xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 7dd65feb) 
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 #include <linux/hugetlb.h>
36 
37 #include <asm/pgalloc.h>
38 #include <asm/prom.h>
39 #include <asm/io.h>
40 #include <asm/mmu_context.h>
41 #include <asm/pgtable.h>
42 #include <asm/mmu.h>
43 #include <asm/smp.h>
44 #include <asm/machdep.h>
45 #include <asm/btext.h>
46 #include <asm/tlb.h>
47 #include <asm/sections.h>
48 #include <asm/sparsemem.h>
49 #include <asm/vdso.h>
50 #include <asm/fixmap.h>
51 
52 #include "mmu_decl.h"
53 
54 #ifndef CPU_FTR_COHERENT_ICACHE
55 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
56 #define CPU_FTR_NOEXECUTE	0
57 #endif
58 
59 int init_bootmem_done;
60 int mem_init_done;
61 phys_addr_t memory_limit;
62 
63 #ifdef CONFIG_HIGHMEM
64 pte_t *kmap_pte;
65 pgprot_t kmap_prot;
66 
67 EXPORT_SYMBOL(kmap_prot);
68 EXPORT_SYMBOL(kmap_pte);
69 
70 static inline pte_t *virt_to_kpte(unsigned long vaddr)
71 {
72 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
73 			vaddr), vaddr), vaddr);
74 }
75 #endif
76 
77 int page_is_ram(unsigned long pfn)
78 {
79 #ifndef CONFIG_PPC64	/* XXX for now */
80 	return pfn < max_pfn;
81 #else
82 	unsigned long paddr = (pfn << PAGE_SHIFT);
83 	int i;
84 	for (i=0; i < lmb.memory.cnt; i++) {
85 		unsigned long base;
86 
87 		base = lmb.memory.region[i].base;
88 
89 		if ((paddr >= base) &&
90 			(paddr < (base + lmb.memory.region[i].size))) {
91 			return 1;
92 		}
93 	}
94 
95 	return 0;
96 #endif
97 }
98 
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
115 int memory_add_physaddr_to_nid(u64 start)
116 {
117 	return hot_add_scn_to_nid(start);
118 }
119 #endif
120 
121 int arch_add_memory(int nid, u64 start, u64 size)
122 {
123 	struct pglist_data *pgdata;
124 	struct zone *zone;
125 	unsigned long start_pfn = start >> PAGE_SHIFT;
126 	unsigned long nr_pages = size >> PAGE_SHIFT;
127 
128 	pgdata = NODE_DATA(nid);
129 
130 	start = (unsigned long)__va(start);
131 	create_section_mapping(start, start + size);
132 
133 	/* this should work for most non-highmem platforms */
134 	zone = pgdata->node_zones;
135 
136 	return __add_pages(nid, zone, start_pfn, nr_pages);
137 }
138 #endif /* CONFIG_MEMORY_HOTPLUG */
139 
140 /*
141  * walk_memory_resource() needs to make sure there is no holes in a given
142  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
143  * Instead it maintains it in lmb.memory structures.  Walk through the
144  * memory regions, find holes and callback for contiguous regions.
145  */
146 int
147 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
148 		void *arg, int (*func)(unsigned long, unsigned long, void *))
149 {
150 	struct lmb_property res;
151 	unsigned long pfn, len;
152 	u64 end;
153 	int ret = -1;
154 
155 	res.base = (u64) start_pfn << PAGE_SHIFT;
156 	res.size = (u64) nr_pages << PAGE_SHIFT;
157 
158 	end = res.base + res.size - 1;
159 	while ((res.base < end) && (lmb_find(&res) >= 0)) {
160 		pfn = (unsigned long)(res.base >> PAGE_SHIFT);
161 		len = (unsigned long)(res.size >> PAGE_SHIFT);
162 		ret = (*func)(pfn, len, arg);
163 		if (ret)
164 			break;
165 		res.base += (res.size + 1);
166 		res.size = (end - res.base + 1);
167 	}
168 	return ret;
169 }
170 EXPORT_SYMBOL_GPL(walk_system_ram_range);
171 
172 /*
173  * Initialize the bootmem system and give it all the memory we
174  * have available.  If we are using highmem, we only put the
175  * lowmem into the bootmem system.
