xref: /linux/arch/powerpc/mm/mem.c (revision 55f3538c4923e9dfca132e99ebec370e8094afda) 
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 #include <linux/memremap.h>
40 
41 #include <asm/pgalloc.h>
42 #include <asm/prom.h>
43 #include <asm/io.h>
44 #include <asm/mmu_context.h>
45 #include <asm/pgtable.h>
46 #include <asm/mmu.h>
47 #include <asm/smp.h>
48 #include <asm/machdep.h>
49 #include <asm/btext.h>
50 #include <asm/tlb.h>
51 #include <asm/sections.h>
52 #include <asm/sparsemem.h>
53 #include <asm/vdso.h>
54 #include <asm/fixmap.h>
55 #include <asm/swiotlb.h>
56 #include <asm/rtas.h>
57 
58 #include "mmu_decl.h"
59 
60 #ifndef CPU_FTR_COHERENT_ICACHE
61 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
62 #define CPU_FTR_NOEXECUTE	0
63 #endif
64 
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 #define TOP_ZONE ZONE_HIGHMEM
73 
74 static inline pte_t *virt_to_kpte(unsigned long vaddr)
75 {
76 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
77 			vaddr), vaddr), vaddr);
78 }
79 #else
80 #define TOP_ZONE ZONE_NORMAL
81 #endif
82 
83 int page_is_ram(unsigned long pfn)
84 {
85 #ifndef CONFIG_PPC64	/* XXX for now */
86 	return pfn < max_pfn;
87 #else
88 	unsigned long paddr = (pfn << PAGE_SHIFT);
89 	struct memblock_region *reg;
90 
91 	for_each_memblock(memory, reg)
92 		if (paddr >= reg->base && paddr < (reg->base + reg->size))
93 			return 1;
94 	return 0;
95 #endif
96 }
97 
98 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
99 			      unsigned long size, pgprot_t vma_prot)
100 {
101 	if (ppc_md.phys_mem_access_prot)
102 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
103 
104 	if (!page_is_ram(pfn))
105 		vma_prot = pgprot_noncached(vma_prot);
106 
107 	return vma_prot;
108 }
109 EXPORT_SYMBOL(phys_mem_access_prot);
110 
111 #ifdef CONFIG_MEMORY_HOTPLUG
112 
113 #ifdef CONFIG_NUMA
114 int memory_add_physaddr_to_nid(u64 start)
115 {
116 	return hot_add_scn_to_nid(start);
117 }
118 #endif
119 
120 int __weak create_section_mapping(unsigned long start, unsigned long end)
121 {
122 	return -ENODEV;
123 }
124 
125 int __weak remove_section_mapping(unsigned long start, unsigned long end)
126 {
127 	return -ENODEV;
128 }
129 
130 int arch_add_memory(int nid, u64 start, u64 size, bool want_memblock)
131 {
132 	unsigned long start_pfn = start >> PAGE_SHIFT;
133 	unsigned long nr_pages = size >> PAGE_SHIFT;
134 	int rc;
135 
136 	resize_hpt_for_hotplug(memblock_phys_mem_size());
137 
138 	start = (unsigned long)__va(start);
139 	rc = create_section_mapping(start, start + size);
140 	if (rc) {
141 		pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
142 			start, start + size, rc);
143 		return -EFAULT;
144 	}
145 
146 	return __add_pages(nid, start_pfn, nr_pages, want_memblock);
147 }
148 
149 #ifdef CONFIG_MEMORY_HOTREMOVE
150 int arch_remove_memory(u64 start, u64 size)
151 {
152 	unsigned long start_pfn = start >> PAGE_SHIFT;
153 	unsigned long nr_pages = size >> PAGE_SHIFT;
154 	struct vmem_altmap *altmap;
155 	struct page *page;
156 	int ret;
157 
158 	/*
159 	 * If we have an altmap then we need to skip over any reserved PFNs
160 	 * when querying the zone.
