Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Initialize MMU support.
4 *
5 * Copyright (C) 1998-2003 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
7 */
8#include <linux/kernel.h>
9#include <linux/init.h>
10
11#include <linux/dma-noncoherent.h>
12#include <linux/dmar.h>
13#include <linux/efi.h>
14#include <linux/elf.h>
15#include <linux/memblock.h>
16#include <linux/mm.h>
17#include <linux/sched/signal.h>
18#include <linux/mmzone.h>
19#include <linux/module.h>
20#include <linux/personality.h>
21#include <linux/reboot.h>
22#include <linux/slab.h>
23#include <linux/swap.h>
24#include <linux/proc_fs.h>
25#include <linux/bitops.h>
26#include <linux/kexec.h>
27#include <linux/swiotlb.h>
28
29#include <asm/dma.h>
30#include <asm/io.h>
31#include <asm/numa.h>
32#include <asm/patch.h>
33#include <asm/pgalloc.h>
34#include <asm/sal.h>
35#include <asm/sections.h>
36#include <asm/tlb.h>
37#include <linux/uaccess.h>
38#include <asm/unistd.h>
39#include <asm/mca.h>
40
41extern void ia64_tlb_init (void);
42
43unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
44
45#ifdef CONFIG_VIRTUAL_MEM_MAP
46unsigned long VMALLOC_END = VMALLOC_END_INIT;
47EXPORT_SYMBOL(VMALLOC_END);
48struct page *vmem_map;
49EXPORT_SYMBOL(vmem_map);
50#endif
51
52struct page *zero_page_memmap_ptr; /* map entry for zero page */
53EXPORT_SYMBOL(zero_page_memmap_ptr);
54
55void
56__ia64_sync_icache_dcache (pte_t pte)
57{
58 unsigned long addr;
59 struct page *page;
60
61 page = pte_page(pte);
62 addr = (unsigned long) page_address(page);
63
64 if (test_bit(PG_arch_1, &page->flags))
65 return; /* i-cache is already coherent with d-cache */
66
67 flush_icache_range(addr, addr + page_size(page));
68 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
69}
70
71/*
72 * Since DMA is i-cache coherent, any (complete) pages that were written via
73 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
74 * flush them when they get mapped into an executable vm-area.
75 */
76void arch_sync_dma_for_cpu(phys_addr_t paddr, size_t size,
77 enum dma_data_direction dir)
78{
79 unsigned long pfn = PHYS_PFN(paddr);
80
81 do {
82 set_bit(PG_arch_1, &pfn_to_page(pfn)->flags);
83 } while (++pfn <= PHYS_PFN(paddr + size - 1));
84}
85
86inline void
87ia64_set_rbs_bot (void)
88{
89 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
90
91 if (stack_size > MAX_USER_STACK_SIZE)
92 stack_size = MAX_USER_STACK_SIZE;
93 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
94}
95
96/*
97 * This performs some platform-dependent address space initialization.
98 * On IA-64, we want to setup the VM area for the register backing
99 * store (which grows upwards) and install the gateway page which is
100 * used for signal trampolines, etc.
101 */
102void
103ia64_init_addr_space (void)
104{
105 struct vm_area_struct *vma;
106
107 ia64_set_rbs_bot();
108
109 /*
110 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
111 * the problem. When the process attempts to write to the register backing store
112 * for the first time, it will get a SEGFAULT in this case.
113 */
114 vma = vm_area_alloc(current->mm);
115 if (vma) {
116 vma_set_anonymous(vma);
117 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
118 vma->vm_end = vma->vm_start + PAGE_SIZE;
119 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
120 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
121 mmap_write_lock(current->mm);
122 if (insert_vm_struct(current->mm, vma)) {
123 mmap_write_unlock(current->mm);
124 vm_area_free(vma);
125 return;
126 }
127 mmap_write_unlock(current->mm);
128 }
129
130 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
131 if (!(current->personality & MMAP_PAGE_ZERO)) {
132 vma = vm_area_alloc(current->mm);
133 if (vma) {
134 vma_set_anonymous(vma);
135 vma->vm_end = PAGE_SIZE;
136 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
137 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO |
138 VM_DONTEXPAND | VM_DONTDUMP;
139 mmap_write_lock(current->mm);
140 if (insert_vm_struct(current->mm, vma)) {
141 mmap_write_unlock(current->mm);
142 vm_area_free(vma);
143 return;
144 }
145 mmap_write_unlock(current->mm);
146 }
147 }
148}
149
150void
151free_initmem (void)
152{
153 free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end),
154 -1, "unused kernel");
155}
156
157void __init
158free_initrd_mem (unsigned long start, unsigned long end)
159{
160 /*
161 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
162 * Thus EFI and the kernel may have different page sizes. It is
163 * therefore possible to have the initrd share the same page as
164 * the end of the kernel (given current setup).
