~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/arch/x86/mm/pat/set_memory.c

Version: ~ [ linux-5.13-rc1 ] ~ [ linux-5.12.2 ] ~ [ linux-5.11.19 ] ~ [ linux-5.10.35 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.117 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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  */
 36 struct 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 
 51 enum cpa_warn {
 52         CPA_CONFLICT,
 53         CPA_PROTECT,
 54         CPA_DETECT,
 55 };
 56 
 57 static 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  */
 65 static 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 
 72 static inline pgprot_t cachemode2pgprot(enum page_cache_mode pcm)
 73 {
 74         return __pgprot(cachemode2protval(pcm));
 75 }
 76 
 77 #ifdef CONFIG_PROC_FS
 78 static unsigned long direct_pages_count[PG_LEVEL_NUM];
 79 
 80 void 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 
 88 static 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 
 97 void 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
113 static inline void split_page_count(int level) { }
114 #endif
115 
116 #ifdef CONFIG_X86_CPA_STATISTICS
117 
118 static unsigned long cpa_1g_checked;
119 static unsigned long cpa_1g_sameprot;
120 static unsigned long cpa_1g_preserved;
121 static unsigned long cpa_2m_checked;
122 static unsigned long cpa_2m_sameprot;
123 static unsigned long cpa_2m_preserved;
124 static unsigned long cpa_4k_install;
125 
126 static inline void cpa_inc_1g_checked(void)
127 {
128         cpa_1g_checked++;
129 }
130 
131 static inline void cpa_inc_2m_checked(void)
132 {
133         cpa_2m_checked++;
134 }
135 
136 static inline void cpa_inc_4k_install(void)
137 {
138         data_race(cpa_4k_install++);
139 }
140 
141 static 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 
149 static 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 
157 static 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 
169 static int cpastats_open(struct inode *inode, struct file *file)
170 {
171         return single_open(file, cpastats_show, NULL);
172 }
173 
174 static const struct file_operations cpastats_fops = {
175         .open           = cpastats_open,
176         .read           = seq_read,
177         .llseek         = seq_lseek,
178         .release        = single_release,
179 };
180 
181 static 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 }
187 late_initcall(cpa_stats_init);
188 #else
189 static inline void cpa_inc_1g_checked(void) { }
190 static inline void cpa_inc_2m_checked(void) { }
191 static inline void cpa_inc_4k_install(void) { }
192 static inline void cpa_inc_lp_sameprot(int level) { }
193 static inline void cpa_inc_lp_preserved(int level) { }
194 #endif
195 
196 
197 static inline int
198 within(unsigned long addr, unsigned long start, unsigned long end)
199 {
200         return addr >= start && addr < end;
201 }
202 
203 static inline int
204 within_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 
211 static inline unsigned long highmap_start_pfn(void)
212 {
213         return __pa_symbol(_text) >> PAGE_SHIFT;
214 }
215 
216 static 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 
222 static 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 
233 static 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  */
255 static 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 
264 static 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 
285 static 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  */
306 void clflush_cache_range(void *vaddr, unsigned int size)
307 {
308         mb();
309         clflush_cache_range_opt(vaddr, size);
310         mb();
311 }
312 EXPORT_SYMBOL_GPL(clflush_cache_range);
313 
314 #ifdef CONFIG_ARCH_HAS_PMEM_API
315 void arch_invalidate_pmem(void *addr, size_t size)
316 {
317         clflush_cache_range(addr, size);
318 }
319 EXPORT_SYMBOL_GPL(arch_invalidate_pmem);
320 #endif
321 
322 static 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 
336 static 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 
343 static 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 
352 static 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 
388 static 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 
403 static 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
410 static 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  */
421 static 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  */
444 static 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  */
464 static 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
490 static pgprotval_t protect_kernel_text_ro(unsigned long start,
491                                           unsigned long end)
492 {
493         return 0;
494 }
495 #endif
496 
497 static inline bool conflicts(pgprot_t prot, pgprotval_t val)
498 {
499         return (pgprot_val(prot) & ~val) != pgprot_val(prot);
500 }
501 
502 static 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  */
526 static 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  */
575 pte_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  */
624 pte_t *lookup_address(unsigned long address, unsigned int *level)
625 {
626         return lookup_address_in_pgd(pgd_offset_k(address), address, level);
627 }
628 EXPORT_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  */
634 pte_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 }
639 EXPORT_SYMBOL_GPL(lookup_address_in_mm);
640 
641 static 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  */
655 pmd_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  */
687 phys_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 }
719 EXPORT_SYMBOL_GPL(slow_virt_to_phys);
720 
721 /*
722  * Set the new pmd in all the pgds we know about:
723  */
724 static 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 
748 static 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 
765 static 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 
908 static 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 
923 static 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");
955 set:
956         set_pte(pte, pfn_pte(pfn, ref_prot));
