xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 6aa7de05) 
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_warning(
142 			"Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
143 			start, start + size, rc);
144 		return -EFAULT;
145 	}
146 
147 	return __add_pages(nid, start_pfn, nr_pages, want_memblock);
148 }
149 
150 #ifdef CONFIG_MEMORY_HOTREMOVE
151 int arch_remove_memory(u64 start, u64 size)
152 {
153 	unsigned long start_pfn = start >> PAGE_SHIFT;
154 	unsigned long nr_pages = size >> PAGE_SHIFT;
155 	struct vmem_altmap *altmap;
156 	struct page *page;
157 	int ret;
158 
159 	/*
160 	 * If we have an altmap then we need to skip over any reserved PFNs
161 	 * when querying the zone.
162 	 */
163 	page = pfn_to_page(start_pfn);
164 	altmap = to_vmem_altmap((unsigned long) page);
165 	if (altmap)
166 		page += vmem_altmap_offset(altmap);
167 
168 	ret = __remove_pages(page_zone(page), start_pfn, nr_pages);
169 	if (ret)
170 		return ret;
171 
172 	/* Remove htab bolted mappings for this section of memory */
173 	start = (unsigned long)__va(start);
174 	ret = remove_section_mapping(start, start + size);
175 
176 	/* Ensure all vmalloc mappings are flushed in case they also
177 	 * hit that section of memory
178 	 */
179 	vm_unmap_aliases();
180 
181 	resize_hpt_for_hotplug(memblock_phys_mem_size());
182 
183 	return ret;
184 }
185 #endif
186 #endif /* CONFIG_MEMORY_HOTPLUG */
187 
188 /*
189  * walk_memory_resource() needs to make sure there is no holes in a given
190  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
191  * Instead it maintains it in memblock.memory structures.  Walk through the
192  * memory regions, find holes and callback for contiguous regions.
193  */
194 int
195 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
196 		void *arg, int (*func)(unsigned long, unsigned long, void *))
197 {
198 	struct memblock_region *reg;
199 	unsigned long end_pfn = start_pfn + nr_pages;
200 	unsigned long tstart, tend;
201 	int ret = -1;
202 
203 	for_each_memblock(memory, reg) {
204 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
205 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
206 		if (tstart >= tend)
207 			continue;
208 		ret = (*func)(tstart, tend - tstart, arg);
209 		if (ret)
210 			break;
211 	}
212 	return ret;
213 }
214 EXPORT_SYMBOL_GPL(walk_system_ram_range);
215 
216 #ifndef CONFIG_NEED_MULTIPLE_NODES
217 void __init initmem_init(void)
218 {
219 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
220 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
221 #ifdef CONFIG_HIGHMEM
222 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
223 #endif
224 
225 	/* Place all memblock_regions in the same node and merge contiguous
226 	 * memblock_regions
227 	 */
228 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
229 
230 	/* XXX need to clip this if using highmem? */
231 	sparse_memory_present_with_active_regions(0);
232 	sparse_init();
233 }
234 
235 /* mark pages that don't exist as nosave */
236 static int __init mark_nonram_nosave(void)
237 {
238 	struct memblock_region *reg, *prev = NULL;
239 
240 	for_each_memblock(memory, reg) {
241 		if (prev &&
242 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
243 			register_nosave_region(memblock_region_memory_end_pfn(prev),
244 					       memblock_region_memory_base_pfn(reg));
245 		prev = reg;
246 	}
247 	return 0;
248 }
249 #else /* CONFIG_NEED_MULTIPLE_NODES */
250 static int __init mark_nonram_nosave(void)
251 {
252 	return 0;
253 }
254 #endif
255 
256 static bool zone_limits_final;
257 
258 /*
259  * The memory zones past TOP_ZONE are managed by generic mm code.
260  * These should be set to zero since that's what every other
261  * architecture does.
262  */
263 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
264 	[0            ... TOP_ZONE        ] = ~0UL,
265 	[TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
266 };
267 
268 /*
269  * Restrict the specified zone and all more restrictive zones
270  * to be below the specified pfn.  May not be called after
271  * paging_init().
272  */
273 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
274 {
275 	int i;
276 
277 	if (WARN_ON(zone_limits_final))
278 		return;
279 
280 	for (i = zone; i >= 0; i--) {
281 		if (max_zone_pfns[i] > pfn_limit)
282 			max_zone_pfns[i] = pfn_limit;
283 	}
284 }
285 
286 /*
287  * Find the least restrictive zone that is entirely below the
288  * specified pfn limit.  Returns < 0 if no suitable zone is found.
