]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * Initialize MMU support. | |
4 | * | |
5 | * Copyright (C) 1998-2003 Hewlett-Packard Co | |
6 | * David Mosberger-Tang <davidm@hpl.hp.com> | |
7 | */ | |
1da177e4 LT |
8 | #include <linux/kernel.h> |
9 | #include <linux/init.h> | |
10 | ||
68c60834 | 11 | #include <linux/dma-noncoherent.h> |
1da177e4 LT |
12 | #include <linux/efi.h> |
13 | #include <linux/elf.h> | |
98e4ae8a | 14 | #include <linux/memblock.h> |
1da177e4 | 15 | #include <linux/mm.h> |
3f07c014 | 16 | #include <linux/sched/signal.h> |
1da177e4 LT |
17 | #include <linux/mmzone.h> |
18 | #include <linux/module.h> | |
19 | #include <linux/personality.h> | |
20 | #include <linux/reboot.h> | |
21 | #include <linux/slab.h> | |
22 | #include <linux/swap.h> | |
23 | #include <linux/proc_fs.h> | |
24 | #include <linux/bitops.h> | |
139b8304 | 25 | #include <linux/kexec.h> |
1da177e4 | 26 | |
1da177e4 | 27 | #include <asm/dma.h> |
1da177e4 LT |
28 | #include <asm/io.h> |
29 | #include <asm/machvec.h> | |
30 | #include <asm/numa.h> | |
31 | #include <asm/patch.h> | |
32 | #include <asm/pgalloc.h> | |
33 | #include <asm/sal.h> | |
34 | #include <asm/sections.h> | |
1da177e4 | 35 | #include <asm/tlb.h> |
7c0f6ba6 | 36 | #include <linux/uaccess.h> |
1da177e4 LT |
37 | #include <asm/unistd.h> |
38 | #include <asm/mca.h> | |
39 | ||
1da177e4 LT |
40 | extern void ia64_tlb_init (void); |
41 | ||
42 | unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL; | |
43 | ||
44 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
126b3fcd TH |
45 | unsigned long VMALLOC_END = VMALLOC_END_INIT; |
46 | EXPORT_SYMBOL(VMALLOC_END); | |
1da177e4 LT |
47 | struct page *vmem_map; |
48 | EXPORT_SYMBOL(vmem_map); | |
49 | #endif | |
50 | ||
fde740e4 | 51 | struct page *zero_page_memmap_ptr; /* map entry for zero page */ |
1da177e4 LT |
52 | EXPORT_SYMBOL(zero_page_memmap_ptr); |
53 | ||
1da177e4 | 54 | void |
954ffcb3 | 55 | __ia64_sync_icache_dcache (pte_t pte) |
1da177e4 LT |
56 | { |
57 | unsigned long addr; | |
58 | struct page *page; | |
59 | ||
1da177e4 LT |
60 | page = pte_page(pte); |
61 | addr = (unsigned long) page_address(page); | |
62 | ||
63 | if (test_bit(PG_arch_1, &page->flags)) | |
64 | return; /* i-cache is already coherent with d-cache */ | |
65 | ||
273988fa | 66 | flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page))); |
1da177e4 LT |
67 | set_bit(PG_arch_1, &page->flags); /* mark page as clean */ |
68 | } | |
69 | ||
3fed6ae4 | 70 | #ifdef CONFIG_SWIOTLB |
cde14bbf JB |
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 | */ | |
68c60834 CH |
76 | void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr, |
77 | size_t size, enum dma_data_direction dir) | |
cde14bbf | 78 | { |
68c60834 CH |
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)); | |
cde14bbf | 84 | } |
3fed6ae4 | 85 | #endif |
cde14bbf | 86 | |
1da177e4 LT |
87 | inline void |
88 | ia64_set_rbs_bot (void) | |
89 | { | |
02b763b8 | 90 | unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16; |
1da177e4 LT |
91 | |
92 | if (stack_size > MAX_USER_STACK_SIZE) | |
93 | stack_size = MAX_USER_STACK_SIZE; | |
83d2cd3d | 94 | current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size); |
1da177e4 LT |
95 | } |
96 | ||
97 | /* | |
98 | * This performs some platform-dependent address space initialization. | |
99 | * On IA-64, we want to setup the VM area for the register backing | |
100 | * store (which grows upwards) and install the gateway page which is | |
101 | * used for signal trampolines, etc. | |
102 | */ | |
103 | void | |
104 | ia64_init_addr_space (void) | |
105 | { | |
