xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 56d06fa2) 
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 unsigned long long memory_limit;
65 
66 #ifdef CONFIG_HIGHMEM
67 pte_t *kmap_pte;
68 EXPORT_SYMBOL(kmap_pte);
69 pgprot_t kmap_prot;
70 EXPORT_SYMBOL(kmap_prot);
71 
72 static inline pte_t *virt_to_kpte(unsigned long vaddr)
73 {
74 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
75 			vaddr), vaddr), vaddr);
76 }
77 #endif
78 
79 int page_is_ram(unsigned long pfn)
80 {
81 #ifndef CONFIG_PPC64	/* XXX for now */
82 	return pfn < max_pfn;
83 #else
84 	unsigned long paddr = (pfn << PAGE_SHIFT);
85 	struct memblock_region *reg;
86 
87 	for_each_memblock(memory, reg)
88 		if (paddr >= reg->base && paddr < (reg->base + reg->size))
89 			return 1;
90 	return 0;
91 #endif
92 }
93 
94 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
95 			      unsigned long size, pgprot_t vma_prot)
96 {
97 	if (ppc_md.phys_mem_access_prot)
98 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
99 
100 	if (!page_is_ram(pfn))
101 		vma_prot = pgprot_noncached(vma_prot);
102 
103 	return vma_prot;
104 }
105 EXPORT_SYMBOL(phys_mem_access_prot);
106 
107 #ifdef CONFIG_MEMORY_HOTPLUG
108 
109 #ifdef CONFIG_NUMA
110 int memory_add_physaddr_to_nid(u64 start)
111 {
112 	return hot_add_scn_to_nid(start);
113 }
114 #endif
115 
116 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
117 {
118 	struct pglist_data *pgdata;
119 	struct zone *zone;
120 	unsigned long start_pfn = start >> PAGE_SHIFT;
121 	unsigned long nr_pages = size >> PAGE_SHIFT;
122 	int rc;
123 
124 	pgdata = NODE_DATA(nid);
125 
126 	start = (unsigned long)__va(start);
127 	rc = create_section_mapping(start, start + size);
128 	if (rc) {
129 		pr_warning(
130 			"Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
131 			start, start + size, rc);
132 		return -EFAULT;
133 	}
134 
135 	/* this should work for most non-highmem platforms */
136 	zone = pgdata->node_zones +
137 		zone_for_memory(nid, start, size, 0, for_device);
138 
139 	return __add_pages(nid, zone, start_pfn, nr_pages);
140 }
141 
142 #ifdef CONFIG_MEMORY_HOTREMOVE
143 int arch_remove_memory(u64 start, u64 size)
144 {
145 	unsigned long start_pfn = start >> PAGE_SHIFT;
146 	unsigned long nr_pages = size >> PAGE_SHIFT;
147 	struct zone *zone;
148 	int ret;
149 
150 	zone = page_zone(pfn_to_page(start_pfn));
151 	ret = __remove_pages(zone, start_pfn, nr_pages);
152 	if (ret)
153 		return ret;
154 
155 	/* Remove htab bolted mappings for this section of memory */
156 	start = (unsigned long)__va(start);
157 	ret = remove_section_mapping(start, start + size);
158 
159 	/* Ensure all vmalloc mappings are flushed in case they also
160 	 * hit that section of memory
161 	 */
162 	vm_unmap_aliases();
163 
164 	return ret;
165 }
166 #endif
167 #endif /* CONFIG_MEMORY_HOTPLUG */
168 
169 /*
170  * walk_memory_resource() needs to make sure there is no holes in a given
171  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
172  * Instead it maintains it in memblock.memory structures.  Walk through the
173  * memory regions, find holes and callback for contiguous regions.