289  *
290  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
291  * systems -- the DMA limit can be higher than any possible real pfn.
292  */
293 int dma_pfn_limit_to_zone(u64 pfn_limit)
294 {
295 	int i;
296 
297 	for (i = TOP_ZONE; i >= 0; i--) {
298 		if (max_zone_pfns[i] <= pfn_limit)
299 			return i;
300 	}
301 
302 	return -EPERM;
303 }
304 
305 /*
306  * paging_init() sets up the page tables - in fact we've already done this.
307  */
308 void __init paging_init(void)
309 {
310 	unsigned long long total_ram = memblock_phys_mem_size();
311 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
312 
313 #ifdef CONFIG_PPC32
314 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
315 	unsigned long end = __fix_to_virt(FIX_HOLE);
316 
317 	for (; v < end; v += PAGE_SIZE)
318 		map_kernel_page(v, 0, 0); /* XXX gross */
319 #endif
320 
321 #ifdef CONFIG_HIGHMEM
322 	map_kernel_page(PKMAP_BASE, 0, 0);	/* XXX gross */
323 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
324 
325 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
326 	kmap_prot = PAGE_KERNEL;
327 #endif /* CONFIG_HIGHMEM */
328 
329 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
330 	       (unsigned long long)top_of_ram, total_ram);
331 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
332 	       (long int)((top_of_ram - total_ram) >> 20));
333 
334 #ifdef CONFIG_HIGHMEM
335 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
336 #endif
337 	limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
338 	zone_limits_final = true;
339 	free_area_init_nodes(max_zone_pfns);
340 
341 	mark_nonram_nosave();
342 }
343 
344 void __init mem_init(void)
345 {
346 	/*
347 	 * book3s is limited to 16 page sizes due to encoding this in
348 	 * a 4-bit field for slices.
349 	 */
350 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
351 
352 #ifdef CONFIG_SWIOTLB
353 	swiotlb_init(0);
354 #endif
355 
356 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
357 	set_max_mapnr(max_pfn);
358 	free_all_bootmem();
359 
360 #ifdef CONFIG_HIGHMEM
361 	{
362 		unsigned long pfn, highmem_mapnr;
363 
364 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
365 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
366 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
367 			struct page *page = pfn_to_page(pfn);
368 			if (!memblock_is_reserved(paddr))
369 				free_highmem_page(page);
370 		}
371 	}
372 #endif /* CONFIG_HIGHMEM */
373 
374 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
375 	/*
376 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
377 	 * functions.... do it here for the non-smp case.
378 	 */
379 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
380 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
381 #endif
382 
383 	mem_init_print_info(NULL);
384 #ifdef CONFIG_PPC32
385 	pr_info("Kernel virtual memory layout:\n");
386 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
387 #ifdef CONFIG_HIGHMEM
388 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
389 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
390 #endif /* CONFIG_HIGHMEM */
391 #ifdef CONFIG_NOT_COHERENT_CACHE
392 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
393 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
394 #endif /* CONFIG_NOT_COHERENT_CACHE */
395 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
396 		ioremap_bot, IOREMAP_TOP);
397 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
398 		VMALLOC_START, VMALLOC_END);
399 #endif /* CONFIG_PPC32 */
400 }
401 
402 void free_initmem(void)
403 {
404 	ppc_md.progress = ppc_printk_progress;
405 	mark_initmem_nx();
406 	free_initmem_default(POISON_FREE_INITMEM);
407 }
408 
409 #ifdef CONFIG_BLK_DEV_INITRD
410 void __init free_initrd_mem(unsigned long start, unsigned long end)
411 {
412 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
413 }
414 #endif
415 
416 /*
417  * This is called when a page has been modified by the kernel.
418  * It just marks the page as not i-cache clean.  We do the i-cache
419  * flush later when the page is given to a user process, if necessary.
420  */
421 void flush_dcache_page(struct page *page)
422 {
423 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
424 		return;
425 	/* avoid an atomic op if possible */
426 	if (test_bit(PG_arch_1, &page->flags))
427 		clear_bit(PG_arch_1, &page->flags);
428 }
429 EXPORT_SYMBOL(flush_dcache_page);
430 
431 void flush_dcache_icache_page(struct page *page)
432 {
433 #ifdef CONFIG_HUGETLB_PAGE
434 	if (PageCompound(page)) {
435 		flush_dcache_icache_hugepage(page);
436 		return;
437 	}
438 #endif
439 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
440 	/* On 8xx there is no need to kmap since highmem is not supported */
441 	__flush_dcache_icache(page_address(page));
442 #else
443 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
444 		void *start = kmap_atomic(page);
445 		__flush_dcache_icache(start);
446 		kunmap_atomic(start);
447 	} else {
448 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
449 	}
450 #endif
451 }
452 EXPORT_SYMBOL(flush_dcache_icache_page);
453 
454 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
455 {
456 	clear_page(page);
457 
458 	/*
459 	 * We shouldn't have to do this, but some versions of glibc
460 	 * require it (ld.so assumes zero filled pages are icache clean)
461 	 * - Anton
462 	 */
463 	flush_dcache_page(pg);
464 }
465 EXPORT_SYMBOL(clear_user_page);
466 
467 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
468 		    struct page *pg)
469 {
470 	copy_page(vto, vfrom);
471 
472 	/*
473 	 * We should be able to use the following optimisation, however
474 	 * there are two problems.
