]>
Commit | Line | Data |
---|---|---|
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 | ||
41 | extern void ia64_tlb_init (void); | |
42 | ||
43 | unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL; | |
44 | ||
45 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
46 | unsigned long VMALLOC_END = VMALLOC_END_INIT; | |
47 | EXPORT_SYMBOL(VMALLOC_END); | |
48 | struct page *vmem_map; | |
49 | EXPORT_SYMBOL(vmem_map); | |
50 | #endif | |
51 | ||
52 | struct page *zero_page_memmap_ptr; /* map entry for zero page */ | |
53 | EXPORT_SYMBOL(zero_page_memmap_ptr); | |
54 | ||
55 | void | |
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 | */ | |
76 | void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr, | |
77 | size_t size, 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 | ||
86 | inline void | |
87 | ia64_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 | */ | |
102 | void | |
103 | ia64_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 | down_write(¤t->mm->mmap_sem); | |
122 | if (insert_vm_struct(current->mm, vma)) { | |
123 | up_write(¤t->mm->mmap_sem); | |
124 | vm_area_free(vma); | |
125 | return; | |
126 | } | |
127 | up_write(¤t->mm->mmap_sem); | |
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 | down_write(¤t->mm->mmap_sem); | |
140 | if (insert_vm_struct(current->mm, vma)) { | |
141 | up_write(¤t->mm->mmap_sem); | |
142 | vm_area_free(vma); | |
143 | return; | |
144 | } | |
145 | up_write(¤t->mm->mmap_sem); | |
146 | } | |
147 | } | |
148 | } | |
149 | ||
150 | void | |
151 | free_initmem (void) | |
152 | { | |
153 | free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end), | |
154 | -1, "unused kernel"); | |
155 | } | |
156 | ||
157 | void __init | |
158 | free_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 | */ | |
207 | static struct page * __init | |
208 | put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot) | |
209 | { | |
210 | pgd_t *pgd; | |
211 | pud_t *pud; | |
212 | pmd_t *pmd; | |
213 | pte_t *pte; | |
214 | ||
215 | pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */ | |
216 | ||
217 | { | |
218 | pud = pud_alloc(&init_mm, pgd, address); | |
219 | if (!pud) | |
220 | goto out; | |
221 | pmd = pmd_alloc(&init_mm, pud, address); | |
222 | if (!pmd) | |
223 | goto out; | |
224 | pte = pte_alloc_kernel(pmd, address); | |
225 | if (!pte) | |
226 | goto out; | |
227 | if (!pte_none(*pte)) | |
228 | goto out; | |
229 | set_pte(pte, mk_pte(page, pgprot)); | |
230 | } | |
231 | out: | |
232 | /* no need for flush_tlb */ | |
233 | return page; | |
234 | } | |
235 | ||
236 | static void __init | |
237 | setup_gate (void) | |
238 | { | |
239 | struct page *page; | |
240 | ||
241 | /* | |
242 | * Map the gate page twice: once read-only to export the ELF | |
243 | * headers etc. and once execute-only page to enable | |
244 | * privilege-promotion via "epc": | |
245 | */ | |
246 | page = virt_to_page(ia64_imva(__start_gate_section)); | |
247 | put_kernel_page(page, GATE_ADDR, PAGE_READONLY); | |
248 | #ifdef HAVE_BUGGY_SEGREL | |
249 | page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE)); | |
250 | put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE); | |
251 | #else | |
252 | put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE); | |
253 | /* Fill in the holes (if any) with read-only zero pages: */ | |
254 | { | |
255 | unsigned long addr; | |
256 | ||
257 | for (addr = GATE_ADDR + PAGE_SIZE; | |
258 | addr < GATE_ADDR + PERCPU_PAGE_SIZE; | |
259 | addr += PAGE_SIZE) | |
260 | { | |
261 | put_kernel_page(ZERO_PAGE(0), addr, | |
262 | PAGE_READONLY); | |
263 | put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE, | |
264 | PAGE_READONLY); | |
265 | } | |
266 | } | |
267 | #endif | |