165 *
166 * To avoid freeing/using the wrong page (kernel sized) we:
167 * - align up the beginning of initrd
168 * - align down the end of initrd
169 *
170 * | |
171 * |=============| a000
172 * | |
173 * | |
174 * | | 9000
175 * |/////////////|
176 * |/////////////|
177 * |=============| 8000
178 * |///INITRD////|
179 * |/////////////|
180 * |/////////////| 7000
181 * | |
182 * |KKKKKKKKKKKKK|
183 * |=============| 6000
184 * |KKKKKKKKKKKKK|
185 * |KKKKKKKKKKKKK|
186 * K=kernel using 8KB pages
187 *
188 * In this example, we must free page 8000 ONLY. So we must align up
189 * initrd_start and keep initrd_end as is.
190 */
191 start = PAGE_ALIGN(start);
192 end = end & PAGE_MASK;
193
194 if (start < end)
195 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
196
197 for (; start < end; start += PAGE_SIZE) {
198 if (!virt_addr_valid(start))
199 continue;
200 free_reserved_page(virt_to_page(start));
201 }
202}
203
204/*
205 * This installs a clean page in the kernel's page table.
206 */
207static struct page * __init
208put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
209{
210 pgd_t *pgd;
211 p4d_t *p4d;
212 pud_t *pud;
213 pmd_t *pmd;
214 pte_t *pte;
215
216 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
217
218 {
219 p4d = p4d_alloc(&init_mm, pgd, address);
220 if (!p4d)
221 goto out;
222 pud = pud_alloc(&init_mm, p4d, address);
223 if (!pud)
224 goto out;
225 pmd = pmd_alloc(&init_mm, pud, address);
226 if (!pmd)
227 goto out;
228 pte = pte_alloc_kernel(pmd, address);
229 if (!pte)
230 goto out;
231 if (!pte_none(*pte))
232 goto out;
233 set_pte(pte, mk_pte(page, pgprot));
234 }
235 out:
236 /* no need for flush_tlb */
237 return page;
238}
239
240static void __init
241setup_gate (void)
242{
243 struct page *page;
244
245 /*
246 * Map the gate page twice: once read-only to export the ELF
247 * headers etc. and once execute-only page to enable
248 * privilege-promotion via "epc":
249 */
250 page = virt_to_page(ia64_imva(__start_gate_section));
251 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
252#ifdef HAVE_BUGGY_SEGREL
253 page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
254 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
255#else
256 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
257 /* Fill in the holes (if any) with read-only zero pages: */
258 {
259 unsigned long addr;
260
261 for (addr = GATE_ADDR + PAGE_SIZE;
262 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
263 addr += PAGE_SIZE)
264 {
265 put_kernel_page(ZERO_PAGE(0), addr,
266 PAGE_READONLY);
267 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
268 PAGE_READONLY);
269 }
270 }
271#endif
272 ia64_patch_gate();
273}
274
275static struct vm_area_struct gate_vma;
276
277static int __init gate_vma_init(void)
278{
279 vma_init(&gate_vma, NULL);
280 gate_vma.vm_start = FIXADDR_USER_START;
281 gate_vma.vm_end = FIXADDR_USER_END;
282 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
283 gate_vma.vm_page_prot = __P101;
284
285 return 0;
286}
287__initcall(gate_vma_init);
288
289struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
290{
291 return &gate_vma;
292}
293
294int in_gate_area_no_mm(unsigned long addr)
295{
296 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
297 return 1;
298 return 0;
299}
300
301int in_gate_area(struct mm_struct *mm, unsigned long addr)
302{
303 return in_gate_area_no_mm(addr);
304}
305
306void ia64_mmu_init(void *my_cpu_data)
307{
308 unsigned long pta, impl_va_bits;
309 extern void tlb_init(void);
310
311#ifdef CONFIG_DISABLE_VHPT
312# define VHPT_ENABLE_BIT 0
313#else
314# define VHPT_ENABLE_BIT 1
315#endif
316
317 /*
318 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
319 * address space. The IA-64 architecture guarantees that at least 50 bits of
320 * virtual address space are implemented but if we pick a large enough page size
321 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
322 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
323 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
324 * problem in practice. Alternatively, we could truncate the top of the mapped
325 * address space to not permit mappings that would overlap with the VMLPT.
326 * --davidm 00/12/06
327 */
328# define pte_bits 3
329# define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
330 /*
331 * The virtual page table has to cover the entire implemented address space within
332 * a region even though not all of this space may be mappable. The reason for
333 * this is that the Access bit and Dirty bit fault handlers perform
334 * non-speculative accesses to the virtual page table, so the address range of the
335 * virtual page table itself needs to be covered by virtual page table.