106 | struct vm_area_struct *vma; | |
107 | ||
108 | ia64_set_rbs_bot(); | |
109 | ||
110 | /* | |
111 | * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore | |
112 | * the problem. When the process attempts to write to the register backing store | |
113 | * for the first time, it will get a SEGFAULT in this case. | |
114 | */ | |
490fc053 | 115 | vma = vm_area_alloc(current->mm); |
1da177e4 | 116 | if (vma) { |
ebad825c | 117 | vma_set_anonymous(vma); |
1da177e4 LT |
118 | vma->vm_start = current->thread.rbs_bot & PAGE_MASK; |
119 | vma->vm_end = vma->vm_start + PAGE_SIZE; | |
46dea3d0 | 120 | vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT; |
3ed75eb8 | 121 | vma->vm_page_prot = vm_get_page_prot(vma->vm_flags); |
1da177e4 LT |
122 | down_write(¤t->mm->mmap_sem); |
123 | if (insert_vm_struct(current->mm, vma)) { | |
124 | up_write(¤t->mm->mmap_sem); | |
3928d4f5 | 125 | vm_area_free(vma); |
1da177e4 LT |
126 | return; |
127 | } | |
128 | up_write(¤t->mm->mmap_sem); | |
129 | } | |
130 | ||
131 | /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */ | |
132 | if (!(current->personality & MMAP_PAGE_ZERO)) { | |
490fc053 | 133 | vma = vm_area_alloc(current->mm); |
1da177e4 | 134 | if (vma) { |
ebad825c | 135 | vma_set_anonymous(vma); |
1da177e4 LT |
136 | vma->vm_end = PAGE_SIZE; |
137 | vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT); | |
314e51b9 KK |
138 | vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO | |
139 | VM_DONTEXPAND | VM_DONTDUMP; | |
1da177e4 LT |
140 | down_write(¤t->mm->mmap_sem); |
141 | if (insert_vm_struct(current->mm, vma)) { | |
142 | up_write(¤t->mm->mmap_sem); | |
3928d4f5 | 143 | vm_area_free(vma); |
1da177e4 LT |
144 | return; |
145 | } | |
146 | up_write(¤t->mm->mmap_sem); | |
147 | } | |
148 | } | |
149 | } | |
150 | ||
151 | void | |
152 | free_initmem (void) | |
153 | { | |
11199692 | 154 | free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end), |
dbe67df4 | 155 | -1, "unused kernel"); |
1da177e4 LT |
156 | } |
157 | ||
dae28066 | 158 | void __init |
1da177e4 LT |
159 | free_initrd_mem (unsigned long start, unsigned long end) |
160 | { | |
1da177e4 LT |
161 | /* |
162 | * EFI uses 4KB pages while the kernel can use 4KB or bigger. | |
163 | * Thus EFI and the kernel may have different page sizes. It is | |
164 | * therefore possible to have the initrd share the same page as | |
165 | * the end of the kernel (given current setup). | |
166 | * | |
167 | * To avoid freeing/using the wrong page (kernel sized) we: | |
168 | * - align up the beginning of initrd | |
169 | * - align down the end of initrd | |
170 | * | |
171 | * | | | |
172 | * |=============| a000 | |
173 | * | | | |
174 | * | | | |
175 | * | | 9000 | |
176 | * |/////////////| | |
177 | * |/////////////| | |
178 | * |=============| 8000 | |
179 | * |///INITRD////| | |
180 | * |/////////////| | |
181 | * |/////////////| 7000 | |
182 | * | | | |
183 | * |KKKKKKKKKKKKK| | |
184 | * |=============| 6000 | |
185 | * |KKKKKKKKKKKKK| | |
186 | * |KKKKKKKKKKKKK| | |
187 | * K=kernel using 8KB pages | |
188 | * | |
189 | * In this example, we must free page 8000 ONLY. So we must align up | |
190 | * initrd_start and keep initrd_end as is. | |
191 | */ | |
192 | start = PAGE_ALIGN(start); | |
193 | end = end & PAGE_MASK; | |
194 | ||
195 | if (start < end) | |
196 | printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10); | |
197 | ||
198 | for (; start < end; start += PAGE_SIZE) { | |
199 | if (!virt_addr_valid(start)) | |
200 | continue; | |
66f62594 | 201 | free_reserved_page(virt_to_page(start)); |
1da177e4 LT |
202 | } |
203 | } | |
204 | ||
205 | /* | |
