1 /*
2 * PowerPC version
3 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4 *
5 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
6 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
7 * Copyright (C) 1996 Paul Mackerras
8 * PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
9 *
10 * Derived from "arch/i386/mm/init.c"
11 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 *
18 */
19
20 #include <linux/export.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/string.h>
25 #include <linux/gfp.h>
26 #include <linux/types.h>
27 #include <linux/mm.h>
28 #include <linux/stddef.h>
29 #include <linux/init.h>
30 #include <linux/bootmem.h>
31 #include <linux/highmem.h>
32 #include <linux/initrd.h>
33 #include <linux/pagemap.h>
34 #include <linux/suspend.h>
35 #include <linux/memblock.h>
36 #include <linux/hugetlb.h>
37 #include <linux/slab.h>
38 #include <linux/vmalloc.h>
39 #include <linux/memremap.h>
40
41 #include <asm/pgalloc.h>
42 #include <asm/prom.h>
43 #include <asm/io.h>
44 #include <asm/mmu_context.h>
45 #include <asm/pgtable.h>
46 #include <asm/mmu.h>
47 #include <asm/smp.h>
48 #include <asm/machdep.h>
49 #include <asm/btext.h>
50 #include <asm/tlb.h>
51 #include <asm/sections.h>
52 #include <asm/sparsemem.h>
53 #include <asm/vdso.h>
54 #include <asm/fixmap.h>
55 #include <asm/swiotlb.h>
56 #include <asm/rtas.h>
57
58 #include "mmu_decl.h"
59
60 #ifndef CPU_FTR_COHERENT_ICACHE
61 #define CPU_FTR_COHERENT_ICACHE 0 /* XXX for now */
62 #define CPU_FTR_NOEXECUTE 0
63 #endif
64
65 unsigned long long memory_limit;
66 bool init_mem_is_free;
67
68 #ifdef CONFIG_HIGHMEM
69 pte_t *kmap_pte;
70 EXPORT_SYMBOL(kmap_pte);
71 pgprot_t kmap_prot;
72 EXPORT_SYMBOL(kmap_prot);
73 #define TOP_ZONE ZONE_HIGHMEM
74
virt_to_kpte(unsigned long vaddr)75 static inline pte_t *virt_to_kpte(unsigned long vaddr)
76 {
77 return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
78 vaddr), vaddr), vaddr);
79 }
80 #else
81 #define TOP_ZONE ZONE_NORMAL
82 #endif
83
page_is_ram(unsigned long pfn)84 int page_is_ram(unsigned long pfn)
85 {
86 #ifndef CONFIG_PPC64 /* XXX for now */
87 return pfn < max_pfn;
88 #else
89 unsigned long paddr = (pfn << PAGE_SHIFT);
90 struct memblock_region *reg;
91
92 for_each_memblock(memory, reg)
93 if (paddr >= reg->base && paddr < (reg->base + reg->size))
94 return 1;
95 return 0;
96 #endif
97 }
98
phys_mem_access_prot(struct file * file,unsigned long pfn,unsigned long size,pgprot_t vma_prot)99 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
100 unsigned long size, pgprot_t vma_prot)
101 {
102 if (ppc_md.phys_mem_access_prot)
103 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
104
105 if (!page_is_ram(pfn))
106 vma_prot = pgprot_noncached(vma_prot);
107
108 return vma_prot;
109 }
110 EXPORT_SYMBOL(phys_mem_access_prot);
111
112 #ifdef CONFIG_MEMORY_HOTPLUG
113
114 #ifdef CONFIG_NUMA
memory_add_physaddr_to_nid(u64 start)115 int memory_add_physaddr_to_nid(u64 start)
116 {
117 return hot_add_scn_to_nid(start);
118 }
119 #endif
120
create_section_mapping(unsigned long start,unsigned long end)121 int __weak create_section_mapping(unsigned long start, unsigned long end)
122 {
123 return -ENODEV;
124 }
125
remove_section_mapping(unsigned long start,unsigned long end)126 int __weak remove_section_mapping(unsigned long start, unsigned long end)
127 {
128 return -ENODEV;
129 }
130
arch_add_memory(int nid,u64 start,u64 size,bool want_memblock)131 int arch_add_memory(int nid, u64 start, u64 size, bool want_memblock)
132 {
133 unsigned long start_pfn = start >> PAGE_SHIFT;
134 unsigned long nr_pages = size >> PAGE_SHIFT;
135 int rc;
136
137 resize_hpt_for_hotplug(memblock_phys_mem_size());
138
139 start = (unsigned long)__va(start);
140 rc = create_section_mapping(start, start + size);