174  */
175 int
176 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
177 		void *arg, int (*func)(unsigned long, unsigned long, void *))
178 {
179 	struct memblock_region *reg;
180 	unsigned long end_pfn = start_pfn + nr_pages;
181 	unsigned long tstart, tend;
182 	int ret = -1;
183 
184 	for_each_memblock(memory, reg) {
185 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
186 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
187 		if (tstart >= tend)
188 			continue;
189 		ret = (*func)(tstart, tend - tstart, arg);
190 		if (ret)
191 			break;
192 	}
193 	return ret;
194 }
195 EXPORT_SYMBOL_GPL(walk_system_ram_range);
196 
197 #ifndef CONFIG_NEED_MULTIPLE_NODES
198 void __init initmem_init(void)
199 {
200 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
201 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
202 #ifdef CONFIG_HIGHMEM
203 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
204 #endif
205 
206 	/* Place all memblock_regions in the same node and merge contiguous
207 	 * memblock_regions
208 	 */
209 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
210 
211 	/* XXX need to clip this if using highmem? */
212 	sparse_memory_present_with_active_regions(0);
213 	sparse_init();
214 }
215 
216 /* mark pages that don't exist as nosave */
217 static int __init mark_nonram_nosave(void)
218 {
219 	struct memblock_region *reg, *prev = NULL;
220 
221 	for_each_memblock(memory, reg) {
222 		if (prev &&
223 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
224 			register_nosave_region(memblock_region_memory_end_pfn(prev),
225 					       memblock_region_memory_base_pfn(reg));
226 		prev = reg;
227 	}
228 	return 0;
229 }
230 #else /* CONFIG_NEED_MULTIPLE_NODES */
231 static int __init mark_nonram_nosave(void)
232 {
233 	return 0;
234 }
235 #endif
236 
237 static bool zone_limits_final;
238 
239 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
240 	[0 ... MAX_NR_ZONES - 1] = ~0UL
241 };
242 
243 /*
244  * Restrict the specified zone and all more restrictive zones
245  * to be below the specified pfn.  May not be called after
246  * paging_init().
247  */
248 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
249 {
250 	int i;
251 
252 	if (WARN_ON(zone_limits_final))
253 		return;
254 
255 	for (i = zone; i >= 0; i--) {
256 		if (max_zone_pfns[i] > pfn_limit)
257 			max_zone_pfns[i] = pfn_limit;
258 	}
259 }
260 
261 /*
262  * Find the least restrictive zone that is entirely below the
263  * specified pfn limit.  Returns < 0 if no suitable zone is found.
264  *
265  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
266  * systems -- the DMA limit can be higher than any possible real pfn.
267  */
268 int dma_pfn_limit_to_zone(u64 pfn_limit)
269 {
270 	enum zone_type top_zone = ZONE_NORMAL;
271 	int i;
272 
273 #ifdef CONFIG_HIGHMEM
274 	top_zone = ZONE_HIGHMEM;
275 #endif
276 
277 	for (i = top_zone; i >= 0; i--) {
278 		if (max_zone_pfns[i] <= pfn_limit)
279 			return i;
280 	}
281 
282 	return -EPERM;
283 }
284 
285 /*
286  * paging_init() sets up the page tables - in fact we've already done this.
287  */
288 void __init paging_init(void)
289 {
290 	unsigned long long total_ram = memblock_phys_mem_size();
291 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
292 	enum zone_type top_zone;
293 
294 #ifdef CONFIG_PPC32
295 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
296 	unsigned long end = __fix_to_virt(FIX_HOLE);
297 
298 	for (; v < end; v += PAGE_SIZE)
299 		map_page(v, 0, 0); /* XXX gross */
300 #endif
301 
302 #ifdef CONFIG_HIGHMEM
303 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
304 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
305 
306 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
307 	kmap_prot = PAGE_KERNEL;
308 #endif /* CONFIG_HIGHMEM */
309 
310 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
311 	       (unsigned long long)top_of_ram, total_ram);
312 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
313 	       (long int)((top_of_ram - total_ram) >> 20));
314 
315 #ifdef CONFIG_HIGHMEM
316 	top_zone = ZONE_HIGHMEM;
317 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
318 #else
319 	top_zone = ZONE_NORMAL;
320 #endif
321 
322 	limit_zone_pfn(top_zone, top_of_ram >> PAGE_SHIFT);
323 	zone_limits_final = true;
324 	free_area_init_nodes(max_zone_pfns);
325 
326 	mark_nonram_nosave();
327 }
328 
329 void __init mem_init(void)
330 {
331 	/*
332 	 * book3s is limited to 16 page sizes due to encoding this in
333 	 * a 4-bit field for slices.
