xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 867a0e05) 
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 
40 #include <asm/pgalloc.h>
41 #include <asm/prom.h>
42 #include <asm/io.h>
43 #include <asm/mmu_context.h>
44 #include <asm/pgtable.h>
45 #include <asm/mmu.h>
46 #include <asm/smp.h>
47 #include <asm/machdep.h>
48 #include <asm/btext.h>
49 #include <asm/tlb.h>
50 #include <asm/sections.h>
51 #include <asm/sparsemem.h>
52 #include <asm/vdso.h>
53 #include <asm/fixmap.h>
54 #include <asm/swiotlb.h>
55 #include <asm/rtas.h>
56 
57 #include "mmu_decl.h"
58 
59 #ifndef CPU_FTR_COHERENT_ICACHE
60 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
61 #define CPU_FTR_NOEXECUTE	0
62 #endif
63 
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 		zone_for_memory(nid, start, size, 0);
133 
134 	return __add_pages(nid, zone, start_pfn, nr_pages);
135 }
136 
137 #ifdef CONFIG_MEMORY_HOTREMOVE
138 int arch_remove_memory(u64 start, u64 size)
139 {
140 	unsigned long start_pfn = start >> PAGE_SHIFT;
141 	unsigned long nr_pages = size >> PAGE_SHIFT;
142 	struct zone *zone;
143 	int ret;
144 
145 	zone = page_zone(pfn_to_page(start_pfn));
146 	ret = __remove_pages(zone, start_pfn, nr_pages);
147 	if (ret)
148 		return ret;
149 
150 	/* Remove htab bolted mappings for this section of memory */
151 	start = (unsigned long)__va(start);
152 	ret = remove_section_mapping(start, start + size);
153 
154 	/* Ensure all vmalloc mappings are flushed in case they also
155 	 * hit that section of memory
156 	 */
157 	vm_unmap_aliases();
158 
159 	return ret;
160 }
161 #endif
162 #endif /* CONFIG_MEMORY_HOTPLUG */
163 
164 /*
165  * walk_memory_resource() needs to make sure there is no holes in a given
166  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
167  * Instead it maintains it in memblock.memory structures.  Walk through the
168  * memory regions, find holes and callback for contiguous regions.
169  */
170 int
171 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
172 		void *arg, int (*func)(unsigned long, unsigned long, void *))
173 {
174 	struct memblock_region *reg;
175 	unsigned long end_pfn = start_pfn + nr_pages;
176 	unsigned long tstart, tend;
177 	int ret = -1;
178 
179 	for_each_memblock(memory, reg) {
180 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
181 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
182 		if (tstart >= tend)
183 			continue;
184 		ret = (*func)(tstart, tend - tstart, arg);
185 		if (ret)
186 			break;
187 	}
188 	return ret;
189 }
190 EXPORT_SYMBOL_GPL(walk_system_ram_range);
191 
192 #ifndef CONFIG_NEED_MULTIPLE_NODES
193 void __init initmem_init(void)
194 {
195 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
196 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
197 #ifdef CONFIG_HIGHMEM
198 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
199 #endif
200 
201 	/* Place all memblock_regions in the same node and merge contiguous
202 	 * memblock_regions
203 	 */
204 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
205 
206 	/* XXX need to clip this if using highmem? */
207 	sparse_memory_present_with_active_regions(0);
208 	sparse_init();
209 }
210 
211 /* mark pages that don't exist as nosave */
212 static int __init mark_nonram_nosave(void)
213 {
214 	struct memblock_region *reg, *prev = NULL;
215 
216 	for_each_memblock(memory, reg) {
217 		if (prev &&
218 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
219 			register_nosave_region(memblock_region_memory_end_pfn(prev),
220 					       memblock_region_memory_base_pfn(reg));
221 		prev = reg;
222 	}
223 	return 0;
224 }
225 #else /* CONFIG_NEED_MULTIPLE_NODES */
226 static int __init mark_nonram_nosave(void)
227 {
228 	return 0;
229 }
230 #endif
231 
232 static bool zone_limits_final;
233 
234 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
235 	[0 ... MAX_NR_ZONES - 1] = ~0UL
236 };
237 
238 /*
239  * Restrict the specified zone and all more restrictive zones
240  * to be below the specified pfn.  May not be called after
241  * paging_init().
242  */
243 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
244 {
245 	int i;
246 
247 	if (WARN_ON(zone_limits_final))
248 		return;
249 
250 	for (i = zone; i >= 0; i--) {
251 		if (max_zone_pfns[i] > pfn_limit)
252 			max_zone_pfns[i] = pfn_limit;
253 	}
254 }
255 
256 /*
257  * Find the least restrictive zone that is entirely below the
258  * specified pfn limit.  Returns < 0 if no suitable zone is found.
259  *
260  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
261  * systems -- the DMA limit can be higher than any possible real pfn.
262  */
263 int dma_pfn_limit_to_zone(u64 pfn_limit)
264 {
265 	enum zone_type top_zone = ZONE_NORMAL;
266 	int i;
267 
268 #ifdef CONFIG_HIGHMEM
269 	top_zone = ZONE_HIGHMEM;
270 #endif
271 
272 	for (i = top_zone; i >= 0; i--) {
273 		if (max_zone_pfns[i] <= pfn_limit)
274 			return i;
275 	}
276 
277 	return -EPERM;
278 }
279 
280 /*
281  * paging_init() sets up the page tables - in fact we've already done this.
282  */
283 void __init paging_init(void)
284 {
285 	unsigned long long total_ram = memblock_phys_mem_size();
286 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
287 	enum zone_type top_zone;
288 
289 #ifdef CONFIG_PPC32
290 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
291 	unsigned long end = __fix_to_virt(FIX_HOLE);
292 
293 	for (; v < end; v += PAGE_SIZE)
294 		map_page(v, 0, 0); /* XXX gross */
295 #endif
296 
297 #ifdef CONFIG_HIGHMEM
298 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
299 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
300 
301 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
302 	kmap_prot = PAGE_KERNEL;
303 #endif /* CONFIG_HIGHMEM */
304 
305 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
306 	       (unsigned long long)top_of_ram, total_ram);
307 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
308 	       (long int)((top_of_ram - total_ram) >> 20));
309 
310 #ifdef CONFIG_HIGHMEM
311 	top_zone = ZONE_HIGHMEM;
312 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
313 #else
314 	top_zone = ZONE_NORMAL;
315 #endif
316 
317 	limit_zone_pfn(top_zone, top_of_ram >> PAGE_SHIFT);
318 	zone_limits_final = true;
319 	free_area_init_nodes(max_zone_pfns);
320 
321 	mark_nonram_nosave();
322 }
323 
324 void __init mem_init(void)
325 {
326 	/*
327 	 * book3s is limited to 16 page sizes due to encoding this in
328 	 * a 4-bit field for slices.
329 	 */
330 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
331 
332 #ifdef CONFIG_SWIOTLB
333 	swiotlb_init(0);
334 #endif
335 
336 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
337 	set_max_mapnr(max_pfn);
338 	free_all_bootmem();
339 
340 #ifdef CONFIG_HIGHMEM
341 	{
342 		unsigned long pfn, highmem_mapnr;
343 
344 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
345 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
346 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
347 			struct page *page = pfn_to_page(pfn);
348 			if (!memblock_is_reserved(paddr))
349 				free_highmem_page(page);
350 		}
351 	}
352 #endif /* CONFIG_HIGHMEM */
353 
354 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
355 	/*
356 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
357 	 * functions.... do it here for the non-smp case.
358 	 */
359 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
360 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
361 #endif
362 
363 	mem_init_print_info(NULL);
364 #ifdef CONFIG_PPC32
365 	pr_info("Kernel virtual memory layout:\n");
366 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
367 #ifdef CONFIG_HIGHMEM
368 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
369 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
370 #endif /* CONFIG_HIGHMEM */
371 #ifdef CONFIG_NOT_COHERENT_CACHE
372 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
373 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
374 #endif /* CONFIG_NOT_COHERENT_CACHE */
375 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
376 		ioremap_bot, IOREMAP_TOP);
377 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
378 		VMALLOC_START, VMALLOC_END);
379 #endif /* CONFIG_PPC32 */
380 
381 	mem_init_done = 1;
382 }
383 
384 void free_initmem(void)
385 {
386 	ppc_md.progress = ppc_printk_progress;
387 	free_initmem_default(POISON_FREE_INITMEM);
388 }
389 
390 #ifdef CONFIG_BLK_DEV_INITRD
391 void __init free_initrd_mem(unsigned long start, unsigned long end)
392 {
393 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
394 }
395 #endif
396 
397 /*
398  * This is called when a page has been modified by the kernel.
399  * It just marks the page as not i-cache clean.  We do the i-cache
400  * flush later when the page is given to a user process, if necessary.
