xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision aac5987a) 
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 #define TOP_ZONE ZONE_HIGHMEM
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 #else
79 #define TOP_ZONE ZONE_NORMAL
80 #endif
81 
82 int page_is_ram(unsigned long pfn)
83 {
84 #ifndef CONFIG_PPC64	/* XXX for now */
85 	return pfn < max_pfn;
86 #else
87 	unsigned long paddr = (pfn << PAGE_SHIFT);
88 	struct memblock_region *reg;
89 
90 	for_each_memblock(memory, reg)
91 		if (paddr >= reg->base && paddr < (reg->base + reg->size))
92 			return 1;
93 	return 0;
94 #endif
95 }
96 
97 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
98 			      unsigned long size, pgprot_t vma_prot)
99 {
100 	if (ppc_md.phys_mem_access_prot)
101 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
102 
103 	if (!page_is_ram(pfn))
104 		vma_prot = pgprot_noncached(vma_prot);
105 
106 	return vma_prot;
107 }
108 EXPORT_SYMBOL(phys_mem_access_prot);
109 
110 #ifdef CONFIG_MEMORY_HOTPLUG
111 
112 #ifdef CONFIG_NUMA
113 int memory_add_physaddr_to_nid(u64 start)
114 {
115 	return hot_add_scn_to_nid(start);
116 }
117 #endif
118 
119 int __weak create_section_mapping(unsigned long start, unsigned long end)
120 {
121 	return -ENODEV;
122 }
123 
124 int __weak remove_section_mapping(unsigned long start, unsigned long end)
125 {
126 	return -ENODEV;
127 }
128 
129 int arch_add_memory(int nid, u64 start, u64 size, bool for_device)
130 {
131 	struct pglist_data *pgdata;
132 	struct zone *zone;
133 	unsigned long start_pfn = start >> PAGE_SHIFT;
134 	unsigned long nr_pages = size >> PAGE_SHIFT;
135 	int rc;
136 
137 	resize_hpt_for_hotplug(memblock_phys_mem_size());
138 
139 	pgdata = NODE_DATA(nid);
140 
141 	start = (unsigned long)__va(start);
142 	rc = create_section_mapping(start, start + size);
143 	if (rc) {
144 		pr_warning(
145 			"Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
146 			start, start + size, rc);
147 		return -EFAULT;
148 	}
149 
150 	/* this should work for most non-highmem platforms */
151 	zone = pgdata->node_zones +
152 		zone_for_memory(nid, start, size, 0, for_device);
153 
154 	return __add_pages(nid, zone, start_pfn, nr_pages);
155 }
156 
157 #ifdef CONFIG_MEMORY_HOTREMOVE
158 int arch_remove_memory(u64 start, u64 size)
159 {
160 	unsigned long start_pfn = start >> PAGE_SHIFT;
161 	unsigned long nr_pages = size >> PAGE_SHIFT;
162 	struct zone *zone;
163 	int ret;
164 
165 	zone = page_zone(pfn_to_page(start_pfn));
166 	ret = __remove_pages(zone, start_pfn, nr_pages);
167 	if (ret)
168 		return ret;
169 
170 	/* Remove htab bolted mappings for this section of memory */
171 	start = (unsigned long)__va(start);
172 	ret = remove_section_mapping(start, start + size);
173 
174 	/* Ensure all vmalloc mappings are flushed in case they also
175 	 * hit that section of memory
176 	 */
177 	vm_unmap_aliases();
178 
179 	resize_hpt_for_hotplug(memblock_phys_mem_size());
180 
181 	return ret;
182 }
183 #endif
184 #endif /* CONFIG_MEMORY_HOTPLUG */
185 
186 /*
187  * walk_memory_resource() needs to make sure there is no holes in a given
188  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
189  * Instead it maintains it in memblock.memory structures.  Walk through the
190  * memory regions, find holes and callback for contiguous regions.
