xref: /linux/arch/powerpc/mm/mem.c (revision 5f4123be3cdb1dbd77fa9d6d2bb96bb9689a0a19)
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/module.h>
21 #include <linux/sched.h>
22 #include <linux/kernel.h>
23 #include <linux/errno.h>
24 #include <linux/string.h>
25 #include <linux/types.h>
26 #include <linux/mm.h>
27 #include <linux/stddef.h>
28 #include <linux/init.h>
29 #include <linux/bootmem.h>
30 #include <linux/highmem.h>
31 #include <linux/initrd.h>
32 #include <linux/pagemap.h>
33 #include <linux/suspend.h>
34 #include <linux/lmb.h>
35 
36 #include <asm/pgalloc.h>
37 #include <asm/prom.h>
38 #include <asm/io.h>
39 #include <asm/mmu_context.h>
40 #include <asm/pgtable.h>
41 #include <asm/mmu.h>
42 #include <asm/smp.h>
43 #include <asm/machdep.h>
44 #include <asm/btext.h>
45 #include <asm/tlb.h>
46 #include <asm/sections.h>
47 #include <asm/sparsemem.h>
48 #include <asm/vdso.h>
49 #include <asm/fixmap.h>
50 
51 #include "mmu_decl.h"
52 
53 #ifndef CPU_FTR_COHERENT_ICACHE
54 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
55 #define CPU_FTR_NOEXECUTE	0
56 #endif
57 
58 int init_bootmem_done;
59 int mem_init_done;
60 unsigned long memory_limit;
61 
62 #ifdef CONFIG_HIGHMEM
63 pte_t *kmap_pte;
64 pgprot_t kmap_prot;
65 
66 EXPORT_SYMBOL(kmap_prot);
67 EXPORT_SYMBOL(kmap_pte);
68 
69 static inline pte_t *virt_to_kpte(unsigned long vaddr)
70 {
71 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
72 			vaddr), vaddr), vaddr);
73 }
74 #endif
75 
76 int page_is_ram(unsigned long pfn)
77 {
78 #ifndef CONFIG_PPC64	/* XXX for now */
79 	return pfn < max_pfn;
80 #else
81 	unsigned long paddr = (pfn << PAGE_SHIFT);
82 	int i;
83 	for (i=0; i < lmb.memory.cnt; i++) {
84 		unsigned long base;
85 
86 		base = lmb.memory.region[i].base;
87 
88 		if ((paddr >= base) &&
89 			(paddr < (base + lmb.memory.region[i].size))) {
90 			return 1;
91 		}
92 	}
93 
94 	return 0;
95 #endif
96 }
97 
98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
99 			      unsigned long size, pgprot_t vma_prot)
100 {
101 	if (ppc_md.phys_mem_access_prot)
102 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
103 
104 	if (!page_is_ram(pfn))
105 		vma_prot = __pgprot(pgprot_val(vma_prot)
106 				    | _PAGE_GUARDED | _PAGE_NO_CACHE);
107 	return vma_prot;
108 }
109 EXPORT_SYMBOL(phys_mem_access_prot);
110 
111 #ifdef CONFIG_MEMORY_HOTPLUG
112 
113 #ifdef CONFIG_NUMA
114 int memory_add_physaddr_to_nid(u64 start)
115 {
116 	return hot_add_scn_to_nid(start);
117 }
118 #endif
119 
120 int arch_add_memory(int nid, u64 start, u64 size)
121 {
122 	struct pglist_data *pgdata;
123 	struct zone *zone;
124 	unsigned long start_pfn = start >> PAGE_SHIFT;
125 	unsigned long nr_pages = size >> PAGE_SHIFT;
126 
127 	pgdata = NODE_DATA(nid);
128 
129 	start = (unsigned long)__va(start);
130 	create_section_mapping(start, start + size);
131 
132 	/* this should work for most non-highmem platforms */
133 	zone = pgdata->node_zones;
134 
135 	return __add_pages(zone, start_pfn, nr_pages);
136 }
137 
138 #ifdef CONFIG_MEMORY_HOTREMOVE
139 int remove_memory(u64 start, u64 size)
140 {
141 	unsigned long start_pfn, end_pfn;
142 	int ret;
143 
144 	start_pfn = start >> PAGE_SHIFT;
145 	end_pfn = start_pfn + (size >> PAGE_SHIFT);
146 	ret = offline_pages(start_pfn, end_pfn, 120 * HZ);
147 	if (ret)
148 		goto out;
149 	/* Arch-specific calls go here - next patch */
150 out:
151 	return ret;
152 }
153 #endif /* CONFIG_MEMORY_HOTREMOVE */
154 #endif /* CONFIG_MEMORY_HOTPLUG */
155 
156 /*
157  * walk_memory_resource() needs to make sure there is no holes in a given
158  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
159  * Instead it maintains it in lmb.memory structures.  Walk through the
160  * memory regions, find holes and callback for contiguous regions.
