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/proc_fs.h>
13 #include <linux/debugfs.h>
14 #include <linux/pfn.h>
15 #include <linux/percpu.h>
16 #include <linux/gfp.h>
17 #include <linux/pci.h>
18 #include <linux/vmalloc.h>
19 #include <linux/libnvdimm.h>
20 #include <linux/vmstat.h>
21 #include <linux/kernel.h>
22 #include <linux/cc_platform.h>
23 #include <linux/set_memory.h>
24 #include <linux/memregion.h>
25 
26 #include <asm/e820/api.h>
27 #include <asm/processor.h>
28 #include <asm/tlbflush.h>
29 #include <asm/sections.h>
30 #include <asm/setup.h>
31 #include <linux/uaccess.h>
32 #include <asm/pgalloc.h>
33 #include <asm/proto.h>
34 #include <asm/memtype.h>
35 #include <asm/hyperv-tlfs.h>
36 #include <asm/mshyperv.h>
37 
38 #include "../mm_internal.h"
39 
40 /*
41  * The current flushing context - we pass it instead of 5 arguments:
42  */
43 struct cpa_data {
44 	unsigned long	*vaddr;
45 	pgd_t		*pgd;
46 	pgprot_t	mask_set;
47 	pgprot_t	mask_clr;
48 	unsigned long	numpages;
49 	unsigned long	curpage;
50 	unsigned long	pfn;
51 	unsigned int	flags;
52 	unsigned int	force_split		: 1,
53 			force_static_prot	: 1,
54 			force_flush_all		: 1;
55 	struct page	**pages;
56 };
57 
58 enum cpa_warn {
59 	CPA_CONFLICT,
60 	CPA_PROTECT,
61 	CPA_DETECT,
62 };
63 
64 static const int cpa_warn_level = CPA_PROTECT;
65 
66 /*
67  * Serialize cpa() (for !DEBUG_PAGEALLOC which uses large identity mappings)
68  * using cpa_lock. So that we don't allow any other cpu, with stale large tlb
69  * entries change the page attribute in parallel to some other cpu
70  * splitting a large page entry along with changing the attribute.
71  */
72 static DEFINE_SPINLOCK(cpa_lock);
73 
74 #define CPA_FLUSHTLB 1
75 #define CPA_ARRAY 2
76 #define CPA_PAGES_ARRAY 4
77 #define CPA_NO_CHECK_ALIAS 8 /* Do not search for aliases */
78 
cachemode2pgprot(enum page_cache_mode pcm)79 static inline pgprot_t cachemode2pgprot(enum page_cache_mode pcm)
80 {
81 	return __pgprot(cachemode2protval(pcm));
82 }
83 
84 #ifdef CONFIG_PROC_FS
85 static unsigned long direct_pages_count[PG_LEVEL_NUM];
86 
update_page_count(int level,unsigned long pages)87 void update_page_count(int level, unsigned long pages)
88 {
89 	/* Protect against CPA */
90 	spin_lock(&pgd_lock);
91 	direct_pages_count[level] += pages;
92 	spin_unlock(&pgd_lock);
93 }
94 
split_page_count(int level)95 static void split_page_count(int level)
96 {
97 	if (direct_pages_count[level] == 0)
98 		return;
99 
100 	direct_pages_count[level]--;
101 	if (system_state == SYSTEM_RUNNING) {
102 		if (level == PG_LEVEL_2M)
103 			count_vm_event(DIRECT_MAP_LEVEL2_SPLIT);
104 		else if (level == PG_LEVEL_1G)
105 			count_vm_event(DIRECT_MAP_LEVEL3_SPLIT);
106 	}
107 	direct_pages_count[level - 1] += PTRS_PER_PTE;
108 }
109 
arch_report_meminfo(struct seq_file * m)110 void arch_report_meminfo(struct seq_file *m)
111 {
112 	seq_printf(m, "DirectMap4k:    %8lu kB\n",
113 			direct_pages_count[PG_LEVEL_4K] << 2);
114 #if defined(CONFIG_X86_64) || defined(CONFIG_X86_PAE)
115 	seq_printf(m, "DirectMap2M:    %8lu kB\n",
116 			direct_pages_count[PG_LEVEL_2M] << 11);
117 #else
118 	seq_printf(m, "DirectMap4M:    %8lu kB\n",
119 			direct_pages_count[PG_LEVEL_2M] << 12);
120 #endif
121 	if (direct_gbpages)
122 		seq_printf(m, "DirectMap1G:    %8lu kB\n",
123 			direct_pages_count[PG_LEVEL_1G] << 20);
124 }
125 #else
split_page_count(int level)126 static inline void split_page_count(int level) { }
127 #endif
128 
129 #ifdef CONFIG_X86_CPA_STATISTICS
130 
131 static unsigned long cpa_1g_checked;
132 static unsigned long cpa_1g_sameprot;
133 static unsigned long cpa_1g_preserved;
134 static unsigned long cpa_2m_checked;
135 static unsigned long cpa_2m_sameprot;
136 static unsigned long cpa_2m_preserved;
137 static unsigned long cpa_4k_install;
138 
cpa_inc_1g_checked(void)139 static inline void cpa_inc_1g_checked(void)
140 {
141 	cpa_1g_checked++;
142 }
143 
cpa_inc_2m_checked(void)144 static inline void cpa_inc_2m_checked(void)
145 {
146 	cpa_2m_checked++;
147 }
148 
cpa_inc_4k_install(void)149 static inline void cpa_inc_4k_install(void)
150 {
151 	data_race(cpa_4k_install++);
152 }
153 
cpa_inc_lp_sameprot(int level)154 static inline void cpa_inc_lp_sameprot(int level)
155 {
156 	if (level == PG_LEVEL_1G)
157 		cpa_1g_sameprot++;
158 	else
159 		cpa_2m_sameprot++;
160 }
161 
cpa_inc_lp_preserved(int level)162 static inline void cpa_inc_lp_preserved(int level)
163 {
164 	if (level == PG_LEVEL_1G)
165 		cpa_1g_preserved++;
166 	else
167 		cpa_2m_preserved++;
168 }
169 
cpastats_show(struct seq_file * m,void * p)170 static int cpastats_show(struct seq_file *m, void *p)
171 {
172 	seq_printf(m, "1G pages checked:     %16lu\n", cpa_1g_checked);
173 	seq_printf(m, "1G pages sameprot:    %16lu\n", cpa_1g_sameprot);
174 	seq_printf(m, "1G pages preserved:   %16lu\n", cpa_1g_preserved);
175 	seq_printf(m, "2M pages checked:     %16lu\n", cpa_2m_checked);
176 	seq_printf(m, "2M pages sameprot:    %16lu\n", cpa_2m_sameprot);
177 	seq_printf(m, "2M pages preserved:   %16lu\n", cpa_2m_preserved);
178 	seq_printf(m, "4K pages set-checked: %16lu\n", cpa_4k_install);
179 	return 0;
180 }
181 
cpastats_open(struct inode * inode,struct file * file)182 static int cpastats_open(struct inode *inode, struct file *file)
183 {
184 	return single_open(file, cpastats_show, NULL);
185 }
186 
187 static const struct file_operations cpastats_fops = {
188 	.open		= cpastats_open,
189 	.read		= seq_read,
190 	.llseek		= seq_lseek,
191 	.release	= single_release,
192 };
193 
cpa_stats_init(void)194 static int __init cpa_stats_init(void)
195 {
196 	debugfs_create_file("cpa_stats", S_IRUSR, arch_debugfs_dir, NULL,
197 			    &cpastats_fops);
198 	return 0;
199 }
200 late_initcall(cpa_stats_init);
201 #else
cpa_inc_1g_checked(void)202 static inline void cpa_inc_1g_checked(void) { }
cpa_inc_2m_checked(void)203 static inline void cpa_inc_2m_checked(void) { }
cpa_inc_4k_install(void)204 static inline void cpa_inc_4k_install(void) { }
cpa_inc_lp_sameprot(int level)205 static inline void cpa_inc_lp_sameprot(int level) { }
cpa_inc_lp_preserved(int level)206 static inline void cpa_inc_lp_preserved(int level) { }
207 #endif
208 
209 
210 static inline int
within(unsigned long addr,unsigned long start,unsigned long end)211 within(unsigned long addr, unsigned long start, unsigned long end)
212 {
213 	return addr >= start && addr < end;
214 }
215 
216 static inline int
within_inclusive(unsigned long addr,unsigned long start,unsigned long end)217 within_inclusive(unsigned long addr, unsigned long start, unsigned long end)
218 {
219 	return addr >= start && addr <= end;
220 }
221 
222 #ifdef CONFIG_X86_64
223 
224 /*
225  * The kernel image is mapped into two places in the virtual address space
226  * (addresses without KASLR, of course):
227  *
228  * 1. The kernel direct map (0xffff880000000000)
229  * 2. The "high kernel map" (0xffffffff81000000)
230  *
231  * We actually execute out of #2. If we get the address of a kernel symbol, it
232  * points to #2, but almost all physical-to-virtual translations point to #1.
233  *
234  * This is so that we can have both a directmap of all physical memory *and*
235  * take full advantage of the limited (s32) immediate addressing range (2G)
236  * of x86_64.
237  *
238  * See Documentation/arch/x86/x86_64/mm.rst for more detail.
239  */
240 
highmap_start_pfn(void)241 static inline unsigned long highmap_start_pfn(void)
242 {
243 	return __pa_symbol(_text) >> PAGE_SHIFT;
244 }
245 
highmap_end_pfn(void)246 static inline unsigned long highmap_end_pfn(void)
247 {
248 	/* Do not reference physical address outside the kernel. */
249 	return __pa_symbol(roundup(_brk_end, PMD_SIZE) - 1) >> PAGE_SHIFT;
250 }
251 
__cpa_pfn_in_highmap(unsigned long pfn)252 static bool __cpa_pfn_in_highmap(unsigned long pfn)
253 {
254 	/*
255 	 * Kernel text has an alias mapping at a high address, known
256 	 * here as "highmap".
257 	 */
258 	return within_inclusive(pfn, highmap_start_pfn(), highmap_end_pfn());
259 }
260 
261 #else
262 
__cpa_pfn_in_highmap(unsigned long pfn)263 static bool __cpa_pfn_in_highmap(unsigned long pfn)
264 {
265 	/* There is no highmap on 32-bit */
266 	return false;
267 }
268 
269 #endif
270 
271 /*
272  * See set_mce_nospec().
273  *
274  * Machine check recovery code needs to change cache mode of poisoned pages to
275  * UC to avoid speculative access logging another error. But passing the
276  * address of the 1:1 mapping to set_memory_uc() is a fine way to encourage a
277  * speculative access. So we cheat and flip the top bit of the address. This
278  * works fine for the code that updates the page tables. But at the end of the
279  * process we need to flush the TLB and cache and the non-canonical address
280  * causes a #GP fault when used by the INVLPG and CLFLUSH instructions.
281  *
282  * But in the common case we already have a canonical address. This code
283  * will fix the top bit if needed and is a no-op otherwise.
284  */
fix_addr(unsigned long addr)285 static inline unsigned long fix_addr(unsigned long addr)
286 {
287 #ifdef CONFIG_X86_64
288 	return (long)(addr << 1) >> 1;
289 #else
290 	return addr;
291 #endif
292 }
293 
__cpa_addr(struct cpa_data * cpa,unsigned long idx)294 static unsigned long __cpa_addr(struct cpa_data *cpa, unsigned long idx)
295 {
296 	if (cpa->flags & CPA_PAGES_ARRAY) {
297 		struct page *page = cpa->pages[idx];
298 
299 		if (unlikely(PageHighMem(page)))
300 			return 0;
301 
302 		return (unsigned long)page_address(page);
303 	}
304 
305 	if (cpa->flags & CPA_ARRAY)
306 		return cpa->vaddr[idx];
307 
308 	return *cpa->vaddr + idx * PAGE_SIZE;
309 }
310 
311 /*
312  * Flushing functions
313  */
314 
clflush_cache_range_opt(void * vaddr,unsigned int size)315 static void clflush_cache_range_opt(void *vaddr, unsigned int size)
316 {
317 	const unsigned long clflush_size = boot_cpu_data.x86_clflush_size;
318 	void *p = (void *)((unsigned long)vaddr & ~(clflush_size - 1));
319 	void *vend = vaddr + size;
320 
321 	if (p >= vend)
322 		return;
323 
324 	for (; p < vend; p += clflush_size)
325 		clflushopt(p);
326 }
327 
328 /**
329  * clflush_cache_range - flush a cache range with clflush
330  * @vaddr:	virtual start address
331  * @size:	number of bytes to flush
332  *
333  * CLFLUSHOPT is an unordered instruction which needs fencing with MFENCE or
334  * SFENCE to avoid ordering issues.
