Linux Audio

Check our new training course

Linux BSP upgrade and security maintenance

Need help to get security updates for your Linux BSP?
Loading...
v6.2
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *
  4 *  Copyright (C) 1995  Linus Torvalds
  5 *
  6 *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  7 */
  8
  9#include <linux/signal.h>
 10#include <linux/sched.h>
 11#include <linux/kernel.h>
 12#include <linux/errno.h>
 13#include <linux/string.h>
 14#include <linux/types.h>
 15#include <linux/ptrace.h>
 16#include <linux/mman.h>
 17#include <linux/mm.h>
 18#include <linux/hugetlb.h>
 19#include <linux/swap.h>
 20#include <linux/smp.h>
 21#include <linux/init.h>
 22#include <linux/highmem.h>
 23#include <linux/pagemap.h>
 24#include <linux/pci.h>
 25#include <linux/pfn.h>
 26#include <linux/poison.h>
 27#include <linux/memblock.h>
 28#include <linux/proc_fs.h>
 29#include <linux/memory_hotplug.h>
 30#include <linux/initrd.h>
 31#include <linux/cpumask.h>
 32#include <linux/gfp.h>
 33
 34#include <asm/asm.h>
 35#include <asm/bios_ebda.h>
 36#include <asm/processor.h>
 37#include <linux/uaccess.h>
 38#include <asm/dma.h>
 39#include <asm/fixmap.h>
 40#include <asm/e820/api.h>
 41#include <asm/apic.h>
 42#include <asm/bugs.h>
 43#include <asm/tlb.h>
 44#include <asm/tlbflush.h>
 45#include <asm/olpc_ofw.h>
 46#include <asm/pgalloc.h>
 47#include <asm/sections.h>
 48#include <asm/paravirt.h>
 49#include <asm/setup.h>
 50#include <asm/set_memory.h>
 51#include <asm/page_types.h>
 52#include <asm/cpu_entry_area.h>
 53#include <asm/init.h>
 54#include <asm/pgtable_areas.h>
 55#include <asm/numa.h>
 56
 57#include "mm_internal.h"
 58
 59unsigned long highstart_pfn, highend_pfn;
 60
 61bool __read_mostly __vmalloc_start_set = false;
 62
 63/*
 64 * Creates a middle page table and puts a pointer to it in the
 65 * given global directory entry. This only returns the gd entry
 66 * in non-PAE compilation mode, since the middle layer is folded.
 67 */
 68static pmd_t * __init one_md_table_init(pgd_t *pgd)
 69{
 70	p4d_t *p4d;
 71	pud_t *pud;
 72	pmd_t *pmd_table;
 73
 74#ifdef CONFIG_X86_PAE
 75	if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
 76		pmd_table = (pmd_t *)alloc_low_page();
 77		paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
 78		set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 79		p4d = p4d_offset(pgd, 0);
 80		pud = pud_offset(p4d, 0);
 81		BUG_ON(pmd_table != pmd_offset(pud, 0));
 82
 83		return pmd_table;
 84	}
 85#endif
 86	p4d = p4d_offset(pgd, 0);
 87	pud = pud_offset(p4d, 0);
 88	pmd_table = pmd_offset(pud, 0);
 89
 90	return pmd_table;
 91}
 92
 93/*
 94 * Create a page table and place a pointer to it in a middle page
 95 * directory entry:
 96 */
 97static pte_t * __init one_page_table_init(pmd_t *pmd)
 98{
 99	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
100		pte_t *page_table = (pte_t *)alloc_low_page();
101
102		paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
103		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
104		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
105	}
106
107	return pte_offset_kernel(pmd, 0);
108}
109
110pmd_t * __init populate_extra_pmd(unsigned long vaddr)
111{
112	int pgd_idx = pgd_index(vaddr);
113	int pmd_idx = pmd_index(vaddr);
114
115	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
116}
117
118pte_t * __init populate_extra_pte(unsigned long vaddr)
119{
120	int pte_idx = pte_index(vaddr);
121	pmd_t *pmd;
122
123	pmd = populate_extra_pmd(vaddr);
124	return one_page_table_init(pmd) + pte_idx;
125}
126
127static unsigned long __init
128page_table_range_init_count(unsigned long start, unsigned long end)
129{
130	unsigned long count = 0;
131#ifdef CONFIG_HIGHMEM
132	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
133	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
134	int pgd_idx, pmd_idx;
135	unsigned long vaddr;
136
137	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
138		return 0;
139
140	vaddr = start;
141	pgd_idx = pgd_index(vaddr);
142	pmd_idx = pmd_index(vaddr);
143
144	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
145		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
146							pmd_idx++) {
147			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
148			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
149				count++;
150			vaddr += PMD_SIZE;
151		}
152		pmd_idx = 0;
153	}
154#endif
155	return count;
156}
157
158static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
159					   unsigned long vaddr, pte_t *lastpte,
160					   void **adr)
161{
162#ifdef CONFIG_HIGHMEM
163	/*
164	 * Something (early fixmap) may already have put a pte
165	 * page here, which causes the page table allocation
166	 * to become nonlinear. Attempt to fix it, and if it
167	 * is still nonlinear then we have to bug.
168	 */
169	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
170	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
171
172	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
173	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
174	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
175		pte_t *newpte;
176		int i;
177
178		BUG_ON(after_bootmem);
179		newpte = *adr;
180		for (i = 0; i < PTRS_PER_PTE; i++)
181			set_pte(newpte + i, pte[i]);
182		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
183
184		paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
185		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
186		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
187		__flush_tlb_all();
188
189		paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
190		pte = newpte;
191	}
192	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
193	       && vaddr > fix_to_virt(FIX_KMAP_END)
194	       && lastpte && lastpte + PTRS_PER_PTE != pte);
195#endif
196	return pte;
197}
198
199/*
200 * This function initializes a certain range of kernel virtual memory
201 * with new bootmem page tables, everywhere page tables are missing in
202 * the given range.
203 *
204 * NOTE: The pagetables are allocated contiguous on the physical space
205 * so we can cache the place of the first one and move around without
206 * checking the pgd every time.
