[] = {
 507		[CPA_CONFLICT]	= "conflict",
 508		[CPA_PROTECT]	= "protect",
 509		[CPA_DETECT]	= "detect",
 510	};
 511
 512	if (warnlvl > cpa_warn_level || !conflicts(prot, val))
 513		return;
 514
 515	pr_warn("CPA %8s %10s: 0x%016lx - 0x%016lx PFN %lx req %016llx prevent %016llx\n",
 516		lvltxt[warnlvl], txt, start, end, pfn, (unsigned long long)pgprot_val(prot),
 517		(unsigned long long)val);
 518}
 519
 520/*
 521 * Certain areas of memory on x86 require very specific protection flags,
 522 * for example the BIOS area or kernel text. Callers don't always get this
 523 * right (again, ioremap() on BIOS memory is not uncommon) so this function
 524 * checks and fixes these known static required protection bits.
 525 */
 526static inline pgprot_t static_protections(pgprot_t prot, unsigned long start,
 527					  unsigned long pfn, unsigned long npg,
 528					  unsigned long lpsize, int warnlvl)
 529{
 530	pgprotval_t forbidden, res;
 531	unsigned long end;
 532
 533	/*
 534	 * There is no point in checking RW/NX conflicts when the requested
 535	 * mapping is setting the page !PRESENT.
 536	 */
 537	if (!(pgprot_val(prot) & _PAGE_PRESENT))
 538		return prot;
 539
 540	/* Operate on the virtual address */
 541	end = start + npg * PAGE_SIZE - 1;
 542
 543	res = protect_kernel_text(start, end);
 544	check_conflict(warnlvl, prot, res, start, end, pfn, "Text NX");
 545	forbidden = res;
 546
 547	/*
 548	 * Special case to preserve a large page. If the change spawns the
 549	 * full large page mapping then there is no point to split it
 550	 * up. Happens with ftrace and is going to be removed once ftrace
 551	 * switched to text_poke().
 552	 */
 553	if (lpsize != (npg * PAGE_SIZE) || (start & (lpsize - 1))) {
 554		res = protect_kernel_text_ro(start, end);
 555		check_conflict(warnlvl, prot, res, start, end, pfn, "Text RO");
 556		forbidden |= res;
 557	}
 558
 559	/* Check the PFN directly */
 560	res = protect_pci_bios(pfn, pfn + npg - 1);
 561	check_conflict(warnlvl, prot, res, start, end, pfn, "PCIBIOS NX");
 562	forbidden |= res;
 563
 564	res = protect_rodata(pfn, pfn + npg - 1);
 565	check_conflict(warnlvl, prot, res, start, end, pfn, "Rodata RO");
 566	forbidden |= res;
 567
 568	return __pgprot(pgprot_val(prot) & ~forbidden);
 569}
 570
 571/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 572 * Lookup the page table entry for a virtual address in a specific pgd.
 573 * Return a pointer to the entry and the level of the mapping.
 574 */
 575pte_t *lookup_address_in_pgd(pgd_t *pgd, unsigned long address,
 576			     unsigned int *level)
 577{
 578	p4d_t *p4d;
 579	pud_t *pud;
 580	pmd_t *pmd;
 581
 582	*level = PG_LEVEL_NONE;
 583
 584	if (pgd_none(*pgd))
 585		return NULL;
 586
 587	p4d = p4d_offset(pgd, address);
 588	if (p4d_none(*p4d))
 589		return NULL;
 590
 591	*level = PG_LEVEL_512G;
 592	if (p4d_large(*p4d) || !p4d_present(*p4d))
 593		return (pte_t *)p4d;
 594
 595	pud = pud_offset(p4d, address);
 596	if (pud_none(*pud))
 597		return NULL;
 598
 599	*level = PG_LEVEL_1G;
 600	if (pud_large(*pud) || !pud_present(*pud))
 601		return (pte_t *)pud;
 602
 603	pmd = pmd_offset(pud, address);
 604	if (pmd_none(*pmd))
 605		return NULL;
 606
 607	*level = PG_LEVEL_2M;
 608	if (pmd_large(*pmd) || !pmd_present(*pmd))
 609		return (pte_t *)pmd;
 610
 611	*level = PG_LEVEL_4K;
 612
 613	return pte_offset_kernel(pmd, address);
 614}
 615
 616/*
 617 * Lookup the page table entry for a virtual address. Return a pointer
 618 * to the entry and the level of the mapping.
 619 *
 620 * Note: We return pud and pmd either when the entry is marked large
 621 * or when the present bit is not set. Otherwise we would return a
 622 * pointer to a nonexisting mapping.
 623 */
 624pte_t *lookup_address(unsigned long address, unsigned int *level)
 625{
 626	return lookup_address_in_pgd(pgd_offset_k(address), address, level);
 627}
 628EXPORT_SYMBOL_GPL(lookup_address);
 629
 630/*
 631 * Lookup the page table entry for a virtual address in a given mm. Return a
 632 * pointer to the entry and the level of the mapping.
 633 */
 634pte_t *lookup_address_in_mm(struct mm_struct *mm, unsigned long address,
 635			    unsigned int *level)
 636{
 637	return lookup_address_in_pgd(pgd_offset(mm, address), address, level);
 638}
 639EXPORT_SYMBOL_GPL(lookup_address_in_mm);
 640
 641static pte_t *_lookup_address_cpa(struct cpa_data *cpa, unsigned long address,
 642				  unsigned int *level)
 643{
 644	if (cpa->pgd)
 645		return lookup_address_in_pgd(cpa->pgd + pgd_index(address),
 646					       address, level);
 647
 648	return lookup_address(address, level);
 649}
 650
 651/*
 652 * Lookup the PMD entry for a virtual address. Return a pointer to the entry
 653 * or NULL if not present.
 654 */
 655pmd_t *lookup_pmd_address(unsigned long address)
 656{
 657	pgd_t *pgd;
 658	p4d_t *p4d;
 659	pud_t *pud;
 660
 661	pgd = pgd_offset_k(address);
 662	if (pgd_none(*pgd))
 663		return NULL;
 664
 665	p4d = p4d_offset(pgd, address);
 666	if (p4d_none(*p4d) || p4d_large(*p4d) || !p4d_present(*p4d))
 667		return NULL;
 668
 669	pud = pud_offset(p4d, address);
 670	if (pud_none(*pud) || pud_large(*pud) || !pud_present(*pud))
 671		return NULL;
 672
 673	return pmd_offset(pud, address);
 674}
 675
 676/*
 677 * This is necessary because __pa() does not work on some
 678 * kinds of memory, like vmalloc() or the alloc_remap()
 679 * areas on 32-bit NUMA systems.  The percpu areas can
 680 * end up in this kind of memory, for instance.
 681 *
 682 * This could be optimized, but it is only intended to be
 683 * used at inititalization time, and keeping it
 684 * unoptimized should increase the testing coverage for
 685 * the more obscure platforms.
 