334 	 */
335 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
336 
337 #ifdef CONFIG_SWIOTLB
338 	swiotlb_init(0);
339 #endif
340 
341 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
342 	set_max_mapnr(max_pfn);
343 	free_all_bootmem();
344 
345 #ifdef CONFIG_HIGHMEM
346 	{
347 		unsigned long pfn, highmem_mapnr;
348 
349 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
350 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
351 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
352 			struct page *page = pfn_to_page(pfn);
353 			if (!memblock_is_reserved(paddr))
354 				free_highmem_page(page);
355 		}
356 	}
357 #endif /* CONFIG_HIGHMEM */
358 
359 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
360 	/*
361 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
362 	 * functions.... do it here for the non-smp case.
363 	 */
364 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
365 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
366 #endif
367 
368 	mem_init_print_info(NULL);
369 #ifdef CONFIG_PPC32
370 	pr_info("Kernel virtual memory layout:\n");
371 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
372 #ifdef CONFIG_HIGHMEM
373 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
374 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
375 #endif /* CONFIG_HIGHMEM */
376 #ifdef CONFIG_NOT_COHERENT_CACHE
377 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
378 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
379 #endif /* CONFIG_NOT_COHERENT_CACHE */
380 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
381 		ioremap_bot, IOREMAP_TOP);
382 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
383 		VMALLOC_START, VMALLOC_END);
384 #endif /* CONFIG_PPC32 */
385 }
386 
387 void free_initmem(void)
388 {
389 	ppc_md.progress = ppc_printk_progress;
390 	free_initmem_default(POISON_FREE_INITMEM);
391 }
392 
393 #ifdef CONFIG_BLK_DEV_INITRD
394 void __init free_initrd_mem(unsigned long start, unsigned long end)
395 {
396 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
397 }
398 #endif
399 
400 /*
401  * This is called when a page has been modified by the kernel.
402  * It just marks the page as not i-cache clean.  We do the i-cache
403  * flush later when the page is given to a user process, if necessary.
404  */
405 void flush_dcache_page(struct page *page)
406 {
407 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
408 		return;
409 	/* avoid an atomic op if possible */
410 	if (test_bit(PG_arch_1, &page->flags))
411 		clear_bit(PG_arch_1, &page->flags);
412 }
413 EXPORT_SYMBOL(flush_dcache_page);
414 
415 void flush_dcache_icache_page(struct page *page)
416 {
417 #ifdef CONFIG_HUGETLB_PAGE
418 	if (PageCompound(page)) {
419 		flush_dcache_icache_hugepage(page);
420 		return;
421 	}
422 #endif
423 #if defined(CONFIG_8xx) || defined(CONFIG_PPC64)
424 	/* On 8xx there is no need to kmap since highmem is not supported */
425 	__flush_dcache_icache(page_address(page));
426 #else
427 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
428 		void *start = kmap_atomic(page);
429 		__flush_dcache_icache(start);
430 		kunmap_atomic(start);
431 	} else {
432 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
433 	}
434 #endif
435 }
436 EXPORT_SYMBOL(flush_dcache_icache_page);
437 
438 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
439 {
440 	clear_page(page);
441 
442 	/*
443 	 * We shouldn't have to do this, but some versions of glibc
444 	 * require it (ld.so assumes zero filled pages are icache clean)
445 	 * - Anton
446 	 */
447 	flush_dcache_page(pg);
448 }
449 EXPORT_SYMBOL(clear_user_page);
450 
451 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
452 		    struct page *pg)
453 {
454 	copy_page(vto, vfrom);
455 
456 	/*
457 	 * We should be able to use the following optimisation, however
458 	 * there are two problems.