335  */
clflush_cache_range(void * vaddr,unsigned int size)336 void clflush_cache_range(void *vaddr, unsigned int size)
337 {
338 	mb();
339 	clflush_cache_range_opt(vaddr, size);
340 	mb();
341 }
342 EXPORT_SYMBOL_GPL(clflush_cache_range);
343 
344 #ifdef CONFIG_ARCH_HAS_PMEM_API
arch_invalidate_pmem(void * addr,size_t size)345 void arch_invalidate_pmem(void *addr, size_t size)
346 {
347 	clflush_cache_range(addr, size);
348 }
349 EXPORT_SYMBOL_GPL(arch_invalidate_pmem);
350 #endif
351 
352 #ifdef CONFIG_ARCH_HAS_CPU_CACHE_INVALIDATE_MEMREGION
cpu_cache_has_invalidate_memregion(void)353 bool cpu_cache_has_invalidate_memregion(void)
354 {
355 	return !cpu_feature_enabled(X86_FEATURE_HYPERVISOR);
356 }
357 EXPORT_SYMBOL_NS_GPL(cpu_cache_has_invalidate_memregion, DEVMEM);
358 
cpu_cache_invalidate_memregion(int res_desc)359 int cpu_cache_invalidate_memregion(int res_desc)
360 {
361 	if (WARN_ON_ONCE(!cpu_cache_has_invalidate_memregion()))
362 		return -ENXIO;
363 	wbinvd_on_all_cpus();
364 	return 0;
365 }
366 EXPORT_SYMBOL_NS_GPL(cpu_cache_invalidate_memregion, DEVMEM);
367 #endif
368 
__cpa_flush_all(void * arg)369 static void __cpa_flush_all(void *arg)
370 {
371 	unsigned long cache = (unsigned long)arg;
372 
373 	/*
374 	 * Flush all to work around Errata in early athlons regarding
375 	 * large page flushing.
376 	 */
377 	__flush_tlb_all();
378 
379 	if (cache && boot_cpu_data.x86 >= 4)
380 		wbinvd();
381 }
382 
cpa_flush_all(unsigned long cache)383 static void cpa_flush_all(unsigned long cache)
384 {
385 	BUG_ON(irqs_disabled() && !early_boot_irqs_disabled);
386 
387 	on_each_cpu(__cpa_flush_all, (void *) cache, 1);
388 }
389 
__cpa_flush_tlb(void * data)390 static void __cpa_flush_tlb(void *data)
391 {
392 	struct cpa_data *cpa = data;
393 	unsigned int i;
394 
395 	for (i = 0; i < cpa->numpages; i++)
396 		flush_tlb_one_kernel(fix_addr(__cpa_addr(cpa, i)));
397 }
398 
cpa_flush(struct cpa_data * data,int cache)399 static void cpa_flush(struct cpa_data *data, int cache)
400 {
401 	struct cpa_data *cpa = data;
402 	unsigned int i;
403 
404 	BUG_ON(irqs_disabled() && !early_boot_irqs_disabled);
405 
406 	if (cache && !static_cpu_has(X86_FEATURE_CLFLUSH)) {
407 		cpa_flush_all(cache);
408 		return;
409 	}
410 
411 	if (cpa->force_flush_all || cpa->numpages > tlb_single_page_flush_ceiling)
412 		flush_tlb_all();
413 	else
414 		on_each_cpu(__cpa_flush_tlb, cpa, 1);
415 
416 	if (!cache)
417 		return;
418 
419 	mb();
420 	for (i = 0; i < cpa->numpages; i++) {
421 		unsigned long addr = __cpa_addr(cpa, i);
422 		unsigned int level;
423 
424 		pte_t *pte = lookup_address(addr, &level);
425 
426 		/*
427 		 * Only flush present addresses:
428 		 */
429 		if (pte && (pte_val(*pte) & _PAGE_PRESENT))
430 			clflush_cache_range_opt((void *)fix_addr(addr), PAGE_SIZE);
431 	}
432 	mb();
433 }
434 
overlaps(unsigned long r1_start,unsigned long r1_end,unsigned long r2_start,unsigned long r2_end)435 static bool overlaps(unsigned long r1_start, unsigned long r1_end,
436 		     unsigned long r2_start, unsigned long r2_end)
437 {
438 	return (r1_start <= r2_end && r1_end >= r2_start) ||
439 		(r2_start <= r1_end && r2_end >= r1_start);
440 }
441 
442 #ifdef CONFIG_PCI_BIOS
443 /*
444  * The BIOS area between 640k and 1Mb needs to be executable for PCI BIOS
445  * based config access (CONFIG_PCI_GOBIOS) support.
446  */
447 #define BIOS_PFN	PFN_DOWN(BIOS_BEGIN)
448 #define BIOS_PFN_END	PFN_DOWN(BIOS_END - 1)
449 
protect_pci_bios(unsigned long spfn,unsigned long epfn)450 static pgprotval_t protect_pci_bios(unsigned long spfn, unsigned long epfn)
451 {
452 	if (pcibios_enabled && overlaps(spfn, epfn, BIOS_PFN, BIOS_PFN_END))
453 		return _PAGE_NX;
454 	return 0;
455 }
456 #else
protect_pci_bios(unsigned long spfn,unsigned long epfn)457 static pgprotval_t protect_pci_bios(unsigned long spfn, unsigned long epfn)
458 {
459 	return 0;
460 }
461 #endif
462 
463 /*
464  * The .rodata section needs to be read-only. Using the pfn catches all
465  * aliases.  This also includes __ro_after_init, so do not enforce until
466  * kernel_set_to_readonly is true.
467  */
protect_rodata(unsigned long spfn,unsigned long epfn)468 static pgprotval_t protect_rodata(unsigned long spfn, unsigned long epfn)
469 {
470 	unsigned long epfn_ro, spfn_ro = PFN_DOWN(__pa_symbol(__start_rodata));
471 
472 	/*
473 	 * Note: __end_rodata is at page aligned and not inclusive, so
474 	 * subtract 1 to get the last enforced PFN in the rodata area.
475 	 */
476 	epfn_ro = PFN_DOWN(__pa_symbol(__end_rodata)) - 1;
477 
478 	if (kernel_set_to_readonly && overlaps(spfn, epfn, spfn_ro, epfn_ro))
479 		return _PAGE_RW;
480 	return 0;
481 }
482 
483 /*
484  * Protect kernel text against becoming non executable by forbidding
485  * _PAGE_NX.  This protects only the high kernel mapping (_text -> _etext)
486  * out of which the kernel actually executes.  Do not protect the low
487  * mapping.
488  *
489  * This does not cover __inittext since that is gone after boot.
490  */
protect_kernel_text(unsigned long start,unsigned long end)491 static pgprotval_t protect_kernel_text(unsigned long start, unsigned long end)
492 {
493 	unsigned long t_end = (unsigned long)_etext - 1;
494 	unsigned long t_start = (unsigned long)_text;
495 
496 	if (overlaps(start, end, t_start, t_end))
497 		return _PAGE_NX;
498 	return 0;
499 }
500 
501 #if defined(CONFIG_X86_64)
502 /*
503  * Once the kernel maps the text as RO (kernel_set_to_readonly is set),
504  * kernel text mappings for the large page aligned text, rodata sections
505  * will be always read-only. For the kernel identity mappings covering the
506  * holes caused by this alignment can be anything that user asks.
507  *
508  * This will preserve the large page mappings for kernel text/data at no
509  * extra cost.
510  */
protect_kernel_text_ro(unsigned long start,unsigned long end)511 static pgprotval_t protect_kernel_text_ro(unsigned long start,
512 					  unsigned long end)
513 {
514 	unsigned long t_end = (unsigned long)__end_rodata_hpage_align - 1;
515 	unsigned long t_start = (unsigned long)_text;
516 	unsigned int level;
517 
518 	if (!kernel_set_to_readonly || !overlaps(start, end, t_start, t_end))
519 		return 0;
520 	/*
521 	 * Don't enforce the !RW mapping for the kernel text mapping, if
522 	 * the current mapping is already using small page mapping.  No
523 	 * need to work hard to preserve large page mappings in this case.
524 	 *
525 	 * This also fixes the Linux Xen paravirt guest boot failure caused
526 	 * by unexpected read-only mappings for kernel identity
527 	 * mappings. In this paravirt guest case, the kernel text mapping
528 	 * and the kernel identity mapping share the same page-table pages,
529 	 * so the protections for kernel text and identity mappings have to
530 	 * be the same.
531 	 */
532 	if (lookup_address(start, &level) && (level != PG_LEVEL_4K))
533 		return _PAGE_RW;
534 	return 0;
535 }
536 #else
protect_kernel_text_ro(unsigned long start,unsigned long end)537 static pgprotval_t protect_kernel_text_ro(unsigned long start,
538 					  unsigned long end)
539 {
540 	return 0;
541 }
542 #endif
543 
conflicts(pgprot_t prot,pgprotval_t val)544 static inline bool conflicts(pgprot_t prot, pgprotval_t val)
545 {
546 	return (pgprot_val(prot) & ~val) != pgprot_val(prot);
547 }
548 
check_conflict(int warnlvl,pgprot_t prot,pgprotval_t val,unsigned long start,unsigned long end,unsigned long pfn,const char * txt)549 static inline void check_conflict(int warnlvl, pgprot_t prot, pgprotval_t val,
550 				  unsigned long start, unsigned long end,
551 				  unsigned long pfn, const char *txt)
552 {
553 	static const char *lvltxt[] = {
554 		[CPA_CONFLICT]	= "conflict",
555 		[CPA_PROTECT]	= "protect",
556 		[CPA_DETECT]	= "detect",
557 	};
558 
559 	if (warnlvl > cpa_warn_level || !conflicts(prot, val))
560 		return;
561 
562 	pr_warn("CPA %8s %10s: 0x%016lx - 0x%016lx PFN %lx req %016llx prevent %016llx\n",
563 		lvltxt[warnlvl], txt, start, end, pfn, (unsigned long long)pgprot_val(prot),
564 		(unsigned long long)val);
565 }
566 
567 /*
568  * Certain areas of memory on x86 require very specific protection flags,
569  * for example the BIOS area or kernel text. Callers don't always get this
570  * right (again, ioremap() on BIOS memory is not uncommon) so this function
571  * checks and fixes these known static required protection bits.
572  */
static_protections(pgprot_t prot,unsigned long start,unsigned long pfn,unsigned long npg,unsigned long lpsize,int warnlvl)573 static inline pgprot_t static_protections(pgprot_t prot, unsigned long start,
574 					  unsigned long pfn, unsigned long npg,
575 					  unsigned long lpsize, int warnlvl)
576 {
577 	pgprotval_t forbidden, res;
578 	unsigned long end;
579 
580 	/*
581 	 * There is no point in checking RW/NX conflicts when the requested
582 	 * mapping is setting the page !PRESENT.
583 	 */
584 	if (!(pgprot_val(prot) & _PAGE_PRESENT))
585 		return prot;
586 
587 	/* Operate on the virtual address */
588 	end = start + npg * PAGE_SIZE - 1;
589 
590 	res = protect_kernel_text(start, end);
591 	check_conflict(warnlvl, prot, res, start, end, pfn, "Text NX");
592 	forbidden = res;
593 
594 	/*
595 	 * Special case to preserve a large page. If the change spawns the
596 	 * full large page mapping then there is no point to split it
597 	 * up. Happens with ftrace and is going to be removed once ftrace
598 	 * switched to text_poke().
599 	 */
600 	if (lpsize != (npg * PAGE_SIZE) || (start & (lpsize - 1))) {
601 		res = protect_kernel_text_ro(start, end);
602 		check_conflict(warnlvl, prot, res, start, end, pfn, "Text RO");
603 		forbidden |= res;
604 	}
605 
606 	/* Check the PFN directly */
607 	res = protect_pci_bios(pfn, pfn + npg - 1);
608 	check_conflict(warnlvl, prot, res, start, end, pfn, "PCIBIOS NX");
609 	forbidden |= res;
610 
611 	res = protect_rodata(pfn, pfn + npg - 1);
612 	check_conflict(warnlvl, prot, res, start, end, pfn, "Rodata RO");
613 	forbidden |= res;
614 
615 	return __pgprot(pgprot_val(prot) & ~forbidden);
616 }
617 
618 /*
619  * Validate strict W^X semantics.