161  */
162 int
163 walk_memory_resource(unsigned long start_pfn, unsigned long nr_pages, void *arg,
164 			int (*func)(unsigned long, unsigned long, void *))
165 {
166 	struct lmb_property res;
167 	unsigned long pfn, len;
168 	u64 end;
169 	int ret = -1;
170 
171 	res.base = (u64) start_pfn << PAGE_SHIFT;
172 	res.size = (u64) nr_pages << PAGE_SHIFT;
173 
174 	end = res.base + res.size - 1;
175 	while ((res.base < end) && (lmb_find(&res) >= 0)) {
176 		pfn = (unsigned long)(res.base >> PAGE_SHIFT);
177 		len = (unsigned long)(res.size >> PAGE_SHIFT);
178 		ret = (*func)(pfn, len, arg);
179 		if (ret)
180 			break;
181 		res.base += (res.size + 1);
182 		res.size = (end - res.base + 1);
183 	}
184 	return ret;
185 }
186 EXPORT_SYMBOL_GPL(walk_memory_resource);
187 
188 /*
189  * Initialize the bootmem system and give it all the memory we
190  * have available.  If we are using highmem, we only put the
191  * lowmem into the bootmem system.
192  */
193 #ifndef CONFIG_NEED_MULTIPLE_NODES
194 void __init do_init_bootmem(void)
195 {
196 	unsigned long i;
197 	unsigned long start, bootmap_pages;
198 	unsigned long total_pages;
199 	int boot_mapsize;
200 
201 	max_low_pfn = max_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
202 	total_pages = (lmb_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
203 #ifdef CONFIG_HIGHMEM
204 	total_pages = total_lowmem >> PAGE_SHIFT;
205 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
206 #endif
207 
208 	/*
209 	 * Find an area to use for the bootmem bitmap.  Calculate the size of
210 	 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
211 	 * Add 1 additional page in case the address isn't page-aligned.
212 	 */
213 	bootmap_pages = bootmem_bootmap_pages(total_pages);
214 
215 	start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
216 
217 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
218 	boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
219 
220 	/* Add active regions with valid PFNs */
221 	for (i = 0; i < lmb.memory.cnt; i++) {
222 		unsigned long start_pfn, end_pfn;
223 		start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
224 		end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
225 		add_active_range(0, start_pfn, end_pfn);
226 	}
227 
228 	/* Add all physical memory to the bootmem map, mark each area
229 	 * present.
230 	 */
231 #ifdef CONFIG_HIGHMEM
232 	free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
233 
234 	/* reserve the sections we're already using */
235 	for (i = 0; i < lmb.reserved.cnt; i++) {
236 		unsigned long addr = lmb.reserved.region[i].base +
237 				     lmb_size_bytes(&lmb.reserved, i) - 1;
238 		if (addr < lowmem_end_addr)
239 			reserve_bootmem(lmb.reserved.region[i].base,
240 					lmb_size_bytes(&lmb.reserved, i),
241 					BOOTMEM_DEFAULT);
242 		else if (lmb.reserved.region[i].base < lowmem_end_addr) {
243 			unsigned long adjusted_size = lowmem_end_addr -
244 				      lmb.reserved.region[i].base;
245 			reserve_bootmem(lmb.reserved.region[i].base,
246 					adjusted_size, BOOTMEM_DEFAULT);
247 		}
248 	}
249 #else
250 	free_bootmem_with_active_regions(0, max_pfn);
251 
252 	/* reserve the sections we're already using */
253 	for (i = 0; i < lmb.reserved.cnt; i++)
254 		reserve_bootmem(lmb.reserved.region[i].base,
255 				lmb_size_bytes(&lmb.reserved, i),
256 				BOOTMEM_DEFAULT);
257 
258 #endif
259 	/* XXX need to clip this if using highmem? */
260 	sparse_memory_present_with_active_regions(0);
261 
262 	init_bootmem_done = 1;
263 }
264 
265 /* mark pages that don't exist as nosave */
266 static int __init mark_nonram_nosave(void)
267 {
268 	unsigned long lmb_next_region_start_pfn,
269 		      lmb_region_max_pfn;
270 	int i;
271 
272 	for (i = 0; i < lmb.memory.cnt - 1; i++) {
273 		lmb_region_max_pfn =
274 			(lmb.memory.region[i].base >> PAGE_SHIFT) +
275 			(lmb.memory.region[i].size >> PAGE_SHIFT);
276 		lmb_next_region_start_pfn =
277 			lmb.memory.region[i+1].base >> PAGE_SHIFT;
278 
279 		if (lmb_region_max_pfn < lmb_next_region_start_pfn)
280 			register_nosave_region(lmb_region_max_pfn,
281 					       lmb_next_region_start_pfn);
282 	}
283 
284 	return 0;
285 }
286 
287 /*
288  * paging_init() sets up the page tables - in fact we've already done this.