459 	 * Firstly a bug in some versions of binutils meant PLT sections
460 	 * were not marked executable.
461 	 * Secondly the first word in the GOT section is blrl, used
462 	 * to establish the GOT address. Until recently the GOT was
463 	 * not marked executable.
464 	 * - Anton
465 	 */
466 #if 0
467 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
468 		return;
469 #endif
470 
471 	flush_dcache_page(pg);
472 }
473 
474 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
475 			     unsigned long addr, int len)
476 {
477 	unsigned long maddr;
478 
479 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
480 	flush_icache_range(maddr, maddr + len);
481 	kunmap(page);
482 }
483 EXPORT_SYMBOL(flush_icache_user_range);
484 
485 /*
486  * This is called at the end of handling a user page fault, when the
487  * fault has been handled by updating a PTE in the linux page tables.
488  * We use it to preload an HPTE into the hash table corresponding to
489  * the updated linux PTE.
490  *
491  * This must always be called with the pte lock held.
492  */
493 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
494 		      pte_t *ptep)
495 {
496 #ifdef CONFIG_PPC_STD_MMU
497 	/*
498 	 * We don't need to worry about _PAGE_PRESENT here because we are
499 	 * called with either mm->page_table_lock held or ptl lock held
500 	 */
501 	unsigned long access = 0, trap;
502 
503 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
504 	if (!pte_young(*ptep) || address >= TASK_SIZE)
505 		return;
506 
507 	/* We try to figure out if we are coming from an instruction
508 	 * access fault and pass that down to __hash_page so we avoid
509 	 * double-faulting on execution of fresh text. We have to test
510 	 * for regs NULL since init will get here first thing at boot
511 	 *
512 	 * We also avoid filling the hash if not coming from a fault
513 	 */
514 	if (current->thread.regs == NULL)
515 		return;
516 	trap = TRAP(current->thread.regs);
517 	if (trap == 0x400)
518 		access |= _PAGE_EXEC;
519 	else if (trap != 0x300)
520 		return;
521 	hash_preload(vma->vm_mm, address, access, trap);
522 #endif /* CONFIG_PPC_STD_MMU */
523 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
524 	&& defined(CONFIG_HUGETLB_PAGE)
525 	if (is_vm_hugetlb_page(vma))
526 		book3e_hugetlb_preload(vma, address, *ptep);
527 #endif
528 }
529 
530 /*
531  * System memory should not be in /proc/iomem but various tools expect it
532  * (eg kdump).
533  */
534 static int __init add_system_ram_resources(void)
535 {
536 	struct memblock_region *reg;
537 
538 	for_each_memblock(memory, reg) {
539 		struct resource *res;
540 		unsigned long base = reg->base;
541 		unsigned long size = reg->size;
542 
543 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
544 		WARN_ON(!res);
545 
546 		if (res) {
547 			res->name = "System RAM";
548 			res->start = base;
549 			res->end = base + size - 1;
550 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
551 			WARN_ON(request_resource(&iomem_resource, res) < 0);
552 		}
553 	}
554 
555 	return 0;
556 }
557 subsys_initcall(add_system_ram_resources);
558 
559 #ifdef CONFIG_STRICT_DEVMEM
560 /*
561  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
562  * is valid. The argument is a physical page number.
563  *
564  * Access has to be given to non-kernel-ram areas as well, these contain the
565  * PCI mmio resources as well as potential bios/acpi data regions.
566  */
567 int devmem_is_allowed(unsigned long pfn)
568 {
569 	if (page_is_rtas_user_buf(pfn))
570 		return 1;
571 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
572 		return 0;
573 	if (!page_is_ram(pfn))
574 		return 1;
575 	return 0;
576 }
577 #endif /* CONFIG_STRICT_DEVMEM */
578