620  */
verify_rwx(pgprot_t old,pgprot_t new,unsigned long start,unsigned long pfn,unsigned long npg)621 static inline pgprot_t verify_rwx(pgprot_t old, pgprot_t new, unsigned long start,
622 				  unsigned long pfn, unsigned long npg)
623 {
624 	unsigned long end;
625 
626 	/*
627 	 * 32-bit has some unfixable W+X issues, like EFI code
628 	 * and writeable data being in the same page.  Disable
629 	 * detection and enforcement there.
630 	 */
631 	if (IS_ENABLED(CONFIG_X86_32))
632 		return new;
633 
634 	/* Only verify when NX is supported: */
635 	if (!(__supported_pte_mask & _PAGE_NX))
636 		return new;
637 
638 	if (!((pgprot_val(old) ^ pgprot_val(new)) & (_PAGE_RW | _PAGE_NX)))
639 		return new;
640 
641 	if ((pgprot_val(new) & (_PAGE_RW | _PAGE_NX)) != _PAGE_RW)
642 		return new;
643 
644 	end = start + npg * PAGE_SIZE - 1;
645 	WARN_ONCE(1, "CPA detected W^X violation: %016llx -> %016llx range: 0x%016lx - 0x%016lx PFN %lx\n",
646 		  (unsigned long long)pgprot_val(old),
647 		  (unsigned long long)pgprot_val(new),
648 		  start, end, pfn);
649 
650 	/*
651 	 * For now, allow all permission change attempts by returning the
652 	 * attempted permissions.  This can 'return old' to actively
653 	 * refuse the permission change at a later time.
654 	 */
655 	return new;
656 }
657 
658 /*
659  * Lookup the page table entry for a virtual address in a specific pgd.
660  * Return a pointer to the entry and the level of the mapping.
661  */
lookup_address_in_pgd(pgd_t * pgd,unsigned long address,unsigned int * level)662 pte_t *lookup_address_in_pgd(pgd_t *pgd, unsigned long address,
663 			     unsigned int *level)
664 {
665 	p4d_t *p4d;
666 	pud_t *pud;
667 	pmd_t *pmd;
668 
669 	*level = PG_LEVEL_NONE;
670 
671 	if (pgd_none(*pgd))
672 		return NULL;
673 
674 	p4d = p4d_offset(pgd, address);
675 	if (p4d_none(*p4d))
676 		return NULL;
677 
678 	*level = PG_LEVEL_512G;
679 	if (p4d_large(*p4d) || !p4d_present(*p4d))
680 		return (pte_t *)p4d;
681 
682 	pud = pud_offset(p4d, address);
683 	if (pud_none(*pud))
684 		return NULL;
685 
686 	*level = PG_LEVEL_1G;
687 	if (pud_large(*pud) || !pud_present(*pud))
688 		return (pte_t *)pud;
689 
690 	pmd = pmd_offset(pud, address);
691 	if (pmd_none(*pmd))
692 		return NULL;
693 
694 	*level = PG_LEVEL_2M;
695 	if (pmd_large(*pmd) || !pmd_present(*pmd))
696 		return (pte_t *)pmd;
697 
698 	*level = PG_LEVEL_4K;
699 
700 	return pte_offset_kernel(pmd, address);
701 }
702 
703 /*
704  * Lookup the page table entry for a virtual address. Return a pointer
705  * to the entry and the level of the mapping.
706  *
707  * Note: We return pud and pmd either when the entry is marked large
708  * or when the present bit is not set. Otherwise we would return a
709  * pointer to a nonexisting mapping.
710  */
lookup_address(unsigned long address,unsigned int * level)711 pte_t *lookup_address(unsigned long address, unsigned int *level)
712 {
713 	return lookup_address_in_pgd(pgd_offset_k(address), address, level);
714 }
715 EXPORT_SYMBOL_GPL(lookup_address);
716 
_lookup_address_cpa(struct cpa_data * cpa,unsigned long address,unsigned int * level)717 static pte_t *_lookup_address_cpa(struct cpa_data *cpa, unsigned long address,
718 				  unsigned int *level)
719 {
720 	if (cpa->pgd)
721 		return lookup_address_in_pgd(cpa->pgd + pgd_index(address),
722 					       address, level);
723 
724 	return lookup_address(address, level);
725 }
726 
727 /*
728  * Lookup the PMD entry for a virtual address. Return a pointer to the entry
729  * or NULL if not present.
730  */
lookup_pmd_address(unsigned long address)731 pmd_t *lookup_pmd_address(unsigned long address)
732 {
733 	pgd_t *pgd;
734 	p4d_t *p4d;
735 	pud_t *pud;
736 
737 	pgd = pgd_offset_k(address);
738 	if (pgd_none(*pgd))
739 		return NULL;
740 
741 	p4d = p4d_offset(pgd, address);
742 	if (p4d_none(*p4d) || p4d_large(*p4d) || !p4d_present(*p4d))
743 		return NULL;
744 
745 	pud = pud_offset(p4d, address);
746 	if (pud_none(*pud) || pud_large(*pud) || !pud_present(*pud))
747 		return NULL;
748 
749 	return pmd_offset(pud, address);
750 }
751 
752 /*
753  * This is necessary because __pa() does not work on some
754  * kinds of memory, like vmalloc() or the alloc_remap()
755  * areas on 32-bit NUMA systems.  The percpu areas can
756  * end up in this kind of memory, for instance.
757  *
758  * This could be optimized, but it is only intended to be
759  * used at initialization time, and keeping it
760  * unoptimized should increase the testing coverage for
761  * the more obscure platforms.
762  */
slow_virt_to_phys(void * __virt_addr)763 phys_addr_t slow_virt_to_phys(void *__virt_addr)
764 {
765 	unsigned long virt_addr = (unsigned long)__virt_addr;
766 	phys_addr_t phys_addr;
767 	unsigned long offset;
768 	enum pg_level level;
769 	pte_t *pte;
770 
771 	pte = lookup_address(virt_addr, &level);
772 	BUG_ON(!pte);
773 
774 	/*
775 	 * pXX_pfn() returns unsigned long, which must be cast to phys_addr_t
776 	 * before being left-shifted PAGE_SHIFT bits -- this trick is to
777 	 * make 32-PAE kernel work correctly.
778 	 */
779 	switch (level) {
780 	case PG_LEVEL_1G:
781 		phys_addr = (phys_addr_t)pud_pfn(*(pud_t *)pte) << PAGE_SHIFT;
782 		offset = virt_addr & ~PUD_MASK;
783 		break;
784 	case PG_LEVEL_2M:
785 		phys_addr = (phys_addr_t)pmd_pfn(*(pmd_t *)pte) << PAGE_SHIFT;
786 		offset = virt_addr & ~PMD_MASK;
787 		break;
788 	default:
789 		phys_addr = (phys_addr_t)pte_pfn(*pte) << PAGE_SHIFT;
790 		offset = virt_addr & ~PAGE_MASK;
791 	}
792 
793 	return (phys_addr_t)(phys_addr | offset);
794 }
795 EXPORT_SYMBOL_GPL(slow_virt_to_phys);
796 
797 /*
798  * Set the new pmd in all the pgds we know about:
799  */
__set_pmd_pte(pte_t * kpte,unsigned long address,pte_t pte)800 static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
801 {
802 	/* change init_mm */
803 	set_pte_atomic(kpte, pte);
804 #ifdef CONFIG_X86_32
805 	if (!SHARED_KERNEL_PMD) {
806 		struct page *page;
807 
808 		list_for_each_entry(page, &pgd_list, lru) {
809 			pgd_t *pgd;
810 			p4d_t *p4d;
811 			pud_t *pud;
812 			pmd_t *pmd;
813 
814 			pgd = (pgd_t *)page_address(page) + pgd_index(address);
815 			p4d = p4d_offset(pgd, address);
816 			pud = pud_offset(p4d, address);
817 			pmd = pmd_offset(pud, address);
818 			set_pte_atomic((pte_t *)pmd, pte);
819 		}
820 	}
821 #endif
822 }
823 
pgprot_clear_protnone_bits(pgprot_t prot)824 static pgprot_t pgprot_clear_protnone_bits(pgprot_t prot)
825 {
826 	/*
827 	 * _PAGE_GLOBAL means "global page" for present PTEs.
828 	 * But, it is also used to indicate _PAGE_PROTNONE
829 	 * for non-present PTEs.
830 	 *
831 	 * This ensures that a _PAGE_GLOBAL PTE going from
832 	 * present to non-present is not confused as
833 	 * _PAGE_PROTNONE.
834 	 */
835 	if (!(pgprot_val(prot) & _PAGE_PRESENT))
836 		pgprot_val(prot) &= ~_PAGE_GLOBAL;
837 
838 	return prot;
839 }
840 
__should_split_large_page(pte_t * kpte,unsigned long address,struct cpa_data * cpa)841 static int __should_split_large_page(pte_t *kpte, unsigned long address,
842 				     struct cpa_data *cpa)
843 {
844 	unsigned long numpages, pmask, psize, lpaddr, pfn, old_pfn;
845 	pgprot_t old_prot, new_prot, req_prot, chk_prot;
846 	pte_t new_pte, *tmp;
847 	enum pg_level level;
848 
849 	/*
850 	 * Check for races, another CPU might have split this page
851 	 * up already:
852 	 */
853 	tmp = _lookup_address_cpa(cpa, address, &level);
854 	if (tmp != kpte)
855 		return 1;
856 
857 	switch (level) {
858 	case PG_LEVEL_2M:
859 		old_prot = pmd_pgprot(*(pmd_t *)kpte);
860 		old_pfn = pmd_pfn(*(pmd_t *)kpte);
861 		cpa_inc_2m_checked();
862 		break;
863 	case PG_LEVEL_1G:
864 		old_prot = pud_pgprot(*(pud_t *)kpte);
865 		old_pfn = pud_pfn(*(pud_t *)kpte);
866 		cpa_inc_1g_checked();
867 		break;
868 	default:
869 		return -EINVAL;
870 	}
871 
872 	psize = page_level_size(level);
873 	pmask = page_level_mask(level);
874 
875 	/*
876 	 * Calculate the number of pages, which fit into this large
877 	 * page starting at address:
878 	 */
879 	lpaddr = (address + psize) & pmask;
880 	numpages = (lpaddr - address) >> PAGE_SHIFT;
881 	if (numpages < cpa->numpages)
882 		cpa->numpages = numpages;
883 
884 	/*
885 	 * We are safe now. Check whether the new pgprot is the same:
886 	 * Convert protection attributes to 4k-format, as cpa->mask* are set
887 	 * up accordingly.
888 	 */
889 
890 	/* Clear PSE (aka _PAGE_PAT) and move PAT bit to correct position */
891 	req_prot = pgprot_large_2_4k(old_prot);
892 
893 	pgprot_val(req_prot) &= ~pgprot_val(cpa->mask_clr);
894 	pgprot_val(req_prot) |= pgprot_val(cpa->mask_set);
895 
896 	/*
897 	 * req_prot is in format of 4k pages. It must be converted to large
898 	 * page format: the caching mode includes the PAT bit located at
899 	 * different bit positions in the two formats.
900 	 */
901 	req_prot = pgprot_4k_2_large(req_prot);
902 	req_prot = pgprot_clear_protnone_bits(req_prot);
903 	if (pgprot_val(req_prot) & _PAGE_PRESENT)
904 		pgprot_val(req_prot) |= _PAGE_PSE;
905 
906 	/*
907 	 * old_pfn points to the large page base pfn. So we need to add the
908 	 * offset of the virtual address:
909 	 */
910 	pfn = old_pfn + ((address & (psize - 1)) >> PAGE_SHIFT);
911 	cpa->pfn = pfn;
912 
913 	/*
914 	 * Calculate the large page base address and the number of 4K pages
915 	 * in the large page
916 	 */
917 	lpaddr = address & pmask;
918 	numpages = psize >> PAGE_SHIFT;
919 
920 	/*
921 	 * Sanity check that the existing mapping is correct versus the static
922 	 * protections. static_protections() guards against !PRESENT, so no
923 	 * extra conditional required here.
924 	 */
925 	chk_prot = static_protections(old_prot, lpaddr, old_pfn, numpages,
926 				      psize, CPA_CONFLICT);
927 
928 	if (WARN_ON_ONCE(pgprot_val(chk_prot) != pgprot_val(old_prot))) {
929 		/*
930 		 * Split the large page and tell the split code to
931 		 * enforce static protections.