207 */
208static void __init
209page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
210{
211	int pgd_idx, pmd_idx;
212	unsigned long vaddr;
213	pgd_t *pgd;
214	pmd_t *pmd;
215	pte_t *pte = NULL;
216	unsigned long count = page_table_range_init_count(start, end);
217	void *adr = NULL;
218
219	if (count)
220		adr = alloc_low_pages(count);
221
222	vaddr = start;
223	pgd_idx = pgd_index(vaddr);
224	pmd_idx = pmd_index(vaddr);
225	pgd = pgd_base + pgd_idx;
226
227	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
228		pmd = one_md_table_init(pgd);
229		pmd = pmd + pmd_index(vaddr);
230		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
231							pmd++, pmd_idx++) {
232			pte = page_table_kmap_check(one_page_table_init(pmd),
233						    pmd, vaddr, pte, &adr);
234
235			vaddr += PMD_SIZE;
236		}
237		pmd_idx = 0;
238	}
239}
240
241static inline int is_x86_32_kernel_text(unsigned long addr)
 
 
 
 
242{
243	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
244		return 1;
245	return 0;
246}
247
248/*
249 * This maps the physical memory to kernel virtual address space, a total
250 * of max_low_pfn pages, by creating page tables starting from address
251 * PAGE_OFFSET:
252 */
253unsigned long __init
254kernel_physical_mapping_init(unsigned long start,
255			     unsigned long end,
256			     unsigned long page_size_mask,
257			     pgprot_t prot)
258{
259	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
260	unsigned long last_map_addr = end;
261	unsigned long start_pfn, end_pfn;
262	pgd_t *pgd_base = swapper_pg_dir;
263	int pgd_idx, pmd_idx, pte_ofs;
264	unsigned long pfn;
265	pgd_t *pgd;
266	pmd_t *pmd;
267	pte_t *pte;
268	unsigned pages_2m, pages_4k;
269	int mapping_iter;
270
271	start_pfn = start >> PAGE_SHIFT;
272	end_pfn = end >> PAGE_SHIFT;
273
274	/*
275	 * First iteration will setup identity mapping using large/small pages
276	 * based on use_pse, with other attributes same as set by
277	 * the early code in head_32.S
278	 *
279	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
280	 * as desired for the kernel identity mapping.
281	 *
282	 * This two pass mechanism conforms to the TLB app note which says:
283	 *
284	 *     "Software should not write to a paging-structure entry in a way
285	 *      that would change, for any linear address, both the page size
286	 *      and either the page frame or attributes."
287	 */
288	mapping_iter = 1;
289
290	if (!boot_cpu_has(X86_FEATURE_PSE))
291		use_pse = 0;
292
293repeat:
294	pages_2m = pages_4k = 0;
295	pfn = start_pfn;
296	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
297	pgd = pgd_base + pgd_idx;
298	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
299		pmd = one_md_table_init(pgd);
300
301		if (pfn >= end_pfn)
302			continue;
303#ifdef CONFIG_X86_PAE
304		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
305		pmd += pmd_idx;
306#else
307		pmd_idx = 0;
308#endif
309		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
310		     pmd++, pmd_idx++) {
311			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
312
313			/*
314			 * Map with big pages if possible, otherwise
315			 * create normal page tables:
316			 */
317			if (use_pse) {
318				unsigned int addr2;
319				pgprot_t prot = PAGE_KERNEL_LARGE;
320				/*
321				 * first pass will use the same initial
322				 * identity mapping attribute + _PAGE_PSE.
323				 */
324				pgprot_t init_prot =
325					__pgprot(PTE_IDENT_ATTR |
326						 _PAGE_PSE);
327
328				pfn &= PMD_MASK >> PAGE_SHIFT;
329				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
330					PAGE_OFFSET + PAGE_SIZE-1;
331
332				if (is_x86_32_kernel_text(addr) ||
333				    is_x86_32_kernel_text(addr2))
334					prot = PAGE_KERNEL_LARGE_EXEC;
335
336				pages_2m++;
337				if (mapping_iter == 1)
338					set_pmd(pmd, pfn_pmd(pfn, init_prot));
339				else
340					set_pmd(pmd, pfn_pmd(pfn, prot));
341
342				pfn += PTRS_PER_PTE;
343				continue;
344			}
345			pte = one_page_table_init(pmd);
346
347			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
348			pte += pte_ofs;
349			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
350			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
351				pgprot_t prot = PAGE_KERNEL;
352				/*
353				 * first pass will use the same initial
354				 * identity mapping attribute.
355				 */
356				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
357
358				if (is_x86_32_kernel_text(addr))
359					prot = PAGE_KERNEL_EXEC;
360
361				pages_4k++;
362				if (mapping_iter == 1) {
363					set_pte(pte, pfn_pte(pfn, init_prot));
364					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
365				} else
366					set_pte(pte, pfn_pte(pfn, prot));
367			}
368		}
369	}
370	if (mapping_iter == 1) {
371		/*
372		 * update direct mapping page count only in the first
373		 * iteration.
374		 */
375		update_page_count(PG_LEVEL_2M, pages_2m);
376		update_page_count(PG_LEVEL_4K, pages_4k);
377
378		/*
379		 * local global flush tlb, which will flush the previous
380		 * mappings present in both small and large page TLB's.
381		 */
382		__flush_tlb_all();
383
384		/*
385		 * Second iteration will set the actual desired PTE attributes.
386		 */
387		mapping_iter = 2;
388		goto repeat;
389	}
390	return last_map_addr;
391}
392
 
 
 
 
 
 
 
 
 
 
 
 
 
393#ifdef CONFIG_HIGHMEM
394static void __init permanent_kmaps_init(pgd_t *pgd_base)
395{
396	unsigned long vaddr = PKMAP_BASE;
397
398	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
399
400	pkmap_page_table = virt_to_kpte(vaddr);
401}
402
403void __init add_highpages_with_active_regions(int nid,
404			 unsigned long start_pfn, unsigned long end_pfn)
405{
406	phys_addr_t start, end;
407	u64 i;
408
409	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
410		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
411					    start_pfn, end_pfn);
412		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
413					      start_pfn, end_pfn);
414		for ( ; pfn < e_pfn; pfn++)
415			if (pfn_valid(pfn))
416				free_highmem_page(pfn_to_page(pfn));
417	}
418}
419#else
420static inline void permanent_kmaps_init(pgd_t *pgd_base)
421{
422}
423#endif /* CONFIG_HIGHMEM */
424
425void __init sync_initial_page_table(void)
426{
427	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
428			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
429			KERNEL_PGD_PTRS);
430
431	/*
432	 * sync back low identity map too.  It is used for example
433	 * in the 32-bit EFI stub.
434	 */
435	clone_pgd_range(initial_page_table,
436			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
437			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
438}
439
440void __init native_pagetable_init(void)
441{
442	unsigned long pfn, va;
443	pgd_t *pgd, *base = swapper_pg_dir;
444	p4d_t *p4d;
445	pud_t *pud;
446	pmd_t *pmd;
447	pte_t *pte;
448
449	/*
450	 * Remove any mappings which extend past the end of physical
451	 * memory from the boot time page table.