206 | * This installs a clean page in the kernel's page table. | |
207 | */ | |
dae28066 | 208 | static struct page * __init |
1da177e4 LT |
209 | put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot) |
210 | { | |
211 | pgd_t *pgd; | |
212 | pud_t *pud; | |
213 | pmd_t *pmd; | |
214 | pte_t *pte; | |
215 | ||
1da177e4 LT |
216 | pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */ |
217 | ||
1da177e4 LT |
218 | { |
219 | pud = pud_alloc(&init_mm, pgd, address); | |
220 | if (!pud) | |
221 | goto out; | |
1da177e4 LT |
222 | pmd = pmd_alloc(&init_mm, pud, address); |
223 | if (!pmd) | |
224 | goto out; | |
872fec16 | 225 | pte = pte_alloc_kernel(pmd, address); |
1da177e4 LT |
226 | if (!pte) |
227 | goto out; | |
872fec16 | 228 | if (!pte_none(*pte)) |
1da177e4 | 229 | goto out; |
1da177e4 | 230 | set_pte(pte, mk_pte(page, pgprot)); |
1da177e4 | 231 | } |
872fec16 | 232 | out: |
1da177e4 LT |
233 | /* no need for flush_tlb */ |
234 | return page; | |
235 | } | |
236 | ||
914a4ea4 | 237 | static void __init |
1da177e4 LT |
238 | setup_gate (void) |
239 | { | |
240 | struct page *page; | |
241 | ||
242 | /* | |
ad597bd5 DMT |
243 | * Map the gate page twice: once read-only to export the ELF |
244 | * headers etc. and once execute-only page to enable | |
245 | * privilege-promotion via "epc": | |
1da177e4 | 246 | */ |
e55645ec | 247 | page = virt_to_page(ia64_imva(__start_gate_section)); |
1da177e4 LT |
248 | put_kernel_page(page, GATE_ADDR, PAGE_READONLY); |
249 | #ifdef HAVE_BUGGY_SEGREL | |
e55645ec | 250 | page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE)); |
1da177e4 LT |
251 | put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE); |
252 | #else | |
253 | put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE); | |
ad597bd5 DMT |
254 | /* Fill in the holes (if any) with read-only zero pages: */ |
255 | { | |
256 | unsigned long addr; | |
257 | ||
258 | for (addr = GATE_ADDR + PAGE_SIZE; | |
259 | addr < GATE_ADDR + PERCPU_PAGE_SIZE; | |
260 | addr += PAGE_SIZE) | |
261 | { | |
262 | put_kernel_page(ZERO_PAGE(0), addr, | |
263 | PAGE_READONLY); | |
264 | put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE, | |
265 | PAGE_READONLY); | |
266 | } | |
267 | } | |
1da177e4 LT |
268 | #endif |
269 | ia64_patch_gate(); | |
270 | } | |
271 | ||
a6c19dfe AL |
272 | static struct vm_area_struct gate_vma; |
273 | ||
274 | static int __init gate_vma_init(void) | |
275 | { | |
2c4541e2 | 276 | vma_init(&gate_vma, NULL); |
a6c19dfe AL |
277 | gate_vma.vm_start = FIXADDR_USER_START; |
278 | gate_vma.vm_end = FIXADDR_USER_END; | |
279 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; | |
280 | gate_vma.vm_page_prot = __P101; | |
281 | ||
282 | return 0; | |
283 | } | |
284 | __initcall(gate_vma_init); | |
285 | ||
286 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) | |
287 | { | |
288 | return &gate_vma; | |
289 | } | |
290 | ||
291 | int in_gate_area_no_mm(unsigned long addr) | |
292 | { | |
293 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
294 | return 1; | |
295 | return 0; | |
296 | } | |
297 | ||
298 | int in_gate_area(struct mm_struct *mm, unsigned long addr) | |
299 | { | |
300 | return in_gate_area_no_mm(addr); | |
301 | } | |
302 | ||
5b5e76e9 | 303 | void ia64_mmu_init(void *my_cpu_data) |
1da177e4 | 304 | { |
00b65985 | 305 | unsigned long pta, impl_va_bits; |
5b5e76e9 | 306 | extern void tlb_init(void); |
1da177e4 LT |
307 | |
308 | #ifdef CONFIG_DISABLE_VHPT | |
309 | # define VHPT_ENABLE_BIT 0 | |
310 | #else | |
311 | # define VHPT_ENABLE_BIT 1 | |
312 | #endif | |
313 | ||
1da177e4 LT |
314 | /* |
315 | * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped | |
316 | * address space. The IA-64 architecture guarantees that at least 50 bits of | |
317 | * virtual address space are implemented but if we pick a large enough page size | |
318 | * (e.g., 64KB), the mapped address space is big enough that it will overlap with | |
319 | * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages, | |
320 | * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a | |
321 | * problem in practice. Alternatively, we could truncate the top of the mapped | |
322 | * address space to not permit mappings that would overlap with the VMLPT. | |
323 | * --davidm 00/12/06 | |
324 | */ | |
325 | # define pte_bits 3 | |
326 | # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT) | |
327 | /* | |
328 | * The virtual page table has to cover the entire implemented address space within | |
329 | * a region even though not all of this space may be mappable. The reason for | |
330 | * this is that the Access bit and Dirty bit fault handlers perform | |
331 | * non-speculative accesses to the virtual page table, so the address range of the | |
332 | * virtual page table itself needs to be covered by virtual page table. | |
333 | */ | |
334 | # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits) | |
335 | # define POW2(n) (1ULL << (n)) | |
336 | ||
337 | impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61))); | |
338 | ||
339 | if (impl_va_bits < 51 || impl_va_bits > 61) | |
340 | panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1); | |
6cf07a8c PC |
341 | /* |
342 | * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need, | |
343 | * which must fit into "vmlpt_bits - pte_bits" slots. Second half of | |
344 | * the test makes sure that our mapped space doesn't overlap the | |
345 | * unimplemented hole in the middle of the region. | |
346 | */ | |
347 | if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) || | |
348 | (mapped_space_bits > impl_va_bits - 1)) | |
349 | panic("Cannot build a big enough virtual-linear page table" | |
350 | " to cover mapped address space.\n" | |
351 | " Try using a smaller page size.\n"); | |
352 | ||
1da177e4 LT |
353 | |
354 | /* place the VMLPT at the end of each page-table mapped region: */ | |
355 | pta = POW2(61) - POW2(vmlpt_bits); | |
356 | ||
1da177e4 LT |
357 | /* |
358 | * Set the (virtually mapped linear) page table address. Bit | |
359 | * 8 selects between the short and long format, bits 2-7 the | |
360 | * size of the table, and bit 0 whether the VHPT walker is | |
361 | * enabled. | |
362 | */ | |
363 | ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT); | |
364 | ||
365 | ia64_tlb_init(); | |
366 | ||
367 | #ifdef CONFIG_HUGETLB_PAGE | |
368 | ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2); | |
369 | ia64_srlz_d(); | |
370 | #endif | |
371 | } | |
372 | ||
373 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
e44e41d0 BP |
374 | int vmemmap_find_next_valid_pfn(int node, int i) |
375 | { | |
376 | unsigned long end_address, hole_next_pfn; | |
377 | unsigned long stop_address; | |
378 | pg_data_t *pgdat = NODE_DATA(node); | |
379 | ||
380 | end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i]; | |
381 | end_address = PAGE_ALIGN(end_address); | |
6408068e | 382 | stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)]; |
e44e41d0 BP |
383 | |
384 | do { | |
385 | pgd_t *pgd; | |
386 | pud_t *pud; | |
387 | pmd_t *pmd; | |
388 | pte_t *pte; | |
389 | ||
390 | pgd = pgd_offset_k(end_address); | |
391 | if (pgd_none(*pgd)) { | |
392 | end_address += PGDIR_SIZE; | |
393 | continue; | |
394 | } | |
395 | ||
396 | pud = pud_offset(pgd, end_address); | |