191  */
192 int
193 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
194 		void *arg, int (*func)(unsigned long, unsigned long, void *))
195 {
196 	struct memblock_region *reg;
197 	unsigned long end_pfn = start_pfn + nr_pages;
198 	unsigned long tstart, tend;
199 	int ret = -1;
200 
201 	for_each_memblock(memory, reg) {
202 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
203 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
204 		if (tstart >= tend)
205 			continue;
206 		ret = (*func)(tstart, tend - tstart, arg);
207 		if (ret)
208 			break;
209 	}
210 	return ret;
211 }
212 EXPORT_SYMBOL_GPL(walk_system_ram_range);
213 
214 #ifndef CONFIG_NEED_MULTIPLE_NODES
215 void __init initmem_init(void)
216 {
217 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
218 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
219 #ifdef CONFIG_HIGHMEM
220 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
221 #endif
222 
223 	/* Place all memblock_regions in the same node and merge contiguous
224 	 * memblock_regions
225 	 */
226 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
227 
228 	/* XXX need to clip this if using highmem? */
229 	sparse_memory_present_with_active_regions(0);
230 	sparse_init();
231 }
232 
233 /* mark pages that don't exist as nosave */
234 static int __init mark_nonram_nosave(void)
235 {
236 	struct memblock_region *reg, *prev = NULL;
237 
238 	for_each_memblock(memory, reg) {
239 		if (prev &&
240 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
241 			register_nosave_region(memblock_region_memory_end_pfn(prev),
242 					       memblock_region_memory_base_pfn(reg));
243 		prev = reg;
244 	}
245 	return 0;
246 }
247 #else /* CONFIG_NEED_MULTIPLE_NODES */
248 static int __init mark_nonram_nosave(void)
249 {
250 	return 0;
251 }
252 #endif
253 
254 static bool zone_limits_final;
255 
256 /*
257  * The memory zones past TOP_ZONE are managed by generic mm code.
258  * These should be set to zero since that's what every other
259  * architecture does.
260  */
261 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
262 	[0            ... TOP_ZONE        ] = ~0UL,
263 	[TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
264 };
265 
266 /*
267  * Restrict the specified zone and all more restrictive zones
268  * to be below the specified pfn.  May not be called after
269  * paging_init().
270  */
271 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
272 {
273 	int i;
274 
275 	if (WARN_ON(zone_limits_final))
276 		return;
277 
278 	for (i = zone; i >= 0; i--) {
279 		if (max_zone_pfns[i] > pfn_limit)
280 			max_zone_pfns[i] = pfn_limit;
281 	}
282 }
283 
284 /*
285  * Find the least restrictive zone that is entirely below the
286  * specified pfn limit.  Returns < 0 if no suitable zone is found.
287  *
288  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
289  * systems -- the DMA limit can be higher than any possible real pfn.
290  */
291 int dma_pfn_limit_to_zone(u64 pfn_limit)
292 {
293 	int i;
294 
295 	for (i = TOP_ZONE; i >= 0; i--) {
296 		if (max_zone_pfns[i] <= pfn_limit)
297 			return i;
298 	}
299 
300 	return -EPERM;
301 }
302 
303 /*
304  * paging_init() sets up the page tables - in fact we've already done this.
305  */
306 void __init paging_init(void)
307 {
308 	unsigned long long total_ram = memblock_phys_mem_size();
309 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
310 
311 #ifdef CONFIG_PPC32
312 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
313 	unsigned long end = __fix_to_virt(FIX_HOLE);
314 
315 	for (; v < end; v += PAGE_SIZE)
316 		map_page(v, 0, 0); /* XXX gross */
317 #endif
318 
319 #ifdef CONFIG_HIGHMEM
320 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
321 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
322 
323 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
324 	kmap_prot = PAGE_KERNEL;
325 #endif /* CONFIG_HIGHMEM */
326 
327 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
328 	       (unsigned long long)top_of_ram, total_ram);
329 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
330 	       (long int)((top_of_ram - total_ram) >> 20));
331 
332 #ifdef CONFIG_HIGHMEM
333 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
334 #endif
335 	limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
336 	zone_limits_final = true;
337 	free_area_init_nodes(max_zone_pfns);
338 
339 	mark_nonram_nosave();
340 }
341 
342 void __init mem_init(void)
343 {
344 	/*
345 	 * book3s is limited to 16 page sizes due to encoding this in
346 	 * a 4-bit field for slices.
