xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision b34e08d5) 
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 	int ret;
143 
144 	zone = page_zone(pfn_to_page(start_pfn));
145 	ret = __remove_pages(zone, start_pfn, nr_pages);
146 	if (!ret && (ppc_md.remove_memory))
147 		ret = ppc_md.remove_memory(start, size);
148 
149 	return ret;
150 }
151 #endif
152 #endif /* CONFIG_MEMORY_HOTPLUG */
153 
154 /*
155  * walk_memory_resource() needs to make sure there is no holes in a given
156  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
157  * Instead it maintains it in memblock.memory structures.  Walk through the
158  * memory regions, find holes and callback for contiguous regions.
159  */
160 int
161 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
162 		void *arg, int (*func)(unsigned long, unsigned long, void *))
163 {
164 	struct memblock_region *reg;
165 	unsigned long end_pfn = start_pfn + nr_pages;
166 	unsigned long tstart, tend;
167 	int ret = -1;
168 
169 	for_each_memblock(memory, reg) {
170 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
171 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
172 		if (tstart >= tend)
173 			continue;
174 		ret = (*func)(tstart, tend - tstart, arg);
175 		if (ret)
176 			break;
177 	}
178 	return ret;
179 }
180 EXPORT_SYMBOL_GPL(walk_system_ram_range);
181 
182 /*
183  * Initialize the bootmem system and give it all the memory we
184  * have available.  If we are using highmem, we only put the
185  * lowmem into the bootmem system.
186  */
187 #ifndef CONFIG_NEED_MULTIPLE_NODES
188 void __init do_init_bootmem(void)
189 {
190 	unsigned long start, bootmap_pages;
191 	unsigned long total_pages;
192 	struct memblock_region *reg;
193 	int boot_mapsize;
194 
195 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
196 	total_pages = (memblock_end_of_DRAM() - memstart_addr) >> PAGE_SHIFT;
197 #ifdef CONFIG_HIGHMEM
198 	total_pages = total_lowmem >> PAGE_SHIFT;
199 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
200 #endif
201 
202 	/*
203 	 * Find an area to use for the bootmem bitmap.  Calculate the size of
204 	 * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
205 	 * Add 1 additional page in case the address isn't page-aligned.
206 	 */
207 	bootmap_pages = bootmem_bootmap_pages(total_pages);
208 
209 	start = memblock_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
210 
211 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
212 	boot_mapsize = init_bootmem_node(NODE_DATA(0), start >> PAGE_SHIFT, min_low_pfn, max_low_pfn);
213 
214 	/* Place all memblock_regions in the same node and merge contiguous
215 	 * memblock_regions
216 	 */
217 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
218 
219 	/* Add all physical memory to the bootmem map, mark each area
220 	 * present.
221 	 */
222 #ifdef CONFIG_HIGHMEM
223 	free_bootmem_with_active_regions(0, lowmem_end_addr >> PAGE_SHIFT);
224 
225 	/* reserve the sections we're already using */
226 	for_each_memblock(reserved, reg) {
227 		unsigned long top = reg->base + reg->size - 1;
228 		if (top < lowmem_end_addr)
229 			reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
230 		else if (reg->base < lowmem_end_addr) {
231 			unsigned long trunc_size = lowmem_end_addr - reg->base;
232 			reserve_bootmem(reg->base, trunc_size, BOOTMEM_DEFAULT);
233 		}
234 	}
235 #else
236 	free_bootmem_with_active_regions(0, max_pfn);
237 
238 	/* reserve the sections we're already using */
239 	for_each_memblock(reserved, reg)
240 		reserve_bootmem(reg->base, reg->size, BOOTMEM_DEFAULT);
241 #endif
242 	/* XXX need to clip this if using highmem? */
243 	sparse_memory_present_with_active_regions(0);
244 
245 	init_bootmem_done = 1;
246 }
247 
248 /* mark pages that don't exist as nosave */
249 static int __init mark_nonram_nosave(void)
250 {
251 	struct memblock_region *reg, *prev = NULL;
252 
253 	for_each_memblock(memory, reg) {
254 		if (prev &&
255 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
256 			register_nosave_region(memblock_region_memory_end_pfn(prev),
257 					       memblock_region_memory_base_pfn(reg));
258 		prev = reg;
259 	}
260 	return 0;
261 }
262 
263 /*
264  * paging_init() sets up the page tables - in fact we've already done this.
