Commit | Line | Data |
---|---|---|
1da177e4 LT |
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/config.h> | |
8 | #include <linux/kernel.h> | |
9 | #include <linux/init.h> | |
10 | ||
11 | #include <linux/bootmem.h> | |
12 | #include <linux/efi.h> | |
13 | #include <linux/elf.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 | ||
24 | #include <asm/a.out.h> | |
25 | #include <asm/dma.h> | |
26 | #include <asm/ia32.h> | |
27 | #include <asm/io.h> | |
28 | #include <asm/machvec.h> | |
29 | #include <asm/numa.h> | |
30 | #include <asm/patch.h> | |
31 | #include <asm/pgalloc.h> | |
32 | #include <asm/sal.h> | |
33 | #include <asm/sections.h> | |
34 | #include <asm/system.h> | |
35 | #include <asm/tlb.h> | |
36 | #include <asm/uaccess.h> | |
37 | #include <asm/unistd.h> | |
38 | #include <asm/mca.h> | |
39 | ||
40 | DEFINE_PER_CPU(struct mmu_gather, mmu_gathers); | |
41 | ||
fde740e4 RH |
42 | DEFINE_PER_CPU(unsigned long *, __pgtable_quicklist); |
43 | DEFINE_PER_CPU(long, __pgtable_quicklist_size); | |
44 | ||
1da177e4 LT |
45 | extern void ia64_tlb_init (void); |
46 | ||
47 | unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL; | |
48 | ||
49 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
50 | unsigned long vmalloc_end = VMALLOC_END_INIT; | |
51 | EXPORT_SYMBOL(vmalloc_end); | |
52 | struct page *vmem_map; | |
53 | EXPORT_SYMBOL(vmem_map); | |
54 | #endif | |
55 | ||
fde740e4 | 56 | struct page *zero_page_memmap_ptr; /* map entry for zero page */ |
1da177e4 LT |
57 | EXPORT_SYMBOL(zero_page_memmap_ptr); |
58 | ||
fde740e4 | 59 | #define MIN_PGT_PAGES 25UL |
e96c9b47 | 60 | #define MAX_PGT_FREES_PER_PASS 16L |
fde740e4 RH |
61 | #define PGT_FRACTION_OF_NODE_MEM 16 |
62 | ||
63 | static inline long | |
64 | max_pgt_pages(void) | |
65 | { | |
66 | u64 node_free_pages, max_pgt_pages; | |
67 | ||
68 | #ifndef CONFIG_NUMA | |
69 | node_free_pages = nr_free_pages(); | |
70 | #else | |
71 | node_free_pages = nr_free_pages_pgdat(NODE_DATA(numa_node_id())); | |
72 | #endif | |
73 | max_pgt_pages = node_free_pages / PGT_FRACTION_OF_NODE_MEM; | |
74 | max_pgt_pages = max(max_pgt_pages, MIN_PGT_PAGES); | |
75 | return max_pgt_pages; | |
76 | } | |
77 | ||
78 | static inline long | |
79 | min_pages_to_free(void) | |
80 | { | |
81 | long pages_to_free; | |
82 | ||
83 | pages_to_free = pgtable_quicklist_size - max_pgt_pages(); | |
84 | pages_to_free = min(pages_to_free, MAX_PGT_FREES_PER_PASS); | |
85 | return pages_to_free; | |
86 | } | |
87 | ||
1da177e4 | 88 | void |
fde740e4 | 89 | check_pgt_cache(void) |
1da177e4 | 90 | { |
fde740e4 | 91 | long pages_to_free; |
1da177e4 | 92 | |
fde740e4 RH |
93 | if (unlikely(pgtable_quicklist_size <= MIN_PGT_PAGES)) |
94 | return; | |
1da177e4 LT |
95 | |
96 | preempt_disable(); | |
fde740e4 RH |
97 | while (unlikely((pages_to_free = min_pages_to_free()) > 0)) { |
98 | while (pages_to_free--) { | |
99 | free_page((unsigned long)pgtable_quicklist_alloc()); | |
100 | } | |
101 | preempt_enable(); | |
102 | preempt_disable(); | |
1da177e4 LT |
103 | } |
104 | preempt_enable(); | |
105 | } | |
106 | ||
107 | void | |
108 | lazy_mmu_prot_update (pte_t pte) | |
109 | { | |
110 | unsigned long addr; | |
111 | struct page *page; | |
112 | ||
113 | if (!pte_exec(pte)) | |
114 | return; /* not an executable page... */ | |
115 | ||
116 | page = pte_page(pte); | |
117 | addr = (unsigned long) page_address(page); | |
118 | ||
119 | if (test_bit(PG_arch_1, &page->flags)) | |
120 | return; /* i-cache is already coherent with d-cache */ | |
121 | ||
122 | flush_icache_range(addr, addr + PAGE_SIZE); | |
123 | set_bit(PG_arch_1, &page->flags); /* mark page as clean */ | |