397 | if (pud_none(*pud)) { | |
398 | end_address += PUD_SIZE; | |
399 | continue; | |
400 | } | |
401 | ||
402 | pmd = pmd_offset(pud, end_address); | |
403 | if (pmd_none(*pmd)) { | |
404 | end_address += PMD_SIZE; | |
405 | continue; | |
406 | } | |
407 | ||
408 | pte = pte_offset_kernel(pmd, end_address); | |
409 | retry_pte: | |
410 | if (pte_none(*pte)) { | |
411 | end_address += PAGE_SIZE; | |
412 | pte++; | |
413 | if ((end_address < stop_address) && | |
414 | (end_address != ALIGN(end_address, 1UL << PMD_SHIFT))) | |
415 | goto retry_pte; | |
416 | continue; | |
417 | } | |
418 | /* Found next valid vmem_map page */ | |
419 | break; | |
420 | } while (end_address < stop_address); | |
421 | ||
422 | end_address = min(end_address, stop_address); | |
423 | end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1; | |
424 | hole_next_pfn = end_address / sizeof(struct page); | |
425 | return hole_next_pfn - pgdat->node_start_pfn; | |
426 | } | |
1da177e4 | 427 | |
e088a4ad | 428 | int __init create_mem_map_page_table(u64 start, u64 end, void *arg) |
1da177e4 LT |
429 | { |
430 | unsigned long address, start_page, end_page; | |
431 | struct page *map_start, *map_end; | |
432 | int node; | |
433 | pgd_t *pgd; | |
434 | pud_t *pud; | |
435 | pmd_t *pmd; | |
436 | pte_t *pte; | |
437 | ||
438 | map_start = vmem_map + (__pa(start) >> PAGE_SHIFT); | |
439 | map_end = vmem_map + (__pa(end) >> PAGE_SHIFT); | |
440 | ||
441 | start_page = (unsigned long) map_start & PAGE_MASK; | |
442 | end_page = PAGE_ALIGN((unsigned long) map_end); | |
443 | node = paddr_to_nid(__pa(start)); | |
444 | ||
445 | for (address = start_page; address < end_page; address += PAGE_SIZE) { | |
446 | pgd = pgd_offset_k(address); | |
d80db5c1 MR |
447 | if (pgd_none(*pgd)) { |
448 | pud = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node); | |
449 | if (!pud) | |
450 | goto err_alloc; | |
451 | pgd_populate(&init_mm, pgd, pud); | |
452 | } | |
1da177e4 LT |
453 | pud = pud_offset(pgd, address); |
454 | ||
d80db5c1 MR |
455 | if (pud_none(*pud)) { |
456 | pmd = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node); | |
457 | if (!pmd) | |
458 | goto err_alloc; | |
459 | pud_populate(&init_mm, pud, pmd); | |
460 | } | |
1da177e4 LT |
461 | pmd = pmd_offset(pud, address); |
462 | ||
d80db5c1 MR |
463 | if (pmd_none(*pmd)) { |
464 | pte = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node); | |
465 | if (!pte) | |
466 | goto err_alloc; | |
467 | pmd_populate_kernel(&init_mm, pmd, pte); | |
468 | } | |
1da177e4 LT |
469 | pte = pte_offset_kernel(pmd, address); |
470 | ||
d80db5c1 MR |
471 | if (pte_none(*pte)) { |
472 | void *page = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, | |
473 | node); | |
474 | if (!page) | |
475 | goto err_alloc; | |
476 | set_pte(pte, pfn_pte(__pa(page) >> PAGE_SHIFT, | |
1da177e4 | 477 | PAGE_KERNEL)); |
d80db5c1 | 478 | } |
1da177e4 LT |
479 | } |
480 | return 0; | |
d80db5c1 MR |
481 | |
482 | err_alloc: | |
483 | panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d\n", | |
484 | __func__, PAGE_SIZE, PAGE_SIZE, node); | |
485 | return -ENOMEM; | |
1da177e4 LT |
486 | } |
487 | ||
488 | struct memmap_init_callback_data { | |
489 | struct page *start; | |
490 | struct page *end; | |
491 | int nid; | |
492 | unsigned long zone; | |
493 | }; | |
494 | ||
18b8befd | 495 | static int __meminit |
e088a4ad | 496 | virtual_memmap_init(u64 start, u64 end, void *arg) |
1da177e4 LT |
497 | { |
498 | struct memmap_init_callback_data *args; | |
499 | struct page *map_start, *map_end; | |
500 | ||
501 | args = (struct memmap_init_callback_data *) arg; | |
