• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Initialize MMU support.
3  *
4  * Copyright (C) 1998-2003 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  */
7 #include <linux/kernel.h>
8 #include <linux/init.h>
9 
10 #include <linux/bootmem.h>
11 #include <linux/efi.h>
12 #include <linux/elf.h>
13 #include <linux/memblock.h>
14 #include <linux/mm.h>
15 #include <linux/mmzone.h>
16 #include <linux/module.h>
17 #include <linux/personality.h>
18 #include <linux/reboot.h>
19 #include <linux/slab.h>
20 #include <linux/swap.h>
21 #include <linux/proc_fs.h>
22 #include <linux/bitops.h>
23 #include <linux/kexec.h>
24 
25 #include <asm/dma.h>
26 #include <asm/io.h>
27 #include <asm/machvec.h>
28 #include <asm/numa.h>
29 #include <asm/patch.h>
30 #include <asm/pgalloc.h>
31 #include <asm/sal.h>
32 #include <asm/sections.h>
33 #include <asm/tlb.h>
34 #include <asm/uaccess.h>
35 #include <asm/unistd.h>
36 #include <asm/mca.h>
37 #include <asm/paravirt.h>
38 
39 extern void ia64_tlb_init (void);
40 
41 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
42 
43 #ifdef CONFIG_VIRTUAL_MEM_MAP
44 unsigned long VMALLOC_END = VMALLOC_END_INIT;
45 EXPORT_SYMBOL(VMALLOC_END);
46 struct page *vmem_map;
47 EXPORT_SYMBOL(vmem_map);
48 #endif
49 
50 struct page *zero_page_memmap_ptr;	/* map entry for zero page */
51 EXPORT_SYMBOL(zero_page_memmap_ptr);
52 
53 void
__ia64_sync_icache_dcache(pte_t pte)54 __ia64_sync_icache_dcache (pte_t pte)
55 {
56 	unsigned long addr;
57 	struct page *page;
58 
59 	page = pte_page(pte);
60 	addr = (unsigned long) page_address(page);
61 
62 	if (test_bit(PG_arch_1, &page->flags))
63 		return;				/* i-cache is already coherent with d-cache */
64 
65 	flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
66 	set_bit(PG_arch_1, &page->flags);	/* mark page as clean */
67 }
68 
69 /*
70  * Since DMA is i-cache coherent, any (complete) pages that were written via
71  * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
72  * flush them when they get mapped into an executable vm-area.
73  */
74 void
dma_mark_clean(void * addr,size_t size)75 dma_mark_clean(void *addr, size_t size)
76 {
77 	unsigned long pg_addr, end;
78 
79 	pg_addr = PAGE_ALIGN((unsigned long) addr);
80 	end = (unsigned long) addr + size;
81 	while (pg_addr + PAGE_SIZE <= end) {
82 		struct page *page = virt_to_page(pg_addr);
83 		set_bit(PG_arch_1, &page->flags);
84 		pg_addr += PAGE_SIZE;
85 	}
86 }
87 
88 inline void
ia64_set_rbs_bot(void)89 ia64_set_rbs_bot (void)
90 {
91 	unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
92 
93 	if (stack_size > MAX_USER_STACK_SIZE)
94 		stack_size = MAX_USER_STACK_SIZE;
95 	current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
96 }
97 
98 /*
99  * This performs some platform-dependent address space initialization.
100  * On IA-64, we want to setup the VM area for the register backing
101  * store (which grows upwards) and install the gateway page which is
102  * used for signal trampolines, etc.
103  */
104 void
ia64_init_addr_space(void)105 ia64_init_addr_space (void)
106 {
107 	struct vm_area_struct *vma;
108 
109 	ia64_set_rbs_bot();
110 
111 	/*
112 	 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
113 	 * the problem.  When the process attempts to write to the register backing store
114 	 * for the first time, it will get a SEGFAULT in this case.