124 | } | |
125 | ||
126 | inline void | |
127 | ia64_set_rbs_bot (void) | |
128 | { | |
129 | unsigned long stack_size = current->signal->rlim[RLIMIT_STACK].rlim_max & -16; | |
130 | ||
131 | if (stack_size > MAX_USER_STACK_SIZE) | |
132 | stack_size = MAX_USER_STACK_SIZE; | |
133 | current->thread.rbs_bot = STACK_TOP - stack_size; | |
134 | } | |
135 | ||
136 | /* | |
137 | * This performs some platform-dependent address space initialization. | |
138 | * On IA-64, we want to setup the VM area for the register backing | |
139 | * store (which grows upwards) and install the gateway page which is | |
140 | * used for signal trampolines, etc. | |
141 | */ | |
142 | void | |
143 | ia64_init_addr_space (void) | |
144 | { | |
145 | struct vm_area_struct *vma; | |
146 | ||
147 | ia64_set_rbs_bot(); | |
148 | ||
149 | /* | |
150 | * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore | |
151 | * the problem. When the process attempts to write to the register backing store | |
152 | * for the first time, it will get a SEGFAULT in this case. | |
153 | */ | |
154 | vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
155 | if (vma) { | |
156 | memset(vma, 0, sizeof(*vma)); | |
157 | vma->vm_mm = current->mm; | |
158 | vma->vm_start = current->thread.rbs_bot & PAGE_MASK; | |
159 | vma->vm_end = vma->vm_start + PAGE_SIZE; | |
160 | vma->vm_page_prot = protection_map[VM_DATA_DEFAULT_FLAGS & 0x7]; | |
161 | vma->vm_flags = VM_DATA_DEFAULT_FLAGS | VM_GROWSUP; | |
162 | down_write(¤t->mm->mmap_sem); | |
163 | if (insert_vm_struct(current->mm, vma)) { | |
164 | up_write(¤t->mm->mmap_sem); | |
165 | kmem_cache_free(vm_area_cachep, vma); | |
166 | return; | |
167 | } | |
168 | up_write(¤t->mm->mmap_sem); | |
169 | } | |
170 | ||
171 | /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */ | |
172 | if (!(current->personality & MMAP_PAGE_ZERO)) { | |
173 | vma = kmem_cache_alloc(vm_area_cachep, SLAB_KERNEL); | |
174 | if (vma) { | |
175 | memset(vma, 0, sizeof(*vma)); | |
176 | vma->vm_mm = current->mm; | |
177 | vma->vm_end = PAGE_SIZE; | |
178 | vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT); | |
179 | vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | VM_RESERVED; | |
180 | down_write(¤t->mm->mmap_sem); | |
181 | if (insert_vm_struct(current->mm, vma)) { | |
182 | up_write(¤t->mm->mmap_sem); | |
183 | kmem_cache_free(vm_area_cachep, vma); | |
184 | return; | |
185 | } | |
186 | up_write(¤t->mm->mmap_sem); | |
187 | } | |
188 | } | |
189 | } | |
190 | ||
191 | void | |
192 | free_initmem (void) | |
193 | { | |
194 | unsigned long addr, eaddr; | |
195 | ||
196 | addr = (unsigned long) ia64_imva(__init_begin); | |
197 | eaddr = (unsigned long) ia64_imva(__init_end); | |
198 | while (addr < eaddr) { | |
199 | ClearPageReserved(virt_to_page(addr)); | |
200 | set_page_count(virt_to_page(addr), 1); | |
201 | free_page(addr); | |
202 | ++totalram_pages; | |
203 | addr += PAGE_SIZE; | |
204 | } | |
205 | printk(KERN_INFO "Freeing unused kernel memory: %ldkB freed\n", | |
206 | (__init_end - __init_begin) >> 10); | |
207 | } | |
208 | ||
209 | void | |
210 | free_initrd_mem (unsigned long start, unsigned long end) | |
211 | { | |
212 | struct page *page; | |
213 | /* | |
214 | * EFI uses 4KB pages while the kernel can use 4KB or bigger. | |
215 | * Thus EFI and the kernel may have different page sizes. It is | |
216 | * therefore possible to have the initrd share the same page as | |
217 | * the end of the kernel (given current setup). | |
218 | * | |
219 | * To avoid freeing/using the wrong page (kernel sized) we: | |
220 | * - align up the beginning of initrd | |