176  */
177 #ifndef CONFIG_NEED_MULTIPLE_NODES
178 void __init do_init_bootmem(void)
179 {
180 	unsigned long i;
181 	unsigned long start, bootmap_pages;
182 	unsigned long total_pages;
183 	int boot_mapsize;
184 
185 	max_low_pfn = max_pfn = lmb_end_of_DRAM() >> PAGE_SHIFT;
186 	total_pages = (lmb_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
187 #ifdef CONFIG_HIGHMEM
188 	total_pages = total_lowmem >> PAGE_SHIFT;
189 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
190 #endif
191 
192 	/*
193 	 * Find an area to use for the bootmem bitmap.  Calculate the size of
194 	 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
195 	 * Add 1 additional page in case the address isn't page-aligned.
196 	 */
197 	bootmap_pages = bootmem_bootmap_pages(total_pages);
198 
199 	start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
200 
201 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
202 	boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
203 
204 	/* Add active regions with valid PFNs */
205 	for (i = 0; i < lmb.memory.cnt; i++) {
206 		unsigned long start_pfn, end_pfn;
207 		start_pfn = lmb.memory.region[i].base >> PAGE_SHIFT;
208 		end_pfn = start_pfn + lmb_size_pages(&lmb.memory, i);
209 		add_active_range(0, start_pfn, end_pfn);
210 	}
211 
212 	/* Add all physical memory to the bootmem map, mark each area
213 	 * present.
214 	 */
215 #ifdef CONFIG_HIGHMEM
216 	free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
217 
218 	/* reserve the sections we're already using */
219 	for (i = 0; i < lmb.reserved.cnt; i++) {
220 		unsigned long addr = lmb.reserved.region[i].base +
221 				     lmb_size_bytes(&lmb.reserved, i) - 1;
222 		if (addr < lowmem_end_addr)
223 			reserve_bootmem(lmb.reserved.region[i].base,
224 					lmb_size_bytes(&lmb.reserved, i),
225 					BOOTMEM_DEFAULT);
226 		else if (lmb.reserved.region[i].base < lowmem_end_addr) {
227 			unsigned long adjusted_size = lowmem_end_addr -
228 				      lmb.reserved.region[i].base;
229 			reserve_bootmem(lmb.reserved.region[i].base,
230 					adjusted_size, BOOTMEM_DEFAULT);
231 		}
232 	}
233 #else
234 	free_bootmem_with_active_regions(0, max_pfn);
235 
236 	/* reserve the sections we're already using */
237 	for (i = 0; i < lmb.reserved.cnt; i++)
238 		reserve_bootmem(lmb.reserved.region[i].base,
239 				lmb_size_bytes(&lmb.reserved, i),
240 				BOOTMEM_DEFAULT);
241 
242 #endif
243 	/* XXX need to clip this if using highmem? */
244 	sparse_memory_present_with_active_regions(0);
245 
246 	init_bootmem_done = 1;
247 }
248 
249 /* mark pages that don't exist as nosave */
250 static int __init mark_nonram_nosave(void)
251 {
252 	unsigned long lmb_next_region_start_pfn,
253 		      lmb_region_max_pfn;
254 	int i;
255 
256 	for (i = 0; i < lmb.memory.cnt - 1; i++) {
257 		lmb_region_max_pfn =
258 			(lmb.memory.region[i].base >> PAGE_SHIFT) +
259 			(lmb.memory.region[i].size >> PAGE_SHIFT);
260 		lmb_next_region_start_pfn =
261 			lmb.memory.region[i+1].base >> PAGE_SHIFT;
262 
263 		if (lmb_region_max_pfn < lmb_next_region_start_pfn)
264 			register_nosave_region(lmb_region_max_pfn,
265 					       lmb_next_region_start_pfn);
266 	}
267 
268 	return 0;
269 }
270 
271 /*
272  * paging_init() sets up the page tables - in fact we've already done this.