115 	 */
116 	vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
117 	if (vma) {
118 		INIT_LIST_HEAD(&vma->anon_vma_chain);
119 		vma->vm_mm = current->mm;
120 		vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
121 		vma->vm_end = vma->vm_start + PAGE_SIZE;
122 		vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
123 		vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
124 		down_write(&current->mm->mmap_sem);
125 		if (insert_vm_struct(current->mm, vma)) {
126 			up_write(&current->mm->mmap_sem);
127 			kmem_cache_free(vm_area_cachep, vma);
128 			return;
129 		}
130 		up_write(&current->mm->mmap_sem);
131 	}
132 
133 	/* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
134 	if (!(current->personality & MMAP_PAGE_ZERO)) {
135 		vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
136 		if (vma) {
137 			INIT_LIST_HEAD(&vma->anon_vma_chain);
138 			vma->vm_mm = current->mm;
139 			vma->vm_end = PAGE_SIZE;
140 			vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
141 			vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO |
142 					VM_DONTEXPAND | VM_DONTDUMP;
143 			down_write(&current->mm->mmap_sem);
144 			if (insert_vm_struct(current->mm, vma)) {
145 				up_write(&current->mm->mmap_sem);
146 				kmem_cache_free(vm_area_cachep, vma);
147 				return;
148 			}
149 			up_write(&current->mm->mmap_sem);
150 		}
151 	}
152 }
153 
154 void
free_initmem(void)155 free_initmem (void)
156 {
157 	free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end),
158 			   -1, "unused kernel");
159 }
160 
161 void __init
free_initrd_mem(unsigned long start,unsigned long end)162 free_initrd_mem (unsigned long start, unsigned long end)
163 {
164 	/*
165 	 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
166 	 * Thus EFI and the kernel may have different page sizes. It is
167 	 * therefore possible to have the initrd share the same page as
168 	 * the end of the kernel (given current setup).
169 	 *
170 	 * To avoid freeing/using the wrong page (kernel sized) we:
171 	 *	- align up the beginning of initrd
172 	 *	- align down the end of initrd
173 	 *
174 	 *  |             |
175 	 *  |=============| a000
176 	 *  |             |
177 	 *  |             |
178 	 *  |             | 9000
179 	 *  |/////////////|
180 	 *  |/////////////|
181 	 *  |=============| 8000
182 	 *  |///INITRD////|
183 	 *  |/////////////|
184 	 *  |/////////////| 7000
185 	 *  |             |
186 	 *  |KKKKKKKKKKKKK|
187 	 *  |=============| 6000
188 	 *  |KKKKKKKKKKKKK|
189 	 *  |KKKKKKKKKKKKK|
190 	 *  K=kernel using 8KB pages
191 	 *
192 	 * In this example, we must free page 8000 ONLY. So we must align up
193 	 * initrd_start and keep initrd_end as is.
194 	 */
195 	start = PAGE_ALIGN(start);
196 	end = end & PAGE_MASK;
197 
198 	if (start < end)
199 		printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
200 
201 	for (; start < end; start += PAGE_SIZE) {
202 		if (!virt_addr_valid(start))
203 			continue;
204 		free_reserved_page(virt_to_page(start));
205 	}
206 }
207 
208 /*
209  * This installs a clean page in the kernel's page table.
210  */
211 static struct page * __init
put_kernel_page(struct page * page,unsigned long address,pgprot_t pgprot)212 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
213 {
214 	pgd_t *pgd;
215 	pud_t *pud;
216 	pmd_t *pmd;
217 	pte_t *pte;
218 
219 	if (!PageReserved(page))
220 		printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
221 		       page_address(page));
222 
223 	pgd = pgd_offset_k(address);		/* note: this is NOT pgd_offset()! */
224 
225 	{
226 		pud = pud_alloc(&init_mm, pgd, address);
227 		if (!pud)
228 			goto out;
229 		pmd = pmd_alloc(&init_mm, pud, address);
230 		if (!pmd)
231 			goto out;
232 		pte = pte_alloc_kernel(pmd, address);
233 		if (!pte)
234 			goto out;
235 		if (!pte_none(*pte))
236 			goto out;
237 		set_pte(pte, mk_pte(page, pgprot));
238 	}
239   out:
240 	/* no need for flush_tlb */
241 	return page;
242 }
243 
244 static void __init
setup_gate(void)245 setup_gate (void)
246 {
247 	void *gate_section;
248 	struct page *page;
249 
250 	/*
251 	 * Map the gate page twice: once read-only to export the ELF
252 	 * headers etc. and once execute-only page to enable
253 	 * privilege-promotion via "epc":
254 	 */
255 	gate_section = paravirt_get_gate_section();
256 	page = virt_to_page(ia64_imva(gate_section));
257 	put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
258 #ifdef HAVE_BUGGY_SEGREL
259 	page = virt_to_page(ia64_imva(gate_section + PAGE_SIZE));
260 	put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
261 #else
262 	put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
263 	/* Fill in the holes (if any) with read-only zero pages: */
264 	{
265 		unsigned long addr;
266 
267 		for (addr = GATE_ADDR + PAGE_SIZE;
268 		     addr < GATE_ADDR + PERCPU_PAGE_SIZE;
269 		     addr += PAGE_SIZE)
270 		{
271 			put_kernel_page(ZERO_PAGE(0), addr,
272 					PAGE_READONLY);
273 			put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
274 					PAGE_READONLY);
275 		}
276 	}
277 #endif
278 	ia64_patch_gate();
279 }
280 
281 static struct vm_area_struct gate_vma;
282 
gate_vma_init(void)283 static int __init gate_vma_init(void)
284 {
285 	gate_vma.vm_mm = NULL;
286 	gate_vma.vm_start = FIXADDR_USER_START;
287 	gate_vma.vm_end = FIXADDR_USER_END;
288 	gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
289 	gate_vma.vm_page_prot = __P101;
290 
291 	return 0;
292 }
293 __initcall(gate_vma_init);
294 
get_gate_vma(struct mm_struct * mm)295 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
296 {
297 	return &gate_vma;
298 }
299 
in_gate_area_no_mm(unsigned long addr)300 int in_gate_area_no_mm(unsigned long addr)
301 {
302 	if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
303 		return 1;
304 	return 0;
305 }
306 
in_gate_area(struct mm_struct * mm,unsigned long addr)307 int in_gate_area(struct mm_struct *mm, unsigned long addr)
308 {
309 	return in_gate_area_no_mm(addr);
310 }
311 
ia64_mmu_init(void * my_cpu_data)312 void ia64_mmu_init(void *my_cpu_data)
313 {
314 	unsigned long pta, impl_va_bits;
315 	extern void tlb_init(void);
316 
317 #ifdef CONFIG_DISABLE_VHPT
318 #	define VHPT_ENABLE_BIT	0
319 #else
320 #	define VHPT_ENABLE_BIT	1
321 #endif
322 
323 	/*
324 	 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
325 	 * address space.  The IA-64 architecture guarantees that at least 50 bits of
326 	 * virtual address space are implemented but if we pick a large enough page size
327 	 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
328 	 * VMLPT.  I assume that once we run on machines big enough to warrant 64KB pages,
329 	 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
330 	 * problem in practice.  Alternatively, we could truncate the top of the mapped
331 	 * address space to not permit mappings that would overlap with the VMLPT.
332 	 * --davidm 00/12/06
333 	 */
334 #	define pte_bits			3
335 #	define mapped_space_bits	(3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
336 	/*
337 	 * The virtual page table has to cover the entire implemented address space within
338 	 * a region even though not all of this space may be mappable.  The reason for
339 	 * this is that the Access bit and Dirty bit fault handlers perform
340 	 * non-speculative accesses to the virtual page table, so the address range of the
341 	 * virtual page table itself needs to be covered by virtual page table.
342 	 */
343 #	define vmlpt_bits		(impl_va_bits - PAGE_SHIFT + pte_bits)
344 #	define POW2(n)			(1ULL << (n))
345 
346 	impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
347 
348 	if (impl_va_bits < 51 || impl_va_bits > 61)
349 		panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
350 	/*
351 	 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
352 	 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
353 	 * the test makes sure that our mapped space doesn't overlap the
354 	 * unimplemented hole in the middle of the region.
355 	 */
356 	if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
357 	    (mapped_space_bits > impl_va_bits - 1))
358 		panic("Cannot build a big enough virtual-linear page table"
359 		      " to cover mapped address space.\n"
360 		      " Try using a smaller page size.\n");
361 
362 
363 	/* place the VMLPT at the end of each page-table mapped region: */
364 	pta = POW2(61) - POW2(vmlpt_bits);
365 
366 	/*
367 	 * Set the (virtually mapped linear) page table address.  Bit
368 	 * 8 selects between the short and long format, bits 2-7 the
369 	 * size of the table, and bit 0 whether the VHPT walker is
370 	 * enabled.
371 	 */
372 	ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
373 
374 	ia64_tlb_init();
375 
376 #ifdef	CONFIG_HUGETLB_PAGE
377 	ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
378 	ia64_srlz_d();
379 #endif
380 }
381 
382 #ifdef CONFIG_VIRTUAL_MEM_MAP
vmemmap_find_next_valid_pfn(int node,int i)383 int vmemmap_find_next_valid_pfn(int node, int i)
384 {
385 	unsigned long end_address, hole_next_pfn;
386 	unsigned long stop_address;
387 	pg_data_t *pgdat = NODE_DATA(node);
388 
389 	end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
390 	end_address = PAGE_ALIGN(end_address);
391 	stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
392 
393 	do {
394 		pgd_t *pgd;
395 		pud_t *pud;
396 		pmd_t *pmd;
397 		pte_t *pte;
398 
399 		pgd = pgd_offset_k(end_address);
400 		if (pgd_none(*pgd)) {
401 			end_address += PGDIR_SIZE;
402 			continue;
403 		}
404 
405 		pud = pud_offset(pgd, end_address);
406 		if (pud_none(*pud)) {
407 			end_address += PUD_SIZE;
408 			continue;
409 		}
410 
411 		pmd = pmd_offset(pud, end_address);
412 		if (pmd_none(*pmd)) {
413 			end_address += PMD_SIZE;
414 			continue;
415 		}
416 
417 		pte = pte_offset_kernel(pmd, end_address);
418 retry_pte:
419 		if (pte_none(*pte)) {
420 			end_address += PAGE_SIZE;
421 			pte++;
422 			if ((end_address < stop_address) &&
423 			    (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
424 				goto retry_pte;
425 			continue;
426 		}
427 		/* Found next valid vmem_map page */
428 		break;
429 	} while (end_address < stop_address);
430 
431 	end_address = min(end_address, stop_address);
432 	end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
433 	hole_next_pfn = end_address / sizeof(struct page);
434 	return hole_next_pfn - pgdat->node_start_pfn;
435 }
436 
create_mem_map_page_table(u64 start,u64 end,void * arg)437 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
438 {
439 	unsigned long address, start_page, end_page;
440 	struct page *map_start, *map_end;
441 	int node;
442 	pgd_t *pgd;
443 	pud_t *pud;
444 	pmd_t *pmd;
445 	pte_t *pte;
446 
447 	map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
448 	map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
449 
450 	start_page = (unsigned long) map_start & PAGE_MASK;
451 	end_page = PAGE_ALIGN((unsigned long) map_end);
452 	node = paddr_to_nid(__pa(start));
453 
454 	for (address = start_page; address < end_page; address += PAGE_SIZE) {
455 		pgd = pgd_offset_k(address);
456 		if (pgd_none(*pgd))
457 			pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
458 		pud = pud_offset(pgd, address);
459 
460 		if (pud_none(*pud))
461 			pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
462 		pmd = pmd_offset(pud, address);
463 
464 		if (pmd_none(*pmd))
465 			pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
466 		pte = pte_offset_kernel(pmd, address);
467 
468 		if (pte_none(*pte))
469 			set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
470 					     PAGE_KERNEL));
471 	}
472 	return 0;
473 }
474 
475 struct memmap_init_callback_data {
476 	struct page *start;
477 	struct page *end;
478 	int nid;
479 	unsigned long zone;
480 };
481 
482 static int __meminit
virtual_memmap_init(u64 start,u64 end,void * arg)483 virtual_memmap_init(u64 start, u64 end, void *arg)
484 {
485 	struct memmap_init_callback_data *args;
486 	struct page *map_start, *map_end;
487 
488 	args = (struct memmap_init_callback_data *) arg;
489 	map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
490 	map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
491 
492 	if (map_start < args->start)
493 		map_start = args->start;
494 	if (map_end > args->end)
495 		map_end = args->end;
496 
497 	/*
498 	 * We have to initialize "out of bounds" struct page elements that fit completely
499 	 * on the same pages that were allocated for the "in bounds" elements because they
500 	 * may be referenced later (and found to be "reserved").