452	 * In virtual address space, we should have at least two pages
453	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
454	 * definition. And max_low_pfn is set to VMALLOC_END physical
455	 * address. If initial memory mapping is doing right job, we
456	 * should have pte used near max_low_pfn or one pmd is not present.
457	 */
458	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
459		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
460		pgd = base + pgd_index(va);
461		if (!pgd_present(*pgd))
462			break;
463
464		p4d = p4d_offset(pgd, va);
465		pud = pud_offset(p4d, va);
466		pmd = pmd_offset(pud, va);
467		if (!pmd_present(*pmd))
468			break;
469
470		/* should not be large page here */
471		if (pmd_large(*pmd)) {
472			pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
473				pfn, pmd, __pa(pmd));
474			BUG_ON(1);
475		}
476
477		pte = pte_offset_kernel(pmd, va);
478		if (!pte_present(*pte))
479			break;
480
481		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
482				pfn, pmd, __pa(pmd), pte, __pa(pte));
483		pte_clear(NULL, va, pte);
484	}
485	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
486	paging_init();
487}
488
489/*
490 * Build a proper pagetable for the kernel mappings.  Up until this
491 * point, we've been running on some set of pagetables constructed by
492 * the boot process.
493 *
494 * If we're booting on native hardware, this will be a pagetable
495 * constructed in arch/x86/kernel/head_32.S.  The root of the
496 * pagetable will be swapper_pg_dir.
497 *
498 * If we're booting paravirtualized under a hypervisor, then there are
499 * more options: we may already be running PAE, and the pagetable may
500 * or may not be based in swapper_pg_dir.  In any case,
501 * paravirt_pagetable_init() will set up swapper_pg_dir
502 * appropriately for the rest of the initialization to work.
503 *
504 * In general, pagetable_init() assumes that the pagetable may already
505 * be partially populated, and so it avoids stomping on any existing
506 * mappings.
507 */
508void __init early_ioremap_page_table_range_init(void)
509{
510	pgd_t *pgd_base = swapper_pg_dir;
511	unsigned long vaddr, end;
512
513	/*
514	 * Fixed mappings, only the page table structure has to be
515	 * created - mappings will be set by set_fixmap():
516	 */
517	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
518	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
519	page_table_range_init(vaddr, end, pgd_base);
520	early_ioremap_reset();
521}
522
523static void __init pagetable_init(void)
524{
525	pgd_t *pgd_base = swapper_pg_dir;
526
527	permanent_kmaps_init(pgd_base);
528}
529
530#define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
531/* Bits supported by the hardware: */
532pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
533/* Bits allowed in normal kernel mappings: */
534pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
535EXPORT_SYMBOL_GPL(__supported_pte_mask);
536/* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
537EXPORT_SYMBOL(__default_kernel_pte_mask);
538
539/* user-defined highmem size */
540static unsigned int highmem_pages = -1;
541
542/*
543 * highmem=size forces highmem to be exactly 'size' bytes.
544 * This works even on boxes that have no highmem otherwise.
545 * This also works to reduce highmem size on bigger boxes.
546 */
547static int __init parse_highmem(char *arg)
548{
549	if (!arg)
550		return -EINVAL;
551
552	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
553	return 0;
554}
555early_param("highmem", parse_highmem);
556
557#define MSG_HIGHMEM_TOO_BIG \
558	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
559
560#define MSG_LOWMEM_TOO_SMALL \
561	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
562/*
563 * All of RAM fits into lowmem - but if user wants highmem
564 * artificially via the highmem=x boot parameter then create
565 * it:
566 */
567static void __init lowmem_pfn_init(void)
568{
569	/* max_low_pfn is 0, we already have early_res support */
570	max_low_pfn = max_pfn;
571
572	if (highmem_pages == -1)
573		highmem_pages = 0;
574#ifdef CONFIG_HIGHMEM
575	if (highmem_pages >= max_pfn) {
576		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
577			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
578		highmem_pages = 0;
579	}
580	if (highmem_pages) {
581		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
582			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
583				pages_to_mb(highmem_pages));
584			highmem_pages = 0;
585		}
586		max_low_pfn -= highmem_pages;
587	}
588#else
589	if (highmem_pages)
590		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
591#endif
592}
593
594#define MSG_HIGHMEM_TOO_SMALL \
595	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
596
597#define MSG_HIGHMEM_TRIMMED \
598	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
599/*
600 * We have more RAM than fits into lowmem - we try to put it into
601 * highmem, also taking the highmem=x boot parameter into account:
602 */
603static void __init highmem_pfn_init(void)
604{
605	max_low_pfn = MAXMEM_PFN;
606
607	if (highmem_pages == -1)
608		highmem_pages = max_pfn - MAXMEM_PFN;
609
610	if (highmem_pages + MAXMEM_PFN < max_pfn)
611		max_pfn = MAXMEM_PFN + highmem_pages;
612
613	if (highmem_pages + MAXMEM_PFN > max_pfn) {
614		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
615			pages_to_mb(max_pfn - MAXMEM_PFN),
616			pages_to_mb(highmem_pages));
617		highmem_pages = 0;
618	}
619#ifndef CONFIG_HIGHMEM
620	/* Maximum memory usable is what is directly addressable */
621	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
622	if (max_pfn > MAX_NONPAE_PFN)
623		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
624	else
625		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
626	max_pfn = MAXMEM_PFN;
627#else /* !CONFIG_HIGHMEM */
628#ifndef CONFIG_HIGHMEM64G
629	if (max_pfn > MAX_NONPAE_PFN) {
630		max_pfn = MAX_NONPAE_PFN;
631		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
632	}
633#endif /* !CONFIG_HIGHMEM64G */
634#endif /* !CONFIG_HIGHMEM */
635}
636
637/*
638 * Determine low and high memory ranges:
639 */
640void __init find_low_pfn_range(void)
641{
642	/* it could update max_pfn */
643
644	if (max_pfn <= MAXMEM_PFN)
645		lowmem_pfn_init();
646	else
647		highmem_pfn_init();
648}
649
650#ifndef CONFIG_NUMA
651void __init initmem_init(void)
652{
653#ifdef CONFIG_HIGHMEM
654	highstart_pfn = highend_pfn = max_pfn;
655	if (max_pfn > max_low_pfn)
656		highstart_pfn = max_low_pfn;
657	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
658		pages_to_mb(highend_pfn - highstart_pfn));
659	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
660#else
661	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
662#endif
663
664	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
665
666#ifdef CONFIG_FLATMEM
667	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
668#endif
669	__vmalloc_start_set = true;
670
671	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
672			pages_to_mb(max_low_pfn));
673
674	setup_bootmem_allocator();
675}
676#endif /* !CONFIG_NUMA */
677
678void __init setup_bootmem_allocator(void)
679{
680	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
681		 max_pfn_mapped<<PAGE_SHIFT);
682	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
683}
684
685/*
686 * paging_init() sets up the page tables - note that the first 8MB are
687 * already mapped by head.S.