273  */
274 void __init paging_init(void)
275 {
276 	unsigned long total_ram = lmb_phys_mem_size();
277 	phys_addr_t top_of_ram = lmb_end_of_DRAM();
278 	unsigned long max_zone_pfns[MAX_NR_ZONES];
279 
280 #ifdef CONFIG_PPC32
281 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
282 	unsigned long end = __fix_to_virt(FIX_HOLE);
283 
284 	for (; v < end; v += PAGE_SIZE)
285 		map_page(v, 0, 0); /* XXX gross */
286 #endif
287 
288 #ifdef CONFIG_HIGHMEM
289 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
290 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
291 
292 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
293 	kmap_prot = PAGE_KERNEL;
294 #endif /* CONFIG_HIGHMEM */
295 
296 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%lx\n",
297 	       (unsigned long long)top_of_ram, total_ram);
298 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
299 	       (long int)((top_of_ram - total_ram) >> 20));
300 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
301 #ifdef CONFIG_HIGHMEM
302 	max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
303 	max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
304 #else
305 	max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
306 #endif
307 	free_area_init_nodes(max_zone_pfns);
308 
309 	mark_nonram_nosave();
310 }
311 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
312 
313 void __init mem_init(void)
314 {
315 #ifdef CONFIG_NEED_MULTIPLE_NODES
316 	int nid;
317 #endif
318 	pg_data_t *pgdat;
319 	unsigned long i;
320 	struct page *page;
321 	unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
322 
323 	num_physpages = lmb.memory.size >> PAGE_SHIFT;
324 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
325 
326 #ifdef CONFIG_NEED_MULTIPLE_NODES
327         for_each_online_node(nid) {
328 		if (NODE_DATA(nid)->node_spanned_pages != 0) {
329 			printk("freeing bootmem node %d\n", nid);
330 			totalram_pages +=
331 				free_all_bootmem_node(NODE_DATA(nid));
332 		}
333 	}
334 #else
335 	max_mapnr = max_pfn;
336 	totalram_pages += free_all_bootmem();
337 #endif
338 	for_each_online_pgdat(pgdat) {
339 		for (i = 0; i < pgdat->node_spanned_pages; i++) {
340 			if (!pfn_valid(pgdat->node_start_pfn + i))
341 				continue;
342 			page = pgdat_page_nr(pgdat, i);
343 			if (PageReserved(page))
344 				reservedpages++;
345 		}
346 	}
347 
348 	codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
349 	datasize = (unsigned long)&_edata - (unsigned long)&_sdata;
350 	initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
351 	bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
352 
353 #ifdef CONFIG_HIGHMEM
354 	{
355 		unsigned long pfn, highmem_mapnr;
356 
357 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
358 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
359 			struct page *page = pfn_to_page(pfn);
360 			if (lmb_is_reserved(pfn << PAGE_SHIFT))
361 				continue;
362 			ClearPageReserved(page);
363 			init_page_count(page);
364 			__free_page(page);
365 			totalhigh_pages++;
366 			reservedpages--;
367 		}
368 		totalram_pages += totalhigh_pages;
369 		printk(KERN_DEBUG "High memory: %luk\n",
370 		       totalhigh_pages << (PAGE_SHIFT-10));
371 	}
372 #endif /* CONFIG_HIGHMEM */
373 
374 	printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
375 	       "%luk reserved, %luk data, %luk bss, %luk init)\n",
376 		nr_free_pages() << (PAGE_SHIFT-10),
377 		num_physpages << (PAGE_SHIFT-10),
378 		codesize >> 10,
379 		reservedpages << (PAGE_SHIFT-10),
380 		datasize >> 10,
381 		bsssize >> 10,
382 		initsize >> 10);
383 
384 #ifdef CONFIG_PPC32
385 	pr_info("Kernel virtual memory layout:\n");
386 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
387 #ifdef CONFIG_HIGHMEM
388 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
389 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
390 #endif /* CONFIG_HIGHMEM */
391 #ifdef CONFIG_NOT_COHERENT_CACHE
392 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
393 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
394 #endif /* CONFIG_NOT_COHERENT_CACHE */
395 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
396 		ioremap_bot, IOREMAP_TOP);
397 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
398 		VMALLOC_START, VMALLOC_END);
399 #endif /* CONFIG_PPC32 */
400 
401 	mem_init_done = 1;
402 }
403 
404 /*
405  * This is called when a page has been modified by the kernel.