957 }
958 
959 static 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 
1064 static 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 
1083 static 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 
1095 static 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 
1107 static 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 
1125 static 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 
1133 static 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 
1177 static 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 
1220 static 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 
1230 static 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 
1240 static 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 
1259 static 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 
1333 static 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  */
1415 static 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 
1464 static 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 
1509 static 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);
1517 repeat:
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 
1582 static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias);
1583 
1584 static 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 
1641 static 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 
1681 out:
1682         /* Restore the original numpages */
1683         cpa->numpages = numpages;
1684         return ret;
1685 }
1686 
1687 static 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);
1775 out:
1776         return ret;
1777 }
1778 
1779 static 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 
1786 static 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 
1793 static 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 
1800 static 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  */
1813 int __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 
1820 int _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 
1833 int 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 
1851 out_free:
1852         memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1853 out_err:
1854         return ret;
1855 }
1856 EXPORT_SYMBOL(set_memory_uc);
1857 
1858 int _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 
1874 int 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 }
1889 EXPORT_SYMBOL(set_memory_wc);
1890 
1891 int _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 
1897 int _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 
1904 int 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 }
1915 EXPORT_SYMBOL(set_memory_wb);
1916 
1917 int 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 
1925 int 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 
1933 int set_memory_ro(unsigned long addr, int numpages)
1934 {
1935         return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW), 0);
1936 }
1937 
1938 int set_memory_rw(unsigned long addr, int numpages)
1939 {
1940         return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
1941 }
1942 
1943 int set_memory_np(unsigned long addr, int numpages)
1944 {
1945         return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
1946 }
1947 
1948 int 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 
1957 int 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 
1963 int set_memory_nonglobal(unsigned long addr, int numpages)
1964 {
1965         return change_page_attr_clear(&addr, numpages,
1966                                       __pgprot(_PAGE_GLOBAL), 0);
1967 }
1968 
1969 int set_memory_global(unsigned long addr, int numpages)
1970 {
1971         return change_page_attr_set(&addr, numpages,
1972                                     __pgprot(_PAGE_GLOBAL), 0);
1973 }
1974 
1975 static 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 
2018 int set_memory_encrypted(unsigned long addr, int numpages)
2019 {
2020         return __set_memory_enc_dec(addr, numpages, true);
2021 }
2022 EXPORT_SYMBOL_GPL(set_memory_encrypted);
2023 
2024 int set_memory_decrypted(unsigned long addr, int numpages)
2025 {
2026         return __set_memory_enc_dec(addr, numpages, false);
2027 }
2028 EXPORT_SYMBOL_GPL(set_memory_decrypted);
2029 
2030 int 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 }
2036 EXPORT_SYMBOL(set_pages_uc);
2037 
2038 static 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 */
2072 err_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 
2084 int set_pages_array_uc(struct page **pages, int numpages)
2085 {
2086         return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_UC_MINUS);
2087 }
2088 EXPORT_SYMBOL(set_pages_array_uc);
2089 
2090 int set_pages_array_wc(struct page **pages, int numpages)
2091 {
2092         return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WC);
2093 }
2094 EXPORT_SYMBOL(set_pages_array_wc);
2095 
2096 int set_pages_array_wt(struct page **pages, int numpages)
2097 {
2098         return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WT);
2099 }
2100 EXPORT_SYMBOL_GPL(set_pages_array_wt);
2101 
2102 int 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 }
2108 EXPORT_SYMBOL(set_pages_wb);
2109 
2110 int 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 }
2133 EXPORT_SYMBOL(set_pages_array_wb);
2134 
2135 int 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 
2142 int 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 
2149 static 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 
2168 static 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 
2187 int set_direct_map_invalid_noflush(struct page *page)
2188 {
2189         return __set_pages_np(page, 1);
2190 }
2191 
2192 int set_direct_map_default_noflush(struct page *page)
2193 {
2194         return __set_pages_p(page, 1);
2195 }
2196 
2197 void __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
2230 bool 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 
2243 int __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 
2271 out:
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  */
2280 int __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
2316 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp