xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision e23feb16) 
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 
39 #include <asm/pgalloc.h>
40 #include <asm/prom.h>
41 #include <asm/io.h>
42 #include <asm/mmu_context.h>
43 #include <asm/pgtable.h>
44 #include <asm/mmu.h>
45 #include <asm/smp.h>
46 #include <asm/machdep.h>
47 #include <asm/btext.h>
48 #include <asm/tlb.h>
49 #include <asm/sections.h>
50 #include <asm/sparsemem.h>
51 #include <asm/vdso.h>
52 #include <asm/fixmap.h>
53 #include <asm/swiotlb.h>
54 #include <asm/rtas.h>
55 
56 #include "mmu_decl.h"
57 
58 #ifndef CPU_FTR_COHERENT_ICACHE
59 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
60 #define CPU_FTR_NOEXECUTE	0
61 #endif
62 
63 int init_bootmem_done;
64 int mem_init_done;
65 unsigned long long memory_limit;
66 
67 #ifdef CONFIG_HIGHMEM
68 pte_t *kmap_pte;
69 EXPORT_SYMBOL(kmap_pte);
70 pgprot_t kmap_prot;
71 EXPORT_SYMBOL(kmap_prot);
72 
73 static inline pte_t *virt_to_kpte(unsigned long vaddr)
74 {
75 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
76 			vaddr), vaddr), vaddr);
77 }
78 #endif
79 
80 int page_is_ram(unsigned long pfn)
81 {
82 #ifndef CONFIG_PPC64	/* XXX for now */
83 	return pfn < max_pfn;
84 #else
85 	unsigned long paddr = (pfn << PAGE_SHIFT);
86 	struct memblock_region *reg;
87 
88 	for_each_memblock(memory, reg)
89 		if (paddr >= reg->base && paddr < (reg->base + reg->size))
90 			return 1;
91 	return 0;
92 #endif
93 }
94 
95 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
96 			      unsigned long size, pgprot_t vma_prot)
97 {
98 	if (ppc_md.phys_mem_access_prot)
99 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
100 
101 	if (!page_is_ram(pfn))
102 		vma_prot = pgprot_noncached(vma_prot);
103 
104 	return vma_prot;
105 }
106 EXPORT_SYMBOL(phys_mem_access_prot);
107 
108 #ifdef CONFIG_MEMORY_HOTPLUG
109 
110 #ifdef CONFIG_NUMA
111 int memory_add_physaddr_to_nid(u64 start)
112 {
113 	return hot_add_scn_to_nid(start);
114 }
115 #endif
116 
117 int arch_add_memory(int nid, u64 start, u64 size)
118 {
119 	struct pglist_data *pgdata;
120 	struct zone *zone;
121 	unsigned long start_pfn = start >> PAGE_SHIFT;
122 	unsigned long nr_pages = size >> PAGE_SHIFT;
123 
124 	pgdata = NODE_DATA(nid);
125 
126 	start = (unsigned long)__va(start);
127 	if (create_section_mapping(start, start + size))
128 		return -EINVAL;
129 
130 	/* this should work for most non-highmem platforms */
131 	zone = pgdata->node_zones;
132 
133 	return __add_pages(nid, zone, start_pfn, nr_pages);
134 }
135 
136 #ifdef CONFIG_MEMORY_HOTREMOVE
137 int arch_remove_memory(u64 start, u64 size)
138 {
139 	unsigned long start_pfn = start >> PAGE_SHIFT;
140 	unsigned long nr_pages = size >> PAGE_SHIFT;
141 	struct zone *zone;
142 
143 	zone = page_zone(pfn_to_page(start_pfn));
144 	return __remove_pages(zone, start_pfn, nr_pages);
145 }
146 #endif
147 #endif /* CONFIG_MEMORY_HOTPLUG */
148 
149 /*
150  * walk_memory_resource() needs to make sure there is no holes in a given
151  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
152  * Instead it maintains it in memblock.memory structures.  Walk through the
153  * memory regions, find holes and callback for contiguous regions.
154  */
155 int
156 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
157 		void *arg, int (*func)(unsigned long, unsigned long, void *))
158 {
159 	struct memblock_region *reg;
160 	unsigned long end_pfn = start_pfn + nr_pages;
161 	unsigned long tstart, tend;
162 	int ret = -1;
163 
164 	for_each_memblock(memory, reg) {
165 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
166 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
167 		if (tstart >= tend)
168 			continue;
169 		ret = (*func)(tstart, tend - tstart, arg);
170 		if (ret)
171 			break;
172 	}
173 	return ret;
174 }
175 EXPORT_SYMBOL_GPL(walk_system_ram_range);
176 
177 /*
178  * Initialize the bootmem system and give it all the memory we
179  * have available.  If we are using highmem, we only put the
180  * lowmem into the bootmem system.
181  */
182 #ifndef CONFIG_NEED_MULTIPLE_NODES
183 void __init do_init_bootmem(void)
184 {
185 	unsigned long start, bootmap_pages;
186 	unsigned long total_pages;
187 	struct memblock_region *reg;
188 	int boot_mapsize;
189 
190 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
191 	total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
192 #ifdef CONFIG_HIGHMEM
193 	total_pages = total_lowmem >> PAGE_SHIFT;
194 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
195 #endif
196 
197 	/*
198 	 * Find an area to use for the bootmem bitmap.  Calculate the size of
199 	 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
200 	 * Add 1 additional page in case the address isn't page-aligned.
201 	 */
202 	bootmap_pages = bootmem_bootmap_pages(total_pages);
203 
204 	start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
205 
206 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
207 	boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
208 
209 	/* Place all memblock_regions in the same node and merge contiguous
210 	 * memblock_regions
211 	 */
212 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, 0);
213 
214 	/* Add all physical memory to the bootmem map, mark each area
215 	 * present.
216 	 */
217 #ifdef CONFIG_HIGHMEM
218 	free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
219 
220 	/* reserve the sections we're already using */
221 	for_each_memblock(reserved, reg) {
222 		unsigned long top = reg->base + reg->size - 1;
223 		if (top < lowmem_end_addr)
224 			reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
225 		else if (reg->base < lowmem_end_addr) {
226 			unsigned long trunc_size = lowmem_end_addr - reg->base;
227 			reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
228 		}
229 	}
230 #else
231 	free_bootmem_with_active_regions(0, max_pfn);
232 
233 	/* reserve the sections we're already using */
234 	for_each_memblock(reserved, reg)
235 		reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
236 #endif
237 	/* XXX need to clip this if using highmem? */
238 	sparse_memory_present_with_active_regions(0);
239 
240 	init_bootmem_done = 1;
241 }
242 
243 /* mark pages that don't exist as nosave */
244 static int __init mark_nonram_nosave(void)
245 {
246 	struct memblock_region *reg, *prev = NULL;
247 
248 	for_each_memblock(memory, reg) {
249 		if (prev &&
250 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
251 			register_nosave_region(memblock_region_memory_end_pfn(prev),
252 					       memblock_region_memory_base_pfn(reg));
253 		prev = reg;
254 	}
255 	return 0;
256 }
257 
258 /*
259  * paging_init() sets up the page tables - in fact we've already done this.
260  */
261 void __init paging_init(void)
262 {
263 	unsigned long long total_ram = memblock_phys_mem_size();
264 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
265 	unsigned long max_zone_pfns[MAX_NR_ZONES];
266 
267 #ifdef CONFIG_PPC32
268 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
269 	unsigned long end = __fix_to_virt(FIX_HOLE);
270 
271 	for (; v < end; v += PAGE_SIZE)
272 		map_page(v, 0, 0); /* XXX gross */
273 #endif
274 
275 #ifdef CONFIG_HIGHMEM
276 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
277 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
278 
279 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
280 	kmap_prot = PAGE_KERNEL;
281 #endif /* CONFIG_HIGHMEM */
282 
283 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
284 	       (unsigned long long)top_of_ram, total_ram);
285 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
286 	       (long int)((top_of_ram - total_ram) >> 20));
287 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
288 #ifdef CONFIG_HIGHMEM
289 	max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
290 	max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
291 #else
292 	max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
293 #endif
294 	free_area_init_nodes(max_zone_pfns);
295 
296 	mark_nonram_nosave();
297 }
298 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
299 
300 static void __init register_page_bootmem_info(void)
301 {
302 	int i;
303 
304 	for_each_online_node(i)
305 		register_page_bootmem_info_node(NODE_DATA(i));
306 }
307 
308 void __init mem_init(void)
309 {
310 #ifdef CONFIG_SWIOTLB
311 	swiotlb_init(0);
312 #endif
313 
314 	register_page_bootmem_info();
315 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
316 	set_max_mapnr(max_pfn);
317 	free_all_bootmem();
318 
319 #ifdef CONFIG_HIGHMEM
320 	{
321 		unsigned long pfn, highmem_mapnr;
322 
323 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
324 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
325 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
326 			struct page *page = pfn_to_page(pfn);
327 			if (!memblock_is_reserved(paddr))
328 				free_highmem_page(page);
329 		}
330 	}
331 #endif /* CONFIG_HIGHMEM */
332 
333 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
334 	/*
335 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
336 	 * functions.... do it here for the non-smp case.
337 	 */
338 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
339 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
340 #endif
341 
342 	mem_init_print_info(NULL);
343 #ifdef CONFIG_PPC32
344 	pr_info("Kernel virtual memory layout:\n");
345 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
346 #ifdef CONFIG_HIGHMEM
347 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
348 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
349 #endif /* CONFIG_HIGHMEM */
350 #ifdef CONFIG_NOT_COHERENT_CACHE
351 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
352 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
353 #endif /* CONFIG_NOT_COHERENT_CACHE */
354 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
355 		ioremap_bot, IOREMAP_TOP);
356 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
357 		VMALLOC_START, VMALLOC_END);
358 #endif /* CONFIG_PPC32 */
359 
360 	mem_init_done = 1;
361 }
362 
363 void free_initmem(void)
364 {
365 	ppc_md.progress = ppc_printk_progress;
366 	free_initmem_default(POISON_FREE_INITMEM);
367 }
368 
369 #ifdef CONFIG_BLK_DEV_INITRD
370 void __init free_initrd_mem(unsigned long start, unsigned long end)
371 {
372 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
373 }
374 #endif
375 
376 /*
377  * This is called when a page has been modified by the kernel.
378  * It just marks the page as not i-cache clean.  We do the i-cache
379  * flush later when the page is given to a user process, if necessary.
380  */
381 void flush_dcache_page(struct page *page)
382 {
383 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
384 		return;
385 	/* avoid an atomic op if possible */
386 	if (test_bit(PG_arch_1, &page->flags))
387 		clear_bit(PG_arch_1, &page->flags);
388 }
389 EXPORT_SYMBOL(flush_dcache_page);
390 
391 void flush_dcache_icache_page(struct page *page)
392 {
393 #ifdef CONFIG_HUGETLB_PAGE
394 	if (PageCompound(page)) {
395 		flush_dcache_icache_hugepage(page);
396 		return;
397 	}
398 #endif
399 #ifdef CONFIG_BOOKE
400 	{
401 		void *start = kmap_atomic(page);
402 		__flush_dcache_icache(start);
403 		kunmap_atomic(start);
404 	}
405 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
406 	/* On 8xx there is no need to kmap since highmem is not supported */
407 	__flush_dcache_icache(page_address(page));
408 #else
409 	__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
410 #endif
411 }
412 EXPORT_SYMBOL(flush_dcache_icache_page);
413 
414 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
415 {
416 	clear_page(page);
417 
418 	/*
419 	 * We shouldn't have to do this, but some versions of glibc
420 	 * require it (ld.so assumes zero filled pages are icache clean)
421 	 * - Anton
422 	 */
423 	flush_dcache_page(pg);
424 }
425 EXPORT_SYMBOL(clear_user_page);
426 
427 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
428 		    struct page *pg)
429 {
430 	copy_page(vto, vfrom);
431 
432 	/*
433 	 * We should be able to use the following optimisation, however
434 	 * there are two problems.