268 | ia64_patch_gate(); | |
269 | } | |
270 | ||
271 | static struct vm_area_struct gate_vma; | |
272 | ||
273 | static int __init gate_vma_init(void) | |
274 | { | |
275 | vma_init(&gate_vma, NULL); | |
276 | gate_vma.vm_start = FIXADDR_USER_START; | |
277 | gate_vma.vm_end = FIXADDR_USER_END; | |
278 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; | |
279 | gate_vma.vm_page_prot = __P101; | |
280 | ||
281 | return 0; | |
282 | } | |
283 | __initcall(gate_vma_init); | |
284 | ||
285 | struct vm_area_struct *get_gate_vma(struct mm_struct *mm) | |
286 | { | |
287 | return &gate_vma; | |
288 | } | |
289 | ||
290 | int in_gate_area_no_mm(unsigned long addr) | |
291 | { | |
292 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) | |
293 | return 1; | |
294 | return 0; | |
295 | } | |
296 | ||
297 | int in_gate_area(struct mm_struct *mm, unsigned long addr) | |
298 | { | |
299 | return in_gate_area_no_mm(addr); | |
300 | } | |
301 | ||
302 | void ia64_mmu_init(void *my_cpu_data) | |
303 | { | |
304 | unsigned long pta, impl_va_bits; | |
305 | extern void tlb_init(void); | |
306 | ||
307 | #ifdef CONFIG_DISABLE_VHPT | |
308 | # define VHPT_ENABLE_BIT 0 | |
309 | #else | |
310 | # define VHPT_ENABLE_BIT 1 | |
311 | #endif | |
312 | ||
313 | /* | |
314 | * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped | |
315 | * address space. The IA-64 architecture guarantees that at least 50 bits of | |
316 | * virtual address space are implemented but if we pick a large enough page size | |
317 | * (e.g., 64KB), the mapped address space is big enough that it will overlap with | |
318 | * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages, | |
319 | * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a | |
320 | * problem in practice. Alternatively, we could truncate the top of the mapped | |
321 | * address space to not permit mappings that would overlap with the VMLPT. | |
322 | * --davidm 00/12/06 | |
323 | */ | |
324 | # define pte_bits 3 | |
325 | # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT) | |
326 | /* | |
327 | * The virtual page table has to cover the entire implemented address space within | |
328 | * a region even though not all of this space may be mappable. The reason for | |
329 | * this is that the Access bit and Dirty bit fault handlers perform | |
330 | * non-speculative accesses to the virtual page table, so the address range of the | |
331 | * virtual page table itself needs to be covered by virtual page table. | |
332 | */ | |
333 | # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits) | |
334 | # define POW2(n) (1ULL << (n)) | |
335 | ||
336 | impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61))); | |
337 | ||
338 | if (impl_va_bits < 51 || impl_va_bits > 61) | |
339 | panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1); | |
340 | /* | |
341 | * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need, | |
342 | * which must fit into "vmlpt_bits - pte_bits" slots. Second half of | |
343 | * the test makes sure that our mapped space doesn't overlap the | |
344 | * unimplemented hole in the middle of the region. | |
345 | */ | |
346 | if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) || | |
347 | (mapped_space_bits > impl_va_bits - 1)) | |
348 | panic("Cannot build a big enough virtual-linear page table" | |
349 | " to cover mapped address space.\n" | |
350 | " Try using a smaller page size.\n"); | |
351 | ||
352 | ||
353 | /* place the VMLPT at the end of each page-table mapped region: */ | |
354 | pta = POW2(61) - POW2(vmlpt_bits); | |
355 | ||
356 | /* | |
357 | * Set the (virtually mapped linear) page table address. Bit | |
358 | * 8 selects between the short and long format, bits 2-7 the | |