221 | * - align down the end of initrd | |
222 | * | |
223 | * | | | |
224 | * |=============| a000 | |
225 | * | | | |
226 | * | | | |
227 | * | | 9000 | |
228 | * |/////////////| | |
229 | * |/////////////| | |
230 | * |=============| 8000 | |
231 | * |///INITRD////| | |
232 | * |/////////////| | |
233 | * |/////////////| 7000 | |
234 | * | | | |
235 | * |KKKKKKKKKKKKK| | |
236 | * |=============| 6000 | |
237 | * |KKKKKKKKKKKKK| | |
238 | * |KKKKKKKKKKKKK| | |
239 | * K=kernel using 8KB pages | |
240 | * | |
241 | * In this example, we must free page 8000 ONLY. So we must align up | |
242 | * initrd_start and keep initrd_end as is. | |
243 | */ | |
244 | start = PAGE_ALIGN(start); | |
245 | end = end & PAGE_MASK; | |
246 | ||
247 | if (start < end) | |
248 | printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10); | |
249 | ||
250 | for (; start < end; start += PAGE_SIZE) { | |
251 | if (!virt_addr_valid(start)) | |
252 | continue; | |
253 | page = virt_to_page(start); | |
254 | ClearPageReserved(page); | |
255 | set_page_count(page, 1); | |
256 | free_page(start); | |
257 | ++totalram_pages; | |
258 | } | |
259 | } | |
260 | ||
261 | /* | |
262 | * This installs a clean page in the kernel's page table. | |
263 | */ | |
264 | struct page * | |
265 | put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot) | |
266 | { | |
267 | pgd_t *pgd; | |
268 | pud_t *pud; | |
269 | pmd_t *pmd; | |
270 | pte_t *pte; | |
271 | ||
272 | if (!PageReserved(page)) | |
273 | printk(KERN_ERR "put_kernel_page: page at 0x%p not in reserved memory\n", | |
274 | page_address(page)); | |
275 | ||
276 | pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */ | |
277 | ||
278 | spin_lock(&init_mm.page_table_lock); | |
279 | { | |
280 | pud = pud_alloc(&init_mm, pgd, address); | |
281 | if (!pud) | |
282 | goto out; | |
283 | ||
284 | pmd = pmd_alloc(&init_mm, pud, address); | |
285 | if (!pmd) | |
286 | goto out; | |
287 | pte = pte_alloc_map(&init_mm, pmd, address); | |
288 | if (!pte) | |
289 | goto out; | |
290 | if (!pte_none(*pte)) { | |
291 | pte_unmap(pte); | |
292 | goto out; | |
293 | } | |
294 | set_pte(pte, mk_pte(page, pgprot)); | |
295 | pte_unmap(pte); | |
296 | } | |
297 | out: spin_unlock(&init_mm.page_table_lock); | |
298 | /* no need for flush_tlb */ | |
299 | return page; | |
300 | } | |
301 | ||
302 | static void | |
303 | setup_gate (void) | |
304 | { | |
305 | struct page *page; | |
306 | ||
307 | /* | |
ad597bd5 DMT |
308 | * Map the gate page twice: once read-only to export the ELF |
309 | * headers etc. and once execute-only page to enable | |
310 | * privilege-promotion via "epc": | |
1da177e4 LT |
311 | */ |
312 | page = virt_to_page(ia64_imva(__start_gate_section)); | |
313 | put_kernel_page(page, GATE_ADDR, PAGE_READONLY); | |
314 | #ifdef HAVE_BUGGY_SEGREL | |
315 | page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE)); | |
316 | put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE); | |
317 | #else | |
318 | put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE); | |
ad597bd5 DMT |
319 | /* Fill in the holes (if any) with read-only zero pages: */ |
320 | { | |
321 | unsigned long addr; | |
322 | ||
323 | for (addr = GATE_ADDR + PAGE_SIZE; | |
324 | addr < GATE_ADDR + PERCPU_PAGE_SIZE; | |
325 | addr += PAGE_SIZE) | |
326 | { | |
327 | put_kernel_page(ZERO_PAGE(0), addr, | |
328 | PAGE_READONLY); | |
329 | put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE, | |
330 | PAGE_READONLY); | |
331 | } | |
332 | } | |
1da177e4 LT |
333 | #endif |
334 | ia64_patch_gate(); | |
335 | } | |
336 | ||
337 | void __devinit | |
338 | ia64_mmu_init (void *my_cpu_data) | |