688 *
689 * This routines also unmaps the page at virtual kernel address 0, so
690 * that we can trap those pesky NULL-reference errors in the kernel.
691 */
692void __init paging_init(void)
693{
694	pagetable_init();
695
696	__flush_tlb_all();
697
 
 
698	/*
699	 * NOTE: at this point the bootmem allocator is fully available.
700	 */
701	olpc_dt_build_devicetree();
702	sparse_init();
703	zone_sizes_init();
704}
705
706/*
707 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
708 * and also on some strange 486's. All 586+'s are OK. This used to involve
709 * black magic jumps to work around some nasty CPU bugs, but fortunately the
710 * switch to using exceptions got rid of all that.
711 */
712static void __init test_wp_bit(void)
713{
714	char z = 0;
715
716	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
717
718	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
719
720	if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
721		clear_fixmap(FIX_WP_TEST);
722		printk(KERN_CONT "Ok.\n");
723		return;
724	}
725
726	printk(KERN_CONT "No.\n");
727	panic("Linux doesn't support CPUs with broken WP.");
728}
729
730void __init mem_init(void)
731{
732	pci_iommu_alloc();
733
734#ifdef CONFIG_FLATMEM
735	BUG_ON(!mem_map);
736#endif
737	/*
738	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
739	 * be done before memblock_free_all(). Memblock use free low memory for
740	 * temporary data (see find_range_array()) and for this purpose can use
741	 * pages that was already passed to the buddy allocator, hence marked as
742	 * not accessible in the page tables when compiled with
743	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
744	 * important here.
745	 */
746	set_highmem_pages_init();
747
748	/* this will put all low memory onto the freelists */
749	memblock_free_all();
750
751	after_bootmem = 1;
752	x86_init.hyper.init_after_bootmem();
753
 
 
754	/*
755	 * Check boundaries twice: Some fundamental inconsistencies can
756	 * be detected at build time already.
757	 */
758#define __FIXADDR_TOP (-PAGE_SIZE)
759#ifdef CONFIG_HIGHMEM
760	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
761	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
762#endif
763#define high_memory (-128UL << 20)
764	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
765#undef high_memory
766#undef __FIXADDR_TOP
767
768#ifdef CONFIG_HIGHMEM
769	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
770	BUG_ON(VMALLOC_END				> PKMAP_BASE);
771#endif
772	BUG_ON(VMALLOC_START				>= VMALLOC_END);
773	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
774
775	test_wp_bit();
776}
777
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
778int kernel_set_to_readonly __read_mostly;
779
780static void mark_nxdata_nx(void)
781{
782	/*
783	 * When this called, init has already been executed and released,
784	 * so everything past _etext should be NX.
785	 */
786	unsigned long start = PFN_ALIGN(_etext);
787	/*
788	 * This comes from is_x86_32_kernel_text upper limit. Also HPAGE where used:
789	 */
790	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
791
792	if (__supported_pte_mask & _PAGE_NX)
793		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
794	set_memory_nx(start, size >> PAGE_SHIFT);
795}
796
797void mark_rodata_ro(void)
798{
799	unsigned long start = PFN_ALIGN(_text);
800	unsigned long size = (unsigned long)__end_rodata - start;
801
802	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
803	pr_info("Write protecting kernel text and read-only data: %luk\n",
804		size >> 10);
805
806	kernel_set_to_readonly = 1;
807
808#ifdef CONFIG_CPA_DEBUG
809	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
810	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
811
812	pr_info("Testing CPA: write protecting again\n");
813	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
814#endif
815	mark_nxdata_nx();
816	if (__supported_pte_mask & _PAGE_NX)
817		debug_checkwx();
818}
v5.9
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 *
  4 *  Copyright (C) 1995  Linus Torvalds
  5 *
  6 *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  7 */
  8
  9#include <linux/signal.h>
 10#include <linux/sched.h>
 11#include <linux/kernel.h>
 12#include <linux/errno.h>
 13#include <linux/string.h>
 14#include <linux/types.h>
 15#include <linux/ptrace.h>
 16#include <linux/mman.h>
 17#include <linux/mm.h>
 18#include <linux/hugetlb.h>
 19#include <linux/swap.h>
 20#include <linux/smp.h>
 21#include <linux/init.h>
 22#include <linux/highmem.h>
 23#include <linux/pagemap.h>
 24#include <linux/pci.h>
 25#include <linux/pfn.h>
 26#include <linux/poison.h>
 27#include <linux/memblock.h>
 28#include <linux/proc_fs.h>
 29#include <linux/memory_hotplug.h>
 30#include <linux/initrd.h>
 31#include <linux/cpumask.h>
 32#include <linux/gfp.h>
 33
 34#include <asm/asm.h>
 35#include <asm/bios_ebda.h>
 36#include <asm/processor.h>
 37#include <linux/uaccess.h>
 38#include <asm/dma.h>
 39#include <asm/fixmap.h>
 40#include <asm/e820/api.h>
 41#include <asm/apic.h>
 42#include <asm/bugs.h>
 43#include <asm/tlb.h>
 44#include <asm/tlbflush.h>
 45#include <asm/olpc_ofw.h>
 46#include <asm/pgalloc.h>
 47#include <asm/sections.h>
 48#include <asm/paravirt.h>
 49#include <asm/setup.h>
 50#include <asm/set_memory.h>
 51#include <asm/page_types.h>
 52#include <asm/cpu_entry_area.h>
 53#include <asm/init.h>
 54#include <asm/pgtable_areas.h>
 55#include <asm/numa.h>
 56
 57#include "mm_internal.h"
 58
 59unsigned long highstart_pfn, highend_pfn;
 60
 61bool __read_mostly __vmalloc_start_set = false;
 62
 63/*
 64 * Creates a middle page table and puts a pointer to it in the
 65 * given global directory entry. This only returns the gd entry
 66 * in non-PAE compilation mode, since the middle layer is folded.
 67 */
 68static pmd_t * __init one_md_table_init(pgd_t *pgd)
 69{
 70	p4d_t *p4d;
 71	pud_t *pud;
 72	pmd_t *pmd_table;
 73
 74#ifdef CONFIG_X86_PAE
 75	if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
 76		pmd_table = (pmd_t *)alloc_low_page();
 77		paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
 78		set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 79		p4d = p4d_offset(pgd, 0);
 80		pud = pud_offset(p4d, 0);
 81		BUG_ON(pmd_table != pmd_offset(pud, 0));
 82
 83		return pmd_table;
 84	}
 85#endif
 86	p4d = p4d_offset(pgd, 0);
 87	pud = pud_offset(p4d, 0);
 88	pmd_table = pmd_offset(pud, 0);
 89
 90	return pmd_table;
 91}
 92
 93/*
 94 * Create a page table and place a pointer to it in a middle page
 95 * directory entry:
 96 */
 97static pte_t * __init one_page_table_init(pmd_t *pmd)
 98{
 99	if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
100		pte_t *page_table = (pte_t *)alloc_low_page();
101
102		paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
103		set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
104		BUG_ON(page_table != pte_offset_kernel(pmd, 0));
105	}
106
107	return pte_offset_kernel(pmd, 0);
108}
109
110pmd_t * __init populate_extra_pmd(unsigned long vaddr)
111{
112	int pgd_idx = pgd_index(vaddr);
113	int pmd_idx = pmd_index(vaddr);
114
115	return one_md_table_init(swapper_pg_dir + pgd_idx) + pmd_idx;
116}
117
118pte_t * __init populate_extra_pte(unsigned long vaddr)
119{
120	int pte_idx = pte_index(vaddr);
121	pmd_t *pmd;
122
123	pmd = populate_extra_pmd(vaddr);
124	return one_page_table_init(pmd) + pte_idx;
125}
126
127static unsigned long __init
128page_table_range_init_count(unsigned long start, unsigned long end)
129{
130	unsigned long count = 0;
131#ifdef CONFIG_HIGHMEM
132	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
133	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
134	int pgd_idx, pmd_idx;
135	unsigned long vaddr;
136
137	if (pmd_idx_kmap_begin == pmd_idx_kmap_end)
138		return 0;
139
140	vaddr = start;
141	pgd_idx = pgd_index(vaddr);
142	pmd_idx = pmd_index(vaddr);
143
144	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd_idx++) {
145		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
146							pmd_idx++) {
147			if ((vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin &&
148			    (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end)
149				count++;
150			vaddr += PMD_SIZE;
151		}
152		pmd_idx = 0;
153	}
154#endif
155	return count;
156}
157
158static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
159					   unsigned long vaddr, pte_t *lastpte,
160					   void **adr)
161{
162#ifdef CONFIG_HIGHMEM
163	/*
164	 * Something (early fixmap) may already have put a pte
165	 * page here, which causes the page table allocation
166	 * to become nonlinear. Attempt to fix it, and if it
167	 * is still nonlinear then we have to bug.
168	 */
169	int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
170	int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
171
172	if (pmd_idx_kmap_begin != pmd_idx_kmap_end
173	    && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
174	    && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end) {
175		pte_t *newpte;
176		int i;
177
178		BUG_ON(after_bootmem);
179		newpte = *adr;
180		for (i = 0; i < PTRS_PER_PTE; i++)
181			set_pte(newpte + i, pte[i]);
182		*adr = (void *)(((unsigned long)(*adr)) + PAGE_SIZE);
183
184		paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
185		set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
186		BUG_ON(newpte != pte_offset_kernel(pmd, 0));
187		__flush_tlb_all();
188
189		paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
190		pte = newpte;
191	}
192	BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
193	       && vaddr > fix_to_virt(FIX_KMAP_END)
194	       && lastpte && lastpte + PTRS_PER_PTE != pte);
195#endif
196	return pte;
197}
198
199/*
200 * This function initializes a certain range of kernel virtual memory
201 * with new bootmem page tables, everywhere page tables are missing in
202 * the given range.
203 *
204 * NOTE: The pagetables are allocated contiguous on the physical space
205 * so we can cache the place of the first one and move around without
206 * checking the pgd every time.
207 */
208static void __init
209page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
210{
211	int pgd_idx, pmd_idx;
212	unsigned long vaddr;
213	pgd_t *pgd;
214	pmd_t *pmd;
215	pte_t *pte = NULL;
216	unsigned long count = page_table_range_init_count(start, end);
217	void *adr = NULL;
218
219	if (count)
220		adr = alloc_low_pages(count);
221
222	vaddr = start;
223	pgd_idx = pgd_index(vaddr);
224	pmd_idx = pmd_index(vaddr);
225	pgd = pgd_base + pgd_idx;
226
227	for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
228		pmd = one_md_table_init(pgd);
229		pmd = pmd + pmd_index(vaddr);
230		for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
231							pmd++, pmd_idx++) {
232			pte = page_table_kmap_check(one_page_table_init(pmd),
233						    pmd, vaddr, pte, &adr);
234
235			vaddr += PMD_SIZE;
236		}
237		pmd_idx = 0;
238	}
239}
240
241/*
242 * The <linux/kallsyms.h> already defines is_kernel_text,
243 * using '__' prefix not to get in conflict.
244 */
245static inline int __is_kernel_text(unsigned long addr)
246{
247	if (addr >= (unsigned long)_text && addr <= (unsigned long)__init_end)
248		return 1;
249	return 0;
250}
251
252/*
253 * This maps the physical memory to kernel virtual address space, a total
254 * of max_low_pfn pages, by creating page tables starting from address
255 * PAGE_OFFSET:
256 */
257unsigned long __init
258kernel_physical_mapping_init(unsigned long start,
259			     unsigned long end,
260			     unsigned long page_size_mask,
261			     pgprot_t prot)
262{
263	int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
264	unsigned long last_map_addr = end;
265	unsigned long start_pfn, end_pfn;
266	pgd_t *pgd_base = swapper_pg_dir;
267	int pgd_idx, pmd_idx, pte_ofs;
268	unsigned long pfn;
269	pgd_t *pgd;
270	pmd_t *pmd;
271	pte_t *pte;
272	unsigned pages_2m, pages_4k;
273	int mapping_iter;
274
275	start_pfn = start >> PAGE_SHIFT;
276	end_pfn = end >> PAGE_SHIFT;
277
278	/*
279	 * First iteration will setup identity mapping using large/small pages
280	 * based on use_pse, with other attributes same as set by
281	 * the early code in head_32.S
282	 *
283	 * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
284	 * as desired for the kernel identity mapping.
285	 *
286	 * This two pass mechanism conforms to the TLB app note which says:
287	 *
288	 *     "Software should not write to a paging-structure entry in a way
289	 *      that would change, for any linear address, both the page size
290	 *      and either the page frame or attributes."
291	 */
292	mapping_iter = 1;
293
294	if (!boot_cpu_has(X86_FEATURE_PSE))
295		use_pse = 0;
296
297repeat:
298	pages_2m = pages_4k = 0;
299	pfn = start_pfn;
300	pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
301	pgd = pgd_base + pgd_idx;
302	for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
303		pmd = one_md_table_init(pgd);
304
305		if (pfn >= end_pfn)
306			continue;
307#ifdef CONFIG_X86_PAE
308		pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
309		pmd += pmd_idx;
310#else
311		pmd_idx = 0;
312#endif
313		for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
314		     pmd++, pmd_idx++) {
315			unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
316
317			/*
318			 * Map with big pages if possible, otherwise
319			 * create normal page tables:
320			 */
321			if (use_pse) {
322				unsigned int addr2;
323				pgprot_t prot = PAGE_KERNEL_LARGE;
324				/*
325				 * first pass will use the same initial
326				 * identity mapping attribute + _PAGE_PSE.
327				 */
328				pgprot_t init_prot =
329					__pgprot(PTE_IDENT_ATTR |
330						 _PAGE_PSE);
331
332				pfn &= PMD_MASK >> PAGE_SHIFT;
333				addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
334					PAGE_OFFSET + PAGE_SIZE-1;
335
336				if (__is_kernel_text(addr) ||
337				    __is_kernel_text(addr2))
338					prot = PAGE_KERNEL_LARGE_EXEC;
339
340				pages_2m++;
341				if (mapping_iter == 1)
342					set_pmd(pmd, pfn_pmd(pfn, init_prot));
343				else
344					set_pmd(pmd, pfn_pmd(pfn, prot));
345
346				pfn += PTRS_PER_PTE;
347				continue;
348			}
349			pte = one_page_table_init(pmd);
350
351			pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
352			pte += pte_ofs;
353			for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
354			     pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
355				pgprot_t prot = PAGE_KERNEL;
356				/*
357				 * first pass will use the same initial
358				 * identity mapping attribute.
359				 */
360				pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
361
362				if (__is_kernel_text(addr))
363					prot = PAGE_KERNEL_EXEC;
364
365				pages_4k++;
366				if (mapping_iter == 1) {
367					set_pte(pte, pfn_pte(pfn, init_prot));
368					last_map_addr = (pfn << PAGE_SHIFT) + PAGE_SIZE;
369				} else
370					set_pte(pte, pfn_pte(pfn, prot));
371			}
372		}
373	}
374	if (mapping_iter == 1) {
375		/*
376		 * update direct mapping page count only in the first
377		 * iteration.
378		 */
379		update_page_count(PG_LEVEL_2M, pages_2m);
380		update_page_count(PG_LEVEL_4K, pages_4k);
381
382		/*
383		 * local global flush tlb, which will flush the previous
384		 * mappings present in both small and large page TLB's.
385		 */
386		__flush_tlb_all();
387
388		/*
389		 * Second iteration will set the actual desired PTE attributes.
390		 */
391		mapping_iter = 2;
392		goto repeat;
393	}
394	return last_map_addr;
395}
396
397pte_t *kmap_pte;
398
399static void __init kmap_init(void)
400{
401	unsigned long kmap_vstart;
402
403	/*
404	 * Cache the first kmap pte:
405	 */
406	kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
407	kmap_pte = virt_to_kpte(kmap_vstart);
408}
409
410#ifdef CONFIG_HIGHMEM
411static void __init permanent_kmaps_init(pgd_t *pgd_base)
412{
413	unsigned long vaddr = PKMAP_BASE;
414
415	page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
416
417	pkmap_page_table = virt_to_kpte(vaddr);
418}
419
420void __init add_highpages_with_active_regions(int nid,
421			 unsigned long start_pfn, unsigned long end_pfn)
422{
423	phys_addr_t start, end;
424	u64 i;
425
426	for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &start, &end, NULL) {
427		unsigned long pfn = clamp_t(unsigned long, PFN_UP(start),
428					    start_pfn, end_pfn);
429		unsigned long e_pfn = clamp_t(unsigned long, PFN_DOWN(end),
430					      start_pfn, end_pfn);
431		for ( ; pfn < e_pfn; pfn++)
432			if (pfn_valid(pfn))
433				free_highmem_page(pfn_to_page(pfn));
434	}
435}
436#else
437static inline void permanent_kmaps_init(pgd_t *pgd_base)
438{
439}
440#endif /* CONFIG_HIGHMEM */
441
442void __init sync_initial_page_table(void)
443{
444	clone_pgd_range(initial_page_table + KERNEL_PGD_BOUNDARY,
445			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
446			KERNEL_PGD_PTRS);
447
448	/*
449	 * sync back low identity map too.  It is used for example
450	 * in the 32-bit EFI stub.
451	 */
452	clone_pgd_range(initial_page_table,
453			swapper_pg_dir     + KERNEL_PGD_BOUNDARY,
454			min(KERNEL_PGD_PTRS, KERNEL_PGD_BOUNDARY));
455}
456
457void __init native_pagetable_init(void)
458{
459	unsigned long pfn, va;
460	pgd_t *pgd, *base = swapper_pg_dir;
461	p4d_t *p4d;
462	pud_t *pud;
463	pmd_t *pmd;
464	pte_t *pte;
465
466	/*
467	 * Remove any mappings which extend past the end of physical
468	 * memory from the boot time page table.
469	 * In virtual address space, we should have at least two pages
470	 * from VMALLOC_END to pkmap or fixmap according to VMALLOC_END
471	 * definition. And max_low_pfn is set to VMALLOC_END physical
472	 * address. If initial memory mapping is doing right job, we
473	 * should have pte used near max_low_pfn or one pmd is not present.
474	 */
475	for (pfn = max_low_pfn; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
476		va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
477		pgd = base + pgd_index(va);
478		if (!pgd_present(*pgd))
479			break;
480
481		p4d = p4d_offset(pgd, va);
482		pud = pud_offset(p4d, va);
483		pmd = pmd_offset(pud, va);
484		if (!pmd_present(*pmd))
485			break;
486
487		/* should not be large page here */
488		if (pmd_large(*pmd)) {
489			pr_warn("try to clear pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx, but pmd is big page and is not using pte !\n",
490				pfn, pmd, __pa(pmd));
491			BUG_ON(1);
492		}
493
494		pte = pte_offset_kernel(pmd, va);
495		if (!pte_present(*pte))
496			break;
497
498		printk(KERN_DEBUG "clearing pte for ram above max_low_pfn: pfn: %lx pmd: %p pmd phys: %lx pte: %p pte phys: %lx\n",
499				pfn, pmd, __pa(pmd), pte, __pa(pte));
500		pte_clear(NULL, va, pte);
501	}
502	paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
503	paging_init();
504}
505
506/*
507 * Build a proper pagetable for the kernel mappings.  Up until this
508 * point, we've been running on some set of pagetables constructed by
509 * the boot process.
510 *
511 * If we're booting on native hardware, this will be a pagetable
512 * constructed in arch/x86/kernel/head_32.S.  The root of the
513 * pagetable will be swapper_pg_dir.
514 *
515 * If we're booting paravirtualized under a hypervisor, then there are
516 * more options: we may already be running PAE, and the pagetable may
517 * or may not be based in swapper_pg_dir.  In any case,
518 * paravirt_pagetable_init() will set up swapper_pg_dir
519 * appropriately for the rest of the initialization to work.
520 *
521 * In general, pagetable_init() assumes that the pagetable may already
522 * be partially populated, and so it avoids stomping on any existing
523 * mappings.
524 */
525void __init early_ioremap_page_table_range_init(void)
526{
527	pgd_t *pgd_base = swapper_pg_dir;
528	unsigned long vaddr, end;
529
530	/*
531	 * Fixed mappings, only the page table structure has to be
532	 * created - mappings will be set by set_fixmap():
533	 */
534	vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
535	end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
536	page_table_range_init(vaddr, end, pgd_base);
537	early_ioremap_reset();
538}
539
540static void __init pagetable_init(void)
541{
542	pgd_t *pgd_base = swapper_pg_dir;
543
544	permanent_kmaps_init(pgd_base);
545}
546
547#define DEFAULT_PTE_MASK ~(_PAGE_NX | _PAGE_GLOBAL)
548/* Bits supported by the hardware: */
549pteval_t __supported_pte_mask __read_mostly = DEFAULT_PTE_MASK;
550/* Bits allowed in normal kernel mappings: */
551pteval_t __default_kernel_pte_mask __read_mostly = DEFAULT_PTE_MASK;
552EXPORT_SYMBOL_GPL(__supported_pte_mask);
553/* Used in PAGE_KERNEL_* macros which are reasonably used out-of-tree: */
554EXPORT_SYMBOL(__default_kernel_pte_mask);
555
556/* user-defined highmem size */
557static unsigned int highmem_pages = -1;
558
559/*
560 * highmem=size forces highmem to be exactly 'size' bytes.
561 * This works even on boxes that have no highmem otherwise.
562 * This also works to reduce highmem size on bigger boxes.
563 */
564static int __init parse_highmem(char *arg)
565{
566	if (!arg)
567		return -EINVAL;
568
569	highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
570	return 0;
571}
572early_param("highmem", parse_highmem);
573
574#define MSG_HIGHMEM_TOO_BIG \
575	"highmem size (%luMB) is bigger than pages available (%luMB)!\n"
576
577#define MSG_LOWMEM_TOO_SMALL \
578	"highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
579/*
580 * All of RAM fits into lowmem - but if user wants highmem
581 * artificially via the highmem=x boot parameter then create
582 * it:
583 */
584static void __init lowmem_pfn_init(void)
585{
586	/* max_low_pfn is 0, we already have early_res support */
587	max_low_pfn = max_pfn;
588
589	if (highmem_pages == -1)
590		highmem_pages = 0;
591#ifdef CONFIG_HIGHMEM
592	if (highmem_pages >= max_pfn) {
593		printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
594			pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
595		highmem_pages = 0;
596	}
597	if (highmem_pages) {
598		if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
599			printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
600				pages_to_mb(highmem_pages));
601			highmem_pages = 0;
602		}
603		max_low_pfn -= highmem_pages;
604	}
605#else
606	if (highmem_pages)
607		printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
608#endif
609}
610
611#define MSG_HIGHMEM_TOO_SMALL \
612	"only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
613
614#define MSG_HIGHMEM_TRIMMED \
615	"Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
616/*
617 * We have more RAM than fits into lowmem - we try to put it into
618 * highmem, also taking the highmem=x boot parameter into account:
619 */
620static void __init highmem_pfn_init(void)
621{
622	max_low_pfn = MAXMEM_PFN;
623
624	if (highmem_pages == -1)
625		highmem_pages = max_pfn - MAXMEM_PFN;
626
627	if (highmem_pages + MAXMEM_PFN < max_pfn)
628		max_pfn = MAXMEM_PFN + highmem_pages;
629
630	if (highmem_pages + MAXMEM_PFN > max_pfn) {
631		printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
632			pages_to_mb(max_pfn - MAXMEM_PFN),
633			pages_to_mb(highmem_pages));
634		highmem_pages = 0;
635	}
636#ifndef CONFIG_HIGHMEM
637	/* Maximum memory usable is what is directly addressable */
638	printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
639	if (max_pfn > MAX_NONPAE_PFN)
640		printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
641	else
642		printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
643	max_pfn = MAXMEM_PFN;
644#else /* !CONFIG_HIGHMEM */
645#ifndef CONFIG_HIGHMEM64G
646	if (max_pfn > MAX_NONPAE_PFN) {
647		max_pfn = MAX_NONPAE_PFN;
648		printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
649	}
650#endif /* !CONFIG_HIGHMEM64G */
651#endif /* !CONFIG_HIGHMEM */
652}
653
654/*
655 * Determine low and high memory ranges:
656 */
657void __init find_low_pfn_range(void)
658{
659	/* it could update max_pfn */
660
661	if (max_pfn <= MAXMEM_PFN)
662		lowmem_pfn_init();
663	else
664		highmem_pfn_init();
665}
666
667#ifndef CONFIG_NEED_MULTIPLE_NODES
668void __init initmem_init(void)
669{
670#ifdef CONFIG_HIGHMEM
671	highstart_pfn = highend_pfn = max_pfn;
672	if (max_pfn > max_low_pfn)
673		highstart_pfn = max_low_pfn;
674	printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
675		pages_to_mb(highend_pfn - highstart_pfn));
676	high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
677#else
678	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
679#endif
680
681	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
682
683#ifdef CONFIG_FLATMEM
684	max_mapnr = IS_ENABLED(CONFIG_HIGHMEM) ? highend_pfn : max_low_pfn;
685#endif
686	__vmalloc_start_set = true;
687
688	printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
689			pages_to_mb(max_low_pfn));
690
691	setup_bootmem_allocator();
692}
693#endif /* !CONFIG_NEED_MULTIPLE_NODES */
694
695void __init setup_bootmem_allocator(void)
696{
697	printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
698		 max_pfn_mapped<<PAGE_SHIFT);
699	printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
700}
701
702/*
703 * paging_init() sets up the page tables - note that the first 8MB are
704 * already mapped by head.S.
705 *
706 * This routines also unmaps the page at virtual kernel address 0, so
707 * that we can trap those pesky NULL-reference errors in the kernel.
708 */
709void __init paging_init(void)
710{
711	pagetable_init();
712
713	__flush_tlb_all();
714
715	kmap_init();
716
717	/*
718	 * NOTE: at this point the bootmem allocator is fully available.
719	 */
720	olpc_dt_build_devicetree();
721	sparse_init();
722	zone_sizes_init();
723}
724
725/*
726 * Test if the WP bit works in supervisor mode. It isn't supported on 386's
727 * and also on some strange 486's. All 586+'s are OK. This used to involve
728 * black magic jumps to work around some nasty CPU bugs, but fortunately the
729 * switch to using exceptions got rid of all that.
730 */
731static void __init test_wp_bit(void)
732{
733	char z = 0;
734
735	printk(KERN_INFO "Checking if this processor honours the WP bit even in supervisor mode...");
736
737	__set_fixmap(FIX_WP_TEST, __pa_symbol(empty_zero_page), PAGE_KERNEL_RO);
738
739	if (copy_to_kernel_nofault((char *)fix_to_virt(FIX_WP_TEST), &z, 1)) {
740		clear_fixmap(FIX_WP_TEST);
741		printk(KERN_CONT "Ok.\n");
742		return;
743	}
744
745	printk(KERN_CONT "No.\n");
746	panic("Linux doesn't support CPUs with broken WP.");
747}
748
749void __init mem_init(void)
750{
751	pci_iommu_alloc();
752
753#ifdef CONFIG_FLATMEM
754	BUG_ON(!mem_map);
755#endif
756	/*
757	 * With CONFIG_DEBUG_PAGEALLOC initialization of highmem pages has to
758	 * be done before memblock_free_all(). Memblock use free low memory for
759	 * temporary data (see find_range_array()) and for this purpose can use
760	 * pages that was already passed to the buddy allocator, hence marked as
761	 * not accessible in the page tables when compiled with
762	 * CONFIG_DEBUG_PAGEALLOC. Otherwise order of initialization is not
763	 * important here.
764	 */
765	set_highmem_pages_init();
766
767	/* this will put all low memory onto the freelists */
768	memblock_free_all();
769
770	after_bootmem = 1;
771	x86_init.hyper.init_after_bootmem();
772
773	mem_init_print_info(NULL);
774
775	/*
776	 * Check boundaries twice: Some fundamental inconsistencies can
777	 * be detected at build time already.
778	 */
779#define __FIXADDR_TOP (-PAGE_SIZE)
780#ifdef CONFIG_HIGHMEM
781	BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
782	BUILD_BUG_ON(VMALLOC_END			> PKMAP_BASE);
783#endif
784#define high_memory (-128UL << 20)
785	BUILD_BUG_ON(VMALLOC_START			>= VMALLOC_END);
786#undef high_memory
787#undef __FIXADDR_TOP
788
789#ifdef CONFIG_HIGHMEM
790	BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE	> FIXADDR_START);
791	BUG_ON(VMALLOC_END				> PKMAP_BASE);
792#endif
793	BUG_ON(VMALLOC_START				>= VMALLOC_END);
794	BUG_ON((unsigned long)high_memory		> VMALLOC_START);
795
796	test_wp_bit();
797}
798
799#ifdef CONFIG_MEMORY_HOTPLUG
800int arch_add_memory(int nid, u64 start, u64 size,
801		    struct mhp_params *params)
802{
803	unsigned long start_pfn = start >> PAGE_SHIFT;
804	unsigned long nr_pages = size >> PAGE_SHIFT;
805	int ret;
806
807	/*
808	 * The page tables were already mapped at boot so if the caller
809	 * requests a different mapping type then we must change all the
810	 * pages with __set_memory_prot().
811	 */
812	if (params->pgprot.pgprot != PAGE_KERNEL.pgprot) {
813		ret = __set_memory_prot(start, nr_pages, params->pgprot);
814		if (ret)
815			return ret;
816	}
817
818	return __add_pages(nid, start_pfn, nr_pages, params);
819}
820
821void arch_remove_memory(int nid, u64 start, u64 size,
822			struct vmem_altmap *altmap)
823{
824	unsigned long start_pfn = start >> PAGE_SHIFT;
825	unsigned long nr_pages = size >> PAGE_SHIFT;
826
827	__remove_pages(start_pfn, nr_pages, altmap);
828}
829#endif
830
831int kernel_set_to_readonly __read_mostly;
832
833static void mark_nxdata_nx(void)
834{
835	/*
836	 * When this called, init has already been executed and released,
837	 * so everything past _etext should be NX.
838	 */
839	unsigned long start = PFN_ALIGN(_etext);
840	/*
841	 * This comes from __is_kernel_text upper limit. Also HPAGE where used:
842	 */
843	unsigned long size = (((unsigned long)__init_end + HPAGE_SIZE) & HPAGE_MASK) - start;
844
845	if (__supported_pte_mask & _PAGE_NX)
846		printk(KERN_INFO "NX-protecting the kernel data: %luk\n", size >> 10);
847	set_memory_nx(start, size >> PAGE_SHIFT);
848}
849
850void mark_rodata_ro(void)
851{
852	unsigned long start = PFN_ALIGN(_text);
853	unsigned long size = (unsigned long)__end_rodata - start;
854
855	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
856	pr_info("Write protecting kernel text and read-only data: %luk\n",
857		size >> 10);
858
859	kernel_set_to_readonly = 1;
860
861#ifdef CONFIG_CPA_DEBUG
862	pr_info("Testing CPA: Reverting %lx-%lx\n", start, start + size);
863	set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
864
865	pr_info("Testing CPA: write protecting again\n");
866	set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
867#endif
868	mark_nxdata_nx();
869	if (__supported_pte_mask & _PAGE_NX)
870		debug_checkwx();
871}