161 	 */
162 	page = pfn_to_page(start_pfn);
163 	altmap = to_vmem_altmap((unsigned long) page);
164 	if (altmap)
165 		page += vmem_altmap_offset(altmap);
166 
167 	ret = __remove_pages(page_zone(page), start_pfn, nr_pages);
168 	if (ret)
169 		return ret;
170 
171 	/* Remove htab bolted mappings for this section of memory */
172 	start = (unsigned long)__va(start);
173 	ret = remove_section_mapping(start, start + size);
174 
175 	/* Ensure all vmalloc mappings are flushed in case they also
176 	 * hit that section of memory
177 	 */
178 	vm_unmap_aliases();
179 
180 	resize_hpt_for_hotplug(memblock_phys_mem_size());
181 
182 	return ret;
183 }
184 #endif
185 #endif /* CONFIG_MEMORY_HOTPLUG */
186 
187 /*
188  * walk_memory_resource() needs to make sure there is no holes in a given
189  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
190  * Instead it maintains it in memblock.memory structures.  Walk through the
191  * memory regions, find holes and callback for contiguous regions.
192  */
193 int
194 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
195 		void *arg, int (*func)(unsigned long, unsigned long, void *))
196 {
197 	struct memblock_region *reg;
198 	unsigned long end_pfn = start_pfn + nr_pages;
199 	unsigned long tstart, tend;
200 	int ret = -1;
201 
202 	for_each_memblock(memory, reg) {
203 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
204 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
205 		if (tstart >= tend)
206 			continue;
207 		ret = (*func)(tstart, tend - tstart, arg);
208 		if (ret)
209 			break;
210 	}
211 	return ret;
212 }
213 EXPORT_SYMBOL_GPL(walk_system_ram_range);
214 
215 #ifndef CONFIG_NEED_MULTIPLE_NODES
216 void __init initmem_init(void)
217 {
218 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
219 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
220 #ifdef CONFIG_HIGHMEM
221 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
222 #endif
223 
224 	/* Place all memblock_regions in the same node and merge contiguous
225 	 * memblock_regions
226 	 */
227 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
228 
229 	/* XXX need to clip this if using highmem? */
230 	sparse_memory_present_with_active_regions(0);
231 	sparse_init();
232 }
233 
234 /* mark pages that don't exist as nosave */
235 static int __init mark_nonram_nosave(void)
236 {
237 	struct memblock_region *reg, *prev = NULL;
238 
239 	for_each_memblock(memory, reg) {
240 		if (prev &&
241 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
242 			register_nosave_region(memblock_region_memory_end_pfn(prev),
243 					       memblock_region_memory_base_pfn(reg));
244 		prev = reg;
245 	}
246 	return 0;
247 }
248 #else /* CONFIG_NEED_MULTIPLE_NODES */
249 static int __init mark_nonram_nosave(void)
250 {
251 	return 0;
252 }
253 #endif
254 
255 static bool zone_limits_final;
256 
257 /*
258  * The memory zones past TOP_ZONE are managed by generic mm code.
259  * These should be set to zero since that's what every other
260  * architecture does.
261  */
262 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
263 	[0            ... TOP_ZONE        ] = ~0UL,
264 	[TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
265 };
266 
267 /*
268  * Restrict the specified zone and all more restrictive zones
269  * to be below the specified pfn.  May not be called after
270  * paging_init().
271  */
272 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
273 {
274 	int i;
275 
276 	if (WARN_ON(zone_limits_final))
277 		return;
278 
279 	for (i = zone; i >= 0; i--) {
280 		if (max_zone_pfns[i] > pfn_limit)
281 			max_zone_pfns[i] = pfn_limit;
282 	}
283 }
284 
285 /*
286  * Find the least restrictive zone that is entirely below the
287  * specified pfn limit.  Returns < 0 if no suitable zone is found.
288  *
289  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
290  * systems -- the DMA limit can be higher than any possible real pfn.
291  */
292 int dma_pfn_limit_to_zone(u64 pfn_limit)
293 {
294 	int i;
295 
296 	for (i = TOP_ZONE; i >= 0; i--) {
297 		if (max_zone_pfns[i] <= pfn_limit)
298 			return i;
299 	}
300 
301 	return -EPERM;
302 }
303 
304 /*
305  * paging_init() sets up the page tables - in fact we've already done this.