336 */
337# define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
338# define POW2(n) (1ULL << (n))
339
340 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
341
342 if (impl_va_bits < 51 || impl_va_bits > 61)
343 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
344 /*
345 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
346 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
347 * the test makes sure that our mapped space doesn't overlap the
348 * unimplemented hole in the middle of the region.
349 */
350 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
351 (mapped_space_bits > impl_va_bits - 1))
352 panic("Cannot build a big enough virtual-linear page table"
353 " to cover mapped address space.\n"
354 " Try using a smaller page size.\n");
355
356
357 /* place the VMLPT at the end of each page-table mapped region: */
358 pta = POW2(61) - POW2(vmlpt_bits);
359
360 /*
361 * Set the (virtually mapped linear) page table address. Bit
362 * 8 selects between the short and long format, bits 2-7 the
363 * size of the table, and bit 0 whether the VHPT walker is
364 * enabled.
365 */
366 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
367
368 ia64_tlb_init();
369
370#ifdef CONFIG_HUGETLB_PAGE
371 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
372 ia64_srlz_d();
373#endif
374}
375
376#ifdef CONFIG_VIRTUAL_MEM_MAP
377int vmemmap_find_next_valid_pfn(int node, int i)
378{
379 unsigned long end_address, hole_next_pfn;
380 unsigned long stop_address;
381 pg_data_t *pgdat = NODE_DATA(node);
382
383 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
384 end_address = PAGE_ALIGN(end_address);
385 stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
386
387 do {
388 pgd_t *pgd;
389 p4d_t *p4d;
390 pud_t *pud;
391 pmd_t *pmd;
392 pte_t *pte;
393
394 pgd = pgd_offset_k(end_address);
395 if (pgd_none(*pgd)) {
396 end_address += PGDIR_SIZE;
397 continue;
398 }
399
400 p4d = p4d_offset(pgd, end_address);
401 if (p4d_none(*p4d)) {
402 end_address += P4D_SIZE;
403 continue;
404 }
405
406 pud = pud_offset(p4d, end_address);
407 if (pud_none(*pud)) {
408 end_address += PUD_SIZE;
409 continue;
410 }
411
412 pmd = pmd_offset(pud, end_address);
413 if (pmd_none(*pmd)) {
414 end_address += PMD_SIZE;
415 continue;
416 }
417
418 pte = pte_offset_kernel(pmd, end_address);
419retry_pte:
420 if (pte_none(*pte)) {
421 end_address += PAGE_SIZE;
422 pte++;
423 if ((end_address < stop_address) &&
424 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
425 goto retry_pte;
426 continue;
427 }
428 /* Found next valid vmem_map page */
429 break;
430 } while (end_address < stop_address);
431
432 end_address = min(end_address, stop_address);
433 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
434 hole_next_pfn = end_address / sizeof(struct page);
435 return hole_next_pfn - pgdat->node_start_pfn;
436}
437
438int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
439{
440 unsigned long address, start_page, end_page;
441 struct page *map_start, *map_end;
442 int node;
443 pgd_t *pgd;
444 p4d_t *p4d;
445 pud_t *pud;
446 pmd_t *pmd;
447 pte_t *pte;
448
449 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
450 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
451
452 start_page = (unsigned long) map_start & PAGE_MASK;
453 end_page = PAGE_ALIGN((unsigned long) map_end);
454 node = paddr_to_nid(__pa(start));
455
456 for (address = start_page; address < end_page; address += PAGE_SIZE) {
457 pgd = pgd_offset_k(address);
458 if (pgd_none(*pgd)) {
459 p4d = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
460 if (!p4d)
461 goto err_alloc;
462 pgd_populate(&init_mm, pgd, p4d);
463 }
464 p4d = p4d_offset(pgd, address);
465
466 if (p4d_none(*p4d)) {
467 pud = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
468 if (!pud)
469 goto err_alloc;
470 p4d_populate(&init_mm, p4d, pud);
471 }
472 pud = pud_offset(p4d, address);
473
474 if (pud_none(*pud)) {
475 pmd = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
476 if (!pmd)
477 goto err_alloc;
478 pud_populate(&init_mm, pud, pmd);
479 }
480 pmd = pmd_offset(pud, address);
481
482 if (pmd_none(*pmd)) {
483 pte = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
484 if (!pte)
485 goto err_alloc;
486 pmd_populate_kernel(&init_mm, pmd, pte);
487 }
488 pte = pte_offset_kernel(pmd, address);
489
490 if (pte_none(*pte)) {
491 void *page = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE,
492 node);
493 if (!page)
494 goto err_alloc;
495 set_pte(pte, pfn_pte(__pa(page) >> PAGE_SHIFT,
496 PAGE_KERNEL));
497 }
498 }
499 return 0;
500
501err_alloc:
502 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d\n",
503 __func__, PAGE_SIZE, PAGE_SIZE, node);
504 return -ENOMEM;
505}
506
507struct memmap_init_callback_data {
508 struct page *start;
509 struct page *end;
510 int nid;
511 unsigned long zone;
512};
513
514static int __meminit
515virtual_memmap_init(u64 start, u64 end, void *arg)
516{
517 struct memmap_init_callback_data *args;
518 struct page *map_start, *map_end;
519
520 args = (struct memmap_init_callback_data *) arg;
521 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
522 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
523
524 if (map_start < args->start)
525 map_start = args->start;
526 if (map_end > args->end)
527 map_end = args->end;
528
529 /*
530 * We have to initialize "out of bounds" struct page elements that fit completely
531 * on the same pages that were allocated for the "in bounds" elements because they
532 * may be referenced later (and found to be "reserved").