 
 
 
 686 */
 687phys_addr_t slow_virt_to_phys(void *__virt_addr)
 688{
 689	unsigned long virt_addr = (unsigned long)__virt_addr;
 690	phys_addr_t phys_addr;
 691	unsigned long offset;
 692	enum pg_level level;
 693	pte_t *pte;
 694
 695	pte = lookup_address(virt_addr, &level);
 696	BUG_ON(!pte);
 697
 698	/*
 699	 * pXX_pfn() returns unsigned long, which must be cast to phys_addr_t
 700	 * before being left-shifted PAGE_SHIFT bits -- this trick is to
 701	 * make 32-PAE kernel work correctly.
 702	 */
 703	switch (level) {
 704	case PG_LEVEL_1G:
 705		phys_addr = (phys_addr_t)pud_pfn(*(pud_t *)pte) << PAGE_SHIFT;
 706		offset = virt_addr & ~PUD_PAGE_MASK;
 707		break;
 708	case PG_LEVEL_2M:
 709		phys_addr = (phys_addr_t)pmd_pfn(*(pmd_t *)pte) << PAGE_SHIFT;
 710		offset = virt_addr & ~PMD_PAGE_MASK;
 711		break;
 712	default:
 713		phys_addr = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
 714		offset = virt_addr & ~PAGE_MASK;
 715	}
 716
 717	return (phys_addr_t)(phys_addr | offset);
 718}
 719EXPORT_SYMBOL_GPL(slow_virt_to_phys);
 720
 721/*
 722 * Set the new pmd in all the pgds we know about:
 723 */
 724static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
 725{
 726	/* change init_mm */
 727	set_pte_atomic(kpte, pte);
 728#ifdef CONFIG_X86_32
 729	if (!SHARED_KERNEL_PMD) {
 730		struct page *page;
 731
 732		list_for_each_entry(page, &pgd_list, lru) {
 733			pgd_t *pgd;
 734			p4d_t *p4d;
 735			pud_t *pud;
 736			pmd_t *pmd;
 737
 738			pgd = (pgd_t *)page_address(page) + pgd_index(address);
 739			p4d = p4d_offset(pgd, address);
 740			pud = pud_offset(p4d, address);
 741			pmd = pmd_offset(pud, address);
 742			set_pte_atomic((pte_t *)pmd, pte);
 743		}
 744	}
 745#endif
 746}
 747
 748static pgprot_t pgprot_clear_protnone_bits(pgprot_t prot)
 749{
 750	/*
 751	 * _PAGE_GLOBAL means "global page" for present PTEs.
 752	 * But, it is also used to indicate _PAGE_PROTNONE
 753	 * for non-present PTEs.
 754	 *
 755	 * This ensures that a _PAGE_GLOBAL PTE going from
 756	 * present to non-present is not confused as
 757	 * _PAGE_PROTNONE.
 758	 */
 759	if (!(pgprot_val(prot) & _PAGE_PRESENT))
 760		pgprot_val(prot) &= ~_PAGE_GLOBAL;
 761
 762	return prot;
 763}
 764
 765static int __should_split_large_page(pte_t *kpte, unsigned long address,
 766				     struct cpa_data *cpa)
 767{
 768	unsigned long numpages, pmask, psize, lpaddr, pfn, old_pfn;
 769	pgprot_t old_prot, new_prot, req_prot, chk_prot;
 770	pte_t new_pte, *tmp;
 771	enum pg_level level;
 772
 773	/*
 774	 * Check for races, another CPU might have split this page
 775	 * up already:
 776	 */
 777	tmp = _lookup_address_cpa(cpa, address, &level);
 778	if (tmp != kpte)
 779		return 1;
 780
 781	switch (level) {
 782	case PG_LEVEL_2M:
 783		old_prot = pmd_pgprot(*(pmd_t *)kpte);
 784		old_pfn = pmd_pfn(*(pmd_t *)kpte);
 785		cpa_inc_2m_checked();
 786		break;
 787	case PG_LEVEL_1G:
 788		old_prot = pud_pgprot(*(pud_t *)kpte);
 789		old_pfn = pud_pfn(*(pud_t *)kpte);
 790		cpa_inc_1g_checked();
 791		break;
 792	default:
 793		return -EINVAL;
 794	}
 795
 796	psize = page_level_size(level);
 797	pmask = page_level_mask(level);
 798
 799	/*
 800	 * Calculate the number of pages, which fit into this large
 801	 * page starting at address:
 802	 */
 803	lpaddr = (address + psize) & pmask;
 804	numpages = (lpaddr - address) >> PAGE_SHIFT;
 805	if (numpages < cpa->numpages)
 806		cpa->numpages = numpages;
 807
 808	/*
 809	 * We are safe now. Check whether the new pgprot is the same:
 810	 * Convert protection attributes to 4k-format, as cpa->mask* are set
 811	 * up accordingly.
 812	 */
 813
 814	/* Clear PSE (aka _PAGE_PAT) and move PAT bit to correct position */
 815	req_prot = pgprot_large_2_4k(old_prot);
 816
 817	pgprot_val(req_prot) &= ~pgprot_val(cpa->mask_clr);
 818	pgprot_val(req_prot) |= pgprot_val(cpa->mask_set);
 819
 820	/*
 821	 * req_prot is in format of 4k pages. It must be converted to large
 822	 * page format: the caching mode includes the PAT bit located at
 823	 * different bit positions in the two formats.
 824	 */
 825	req_prot = pgprot_4k_2_large(req_prot);
 826	req_prot = pgprot_clear_protnone_bits(req_prot);
 827	if (pgprot_val(req_prot) & _PAGE_PRESENT)
 828		pgprot_val(req_prot) |= _PAGE_PSE;
 829
 830	/*
 831	 * old_pfn points to the large page base pfn. So we need to add the
 832	 * offset of the virtual address:
 833	 */
 834	pfn = old_pfn + ((address & (psize - 1)) >> PAGE_SHIFT);
 835	cpa->pfn = pfn;
 836
 837	/*
 838	 * Calculate the large page base address and the number of 4K pages
 839	 * in the large page
 840	 */
 841	lpaddr = address & pmask;
 842	numpages = psize >> PAGE_SHIFT;
 843
 844	/*
 845	 * Sanity check that the existing mapping is correct versus the static
 846	 * protections. static_protections() guards against !PRESENT, so no
 847	 * extra conditional required here.
 848	 */
 849	chk_prot = static_protections(old_prot, lpaddr, old_pfn, numpages,
 850				      psize, CPA_CONFLICT);
 851
 852	if (WARN_ON_ONCE(pgprot_val(chk_prot) != pgprot_val(old_prot))) {
 853		/*
 854		 * Split the large page and tell the split code to
 855		 * enforce static protections.
 856		 */
 857		cpa->force_static_prot = 1;
 858		return 1;
 859	}
 860
 861	/*
 862	 * Optimization: If the requested pgprot is the same as the current
 863	 * pgprot, then the large page can be preserved and no updates are
 864	 * required independent of alignment and length of the requested
 865	 * range. The above already established that the current pgprot is
 866	 * correct, which in consequence makes the requested pgprot correct
 867	 * as well if it is the same. The static protection scan below will
 868	 * not come to a different conclusion.
 869	 */
 870	if (pgprot_val(req_prot) == pgprot_val(old_prot)) {
 871		cpa_inc_lp_sameprot(level);
 872		return 0;
 873	}
 874
 875	/*
 876	 * If the requested range does not cover the full page, split it up
 877	 */
 878	if (address != lpaddr || cpa->numpages != numpages)
 879		return 1;
 880
 881	/*
 882	 * Check whether the requested pgprot is conflicting with a static
 883	 * protection requirement in the large page.
 884	 */
 885	new_prot = static_protections(req_prot, lpaddr, old_pfn, numpages,
 886				      psize, CPA_DETECT);
 887
 