306  */
307 void __init paging_init(void)
308 {
309 	unsigned long long total_ram = memblock_phys_mem_size();
310 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
311 
312 #ifdef CONFIG_PPC32
313 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
314 	unsigned long end = __fix_to_virt(FIX_HOLE);
315 
316 	for (; v < end; v += PAGE_SIZE)
317 		map_kernel_page(v, 0, 0); /* XXX gross */
318 #endif
319 
320 #ifdef CONFIG_HIGHMEM
321 	map_kernel_page(PKMAP_BASE, 0, 0);	/* XXX gross */
322 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
323 
324 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
325 	kmap_prot = PAGE_KERNEL;
326 #endif /* CONFIG_HIGHMEM */
327 
328 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
329 	       (unsigned long long)top_of_ram, total_ram);
330 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
331 	       (long int)((top_of_ram - total_ram) >> 20));
332 
333 #ifdef CONFIG_HIGHMEM
334 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
335 #endif
336 	limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
337 	zone_limits_final = true;
338 	free_area_init_nodes(max_zone_pfns);
339 
340 	mark_nonram_nosave();
341 }
342 
343 void __init mem_init(void)
344 {
345 	/*
346 	 * book3s is limited to 16 page sizes due to encoding this in
347 	 * a 4-bit field for slices.
348 	 */
349 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
350 
351 #ifdef CONFIG_SWIOTLB
352 	swiotlb_init(0);
353 #endif
354 
355 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
356 	set_max_mapnr(max_pfn);
357 	free_all_bootmem();
358 
359 #ifdef CONFIG_HIGHMEM
360 	{
361 		unsigned long pfn, highmem_mapnr;
362 
363 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
364 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
365 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
366 			struct page *page = pfn_to_page(pfn);
367 			if (!memblock_is_reserved(paddr))
368 				free_highmem_page(page);
369 		}
370 	}
371 #endif /* CONFIG_HIGHMEM */
372 
373 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
374 	/*
375 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
376 	 * functions.... do it here for the non-smp case.
377 	 */
378 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
379 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
380 #endif
381 
382 	mem_init_print_info(NULL);
383 #ifdef CONFIG_PPC32
384 	pr_info("Kernel virtual memory layout:\n");
385 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
386 #ifdef CONFIG_HIGHMEM
387 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
388 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
389 #endif /* CONFIG_HIGHMEM */
390 #ifdef CONFIG_NOT_COHERENT_CACHE
391 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
392 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
393 #endif /* CONFIG_NOT_COHERENT_CACHE */
394 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
395 		ioremap_bot, IOREMAP_TOP);
396 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
397 		VMALLOC_START, VMALLOC_END);
398 #endif /* CONFIG_PPC32 */
399 }
400 
401 void free_initmem(void)
402 {
403 	ppc_md.progress = ppc_printk_progress;
404 	mark_initmem_nx();
405 	free_initmem_default(POISON_FREE_INITMEM);
406 }
407 
408 #ifdef CONFIG_BLK_DEV_INITRD
409 void __init free_initrd_mem(unsigned long start, unsigned long end)
410 {
411 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
412 }
413 #endif
414 
415 /*
416  * This is called when a page has been modified by the kernel.
417  * It just marks the page as not i-cache clean.  We do the i-cache
418  * flush later when the page is given to a user process, if necessary.
419  */
420 void flush_dcache_page(struct page *page)
421 {
422 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
423 		return;
424 	/* avoid an atomic op if possible */
425 	if (test_bit(PG_arch_1, &page->flags))
426 		clear_bit(PG_arch_1, &page->flags);
427 }
428 EXPORT_SYMBOL(flush_dcache_page);
429 
430 void flush_dcache_icache_page(struct page *page)
431 {
432 #ifdef CONFIG_HUGETLB_PAGE
433 	if (PageCompound(page)) {
434 		flush_dcache_icache_hugepage(page);
435 		return;
436 	}
437 #endif
438 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
439 	/* On 8xx there is no need to kmap since highmem is not supported */
440 	__flush_dcache_icache(page_address(page));
441 #else
442 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
443 		void *start = kmap_atomic(page);
444 		__flush_dcache_icache(start);
445 		kunmap_atomic(start);
446 	} else {
447 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
448 	}
449 #endif
450 }
451 EXPORT_SYMBOL(flush_dcache_icache_page);
452 
453 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
454 {
455 	clear_page(page);
456 
457 	/*
458 	 * We shouldn't have to do this, but some versions of glibc
459 	 * require it (ld.so assumes zero filled pages are icache clean)
460 	 * - Anton
461 	 */
462 	flush_dcache_page(pg);
463 }
464 EXPORT_SYMBOL(clear_user_page);
465 
466 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
467 		    struct page *pg)
468 {
469 	copy_page(vto, vfrom);
470 
471 	/*
472 	 * We should be able to use the following optimisation, however
473 	 * there are two problems.