533 */
534 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
535 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
536 / sizeof(struct page));
537
538 if (map_start < map_end)
539 memmap_init_zone((unsigned long)(map_end - map_start),
540 args->nid, args->zone, page_to_pfn(map_start),
541 MEMINIT_EARLY, NULL);
542 return 0;
543}
544
545void __meminit
546memmap_init (unsigned long size, int nid, unsigned long zone,
547 unsigned long start_pfn)
548{
549 if (!vmem_map) {
550 memmap_init_zone(size, nid, zone, start_pfn,
551 MEMINIT_EARLY, NULL);
552 } else {
553 struct page *start;
554 struct memmap_init_callback_data args;
555
556 start = pfn_to_page(start_pfn);
557 args.start = start;
558 args.end = start + size;
559 args.nid = nid;
560 args.zone = zone;
561
562 efi_memmap_walk(virtual_memmap_init, &args);
563 }
564}
565
566int
567ia64_pfn_valid (unsigned long pfn)
568{
569 char byte;
570 struct page *pg = pfn_to_page(pfn);
571
572 return (__get_user(byte, (char __user *) pg) == 0)
573 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
574 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
575}
576EXPORT_SYMBOL(ia64_pfn_valid);
577
578int __init find_largest_hole(u64 start, u64 end, void *arg)
579{
580 u64 *max_gap = arg;
581
582 static u64 last_end = PAGE_OFFSET;
583
584 /* NOTE: this algorithm assumes efi memmap table is ordered */
585
586 if (*max_gap < (start - last_end))
587 *max_gap = start - last_end;
588 last_end = end;
589 return 0;
590}
591
592#endif /* CONFIG_VIRTUAL_MEM_MAP */
593
594int __init register_active_ranges(u64 start, u64 len, int nid)
595{
596 u64 end = start + len;
597
598#ifdef CONFIG_KEXEC
599 if (start > crashk_res.start && start < crashk_res.end)
600 start = crashk_res.end;
601 if (end > crashk_res.start && end < crashk_res.end)
602 end = crashk_res.start;
603#endif
604
605 if (start < end)
606 memblock_add_node(__pa(start), end - start, nid);
607 return 0;
608}
609
610int
611find_max_min_low_pfn (u64 start, u64 end, void *arg)
612{
613 unsigned long pfn_start, pfn_end;
614#ifdef CONFIG_FLATMEM
615 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
616 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
617#else
618 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
619 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
620#endif
621 min_low_pfn = min(min_low_pfn, pfn_start);
622 max_low_pfn = max(max_low_pfn, pfn_end);
623 return 0;
624}
625
626/*
627 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
628 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
629 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
630 * useful for performance testing, but conceivably could also come in handy for debugging
631 * purposes.
632 */
633
634static int nolwsys __initdata;
635
636static int __init
637nolwsys_setup (char *s)
638{
639 nolwsys = 1;
640 return 1;
641}
642
643__setup("nolwsys", nolwsys_setup);
644
645void __init
646mem_init (void)
647{
648 int i;
649
650 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
651 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
652 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
653
654 /*
655 * This needs to be called _after_ the command line has been parsed but
656 * _before_ any drivers that may need the PCI DMA interface are
657 * initialized or bootmem has been freed.
658 */
659#ifdef CONFIG_INTEL_IOMMU
660 detect_intel_iommu();
661 if (!iommu_detected)
662#endif
663#ifdef CONFIG_SWIOTLB
664 swiotlb_init(1);
665#endif
666
667#ifdef CONFIG_FLATMEM
668 BUG_ON(!mem_map);
669#endif
670
671 set_max_mapnr(max_low_pfn);
672 high_memory = __va(max_low_pfn * PAGE_SIZE);
673 memblock_free_all();
674 mem_init_print_info(NULL);
675
676 /*
677 * For fsyscall entrpoints with no light-weight handler, use the ordinary
678 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
679 * code can tell them apart.