 
 888	/*
 889	 * If there is a conflict, split the large page.
 890	 *
 891	 * There used to be a 4k wise evaluation trying really hard to
 892	 * preserve the large pages, but experimentation has shown, that this
 893	 * does not help at all. There might be corner cases which would
 894	 * preserve one large page occasionally, but it's really not worth the
 895	 * extra code and cycles for the common case.
 896	 */
 897	if (pgprot_val(req_prot) != pgprot_val(new_prot))
 898		return 1;
 899
 900	/* All checks passed. Update the large page mapping. */
 901	new_pte = pfn_pte(old_pfn, new_prot);
 902	__set_pmd_pte(kpte, address, new_pte);
 903	cpa->flags |= CPA_FLUSHTLB;
 904	cpa_inc_lp_preserved(level);
 905	return 0;
 906}
 907
 908static int should_split_large_page(pte_t *kpte, unsigned long address,
 909				   struct cpa_data *cpa)
 910{
 911	int do_split;
 912
 913	if (cpa->force_split)
 914		return 1;
 915
 916	spin_lock(&pgd_lock);
 917	do_split = __should_split_large_page(kpte, address, cpa);
 918	spin_unlock(&pgd_lock);
 919
 920	return do_split;
 921}
 922
 923static void split_set_pte(struct cpa_data *cpa, pte_t *pte, unsigned long pfn,
 924			  pgprot_t ref_prot, unsigned long address,
 925			  unsigned long size)
 926{
 927	unsigned int npg = PFN_DOWN(size);
 928	pgprot_t prot;
 929
 930	/*
 931	 * If should_split_large_page() discovered an inconsistent mapping,
 932	 * remove the invalid protection in the split mapping.
 933	 */
 934	if (!cpa->force_static_prot)
 935		goto set;
 936
 937	/* Hand in lpsize = 0 to enforce the protection mechanism */
 938	prot = static_protections(ref_prot, address, pfn, npg, 0, CPA_PROTECT);
 939
 940	if (pgprot_val(prot) == pgprot_val(ref_prot))
 941		goto set;
 942
 943	/*
 944	 * If this is splitting a PMD, fix it up. PUD splits cannot be
 945	 * fixed trivially as that would require to rescan the newly
 946	 * installed PMD mappings after returning from split_large_page()
 947	 * so an eventual further split can allocate the necessary PTE
 948	 * pages. Warn for now and revisit it in case this actually
 949	 * happens.
 950	 */
 951	if (size == PAGE_SIZE)
 952		ref_prot = prot;
 953	else
 954		pr_warn_once("CPA: Cannot fixup static protections for PUD split\n");
 955set:
 956	set_pte(pte, pfn_pte(pfn, ref_prot));
 957}
 958
 959static int
 960__split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
 961		   struct page *base)
 962{
 963	unsigned long lpaddr, lpinc, ref_pfn, pfn, pfninc = 1;
 964	pte_t *pbase = (pte_t *)page_address(base);
 965	unsigned int i, level;
 966	pgprot_t ref_prot;
 967	pte_t *tmp;
 968
 969	spin_lock(&pgd_lock);
 970	/*
 971	 * Check for races, another CPU might have split this page
 972	 * up for us already:
 973	 */
 974	tmp = _lookup_address_cpa(cpa, address, &level);
 975	if (tmp != kpte) {
 976		spin_unlock(&pgd_lock);
 977		return 1;
 978	}
 979
 980	paravirt_alloc_pte(&init_mm, page_to_pfn(base));
 981
 982	switch (level) {
 983	case PG_LEVEL_2M:
 984		ref_prot = pmd_pgprot(*(pmd_t *)kpte);
 985		/*
 986		 * Clear PSE (aka _PAGE_PAT) and move
 987		 * PAT bit to correct position.
 988		 */
 989		ref_prot = pgprot_large_2_4k(ref_prot);
 990		ref_pfn = pmd_pfn(*(pmd_t *)kpte);
 991		lpaddr = address & PMD_MASK;
 992		lpinc = PAGE_SIZE;
 993		break;
 994
 995	case PG_LEVEL_1G:
 996		ref_prot = pud_pgprot(*(pud_t *)kpte);
 997		ref_pfn = pud_pfn(*(pud_t *)kpte);
 998		pfninc = PMD_PAGE_SIZE >> PAGE_SHIFT;
 999		lpaddr = address & PUD_MASK;
1000		lpinc = PMD_SIZE;
1001		/*
1002		 * Clear the PSE flags if the PRESENT flag is not set
1003		 * otherwise pmd_present/pmd_huge will return true
1004		 * even on a non present pmd.
1005		 */
1006		if (!(pgprot_val(ref_prot) & _PAGE_PRESENT))
1007			pgprot_val(ref_prot) &= ~_PAGE_PSE;
1008		break;
1009
1010	default:
1011		spin_unlock(&pgd_lock);
1012		return 1;
1013	}
1014
1015	ref_prot = pgprot_clear_protnone_bits(ref_prot);
1016
1017	/*
1018	 * Get the target pfn from the original entry:
1019	 */
1020	pfn = ref_pfn;
1021	for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc, lpaddr += lpinc)
1022		split_set_pte(cpa, pbase + i, pfn, ref_prot, lpaddr, lpinc);
1023
1024	if (virt_addr_valid(address)) {
1025		unsigned long pfn = PFN_DOWN(__pa(address));
1026
1027		if (pfn_range_is_mapped(pfn, pfn + 1))
1028			split_page_count(level);
1029	}
1030
1031	/*
1032	 * Install the new, split up pagetable.
1033	 *
1034	 * We use the standard kernel pagetable protections for the new
1035	 * pagetable protections, the actual ptes set above control the
1036	 * primary protection behavior:
1037	 */
1038	__set_pmd_pte(kpte, address, mk_pte(base, __pgprot(_KERNPG_TABLE)));
1039
1040	/*
1041	 * Do a global flush tlb after splitting the large page
1042	 * and before we do the actual change page attribute in the PTE.
1043	 *
1044	 * Without this, we violate the TLB application note, that says:
1045	 * "The TLBs may contain both ordinary and large-page
1046	 *  translations for a 4-KByte range of linear addresses. This
1047	 *  may occur if software modifies the paging structures so that
1048	 *  the page size used for the address range changes. If the two
1049	 *  translations differ with respect to page frame or attributes
1050	 *  (e.g., permissions), processor behavior is undefined and may
1051	 *  be implementation-specific."
1052	 *
1053	 * We do this global tlb flush inside the cpa_lock, so that we
1054	 * don't allow any other cpu, with stale tlb entries change the
1055	 * page attribute in parallel, that also falls into the
1056	 * just split large page entry.
1057	 */
1058	flush_tlb_all();
1059	spin_unlock(&pgd_lock);
1060
1061	return 0;
1062}
1063
1064static int split_large_page(struct cpa_data *cpa, pte_t *kpte,
1065			    unsigned long address)
1066{
1067	struct page *base;
1068
1069	if (!debug_pagealloc_enabled())
1070		spin_unlock(&cpa_lock);
1071	base = alloc_pages(GFP_KERNEL, 0);
1072	if (!debug_pagealloc_enabled())
1073		spin_lock(&cpa_lock);
1074	if (!base)
1075		return -ENOMEM;
1076
1077	if (__split_large_page(cpa, kpte, address, base))
1078		__free_page(base);
1079
1080	return 0;
1081}
1082
1083static bool try_to_free_pte_page(pte_t *pte)
1084{
1085	int i;
1086
1087	for (i = 0; i < PTRS_PER_PTE; i++)
1088		if (!pte_none(pte[i]))
1089			return false;
1090
1091	free_page((unsigned long)pte);
1092	return true;
1093}
1094
1095static bool try_to_free_pmd_page(pmd_t *pmd)
1096{
1097	int i;
1098
1099	for (i = 0; i < PTRS_PER_PMD; i++)
1100		if (!pmd_none(pmd[i]))
1101			return false;
1102
1103	free_page((unsigned long)pmd);
1104	return true;
1105}
1106
1107static bool unmap_pte_range(pmd_t *pmd, unsigned long start, unsigned long end)
1108{
1109	pte_t *pte = pte_offset_kernel(pmd, start);
1110
1111	while (start < end) {
1112		set_pte(pte, __pte(0));
1113
1114		start += PAGE_SIZE;
1115		pte++;
1116	}
1117
1118	if (try_to_free_pte_page((pte_t *)pmd_page_vaddr(*pmd))) {
1119		pmd_clear(pmd);
1120		return true;
1121	}
1122	return false;
1123}
1124
1125static void __unmap_pmd_range(pud_t *pud, pmd_t *pmd,
1126			      unsigned long start, unsigned long end)
1127{
1128	if (unmap_pte_range(pmd, start, end))
1129		if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud)))
1130			pud_clear(pud);
1131}
1132
1133static void unmap_pmd_range(pud_t *pud, unsigned long start, unsigned long end)
1134{
1135	pmd_t *pmd = pmd_offset(pud, start);
1136
1137	/*
1138	 * Not on a 2MB page boundary?
1139	 */
1140	if (start & (PMD_SIZE - 1)) {
1141		unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
1142		unsigned long pre_end = min_t(unsigned long, end, next_page);
1143
1144		__unmap_pmd_range(pud, pmd, start, pre_end);
1145
1146		start = pre_end;
1147		pmd++;
1148	}
1149
1150	/*
1151	 * Try to unmap in 2M chunks.
1152	 */
1153	while (end - start >= PMD_SIZE) {
1154		if (pmd_large(*pmd))
1155			pmd_clear(pmd);
1156		else
1157			__unmap_pmd_range(pud, pmd, start, start + PMD_SIZE);
1158
1159		start += PMD_SIZE;
1160		pmd++;
1161	}
1162
1163	/*
1164	 * 4K leftovers?
1165	 */
1166	if (start < end)
1167		return __unmap_pmd_range(pud, pmd, start, end);
1168
1169	/*
1170	 * Try again to free the PMD page if haven't succeeded above.
1171	 */
1172	if (!pud_none(*pud))
1173		if (try_to_free_pmd_page((pmd_t *)pud_page_vaddr(*pud)))
1174			pud_clear(pud);
1175}
1176
1177static void unmap_pud_range(p4d_t *p4d, unsigned long start, unsigned long end)
1178{
1179	pud_t *pud = pud_offset(p4d, start);
1180
1181	/*
1182	 * Not on a GB page boundary?
1183	 */
1184	if (start & (PUD_SIZE - 1)) {
1185		unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1186		unsigned long pre_end	= min_t(unsigned long, end, next_page);
1187