347 	 */
348 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
349 
350 #ifdef CONFIG_SWIOTLB
351 	swiotlb_init(0);
352 #endif
353 
354 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
355 	set_max_mapnr(max_pfn);
356 	free_all_bootmem();
357 
358 #ifdef CONFIG_HIGHMEM
359 	{
360 		unsigned long pfn, highmem_mapnr;
361 
362 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
363 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
364 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
365 			struct page *page = pfn_to_page(pfn);
366 			if (!memblock_is_reserved(paddr))
367 				free_highmem_page(page);
368 		}
369 	}
370 #endif /* CONFIG_HIGHMEM */
371 
372 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
373 	/*
374 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
375 	 * functions.... do it here for the non-smp case.
376 	 */
377 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
378 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
379 #endif
380 
381 	mem_init_print_info(NULL);
382 #ifdef CONFIG_PPC32
383 	pr_info("Kernel virtual memory layout:\n");
384 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
385 #ifdef CONFIG_HIGHMEM
386 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
387 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
388 #endif /* CONFIG_HIGHMEM */
389 #ifdef CONFIG_NOT_COHERENT_CACHE
390 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
391 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
392 #endif /* CONFIG_NOT_COHERENT_CACHE */
393 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
394 		ioremap_bot, IOREMAP_TOP);
395 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
396 		VMALLOC_START, VMALLOC_END);
397 #endif /* CONFIG_PPC32 */
398 }
399 
400 void free_initmem(void)
401 {
402 	ppc_md.progress = ppc_printk_progress;
403 	free_initmem_default(POISON_FREE_INITMEM);
404 }
405 
406 #ifdef CONFIG_BLK_DEV_INITRD
407 void __init free_initrd_mem(unsigned long start, unsigned long end)
408 {
409 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
410 }
411 #endif
412 
413 /*
414  * This is called when a page has been modified by the kernel.
415  * It just marks the page as not i-cache clean.  We do the i-cache
416  * flush later when the page is given to a user process, if necessary.
417  */
418 void flush_dcache_page(struct page *page)
419 {
420 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
421 		return;
422 	/* avoid an atomic op if possible */
423 	if (test_bit(PG_arch_1, &page->flags))
424 		clear_bit(PG_arch_1, &page->flags);
425 }
426 EXPORT_SYMBOL(flush_dcache_page);
427 
428 void flush_dcache_icache_page(struct page *page)
429 {
430 #ifdef CONFIG_HUGETLB_PAGE
431 	if (PageCompound(page)) {
432 		flush_dcache_icache_hugepage(page);
433 		return;
434 	}
435 #endif
436 #if defined(CONFIG_8xx) || defined(CONFIG_PPC64)
437 	/* On 8xx there is no need to kmap since highmem is not supported */
438 	__flush_dcache_icache(page_address(page));
439 #else
440 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
441 		void *start = kmap_atomic(page);
442 		__flush_dcache_icache(start);
443 		kunmap_atomic(start);
444 	} else {
445 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
446 	}
447 #endif
448 }
449 EXPORT_SYMBOL(flush_dcache_icache_page);
450 
451 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
452 {
453 	clear_page(page);
454 
455 	/*
456 	 * We shouldn't have to do this, but some versions of glibc
457 	 * require it (ld.so assumes zero filled pages are icache clean)
458 	 * - Anton
459 	 */
460 	flush_dcache_page(pg);
461 }
462 EXPORT_SYMBOL(clear_user_page);
463 
464 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
465 		    struct page *pg)
466 {
467 	copy_page(vto, vfrom);
468 
469 	/*
470 	 * We should be able to use the following optimisation, however
471 	 * there are two problems.
472 	 * Firstly a bug in some versions of binutils meant PLT sections
473 	 * were not marked executable.
474 	 * Secondly the first word in the GOT section is blrl, used
475 	 * to establish the GOT address. Until recently the GOT was
476 	 * not marked executable.
477 	 * - Anton
478 	 */
479 #if 0
480 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
481 		return;
482 #endif
483 
484 	flush_dcache_page(pg);
485 }
486 
487 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
488 			     unsigned long addr, int len)
489 {
490 	unsigned long maddr;
491 
492 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
493 	flush_icache_range(maddr, maddr + len);
494 	kunmap(page);
495 }
496 EXPORT_SYMBOL(flush_icache_user_range);
497 
498 /*
499  * This is called at the end of handling a user page fault, when the
500  * fault has been handled by updating a PTE in the linux page tables.
501  * We use it to preload an HPTE into the hash table corresponding to
502  * the updated linux PTE.
503  *
504  * This must always be called with the pte lock held.
505  */
506 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
507 		      pte_t *ptep)
508 {
509 #ifdef CONFIG_PPC_STD_MMU
510 	/*
511 	 * We don't need to worry about _PAGE_PRESENT here because we are
512 	 * called with either mm->page_table_lock held or ptl lock held
513 	 */
514 	unsigned long access, trap;
515 
516 	if (radix_enabled())
517 		return;
518 
519 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
520 	if (!pte_young(*ptep) || address >= TASK_SIZE)
521 		return;
522 
523 	/* We try to figure out if we are coming from an instruction
524 	 * access fault and pass that down to __hash_page so we avoid
525 	 * double-faulting on execution of fresh text. We have to test
526 	 * for regs NULL since init will get here first thing at boot
527 	 *
528 	 * We also avoid filling the hash if not coming from a fault
529 	 */
530 
531 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
532 	switch (trap) {
533 	case 0x300:
534 		access = 0UL;
535 		break;
536 	case 0x400:
537 		access = _PAGE_EXEC;
538 		break;
539 	default:
540 		return;
541 	}
542 
543 	hash_preload(vma->vm_mm, address, access, trap);
544 #endif /* CONFIG_PPC_STD_MMU */
545 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
546 	&& defined(CONFIG_HUGETLB_PAGE)
547 	if (is_vm_hugetlb_page(vma))
548 		book3e_hugetlb_preload(vma, address, *ptep);
549 #endif
550 }
551 
552 /*
553  * System memory should not be in /proc/iomem but various tools expect it
554  * (eg kdump).
555  */
556 static int __init add_system_ram_resources(void)
557 {
558 	struct memblock_region *reg;
559 
560 	for_each_memblock(memory, reg) {
561 		struct resource *res;
562 		unsigned long base = reg->base;
563 		unsigned long size = reg->size;
564 
565 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
566 		WARN_ON(!res);
567 
568 		if (res) {
569 			res->name = "System RAM";
570 			res->start = base;
571 			res->end = base + size - 1;
572 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
573 			WARN_ON(request_resource(&iomem_resource, res) < 0);
574 		}
575 	}
576 
577 	return 0;
578 }
579 subsys_initcall(add_system_ram_resources);
580 
581 #ifdef CONFIG_STRICT_DEVMEM
582 /*
583  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
584  * is valid. The argument is a physical page number.
585  *
586  * Access has to be given to non-kernel-ram areas as well, these contain the
587  * PCI mmio resources as well as potential bios/acpi data regions.
588  */
589 int devmem_is_allowed(unsigned long pfn)
590 {
591 	if (page_is_rtas_user_buf(pfn))
592 		return 1;
593 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
594 		return 0;
595 	if (!page_is_ram(pfn))
596 		return 1;
597 	return 0;
598 }
599 #endif /* CONFIG_STRICT_DEVMEM */
600