339 | { | |
340 | unsigned long psr, pta, impl_va_bits; | |
341 | extern void __devinit tlb_init (void); | |
342 | ||
343 | #ifdef CONFIG_DISABLE_VHPT | |
344 | # define VHPT_ENABLE_BIT 0 | |
345 | #else | |
346 | # define VHPT_ENABLE_BIT 1 | |
347 | #endif | |
348 | ||
349 | /* Pin mapping for percpu area into TLB */ | |
350 | psr = ia64_clear_ic(); | |
351 | ia64_itr(0x2, IA64_TR_PERCPU_DATA, PERCPU_ADDR, | |
352 | pte_val(pfn_pte(__pa(my_cpu_data) >> PAGE_SHIFT, PAGE_KERNEL)), | |
353 | PERCPU_PAGE_SHIFT); | |
354 | ||
355 | ia64_set_psr(psr); | |
356 | ia64_srlz_i(); | |
357 | ||
358 | /* | |
359 | * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped | |
360 | * address space. The IA-64 architecture guarantees that at least 50 bits of | |
361 | * virtual address space are implemented but if we pick a large enough page size | |
362 | * (e.g., 64KB), the mapped address space is big enough that it will overlap with | |
363 | * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages, | |
364 | * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a | |
365 | * problem in practice. Alternatively, we could truncate the top of the mapped | |
366 | * address space to not permit mappings that would overlap with the VMLPT. | |
367 | * --davidm 00/12/06 | |
368 | */ | |
369 | # define pte_bits 3 | |
370 | # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT) | |
371 | /* | |
372 | * The virtual page table has to cover the entire implemented address space within | |
373 | * a region even though not all of this space may be mappable. The reason for | |
374 | * this is that the Access bit and Dirty bit fault handlers perform | |
375 | * non-speculative accesses to the virtual page table, so the address range of the | |
376 | * virtual page table itself needs to be covered by virtual page table. | |
377 | */ | |
378 | # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits) | |
379 | # define POW2(n) (1ULL << (n)) | |
380 | ||
381 | impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61))); | |
382 | ||
383 | if (impl_va_bits < 51 || impl_va_bits > 61) | |
384 | panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1); | |
385 | ||
386 | /* place the VMLPT at the end of each page-table mapped region: */ | |
387 | pta = POW2(61) - POW2(vmlpt_bits); | |
388 | ||
389 | if (POW2(mapped_space_bits) >= pta) | |
390 | panic("mm/init: overlap between virtually mapped linear page table and " | |
391 | "mapped kernel space!"); | |
392 | /* | |
393 | * Set the (virtually mapped linear) page table address. Bit | |
394 | * 8 selects between the short and long format, bits 2-7 the | |
395 | * size of the table, and bit 0 whether the VHPT walker is | |
396 | * enabled. | |
397 | */ | |
398 | ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT); | |
399 | ||
400 | ia64_tlb_init(); | |
401 | ||
402 | #ifdef CONFIG_HUGETLB_PAGE | |
403 | ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2); | |
404 | ia64_srlz_d(); | |
405 | #endif | |
406 | } | |
407 | ||
408 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
409 | ||
410 | int | |
411 | create_mem_map_page_table (u64 start, u64 end, void *arg) | |
412 | { | |
413 | unsigned long address, start_page, end_page; | |
414 | struct page *map_start, *map_end; | |
415 | int node; | |
416 | pgd_t *pgd; | |
417 | pud_t *pud; | |
418 | pmd_t *pmd; | |
419 | pte_t *pte; | |
420 | ||
421 | map_start = vmem_map + (__pa(start) >> PAGE_SHIFT); | |
422 | map_end = vmem_map + (__pa(end) >> PAGE_SHIFT); | |
423 | ||
424 | start_page = (unsigned long) map_start & PAGE_MASK; | |
425 | end_page = PAGE_ALIGN((unsigned long) map_end); | |