289  */
290 void __init paging_init(void)
291 {
292 	unsigned long total_ram = lmb_phys_mem_size();
293 	phys_addr_t top_of_ram = lmb_end_of_DRAM();
294 	unsigned long max_zone_pfns[MAX_NR_ZONES];
295 
296 #ifdef CONFIG_PPC32
297 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
298 	unsigned long end = __fix_to_virt(FIX_HOLE);
299 
300 	for (; v < end; v += PAGE_SIZE)
301 		map_page(v, 0, 0); /* XXX gross */
302 #endif
303 
304 #ifdef CONFIG_HIGHMEM
305 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
306 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
307 
308 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
309 	kmap_prot = PAGE_KERNEL;
310 #endif /* CONFIG_HIGHMEM */
311 
312 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
313 	       (unsigned long long)top_of_ram, total_ram);
314 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
315 	       (long int)((top_of_ram - total_ram) >> 20));
316 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
317 #ifdef CONFIG_HIGHMEM
318 	max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
319 	max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
320 #else
321 	max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
322 #endif
323 	free_area_init_nodes(max_zone_pfns);
324 
325 	mark_nonram_nosave();
326 }
327 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
328 
329 void __init mem_init(void)
330 {
331 #ifdef CONFIG_NEED_MULTIPLE_NODES
332 	int nid;
333 #endif
334 	pg_data_t *pgdat;
335 	unsigned long i;
336 	struct page *page;
337 	unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
338 
339 	num_physpages = lmb.memory.size >> PAGE_SHIFT;
340 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
341 
342 #ifdef CONFIG_NEED_MULTIPLE_NODES
343         for_each_online_node(nid) {
344 		if (NODE_DATA(nid)->node_spanned_pages != 0) {
345 			printk("freeing bootmem node %d\n", nid);
346 			totalram_pages +=
347 				free_all_bootmem_node(NODE_DATA(nid));
348 		}
349 	}
350 #else
351 	max_mapnr = max_pfn;
352 	totalram_pages += free_all_bootmem();
353 #endif
354 	for_each_online_pgdat(pgdat) {
355 		for (i = 0; i < pgdat->node_spanned_pages; i++) {
356 			if (!pfn_valid(pgdat->node_start_pfn + i))
357 				continue;
358 			page = pgdat_page_nr(pgdat, i);
359 			if (PageReserved(page))
360 				reservedpages++;
361 		}
362 	}
363 
364 	codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
365 	datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
366 	initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
367 	bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
368 
369 #ifdef CONFIG_HIGHMEM
370 	{
371 		unsigned long pfn, highmem_mapnr;
372 
373 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
374 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
375 			struct page *page = pfn_to_page(pfn);
376 			if (lmb_is_reserved(pfn << PAGE_SHIFT))
377 				continue;
378 			ClearPageReserved(page);
379 			init_page_count(page);
380 			__free_page(page);
381 			totalhigh_pages++;
382 			reservedpages--;
383 		}
384 		totalram_pages += totalhigh_pages;
385 		printk(KERN_DEBUG "High memory: %luk\n",
386 		       totalhigh_pages << (PAGE_SHIFT-10));
387 	}
388 #endif /* CONFIG_HIGHMEM */
389 
390 	printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
391 	       "%luk reserved, %luk data, %luk bss, %luk init)\n",
392 		(unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
393 		num_physpages << (PAGE_SHIFT-10),
394 		codesize >> 10,
395 		reservedpages << (PAGE_SHIFT-10),
396 		datasize >> 10,
397 		bsssize >> 10,
398 		initsize >> 10);
399 
400 	mem_init_done = 1;
401 }
402 
403 /*
404  * This is called when a page has been modified by the kernel.
405  * It just marks the page as not i-cache clean.  We do the i-cache
406  * flush later when the page is given to a user process, if necessary.