932 		 */
933 		cpa->force_static_prot = 1;
934 		return 1;
935 	}
936 
937 	/*
938 	 * Optimization: If the requested pgprot is the same as the current
939 	 * pgprot, then the large page can be preserved and no updates are
940 	 * required independent of alignment and length of the requested
941 	 * range. The above already established that the current pgprot is
942 	 * correct, which in consequence makes the requested pgprot correct
943 	 * as well if it is the same. The static protection scan below will
944 	 * not come to a different conclusion.
945 	 */
946 	if (pgprot_val(req_prot) == pgprot_val(old_prot)) {
947 		cpa_inc_lp_sameprot(level);
948 		return 0;
949 	}
950 
951 	/*
952 	 * If the requested range does not cover the full page, split it up
953 	 */
954 	if (address != lpaddr || cpa->numpages != numpages)
955 		return 1;
956 
957 	/*
958 	 * Check whether the requested pgprot is conflicting with a static
959 	 * protection requirement in the large page.
960 	 */
961 	new_prot = static_protections(req_prot, lpaddr, old_pfn, numpages,
962 				      psize, CPA_DETECT);
963 
964 	new_prot = verify_rwx(old_prot, new_prot, lpaddr, old_pfn, numpages);
965 
966 	/*
967 	 * If there is a conflict, split the large page.
968 	 *
969 	 * There used to be a 4k wise evaluation trying really hard to
970 	 * preserve the large pages, but experimentation has shown, that this
971 	 * does not help at all. There might be corner cases which would
972 	 * preserve one large page occasionally, but it's really not worth the
973 	 * extra code and cycles for the common case.
974 	 */
975 	if (pgprot_val(req_prot) != pgprot_val(new_prot))
976 		return 1;
977 
978 	/* All checks passed. Update the large page mapping. */
979 	new_pte = pfn_pte(old_pfn, new_prot);
980 	__set_pmd_pte(kpte, address, new_pte);
981 	cpa->flags |= CPA_FLUSHTLB;
982 	cpa_inc_lp_preserved(level);
983 	return 0;
984 }
985 
should_split_large_page(pte_t * kpte,unsigned long address,struct cpa_data * cpa)986 static int should_split_large_page(pte_t *kpte, unsigned long address,
987 				   struct cpa_data *cpa)
988 {
989 	int do_split;
990 
991 	if (cpa->force_split)
992 		return 1;
993 
994 	spin_lock(&pgd_lock);
995 	do_split = __should_split_large_page(kpte, address, cpa);
996 	spin_unlock(&pgd_lock);
997 
998 	return do_split;
999 }
1000 
split_set_pte(struct cpa_data * cpa,pte_t * pte,unsigned long pfn,pgprot_t ref_prot,unsigned long address,unsigned long size)1001 static void split_set_pte(struct cpa_data *cpa, pte_t *pte, unsigned long pfn,
1002 			  pgprot_t ref_prot, unsigned long address,
1003 			  unsigned long size)
1004 {
1005 	unsigned int npg = PFN_DOWN(size);
1006 	pgprot_t prot;
1007 
1008 	/*
1009 	 * If should_split_large_page() discovered an inconsistent mapping,
1010 	 * remove the invalid protection in the split mapping.
1011 	 */
1012 	if (!cpa->force_static_prot)
1013 		goto set;
1014 
1015 	/* Hand in lpsize = 0 to enforce the protection mechanism */
1016 	prot = static_protections(ref_prot, address, pfn, npg, 0, CPA_PROTECT);
1017 
1018 	if (pgprot_val(prot) == pgprot_val(ref_prot))
1019 		goto set;
1020 
1021 	/*
1022 	 * If this is splitting a PMD, fix it up. PUD splits cannot be
1023 	 * fixed trivially as that would require to rescan the newly
1024 	 * installed PMD mappings after returning from split_large_page()
1025 	 * so an eventual further split can allocate the necessary PTE
1026 	 * pages. Warn for now and revisit it in case this actually
1027 	 * happens.
1028 	 */
1029 	if (size == PAGE_SIZE)
1030 		ref_prot = prot;
1031 	else
1032 		pr_warn_once("CPA: Cannot fixup static protections for PUD split\n");
1033 set:
1034 	set_pte(pte, pfn_pte(pfn, ref_prot));
1035 }
1036 
1037 static int
__split_large_page(struct cpa_data * cpa,pte_t * kpte,unsigned long address,struct page * base)1038 __split_large_page(struct cpa_data *cpa, pte_t *kpte, unsigned long address,
1039 		   struct page *base)
1040 {
1041 	unsigned long lpaddr, lpinc, ref_pfn, pfn, pfninc = 1;
1042 	pte_t *pbase = (pte_t *)page_address(base);
1043 	unsigned int i, level;
1044 	pgprot_t ref_prot;
1045 	pte_t *tmp;
1046 
1047 	spin_lock(&pgd_lock);
1048 	/*
1049 	 * Check for races, another CPU might have split this page
1050 	 * up for us already:
1051 	 */
1052 	tmp = _lookup_address_cpa(cpa, address, &level);
1053 	if (tmp != kpte) {
1054 		spin_unlock(&pgd_lock);
1055 		return 1;
1056 	}
1057 
1058 	paravirt_alloc_pte(&init_mm, page_to_pfn(base));
1059 
1060 	switch (level) {
1061 	case PG_LEVEL_2M:
1062 		ref_prot = pmd_pgprot(*(pmd_t *)kpte);
1063 		/*
1064 		 * Clear PSE (aka _PAGE_PAT) and move
1065 		 * PAT bit to correct position.
1066 		 */
1067 		ref_prot = pgprot_large_2_4k(ref_prot);
1068 		ref_pfn = pmd_pfn(*(pmd_t *)kpte);
1069 		lpaddr = address & PMD_MASK;
1070 		lpinc = PAGE_SIZE;
1071 		break;
1072 
1073 	case PG_LEVEL_1G:
1074 		ref_prot = pud_pgprot(*(pud_t *)kpte);
1075 		ref_pfn = pud_pfn(*(pud_t *)kpte);
1076 		pfninc = PMD_SIZE >> PAGE_SHIFT;
1077 		lpaddr = address & PUD_MASK;
1078 		lpinc = PMD_SIZE;
1079 		/*
1080 		 * Clear the PSE flags if the PRESENT flag is not set
1081 		 * otherwise pmd_present/pmd_huge will return true
1082 		 * even on a non present pmd.
1083 		 */
1084 		if (!(pgprot_val(ref_prot) & _PAGE_PRESENT))
1085 			pgprot_val(ref_prot) &= ~_PAGE_PSE;
1086 		break;
1087 
1088 	default:
1089 		spin_unlock(&pgd_lock);
1090 		return 1;
1091 	}
1092 
1093 	ref_prot = pgprot_clear_protnone_bits(ref_prot);
1094 
1095 	/*
1096 	 * Get the target pfn from the original entry:
1097 	 */
1098 	pfn = ref_pfn;
1099 	for (i = 0; i < PTRS_PER_PTE; i++, pfn += pfninc, lpaddr += lpinc)
1100 		split_set_pte(cpa, pbase + i, pfn, ref_prot, lpaddr, lpinc);
1101 
1102 	if (virt_addr_valid(address)) {
1103 		unsigned long pfn = PFN_DOWN(__pa(address));
1104 
1105 		if (pfn_range_is_mapped(pfn, pfn + 1))
1106 			split_page_count(level);
1107 	}
1108 
1109 	/*
1110 	 * Install the new, split up pagetable.
1111 	 *
1112 	 * We use the standard kernel pagetable protections for the new
1113 	 * pagetable protections, the actual ptes set above control the
1114 	 * primary protection behavior:
1115 	 */
1116 	__set_pmd_pte(kpte, address, mk_pte(base, __pgprot(_KERNPG_TABLE)));
1117 
1118 	/*
1119 	 * Do a global flush tlb after splitting the large page
1120 	 * and before we do the actual change page attribute in the PTE.
1121 	 *
1122 	 * Without this, we violate the TLB application note, that says:
1123 	 * "The TLBs may contain both ordinary and large-page
1124 	 *  translations for a 4-KByte range of linear addresses. This
1125 	 *  may occur if software modifies the paging structures so that
1126 	 *  the page size used for the address range changes. If the two
1127 	 *  translations differ with respect to page frame or attributes
1128 	 *  (e.g., permissions), processor behavior is undefined and may
1129 	 *  be implementation-specific."
1130 	 *
1131 	 * We do this global tlb flush inside the cpa_lock, so that we
1132 	 * don't allow any other cpu, with stale tlb entries change the
1133 	 * page attribute in parallel, that also falls into the
1134 	 * just split large page entry.
1135 	 */
1136 	flush_tlb_all();
1137 	spin_unlock(&pgd_lock);
1138 
1139 	return 0;
1140 }
1141 
split_large_page(struct cpa_data * cpa,pte_t * kpte,unsigned long address)1142 static int split_large_page(struct cpa_data *cpa, pte_t *kpte,
1143 			    unsigned long address)
1144 {
1145 	struct page *base;
1146 
1147 	if (!debug_pagealloc_enabled())
1148 		spin_unlock(&cpa_lock);
1149 	base = alloc_pages(GFP_KERNEL, 0);
1150 	if (!debug_pagealloc_enabled())
1151 		spin_lock(&cpa_lock);
1152 	if (!base)
1153 		return -ENOMEM;
1154 
1155 	if (__split_large_page(cpa, kpte, address, base))
1156 		__free_page(base);
1157 
1158 	return 0;
1159 }
1160 
try_to_free_pte_page(pte_t * pte)1161 static bool try_to_free_pte_page(pte_t *pte)
1162 {
1163 	int i;
1164 
1165 	for (i = 0; i < PTRS_PER_PTE; i++)
1166 		if (!pte_none(pte[i]))
1167 			return false;
1168 
1169 	free_page((unsigned long)pte);
1170 	return true;
1171 }
1172 
try_to_free_pmd_page(pmd_t * pmd)1173 static bool try_to_free_pmd_page(pmd_t *pmd)
1174 {
1175 	int i;
1176 
1177 	for (i = 0; i < PTRS_PER_PMD; i++)
1178 		if (!pmd_none(pmd[i]))
1179 			return false;
1180 
1181 	free_page((unsigned long)pmd);
1182 	return true;
1183 }
1184 
unmap_pte_range(pmd_t * pmd,unsigned long start,unsigned long end)1185 static bool unmap_pte_range(pmd_t *pmd, unsigned long start, unsigned long end)
1186 {
1187 	pte_t *pte = pte_offset_kernel(pmd, start);
1188 
1189 	while (start < end) {
1190 		set_pte(pte, __pte(0));
1191 
1192 		start += PAGE_SIZE;
1193 		pte++;
1194 	}
1195 
1196 	if (try_to_free_pte_page((pte_t *)pmd_page_vaddr(*pmd))) {
1197 		pmd_clear(pmd);
1198 		return true;
1199 	}
1200 	return false;
1201 }
1202 
__unmap_pmd_range(pud_t * pud,pmd_t * pmd,unsigned long start,unsigned long end)1203 static void __unmap_pmd_range(pud_t *pud, pmd_t *pmd,
1204 			      unsigned long start, unsigned long end)
1205 {
1206 	if (unmap_pte_range(pmd, start, end))
1207 		if (try_to_free_pmd_page(pud_pgtable(*pud)))
1208 			pud_clear(pud);
1209 }
1210 
unmap_pmd_range(pud_t * pud,unsigned long start,unsigned long end)1211 static void unmap_pmd_range(pud_t *pud, unsigned long start, unsigned long end)
1212 {
1213 	pmd_t *pmd = pmd_offset(pud, start);
1214 
1215 	/*
1216 	 * Not on a 2MB page boundary?
1217 	 */
1218 	if (start & (PMD_SIZE - 1)) {
1219 		unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
1220 		unsigned long pre_end = min_t(unsigned long, end, next_page);
1221 
1222 		__unmap_pmd_range(pud, pmd, start, pre_end);
1223 
1224 		start = pre_end;
1225 		pmd++;
1226 	}
1227 
1228 	/*
1229 	 * Try to unmap in 2M chunks.
1230 	 */
1231 	while (end - start >= PMD_SIZE) {
1232 		if (pmd_large(*pmd))
1233 			pmd_clear(pmd);
1234 		else
1235 			__unmap_pmd_range(pud, pmd, start, start + PMD_SIZE);
1236 
1237 		start += PMD_SIZE;
1238 		pmd++;
1239 	}
1240 
1241 	/*
1242 	 * 4K leftovers?
1243 	 */
1244 	if (start < end)
1245 		return __unmap_pmd_range(pud, pmd, start, end);
1246 
1247 	/*
1248 	 * Try again to free the PMD page if haven't succeeded above.