359 | * size of the table, and bit 0 whether the VHPT walker is | |
360 | * enabled. | |
361 | */ | |
362 | ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT); | |
363 | ||
364 | ia64_tlb_init(); | |
365 | ||
366 | #ifdef CONFIG_HUGETLB_PAGE | |
367 | ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2); | |
368 | ia64_srlz_d(); | |
369 | #endif | |
370 | } | |
371 | ||
372 | #ifdef CONFIG_VIRTUAL_MEM_MAP | |
373 | int vmemmap_find_next_valid_pfn(int node, int i) | |
374 | { | |
375 | unsigned long end_address, hole_next_pfn; | |
376 | unsigned long stop_address; | |
377 | pg_data_t *pgdat = NODE_DATA(node); | |
378 | ||
379 | end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i]; | |
380 | end_address = PAGE_ALIGN(end_address); | |
381 | stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)]; | |
382 | ||
383 | do { | |
384 | pgd_t *pgd; | |
385 | pud_t *pud; | |
386 | pmd_t *pmd; | |
387 | pte_t *pte; | |
388 | ||
389 | pgd = pgd_offset_k(end_address); | |
390 | if (pgd_none(*pgd)) { | |
391 | end_address += PGDIR_SIZE; | |
392 | continue; | |
393 | } | |
394 | ||
395 | pud = pud_offset(pgd, end_address); | |
396 | if (pud_none(*pud)) { | |
397 | end_address += PUD_SIZE; | |
398 | continue; | |
399 | } | |
400 | ||
401 | pmd = pmd_offset(pud, end_address); | |
402 | if (pmd_none(*pmd)) { | |
403 | end_address += PMD_SIZE; | |
404 | continue; | |
405 | } | |
406 | ||
407 | pte = pte_offset_kernel(pmd, end_address); | |
408 | retry_pte: | |
409 | if (pte_none(*pte)) { | |
410 | end_address += PAGE_SIZE; | |
411 | pte++; | |
412 | if ((end_address < stop_address) && | |
413 | (end_address != ALIGN(end_address, 1UL << PMD_SHIFT))) | |
414 | goto retry_pte; | |
415 | continue; | |
416 | } | |
417 | /* Found next valid vmem_map page */ | |
418 | break; | |
419 | } while (end_address < stop_address); | |
420 | ||
421 | end_address = min(end_address, stop_address); | |
422 | end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1; | |
423 | hole_next_pfn = end_address / sizeof(struct page); | |
424 | return hole_next_pfn - pgdat->node_start_pfn; | |
425 | } | |
426 | ||
427 | int __init create_mem_map_page_table(u64 start, u64 end, void *arg) | |
428 | { | |
429 | unsigned long address, start_page, end_page; | |
430 | struct page *map_start, *map_end; | |
431 | int node; | |
432 | pgd_t *pgd; | |
433 | pud_t *pud; | |
434 | pmd_t *pmd; | |
435 | pte_t *pte; | |
436 | ||
437 | map_start = vmem_map + (__pa(start) >> PAGE_SHIFT); | |
438 | map_end = vmem_map + (__pa(end) >> PAGE_SHIFT); | |
439 | ||
440 | start_page = (unsigned long) map_start & PAGE_MASK; | |
441 | end_page = PAGE_ALIGN((unsigned long) map_end); | |
442 | node = paddr_to_nid(__pa(start)); | |
443 | ||
444 | for (address = start_page; address < end_page; address += PAGE_SIZE) { | |
445 | pgd = pgd_offset_k(address); | |
446 | if (pgd_none(*pgd)) { | |
447 | pud = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node); | |
448 | if (!pud) | |
449 | goto err_alloc; | |
450 | pgd_populate(&init_mm, pgd, pud); | |
451 | } | |
452 | pud = pud_offset(pgd, address); | |
453 | ||
454 | if (pud_none(*pud)) { | |
455 | pmd = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node); | |
456 | if (!pmd) | |
457 | goto err_alloc; | |
458 | pud_populate(&init_mm, pud, pmd); | |
459 | } | |
460 | pmd = pmd_offset(pud, address); | |
461 | ||
462 | if (pmd_none(*pmd)) { | |
463 | pte = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node); | |
464 | if (!pte) | |
465 | goto err_alloc; | |
466 | pmd_populate_kernel(&init_mm, pmd, pte); | |
467 | } | |
468 | pte = pte_offset_kernel(pmd, address); | |
469 | ||
470 | if (pte_none(*pte)) { | |
471 | void *page = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, | |