401  */
402 void flush_dcache_page(struct page *page)
403 {
404 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
405 		return;
406 	/* avoid an atomic op if possible */
407 	if (test_bit(PG_arch_1, &page->flags))
408 		clear_bit(PG_arch_1, &page->flags);
409 }
410 EXPORT_SYMBOL(flush_dcache_page);
411 
412 void flush_dcache_icache_page(struct page *page)
413 {
414 #ifdef CONFIG_HUGETLB_PAGE
415 	if (PageCompound(page)) {
416 		flush_dcache_icache_hugepage(page);
417 		return;
418 	}
419 #endif
420 #ifdef CONFIG_BOOKE
421 	{
422 		void *start = kmap_atomic(page);
423 		__flush_dcache_icache(start);
424 		kunmap_atomic(start);
425 	}
426 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
427 	/* On 8xx there is no need to kmap since highmem is not supported */
428 	__flush_dcache_icache(page_address(page));
429 #else
430 	__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
431 #endif
432 }
433 EXPORT_SYMBOL(flush_dcache_icache_page);
434 
435 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
436 {
437 	clear_page(page);
438 
439 	/*
440 	 * We shouldn't have to do this, but some versions of glibc
441 	 * require it (ld.so assumes zero filled pages are icache clean)
442 	 * - Anton
443 	 */
444 	flush_dcache_page(pg);
445 }
446 EXPORT_SYMBOL(clear_user_page);
447 
448 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
449 		    struct page *pg)
450 {
451 	copy_page(vto, vfrom);
452 
453 	/*
454 	 * We should be able to use the following optimisation, however
455 	 * there are two problems.
456 	 * Firstly a bug in some versions of binutils meant PLT sections
457 	 * were not marked executable.
458 	 * Secondly the first word in the GOT section is blrl, used
459 	 * to establish the GOT address. Until recently the GOT was
460 	 * not marked executable.
461 	 * - Anton
462 	 */
463 #if 0
464 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
465 		return;
466 #endif
467 
468 	flush_dcache_page(pg);
469 }
470 
471 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
472 			     unsigned long addr, int len)
473 {
474 	unsigned long maddr;
475 
476 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
477 	flush_icache_range(maddr, maddr + len);
478 	kunmap(page);
479 }
480 EXPORT_SYMBOL(flush_icache_user_range);
481 
482 /*
483  * This is called at the end of handling a user page fault, when the
484  * fault has been handled by updating a PTE in the linux page tables.
485  * We use it to preload an HPTE into the hash table corresponding to
486  * the updated linux PTE.
487  *
488  * This must always be called with the pte lock held.
489  */
490 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
491 		      pte_t *ptep)
492 {
493 #ifdef CONFIG_PPC_STD_MMU
494 	/*
495 	 * We don't need to worry about _PAGE_PRESENT here because we are
496 	 * called with either mm->page_table_lock held or ptl lock held
497 	 */
498 	unsigned long access = 0, trap;
499 
500 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
501 	if (!pte_young(*ptep) || address >= TASK_SIZE)
502 		return;
503 
504 	/* We try to figure out if we are coming from an instruction
505 	 * access fault and pass that down to __hash_page so we avoid
506 	 * double-faulting on execution of fresh text. We have to test
507 	 * for regs NULL since init will get here first thing at boot
508 	 *
509 	 * We also avoid filling the hash if not coming from a fault
510 	 */
511 	if (current->thread.regs == NULL)
512 		return;
513 	trap = TRAP(current->thread.regs);
514 	if (trap == 0x400)
515 		access |= _PAGE_EXEC;
516 	else if (trap != 0x300)
517 		return;
518 	hash_preload(vma->vm_mm, address, access, trap);
519 #endif /* CONFIG_PPC_STD_MMU */
520 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
521 	&& defined(CONFIG_HUGETLB_PAGE)
522 	if (is_vm_hugetlb_page(vma))
523 		book3e_hugetlb_preload(vma, address, *ptep);
524 #endif
525 }
526 
527 /*
528  * System memory should not be in /proc/iomem but various tools expect it
529  * (eg kdump).
530  */
531 static int __init add_system_ram_resources(void)
532 {
533 	struct memblock_region *reg;
534 
535 	for_each_memblock(memory, reg) {
536 		struct resource *res;
537 		unsigned long base = reg->base;
538 		unsigned long size = reg->size;
539 
540 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
541 		WARN_ON(!res);
542 
543 		if (res) {
544 			res->name = "System RAM";
545 			res->start = base;
546 			res->end = base + size - 1;
547 			res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
548 			WARN_ON(request_resource(&iomem_resource, res) < 0);
549 		}
550 	}
551 
552 	return 0;
553 }
554 subsys_initcall(add_system_ram_resources);
555 
556 #ifdef CONFIG_STRICT_DEVMEM
557 /*
558  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
559  * is valid. The argument is a physical page number.
560  *
561  * Access has to be given to non-kernel-ram areas as well, these contain the
562  * PCI mmio resources as well as potential bios/acpi data regions.
563  */
564 int devmem_is_allowed(unsigned long pfn)
565 {
566 	if (iomem_is_exclusive(pfn << PAGE_SHIFT))
567 		return 0;
568 	if (!page_is_ram(pfn))
569 		return 1;
570 	if (page_is_rtas_user_buf(pfn))
571 		return 1;
572 	return 0;
573 }
574 #endif /* CONFIG_STRICT_DEVMEM */
575