1188		unmap_pmd_range(pud, start, pre_end);
1189
1190		start = pre_end;
1191		pud++;
1192	}
1193
1194	/*
1195	 * Try to unmap in 1G chunks?
1196	 */
1197	while (end - start >= PUD_SIZE) {
1198
1199		if (pud_large(*pud))
1200			pud_clear(pud);
1201		else
1202			unmap_pmd_range(pud, start, start + PUD_SIZE);
1203
1204		start += PUD_SIZE;
1205		pud++;
1206	}
1207
1208	/*
1209	 * 2M leftovers?
1210	 */
1211	if (start < end)
1212		unmap_pmd_range(pud, start, end);
1213
1214	/*
1215	 * No need to try to free the PUD page because we'll free it in
1216	 * populate_pgd's error path
1217	 */
1218}
1219
1220static int alloc_pte_page(pmd_t *pmd)
1221{
1222	pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
1223	if (!pte)
1224		return -1;
1225
1226	set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
1227	return 0;
1228}
1229
1230static int alloc_pmd_page(pud_t *pud)
1231{
1232	pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
1233	if (!pmd)
1234		return -1;
1235
1236	set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
1237	return 0;
1238}
1239
1240static void populate_pte(struct cpa_data *cpa,
1241			 unsigned long start, unsigned long end,
1242			 unsigned num_pages, pmd_t *pmd, pgprot_t pgprot)
1243{
1244	pte_t *pte;
1245
1246	pte = pte_offset_kernel(pmd, start);
1247
1248	pgprot = pgprot_clear_protnone_bits(pgprot);
1249
1250	while (num_pages-- && start < end) {
1251		set_pte(pte, pfn_pte(cpa->pfn, pgprot));
1252
1253		start	 += PAGE_SIZE;
1254		cpa->pfn++;
1255		pte++;
1256	}
1257}
1258
1259static long populate_pmd(struct cpa_data *cpa,
1260			 unsigned long start, unsigned long end,
1261			 unsigned num_pages, pud_t *pud, pgprot_t pgprot)
1262{
1263	long cur_pages = 0;
1264	pmd_t *pmd;
1265	pgprot_t pmd_pgprot;
1266
1267	/*
1268	 * Not on a 2M boundary?
1269	 */
1270	if (start & (PMD_SIZE - 1)) {
1271		unsigned long pre_end = start + (num_pages << PAGE_SHIFT);
1272		unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
1273
1274		pre_end   = min_t(unsigned long, pre_end, next_page);
1275		cur_pages = (pre_end - start) >> PAGE_SHIFT;
1276		cur_pages = min_t(unsigned int, num_pages, cur_pages);
1277
1278		/*
1279		 * Need a PTE page?
1280		 */
1281		pmd = pmd_offset(pud, start);
1282		if (pmd_none(*pmd))
1283			if (alloc_pte_page(pmd))
1284				return -1;
1285
1286		populate_pte(cpa, start, pre_end, cur_pages, pmd, pgprot);
1287
1288		start = pre_end;
1289	}
1290
1291	/*
1292	 * We mapped them all?
1293	 */
1294	if (num_pages == cur_pages)
1295		return cur_pages;
1296
1297	pmd_pgprot = pgprot_4k_2_large(pgprot);
1298
1299	while (end - start >= PMD_SIZE) {
1300
1301		/*
1302		 * We cannot use a 1G page so allocate a PMD page if needed.
1303		 */
1304		if (pud_none(*pud))
1305			if (alloc_pmd_page(pud))
1306				return -1;
1307
1308		pmd = pmd_offset(pud, start);
1309
1310		set_pmd(pmd, pmd_mkhuge(pfn_pmd(cpa->pfn,
1311					canon_pgprot(pmd_pgprot))));
1312
1313		start	  += PMD_SIZE;
1314		cpa->pfn  += PMD_SIZE >> PAGE_SHIFT;
1315		cur_pages += PMD_SIZE >> PAGE_SHIFT;
1316	}
1317
1318	/*
1319	 * Map trailing 4K pages.
1320	 */
1321	if (start < end) {
1322		pmd = pmd_offset(pud, start);
1323		if (pmd_none(*pmd))
1324			if (alloc_pte_page(pmd))
1325				return -1;
1326
1327		populate_pte(cpa, start, end, num_pages - cur_pages,
1328			     pmd, pgprot);
1329	}
1330	return num_pages;
1331}
1332
1333static int populate_pud(struct cpa_data *cpa, unsigned long start, p4d_t *p4d,
1334			pgprot_t pgprot)
1335{
1336	pud_t *pud;
1337	unsigned long end;
1338	long cur_pages = 0;
1339	pgprot_t pud_pgprot;
1340
1341	end = start + (cpa->numpages << PAGE_SHIFT);
1342
1343	/*
1344	 * Not on a Gb page boundary? => map everything up to it with
1345	 * smaller pages.
1346	 */
1347	if (start & (PUD_SIZE - 1)) {
1348		unsigned long pre_end;
1349		unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1350
1351		pre_end   = min_t(unsigned long, end, next_page);
1352		cur_pages = (pre_end - start) >> PAGE_SHIFT;
1353		cur_pages = min_t(int, (int)cpa->numpages, cur_pages);
1354
1355		pud = pud_offset(p4d, start);
1356
1357		/*
1358		 * Need a PMD page?
1359		 */
1360		if (pud_none(*pud))
1361			if (alloc_pmd_page(pud))
1362				return -1;
1363
1364		cur_pages = populate_pmd(cpa, start, pre_end, cur_pages,
1365					 pud, pgprot);
1366		if (cur_pages < 0)
1367			return cur_pages;
1368
1369		start = pre_end;
1370	}
1371
1372	/* We mapped them all? */
1373	if (cpa->numpages == cur_pages)
1374		return cur_pages;
1375
1376	pud = pud_offset(p4d, start);
1377	pud_pgprot = pgprot_4k_2_large(pgprot);
1378
1379	/*
1380	 * Map everything starting from the Gb boundary, possibly with 1G pages
1381	 */
1382	while (boot_cpu_has(X86_FEATURE_GBPAGES) && end - start >= PUD_SIZE) {
1383		set_pud(pud, pud_mkhuge(pfn_pud(cpa->pfn,
1384				   canon_pgprot(pud_pgprot))));
1385
1386		start	  += PUD_SIZE;
1387		cpa->pfn  += PUD_SIZE >> PAGE_SHIFT;
1388		cur_pages += PUD_SIZE >> PAGE_SHIFT;
1389		pud++;
1390	}
1391
1392	/* Map trailing leftover */
1393	if (start < end) {
1394		long tmp;
1395
1396		pud = pud_offset(p4d, start);
1397		if (pud_none(*pud))
1398			if (alloc_pmd_page(pud))
1399				return -1;
1400
1401		tmp = populate_pmd(cpa, start, end, cpa->numpages - cur_pages,
1402				   pud, pgprot);
1403		if (tmp < 0)
1404			return cur_pages;
1405
1406		cur_pages += tmp;
1407	}
1408	return cur_pages;
1409}
1410
1411/*
1412 * Restrictions for kernel page table do not necessarily apply when mapping in
1413 * an alternate PGD.
1414 */
1415static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
1416{
1417	pgprot_t pgprot = __pgprot(_KERNPG_TABLE);
1418	pud_t *pud = NULL;	/* shut up gcc */
1419	p4d_t *p4d;
1420	pgd_t *pgd_entry;
1421	long ret;
1422
1423	pgd_entry = cpa->pgd + pgd_index(addr);
1424
1425	if (pgd_none(*pgd_entry)) {
1426		p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
1427		if (!p4d)
1428			return -1;
1429
1430		set_pgd(pgd_entry, __pgd(__pa(p4d) | _KERNPG_TABLE));
1431	}
1432
1433	/*
1434	 * Allocate a PUD page and hand it down for mapping.
1435	 */
1436	p4d = p4d_offset(pgd_entry, addr);
1437	if (p4d_none(*p4d)) {
1438		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
1439		if (!pud)
1440			return -1;
1441
1442		set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
1443	}
1444
1445	pgprot_val(pgprot) &= ~pgprot_val(cpa->mask_clr);
1446	pgprot_val(pgprot) |=  pgprot_val(cpa->mask_set);
1447
1448	ret = populate_pud(cpa, addr, p4d, pgprot);
1449	if (ret < 0) {
1450		/*
1451		 * Leave the PUD page in place in case some other CPU or thread
1452		 * already found it, but remove any useless entries we just
1453		 * added to it.
1454		 */
1455		unmap_pud_range(p4d, addr,
1456				addr + (cpa->numpages << PAGE_SHIFT));
1457		return ret;
1458	}
1459
1460	cpa->numpages = ret;
1461	return 0;
1462}
1463
1464static int __cpa_process_fault(struct cpa_data *cpa, unsigned long vaddr,
1465			       int primary)
1466{
1467	if (cpa->pgd) {
1468		/*
1469		 * Right now, we only execute this code path when mapping
1470		 * the EFI virtual memory map regions, no other users
1471		 * provide a ->pgd value. This may change in the future.
1472		 */
1473		return populate_pgd(cpa, vaddr);
1474	}
1475
1476	/*
1477	 * Ignore all non primary paths.
1478	 */
1479	if (!primary) {
1480		cpa->numpages = 1;
1481		return 0;
1482	}
1483
1484	/*
1485	 * Ignore the NULL PTE for kernel identity mapping, as it is expected
1486	 * to have holes.
1487	 * Also set numpages to '1' indicating that we processed cpa req for
1488	 * one virtual address page and its pfn. TBD: numpages can be set based
1489	 * on the initial value and the level returned by lookup_address().
1490	 */
1491	if (within(vaddr, PAGE_OFFSET,
1492		   PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))) {
1493		cpa->numpages = 1;
1494		cpa->pfn = __pa(vaddr) >> PAGE_SHIFT;
1495		return 0;
1496
1497	} else if (__cpa_pfn_in_highmap(cpa->pfn)) {
1498		/* Faults in the highmap are OK, so do not warn: */
1499		return -EFAULT;
1500	} else {
1501		WARN(1, KERN_WARNING "CPA: called for zero pte. "
1502			"vaddr = %lx cpa->vaddr = %lx\n", vaddr,
1503			*cpa->vaddr);
1504
1505		return -EFAULT;
1506	}
1507}
1508
1509static int __change_page_attr(struct cpa_data *cpa, int primary)
1510{
1511	unsigned long address;
1512	int do_split, err;
1513	unsigned int level;
1514	pte_t *kpte, old_pte;
1515
1516	address = __cpa_addr(cpa, cpa->curpage);
1517repeat:
1518	kpte = _lookup_address_cpa(cpa, address, &level);
1519	if (!kpte)
1520		return __cpa_process_fault(cpa, address, primary);
1521
1522	old_pte = *kpte;
1523	if (pte_none(old_pte))
1524		return __cpa_process_fault(cpa, address, primary);
1525
1526	if (level == PG_LEVEL_4K) {
1527		pte_t new_pte;
 