472 | node); | |
473 | if (!page) | |
474 | goto err_alloc; | |
475 | set_pte(pte, pfn_pte(__pa(page) >> PAGE_SHIFT, | |
476 | PAGE_KERNEL)); | |
477 | } | |
478 | } | |
479 | return 0; | |
480 | ||
481 | err_alloc: | |
482 | panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d\n", | |
483 | __func__, PAGE_SIZE, PAGE_SIZE, node); | |
484 | return -ENOMEM; | |
485 | } | |
486 | ||
487 | struct memmap_init_callback_data { | |
488 | struct page *start; | |
489 | struct page *end; | |
490 | int nid; | |
491 | unsigned long zone; | |
492 | }; | |
493 | ||
494 | static int __meminit | |
495 | virtual_memmap_init(u64 start, u64 end, void *arg) | |
496 | { | |
497 | struct memmap_init_callback_data *args; | |
498 | struct page *map_start, *map_end; | |
499 | ||
500 | args = (struct memmap_init_callback_data *) arg; | |
501 | map_start = vmem_map + (__pa(start) >> PAGE_SHIFT); | |
502 | map_end = vmem_map + (__pa(end) >> PAGE_SHIFT); | |
503 | ||
504 | if (map_start < args->start) | |
505 | map_start = args->start; | |
506 | if (map_end > args->end) | |
507 | map_end = args->end; | |
508 | ||
509 | /* | |
510 | * We have to initialize "out of bounds" struct page elements that fit completely | |
511 | * on the same pages that were allocated for the "in bounds" elements because they | |
512 | * may be referenced later (and found to be "reserved"). | |
513 | */ | |
514 | map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page); | |
515 | map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end) | |
516 | / sizeof(struct page)); | |
517 | ||
518 | if (map_start < map_end) | |
519 | memmap_init_zone((unsigned long)(map_end - map_start), | |
520 | args->nid, args->zone, page_to_pfn(map_start), | |
521 | MEMINIT_EARLY, NULL); | |
522 | return 0; | |
523 | } | |
524 | ||
525 | void __meminit | |
526 | memmap_init (unsigned long size, int nid, unsigned long zone, | |
527 | unsigned long start_pfn) | |
528 | { | |
529 | if (!vmem_map) { | |
530 | memmap_init_zone(size, nid, zone, start_pfn, | |
531 | MEMINIT_EARLY, NULL); | |
532 | } else { | |
533 | struct page *start; | |
534 | struct memmap_init_callback_data args; | |
535 | ||
536 | start = pfn_to_page(start_pfn); | |
537 | args.start = start; | |
538 | args.end = start + size; | |
539 | args.nid = nid; | |
540 | args.zone = zone; | |
541 | ||
542 | efi_memmap_walk(virtual_memmap_init, &args); | |
543 | } | |
544 | } | |
545 | ||
546 | int | |
547 | ia64_pfn_valid (unsigned long pfn) | |
548 | { | |
549 | char byte; | |
550 | struct page *pg = pfn_to_page(pfn); | |
551 | ||
552 | return (__get_user(byte, (char __user *) pg) == 0) | |
553 | && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK)) | |
554 | || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0)); | |
555 | } | |
556 | EXPORT_SYMBOL(ia64_pfn_valid); | |
557 | ||
558 | int __init find_largest_hole(u64 start, u64 end, void *arg) | |
559 | { | |
560 | u64 *max_gap = arg; | |
561 | ||
562 | static u64 last_end = PAGE_OFFSET; | |
563 | ||
564 | /* NOTE: this algorithm assumes efi memmap table is ordered */ | |
565 | ||
566 | if (*max_gap < (start - last_end)) | |
567 | *max_gap = start - last_end; | |
568 | last_end = end; | |
569 | return 0; | |
570 | } | |
571 | ||
572 | #endif /* CONFIG_VIRTUAL_MEM_MAP */ | |
573 | ||
574 | int __init register_active_ranges(u64 start, u64 len, int nid) | |
575 | { | |
576 | u64 end = start + len; | |
577 | ||
578 | #ifdef CONFIG_KEXEC | |
579 | if (start > crashk_res.start && start < crashk_res.end) | |
580 | start = crashk_res.end; | |
581 | if (end > crashk_res.start && end < crashk_res.end) | |
582 | end = crashk_res.start; | |
583 | #endif | |
584 | ||
585 | if (start < end) | |
586 | memblock_add_node(__pa(start), end - start, nid); | |