426 | node = paddr_to_nid(__pa(start)); | |
427 | ||
428 | for (address = start_page; address < end_page; address += PAGE_SIZE) { | |
429 | pgd = pgd_offset_k(address); | |
430 | if (pgd_none(*pgd)) | |
431 | pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)); | |
432 | pud = pud_offset(pgd, address); | |
433 | ||
434 | if (pud_none(*pud)) | |
435 | pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)); | |
436 | pmd = pmd_offset(pud, address); | |
437 | ||
438 | if (pmd_none(*pmd)) | |
439 | pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)); | |
440 | pte = pte_offset_kernel(pmd, address); | |
441 | ||
442 | if (pte_none(*pte)) | |
443 | set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT, | |
444 | PAGE_KERNEL)); | |
445 | } | |
446 | return 0; | |
447 | } | |
448 | ||
449 | struct memmap_init_callback_data { | |
450 | struct page *start; | |
451 | struct page *end; | |
452 | int nid; | |
453 | unsigned long zone; | |
454 | }; | |
455 | ||
456 | static int | |
457 | virtual_memmap_init (u64 start, u64 end, void *arg) | |
458 | { | |
459 | struct memmap_init_callback_data *args; | |
460 | struct page *map_start, *map_end; | |
461 | ||
462 | args = (struct memmap_init_callback_data *) arg; | |
463 | map_start = vmem_map + (__pa(start) >> PAGE_SHIFT); | |
464 | map_end = vmem_map + (__pa(end) >> PAGE_SHIFT); | |
465 | ||
466 | if (map_start < args->start) | |
467 | map_start = args->start; | |
468 | if (map_end > args->end) | |
469 | map_end = args->end; | |
470 | ||
471 | /* | |
472 | * We have to initialize "out of bounds" struct page elements that fit completely | |
473 | * on the same pages that were allocated for the "in bounds" elements because they | |
474 | * may be referenced later (and found to be "reserved"). | |
475 | */ | |
476 | map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page); | |
477 | map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end) | |
478 | / sizeof(struct page)); | |
479 | ||
480 | if (map_start < map_end) | |
481 | memmap_init_zone((unsigned long)(map_end - map_start), | |
482 | args->nid, args->zone, page_to_pfn(map_start)); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | void | |
487 | memmap_init (unsigned long size, int nid, unsigned long zone, | |
488 | unsigned long start_pfn) | |
489 | { | |
490 | if (!vmem_map) | |
491 | memmap_init_zone(size, nid, zone, start_pfn); | |
492 | else { | |
493 | struct page *start; | |
494 | struct memmap_init_callback_data args; | |
495 | ||
496 | start = pfn_to_page(start_pfn); | |
497 | args.start = start; | |
498 | args.end = start + size; | |
499 | args.nid = nid; | |
500 | args.zone = zone; | |
501 | ||
502 | efi_memmap_walk(virtual_memmap_init, &args); | |
503 | } | |
504 | } | |
505 | ||
506 | int | |
507 | ia64_pfn_valid (unsigned long pfn) | |
508 | { | |
509 | char byte; | |
510 | struct page *pg = pfn_to_page(pfn); | |
511 | ||
512 | return (__get_user(byte, (char __user *) pg) == 0) | |
513 | && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK)) | |
514 | || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0)); | |
515 | } | |
516 | EXPORT_SYMBOL(ia64_pfn_valid); | |
517 | ||
518 | int | |
519 | find_largest_hole (u64 start, u64 end, void *arg) | |
520 | { | |
521 | u64 *max_gap = arg; | |
522 | ||
523 | static u64 last_end = PAGE_OFFSET; | |
524 | ||
525 | /* NOTE: this algorithm assumes efi memmap table is ordered */ | |
526 | ||
527 | if (*max_gap < (start - last_end)) | |
528 | *max_gap = start - last_end; | |
529 | last_end = end; | |
530 | return 0; | |
531 | } | |
532 | #endif /* CONFIG_VIRTUAL_MEM_MAP */ | |
533 | ||
534 | static int | |
