xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 23c2b932) 
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 arch_add_memory(int nid, u64 start, u64 size, bool for_device)
120 {
121 	struct pglist_data *pgdata;
122 	struct zone *zone;
123 	unsigned long start_pfn = start >> PAGE_SHIFT;
124 	unsigned long nr_pages = size >> PAGE_SHIFT;
125 	int rc;
126 
127 	pgdata = NODE_DATA(nid);
128 
129 	start = (unsigned long)__va(start);
130 	rc = create_section_mapping(start, start + size);
131 	if (rc) {
132 		pr_warning(
133 			"Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
134 			start, start + size, rc);
135 		return -EFAULT;
136 	}
137 
138 	/* this should work for most non-highmem platforms */
139 	zone = pgdata->node_zones +
140 		zone_for_memory(nid, start, size, 0, for_device);
141 
142 	return __add_pages(nid, zone, start_pfn, nr_pages);
143 }
144 
145 #ifdef CONFIG_MEMORY_HOTREMOVE
146 int arch_remove_memory(u64 start, u64 size)
147 {
148 	unsigned long start_pfn = start >> PAGE_SHIFT;
149 	unsigned long nr_pages = size >> PAGE_SHIFT;
150 	struct zone *zone;
151 	int ret;
152 
153 	zone = page_zone(pfn_to_page(start_pfn));
154 	ret = __remove_pages(zone, start_pfn, nr_pages);
155 	if (ret)
156 		return ret;
157 
158 	/* Remove htab bolted mappings for this section of memory */
159 	start = (unsigned long)__va(start);
160 	ret = remove_section_mapping(start, start + size);
161 
162 	/* Ensure all vmalloc mappings are flushed in case they also
163 	 * hit that section of memory
164 	 */
165 	vm_unmap_aliases();
166 
167 	return ret;
168 }
169 #endif
170 #endif /* CONFIG_MEMORY_HOTPLUG */
171 
172 /*
173  * walk_memory_resource() needs to make sure there is no holes in a given
174  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
175  * Instead it maintains it in memblock.memory structures.  Walk through the
176  * memory regions, find holes and callback for contiguous regions.
177  */
178 int
179 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
180 		void *arg, int (*func)(unsigned long, unsigned long, void *))
181 {
182 	struct memblock_region *reg;
183 	unsigned long end_pfn = start_pfn + nr_pages;
184 	unsigned long tstart, tend;
185 	int ret = -1;
186 
187 	for_each_memblock(memory, reg) {
188 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
189 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
190 		if (tstart >= tend)
191 			continue;
192 		ret = (*func)(tstart, tend - tstart, arg);
193 		if (ret)
194 			break;
195 	}
196 	return ret;
197 }
198 EXPORT_SYMBOL_GPL(walk_system_ram_range);
199 
200 #ifndef CONFIG_NEED_MULTIPLE_NODES
201 void __init initmem_init(void)
202 {
203 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
204 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
205 #ifdef CONFIG_HIGHMEM
206 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
207 #endif
208 
209 	/* Place all memblock_regions in the same node and merge contiguous
210 	 * memblock_regions
211 	 */
212 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
213 
214 	/* XXX need to clip this if using highmem? */
215 	sparse_memory_present_with_active_regions(0);
216 	sparse_init();
217 }
218 
219 /* mark pages that don't exist as nosave */
220 static int __init mark_nonram_nosave(void)
221 {
222 	struct memblock_region *reg, *prev = NULL;
223 
224 	for_each_memblock(memory, reg) {
225 		if (prev &&
226 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
227 			register_nosave_region(memblock_region_memory_end_pfn(prev),
228 					       memblock_region_memory_base_pfn(reg));
229 		prev = reg;
230 	}
231 	return 0;
232 }
233 #else /* CONFIG_NEED_MULTIPLE_NODES */
234 static int __init mark_nonram_nosave(void)
235 {
236 	return 0;
237 }
238 #endif
239 
240 static bool zone_limits_final;
241 
242 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
243 	[0 ... MAX_NR_ZONES - 1] = ~0UL
244 };
245 
246 /*
247  * Restrict the specified zone and all more restrictive zones
248  * to be below the specified pfn.  May not be called after
249  * paging_init().
250  */
251 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
252 {
253 	int i;
254 
255 	if (WARN_ON(zone_limits_final))
256 		return;
257 
258 	for (i = zone; i >= 0; i--) {
259 		if (max_zone_pfns[i] > pfn_limit)
260 			max_zone_pfns[i] = pfn_limit;
261 	}
262 }
263 
264 /*
265  * Find the least restrictive zone that is entirely below the
266  * specified pfn limit.  Returns < 0 if no suitable zone is found.