1249 	 */
1250 	if (!pud_none(*pud))
1251 		if (try_to_free_pmd_page(pud_pgtable(*pud)))
1252 			pud_clear(pud);
1253 }
1254 
unmap_pud_range(p4d_t * p4d,unsigned long start,unsigned long end)1255 static void unmap_pud_range(p4d_t *p4d, unsigned long start, unsigned long end)
1256 {
1257 	pud_t *pud = pud_offset(p4d, start);
1258 
1259 	/*
1260 	 * Not on a GB page boundary?
1261 	 */
1262 	if (start & (PUD_SIZE - 1)) {
1263 		unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1264 		unsigned long pre_end	= min_t(unsigned long, end, next_page);
1265 
1266 		unmap_pmd_range(pud, start, pre_end);
1267 
1268 		start = pre_end;
1269 		pud++;
1270 	}
1271 
1272 	/*
1273 	 * Try to unmap in 1G chunks?
1274 	 */
1275 	while (end - start >= PUD_SIZE) {
1276 
1277 		if (pud_large(*pud))
1278 			pud_clear(pud);
1279 		else
1280 			unmap_pmd_range(pud, start, start + PUD_SIZE);
1281 
1282 		start += PUD_SIZE;
1283 		pud++;
1284 	}
1285 
1286 	/*
1287 	 * 2M leftovers?
1288 	 */
1289 	if (start < end)
1290 		unmap_pmd_range(pud, start, end);
1291 
1292 	/*
1293 	 * No need to try to free the PUD page because we'll free it in
1294 	 * populate_pgd's error path
1295 	 */
1296 }
1297 
alloc_pte_page(pmd_t * pmd)1298 static int alloc_pte_page(pmd_t *pmd)
1299 {
1300 	pte_t *pte = (pte_t *)get_zeroed_page(GFP_KERNEL);
1301 	if (!pte)
1302 		return -1;
1303 
1304 	set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE));
1305 	return 0;
1306 }
1307 
alloc_pmd_page(pud_t * pud)1308 static int alloc_pmd_page(pud_t *pud)
1309 {
1310 	pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL);
1311 	if (!pmd)
1312 		return -1;
1313 
1314 	set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE));
1315 	return 0;
1316 }
1317 
populate_pte(struct cpa_data * cpa,unsigned long start,unsigned long end,unsigned num_pages,pmd_t * pmd,pgprot_t pgprot)1318 static void populate_pte(struct cpa_data *cpa,
1319 			 unsigned long start, unsigned long end,
1320 			 unsigned num_pages, pmd_t *pmd, pgprot_t pgprot)
1321 {
1322 	pte_t *pte;
1323 
1324 	pte = pte_offset_kernel(pmd, start);
1325 
1326 	pgprot = pgprot_clear_protnone_bits(pgprot);
1327 
1328 	while (num_pages-- && start < end) {
1329 		set_pte(pte, pfn_pte(cpa->pfn, pgprot));
1330 
1331 		start	 += PAGE_SIZE;
1332 		cpa->pfn++;
1333 		pte++;
1334 	}
1335 }
1336 
populate_pmd(struct cpa_data * cpa,unsigned long start,unsigned long end,unsigned num_pages,pud_t * pud,pgprot_t pgprot)1337 static long populate_pmd(struct cpa_data *cpa,
1338 			 unsigned long start, unsigned long end,
1339 			 unsigned num_pages, pud_t *pud, pgprot_t pgprot)
1340 {
1341 	long cur_pages = 0;
1342 	pmd_t *pmd;
1343 	pgprot_t pmd_pgprot;
1344 
1345 	/*
1346 	 * Not on a 2M boundary?
1347 	 */
1348 	if (start & (PMD_SIZE - 1)) {
1349 		unsigned long pre_end = start + (num_pages << PAGE_SHIFT);
1350 		unsigned long next_page = (start + PMD_SIZE) & PMD_MASK;
1351 
1352 		pre_end   = min_t(unsigned long, pre_end, next_page);
1353 		cur_pages = (pre_end - start) >> PAGE_SHIFT;
1354 		cur_pages = min_t(unsigned int, num_pages, cur_pages);
1355 
1356 		/*
1357 		 * Need a PTE page?
1358 		 */
1359 		pmd = pmd_offset(pud, start);
1360 		if (pmd_none(*pmd))
1361 			if (alloc_pte_page(pmd))
1362 				return -1;
1363 
1364 		populate_pte(cpa, start, pre_end, cur_pages, pmd, pgprot);
1365 
1366 		start = pre_end;
1367 	}
1368 
1369 	/*
1370 	 * We mapped them all?
1371 	 */
1372 	if (num_pages == cur_pages)
1373 		return cur_pages;
1374 
1375 	pmd_pgprot = pgprot_4k_2_large(pgprot);
1376 
1377 	while (end - start >= PMD_SIZE) {
1378 
1379 		/*
1380 		 * We cannot use a 1G page so allocate a PMD page if needed.
1381 		 */
1382 		if (pud_none(*pud))
1383 			if (alloc_pmd_page(pud))
1384 				return -1;
1385 
1386 		pmd = pmd_offset(pud, start);
1387 
1388 		set_pmd(pmd, pmd_mkhuge(pfn_pmd(cpa->pfn,
1389 					canon_pgprot(pmd_pgprot))));
1390 
1391 		start	  += PMD_SIZE;
1392 		cpa->pfn  += PMD_SIZE >> PAGE_SHIFT;
1393 		cur_pages += PMD_SIZE >> PAGE_SHIFT;
1394 	}
1395 
1396 	/*
1397 	 * Map trailing 4K pages.
1398 	 */
1399 	if (start < end) {
1400 		pmd = pmd_offset(pud, start);
1401 		if (pmd_none(*pmd))
1402 			if (alloc_pte_page(pmd))
1403 				return -1;
1404 
1405 		populate_pte(cpa, start, end, num_pages - cur_pages,
1406 			     pmd, pgprot);
1407 	}
1408 	return num_pages;
1409 }
1410 
populate_pud(struct cpa_data * cpa,unsigned long start,p4d_t * p4d,pgprot_t pgprot)1411 static int populate_pud(struct cpa_data *cpa, unsigned long start, p4d_t *p4d,
1412 			pgprot_t pgprot)
1413 {
1414 	pud_t *pud;
1415 	unsigned long end;
1416 	long cur_pages = 0;
1417 	pgprot_t pud_pgprot;
1418 
1419 	end = start + (cpa->numpages << PAGE_SHIFT);
1420 
1421 	/*
1422 	 * Not on a Gb page boundary? => map everything up to it with
1423 	 * smaller pages.
1424 	 */
1425 	if (start & (PUD_SIZE - 1)) {
1426 		unsigned long pre_end;
1427 		unsigned long next_page = (start + PUD_SIZE) & PUD_MASK;
1428 
1429 		pre_end   = min_t(unsigned long, end, next_page);
1430 		cur_pages = (pre_end - start) >> PAGE_SHIFT;
1431 		cur_pages = min_t(int, (int)cpa->numpages, cur_pages);
1432 
1433 		pud = pud_offset(p4d, start);
1434 
1435 		/*
1436 		 * Need a PMD page?
1437 		 */
1438 		if (pud_none(*pud))
1439 			if (alloc_pmd_page(pud))
1440 				return -1;
1441 
1442 		cur_pages = populate_pmd(cpa, start, pre_end, cur_pages,
1443 					 pud, pgprot);
1444 		if (cur_pages < 0)
1445 			return cur_pages;
1446 
1447 		start = pre_end;
1448 	}
1449 
1450 	/* We mapped them all? */
1451 	if (cpa->numpages == cur_pages)
1452 		return cur_pages;
1453 
1454 	pud = pud_offset(p4d, start);
1455 	pud_pgprot = pgprot_4k_2_large(pgprot);
1456 
1457 	/*
1458 	 * Map everything starting from the Gb boundary, possibly with 1G pages
1459 	 */
1460 	while (boot_cpu_has(X86_FEATURE_GBPAGES) && end - start >= PUD_SIZE) {
1461 		set_pud(pud, pud_mkhuge(pfn_pud(cpa->pfn,
1462 				   canon_pgprot(pud_pgprot))));
1463 
1464 		start	  += PUD_SIZE;
1465 		cpa->pfn  += PUD_SIZE >> PAGE_SHIFT;
1466 		cur_pages += PUD_SIZE >> PAGE_SHIFT;
1467 		pud++;
1468 	}
1469 
1470 	/* Map trailing leftover */
1471 	if (start < end) {
1472 		long tmp;
1473 
1474 		pud = pud_offset(p4d, start);
1475 		if (pud_none(*pud))
1476 			if (alloc_pmd_page(pud))
1477 				return -1;
1478 
1479 		tmp = populate_pmd(cpa, start, end, cpa->numpages - cur_pages,
1480 				   pud, pgprot);
1481 		if (tmp < 0)
1482 			return cur_pages;
1483 
1484 		cur_pages += tmp;
1485 	}
1486 	return cur_pages;
1487 }
1488 
1489 /*
1490  * Restrictions for kernel page table do not necessarily apply when mapping in
1491  * an alternate PGD.
1492  */
populate_pgd(struct cpa_data * cpa,unsigned long addr)1493 static int populate_pgd(struct cpa_data *cpa, unsigned long addr)
1494 {
1495 	pgprot_t pgprot = __pgprot(_KERNPG_TABLE);
1496 	pud_t *pud = NULL;	/* shut up gcc */
1497 	p4d_t *p4d;
1498 	pgd_t *pgd_entry;
1499 	long ret;
1500 
1501 	pgd_entry = cpa->pgd + pgd_index(addr);
1502 
1503 	if (pgd_none(*pgd_entry)) {
1504 		p4d = (p4d_t *)get_zeroed_page(GFP_KERNEL);
1505 		if (!p4d)
1506 			return -1;
1507 
1508 		set_pgd(pgd_entry, __pgd(__pa(p4d) | _KERNPG_TABLE));
1509 	}
1510 
1511 	/*
1512 	 * Allocate a PUD page and hand it down for mapping.
1513 	 */
1514 	p4d = p4d_offset(pgd_entry, addr);
1515 	if (p4d_none(*p4d)) {
1516 		pud = (pud_t *)get_zeroed_page(GFP_KERNEL);
1517 		if (!pud)
1518 			return -1;
1519 
1520 		set_p4d(p4d, __p4d(__pa(pud) | _KERNPG_TABLE));
1521 	}
1522 
1523 	pgprot_val(pgprot) &= ~pgprot_val(cpa->mask_clr);
1524 	pgprot_val(pgprot) |=  pgprot_val(cpa->mask_set);
1525 
1526 	ret = populate_pud(cpa, addr, p4d, pgprot);
1527 	if (ret < 0) {
1528 		/*
1529 		 * Leave the PUD page in place in case some other CPU or thread
1530 		 * already found it, but remove any useless entries we just
1531 		 * added to it.
1532 		 */
1533 		unmap_pud_range(p4d, addr,
1534 				addr + (cpa->numpages << PAGE_SHIFT));
1535 		return ret;
1536 	}
1537 
1538 	cpa->numpages = ret;
1539 	return 0;
1540 }
1541 
__cpa_process_fault(struct cpa_data * cpa,unsigned long vaddr,int primary)1542 static int __cpa_process_fault(struct cpa_data *cpa, unsigned long vaddr,
1543 			       int primary)
1544 {
1545 	if (cpa->pgd) {
1546 		/*
1547 		 * Right now, we only execute this code path when mapping
1548 		 * the EFI virtual memory map regions, no other users
1549 		 * provide a ->pgd value. This may change in the future.
1550 		 */
1551 		return populate_pgd(cpa, vaddr);
1552 	}
1553 
1554 	/*
1555 	 * Ignore all non primary paths.
1556 	 */
1557 	if (!primary) {
1558 		cpa->numpages = 1;
1559 		return 0;
1560 	}
1561 
1562 	/*
1563 	 * Ignore the NULL PTE for kernel identity mapping, as it is expected
1564 	 * to have holes.
1565 	 * Also set numpages to '1' indicating that we processed cpa req for
1566 	 * one virtual address page and its pfn. TBD: numpages can be set based
1567 	 * on the initial value and the level returned by lookup_address().