680 */
681 for (i = 0; i < NR_syscalls; ++i) {
682 extern unsigned long fsyscall_table[NR_syscalls];
683 extern unsigned long sys_call_table[NR_syscalls];
684
685 if (!fsyscall_table[i] || nolwsys)
686 fsyscall_table[i] = sys_call_table[i] | 1;
687 }
688 setup_gate();
689}
690
691#ifdef CONFIG_MEMORY_HOTPLUG
692int arch_add_memory(int nid, u64 start, u64 size,
693 struct mhp_params *params)
694{
695 unsigned long start_pfn = start >> PAGE_SHIFT;
696 unsigned long nr_pages = size >> PAGE_SHIFT;
697 int ret;
698
699 if (WARN_ON_ONCE(params->pgprot.pgprot != PAGE_KERNEL.pgprot))
700 return -EINVAL;
701
702 ret = __add_pages(nid, start_pfn, nr_pages, params);
703 if (ret)
704 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
705 __func__, ret);
706
707 return ret;
708}
709
710void arch_remove_memory(int nid, u64 start, u64 size,
711 struct vmem_altmap *altmap)
712{
713 unsigned long start_pfn = start >> PAGE_SHIFT;
714 unsigned long nr_pages = size >> PAGE_SHIFT;
715
716 __remove_pages(start_pfn, nr_pages, altmap);
717}
718#endif
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/mm.h>
14#include <linux/mmzone.h>
15#include <linux/module.h>
16#include <linux/personality.h>
17#include <linux/reboot.h>
18#include <linux/slab.h>
19#include <linux/swap.h>
20#include <linux/proc_fs.h>
21#include <linux/bitops.h>
22#include <linux/kexec.h>
23
24#include <asm/dma.h>
25#include <asm/io.h>
26#include <asm/machvec.h>
27#include <asm/numa.h>
28#include <asm/patch.h>
29#include <asm/pgalloc.h>
30#include <asm/sal.h>
31#include <asm/sections.h>
32#include <asm/system.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
39extern void ia64_tlb_init (void);
40
41unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
42
43#ifdef CONFIG_VIRTUAL_MEM_MAP
44unsigned long VMALLOC_END = VMALLOC_END_INIT;
45EXPORT_SYMBOL(VMALLOC_END);
46struct page *vmem_map;
47EXPORT_SYMBOL(vmem_map);
48#endif
49
50struct page *zero_page_memmap_ptr; /* map entry for zero page */
51EXPORT_SYMBOL(zero_page_memmap_ptr);
52
53void
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 */
74void
75dma_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
88inline void
89ia64_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 */
104void
105ia64_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(¤t->mm->mmap_sem);
125 if (insert_vm_struct(current->mm, vma)) {
126 up_write(¤t->mm->mmap_sem);
127 kmem_cache_free(vm_area_cachep, vma);
128 return;
129 }
130 up_write(¤t->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 | VM_RESERVED;
142 down_write(¤t->mm->mmap_sem);
143 if (insert_vm_struct(current->mm, vma)) {
144 up_write(¤t->mm->mmap_sem);
145 kmem_cache_free(vm_area_cachep, vma);
146 return;
147 }
148 up_write(¤t->mm->mmap_sem);
149 }
150 }
151}
152
153void
154free_initmem (void)
155{
156 unsigned long addr, eaddr;
157
158 addr = (unsigned long) ia64_imva(__init_begin);
159 eaddr = (unsigned long) ia64_imva(__init_end);
160 while (addr < eaddr) {
161 ClearPageReserved(virt_to_page(addr));
162 init_page_count(virt_to_page(addr));
163 free_page(addr);
164 ++totalram_pages;
165 addr += PAGE_SIZE;
166 }
167 printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n",
168 (__init_end - __init_begin) >> 10);
169}
170
171void __init
172free_initrd_mem (unsigned long start, unsigned long end)
173{
174 struct page *page;
175 /*
176 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
177 * Thus EFI and the kernel may have different page sizes. It is
178 * therefore possible to have the initrd share the same page as
179 * the end of the kernel (given current setup).
180 *
181 * To avoid freeing/using the wrong page (kernel sized) we:
182 * - align up the beginning of initrd
183 * - align down the end of initrd
184 *
185 * | |
186 * |=============| a000
187 * | |
188 * | |
189 * | | 9000
190 * |/////////////|
191 * |/////////////|
192 * |=============| 8000
193 * |///INITRD////|
194 * |/////////////|
195 * |/////////////| 7000
196 * | |
197 * |KKKKKKKKKKKKK|
198 * |=============| 6000
199 * |KKKKKKKKKKKKK|
200 * |KKKKKKKKKKKKK|
201 * K=kernel using 8KB pages
202 *
203 * In this example, we must free page 8000 ONLY. So we must align up
204 * initrd_start and keep initrd_end as is.
205 */
206 start = PAGE_ALIGN(start);
207 end = end & PAGE_MASK;
208
209 if (start < end)
210 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
211
212 for (; start < end; start += PAGE_SIZE) {
213 if (!virt_addr_valid(start))
214 continue;
215 page = virt_to_page(start);
216 ClearPageReserved(page);
217 init_page_count(page);
218 free_page(start);
219 ++totalram_pages;
220 }
221}
222
223/*
224 * This installs a clean page in the kernel's page table.