267  *
268  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
269  * systems -- the DMA limit can be higher than any possible real pfn.
270  */
271 int dma_pfn_limit_to_zone(u64 pfn_limit)
272 {
273 	int i;
274 
275 	for (i = TOP_ZONE; i >= 0; i--) {
276 		if (max_zone_pfns[i] <= pfn_limit)
277 			return i;
278 	}
279 
280 	return -EPERM;
281 }
282 
283 /*
284  * paging_init() sets up the page tables - in fact we've already done this.
285  */
286 void __init paging_init(void)
287 {
288 	unsigned long long total_ram = memblock_phys_mem_size();
289 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
290 
291 #ifdef CONFIG_PPC32
292 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
293 	unsigned long end = __fix_to_virt(FIX_HOLE);
294 
295 	for (; v < end; v += PAGE_SIZE)
296 		map_page(v, 0, 0); /* XXX gross */
297 #endif
298 
299 #ifdef CONFIG_HIGHMEM
300 	map_page(PKMAP_BASE, 0, 0);	/* XXX gross */
301 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
302 
303 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
304 	kmap_prot = PAGE_KERNEL;
305 #endif /* CONFIG_HIGHMEM */
306 
307 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
308 	       (unsigned long long)top_of_ram, total_ram);
309 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
310 	       (long int)((top_of_ram - total_ram) >> 20));
311 
312 #ifdef CONFIG_HIGHMEM
313 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
314 #endif
315 	limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
316 	zone_limits_final = true;
317 	free_area_init_nodes(max_zone_pfns);
318 
319 	mark_nonram_nosave();
320 }
321 
322 void __init mem_init(void)
323 {
324 	/*
325 	 * book3s is limited to 16 page sizes due to encoding this in
326 	 * a 4-bit field for slices.
327 	 */
328 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
329 
330 #ifdef CONFIG_SWIOTLB
331 	swiotlb_init(0);
332 #endif
333 
334 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
335 	set_max_mapnr(max_pfn);
336 	free_all_bootmem();
337 
338 #ifdef CONFIG_HIGHMEM
339 	{
340 		unsigned long pfn, highmem_mapnr;
341 
342 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
343 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
344 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
345 			struct page *page = pfn_to_page(pfn);
346 			if (!memblock_is_reserved(paddr))
347 				free_highmem_page(page);
348 		}
349 	}
350 #endif /* CONFIG_HIGHMEM */
351 
352 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
353 	/*
354 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
355 	 * functions.... do it here for the non-smp case.
356 	 */
357 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
358 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
359 #endif
360 
361 	mem_init_print_info(NULL);
362 #ifdef CONFIG_PPC32
363 	pr_info("Kernel virtual memory layout:\n");
364 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
365 #ifdef CONFIG_HIGHMEM
366 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
367 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
368 #endif /* CONFIG_HIGHMEM */
369 #ifdef CONFIG_NOT_COHERENT_CACHE
370 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
371 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
372 #endif /* CONFIG_NOT_COHERENT_CACHE */
373 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
374 		ioremap_bot, IOREMAP_TOP);
375 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
376 		VMALLOC_START, VMALLOC_END);
377 #endif /* CONFIG_PPC32 */
378 }
379 
380 void free_initmem(void)
381 {
382 	ppc_md.progress = ppc_printk_progress;
383 	free_initmem_default(POISON_FREE_INITMEM);
384 }
385 
386 #ifdef CONFIG_BLK_DEV_INITRD
387 void __init free_initrd_mem(unsigned long start, unsigned long end)
388 {
389 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
390 }
391 #endif
392 
393 /*
394  * This is called when a page has been modified by the kernel.
395  * It just marks the page as not i-cache clean.  We do the i-cache
396  * flush later when the page is given to a user process, if necessary.