474 	 * Firstly a bug in some versions of binutils meant PLT sections
475 	 * were not marked executable.
476 	 * Secondly the first word in the GOT section is blrl, used
477 	 * to establish the GOT address. Until recently the GOT was
478 	 * not marked executable.
479 	 * - Anton
480 	 */
481 #if 0
482 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
483 		return;
484 #endif
485 
486 	flush_dcache_page(pg);
487 }
488 
489 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
490 			     unsigned long addr, int len)
491 {
492 	unsigned long maddr;
493 
494 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
495 	flush_icache_range(maddr, maddr + len);
496 	kunmap(page);
497 }
498 EXPORT_SYMBOL(flush_icache_user_range);
499 
500 /*
501  * This is called at the end of handling a user page fault, when the
502  * fault has been handled by updating a PTE in the linux page tables.
503  * We use it to preload an HPTE into the hash table corresponding to
504  * the updated linux PTE.
505  *
506  * This must always be called with the pte lock held.
507  */
508 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
509 		      pte_t *ptep)
510 {
511 #ifdef CONFIG_PPC_STD_MMU
512 	/*
513 	 * We don't need to worry about _PAGE_PRESENT here because we are
514 	 * called with either mm->page_table_lock held or ptl lock held
515 	 */
516 	unsigned long access, trap;
517 
518 	if (radix_enabled())
519 		return;
520 
521 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
522 	if (!pte_young(*ptep) || address >= TASK_SIZE)
523 		return;
524 
525 	/* We try to figure out if we are coming from an instruction
526 	 * access fault and pass that down to __hash_page so we avoid
527 	 * double-faulting on execution of fresh text. We have to test
528 	 * for regs NULL since init will get here first thing at boot
529 	 *
530 	 * We also avoid filling the hash if not coming from a fault
531 	 */
532 
533 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
534 	switch (trap) {
535 	case 0x300:
536 		access = 0UL;
537 		break;
538 	case 0x400:
539 		access = _PAGE_EXEC;
540 		break;
541 	default:
542 		return;
543 	}
544 
545 	hash_preload(vma->vm_mm, address, access, trap);
546 #endif /* CONFIG_PPC_STD_MMU */
547 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
548 	&& defined(CONFIG_HUGETLB_PAGE)
549 	if (is_vm_hugetlb_page(vma))
550 		book3e_hugetlb_preload(vma, address, *ptep);
551 #endif
552 }
553 
554 /*
555  * System memory should not be in /proc/iomem but various tools expect it
556  * (eg kdump).
557  */
558 static int __init add_system_ram_resources(void)
559 {
560 	struct memblock_region *reg;
561 
562 	for_each_memblock(memory, reg) {
563 		struct resource *res;
564 		unsigned long base = reg->base;
565 		unsigned long size = reg->size;
566 
567 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
568 		WARN_ON(!res);
569 
570 		if (res) {
571 			res->name = "System RAM";
572 			res->start = base;
573 			res->end = base + size - 1;
574 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
575 			WARN_ON(request_resource(&iomem_resource, res) < 0);
576 		}
577 	}
578 
579 	return 0;
580 }
581 subsys_initcall(add_system_ram_resources);
582 
583 #ifdef CONFIG_STRICT_DEVMEM
584 /*
585  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
586  * is valid. The argument is a physical page number.
587  *
588  * Access has to be given to non-kernel-ram areas as well, these contain the
589  * PCI mmio resources as well as potential bios/acpi data regions.
590  */
591 int devmem_is_allowed(unsigned long pfn)
592 {
593 	if (page_is_rtas_user_buf(pfn))
594 		return 1;
595 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
596 		return 0;
597 	if (!page_is_ram(pfn))
598 		return 1;
599 	return 0;
600 }
601 #endif /* CONFIG_STRICT_DEVMEM */
602