535 | count_reserved_pages (u64 start, u64 end, void *arg) | |
536 | { | |
537 | unsigned long num_reserved = 0; | |
538 | unsigned long *count = arg; | |
539 | ||
540 | for (; start < end; start += PAGE_SIZE) | |
541 | if (PageReserved(virt_to_page(start))) | |
542 | ++num_reserved; | |
543 | *count += num_reserved; | |
544 | return 0; | |
545 | } | |
546 | ||
547 | /* | |
548 | * Boot command-line option "nolwsys" can be used to disable the use of any light-weight | |
549 | * system call handler. When this option is in effect, all fsyscalls will end up bubbling | |
550 | * down into the kernel and calling the normal (heavy-weight) syscall handler. This is | |
551 | * useful for performance testing, but conceivably could also come in handy for debugging | |
552 | * purposes. | |
553 | */ | |
554 | ||
555 | static int nolwsys; | |
556 | ||
557 | static int __init | |
558 | nolwsys_setup (char *s) | |
559 | { | |
560 | nolwsys = 1; | |
561 | return 1; | |
562 | } | |
563 | ||
564 | __setup("nolwsys", nolwsys_setup); | |
565 | ||
566 | void | |
567 | mem_init (void) | |
568 | { | |
569 | long reserved_pages, codesize, datasize, initsize; | |
1da177e4 LT |
570 | pg_data_t *pgdat; |
571 | int i; | |
572 | static struct kcore_list kcore_mem, kcore_vmem, kcore_kernel; | |
573 | ||
fde740e4 RH |
574 | BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE); |
575 | BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE); | |
576 | BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE); | |
577 | ||
1da177e4 LT |
578 | #ifdef CONFIG_PCI |
579 | /* | |
580 | * This needs to be called _after_ the command line has been parsed but _before_ | |
581 | * any drivers that may need the PCI DMA interface are initialized or bootmem has | |
582 | * been freed. | |
583 | */ | |
584 | platform_dma_init(); | |
585 | #endif | |
586 | ||
587 | #ifndef CONFIG_DISCONTIGMEM | |
588 | if (!mem_map) | |
589 | BUG(); | |
590 | max_mapnr = max_low_pfn; | |
591 | #endif | |
592 | ||
593 | high_memory = __va(max_low_pfn * PAGE_SIZE); | |
594 | ||
595 | kclist_add(&kcore_mem, __va(0), max_low_pfn * PAGE_SIZE); | |
596 | kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START); | |
597 | kclist_add(&kcore_kernel, _stext, _end - _stext); | |
598 | ||
599 | for_each_pgdat(pgdat) | |
600 | totalram_pages += free_all_bootmem_node(pgdat); | |
601 | ||
602 | reserved_pages = 0; | |
603 | efi_memmap_walk(count_reserved_pages, &reserved_pages); | |
604 | ||
605 | codesize = (unsigned long) _etext - (unsigned long) _stext; | |
606 | datasize = (unsigned long) _edata - (unsigned long) _etext; | |
607 | initsize = (unsigned long) __init_end - (unsigned long) __init_begin; | |
608 | ||
609 | printk(KERN_INFO "Memory: %luk/%luk available (%luk code, %luk reserved, " | |
610 | "%luk data, %luk init)\n", (unsigned long) nr_free_pages() << (PAGE_SHIFT - 10), | |
611 | num_physpages << (PAGE_SHIFT - 10), codesize >> 10, | |
612 | reserved_pages << (PAGE_SHIFT - 10), datasize >> 10, initsize >> 10); | |
613 | ||
1da177e4 LT |
614 | |
615 | /* | |
616 | * For fsyscall entrpoints with no light-weight handler, use the ordinary | |
617 | * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry | |
618 | * code can tell them apart. | |
619 | */ | |
620 | for (i = 0; i < NR_syscalls; ++i) { | |
621 | extern unsigned long fsyscall_table[NR_syscalls]; | |
622 | extern unsigned long sys_call_table[NR_syscalls]; | |
623 | ||
624 | if (!fsyscall_table[i] || nolwsys) | |
625 | fsyscall_table[i] = sys_call_table[i] | 1; | |
626 | } | |
627 | setup_gate(); | |
628 | ||
629 | #ifdef CONFIG_IA32_SUPPORT | |
630 | ia32_mem_init(); | |
631 | #endif | |
632 | } |