501 	 */
502 	map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
503 	map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
504 		    / sizeof(struct page));
505 
506 	if (map_start < map_end)
507 		memmap_init_zone((unsigned long)(map_end - map_start),
508 				 args->nid, args->zone, page_to_pfn(map_start),
509 				 MEMMAP_EARLY);
510 	return 0;
511 }
512 
513 void __meminit
memmap_init(unsigned long size,int nid,unsigned long zone,unsigned long start_pfn)514 memmap_init (unsigned long size, int nid, unsigned long zone,
515 	     unsigned long start_pfn)
516 {
517 	if (!vmem_map)
518 		memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
519 	else {
520 		struct page *start;
521 		struct memmap_init_callback_data args;
522 
523 		start = pfn_to_page(start_pfn);
524 		args.start = start;
525 		args.end = start + size;
526 		args.nid = nid;
527 		args.zone = zone;
528 
529 		efi_memmap_walk(virtual_memmap_init, &args);
530 	}
531 }
532 
533 int
ia64_pfn_valid(unsigned long pfn)534 ia64_pfn_valid (unsigned long pfn)
535 {
536 	char byte;
537 	struct page *pg = pfn_to_page(pfn);
538 
539 	return     (__get_user(byte, (char __user *) pg) == 0)
540 		&& ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
541 			|| (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
542 }
543 EXPORT_SYMBOL(ia64_pfn_valid);
544 
find_largest_hole(u64 start,u64 end,void * arg)545 int __init find_largest_hole(u64 start, u64 end, void *arg)
546 {
547 	u64 *max_gap = arg;
548 
549 	static u64 last_end = PAGE_OFFSET;
550 
551 	/* NOTE: this algorithm assumes efi memmap table is ordered */
552 
553 	if (*max_gap < (start - last_end))
554 		*max_gap = start - last_end;
555 	last_end = end;
556 	return 0;
557 }
558 
559 #endif /* CONFIG_VIRTUAL_MEM_MAP */
560 
register_active_ranges(u64 start,u64 len,int nid)561 int __init register_active_ranges(u64 start, u64 len, int nid)
562 {
563 	u64 end = start + len;
564 
565 #ifdef CONFIG_KEXEC
566 	if (start > crashk_res.start && start < crashk_res.end)
567 		start = crashk_res.end;
568 	if (end > crashk_res.start && end < crashk_res.end)
569 		end = crashk_res.start;
570 #endif
571 
572 	if (start < end)
573 		memblock_add_node(__pa(start), end - start, nid);
574 	return 0;
575 }
576 
577 int
find_max_min_low_pfn(u64 start,u64 end,void * arg)578 find_max_min_low_pfn (u64 start, u64 end, void *arg)
579 {
580 	unsigned long pfn_start, pfn_end;
581 #ifdef CONFIG_FLATMEM
582 	pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
583 	pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
584 #else
585 	pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
586 	pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
587 #endif
588 	min_low_pfn = min(min_low_pfn, pfn_start);
589 	max_low_pfn = max(max_low_pfn, pfn_end);
590 	return 0;
591 }
592 
593 /*
594  * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
595  * system call handler.  When this option is in effect, all fsyscalls will end up bubbling
596  * down into the kernel and calling the normal (heavy-weight) syscall handler.  This is
597  * useful for performance testing, but conceivably could also come in handy for debugging
598  * purposes.
599  */
600 
601 static int nolwsys __initdata;
602 
603 static int __init
nolwsys_setup(char * s)604 nolwsys_setup (char *s)
605 {
606 	nolwsys = 1;
607 	return 1;
608 }
609 
610 __setup("nolwsys", nolwsys_setup);
611 
612 void __init
mem_init(void)613 mem_init (void)
614 {
615 	int i;
616 
617 	BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
618 	BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
619 	BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
620 
621 #ifdef CONFIG_PCI
622 	/*
623 	 * This needs to be called _after_ the command line has been parsed but _before_
624 	 * any drivers that may need the PCI DMA interface are initialized or bootmem has
625 	 * been freed.
626 	 */
627 	platform_dma_init();
628 #endif
629 
630 #ifdef CONFIG_FLATMEM
631 	BUG_ON(!mem_map);
632 #endif
633 
634 	set_max_mapnr(max_low_pfn);
635 	high_memory = __va(max_low_pfn * PAGE_SIZE);
636 	free_all_bootmem();
637 	mem_init_print_info(NULL);
638 
639 	/*
640 	 * For fsyscall entrpoints with no light-weight handler, use the ordinary
641 	 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
642 	 * code can tell them apart.