1528		pgprot_t new_prot = pte_pgprot(old_pte);
1529		unsigned long pfn = pte_pfn(old_pte);
1530
1531		pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
1532		pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
1533
1534		cpa_inc_4k_install();
1535		/* Hand in lpsize = 0 to enforce the protection mechanism */
1536		new_prot = static_protections(new_prot, address, pfn, 1, 0,
1537					      CPA_PROTECT);
1538
 
 
1539		new_prot = pgprot_clear_protnone_bits(new_prot);
1540
1541		/*
1542		 * We need to keep the pfn from the existing PTE,
1543		 * after all we're only going to change it's attributes
1544		 * not the memory it points to
1545		 */
1546		new_pte = pfn_pte(pfn, new_prot);
1547		cpa->pfn = pfn;
1548		/*
1549		 * Do we really change anything ?
1550		 */
1551		if (pte_val(old_pte) != pte_val(new_pte)) {
1552			set_pte_atomic(kpte, new_pte);
1553			cpa->flags |= CPA_FLUSHTLB;
1554		}
1555		cpa->numpages = 1;
1556		return 0;
1557	}
1558
1559	/*
1560	 * Check, whether we can keep the large page intact
1561	 * and just change the pte:
1562	 */
1563	do_split = should_split_large_page(kpte, address, cpa);
1564	/*
1565	 * When the range fits into the existing large page,
1566	 * return. cp->numpages and cpa->tlbflush have been updated in
1567	 * try_large_page:
1568	 */
1569	if (do_split <= 0)
1570		return do_split;
1571
1572	/*
1573	 * We have to split the large page:
1574	 */
1575	err = split_large_page(cpa, kpte, address);
1576	if (!err)
1577		goto repeat;
1578
1579	return err;
1580}
1581
1582static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias);
1583
 
 
 
1584static int cpa_process_alias(struct cpa_data *cpa)
1585{
1586	struct cpa_data alias_cpa;
1587	unsigned long laddr = (unsigned long)__va(cpa->pfn << PAGE_SHIFT);
1588	unsigned long vaddr;
1589	int ret;
1590
1591	if (!pfn_range_is_mapped(cpa->pfn, cpa->pfn + 1))
1592		return 0;
1593
1594	/*
1595	 * No need to redo, when the primary call touched the direct
1596	 * mapping already:
1597	 */
1598	vaddr = __cpa_addr(cpa, cpa->curpage);
1599	if (!(within(vaddr, PAGE_OFFSET,
1600		    PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT)))) {
1601
1602		alias_cpa = *cpa;
1603		alias_cpa.vaddr = &laddr;
1604		alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1605		alias_cpa.curpage = 0;
1606
 
 
 
 
 