1568 	 */
1569 	if (within(vaddr, PAGE_OFFSET,
1570 		   PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT))) {
1571 		cpa->numpages = 1;
1572 		cpa->pfn = __pa(vaddr) >> PAGE_SHIFT;
1573 		return 0;
1574 
1575 	} else if (__cpa_pfn_in_highmap(cpa->pfn)) {
1576 		/* Faults in the highmap are OK, so do not warn: */
1577 		return -EFAULT;
1578 	} else {
1579 		WARN(1, KERN_WARNING "CPA: called for zero pte. "
1580 			"vaddr = %lx cpa->vaddr = %lx\n", vaddr,
1581 			*cpa->vaddr);
1582 
1583 		return -EFAULT;
1584 	}
1585 }
1586 
__change_page_attr(struct cpa_data * cpa,int primary)1587 static int __change_page_attr(struct cpa_data *cpa, int primary)
1588 {
1589 	unsigned long address;
1590 	int do_split, err;
1591 	unsigned int level;
1592 	pte_t *kpte, old_pte;
1593 
1594 	address = __cpa_addr(cpa, cpa->curpage);
1595 repeat:
1596 	kpte = _lookup_address_cpa(cpa, address, &level);
1597 	if (!kpte)
1598 		return __cpa_process_fault(cpa, address, primary);
1599 
1600 	old_pte = *kpte;
1601 	if (pte_none(old_pte))
1602 		return __cpa_process_fault(cpa, address, primary);
1603 
1604 	if (level == PG_LEVEL_4K) {
1605 		pte_t new_pte;
1606 		pgprot_t old_prot = pte_pgprot(old_pte);
1607 		pgprot_t new_prot = pte_pgprot(old_pte);
1608 		unsigned long pfn = pte_pfn(old_pte);
1609 
1610 		pgprot_val(new_prot) &= ~pgprot_val(cpa->mask_clr);
1611 		pgprot_val(new_prot) |= pgprot_val(cpa->mask_set);
1612 
1613 		cpa_inc_4k_install();
1614 		/* Hand in lpsize = 0 to enforce the protection mechanism */
1615 		new_prot = static_protections(new_prot, address, pfn, 1, 0,
1616 					      CPA_PROTECT);
1617 
1618 		new_prot = verify_rwx(old_prot, new_prot, address, pfn, 1);
1619 
1620 		new_prot = pgprot_clear_protnone_bits(new_prot);
1621 
1622 		/*
1623 		 * We need to keep the pfn from the existing PTE,
1624 		 * after all we're only going to change it's attributes
1625 		 * not the memory it points to
1626 		 */
1627 		new_pte = pfn_pte(pfn, new_prot);
1628 		cpa->pfn = pfn;
1629 		/*
1630 		 * Do we really change anything ?
1631 		 */
1632 		if (pte_val(old_pte) != pte_val(new_pte)) {
1633 			set_pte_atomic(kpte, new_pte);
1634 			cpa->flags |= CPA_FLUSHTLB;
1635 		}
1636 		cpa->numpages = 1;
1637 		return 0;
1638 	}
1639 
1640 	/*
1641 	 * Check, whether we can keep the large page intact
1642 	 * and just change the pte:
1643 	 */
1644 	do_split = should_split_large_page(kpte, address, cpa);
1645 	/*
1646 	 * When the range fits into the existing large page,
1647 	 * return. cp->numpages and cpa->tlbflush have been updated in
1648 	 * try_large_page:
1649 	 */
1650 	if (do_split <= 0)
1651 		return do_split;
1652 
1653 	/*
1654 	 * We have to split the large page:
1655 	 */
1656 	err = split_large_page(cpa, kpte, address);
1657 	if (!err)
1658 		goto repeat;
1659 
1660 	return err;
1661 }
1662 
1663 static int __change_page_attr_set_clr(struct cpa_data *cpa, int primary);
1664 
1665 /*
1666  * Check the directmap and "high kernel map" 'aliases'.
1667  */
cpa_process_alias(struct cpa_data * cpa)1668 static int cpa_process_alias(struct cpa_data *cpa)
1669 {
1670 	struct cpa_data alias_cpa;
1671 	unsigned long laddr = (unsigned long)__va(cpa->pfn << PAGE_SHIFT);
1672 	unsigned long vaddr;
1673 	int ret;
1674 
1675 	if (!pfn_range_is_mapped(cpa->pfn, cpa->pfn + 1))
1676 		return 0;
1677 
1678 	/*
1679 	 * No need to redo, when the primary call touched the direct
1680 	 * mapping already:
1681 	 */
1682 	vaddr = __cpa_addr(cpa, cpa->curpage);
1683 	if (!(within(vaddr, PAGE_OFFSET,
1684 		    PAGE_OFFSET + (max_pfn_mapped << PAGE_SHIFT)))) {
1685 
1686 		alias_cpa = *cpa;
1687 		alias_cpa.vaddr = &laddr;
1688 		alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1689 		alias_cpa.curpage = 0;
1690 
1691 		/* Directmap always has NX set, do not modify. */
1692 		if (__supported_pte_mask & _PAGE_NX) {
1693 			alias_cpa.mask_clr.pgprot &= ~_PAGE_NX;
1694 			alias_cpa.mask_set.pgprot &= ~_PAGE_NX;
1695 		}
1696 
1697 		cpa->force_flush_all = 1;
1698 
1699 		ret = __change_page_attr_set_clr(&alias_cpa, 0);
1700 		if (ret)
1701 			return ret;
1702 	}
1703 
1704 #ifdef CONFIG_X86_64
1705 	/*
1706 	 * If the primary call didn't touch the high mapping already
1707 	 * and the physical address is inside the kernel map, we need
1708 	 * to touch the high mapped kernel as well:
1709 	 */
1710 	if (!within(vaddr, (unsigned long)_text, _brk_end) &&
1711 	    __cpa_pfn_in_highmap(cpa->pfn)) {
1712 		unsigned long temp_cpa_vaddr = (cpa->pfn << PAGE_SHIFT) +
1713 					       __START_KERNEL_map - phys_base;
1714 		alias_cpa = *cpa;
1715 		alias_cpa.vaddr = &temp_cpa_vaddr;
1716 		alias_cpa.flags &= ~(CPA_PAGES_ARRAY | CPA_ARRAY);
1717 		alias_cpa.curpage = 0;
1718 
1719 		/*
1720 		 * [_text, _brk_end) also covers data, do not modify NX except
1721 		 * in cases where the highmap is the primary target.
1722 		 */
1723 		if (__supported_pte_mask & _PAGE_NX) {
1724 			alias_cpa.mask_clr.pgprot &= ~_PAGE_NX;
1725 			alias_cpa.mask_set.pgprot &= ~_PAGE_NX;
1726 		}
1727 
1728 		cpa->force_flush_all = 1;
1729 		/*
1730 		 * The high mapping range is imprecise, so ignore the
1731 		 * return value.
1732 		 */
1733 		__change_page_attr_set_clr(&alias_cpa, 0);
1734 	}
1735 #endif
1736 
1737 	return 0;
1738 }
1739 
__change_page_attr_set_clr(struct cpa_data * cpa,int primary)1740 static int __change_page_attr_set_clr(struct cpa_data *cpa, int primary)
1741 {
1742 	unsigned long numpages = cpa->numpages;
1743 	unsigned long rempages = numpages;
1744 	int ret = 0;
1745 
1746 	/*
1747 	 * No changes, easy!
1748 	 */
1749 	if (!(pgprot_val(cpa->mask_set) | pgprot_val(cpa->mask_clr)) &&
1750 	    !cpa->force_split)
1751 		return ret;
1752 
1753 	while (rempages) {
1754 		/*
1755 		 * Store the remaining nr of pages for the large page
1756 		 * preservation check.
1757 		 */
1758 		cpa->numpages = rempages;
1759 		/* for array changes, we can't use large page */
1760 		if (cpa->flags & (CPA_ARRAY | CPA_PAGES_ARRAY))
1761 			cpa->numpages = 1;
1762 
1763 		if (!debug_pagealloc_enabled())
1764 			spin_lock(&cpa_lock);
1765 		ret = __change_page_attr(cpa, primary);
1766 		if (!debug_pagealloc_enabled())
1767 			spin_unlock(&cpa_lock);
1768 		if (ret)
1769 			goto out;
1770 
1771 		if (primary && !(cpa->flags & CPA_NO_CHECK_ALIAS)) {
1772 			ret = cpa_process_alias(cpa);
1773 			if (ret)
1774 				goto out;
1775 		}
1776 
1777 		/*
1778 		 * Adjust the number of pages with the result of the
1779 		 * CPA operation. Either a large page has been
1780 		 * preserved or a single page update happened.
1781 		 */
1782 		BUG_ON(cpa->numpages > rempages || !cpa->numpages);
1783 		rempages -= cpa->numpages;
1784 		cpa->curpage += cpa->numpages;
1785 	}
1786 
1787 out:
1788 	/* Restore the original numpages */
1789 	cpa->numpages = numpages;
1790 	return ret;
1791 }
1792 
change_page_attr_set_clr(unsigned long * addr,int numpages,pgprot_t mask_set,pgprot_t mask_clr,int force_split,int in_flag,struct page ** pages)1793 static int change_page_attr_set_clr(unsigned long *addr, int numpages,
1794 				    pgprot_t mask_set, pgprot_t mask_clr,
1795 				    int force_split, int in_flag,
1796 				    struct page **pages)
1797 {
1798 	struct cpa_data cpa;
1799 	int ret, cache;
1800 
1801 	memset(&cpa, 0, sizeof(cpa));
1802 
1803 	/*
1804 	 * Check, if we are requested to set a not supported
1805 	 * feature.  Clearing non-supported features is OK.
1806 	 */
1807 	mask_set = canon_pgprot(mask_set);
1808 
1809 	if (!pgprot_val(mask_set) && !pgprot_val(mask_clr) && !force_split)
1810 		return 0;
1811 
1812 	/* Ensure we are PAGE_SIZE aligned */
1813 	if (in_flag & CPA_ARRAY) {
1814 		int i;
1815 		for (i = 0; i < numpages; i++) {
1816 			if (addr[i] & ~PAGE_MASK) {
1817 				addr[i] &= PAGE_MASK;
1818 				WARN_ON_ONCE(1);
1819 			}
1820 		}
1821 	} else if (!(in_flag & CPA_PAGES_ARRAY)) {
1822 		/*
1823 		 * in_flag of CPA_PAGES_ARRAY implies it is aligned.
1824 		 * No need to check in that case
1825 		 */
1826 		if (*addr & ~PAGE_MASK) {
1827 			*addr &= PAGE_MASK;
1828 			/*
1829 			 * People should not be passing in unaligned addresses:
1830 			 */
1831 			WARN_ON_ONCE(1);
1832 		}
1833 	}
1834 
1835 	/* Must avoid aliasing mappings in the highmem code */
1836 	kmap_flush_unused();
1837 
1838 	vm_unmap_aliases();
1839 
1840 	cpa.vaddr = addr;
1841 	cpa.pages = pages;
1842 	cpa.numpages = numpages;
1843 	cpa.mask_set = mask_set;
1844 	cpa.mask_clr = mask_clr;
1845 	cpa.flags = in_flag;
1846 	cpa.curpage = 0;
1847 	cpa.force_split = force_split;
1848 
1849 	ret = __change_page_attr_set_clr(&cpa, 1);
1850 
1851 	/*
1852 	 * Check whether we really changed something:
1853 	 */
1854 	if (!(cpa.flags & CPA_FLUSHTLB))
1855 		goto out;
1856 
1857 	/*
1858 	 * No need to flush, when we did not set any of the caching
1859 	 * attributes:
1860 	 */
1861 	cache = !!pgprot2cachemode(mask_set);
1862 
1863 	/*
1864 	 * On error; flush everything to be sure.
1865 	 */
1866 	if (ret) {
1867 		cpa_flush_all(cache);
1868 		goto out;
1869 	}
1870 
1871 	cpa_flush(&cpa, cache);
1872 out:
1873 	return ret;
1874 }
1875 
change_page_attr_set(unsigned long * addr,int numpages,pgprot_t mask,int array)1876 static inline int change_page_attr_set(unsigned long *addr, int numpages,
1877 				       pgprot_t mask, int array)
1878 {
1879 	return change_page_attr_set_clr(addr, numpages, mask, __pgprot(0), 0,
1880 		(array ? CPA_ARRAY : 0), NULL);
1881 }
1882 
change_page_attr_clear(unsigned long * addr,int numpages,pgprot_t mask,int array)1883 static inline int change_page_attr_clear(unsigned long *addr, int numpages,
1884 					 pgprot_t mask, int array)
1885 {
1886 	return change_page_attr_set_clr(addr, numpages, __pgprot(0), mask, 0,
1887 		(array ? CPA_ARRAY : 0), NULL);
1888 }
1889 
cpa_set_pages_array(struct page ** pages,int numpages,pgprot_t mask)1890 static inline int cpa_set_pages_array(struct page **pages, int numpages,
1891 				       pgprot_t mask)
1892 {
1893 	return change_page_attr_set_clr(NULL, numpages, mask, __pgprot(0), 0,
1894 		CPA_PAGES_ARRAY, pages);
1895 }
1896 
cpa_clear_pages_array(struct page ** pages,int numpages,pgprot_t mask)1897 static inline int cpa_clear_pages_array(struct page **pages, int numpages,
1898 					 pgprot_t mask)
1899 {
1900 	return change_page_attr_set_clr(NULL, numpages, __pgprot(0), mask, 0,
1901 		CPA_PAGES_ARRAY, pages);
1902 }
1903 
1904 /*
1905  * __set_memory_prot is an internal helper for callers that have been passed
1906  * a pgprot_t value from upper layers and a reservation has already been taken.