475 	 * Firstly a bug in some versions of binutils meant PLT sections
476 	 * were not marked executable.
477 	 * Secondly the first word in the GOT section is blrl, used
478 	 * to establish the GOT address. Until recently the GOT was
479 	 * not marked executable.
480 	 * - Anton
481 	 */
482 #if 0
483 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
484 		return;
485 #endif
486 
487 	flush_dcache_page(pg);
488 }
489 
490 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
491 			     unsigned long addr, int len)
492 {
493 	unsigned long maddr;
494 
495 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
496 	flush_icache_range(maddr, maddr + len);
497 	kunmap(page);
498 }
499 EXPORT_SYMBOL(flush_icache_user_range);
500 
501 /*
502  * This is called at the end of handling a user page fault, when the
503  * fault has been handled by updating a PTE in the linux page tables.
504  * We use it to preload an HPTE into the hash table corresponding to
505  * the updated linux PTE.
506  *
507  * This must always be called with the pte lock held.
508  */
509 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
510 		      pte_t *ptep)
511 {
512 #ifdef CONFIG_PPC_STD_MMU
513 	/*
514 	 * We don't need to worry about _PAGE_PRESENT here because we are
515 	 * called with either mm->page_table_lock held or ptl lock held
516 	 */
517 	unsigned long access, trap;
518 
519 	if (radix_enabled())
520 		return;
521 
522 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
523 	if (!pte_young(*ptep) || address >= TASK_SIZE)
524 		return;
525 
526 	/* We try to figure out if we are coming from an instruction
527 	 * access fault and pass that down to __hash_page so we avoid
528 	 * double-faulting on execution of fresh text. We have to test
529 	 * for regs NULL since init will get here first thing at boot
530 	 *
531 	 * We also avoid filling the hash if not coming from a fault
532 	 */
533 
534 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
535 	switch (trap) {
536 	case 0x300:
537 		access = 0UL;
538 		break;
539 	case 0x400:
540 		access = _PAGE_EXEC;
541 		break;
542 	default:
543 		return;
544 	}
545 
546 	hash_preload(vma->vm_mm, address, access, trap);
547 #endif /* CONFIG_PPC_STD_MMU */
548 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
549 	&& defined(CONFIG_HUGETLB_PAGE)
550 	if (is_vm_hugetlb_page(vma))
551 		book3e_hugetlb_preload(vma, address, *ptep);
552 #endif
553 }
554 
555 /*
556  * System memory should not be in /proc/iomem but various tools expect it
557  * (eg kdump).
558  */
559 static int __init add_system_ram_resources(void)
560 {
561 	struct memblock_region *reg;
562 
563 	for_each_memblock(memory, reg) {
564 		struct resource *res;
565 		unsigned long base = reg->base;
566 		unsigned long size = reg->size;
567 
568 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
569 		WARN_ON(!res);
570 
571 		if (res) {
572 			res->name = "System RAM";
573 			res->start = base;
574 			res->end = base + size - 1;
575 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
576 			WARN_ON(request_resource(&iomem_resource, res) < 0);
577 		}
578 	}
579 
580 	return 0;
581 }
582 subsys_initcall(add_system_ram_resources);
583 
584 #ifdef CONFIG_STRICT_DEVMEM
585 /*
586  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
587  * is valid. The argument is a physical page number.
588  *
589  * Access has to be given to non-kernel-ram areas as well, these contain the
590  * PCI mmio resources as well as potential bios/acpi data regions.
591  */
592 int devmem_is_allowed(unsigned long pfn)
593 {
594 	if (page_is_rtas_user_buf(pfn))
595 		return 1;
596 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
597 		return 0;
598 	if (!page_is_ram(pfn))
599 		return 1;
600 	return 0;
601 }
602 #endif /* CONFIG_STRICT_DEVMEM */
603