435 	 * Firstly a bug in some versions of binutils meant PLT sections
436 	 * were not marked executable.
437 	 * Secondly the first word in the GOT section is blrl, used
438 	 * to establish the GOT address. Until recently the GOT was
439 	 * not marked executable.
440 	 * - Anton
441 	 */
442 #if 0
443 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
444 		return;
445 #endif
446 
447 	flush_dcache_page(pg);
448 }
449 
450 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
451 			     unsigned long addr, int len)
452 {
453 	unsigned long maddr;
454 
455 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
456 	flush_icache_range(maddr, maddr + len);
457 	kunmap(page);
458 }
459 EXPORT_SYMBOL(flush_icache_user_range);
460 
461 /*
462  * This is called at the end of handling a user page fault, when the
463  * fault has been handled by updating a PTE in the linux page tables.
464  * We use it to preload an HPTE into the hash table corresponding to
465  * the updated linux PTE.
466  *
467  * This must always be called with the pte lock held.
468  */
469 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
470 		      pte_t *ptep)
471 {
472 #ifdef CONFIG_PPC_STD_MMU
473 	/*
474 	 * We don't need to worry about _PAGE_PRESENT here because we are
475 	 * called with either mm->page_table_lock held or ptl lock held
476 	 */
477 	unsigned long access = 0, trap;
478 
479 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
480 	if (!pte_young(*ptep) || address >= TASK_SIZE)
481 		return;
482 
483 	/* We try to figure out if we are coming from an instruction
484 	 * access fault and pass that down to __hash_page so we avoid
485 	 * double-faulting on execution of fresh text. We have to test
486 	 * for regs NULL since init will get here first thing at boot
487 	 *
488 	 * We also avoid filling the hash if not coming from a fault
489 	 */
490 	if (current->thread.regs == NULL)
491 		return;
492 	trap = TRAP(current->thread.regs);
493 	if (trap == 0x400)
494 		access |= _PAGE_EXEC;
495 	else if (trap != 0x300)
496 		return;
497 	hash_preload(vma->vm_mm, address, access, trap);
498 #endif /* CONFIG_PPC_STD_MMU */
499 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
500 	&& defined(CONFIG_HUGETLB_PAGE)
501 	if (is_vm_hugetlb_page(vma))
502 		book3e_hugetlb_preload(vma, address, *ptep);
503 #endif
504 }
505 
506 /*
507  * System memory should not be in /proc/iomem but various tools expect it
508  * (eg kdump).
509  */
510 static int add_system_ram_resources(void)
511 {
512 	struct memblock_region *reg;
513 
514 	for_each_memblock(memory, reg) {
515 		struct resource *res;
516 		unsigned long base = reg->base;
517 		unsigned long size = reg->size;
518 
519 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
520 		WARN_ON(!res);
521 
522 		if (res) {
523 			res->name = "System RAM";
524 			res->start = base;
525 			res->end = base + size - 1;
526 			res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
527 			WARN_ON(request_resource(&iomem_resource, res) < 0);
528 		}
529 	}
530 
531 	return 0;
532 }
533 subsys_initcall(add_system_ram_resources);
534 
535 #ifdef CONFIG_STRICT_DEVMEM
536 /*
537  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
538  * is valid. The argument is a physical page number.
539  *
540  * Access has to be given to non-kernel-ram areas as well, these contain the
541  * PCI mmio resources as well as potential bios/acpi data regions.
542  */
543 int devmem_is_allowed(unsigned long pfn)
544 {
545 	if (iomem_is_exclusive(pfn << PAGE_SHIFT))
546 		return 0;
547 	if (!page_is_ram(pfn))
548 		return 1;
549 	if (page_is_rtas_user_buf(pfn))
550 		return 1;
551 	return 0;
552 }
553 #endif /* CONFIG_STRICT_DEVMEM */
554