1907  * If you want to set the pgprot to a specific page protocol, use the
1908  * set_memory_xx() functions.
1909  */
__set_memory_prot(unsigned long addr,int numpages,pgprot_t prot)1910 int __set_memory_prot(unsigned long addr, int numpages, pgprot_t prot)
1911 {
1912 	return change_page_attr_set_clr(&addr, numpages, prot,
1913 					__pgprot(~pgprot_val(prot)), 0, 0,
1914 					NULL);
1915 }
1916 
_set_memory_uc(unsigned long addr,int numpages)1917 int _set_memory_uc(unsigned long addr, int numpages)
1918 {
1919 	/*
1920 	 * for now UC MINUS. see comments in ioremap()
1921 	 * If you really need strong UC use ioremap_uc(), but note
1922 	 * that you cannot override IO areas with set_memory_*() as
1923 	 * these helpers cannot work with IO memory.
1924 	 */
1925 	return change_page_attr_set(&addr, numpages,
1926 				    cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1927 				    0);
1928 }
1929 
set_memory_uc(unsigned long addr,int numpages)1930 int set_memory_uc(unsigned long addr, int numpages)
1931 {
1932 	int ret;
1933 
1934 	/*
1935 	 * for now UC MINUS. see comments in ioremap()
1936 	 */
1937 	ret = memtype_reserve(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1938 			      _PAGE_CACHE_MODE_UC_MINUS, NULL);
1939 	if (ret)
1940 		goto out_err;
1941 
1942 	ret = _set_memory_uc(addr, numpages);
1943 	if (ret)
1944 		goto out_free;
1945 
1946 	return 0;
1947 
1948 out_free:
1949 	memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1950 out_err:
1951 	return ret;
1952 }
1953 EXPORT_SYMBOL(set_memory_uc);
1954 
_set_memory_wc(unsigned long addr,int numpages)1955 int _set_memory_wc(unsigned long addr, int numpages)
1956 {
1957 	int ret;
1958 
1959 	ret = change_page_attr_set(&addr, numpages,
1960 				   cachemode2pgprot(_PAGE_CACHE_MODE_UC_MINUS),
1961 				   0);
1962 	if (!ret) {
1963 		ret = change_page_attr_set_clr(&addr, numpages,
1964 					       cachemode2pgprot(_PAGE_CACHE_MODE_WC),
1965 					       __pgprot(_PAGE_CACHE_MASK),
1966 					       0, 0, NULL);
1967 	}
1968 	return ret;
1969 }
1970 
set_memory_wc(unsigned long addr,int numpages)1971 int set_memory_wc(unsigned long addr, int numpages)
1972 {
1973 	int ret;
1974 
1975 	ret = memtype_reserve(__pa(addr), __pa(addr) + numpages * PAGE_SIZE,
1976 		_PAGE_CACHE_MODE_WC, NULL);
1977 	if (ret)
1978 		return ret;
1979 
1980 	ret = _set_memory_wc(addr, numpages);
1981 	if (ret)
1982 		memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
1983 
1984 	return ret;
1985 }
1986 EXPORT_SYMBOL(set_memory_wc);
1987 
_set_memory_wt(unsigned long addr,int numpages)1988 int _set_memory_wt(unsigned long addr, int numpages)
1989 {
1990 	return change_page_attr_set(&addr, numpages,
1991 				    cachemode2pgprot(_PAGE_CACHE_MODE_WT), 0);
1992 }
1993 
_set_memory_wb(unsigned long addr,int numpages)1994 int _set_memory_wb(unsigned long addr, int numpages)
1995 {
1996 	/* WB cache mode is hard wired to all cache attribute bits being 0 */
1997 	return change_page_attr_clear(&addr, numpages,
1998 				      __pgprot(_PAGE_CACHE_MASK), 0);
1999 }
2000 
set_memory_wb(unsigned long addr,int numpages)2001 int set_memory_wb(unsigned long addr, int numpages)
2002 {
2003 	int ret;
2004 
2005 	ret = _set_memory_wb(addr, numpages);
2006 	if (ret)
2007 		return ret;
2008 
2009 	memtype_free(__pa(addr), __pa(addr) + numpages * PAGE_SIZE);
2010 	return 0;
2011 }
2012 EXPORT_SYMBOL(set_memory_wb);
2013 
2014 /* Prevent speculative access to a page by marking it not-present */
2015 #ifdef CONFIG_X86_64
set_mce_nospec(unsigned long pfn)2016 int set_mce_nospec(unsigned long pfn)
2017 {
2018 	unsigned long decoy_addr;
2019 	int rc;
2020 
2021 	/* SGX pages are not in the 1:1 map */
2022 	if (arch_is_platform_page(pfn << PAGE_SHIFT))
2023 		return 0;
2024 	/*
2025 	 * We would like to just call:
2026 	 *      set_memory_XX((unsigned long)pfn_to_kaddr(pfn), 1);
2027 	 * but doing that would radically increase the odds of a
2028 	 * speculative access to the poison page because we'd have
2029 	 * the virtual address of the kernel 1:1 mapping sitting
2030 	 * around in registers.
2031 	 * Instead we get tricky.  We create a non-canonical address
2032 	 * that looks just like the one we want, but has bit 63 flipped.
2033 	 * This relies on set_memory_XX() properly sanitizing any __pa()
2034 	 * results with __PHYSICAL_MASK or PTE_PFN_MASK.
2035 	 */
2036 	decoy_addr = (pfn << PAGE_SHIFT) + (PAGE_OFFSET ^ BIT(63));
2037 
2038 	rc = set_memory_np(decoy_addr, 1);
2039 	if (rc)
2040 		pr_warn("Could not invalidate pfn=0x%lx from 1:1 map\n", pfn);
2041 	return rc;
2042 }
2043 
set_memory_p(unsigned long * addr,int numpages)2044 static int set_memory_p(unsigned long *addr, int numpages)
2045 {
2046 	return change_page_attr_set(addr, numpages, __pgprot(_PAGE_PRESENT), 0);
2047 }
2048 
2049 /* Restore full speculative operation to the pfn. */
clear_mce_nospec(unsigned long pfn)2050 int clear_mce_nospec(unsigned long pfn)
2051 {
2052 	unsigned long addr = (unsigned long) pfn_to_kaddr(pfn);
2053 
2054 	return set_memory_p(&addr, 1);
2055 }
2056 EXPORT_SYMBOL_GPL(clear_mce_nospec);
2057 #endif /* CONFIG_X86_64 */
2058 
set_memory_x(unsigned long addr,int numpages)2059 int set_memory_x(unsigned long addr, int numpages)
2060 {
2061 	if (!(__supported_pte_mask & _PAGE_NX))
2062 		return 0;
2063 
2064 	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_NX), 0);
2065 }
2066 
set_memory_nx(unsigned long addr,int numpages)2067 int set_memory_nx(unsigned long addr, int numpages)
2068 {
2069 	if (!(__supported_pte_mask & _PAGE_NX))
2070 		return 0;
2071 
2072 	return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_NX), 0);
2073 }
2074 
set_memory_ro(unsigned long addr,int numpages)2075 int set_memory_ro(unsigned long addr, int numpages)
2076 {
2077 	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_RW | _PAGE_DIRTY), 0);
2078 }
2079 
set_memory_rox(unsigned long addr,int numpages)2080 int set_memory_rox(unsigned long addr, int numpages)
2081 {
2082 	pgprot_t clr = __pgprot(_PAGE_RW | _PAGE_DIRTY);
2083 
2084 	if (__supported_pte_mask & _PAGE_NX)
2085 		clr.pgprot |= _PAGE_NX;
2086 
2087 	return change_page_attr_clear(&addr, numpages, clr, 0);
2088 }
2089 
set_memory_rw(unsigned long addr,int numpages)2090 int set_memory_rw(unsigned long addr, int numpages)
2091 {
2092 	return change_page_attr_set(&addr, numpages, __pgprot(_PAGE_RW), 0);
2093 }
2094 
set_memory_np(unsigned long addr,int numpages)2095 int set_memory_np(unsigned long addr, int numpages)
2096 {
2097 	return change_page_attr_clear(&addr, numpages, __pgprot(_PAGE_PRESENT), 0);
2098 }
2099 
set_memory_np_noalias(unsigned long addr,int numpages)2100 int set_memory_np_noalias(unsigned long addr, int numpages)
2101 {
2102 	return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
2103 					__pgprot(_PAGE_PRESENT), 0,
2104 					CPA_NO_CHECK_ALIAS, NULL);
2105 }
2106 
set_memory_4k(unsigned long addr,int numpages)2107 int set_memory_4k(unsigned long addr, int numpages)
2108 {
2109 	return change_page_attr_set_clr(&addr, numpages, __pgprot(0),
2110 					__pgprot(0), 1, 0, NULL);
2111 }
2112 
set_memory_nonglobal(unsigned long addr,int numpages)2113 int set_memory_nonglobal(unsigned long addr, int numpages)
2114 {
2115 	return change_page_attr_clear(&addr, numpages,
2116 				      __pgprot(_PAGE_GLOBAL), 0);
2117 }
2118 
set_memory_global(unsigned long addr,int numpages)2119 int set_memory_global(unsigned long addr, int numpages)
2120 {
2121 	return change_page_attr_set(&addr, numpages,
2122 				    __pgprot(_PAGE_GLOBAL), 0);
2123 }
2124 
2125 /*
2126  * __set_memory_enc_pgtable() is used for the hypervisors that get
2127  * informed about "encryption" status via page tables.
2128  */
__set_memory_enc_pgtable(unsigned long addr,int numpages,bool enc)2129 static int __set_memory_enc_pgtable(unsigned long addr, int numpages, bool enc)
2130 {
2131 	pgprot_t empty = __pgprot(0);
2132 	struct cpa_data cpa;
2133 	int ret;
2134 
2135 	/* Should not be working on unaligned addresses */
2136 	if (WARN_ONCE(addr & ~PAGE_MASK, "misaligned address: %#lx\n", addr))
2137 		addr &= PAGE_MASK;
2138 
2139 	memset(&cpa, 0, sizeof(cpa));
2140 	cpa.vaddr = &addr;
2141 	cpa.numpages = numpages;
2142 	cpa.mask_set = enc ? pgprot_encrypted(empty) : pgprot_decrypted(empty);
2143 	cpa.mask_clr = enc ? pgprot_decrypted(empty) : pgprot_encrypted(empty);
2144 	cpa.pgd = init_mm.pgd;
2145 
2146 	/* Must avoid aliasing mappings in the highmem code */
2147 	kmap_flush_unused();
2148 	vm_unmap_aliases();
2149 
2150 	/* Flush the caches as needed before changing the encryption attribute. */
2151 	if (x86_platform.guest.enc_tlb_flush_required(enc))
2152 		cpa_flush(&cpa, x86_platform.guest.enc_cache_flush_required());
2153 
2154 	/* Notify hypervisor that we are about to set/clr encryption attribute. */
2155 	if (!x86_platform.guest.enc_status_change_prepare(addr, numpages, enc))
2156 		return -EIO;
2157 
2158 	ret = __change_page_attr_set_clr(&cpa, 1);
2159 
2160 	/*
2161 	 * After changing the encryption attribute, we need to flush TLBs again
2162 	 * in case any speculative TLB caching occurred (but no need to flush
2163 	 * caches again).  We could just use cpa_flush_all(), but in case TLB
2164 	 * flushing gets optimized in the cpa_flush() path use the same logic
2165 	 * as above.