141 if (rc) {
142 pr_warning(
143 "Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
144 start, start + size, rc);
145 return -EFAULT;
146 }
147 flush_inval_dcache_range(start, start + size);
148
149 return __add_pages(nid, start_pfn, nr_pages, want_memblock);
150 }
151
152 #ifdef CONFIG_MEMORY_HOTREMOVE
arch_remove_memory(u64 start,u64 size)153 int arch_remove_memory(u64 start, u64 size)
154 {
155 unsigned long start_pfn = start >> PAGE_SHIFT;
156 unsigned long nr_pages = size >> PAGE_SHIFT;
157 struct vmem_altmap *altmap;
158 struct page *page;
159 int ret;
160
161 /*
162 * If we have an altmap then we need to skip over any reserved PFNs
163 * when querying the zone.
164 */
165 page = pfn_to_page(start_pfn);
166 altmap = to_vmem_altmap((unsigned long) page);
167 if (altmap)
168 page += vmem_altmap_offset(altmap);
169
170 ret = __remove_pages(page_zone(page), start_pfn, nr_pages);
171 if (ret)
172 return ret;
173
174 /* Remove htab bolted mappings for this section of memory */
175 start = (unsigned long)__va(start);
176 flush_inval_dcache_range(start, start + size);
177 ret = remove_section_mapping(start, start + size);
178
179 /* Ensure all vmalloc mappings are flushed in case they also
180 * hit that section of memory
181 */
182 vm_unmap_aliases();
183
184 resize_hpt_for_hotplug(memblock_phys_mem_size());
185
186 return ret;
187 }
188 #endif
189 #endif /* CONFIG_MEMORY_HOTPLUG */
190
191 /*
192 * walk_memory_resource() needs to make sure there is no holes in a given
193 * memory range. PPC64 does not maintain the memory layout in /proc/iomem.
194 * Instead it maintains it in memblock.memory structures. Walk through the
195 * memory regions, find holes and callback for contiguous regions.
196 */
197 int
walk_system_ram_range(unsigned long start_pfn,unsigned long nr_pages,void * arg,int (* func)(unsigned long,unsigned long,void *))198 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
199 void *arg, int (*func)(unsigned long, unsigned long, void *))
200 {
201 struct memblock_region *reg;
202 unsigned long end_pfn = start_pfn + nr_pages;
203 unsigned long tstart, tend;
204 int ret = -1;
205
206 for_each_memblock(memory, reg) {
207 tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
208 tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
209 if (tstart >= tend)
210 continue;
211 ret = (*func)(tstart, tend - tstart, arg);
212 if (ret)
213 break;
214 }
215 return ret;
216 }
217 EXPORT_SYMBOL_GPL(walk_system_ram_range);
218
219 #ifndef CONFIG_NEED_MULTIPLE_NODES
initmem_init(void)220 void __init initmem_init(void)
221 {
222 max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
223 min_low_pfn = MEMORY_START >> PAGE_SHIFT;
224 #ifdef CONFIG_HIGHMEM
225 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
226 #endif
227
228 /* Place all memblock_regions in the same node and merge contiguous
229 * memblock_regions
230 */
231 memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
232
233 /* XXX need to clip this if using highmem? */
234 sparse_memory_present_with_active_regions(0);
235 sparse_init();
236 }
237
238 /* mark pages that don't exist as nosave */
mark_nonram_nosave(void)239 static int __init mark_nonram_nosave(void)
240 {
241 struct memblock_region *reg, *prev = NULL;
242
243 for_each_memblock(memory, reg) {
244 if (prev &&
245 memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
246 register_nosave_region(memblock_region_memory_end_pfn(prev),
247 memblock_region_memory_base_pfn(reg));
248 prev = reg;
249 }
250 return 0;
251 }
252 #else /* CONFIG_NEED_MULTIPLE_NODES */
mark_nonram_nosave(void)253 static int __init mark_nonram_nosave(void)
254 {
255 return 0;
256 }
257 #endif
258
259 static bool zone_limits_final;
260
261 /*
262 * The memory zones past TOP_ZONE are managed by generic mm code.