 
1607		cpa->force_flush_all = 1;
1608
1609		ret = __change_page_attr_set_clr(&alias_cpa, 0);
1610		if (ret)
1611			return ret;
1612	}
1613
1614#ifdef CONFIG_X86_64
1615	/*
1616	 * If the primary call didn't touch the high mapping already
1617	 * and the physical address is inside the kernel map, we need
1618	 * to touch the high mapped kernel as well:
1619	 */
1620	if (!within(vaddr, (unsigned long)_text, _brk_end) &&
1621	    __cpa_pfn_in_highmap(cpa->pfn)) {
1622		unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) +
1623					       __START_KERNEL_map - phys_base;
1624		alias_cpa = *cpa;
1625		alias_cpa.vaddr = &temp_cpa_vaddr;
1626		alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1627		alias_cpa.curpage = 0;
1628
 
 
 
 
 
 
 
 
 
1629		cpa->force_flush_all = 1;
1630		/*
1631		 * The high mapping range is imprecise, so ignore the
1632		 * return value.
1633		 */
1634		__change_page_attr_set_clr(&alias_cpa, 0);
1635	}
1636#endif
1637
1638	return 0;
1639}
1640
1641static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
1642{
1643	unsigned long numpages = cpa->numpages;
1644	unsigned long rempages = numpages;
1645	int ret = 0;
1646
 
 
 
 
 
 
 
1647	while (rempages) {
1648		/*
1649		 * Store the remaining nr of pages for the large page
1650		 * preservation check.
1651		 */
1652		cpa->numpages = rempages;
1653		/* for array changes, we can't use large page */
1654		if (cpa->flags & (CPA_ARRAY | CPA_PAGES_ARRAY))
1655			cpa->numpages = 1;
1656
1657		if (!debug_pagealloc_enabled())
1658			spin_lock(&cpa_lock);
1659		ret = __change_page_attr(cpa, checkalias);
1660		if (!debug_pagealloc_enabled())
1661			spin_unlock(&cpa_lock);
1662		if (ret)
1663			goto out;
1664
1665		if (checkalias) {
1666			ret = cpa_process_alias(cpa);
1667			if (ret)
1668				goto out;
1669		}
1670
1671		/*
1672		 * Adjust the number of pages with the result of the
1673		 * CPA operation. Either a large page has been
1674		 * preserved or a single page update happened.
1675		 */
1676		BUG_ON(cpa->numpages > rempages || !cpa->numpages);
1677		rempages -= cpa->numpages;
1678		cpa->curpage += cpa->numpages;
1679	}
1680
1681out:
1682	/* Restore the original numpages */
1683	cpa->numpages = numpages;
1684	return ret;
1685}
1686
1687static int change_page_attr_set_clr(unsigned long *addr, int numpages,
1688				    pgprot_t mask_set, pgprot_t mask_clr,
1689				    int force_split, int in_flag,
1690				    struct page **pages)
1691{
1692	struct cpa_data cpa;
1693	int ret, cache, checkalias;
1694
1695	memset(&cpa, 0, sizeof(cpa));
1696
1697	/*
1698	 * Check, if we are requested to set a not supported
1699	 * feature.  Clearing non-supported features is OK.
1700	 */
1701	mask_set = canon_pgprot(mask_set);
1702
1703	if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
1704		return 0;
1705
1706	/* Ensure we are PAGE_SIZE aligned */
1707	if (in_flag & CPA_ARRAY) {
1708		int i;
1709		for (i = 0; i < numpages; i++) {
1710			if (addr[i] & ~PAGE_MASK) {
1711				addr[i] &= PAGE_MASK;
1712				WARN_ON_ONCE(1);
1713			}
1714		}
1715	} else if (!(in_flag & CPA_PAGES_ARRAY)) {
1716		/*
1717		 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
1718		 * No need to check in that case
1719		 */
1720		if (*addr & ~PAGE_MASK) {
1721			*addr &= PAGE_MASK;
1722			/*
1723			 * People should not be passing in unaligned addresses:
1724			 */
1725			WARN_ON_ONCE(1);
1726		}
1727	}
1728
1729	/* Must avoid aliasing mappings in the highmem code */
1730	kmap_flush_unused();
1731
1732	vm_unmap_aliases();
1733
1734	cpa.vaddr = addr;
1735	cpa.pages = pages;
1736	cpa.numpages = numpages;
1737	cpa.mask_set = mask_set;
1738	cpa.mask_clr = mask_clr;
1739	cpa.flags = 0;
1740	cpa.curpage = 0;
1741	cpa.force_split = force_split;
1742
1743	if (in_flag & (CPA_ARRAY | CPA_PAGES_ARRAY))
1744		cpa.flags |= in_flag;
1745
1746	/* No alias checking for _NX bit modifications */
1747	checkalias = (pgprot_val(mask_set) | pgprot_val(mask_clr)) != _PAGE_NX;
1748	/* Has caller explicitly disabled alias checking? */
1749	if (in_flag & CPA_NO_CHECK_ALIAS)
1750		checkalias = 0;
1751
1752	ret = __change_page_attr_set_clr(&cpa, checkalias);
1753
1754	/*
1755	 * Check whether we really changed something:
1756	 */
1757	if (!(cpa.flags & CPA_FLUSHTLB))
1758		goto out;
1759
1760	/*
1761	 * No need to flush, when we did not set any of the caching
1762	 * attributes:
1763	 */
1764	cache = !!pgprot2cachemode(mask_set);
1765
1766	/*
1767	 * On error; flush everything to be sure.
1768	 */
1769	if (ret) {
1770		cpa_flush_all(cache);
1771		goto out;
1772	}
1773
1774	cpa_flush(&cpa, cache);
1775out:
1776	return ret;
1777}
1778
1779static inline int change_page_attr_set(unsigned long *addr, int numpages,
1780				       pgprot_t mask, int array)
1781{
1782	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0,
1783		(array ? CPA_ARRAY : 0), NULL);
1784}
1785
1786static inline int change_page_attr_clear(unsigned long *addr, int numpages,
1787					 pgprot_t mask, int array)
1788{
1789	return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0,
1790		(array ? CPA_ARRAY : 0), NULL);
1791}
1792
1793static inline int cpa_set_pages_array(struct page **pages, int numpages,
1794				       pgprot_t mask)
1795{
1796	return change_page_attr_set_clr(NULL, numpages, mask, __pgprot(0), 0,
1797		CPA_PAGES_ARRAY, pages);
1798}
1799
1800static inline int cpa_clear_pages_array(struct page **pages, int numpages,
1801					 pgprot_t mask)
1802{
1803	return change_page_attr_set_clr(NULL, numpages, __pgprot(0), mask, 0,
1804		CPA_PAGES_ARRAY, pages);
1805}
1806
1807/*
1808 * _set_memory_prot is an internal helper for callers that have been passed
1809 * a pgprot_t value from upper layers and a reservation has already been taken.
1810 * If you want to set the pgprot to a specific page protocol, use the
1811 * set_memory_xx() functions.
1812 */
1813int __set_memory_prot(unsigned long addr, int numpages, pgprot_t prot)
1814{
1815	return change_page_attr_set_clr(&addr, numpages, prot,
1816					__pgprot(~pgprot_val(prot)), 0, 0,
1817					NULL);
1818}
1819
1820int _set_memory_uc(unsigned long addr, int numpages)
1821{
1822	/*
1823	 * for now UC MINUS. see comments in ioremap()
1824	 * If you really need strong UC use ioremap_uc(), but note
1825	 * that you cannot override IO areas with set_memory_*() as
1826	 * these helpers cannot work with IO memory.
1827	 */
1828	return change_page_attr_set(&addr, numpages,
1829				    cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1830				    0);
1831}
1832
1833int set_memory_uc(unsigned long addr, int numpages)
1834{
1835	int ret;
1836
1837	/*
1838	 * for now UC MINUS. see comments in ioremap()
1839	 */
1840	ret = memtype_reserve(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1841			      _PAGE_CACHE_MODE_UC_MINUS, NULL);
1842	if (ret)
1843		goto out_err;
1844
1845	ret = _set_memory_uc(addr, numpages);
1846	if (ret)
1847		goto out_free;
1848
1849	return 0;
1850
1851out_free:
1852	memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1853out_err:
1854	return ret;
1855}
1856EXPORT_SYMBOL(set_memory_uc);
1857
1858int _set_memory_wc(unsigned long addr, int numpages)
1859{
1860	int ret;
1861
1862	ret = change_page_attr_set(&addr, numpages,
1863				   cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1864				   0);
1865	if (!ret) {
1866		ret = change_page_attr_set_clr(&addr, numpages,
1867					       cachemode2pgprot(_PAGE_CACHE_MODE_WC),
1868					       __pgprot(_PAGE_CACHE_MASK),
1869					       0, 0, NULL);
1870	}
1871	return ret;
1872}
1873
1874int set_memory_wc(unsigned long addr, int numpages)
1875{
1876	int ret;
1877
1878	ret = memtype_reserve(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1879		_PAGE_CACHE_MODE_WC, NULL);
1880	if (ret)
1881		return ret;
1882
1883	ret = _set_memory_wc(addr, numpages);
1884	if (ret)
1885		memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1886
1887	return ret;
1888}
1889EXPORT_SYMBOL(set_memory_wc);
1890
1891int _set_memory_wt(unsigned long addr, int numpages)
1892{
1893	return change_page_attr_set(&addr, numpages,
1894				    cachemode2pgprot(_PAGE_CACHE_MODE_WT), 0);
1895}
1896
1897int _set_memory_wb(unsigned long addr, int numpages)
1898{
1899	/* WB cache mode is hard wired to all cache attribute bits being 0 */
1900	return change_page_attr_clear(&addr, numpages,
1901				      __pgprot(_PAGE_CACHE_MASK), 0);
1902}
1903
1904int set_memory_wb(unsigned long addr, int numpages)
1905{
1906	int ret;
1907
1908	ret = _set_memory_wb(addr, numpages);
1909	if (ret)
1910		return ret;
1911
1912	memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1913	return 0;
1914}
1915EXPORT_SYMBOL(set_memory_wb);
1916
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1917int set_memory_x(unsigned long addr, int numpages)
1918{
1919	if (!(__supported_pte_mask & _PAGE_NX))
1920		return 0;
1921
1922	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0);
1923}
1924
1925int set_memory_nx(unsigned long addr, int numpages)
1926{
1927	if (!(__supported_pte_mask & _PAGE_NX))
1928		return 0;
1929
1930	return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0);
1931}
1932
1933int set_memory_ro(unsigned long addr, int numpages)
1934{
1935	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0);
 