265  */
266 void __init paging_init(void)
267 {
268 	unsigned long long total_ram = memblock_phys_mem_size();
269 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
270 	unsigned long max_zone_pfns[MAX_NR_ZONES];
271 
272 #ifdef CONFIG_PPC32
273 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
274 	unsigned long end = __fix_to_virt(FIX_HOLE);
275 
276 	for (; v < end; v += PAGE_SIZE)
277 		map_page(v, 0, 0); /* XXX gross */
278 #endif
279 
280 #ifdef CONFIG_HIGHMEM
281 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
282 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
283 
284 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
285 	kmap_prot = PAGE_KERNEL;
286 #endif /* CONFIG_HIGHMEM */
287 
288 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
289 	       (unsigned long long)top_of_ram, total_ram);
290 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
291 	       (long int)((top_of_ram - total_ram) >> 20));
292 	memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
293 #ifdef CONFIG_HIGHMEM
294 	max_zone_pfns[ZONE_DMA] = lowmem_end_addr >> PAGE_SHIFT;
295 	max_zone_pfns[ZONE_HIGHMEM] = top_of_ram >> PAGE_SHIFT;
296 #else
297 	max_zone_pfns[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
298 #endif
299 	free_area_init_nodes(max_zone_pfns);
300 
301 	mark_nonram_nosave();
302 }
303 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
304 
305 static void __init register_page_bootmem_info(void)
306 {
307 	int i;
308 
309 	for_each_online_node(i)
310 		register_page_bootmem_info_node(NODE_DATA(i));
311 }
312 
313 void __init mem_init(void)
314 {
315 	/*
316 	 * book3s is limited to 16 page sizes due to encoding this in
317 	 * a 4-bit field for slices.
318 	 */
319 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
320 
321 #ifdef CONFIG_SWIOTLB
322 	swiotlb_init(0);
323 #endif
324 
325 	register_page_bootmem_info();
326 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
327 	set_max_mapnr(max_pfn);
328 	free_all_bootmem();
329 
330 #ifdef CONFIG_HIGHMEM
331 	{
332 		unsigned long pfn, highmem_mapnr;
333 
334 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
335 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
336 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
337 			struct page *page = pfn_to_page(pfn);
338 			if (!memblock_is_reserved(paddr))
339 				free_highmem_page(page);
340 		}
341 	}
342 #endif /* CONFIG_HIGHMEM */
343 
344 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
345 	/*
346 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
347 	 * functions.... do it here for the non-smp case.
348 	 */
349 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
350 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
351 #endif
352 
353 	mem_init_print_info(NULL);
354 #ifdef CONFIG_PPC32
355 	pr_info("Kernel virtual memory layout:\n");
356 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
357 #ifdef CONFIG_HIGHMEM
358 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
359 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
360 #endif /* CONFIG_HIGHMEM */
361 #ifdef CONFIG_NOT_COHERENT_CACHE
362 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
363 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
364 #endif /* CONFIG_NOT_COHERENT_CACHE */
365 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
366 		ioremap_bot, IOREMAP_TOP);
367 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
368 		VMALLOC_START, VMALLOC_END);
369 #endif /* CONFIG_PPC32 */
370 
371 	mem_init_done = 1;
372 }
373 
374 void free_initmem(void)
375 {
376 	ppc_md.progress = ppc_printk_progress;
377 	free_initmem_default(POISON_FREE_INITMEM);
378 }
379 
380 #ifdef CONFIG_BLK_DEV_INITRD
381 void __init free_initrd_mem(unsigned long start, unsigned long end)
382 {
383 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
384 }
385 #endif
386 
387 /*
388  * This is called when a page has been modified by the kernel.
389  * It just marks the page as not i-cache clean.  We do the i-cache
390  * flush later when the page is given to a user process, if necessary.
391  */
392 void flush_dcache_page(struct page *page)
393 {
394 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
395 		return;
396 	/* avoid an atomic op if possible */
397 	if (test_bit(PG_arch_1, &page->flags))
398 		clear_bit(PG_arch_1, &page->flags);
399 }
400 EXPORT_SYMBOL(flush_dcache_page);
401 
402 void flush_dcache_icache_page(struct page *page)
403 {
404 #ifdef CONFIG_HUGETLB_PAGE
405 	if (PageCompound(page)) {
406 		flush_dcache_icache_hugepage(page);
407 		return;
408 	}
409 #endif
410 #ifdef CONFIG_BOOKE
411 	{
412 		void *start = kmap_atomic(page);
413 		__flush_dcache_icache(start);
414 		kunmap_atomic(start);
415 	}
416 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
417 	/* On 8xx there is no need to kmap since highmem is not supported */
418 	__flush_dcache_icache(page_address(page));
419 #else
420 	__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
421 #endif
422 }
423 EXPORT_SYMBOL(flush_dcache_icache_page);
424 
425 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
426 {
427 	clear_page(page);
428 
429 	/*
430 	 * We shouldn't have to do this, but some versions of glibc
431 	 * require it (ld.so assumes zero filled pages are icache clean)
432 	 * - Anton
433 	 */
434 	flush_dcache_page(pg);
435 }
436 EXPORT_SYMBOL(clear_user_page);
437 
438 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
439 		    struct page *pg)
440 {
441 	copy_page(vto, vfrom);
442 
443 	/*
444 	 * We should be able to use the following optimisation, however
445 	 * there are two problems.
446 	 * Firstly a bug in some versions of binutils meant PLT sections
447 	 * were not marked executable.
448 	 * Secondly the first word in the GOT section is blrl, used
449 	 * to establish the GOT address. Until recently the GOT was
450 	 * not marked executable.
451 	 * - Anton
452 	 */
453 #if 0
454 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
455 		return;
456 #endif
457 
458 	flush_dcache_page(pg);
459 }
460 
461 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
462 			     unsigned long addr, int len)
463 {
464 	unsigned long maddr;
465 
466 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
467 	flush_icache_range(maddr, maddr + len);
468 	kunmap(page);
469 }
470 EXPORT_SYMBOL(flush_icache_user_range);
471 
472 /*
473  * This is called at the end of handling a user page fault, when the
474  * fault has been handled by updating a PTE in the linux page tables.
475  * We use it to preload an HPTE into the hash table corresponding to
476  * the updated linux PTE.
477  *
478  * This must always be called with the pte lock held.
479  */
480 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
481 		      pte_t *ptep)
482 {
483 #ifdef CONFIG_PPC_STD_MMU
484 	/*
485 	 * We don't need to worry about _PAGE_PRESENT here because we are
486 	 * called with either mm->page_table_lock held or ptl lock held
487 	 */
488 	unsigned long access = 0, trap;
489 
490 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
491 	if (!pte_young(*ptep) || address >= TASK_SIZE)
492 		return;
493 
494 	/* We try to figure out if we are coming from an instruction
495 	 * access fault and pass that down to __hash_page so we avoid
496 	 * double-faulting on execution of fresh text. We have to test
497 	 * for regs NULL since init will get here first thing at boot
498 	 *
499 	 * We also avoid filling the hash if not coming from a fault
500 	 */
501 	if (current->thread.regs == NULL)
502 		return;
503 	trap = TRAP(current->thread.regs);
504 	if (trap == 0x400)
505 		access |= _PAGE_EXEC;
506 	else if (trap != 0x300)
507 		return;
508 	hash_preload(vma->vm_mm, address, access, trap);
509 #endif /* CONFIG_PPC_STD_MMU */
510 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
511 	&& defined(CONFIG_HUGETLB_PAGE)
512 	if (is_vm_hugetlb_page(vma))
513 		book3e_hugetlb_preload(vma, address, *ptep);
514 #endif
515 }
516 
517 /*
518  * System memory should not be in /proc/iomem but various tools expect it
519  * (eg kdump).
520  */
521 static int __init add_system_ram_resources(void)
522 {
523 	struct memblock_region *reg;
524 
525 	for_each_memblock(memory, reg) {
526 		struct resource *res;
527 		unsigned long base = reg->base;
528 		unsigned long size = reg->size;
529 
530 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
531 		WARN_ON(!res);
532 
533 		if (res) {
534 			res->name = "System RAM";
535 			res->start = base;
536 			res->end = base + size - 1;
537 			res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
538 			WARN_ON(request_resource(&iomem_resource, res) < 0);
539 		}
540 	}
541 
542 	return 0;
543 }
544 subsys_initcall(add_system_ram_resources);
545 
546 #ifdef CONFIG_STRICT_DEVMEM
547 /*
548  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
549  * is valid. The argument is a physical page number.
550  *
551  * Access has to be given to non-kernel-ram areas as well, these contain the
552  * PCI mmio resources as well as potential bios/acpi data regions.
553  */
554 int devmem_is_allowed(unsigned long pfn)
555 {
556 	if (iomem_is_exclusive(pfn << PAGE_SHIFT))
557 		return 0;
558 	if (!page_is_ram(pfn))
559 		return 1;
560 	if (page_is_rtas_user_buf(pfn))
561 		return 1;
562 	return 0;
563 }
564 #endif /* CONFIG_STRICT_DEVMEM */
565