397  */
398 void flush_dcache_page(struct page *page)
399 {
400 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
401 		return;
402 	/* avoid an atomic op if possible */
403 	if (test_bit(PG_arch_1, &page->flags))
404 		clear_bit(PG_arch_1, &page->flags);
405 }
406 EXPORT_SYMBOL(flush_dcache_page);
407 
408 void flush_dcache_icache_page(struct page *page)
409 {
410 #ifdef CONFIG_HUGETLB_PAGE
411 	if (PageCompound(page)) {
412 		flush_dcache_icache_hugepage(page);
413 		return;
414 	}
415 #endif
416 #if 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 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
421 		void *start = kmap_atomic(page);
422 		__flush_dcache_icache(start);
423 		kunmap_atomic(start);
424 	} else {
425 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
426 	}
427 #endif
428 }
429 EXPORT_SYMBOL(flush_dcache_icache_page);
430 
431 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
432 {
433 	clear_page(page);
434 
435 	/*
436 	 * We shouldn't have to do this, but some versions of glibc
437 	 * require it (ld.so assumes zero filled pages are icache clean)
438 	 * - Anton
439 	 */
440 	flush_dcache_page(pg);
441 }
442 EXPORT_SYMBOL(clear_user_page);
443 
444 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
445 		    struct page *pg)
446 {
447 	copy_page(vto, vfrom);
448 
449 	/*
450 	 * We should be able to use the following optimisation, however
451 	 * there are two problems.
452 	 * Firstly a bug in some versions of binutils meant PLT sections
453 	 * were not marked executable.
454 	 * Secondly the first word in the GOT section is blrl, used
455 	 * to establish the GOT address. Until recently the GOT was
456 	 * not marked executable.
457 	 * - Anton
458 	 */
459 #if 0
460 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
461 		return;
462 #endif
463 
464 	flush_dcache_page(pg);
465 }
466 
467 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
468 			     unsigned long addr, int len)
469 {
470 	unsigned long maddr;
471 
472 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
473 	flush_icache_range(maddr, maddr + len);
474 	kunmap(page);
475 }
476 EXPORT_SYMBOL(flush_icache_user_range);
477 
478 /*
479  * This is called at the end of handling a user page fault, when the
480  * fault has been handled by updating a PTE in the linux page tables.
481  * We use it to preload an HPTE into the hash table corresponding to
482  * the updated linux PTE.
483  *
484  * This must always be called with the pte lock held.
485  */
486 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
487 		      pte_t *ptep)
488 {
489 #ifdef CONFIG_PPC_STD_MMU
490 	/*
491 	 * We don't need to worry about _PAGE_PRESENT here because we are
492 	 * called with either mm->page_table_lock held or ptl lock held
493 	 */
494 	unsigned long access, trap;
495 
496 	if (radix_enabled())
497 		return;
498 
499 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
500 	if (!pte_young(*ptep) || address >= TASK_SIZE)
501 		return;
502 
503 	/* We try to figure out if we are coming from an instruction
504 	 * access fault and pass that down to __hash_page so we avoid
505 	 * double-faulting on execution of fresh text. We have to test
506 	 * for regs NULL since init will get here first thing at boot
507 	 *
508 	 * We also avoid filling the hash if not coming from a fault
509 	 */
510 
511 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
512 	switch (trap) {
513 	case 0x300:
514 		access = 0UL;
515 		break;
516 	case 0x400:
517 		access = _PAGE_EXEC;
518 		break;
519 	default:
520 		return;
521 	}
522 
523 	hash_preload(vma->vm_mm, address, access, trap);
524 #endif /* CONFIG_PPC_STD_MMU */
525 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
526 	&& defined(CONFIG_HUGETLB_PAGE)
527 	if (is_vm_hugetlb_page(vma))
528 		book3e_hugetlb_preload(vma, address, *ptep);
529 #endif
530 }
531 
532 /*
533  * System memory should not be in /proc/iomem but various tools expect it
534  * (eg kdump).
535  */
536 static int __init add_system_ram_resources(void)
537 {
538 	struct memblock_region *reg;
539 
540 	for_each_memblock(memory, reg) {
541 		struct resource *res;
542 		unsigned long base = reg->base;
543 		unsigned long size = reg->size;
544 
545 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
546 		WARN_ON(!res);
547 
548 		if (res) {
549 			res->name = "System RAM";
550 			res->start = base;
551 			res->end = base + size - 1;
552 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
553 			WARN_ON(request_resource(&iomem_resource, res) < 0);
554 		}
555 	}
556 
557 	return 0;
558 }
559 subsys_initcall(add_system_ram_resources);
560 
561 #ifdef CONFIG_STRICT_DEVMEM
562 /*
563  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
564  * is valid. The argument is a physical page number.
565  *
566  * Access has to be given to non-kernel-ram areas as well, these contain the
567  * PCI mmio resources as well as potential bios/acpi data regions.
568  */
569 int devmem_is_allowed(unsigned long pfn)
570 {
571 	if (page_is_rtas_user_buf(pfn))
572 		return 1;
573 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
574 		return 0;
575 	if (!page_is_ram(pfn))
576 		return 1;
577 	return 0;
578 }
579 #endif /* CONFIG_STRICT_DEVMEM */
580