263 * These should be set to zero since that's what every other
264 * architecture does.
265 */
266 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
267 [0 ... TOP_ZONE ] = ~0UL,
268 [TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
269 };
270
271 /*
272 * Restrict the specified zone and all more restrictive zones
273 * to be below the specified pfn. May not be called after
274 * paging_init().
275 */
limit_zone_pfn(enum zone_type zone,unsigned long pfn_limit)276 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
277 {
278 int i;
279
280 if (WARN_ON(zone_limits_final))
281 return;
282
283 for (i = zone; i >= 0; i--) {
284 if (max_zone_pfns[i] > pfn_limit)
285 max_zone_pfns[i] = pfn_limit;
286 }
287 }
288
289 /*
290 * Find the least restrictive zone that is entirely below the
291 * specified pfn limit. Returns < 0 if no suitable zone is found.
292 *
293 * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
294 * systems -- the DMA limit can be higher than any possible real pfn.
295 */
dma_pfn_limit_to_zone(u64 pfn_limit)296 int dma_pfn_limit_to_zone(u64 pfn_limit)
297 {
298 int i;
299
300 for (i = TOP_ZONE; i >= 0; i--) {
301 if (max_zone_pfns[i] <= pfn_limit)
302 return i;
303 }
304
305 return -EPERM;
306 }
307
308 /*
309 * paging_init() sets up the page tables - in fact we've already done this.
310 */
paging_init(void)311 void __init paging_init(void)
312 {
313 unsigned long long total_ram = memblock_phys_mem_size();
314 phys_addr_t top_of_ram = memblock_end_of_DRAM();
315
316 #ifdef CONFIG_PPC32
317 unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
318 unsigned long end = __fix_to_virt(FIX_HOLE);
319
320 for (; v < end; v += PAGE_SIZE)
321 map_kernel_page(v, 0, 0); /* XXX gross */
322 #endif
323
324 #ifdef CONFIG_HIGHMEM
325 map_kernel_page(PKMAP_BASE, 0, 0); /* XXX gross */
326 pkmap_page_table = virt_to_kpte(PKMAP_BASE);
327
328 kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
329 kmap_prot = PAGE_KERNEL;
330 #endif /* CONFIG_HIGHMEM */
331
332 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
333 (unsigned long long)top_of_ram, total_ram);
334 printk(KERN_DEBUG "Memory hole size: %ldMB\n",
335 (long int)((top_of_ram - total_ram) >> 20));
336
337 #ifdef CONFIG_HIGHMEM
338 limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
339 #endif
340 limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
341 zone_limits_final = true;
342 free_area_init_nodes(max_zone_pfns);
343
344 mark_nonram_nosave();
345 }
346
mem_init(void)347 void __init mem_init(void)
348 {
349 /*
350 * book3s is limited to 16 page sizes due to encoding this in
351 * a 4-bit field for slices.
352 */
353 BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
354
355 #ifdef CONFIG_SWIOTLB
356 /*
357 * Some platforms (e.g. 85xx) limit DMA-able memory way below
358 * 4G. We force memblock to bottom-up mode to ensure that the
359 * memory allocated in swiotlb_init() is DMA-able.
360 * As it's the last memblock allocation, no need to reset it
361 * back to to-down.
