xref: /openbmc/linux/arch/ia64/mm/init.c (revision 94c7b6fc)
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
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
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
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
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
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
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
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
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 void ia64_mmu_init(void *my_cpu_data)
282 {
283 	unsigned long pta, impl_va_bits;
284 	extern void tlb_init(void);
285 
286 #ifdef CONFIG_DISABLE_VHPT
287 #	define VHPT_ENABLE_BIT	0
288 #else
289 #	define VHPT_ENABLE_BIT	1
290 #endif
291 
292 	/*
293 	 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
294 	 * address space.  The IA-64 architecture guarantees that at least 50 bits of
295 	 * virtual address space are implemented but if we pick a large enough page size
296 	 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
297 	 * VMLPT.  I assume that once we run on machines big enough to warrant 64KB pages,
298 	 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
299 	 * problem in practice.  Alternatively, we could truncate the top of the mapped
300 	 * address space to not permit mappings that would overlap with the VMLPT.
301 	 * --davidm 00/12/06
302 	 */
303 #	define pte_bits			3
304 #	define mapped_space_bits	(3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
305 	/*
306 	 * The virtual page table has to cover the entire implemented address space within
307 	 * a region even though not all of this space may be mappable.  The reason for
308 	 * this is that the Access bit and Dirty bit fault handlers perform
309 	 * non-speculative accesses to the virtual page table, so the address range of the
310 	 * virtual page table itself needs to be covered by virtual page table.
311 	 */
312 #	define vmlpt_bits		(impl_va_bits - PAGE_SHIFT + pte_bits)
313 #	define POW2(n)			(1ULL << (n))
314 
315 	impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
316 
317 	if (impl_va_bits < 51 || impl_va_bits > 61)
318 		panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
319 	/*
320 	 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
321 	 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
322 	 * the test makes sure that our mapped space doesn't overlap the
323 	 * unimplemented hole in the middle of the region.
324 	 */
325 	if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
326 	    (mapped_space_bits > impl_va_bits - 1))
327 		panic("Cannot build a big enough virtual-linear page table"
328 		      " to cover mapped address space.\n"
329 		      " Try using a smaller page size.\n");
330 
331 
332 	/* place the VMLPT at the end of each page-table mapped region: */
333 	pta = POW2(61) - POW2(vmlpt_bits);
334 
335 	/*
336 	 * Set the (virtually mapped linear) page table address.  Bit
337 	 * 8 selects between the short and long format, bits 2-7 the
338 	 * size of the table, and bit 0 whether the VHPT walker is
339 	 * enabled.
340 	 */
341 	ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
342 
343 	ia64_tlb_init();
344 
345 #ifdef	CONFIG_HUGETLB_PAGE
346 	ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
347 	ia64_srlz_d();
348 #endif
349 }
350 
351 #ifdef CONFIG_VIRTUAL_MEM_MAP
352 int vmemmap_find_next_valid_pfn(int node, int i)
353 {
354 	unsigned long end_address, hole_next_pfn;
355 	unsigned long stop_address;
356 	pg_data_t *pgdat = NODE_DATA(node);
357 
358 	end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
359 	end_address = PAGE_ALIGN(end_address);
360 	stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
361 
362 	do {
363 		pgd_t *pgd;
364 		pud_t *pud;
365 		pmd_t *pmd;
366 		pte_t *pte;
367 
368 		pgd = pgd_offset_k(end_address);
369 		if (pgd_none(*pgd)) {
370 			end_address += PGDIR_SIZE;
371 			continue;
372 		}
373 
374 		pud = pud_offset(pgd, end_address);
375 		if (pud_none(*pud)) {
376 			end_address += PUD_SIZE;
377 			continue;
378 		}
379 
380 		pmd = pmd_offset(pud, end_address);
381 		if (pmd_none(*pmd)) {
382 			end_address += PMD_SIZE;
383 			continue;
384 		}
385 
386 		pte = pte_offset_kernel(pmd, end_address);
387 retry_pte:
388 		if (pte_none(*pte)) {
389 			end_address += PAGE_SIZE;
390 			pte++;
391 			if ((end_address < stop_address) &&
392 			    (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
393 				goto retry_pte;
394 			continue;
395 		}
396 		/* Found next valid vmem_map page */
397 		break;
398 	} while (end_address < stop_address);
399 
400 	end_address = min(end_address, stop_address);
401 	end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
402 	hole_next_pfn = end_address / sizeof(struct page);
403 	return hole_next_pfn - pgdat->node_start_pfn;
404 }
405 
406 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
407 {
408 	unsigned long address, start_page, end_page;
409 	struct page *map_start, *map_end;
410 	int node;
411 	pgd_t *pgd;
412 	pud_t *pud;
413 	pmd_t *pmd;
414 	pte_t *pte;
415 
416 	map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
417 	map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
418 
419 	start_page = (unsigned long) map_start & PAGE_MASK;
420 	end_page = PAGE_ALIGN((unsigned long) map_end);
421 	node = paddr_to_nid(__pa(start));
422 
423 	for (address = start_page; address < end_page; address += PAGE_SIZE) {
424 		pgd = pgd_offset_k(address);
425 		if (pgd_none(*pgd))
426 			pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
427 		pud = pud_offset(pgd, address);
428 
429 		if (pud_none(*pud))
430 			pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
431 		pmd = pmd_offset(pud, address);
432 
433 		if (pmd_none(*pmd))
434 			pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
435 		pte = pte_offset_kernel(pmd, address);
436 
437 		if (pte_none(*pte))
438 			set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
439 					     PAGE_KERNEL));
440 	}
441 	return 0;
442 }
443 
444 struct memmap_init_callback_data {
445 	struct page *start;
446 	struct page *end;
447 	int nid;
448 	unsigned long zone;
449 };
450 
451 static int __meminit
452 virtual_memmap_init(u64 start, u64 end, void *arg)
453 {
454 	struct memmap_init_callback_data *args;
455 	struct page *map_start, *map_end;
456 
457 	args = (struct memmap_init_callback_data *) arg;
458 	map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
459 	map_end   = vmem_map + (__pa(end) >> PAGE_SHIFT);
460 
461 	if (map_start < args->start)
462 		map_start = args->start;
463 	if (map_end > args->end)
464 		map_end = args->end;
465 
466 	/*
467 	 * We have to initialize "out of bounds" struct page elements that fit completely
468 	 * on the same pages that were allocated for the "in bounds" elements because they
469 	 * may be referenced later (and found to be "reserved").