225 */
226static struct page * __init
227put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
228{
229 pgd_t *pgd;
230 pud_t *pud;
231 pmd_t *pmd;
232 pte_t *pte;
233
234 if (!PageReserved(page))
235 printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n",
236 page_address(page));
237
238 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
239
240 {
241 pud = pud_alloc(&init_mm, pgd, address);
242 if (!pud)
243 goto out;
244 pmd = pmd_alloc(&init_mm, pud, address);
245 if (!pmd)
246 goto out;
247 pte = pte_alloc_kernel(pmd, address);
248 if (!pte)
249 goto out;
250 if (!pte_none(*pte))
251 goto out;
252 set_pte(pte, mk_pte(page, pgprot));
253 }
254 out:
255 /* no need for flush_tlb */
256 return page;
257}
258
259static void __init
260setup_gate (void)
261{
262 void *gate_section;
263 struct page *page;
264
265 /*
266 * Map the gate page twice: once read-only to export the ELF
267 * headers etc. and once execute-only page to enable
268 * privilege-promotion via "epc":
269 */
270 gate_section = paravirt_get_gate_section();
271 page = virt_to_page(ia64_imva(gate_section));
272 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
273#ifdef HAVE_BUGGY_SEGREL
274 page = virt_to_page(ia64_imva(gate_section + PAGE_SIZE));
275 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
276#else
277 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
278 /* Fill in the holes (if any) with read-only zero pages: */
279 {
280 unsigned long addr;
281
282 for (addr = GATE_ADDR + PAGE_SIZE;
283 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
284 addr += PAGE_SIZE)
285 {
286 put_kernel_page(ZERO_PAGE(0), addr,
287 PAGE_READONLY);
288 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
289 PAGE_READONLY);
290 }
291 }
292#endif
293 ia64_patch_gate();
294}
295
296void __devinit
297ia64_mmu_init (void *my_cpu_data)
298{
299 unsigned long pta, impl_va_bits;
300 extern void __devinit tlb_init (void);
301
302#ifdef CONFIG_DISABLE_VHPT
303# define VHPT_ENABLE_BIT 0
304#else
305# define VHPT_ENABLE_BIT 1
306#endif
307
308 /*
309 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
310 * address space. The IA-64 architecture guarantees that at least 50 bits of
311 * virtual address space are implemented but if we pick a large enough page size
312 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
313 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
314 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
315 * problem in practice. Alternatively, we could truncate the top of the mapped
316 * address space to not permit mappings that would overlap with the VMLPT.
317 * --davidm 00/12/06
318 */
319# define pte_bits 3
320# define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
321 /*
322 * The virtual page table has to cover the entire implemented address space within
323 * a region even though not all of this space may be mappable. The reason for
324 * this is that the Access bit and Dirty bit fault handlers perform
325 * non-speculative accesses to the virtual page table, so the address range of the
326 * virtual page table itself needs to be covered by virtual page table.
327 */
328# define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
329# define POW2(n) (1ULL << (n))
330
331 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
332
333 if (impl_va_bits < 51 || impl_va_bits > 61)
334 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
335 /*
336 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
337 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
338 * the test makes sure that our mapped space doesn't overlap the
339 * unimplemented hole in the middle of the region.
340 */
341 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
342 (mapped_space_bits > impl_va_bits - 1))
343 panic("Cannot build a big enough virtual-linear page table"
344 " to cover mapped address space.\n"
345 " Try using a smaller page size.\n");
346
347
348 /* place the VMLPT at the end of each page-table mapped region: */
349 pta = POW2(61) - POW2(vmlpt_bits);
350
351 /*
352 * Set the (virtually mapped linear) page table address. Bit
353 * 8 selects between the short and long format, bits 2-7 the
354 * size of the table, and bit 0 whether the VHPT walker is
355 * enabled.