502 | map_start = vmem_map + (__pa(start) >> PAGE_SHIFT); | |
503 | map_end = vmem_map + (__pa(end) >> PAGE_SHIFT); | |
504 | ||
505 | if (map_start < args->start) | |
506 | map_start = args->start; | |
507 | if (map_end > args->end) | |
508 | map_end = args->end; | |
509 | ||
510 | /* | |
511 | * We have to initialize "out of bounds" struct page elements that fit completely | |
512 | * on the same pages that were allocated for the "in bounds" elements because they | |
513 | * may be referenced later (and found to be "reserved"). | |
514 | */ | |
515 | map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page); | |
516 | map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end) | |
517 | / sizeof(struct page)); | |
518 | ||
519 | if (map_start < map_end) | |
520 | memmap_init_zone((unsigned long)(map_end - map_start), | |
a2f3aa02 | 521 | args->nid, args->zone, page_to_pfn(map_start), |
a99583e7 | 522 | MEMMAP_EARLY, NULL); |
1da177e4 LT |
523 | return 0; |
524 | } | |
525 | ||
18b8befd | 526 | void __meminit |
1da177e4 LT |
527 | memmap_init (unsigned long size, int nid, unsigned long zone, |
528 | unsigned long start_pfn) | |
529 | { | |
a99583e7 CH |
530 | if (!vmem_map) { |
531 | memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY, | |
532 | NULL); | |
533 | } else { | |
1da177e4 LT |
534 | struct page *start; |
535 | struct memmap_init_callback_data args; | |
536 | ||
537 | start = pfn_to_page(start_pfn); | |
538 | args.start = start; | |
539 | args.end = start + size; | |
540 | args.nid = nid; | |
541 | args.zone = zone; | |
542 | ||
543 | efi_memmap_walk(virtual_memmap_init, &args); | |
544 | } | |
545 | } | |
546 | ||
547 | int | |
548 | ia64_pfn_valid (unsigned long pfn) | |
549 | { | |
550 | char byte; | |
551 | struct page *pg = pfn_to_page(pfn); | |
552 | ||
553 | return (__get_user(byte, (char __user *) pg) == 0) | |
554 | && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK)) | |
555 | || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0)); | |
556 | } | |
557 | EXPORT_SYMBOL(ia64_pfn_valid); | |
558 | ||
e088a4ad | 559 | int __init find_largest_hole(u64 start, u64 end, void *arg) |
1da177e4 LT |
560 | { |
561 | u64 *max_gap = arg; | |
562 | ||
563 | static u64 last_end = PAGE_OFFSET; | |
564 | ||
565 | /* NOTE: this algorithm assumes efi memmap table is ordered */ | |
566 | ||
567 | if (*max_gap < (start - last_end)) | |
568 | *max_gap = start - last_end; | |
569 | last_end = end; | |
570 | return 0; | |
571 | } | |
05e0caad | 572 | |
139b8304 BP |
573 | #endif /* CONFIG_VIRTUAL_MEM_MAP */ |
574 | ||
e088a4ad | 575 | int __init register_active_ranges(u64 start, u64 len, int nid) |
05e0caad | 576 | { |
98075d24 | 577 | u64 end = start + len; |
139b8304 | 578 | |
139b8304 BP |
579 | #ifdef CONFIG_KEXEC |
580 | if (start > crashk_res.start && start < crashk_res.end) | |
581 | start = crashk_res.end; | |
582 | if (end > crashk_res.start && end < crashk_res.end) | |
583 | end = crashk_res.start; | |
584 | #endif | |
585 | ||
586 | if (start < end) | |
98e4ae8a | 587 | memblock_add_node(__pa(start), end - start, nid); |
05e0caad MG |
588 | return 0; |
589 | } | |
1da177e4 | 590 | |
a3f5c338 | 591 | int |
e088a4ad | 592 | find_max_min_low_pfn (u64 start, u64 end, void *arg) |
a3f5c338 ZN |
593 | { |
594 | unsigned long pfn_start, pfn_end; | |
595 | #ifdef CONFIG_FLATMEM | |
596 | pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT; | |
597 | pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT; | |
598 | #else | |
599 | pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT; | |
600 | pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT; | |