362 */
363 memblock_set_bottom_up(true);
364 swiotlb_init(0);
365 #endif
366
367 high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
368 set_max_mapnr(max_pfn);
369 free_all_bootmem();
370
371 #ifdef CONFIG_HIGHMEM
372 {
373 unsigned long pfn, highmem_mapnr;
374
375 highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
376 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
377 phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
378 struct page *page = pfn_to_page(pfn);
379 if (!memblock_is_reserved(paddr))
380 free_highmem_page(page);
381 }
382 }
383 #endif /* CONFIG_HIGHMEM */
384
385 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
386 /*
387 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
388 * functions.... do it here for the non-smp case.
389 */
390 per_cpu(next_tlbcam_idx, smp_processor_id()) =
391 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
392 #endif
393
394 mem_init_print_info(NULL);
395 #ifdef CONFIG_PPC32
396 pr_info("Kernel virtual memory layout:\n");
397 pr_info(" * 0x%08lx..0x%08lx : fixmap\n", FIXADDR_START, FIXADDR_TOP);
398 #ifdef CONFIG_HIGHMEM
399 pr_info(" * 0x%08lx..0x%08lx : highmem PTEs\n",
400 PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
401 #endif /* CONFIG_HIGHMEM */
402 #ifdef CONFIG_NOT_COHERENT_CACHE
403 pr_info(" * 0x%08lx..0x%08lx : consistent mem\n",
404 IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
405 #endif /* CONFIG_NOT_COHERENT_CACHE */
406 pr_info(" * 0x%08lx..0x%08lx : early ioremap\n",
407 ioremap_bot, IOREMAP_TOP);
408 pr_info(" * 0x%08lx..0x%08lx : vmalloc & ioremap\n",
409 VMALLOC_START, VMALLOC_END);
410 #endif /* CONFIG_PPC32 */
411 }
412
free_initmem(void)413 void free_initmem(void)
414 {
415 ppc_md.progress = ppc_printk_progress;
416 mark_initmem_nx();
417 init_mem_is_free = true;
418 free_initmem_default(POISON_FREE_INITMEM);
419 }
420
421 #ifdef CONFIG_BLK_DEV_INITRD
free_initrd_mem(unsigned long start,unsigned long end)422 void __init free_initrd_mem(unsigned long start, unsigned long end)
423 {
424 free_reserved_area((void *)start, (void *)end, -1, "initrd");
425 }
426 #endif
427
428 /*
429 * This is called when a page has been modified by the kernel.
430 * It just marks the page as not i-cache clean. We do the i-cache
431 * flush later when the page is given to a user process, if necessary.
432 */
flush_dcache_page(struct page * page)433 void flush_dcache_page(struct page *page)
434 {
435 if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
436 return;
437 /* avoid an atomic op if possible */
438 if (test_bit(PG_arch_1, &page->flags))
439 clear_bit(PG_arch_1, &page->flags);
440 }
441 EXPORT_SYMBOL(flush_dcache_page);
442
flush_dcache_icache_page(struct page * page)443 void flush_dcache_icache_page(struct page *page)
444 {
445 #ifdef CONFIG_HUGETLB_PAGE
446 if (PageCompound(page)) {
447 flush_dcache_icache_hugepage(page);
448 return;
449 }
450 #endif
451 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
452 /* On 8xx there is no need to kmap since highmem is not supported */
453 __flush_dcache_icache(page_address(page));
454 #else
455 if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
456 void *start = kmap_atomic(page);
457 __flush_dcache_icache(start);
458 kunmap_atomic(start);
459 } else {
460 __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
461 }
462 #endif
463 }
464 EXPORT_SYMBOL(flush_dcache_icache_page);
465
clear_user_page(void * page,unsigned long vaddr,struct page * pg)466 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
467 {
468 clear_page(page);
469
470 /*
471 * We shouldn't have to do this, but some versions of glibc
472 * require it (ld.so assumes zero filled pages are icache clean)
473 * - Anton
474 */
475 flush_dcache_page(pg);
476 }
477 EXPORT_SYMBOL(clear_user_page);
478
copy_user_page(void * vto,void * vfrom,unsigned long vaddr,struct page * pg)479 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
480 struct page *pg)
481 {
482 copy_page(vto, vfrom);
483
484 /*
485 * We should be able to use the following optimisation, however
486 * there are two problems.