356 */
357 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
358
359 ia64_tlb_init();
360
361#ifdef CONFIG_HUGETLB_PAGE
362 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
363 ia64_srlz_d();
364#endif
365}
366
367#ifdef CONFIG_VIRTUAL_MEM_MAP
368int vmemmap_find_next_valid_pfn(int node, int i)
369{
370 unsigned long end_address, hole_next_pfn;
371 unsigned long stop_address;
372 pg_data_t *pgdat = NODE_DATA(node);
373
374 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
375 end_address = PAGE_ALIGN(end_address);
376
377 stop_address = (unsigned long) &vmem_map[
378 pgdat->node_start_pfn + pgdat->node_spanned_pages];
379
380 do {
381 pgd_t *pgd;
382 pud_t *pud;
383 pmd_t *pmd;
384 pte_t *pte;
385
386 pgd = pgd_offset_k(end_address);
387 if (pgd_none(*pgd)) {
388 end_address += PGDIR_SIZE;
389 continue;
390 }
391
392 pud = pud_offset(pgd, end_address);
393 if (pud_none(*pud)) {
394 end_address += PUD_SIZE;
395 continue;
396 }
397
398 pmd = pmd_offset(pud, end_address);
399 if (pmd_none(*pmd)) {
400 end_address += PMD_SIZE;
401 continue;
402 }
403
404 pte = pte_offset_kernel(pmd, end_address);
405retry_pte:
406 if (pte_none(*pte)) {
407 end_address += PAGE_SIZE;
408 pte++;
409 if ((end_address < stop_address) &&
410 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
411 goto retry_pte;
412 continue;
413 }
414 /* Found next valid vmem_map page */
415 break;
416 } while (end_address < stop_address);
417
418 end_address = min(end_address, stop_address);
419 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
420 hole_next_pfn = end_address / sizeof(struct page);
421 return hole_next_pfn - pgdat->node_start_pfn;
422}
423
424int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
425{
426 unsigned long address, start_page, end_page;
427 struct page *map_start, *map_end;
428 int node;
429 pgd_t *pgd;
430 pud_t *pud;
431 pmd_t *pmd;
432 pte_t *pte;
433
434 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
435 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
436
437 start_page = (unsigned long) map_start & PAGE_MASK;
438 end_page = PAGE_ALIGN((unsigned long) map_end);
439 node = paddr_to_nid(__pa(start));
440
441 for (address = start_page; address < end_page; address += PAGE_SIZE) {
442 pgd = pgd_offset_k(address);
443 if (pgd_none(*pgd))
444 pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
445 pud = pud_offset(pgd, address);
446
447 if (pud_none(*pud))
448 pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
449 pmd = pmd_offset(pud, address);
450
451 if (pmd_none(*pmd))
452 pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
453 pte = pte_offset_kernel(pmd, address);
454
455 if (pte_none(*pte))
456 set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
457 PAGE_KERNEL));
458 }
459 return 0;
460}
461
462struct memmap_init_callback_data {
463 struct page *start;
464 struct page *end;
465 int nid;
466 unsigned long zone;
467};
468
469static int __meminit
470virtual_memmap_init(u64 start, u64 end, void *arg)
471{
472 struct memmap_init_callback_data *args;
473 struct page *map_start, *map_end;
474
475 args = (struct memmap_init_callback_data *) arg;
476 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
477 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
478
479 if (map_start < args->start)
480 map_start = args->start;
481 if (map_end > args->end)
482 map_end = args->end;
483
484 /*
485 * We have to initialize "out of bounds" struct page elements that fit completely
486 * on the same pages that were allocated for the "in bounds" elements because they
487 * may be referenced later (and found to be "reserved").
488 */
489 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
490 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
491 / sizeof(struct page));
492
493 if (map_start < map_end)
494 memmap_init_zone((unsigned long)(map_end - map_start),
495 args->nid, args->zone, page_to_pfn(map_start),
496 MEMMAP_EARLY);
497 return 0;
498}
499
500void __meminit
501memmap_init (unsigned long size, int nid, unsigned long zone,
502 unsigned long start_pfn)
503{
504 if (!vmem_map)
505 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
506 else {
507 struct page *start;
508 struct memmap_init_callback_data args;
509
510 start = pfn_to_page(start_pfn);
511 args.start = start;
512 args.end = start + size;
513 args.nid = nid;
514 args.zone = zone;
515
516 efi_memmap_walk(virtual_memmap_init, &args);
517 }
518}
519
520int
521ia64_pfn_valid (unsigned long pfn)
522{
523 char byte;
524 struct page *pg = pfn_to_page(pfn);
525
526 return (__get_user(byte, (char __user *) pg) == 0)
527 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
528 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
529}
530EXPORT_SYMBOL(ia64_pfn_valid);
531
532int __init find_largest_hole(u64 start, u64 end, void *arg)
533{
534 u64 *max_gap = arg;
535
536 static u64 last_end = PAGE_OFFSET;
537
538 /* NOTE: this algorithm assumes efi memmap table is ordered */
539
540 if (*max_gap < (start - last_end))
541 *max_gap = start - last_end;
542 last_end = end;
543 return 0;
544}
545
546#endif /* CONFIG_VIRTUAL_MEM_MAP */
547
548int __init register_active_ranges(u64 start, u64 len, int nid)
549{
550 u64 end = start + len;
551
552#ifdef CONFIG_KEXEC
553 if (start > crashk_res.start && start < crashk_res.end)
554 start = crashk_res.end;
555 if (end > crashk_res.start && end < crashk_res.end)
556 end = crashk_res.start;
557#endif
558
559 if (start < end)
560 add_active_range(nid, __pa(start) >> PAGE_SHIFT,
561 __pa(end) >> PAGE_SHIFT);
562 return 0;
563}
564
565static int __init
566count_reserved_pages(u64 start, u64 end, void *arg)
567{
568 unsigned long num_reserved = 0;
569 unsigned long *count = arg;
570
571 for (; start < end; start += PAGE_SIZE)
572 if (PageReserved(virt_to_page(start)))
573 ++num_reserved;
574 *count += num_reserved;
575 return 0;
576}
577
578int
579find_max_min_low_pfn (u64 start, u64 end, void *arg)
580{
581 unsigned long pfn_start, pfn_end;
582#ifdef CONFIG_FLATMEM
583 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
584 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
585#else
586 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
587 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
588#endif
589 min_low_pfn = min(min_low_pfn, pfn_start);
590 max_low_pfn = max(max_low_pfn, pfn_end);
591 return 0;
592}
593
594/*
595 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
596 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
597 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
598 * useful for performance testing, but conceivably could also come in handy for debugging
599 * purposes.