601 | #endif | |
602 | min_low_pfn = min(min_low_pfn, pfn_start); | |
603 | max_low_pfn = max(max_low_pfn, pfn_end); | |
604 | return 0; | |
605 | } | |
606 | ||
1da177e4 LT |
607 | /* |
608 | * Boot command-line option "nolwsys" can be used to disable the use of any light-weight | |
609 | * system call handler. When this option is in effect, all fsyscalls will end up bubbling | |
610 | * down into the kernel and calling the normal (heavy-weight) syscall handler. This is | |
611 | * useful for performance testing, but conceivably could also come in handy for debugging | |
612 | * purposes. | |
613 | */ | |
614 | ||
03906ea0 | 615 | static int nolwsys __initdata; |
1da177e4 LT |
616 | |
617 | static int __init | |
618 | nolwsys_setup (char *s) | |
619 | { | |
620 | nolwsys = 1; | |
621 | return 1; | |
622 | } | |
623 | ||
624 | __setup("nolwsys", nolwsys_setup); | |
625 | ||
dae28066 | 626 | void __init |
1da177e4 LT |
627 | mem_init (void) |
628 | { | |
1da177e4 | 629 | int i; |
1da177e4 | 630 | |
fde740e4 RH |
631 | BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE); |
632 | BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE); | |
633 | BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE); | |
634 | ||
1da177e4 LT |
635 | #ifdef CONFIG_PCI |
636 | /* | |
637 | * This needs to be called _after_ the command line has been parsed but _before_ | |
638 | * any drivers that may need the PCI DMA interface are initialized or bootmem has | |
639 | * been freed. | |
640 | */ | |
641 | platform_dma_init(); | |
642 | #endif | |
643 | ||
2d4b1fa2 | 644 | #ifdef CONFIG_FLATMEM |
80a03e29 | 645 | BUG_ON(!mem_map); |
1da177e4 LT |
646 | #endif |
647 | ||
b57b63a2 | 648 | set_max_mapnr(max_low_pfn); |
1da177e4 | 649 | high_memory = __va(max_low_pfn * PAGE_SIZE); |
c6ffc5ca | 650 | memblock_free_all(); |
de4bcddc | 651 | mem_init_print_info(NULL); |
1da177e4 LT |
652 | |
653 | /* | |
654 | * For fsyscall entrpoints with no light-weight handler, use the ordinary | |
655 | * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry | |
656 | * code can tell them apart. | |
657 | */ | |
658 | for (i = 0; i < NR_syscalls; ++i) { | |
e55645ec | 659 | extern unsigned long fsyscall_table[NR_syscalls]; |
1da177e4 LT |
660 | extern unsigned long sys_call_table[NR_syscalls]; |
661 | ||
662 | if (!fsyscall_table[i] || nolwsys) | |
663 | fsyscall_table[i] = sys_call_table[i] | 1; | |
664 | } | |
665 | setup_gate(); | |
1da177e4 | 666 | } |
1681b8e1 YG |
667 | |
668 | #ifdef CONFIG_MEMORY_HOTPLUG | |
940519f0 MH |
669 | int arch_add_memory(int nid, u64 start, u64 size, |
670 | struct mhp_restrictions *restrictions) | |
1681b8e1 | 671 | { |
1681b8e1 YG |
672 | unsigned long start_pfn = start >> PAGE_SHIFT; |
673 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
674 | int ret; | |
675 | ||
940519f0 | 676 | ret = __add_pages(nid, start_pfn, nr_pages, restrictions); |
1681b8e1 YG |
677 | if (ret) |
678 | printk("%s: Problem encountered in __add_pages() as ret=%d\n", | |
d4ed8084 | 679 | __func__, ret); |
1681b8e1 YG |
680 | |
681 | return ret; | |
682 | } | |
24d335ca WC |
683 | |
684 | #ifdef CONFIG_MEMORY_HOTREMOVE | |
ac5c9426 DH |
685 | void arch_remove_memory(int nid, u64 start, u64 size, |
686 | struct vmem_altmap *altmap) | |
24d335ca WC |
687 | { |
688 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
689 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
690 | struct zone *zone; | |
24d335ca WC |
691 | |
692 | zone = page_zone(pfn_to_page(start_pfn)); | |
ac5c9426 | 693 | __remove_pages(zone, start_pfn, nr_pages, altmap); |
24d335ca WC |
694 | } |
695 | #endif | |
1681b8e1 | 696 | #endif |