407  */
408 void flush_dcache_page(struct page *page)
409 {
410 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
411 		return;
412 	/* avoid an atomic op if possible */
413 	if (test_bit(PG_arch_1, &page->flags))
414 		clear_bit(PG_arch_1, &page->flags);
415 }
416 EXPORT_SYMBOL(flush_dcache_page);
417 
418 void flush_dcache_icache_page(struct page *page)
419 {
420 #ifdef CONFIG_BOOKE
421 	void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
422 	__flush_dcache_icache(start);
423 	kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
424 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
425 	/* On 8xx there is no need to kmap since highmem is not supported */
426 	__flush_dcache_icache(page_address(page));
427 #else
428 	__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
429 #endif
430 
431 }
432 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
433 {
434 	clear_page(page);
435 
436 	/*
437 	 * We shouldnt have to do this, but some versions of glibc
438 	 * require it (ld.so assumes zero filled pages are icache clean)
439 	 * - Anton
440 	 */
441 	flush_dcache_page(pg);
442 }
443 EXPORT_SYMBOL(clear_user_page);
444 
445 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
446 		    struct page *pg)
447 {
448 	copy_page(vto, vfrom);
449 
450 	/*
451 	 * We should be able to use the following optimisation, however
452 	 * there are two problems.
453 	 * Firstly a bug in some versions of binutils meant PLT sections
454 	 * were not marked executable.
455 	 * Secondly the first word in the GOT section is blrl, used
456 	 * to establish the GOT address. Until recently the GOT was
457 	 * not marked executable.
458 	 * - Anton
459 	 */
460 #if 0
461 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
462 		return;
463 #endif
464 
465 	flush_dcache_page(pg);
466 }
467 
468 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
469 			     unsigned long addr, int len)
470 {
471 	unsigned long maddr;
472 
473 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
474 	flush_icache_range(maddr, maddr + len);
475 	kunmap(page);
476 }
477 EXPORT_SYMBOL(flush_icache_user_range);
478 
479 /*
480  * This is called at the end of handling a user page fault, when the
481  * fault has been handled by updating a PTE in the linux page tables.
482  * We use it to preload an HPTE into the hash table corresponding to
483  * the updated linux PTE.
484  *
485  * This must always be called with the pte lock held.
486  */
487 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
488 		      pte_t pte)
489 {
490 #ifdef CONFIG_PPC_STD_MMU
491 	unsigned long access = 0, trap;
492 #endif
493 	unsigned long pfn = pte_pfn(pte);
494 
495 	/* handle i-cache coherency */
496 	if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
497 	    !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
498 	    pfn_valid(pfn)) {
499 		struct page *page = pfn_to_page(pfn);
500 #ifdef CONFIG_8xx
501 		/* On 8xx, cache control instructions (particularly
502 		 * "dcbst" from flush_dcache_icache) fault as write
503 		 * operation if there is an unpopulated TLB entry
504 		 * for the address in question. To workaround that,
505 		 * we invalidate the TLB here, thus avoiding dcbst
506 		 * misbehaviour.
507 		 */
508 		_tlbie(address, 0 /* 8xx doesn't care about PID */);
509 #endif
510 		/* The _PAGE_USER test should really be _PAGE_EXEC, but
511 		 * older glibc versions execute some code from no-exec
512 		 * pages, which for now we are supporting.  If exec-only
513 		 * pages are ever implemented, this will have to change.
514 		 */
515 		if (!PageReserved(page) && (pte_val(pte) & _PAGE_USER)
516 		    && !test_bit(PG_arch_1, &page->flags)) {
517 			if (vma->vm_mm == current->active_mm) {
518 				__flush_dcache_icache((void *) address);
519 			} else
520 				flush_dcache_icache_page(page);
521 			set_bit(PG_arch_1, &page->flags);
522 		}
523 	}
524 
525 #ifdef CONFIG_PPC_STD_MMU
526 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
527 	if (!pte_young(pte) || address >= TASK_SIZE)
528 		return;
529 
530 	/* We try to figure out if we are coming from an instruction
531 	 * access fault and pass that down to __hash_page so we avoid
532 	 * double-faulting on execution of fresh text. We have to test
533 	 * for regs NULL since init will get here first thing at boot
534 	 *
535 	 * We also avoid filling the hash if not coming from a fault
536 	 */
537 	if (current->thread.regs == NULL)
538 		return;
539 	trap = TRAP(current->thread.regs);
540 	if (trap == 0x400)
541 		access |= _PAGE_EXEC;
542 	else if (trap != 0x300)
543 		return;
544 	hash_preload(vma->vm_mm, address, access, trap);
545 #endif /* CONFIG_PPC_STD_MMU */
546 }
547