587 | return 0; | |
588 | } | |
589 | ||
590 | int | |
591 | find_max_min_low_pfn (u64 start, u64 end, void *arg) | |
592 | { | |
593 | unsigned long pfn_start, pfn_end; | |
594 | #ifdef CONFIG_FLATMEM | |
595 | pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT; | |
596 | pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT; | |
597 | #else | |
598 | pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT; | |
599 | pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT; | |
600 | #endif | |
601 | min_low_pfn = min(min_low_pfn, pfn_start); | |
602 | max_low_pfn = max(max_low_pfn, pfn_end); | |
603 | return 0; | |
604 | } | |
605 | ||
606 | /* | |
607 | * Boot command-line option "nolwsys" can be used to disable the use of any light-weight | |
608 | * system call handler. When this option is in effect, all fsyscalls will end up bubbling | |
609 | * down into the kernel and calling the normal (heavy-weight) syscall handler. This is | |
610 | * useful for performance testing, but conceivably could also come in handy for debugging | |
611 | * purposes. | |
612 | */ | |
613 | ||
614 | static int nolwsys __initdata; | |
615 | ||
616 | static int __init | |
617 | nolwsys_setup (char *s) | |
618 | { | |
619 | nolwsys = 1; | |
620 | return 1; | |
621 | } | |
622 | ||
623 | __setup("nolwsys", nolwsys_setup); | |
624 | ||
625 | void __init | |
626 | mem_init (void) | |
627 | { | |
628 | int i; | |
629 | ||
630 | BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE); | |
631 | BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE); | |
632 | BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE); | |
633 | ||
634 | /* | |
635 | * This needs to be called _after_ the command line has been parsed but | |
636 | * _before_ any drivers that may need the PCI DMA interface are | |
637 | * initialized or bootmem has been freed. | |
638 | */ | |
639 | #ifdef CONFIG_INTEL_IOMMU | |
640 | detect_intel_iommu(); | |
641 | if (!iommu_detected) | |
642 | #endif | |
643 | #ifdef CONFIG_SWIOTLB | |
644 | swiotlb_init(1); | |
645 | #endif | |
646 | ||
647 | #ifdef CONFIG_FLATMEM | |
648 | BUG_ON(!mem_map); | |
649 | #endif | |
650 | ||
651 | set_max_mapnr(max_low_pfn); | |
652 | high_memory = __va(max_low_pfn * PAGE_SIZE); | |
653 | memblock_free_all(); | |
654 | mem_init_print_info(NULL); | |
655 | ||
656 | /* | |
657 | * For fsyscall entrpoints with no light-weight handler, use the ordinary | |
658 | * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry | |
659 | * code can tell them apart. | |
660 | */ | |
661 | for (i = 0; i < NR_syscalls; ++i) { | |
662 | extern unsigned long fsyscall_table[NR_syscalls]; | |
663 | extern unsigned long sys_call_table[NR_syscalls]; | |
664 | ||
665 | if (!fsyscall_table[i] || nolwsys) | |
666 | fsyscall_table[i] = sys_call_table[i] | 1; | |
667 | } | |
668 | setup_gate(); | |
669 | } | |
670 | ||
671 | #ifdef CONFIG_MEMORY_HOTPLUG | |
672 | int arch_add_memory(int nid, u64 start, u64 size, | |
673 | struct mhp_restrictions *restrictions) | |
674 | { | |
675 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
676 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
677 | int ret; | |
678 | ||
679 | ret = __add_pages(nid, start_pfn, nr_pages, restrictions); | |
680 | if (ret) | |
681 | printk("%s: Problem encountered in __add_pages() as ret=%d\n", | |
682 | __func__, ret); | |
683 | ||
684 | return ret; | |
685 | } | |
686 | ||
687 | void arch_remove_memory(int nid, u64 start, u64 size, | |
688 | struct vmem_altmap *altmap) | |
689 | { | |
690 | unsigned long start_pfn = start >> PAGE_SHIFT; | |
691 | unsigned long nr_pages = size >> PAGE_SHIFT; | |
692 | ||
693 | __remove_pages(start_pfn, nr_pages, altmap); | |
694 | } | |
695 | #endif |