406  * It just marks the page as not i-cache clean.  We do the i-cache
407  * flush later when the page is given to a user process, if necessary.
408  */
409 void flush_dcache_page(struct page *page)
410 {
411 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
412 		return;
413 	/* avoid an atomic op if possible */
414 	if (test_bit(PG_arch_1, &page->flags))
415 		clear_bit(PG_arch_1, &page->flags);
416 }
417 EXPORT_SYMBOL(flush_dcache_page);
418 
419 void flush_dcache_icache_page(struct page *page)
420 {
421 #ifdef CONFIG_HUGETLB_PAGE
422 	if (PageCompound(page)) {
423 		flush_dcache_icache_hugepage(page);
424 		return;
425 	}
426 #endif
427 #ifdef CONFIG_BOOKE
428 	{
429 		void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
430 		__flush_dcache_icache(start);
431 		kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
432 	}
433 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
434 	/* On 8xx there is no need to kmap since highmem is not supported */
435 	__flush_dcache_icache(page_address(page));
436 #else
437 	__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
438 #endif
439 }
440 
441 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
442 {
443 	clear_page(page);
444 
445 	/*
446 	 * We shouldnt have to do this, but some versions of glibc
447 	 * require it (ld.so assumes zero filled pages are icache clean)
448 	 * - Anton
449 	 */
450 	flush_dcache_page(pg);
451 }
452 EXPORT_SYMBOL(clear_user_page);
453 
454 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
455 		    struct page *pg)
456 {
457 	copy_page(vto, vfrom);
458 
459 	/*
460 	 * We should be able to use the following optimisation, however
461 	 * there are two problems.
462 	 * Firstly a bug in some versions of binutils meant PLT sections
463 	 * were not marked executable.
464 	 * Secondly the first word in the GOT section is blrl, used
465 	 * to establish the GOT address. Until recently the GOT was
466 	 * not marked executable.
467 	 * - Anton
468 	 */
469 #if 0
470 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
471 		return;
472 #endif
473 
474 	flush_dcache_page(pg);
475 }
476 
477 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
478 			     unsigned long addr, int len)
479 {
480 	unsigned long maddr;
481 
482 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
483 	flush_icache_range(maddr, maddr + len);
484 	kunmap(page);
485 }
486 EXPORT_SYMBOL(flush_icache_user_range);
487 
488 /*
489  * This is called at the end of handling a user page fault, when the
490  * fault has been handled by updating a PTE in the linux page tables.
491  * We use it to preload an HPTE into the hash table corresponding to
492  * the updated linux PTE.
493  *
494  * This must always be called with the pte lock held.
495  */
496 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
497 		      pte_t pte)
498 {
499 #ifdef CONFIG_PPC_STD_MMU
500 	unsigned long access = 0, trap;
501 
502 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
503 	if (!pte_young(pte) || address >= TASK_SIZE)
504 		return;
505 
506 	/* We try to figure out if we are coming from an instruction
507 	 * access fault and pass that down to __hash_page so we avoid
508 	 * double-faulting on execution of fresh text. We have to test
509 	 * for regs NULL since init will get here first thing at boot
510 	 *
511 	 * We also avoid filling the hash if not coming from a fault
512 	 */
513 	if (current->thread.regs == NULL)
514 		return;
515 	trap = TRAP(current->thread.regs);
516 	if (trap == 0x400)
517 		access |= _PAGE_EXEC;
518 	else if (trap != 0x300)
519 		return;
520 	hash_preload(vma->vm_mm, address, access, trap);
521 #endif /* CONFIG_PPC_STD_MMU */
522 }
523