470 	 */
471 	map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
472 	map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
473 		    / sizeof(struct page));
474 
475 	if (map_start < map_end)
476 		memmap_init_zone((unsigned long)(map_end - map_start),
477 				 args->nid, args->zone, page_to_pfn(map_start),
478 				 MEMMAP_EARLY);
479 	return 0;
480 }
481 
482 void __meminit
483 memmap_init (unsigned long size, int nid, unsigned long zone,
484 	     unsigned long start_pfn)
485 {
486 	if (!vmem_map)
487 		memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
488 	else {
489 		struct page *start;
490 		struct memmap_init_callback_data args;
491 
492 		start = pfn_to_page(start_pfn);
493 		args.start = start;
494 		args.end = start + size;
495 		args.nid = nid;
496 		args.zone = zone;
497 
498 		efi_memmap_walk(virtual_memmap_init, &args);
499 	}
500 }
501 
502 int
503 ia64_pfn_valid (unsigned long pfn)
504 {
505 	char byte;
506 	struct page *pg = pfn_to_page(pfn);
507 
508 	return     (__get_user(byte, (char __user *) pg) == 0)
509 		&& ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
510 			|| (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
511 }
512 EXPORT_SYMBOL(ia64_pfn_valid);
513 
514 int __init find_largest_hole(u64 start, u64 end, void *arg)
515 {
516 	u64 *max_gap = arg;
517 
518 	static u64 last_end = PAGE_OFFSET;
519 
520 	/* NOTE: this algorithm assumes efi memmap table is ordered */
521 
522 	if (*max_gap < (start - last_end))
523 		*max_gap = start - last_end;
524 	last_end = end;
525 	return 0;
526 }
527 
528 #endif /* CONFIG_VIRTUAL_MEM_MAP */
529 
530 int __init register_active_ranges(u64 start, u64 len, int nid)
531 {
532 	u64 end = start + len;
533 
534 #ifdef CONFIG_KEXEC
535 	if (start > crashk_res.start && start < crashk_res.end)
536 		start = crashk_res.end;
537 	if (end > crashk_res.start && end < crashk_res.end)
538 		end = crashk_res.start;
539 #endif
540 
541 	if (start < end)
542 		memblock_add_node(__pa(start), end - start, nid);
543 	return 0;
544 }
545 
546 int
547 find_max_min_low_pfn (u64 start, u64 end, void *arg)
548 {
549 	unsigned long pfn_start, pfn_end;
550 #ifdef CONFIG_FLATMEM
551 	pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
552 	pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
553 #else
554 	pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
555 	pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
556 #endif
557 	min_low_pfn = min(min_low_pfn, pfn_start);
558 	max_low_pfn = max(max_low_pfn, pfn_end);
559 	return 0;
560 }
561 
562 /*
563  * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
564  * system call handler.  When this option is in effect, all fsyscalls will end up bubbling
565  * down into the kernel and calling the normal (heavy-weight) syscall handler.  This is
566  * useful for performance testing, but conceivably could also come in handy for debugging
567  * purposes.
568  */
569 
570 static int nolwsys __initdata;
571 
572 static int __init
573 nolwsys_setup (char *s)
574 {
575 	nolwsys = 1;
576 	return 1;
577 }
578 
579 __setup("nolwsys", nolwsys_setup);
580 
581 void __init
582 mem_init (void)
583 {
584 	int i;
585 
586 	BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
587 	BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
588 	BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
589 
590 #ifdef CONFIG_PCI
591 	/*
592 	 * This needs to be called _after_ the command line has been parsed but _before_
593 	 * any drivers that may need the PCI DMA interface are initialized or bootmem has
594 	 * been freed.