600 */
601
602static int nolwsys __initdata;
603
604static int __init
605nolwsys_setup (char *s)
606{
607 nolwsys = 1;
608 return 1;
609}
610
611__setup("nolwsys", nolwsys_setup);
612
613void __init
614mem_init (void)
615{
616 long reserved_pages, codesize, datasize, initsize;
617 pg_data_t *pgdat;
618 int i;
619
620 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
621 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
622 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
623
624#ifdef CONFIG_PCI
625 /*
626 * This needs to be called _after_ the command line has been parsed but _before_
627 * any drivers that may need the PCI DMA interface are initialized or bootmem has
628 * been freed.
629 */
630 platform_dma_init();
631#endif
632
633#ifdef CONFIG_FLATMEM
634 BUG_ON(!mem_map);
635 max_mapnr = max_low_pfn;
636#endif
637
638 high_memory = __va(max_low_pfn * PAGE_SIZE);
639
640 for_each_online_pgdat(pgdat)
641 if (pgdat->bdata->node_bootmem_map)
642 totalram_pages += free_all_bootmem_node(pgdat);
643
644 reserved_pages = 0;
645 efi_memmap_walk(count_reserved_pages, &reserved_pages);
646
647 codesize = (unsigned long) _etext - (unsigned long) _stext;
648 datasize = (unsigned long) _edata - (unsigned long) _etext;
649 initsize = (unsigned long) __init_end - (unsigned long) __init_begin;
650
651 printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, "
652 "%luk data, %luk init)\n", nr_free_pages() << (PAGE_SHIFT - 10),
653 num_physpages << (PAGE_SHIFT - 10), codesize >> 10,
654 reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10);
655
656
657 /*
658 * For fsyscall entrpoints with no light-weight handler, use the ordinary
659 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
660 * code can tell them apart.
661 */
662 for (i = 0; i < NR_syscalls; ++i) {
663 extern unsigned long sys_call_table[NR_syscalls];
664 unsigned long *fsyscall_table = paravirt_get_fsyscall_table();
665
666 if (!fsyscall_table[i] || nolwsys)
667 fsyscall_table[i] = sys_call_table[i] | 1;
668 }
669 setup_gate();
670}
671
672#ifdef CONFIG_MEMORY_HOTPLUG
673int arch_add_memory(int nid, u64 start, u64 size)
674{
675 pg_data_t *pgdat;
676 struct zone *zone;
677 unsigned long start_pfn = start >> PAGE_SHIFT;
678 unsigned long nr_pages = size >> PAGE_SHIFT;
679 int ret;
680
681 pgdat = NODE_DATA(nid);
682
683 zone = pgdat->node_zones + ZONE_NORMAL;
684 ret = __add_pages(nid, zone, start_pfn, nr_pages);
685
686 if (ret)
687 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
688 __func__, ret);
689
690 return ret;
691}
692#endif
693
694/*
695 * Even when CONFIG_IA32_SUPPORT is not enabled it is
696 * useful to have the Linux/x86 domain registered to
697 * avoid an attempted module load when emulators call
698 * personality(PER_LINUX32). This saves several milliseconds
699 * on each such call.
700 */
701static struct exec_domain ia32_exec_domain;
702
703static int __init
704per_linux32_init(void)
705{
706 ia32_exec_domain.name = "Linux/x86";
707 ia32_exec_domain.handler = NULL;
708 ia32_exec_domain.pers_low = PER_LINUX32;
709 ia32_exec_domain.pers_high = PER_LINUX32;
710 ia32_exec_domain.signal_map = default_exec_domain.signal_map;
711 ia32_exec_domain.signal_invmap = default_exec_domain.signal_invmap;
712 register_exec_domain(&ia32_exec_domain);
713
714 return 0;
715}
716
717__initcall(per_linux32_init);