 
 
 
 
 
 
 
 
 
1936}
1937
1938int set_memory_rw(unsigned long addr, int numpages)
1939{
1940	return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
1941}
1942
1943int set_memory_np(unsigned long addr, int numpages)
1944{
1945	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
1946}
1947
1948int set_memory_np_noalias(unsigned long addr, int numpages)
1949{
1950	int cpa_flags = CPA_NO_CHECK_ALIAS;
1951
1952	return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
1953					__pgprot(_PAGE_PRESENT), 0,
1954					cpa_flags, NULL);
 
 
 
 
 
1955}
1956
1957int set_memory_4k(unsigned long addr, int numpages)
1958{
1959	return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
1960					__pgprot(0), 1, 0, NULL);
1961}
1962
1963int set_memory_nonglobal(unsigned long addr, int numpages)
1964{
1965	return change_page_attr_clear(&addr, numpages,
1966				      __pgprot(_PAGE_GLOBAL), 0);
1967}
1968
1969int set_memory_global(unsigned long addr, int numpages)
1970{
1971	return change_page_attr_set(&addr, numpages,
1972				    __pgprot(_PAGE_GLOBAL), 0);
1973}
1974
1975static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc)
 
 
 
 
1976{
 
1977	struct cpa_data cpa;
1978	int ret;
1979
1980	/* Nothing to do if memory encryption is not active */
1981	if (!mem_encrypt_active())
1982		return 0;
1983
1984	/* Should not be working on unaligned addresses */
1985	if (WARN_ONCE(addr & ~PAGE_MASK, "misaligned address: %#lx\n", addr))
1986		addr &= PAGE_MASK;
1987
1988	memset(&cpa, 0, sizeof(cpa));
1989	cpa.vaddr = &addr;
1990	cpa.numpages = numpages;
1991	cpa.mask_set = enc ? __pgprot(_PAGE_ENC) : __pgprot(0);
1992	cpa.mask_clr = enc ? __pgprot(0) : __pgprot(_PAGE_ENC);
1993	cpa.pgd = init_mm.pgd;
1994
1995	/* Must avoid aliasing mappings in the highmem code */
1996	kmap_flush_unused();
1997	vm_unmap_aliases();
1998
1999	/*
2000	 * Before changing the encryption attribute, we need to flush caches.
2001	 */
2002	cpa_flush(&cpa, 1);
 
 
 
2003
2004	ret = __change_page_attr_set_clr(&cpa, 1);
2005
2006	/*
2007	 * After changing the encryption attribute, we need to flush TLBs again
2008	 * in case any speculative TLB caching occurred (but no need to flush
2009	 * caches again).  We could just use cpa_flush_all(), but in case TLB
2010	 * flushing gets optimized in the cpa_flush() path use the same logic
2011	 * as above.
2012	 */
2013	cpa_flush(&cpa, 0);
2014
 
 
 
 
 
 
2015	return ret;
2016}
2017
 
 
 
 
 
 
 