643 	 */
644 	for (i = 0; i < NR_syscalls; ++i) {
645 		extern unsigned long sys_call_table[NR_syscalls];
646 		unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
647 
648 		if (!fsyscall_table[i] || nolwsys)
649 			fsyscall_table[i] = sys_call_table[i] | 1;
650 	}
651 	setup_gate();
652 }
653 
654 #ifdef CONFIG_MEMORY_HOTPLUG
arch_add_memory(int nid,u64 start,u64 size)655 int arch_add_memory(int nid, u64 start, u64 size)
656 {
657 	pg_data_t *pgdat;
658 	struct zone *zone;
659 	unsigned long start_pfn = start >> PAGE_SHIFT;
660 	unsigned long nr_pages = size >> PAGE_SHIFT;
661 	int ret;
662 
663 	pgdat = NODE_DATA(nid);
664 
665 	zone = pgdat->node_zones +
666 		zone_for_memory(nid, start, size, ZONE_NORMAL);
667 	ret = __add_pages(nid, zone, start_pfn, nr_pages);
668 
669 	if (ret)
670 		printk("%s: Problem encountered in __add_pages() as ret=%d\n",
671 		       __func__,  ret);
672 
673 	return ret;
674 }
675 
676 #ifdef CONFIG_MEMORY_HOTREMOVE
arch_remove_memory(u64 start,u64 size)677 int arch_remove_memory(u64 start, u64 size)
678 {
679 	unsigned long start_pfn = start >> PAGE_SHIFT;
680 	unsigned long nr_pages = size >> PAGE_SHIFT;
681 	struct zone *zone;
682 	int ret;
683 
684 	zone = page_zone(pfn_to_page(start_pfn));
685 	ret = __remove_pages(zone, start_pfn, nr_pages);
686 	if (ret)
687 		pr_warn("%s: Problem encountered in __remove_pages() as"
688 			" ret=%d\n", __func__,  ret);
689 
690 	return ret;
691 }
692 #endif
693 #endif
694 
695 /*
696  * Even when CONFIG_IA32_SUPPORT is not enabled it is
697  * useful to have the Linux/x86 domain registered to
698  * avoid an attempted module load when emulators call
699  * personality(PER_LINUX32). This saves several milliseconds
700  * on each such call.
701  */
702 static struct exec_domain ia32_exec_domain;
703 
704 static int __init
per_linux32_init(void)705 per_linux32_init(void)
706 {
707 	ia32_exec_domain.name = "Linux/x86";
708 	ia32_exec_domain.handler = NULL;
709 	ia32_exec_domain.pers_low = PER_LINUX32;
710 	ia32_exec_domain.pers_high = PER_LINUX32;
711 	ia32_exec_domain.signal_map = default_exec_domain.signal_map;
712 	ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
713 	register_exec_domain(&ia32_exec_domain);
714 
715 	return 0;
716 }
717 
718 __initcall(per_linux32_init);
719 
720 /**
721  * show_mem - give short summary of memory stats
722  *
723  * Shows a simple page count of reserved and used pages in the system.
724  * For discontig machines, it does this on a per-pgdat basis.
725  */
show_mem(unsigned int filter)726 void show_mem(unsigned int filter)
727 {
728 	int total_reserved = 0;
729 	unsigned long total_present = 0;
730 	pg_data_t *pgdat;
731 
732 	printk(KERN_INFO "Mem-info:\n");
733 	show_free_areas(filter);
734 	printk(KERN_INFO "Node memory in pages:\n");
735 	for_each_online_pgdat(pgdat) {
736 		unsigned long present;
737 		unsigned long flags;
738 		int reserved = 0;
739 		int nid = pgdat->node_id;
740 		int zoneid;
741 
742 		if (skip_free_areas_node(filter, nid))
743 			continue;
744 		pgdat_resize_lock(pgdat, &flags);
745 
746 		for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
747 			struct zone *zone = &pgdat->node_zones[zoneid];
748 			if (!populated_zone(zone))
749 				continue;
750 
751 			reserved += zone->present_pages - zone->managed_pages;
752 		}
753 		present = pgdat->node_present_pages;
754 
755 		pgdat_resize_unlock(pgdat, &flags);
756 		total_present += present;
757 		total_reserved += reserved;
758 		printk(KERN_INFO "Node %4d:  RAM: %11ld, rsvd: %8d, ",
759 		       nid, present, reserved);
760 	}
761 	printk(KERN_INFO "%ld pages of RAM\n", total_present);
762 	printk(KERN_INFO "%d reserved pages\n", total_reserved);
763 	printk(KERN_INFO "Total of %ld pages in page table cache\n",
764 	       quicklist_total_size());
765 	printk(KERN_INFO "%ld free buffer pages\n", nr_free_buffer_pages());
766 }
767