2166 	 */
2167 	cpa_flush(&cpa, 0);
2168 
2169 	/* Notify hypervisor that we have successfully set/clr encryption attribute. */
2170 	if (!ret) {
2171 		if (!x86_platform.guest.enc_status_change_finish(addr, numpages, enc))
2172 			ret = -EIO;
2173 	}
2174 
2175 	return ret;
2176 }
2177 
__set_memory_enc_dec(unsigned long addr,int numpages,bool enc)2178 static int __set_memory_enc_dec(unsigned long addr, int numpages, bool enc)
2179 {
2180 	if (cc_platform_has(CC_ATTR_MEM_ENCRYPT))
2181 		return __set_memory_enc_pgtable(addr, numpages, enc);
2182 
2183 	return 0;
2184 }
2185 
set_memory_encrypted(unsigned long addr,int numpages)2186 int set_memory_encrypted(unsigned long addr, int numpages)
2187 {
2188 	return __set_memory_enc_dec(addr, numpages, true);
2189 }
2190 EXPORT_SYMBOL_GPL(set_memory_encrypted);
2191 
set_memory_decrypted(unsigned long addr,int numpages)2192 int set_memory_decrypted(unsigned long addr, int numpages)
2193 {
2194 	return __set_memory_enc_dec(addr, numpages, false);
2195 }
2196 EXPORT_SYMBOL_GPL(set_memory_decrypted);
2197 
set_pages_uc(struct page * page,int numpages)2198 int set_pages_uc(struct page *page, int numpages)
2199 {
2200 	unsigned long addr = (unsigned long)page_address(page);
2201 
2202 	return set_memory_uc(addr, numpages);
2203 }
2204 EXPORT_SYMBOL(set_pages_uc);
2205 
_set_pages_array(struct page ** pages,int numpages,enum page_cache_mode new_type)2206 static int _set_pages_array(struct page **pages, int numpages,
2207 		enum page_cache_mode new_type)
2208 {
2209 	unsigned long start;
2210 	unsigned long end;
2211 	enum page_cache_mode set_type;
2212 	int i;
2213 	int free_idx;
2214 	int ret;
2215 
2216 	for (i = 0; i < numpages; i++) {
2217 		if (PageHighMem(pages[i]))
2218 			continue;
2219 		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2220 		end = start + PAGE_SIZE;
2221 		if (memtype_reserve(start, end, new_type, NULL))
2222 			goto err_out;
2223 	}
2224 
2225 	/* If WC, set to UC- first and then WC */
2226 	set_type = (new_type == _PAGE_CACHE_MODE_WC) ?
2227 				_PAGE_CACHE_MODE_UC_MINUS : new_type;
2228 
2229 	ret = cpa_set_pages_array(pages, numpages,
2230 				  cachemode2pgprot(set_type));
2231 	if (!ret && new_type == _PAGE_CACHE_MODE_WC)
2232 		ret = change_page_attr_set_clr(NULL, numpages,
2233 					       cachemode2pgprot(
2234 						_PAGE_CACHE_MODE_WC),
2235 					       __pgprot(_PAGE_CACHE_MASK),
2236 					       0, CPA_PAGES_ARRAY, pages);
2237 	if (ret)
2238 		goto err_out;
2239 	return 0; /* Success */
2240 err_out:
2241 	free_idx = i;
2242 	for (i = 0; i < free_idx; i++) {
2243 		if (PageHighMem(pages[i]))
2244 			continue;
2245 		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2246 		end = start + PAGE_SIZE;
2247 		memtype_free(start, end);
2248 	}
2249 	return -EINVAL;
2250 }
2251 
set_pages_array_uc(struct page ** pages,int numpages)2252 int set_pages_array_uc(struct page **pages, int numpages)
2253 {
2254 	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_UC_MINUS);
2255 }
2256 EXPORT_SYMBOL(set_pages_array_uc);
2257 
set_pages_array_wc(struct page ** pages,int numpages)2258 int set_pages_array_wc(struct page **pages, int numpages)
2259 {
2260 	return _set_pages_array(pages, numpages, _PAGE_CACHE_MODE_WC);
2261 }
2262 EXPORT_SYMBOL(set_pages_array_wc);
2263 
set_pages_wb(struct page * page,int numpages)2264 int set_pages_wb(struct page *page, int numpages)
2265 {
2266 	unsigned long addr = (unsigned long)page_address(page);
2267 
2268 	return set_memory_wb(addr, numpages);
2269 }
2270 EXPORT_SYMBOL(set_pages_wb);
2271 
set_pages_array_wb(struct page ** pages,int numpages)2272 int set_pages_array_wb(struct page **pages, int numpages)
2273 {
2274 	int retval;
2275 	unsigned long start;
2276 	unsigned long end;
2277 	int i;
2278 
2279 	/* WB cache mode is hard wired to all cache attribute bits being 0 */
2280 	retval = cpa_clear_pages_array(pages, numpages,
2281 			__pgprot(_PAGE_CACHE_MASK));
2282 	if (retval)
2283 		return retval;
2284 
2285 	for (i = 0; i < numpages; i++) {
2286 		if (PageHighMem(pages[i]))
2287 			continue;
2288 		start = page_to_pfn(pages[i]) << PAGE_SHIFT;
2289 		end = start + PAGE_SIZE;
2290 		memtype_free(start, end);
2291 	}
2292 
2293 	return 0;
2294 }
2295 EXPORT_SYMBOL(set_pages_array_wb);
2296 
set_pages_ro(struct page * page,int numpages)2297 int set_pages_ro(struct page *page, int numpages)
2298 {
2299 	unsigned long addr = (unsigned long)page_address(page);
2300 
2301 	return set_memory_ro(addr, numpages);
2302 }
2303 
set_pages_rw(struct page * page,int numpages)2304 int set_pages_rw(struct page *page, int numpages)
2305 {
2306 	unsigned long addr = (unsigned long)page_address(page);
2307 
2308 	return set_memory_rw(addr, numpages);
2309 }
2310 
__set_pages_p(struct page * page,int numpages)2311 static int __set_pages_p(struct page *page, int numpages)
2312 {
2313 	unsigned long tempaddr = (unsigned long) page_address(page);
2314 	struct cpa_data cpa = { .vaddr = &tempaddr,
2315 				.pgd = NULL,
2316 				.numpages = numpages,
2317 				.mask_set = __pgprot(_PAGE_PRESENT | _PAGE_RW),
2318 				.mask_clr = __pgprot(0),
2319 				.flags = CPA_NO_CHECK_ALIAS };
2320 
2321 	/*
2322 	 * No alias checking needed for setting present flag. otherwise,
2323 	 * we may need to break large pages for 64-bit kernel text
2324 	 * mappings (this adds to complexity if we want to do this from
2325 	 * atomic context especially). Let's keep it simple!
2326 	 */
2327 	return __change_page_attr_set_clr(&cpa, 1);
2328 }
2329 
__set_pages_np(struct page * page,int numpages)2330 static int __set_pages_np(struct page *page, int numpages)
2331 {
2332 	unsigned long tempaddr = (unsigned long) page_address(page);
2333 	struct cpa_data cpa = { .vaddr = &tempaddr,
2334 				.pgd = NULL,
2335 				.numpages = numpages,
2336 				.mask_set = __pgprot(0),
2337 				.mask_clr = __pgprot(_PAGE_PRESENT | _PAGE_RW),
2338 				.flags = CPA_NO_CHECK_ALIAS };
2339 
2340 	/*
2341 	 * No alias checking needed for setting not present flag. otherwise,
2342 	 * we may need to break large pages for 64-bit kernel text
2343 	 * mappings (this adds to complexity if we want to do this from
2344 	 * atomic context especially). Let's keep it simple!
2345 	 */
2346 	return __change_page_attr_set_clr(&cpa, 1);
2347 }
2348 
set_direct_map_invalid_noflush(struct page * page)2349 int set_direct_map_invalid_noflush(struct page *page)
2350 {
2351 	return __set_pages_np(page, 1);
2352 }
2353 
set_direct_map_default_noflush(struct page * page)2354 int set_direct_map_default_noflush(struct page *page)
2355 {
2356 	return __set_pages_p(page, 1);
2357 }
2358 
2359 #ifdef CONFIG_DEBUG_PAGEALLOC
__kernel_map_pages(struct page * page,int numpages,int enable)2360 void __kernel_map_pages(struct page *page, int numpages, int enable)
2361 {
2362 	if (PageHighMem(page))
2363 		return;
2364 	if (!enable) {
2365 		debug_check_no_locks_freed(page_address(page),
2366 					   numpages * PAGE_SIZE);
2367 	}
2368 
2369 	/*
2370 	 * The return value is ignored as the calls cannot fail.
2371 	 * Large pages for identity mappings are not used at boot time
2372 	 * and hence no memory allocations during large page split.
2373 	 */
2374 	if (enable)
2375 		__set_pages_p(page, numpages);
2376 	else
2377 		__set_pages_np(page, numpages);
2378 
2379 	/*
2380 	 * We should perform an IPI and flush all tlbs,
2381 	 * but that can deadlock->flush only current cpu.
2382 	 * Preemption needs to be disabled around __flush_tlb_all() due to
2383 	 * CR3 reload in __native_flush_tlb().
2384 	 */
2385 	preempt_disable();
2386 	__flush_tlb_all();
2387 	preempt_enable();
2388 
2389 	arch_flush_lazy_mmu_mode();
2390 }
2391 #endif /* CONFIG_DEBUG_PAGEALLOC */
2392 
kernel_page_present(struct page * page)2393 bool kernel_page_present(struct page *page)
2394 {
2395 	unsigned int level;
2396 	pte_t *pte;
2397 
2398 	if (PageHighMem(page))
2399 		return false;
2400 
2401 	pte = lookup_address((unsigned long)page_address(page), &level);
2402 	return (pte_val(*pte) & _PAGE_PRESENT);
2403 }
2404 
kernel_map_pages_in_pgd(pgd_t * pgd,u64 pfn,unsigned long address,unsigned numpages,unsigned long page_flags)2405 int __init kernel_map_pages_in_pgd(pgd_t *pgd, u64 pfn, unsigned long address,
2406 				   unsigned numpages, unsigned long page_flags)
2407 {
2408 	int retval = -EINVAL;
2409 
2410 	struct cpa_data cpa = {
2411 		.vaddr = &address,
2412 		.pfn = pfn,
2413 		.pgd = pgd,
2414 		.numpages = numpages,
2415 		.mask_set = __pgprot(0),
2416 		.mask_clr = __pgprot(~page_flags & (_PAGE_NX|_PAGE_RW)),
2417 		.flags = CPA_NO_CHECK_ALIAS,
2418 	};
2419 
2420 	WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2421 
2422 	if (!(__supported_pte_mask & _PAGE_NX))
2423 		goto out;
2424 
2425 	if (!(page_flags & _PAGE_ENC))
2426 		cpa.mask_clr = pgprot_encrypted(cpa.mask_clr);
2427 
2428 	cpa.mask_set = __pgprot(_PAGE_PRESENT | page_flags);
2429 
2430 	retval = __change_page_attr_set_clr(&cpa, 1);
2431 	__flush_tlb_all();
2432 
2433 out:
2434 	return retval;
2435 }
2436 
2437 /*
2438  * __flush_tlb_all() flushes mappings only on current CPU and hence this
2439  * function shouldn't be used in an SMP environment. Presently, it's used only
2440  * during boot (way before smp_init()) by EFI subsystem and hence is ok.
2441  */
kernel_unmap_pages_in_pgd(pgd_t * pgd,unsigned long address,unsigned long numpages)2442 int __init kernel_unmap_pages_in_pgd(pgd_t *pgd, unsigned long address,
2443 				     unsigned long numpages)
2444 {
2445 	int retval;
2446 
2447 	/*
2448 	 * The typical sequence for unmapping is to find a pte through
2449 	 * lookup_address_in_pgd() (ideally, it should never return NULL because
2450 	 * the address is already mapped) and change it's protections. As pfn is
2451 	 * the *target* of a mapping, it's not useful while unmapping.
2452 	 */
2453 	struct cpa_data cpa = {
2454 		.vaddr		= &address,
2455 		.pfn		= 0,
2456 		.pgd		= pgd,
2457 		.numpages	= numpages,
2458 		.mask_set	= __pgprot(0),
2459 		.mask_clr	= __pgprot(_PAGE_PRESENT | _PAGE_RW),
2460 		.flags		= CPA_NO_CHECK_ALIAS,
2461 	};
2462 
2463 	WARN_ONCE(num_online_cpus() > 1, "Don't call after initializing SMP");
2464 
2465 	retval = __change_page_attr_set_clr(&cpa, 1);
2466 	__flush_tlb_all();
2467 
2468 	return retval;
2469 }
2470 
2471 /*
2472  * The testcases use internal knowledge of the implementation that shouldn't
2473  * be exposed to the rest of the kernel. Include these directly here.
2474  */
2475 #ifdef CONFIG_CPA_DEBUG
2476 #include "cpa-test.c"
2477 #endif
2478