 
2018int set_memory_encrypted(unsigned long addr, int numpages)
2019{
2020	return __set_memory_enc_dec(addr, numpages, true);
2021}
2022EXPORT_SYMBOL_GPL(set_memory_encrypted);
2023
2024int set_memory_decrypted(unsigned long addr, int numpages)
2025{
2026	return __set_memory_enc_dec(addr, numpages, false);
2027}
2028EXPORT_SYMBOL_GPL(set_memory_decrypted);
2029
2030int set_pages_uc(struct page *page, int numpages)
2031{
2032	unsigned long addr = (unsigned long)page_address(page);
2033
2034	return set_memory_uc(addr, numpages);
2035}
2036EXPORT_SYMBOL(set_pages_uc);
2037
2038static int _set_pages_array(struct page **pages, int numpages,
2039		enum page_cache_mode new_type)
2040{
2041	unsigned long start;
2042	unsigned long end;
2043	enum page_cache_mode set_type;
2044	int i;
2045	int free_idx;
2046	int ret;
2047
2048	for (i = 0; i < numpages; i++) {
2049		if (PageHighMem(pages[i]))
2050			continue;
2051		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2052		end = start + PAGE_SIZE;
2053		if (memtype_reserve(start, end, new_type, NULL))
2054			goto err_out;
2055	}
2056
2057	/* If WC, set to UC- first and then WC */
2058	set_type = (new_type == _PAGE_CACHE_MODE_WC) ?
2059				_PAGE_CACHE_MODE_UC_MINUS : new_type;
2060
2061	ret = cpa_set_pages_array(pages, numpages,
2062				  cachemode2pgprot(set_type));
2063	if (!ret && new_type == _PAGE_CACHE_MODE_WC)
2064		ret = change_page_attr_set_clr(NULL, numpages,
2065					       cachemode2pgprot(
2066						_PAGE_CACHE_MODE_WC),
2067					       __pgprot(_PAGE_CACHE_MASK),
2068					       0, CPA_PAGES_ARRAY, pages);
2069	if (ret)
2070		goto err_out;
2071	return 0; /* Success */
2072err_out:
2073	free_idx = i;
2074	for (i = 0; i < free_idx; i++) {
2075		if (PageHighMem(pages[i]))
2076			continue;
2077		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2078		end = start + PAGE_SIZE;
2079		memtype_free(start, end);
2080	}
2081	return -EINVAL;
2082}
2083
2084int set_pages_array_uc(struct page **pages, int numpages)
2085{
2086	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_UC_MINUS);
2087}
2088EXPORT_SYMBOL(set_pages_array_uc);
2089
2090int set_pages_array_wc(struct page **pages, int numpages)
2091{
2092	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WC);
2093}
2094EXPORT_SYMBOL(set_pages_array_wc);
2095
2096int set_pages_array_wt(struct page **pages, int numpages)
2097{
2098	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WT);
2099}
2100EXPORT_SYMBOL_GPL(set_pages_array_wt);
2101
2102int set_pages_wb(struct page *page, int numpages)
2103{
2104	unsigned long addr = (unsigned long)page_address(page);
2105
2106	return set_memory_wb(addr, numpages);
2107}
2108EXPORT_SYMBOL(set_pages_wb);
2109
2110int set_pages_array_wb(struct page **pages, int numpages)
2111{
2112	int retval;
2113	unsigned long start;
2114	unsigned long end;
2115	int i;
2116
2117	/* WB cache mode is hard wired to all cache attribute bits being 0 */
2118	retval = cpa_clear_pages_array(pages, numpages,
2119			__pgprot(_PAGE_CACHE_MASK));
2120	if (retval)
2121		return retval;
2122
2123	for (i = 0; i < numpages; i++) {
2124		if (PageHighMem(pages[i]))
2125			continue;
2126		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2127		end = start + PAGE_SIZE;
2128		memtype_free(start, end);
2129	}
2130
2131	return 0;
2132}
2133EXPORT_SYMBOL(set_pages_array_wb);
2134
2135int set_pages_ro(struct page *page, int numpages)
2136{
2137	unsigned long addr = (unsigned long)page_address(page);
2138
2139	return set_memory_ro(addr, numpages);
2140}
2141
2142int set_pages_rw(struct page *page, int numpages)
2143{
2144	unsigned long addr = (unsigned long)page_address(page);
2145
2146	return set_memory_rw(addr, numpages);
2147}
2148
2149static int __set_pages_p(struct page *page, int numpages)
2150{
2151	unsigned long tempaddr = (unsigned long) page_address(page);
2152	struct cpa_data cpa = { .vaddr = &tempaddr,
2153				.pgd = NULL,
2154				.numpages = numpages,
2155				.mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
2156				.mask_clr = __pgprot(0),
2157				.flags = 0};
2158
2159	/*
2160	 * No alias checking needed for setting present flag. otherwise,
2161	 * we may need to break large pages for 64-bit kernel text
2162	 * mappings (this adds to complexity if we want to do this from
2163	 * atomic context especially). Let's keep it simple!
2164	 */
2165	return __change_page_attr_set_clr(&cpa, 0);
2166}
2167
2168static int __set_pages_np(struct page *page, int numpages)
2169{
2170	unsigned long tempaddr = (unsigned long) page_address(page);
2171	struct cpa_data cpa = { .vaddr = &tempaddr,
2172				.pgd = NULL,
2173				.numpages = numpages,
2174				.mask_set = __pgprot(0),
2175				.mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW),
2176				.flags = 0};
2177
2178	/*
2179	 * No alias checking needed for setting not present flag. otherwise,
2180	 * we may need to break large pages for 64-bit kernel text
2181	 * mappings (this adds to complexity if we want to do this from
2182	 * atomic context especially). Let's keep it simple!
2183	 */
2184	return __change_page_attr_set_clr(&cpa, 0);
2185}
2186
2187int set_direct_map_invalid_noflush(struct page *page)
2188{
2189	return __set_pages_np(page, 1);
2190}
2191
2192int set_direct_map_default_noflush(struct page *page)
2193{
2194	return __set_pages_p(page, 1);
2195}
2196
 
2197void __kernel_map_pages(struct page *page, int numpages, int enable)
2198{
2199	if (PageHighMem(page))
2200		return;
2201	if (!enable) {
2202		debug_check_no_locks_freed(page_address(page),
2203					   numpages * PAGE_SIZE);
2204	}
2205
2206	/*
2207	 * The return value is ignored as the calls cannot fail.
2208	 * Large pages for identity mappings are not used at boot time
2209	 * and hence no memory allocations during large page split.
2210	 */
2211	if (enable)
2212		__set_pages_p(page, numpages);
2213	else
2214		__set_pages_np(page, numpages);
2215
2216	/*
2217	 * We should perform an IPI and flush all tlbs,
2218	 * but that can deadlock->flush only current cpu.
2219	 * Preemption needs to be disabled around __flush_tlb_all() due to
2220	 * CR3 reload in __native_flush_tlb().
2221	 */
2222	preempt_disable();
2223	__flush_tlb_all();
2224	preempt_enable();
2225
2226	arch_flush_lazy_mmu_mode();
2227}
 
2228
2229#ifdef CONFIG_HIBERNATION
2230bool kernel_page_present(struct page *page)
2231{
2232	unsigned int level;
2233	pte_t *pte;
2234
2235	if (PageHighMem(page))
2236		return false;
2237
2238	pte = lookup_address((unsigned long)page_address(page), &level);
2239	return (pte_val(*pte) & _PAGE_PRESENT);
2240}
2241#endif /* CONFIG_HIBERNATION */
2242
2243int __init kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address,
2244				   unsigned numpages, unsigned long page_flags)
2245{
2246	int retval = -EINVAL;
2247
2248	struct cpa_data cpa = {
2249		.vaddr = &address,
2250		.pfn = pfn,
2251		.pgd = pgd,
2252		.numpages = numpages,
2253		.mask_set = __pgprot(0),
2254		.mask_clr = __pgprot(~page_flags & (_PAGE_NX|_PAGE_RW)),
2255		.flags = 0,
2256	};
2257
2258	WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2259
2260	if (!(__supported_pte_mask & _PAGE_NX))
2261		goto out;
2262
2263	if (!(page_flags & _PAGE_ENC))
2264		cpa.mask_clr = pgprot_encrypted(cpa.mask_clr);
2265
2266	cpa.mask_set = __pgprot(_PAGE_PRESENT | page_flags);
2267
2268	retval = __change_page_attr_set_clr(&cpa, 0);
2269	__flush_tlb_all();
2270
2271out:
2272	return retval;
2273}
2274
2275/*
2276 * __flush_tlb_all() flushes mappings only on current CPU and hence this
2277 * function shouldn't be used in an SMP environment. Presently, it's used only
2278 * during boot (way before smp_init()) by EFI subsystem and hence is ok.
2279 */
2280int __init kernel_unmap_pages_in_pgd(pgd_t *pgd, unsigned long address,
2281				     unsigned long numpages)
2282{
2283	int retval;
2284
2285	/*
2286	 * The typical sequence for unmapping is to find a pte through
2287	 * lookup_address_in_pgd() (ideally, it should never return NULL because
2288	 * the address is already mapped) and change it's protections. As pfn is
2289	 * the *target* of a mapping, it's not useful while unmapping.
2290	 */
2291	struct cpa_data cpa = {
2292		.vaddr		= &address,
2293		.pfn		= 0,
2294		.pgd		= pgd,
2295		.numpages	= numpages,
2296		.mask_set	= __pgprot(0),
2297		.mask_clr	= __pgprot(_PAGE_PRESENT | _PAGE_RW),
2298		.flags		= 0,
2299	};
2300
2301	WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2302
2303	retval = __change_page_attr_set_clr(&cpa, 0);
2304	__flush_tlb_all();
2305
2306	return retval;
2307}
2308
2309/*
2310 * The testcases use internal knowledge of the implementation that shouldn't
2311 * be exposed to the rest of the kernel. Include these directly here.
2312 */
2313#ifdef CONFIG_CPA_DEBUG
2314#include "cpa-test.c"
2315#endif