487 * Firstly a bug in some versions of binutils meant PLT sections
488 * were not marked executable.
489 * Secondly the first word in the GOT section is blrl, used
490 * to establish the GOT address. Until recently the GOT was
491 * not marked executable.
492 * - Anton
493 */
494 #if 0
495 if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
496 return;
497 #endif
498
499 flush_dcache_page(pg);
500 }
501
flush_icache_user_range(struct vm_area_struct * vma,struct page * page,unsigned long addr,int len)502 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
503 unsigned long addr, int len)
504 {
505 unsigned long maddr;
506
507 maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
508 flush_icache_range(maddr, maddr + len);
509 kunmap(page);
510 }
511 EXPORT_SYMBOL(flush_icache_user_range);
512
513 /*
514 * This is called at the end of handling a user page fault, when the
515 * fault has been handled by updating a PTE in the linux page tables.
516 * We use it to preload an HPTE into the hash table corresponding to
517 * the updated linux PTE.
518 *
519 * This must always be called with the pte lock held.
520 */
update_mmu_cache(struct vm_area_struct * vma,unsigned long address,pte_t * ptep)521 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
522 pte_t *ptep)
523 {
524 #ifdef CONFIG_PPC_STD_MMU
525 /*
526 * We don't need to worry about _PAGE_PRESENT here because we are
527 * called with either mm->page_table_lock held or ptl lock held
528 */
529 unsigned long access, trap;
530
531 if (radix_enabled())
532 return;
533
534 /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
535 if (!pte_young(*ptep) || address >= TASK_SIZE)
536 return;
537
538 /* We try to figure out if we are coming from an instruction
539 * access fault and pass that down to __hash_page so we avoid
540 * double-faulting on execution of fresh text. We have to test
541 * for regs NULL since init will get here first thing at boot
542 *
543 * We also avoid filling the hash if not coming from a fault
544 */
545
546 trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
547 switch (trap) {
548 case 0x300:
549 access = 0UL;
550 break;
551 case 0x400:
552 access = _PAGE_EXEC;
553 break;
554 default:
555 return;
556 }
557
558 hash_preload(vma->vm_mm, address, access, trap);
559 #endif /* CONFIG_PPC_STD_MMU */
560 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
561 && defined(CONFIG_HUGETLB_PAGE)
562 if (is_vm_hugetlb_page(vma))
563 book3e_hugetlb_preload(vma, address, *ptep);
564 #endif
565 }
566
567 /*
568 * System memory should not be in /proc/iomem but various tools expect it
569 * (eg kdump).
570 */
add_system_ram_resources(void)571 static int __init add_system_ram_resources(void)
572 {
573 struct memblock_region *reg;
574
575 for_each_memblock(memory, reg) {
576 struct resource *res;
577 unsigned long base = reg->base;
578 unsigned long size = reg->size;
579
580 res = kzalloc(sizeof(struct resource), GFP_KERNEL);
581 WARN_ON(!res);
582
583 if (res) {
584 res->name = "System RAM";
585 res->start = base;
586 res->end = base + size - 1;
587 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
588 WARN_ON(request_resource(&iomem_resource, res) < 0);
589 }
590 }
591
592 return 0;
593 }
594 subsys_initcall(add_system_ram_resources);
595
596 #ifdef CONFIG_STRICT_DEVMEM
597 /*
598 * devmem_is_allowed(): check to see if /dev/mem access to a certain address
599 * is valid. The argument is a physical page number.
600 *
601 * Access has to be given to non-kernel-ram areas as well, these contain the
602 * PCI mmio resources as well as potential bios/acpi data regions.
603 */
devmem_is_allowed(unsigned long pfn)604 int devmem_is_allowed(unsigned long pfn)
605 {
606 if (page_is_rtas_user_buf(pfn))
607 return 1;
608 if (iomem_is_exclusive(PFN_PHYS(pfn)))
609 return 0;
610 if (!page_is_ram(pfn))
611 return 1;
612 return 0;
613 }
614 #endif /* CONFIG_STRICT_DEVMEM */
615