595 	 */
596 	platform_dma_init();
597 #endif
598 
599 #ifdef CONFIG_FLATMEM
600 	BUG_ON(!mem_map);
601 #endif
602 
603 	set_max_mapnr(max_low_pfn);
604 	high_memory = __va(max_low_pfn * PAGE_SIZE);
605 	free_all_bootmem();
606 	mem_init_print_info(NULL);
607 
608 	/*
609 	 * For fsyscall entrpoints with no light-weight handler, use the ordinary
610 	 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
611 	 * code can tell them apart.
612 	 */
613 	for (i = 0; i < NR_syscalls; ++i) {
614 		extern unsigned long sys_call_table[NR_syscalls];
615 		unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
616 
617 		if (!fsyscall_table[i] || nolwsys)
618 			fsyscall_table[i] = sys_call_table[i] | 1;
619 	}
620 	setup_gate();
621 }
622 
623 #ifdef CONFIG_MEMORY_HOTPLUG
624 int arch_add_memory(int nid, u64 start, u64 size)
625 {
626 	pg_data_t *pgdat;
627 	struct zone *zone;
628 	unsigned long start_pfn = start >> PAGE_SHIFT;
629 	unsigned long nr_pages = size >> PAGE_SHIFT;
630 	int ret;
631 
632 	pgdat = NODE_DATA(nid);
633 
634 	zone = pgdat->node_zones + ZONE_NORMAL;
635 	ret = __add_pages(nid, zone, start_pfn, nr_pages);
636 
637 	if (ret)
638 		printk("%s: Problem encountered in __add_pages() as ret=%d\n",
639 		       __func__,  ret);
640 
641 	return ret;
642 }
643 
644 #ifdef CONFIG_MEMORY_HOTREMOVE
645 int arch_remove_memory(u64 start, u64 size)
646 {
647 	unsigned long start_pfn = start >> PAGE_SHIFT;
648 	unsigned long nr_pages = size >> PAGE_SHIFT;
649 	struct zone *zone;
650 	int ret;
651 
652 	zone = page_zone(pfn_to_page(start_pfn));
653 	ret = __remove_pages(zone, start_pfn, nr_pages);
654 	if (ret)
655 		pr_warn("%s: Problem encountered in __remove_pages() as"
656 			" ret=%d\n", __func__,  ret);
657 
658 	return ret;
659 }
660 #endif
661 #endif
662 
663 /*
664  * Even when CONFIG_IA32_SUPPORT is not enabled it is
665  * useful to have the Linux/x86 domain registered to
666  * avoid an attempted module load when emulators call
667  * personality(PER_LINUX32). This saves several milliseconds
668  * on each such call.
669  */
670 static struct exec_domain ia32_exec_domain;
671 
672 static int __init
673 per_linux32_init(void)
674 {
675 	ia32_exec_domain.name = "Linux/x86";
676 	ia32_exec_domain.handler = NULL;
677 	ia32_exec_domain.pers_low = PER_LINUX32;
678 	ia32_exec_domain.pers_high = PER_LINUX32;
679 	ia32_exec_domain.signal_map = default_exec_domain.signal_map;
680 	ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
681 	register_exec_domain(&ia32_exec_domain);
682 
683 	return 0;
684 }
685 
686 __initcall(per_linux32_init);
687 
688 /**
689  * show_mem - give short summary of memory stats
690  *
691  * Shows a simple page count of reserved and used pages in the system.
692  * For discontig machines, it does this on a per-pgdat basis.
693  */
694 void show_mem(unsigned int filter)
695 {
696 	int total_reserved = 0;
697 	unsigned long total_present = 0;
698 	pg_data_t *pgdat;
699 
700 	printk(KERN_INFO "Mem-info:\n");
701 	show_free_areas(filter);
702 	printk(KERN_INFO "Node memory in pages:\n");
703 	for_each_online_pgdat(pgdat) {
704 		unsigned long present;
705 		unsigned long flags;
706 		int reserved = 0;
707 		int nid = pgdat->node_id;
708 		int zoneid;
709 
710 		if (skip_free_areas_node(filter, nid))
711 			continue;
712 		pgdat_resize_lock(pgdat, &flags);
713 
714 		for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
715 			struct zone *zone = &pgdat->node_zones[zoneid];
716 			if (!populated_zone(zone))
717 				continue;
718 
719 			reserved += zone->present_pages - zone->managed_pages;
720 		}
721 		present = pgdat->node_present_pages;
722 
723 		pgdat_resize_unlock(pgdat, &flags);
724 		total_present += present;
725 		total_reserved += reserved;
726 		printk(KERN_INFO "Node %4d:  RAM: %11ld, rsvd: %8d, ",
727 		       nid, present, reserved);
728 	}
729 	printk(KERN_INFO "%ld pages of RAM\n", total_present);
730 	printk(KERN_INFO "%d reserved pages\n", total_reserved);
731 	printk(KERN_INFO "Total of %ld pages in page table cache\n",
732 	       quicklist_total_size());
733 	printk(KERN_INFO "%ld free buffer pages\n", nr_free_buffer_pages());
734 }
735