xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision d2ba09c1) 
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 	return memblock_is_memory(__pfn_to_phys(pfn));
86 }
87 
88 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
89 			      unsigned long size, pgprot_t vma_prot)
90 {
91 	if (ppc_md.phys_mem_access_prot)
92 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
93 
94 	if (!page_is_ram(pfn))
95 		vma_prot = pgprot_noncached(vma_prot);
96 
97 	return vma_prot;
98 }
99 EXPORT_SYMBOL(phys_mem_access_prot);
100 
101 #ifdef CONFIG_MEMORY_HOTPLUG
102 
103 #ifdef CONFIG_NUMA
104 int memory_add_physaddr_to_nid(u64 start)
105 {
106 	return hot_add_scn_to_nid(start);
107 }
108 #endif
109 
110 int __weak create_section_mapping(unsigned long start, unsigned long end, int nid)
111 {
112 	return -ENODEV;
113 }
114 
115 int __weak remove_section_mapping(unsigned long start, unsigned long end)
116 {
117 	return -ENODEV;
118 }
119 
120 int __meminit arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
121 		bool want_memblock)
122 {
123 	unsigned long start_pfn = start >> PAGE_SHIFT;
124 	unsigned long nr_pages = size >> PAGE_SHIFT;
125 	int rc;
126 
127 	resize_hpt_for_hotplug(memblock_phys_mem_size());
128 
129 	start = (unsigned long)__va(start);
130 	rc = create_section_mapping(start, start + size, nid);
131 	if (rc) {
132 		pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
133 			start, start + size, rc);
134 		return -EFAULT;
135 	}
136 	flush_inval_dcache_range(start, start + size);
137 
138 	return __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
139 }
140 
141 #ifdef CONFIG_MEMORY_HOTREMOVE
142 int __meminit arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
143 {
144 	unsigned long start_pfn = start >> PAGE_SHIFT;
145 	unsigned long nr_pages = size >> PAGE_SHIFT;
146 	struct page *page;
147 	int ret;
148 
149 	/*
150 	 * If we have an altmap then we need to skip over any reserved PFNs
151 	 * when querying the zone.
152 	 */
153 	page = pfn_to_page(start_pfn);
154 	if (altmap)
155 		page += vmem_altmap_offset(altmap);
156 
157 	ret = __remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
158 	if (ret)
159 		return ret;
160 
161 	/* Remove htab bolted mappings for this section of memory */
162 	start = (unsigned long)__va(start);
163 	flush_inval_dcache_range(start, start + size);
164 	ret = remove_section_mapping(start, start + size);
165 
166 	/* Ensure all vmalloc mappings are flushed in case they also
167 	 * hit that section of memory
168 	 */
169 	vm_unmap_aliases();
170 
171 	resize_hpt_for_hotplug(memblock_phys_mem_size());
172 
173 	return ret;
174 }
175 #endif
176 #endif /* CONFIG_MEMORY_HOTPLUG */
177 
178 /*
179  * walk_memory_resource() needs to make sure there is no holes in a given
180  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
181  * Instead it maintains it in memblock.memory structures.  Walk through the
182  * memory regions, find holes and callback for contiguous regions.
183  */
184 int
185 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
186 		void *arg, int (*func)(unsigned long, unsigned long, void *))
187 {
188 	struct memblock_region *reg;
189 	unsigned long end_pfn = start_pfn + nr_pages;
190 	unsigned long tstart, tend;
191 	int ret = -1;
192 
193 	for_each_memblock(memory, reg) {
194 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
195 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
196 		if (tstart >= tend)
197 			continue;
198 		ret = (*func)(tstart, tend - tstart, arg);
199 		if (ret)
200 			break;
201 	}
202 	return ret;
203 }
204 EXPORT_SYMBOL_GPL(walk_system_ram_range);
205 
206 #ifndef CONFIG_NEED_MULTIPLE_NODES
207 void __init mem_topology_setup(void)
208 {
209 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
210 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
211 #ifdef CONFIG_HIGHMEM
212 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
213 #endif
214 
215 	/* Place all memblock_regions in the same node and merge contiguous
216 	 * memblock_regions
217 	 */
218 	memblock_set_node(0, (phys_addr_t)ULLONG_MAX, &memblock.memory, 0);
219 }
220 
221 void __init initmem_init(void)
222 {
223 	/* XXX need to clip this if using highmem? */
224 	sparse_memory_present_with_active_regions(0);
225 	sparse_init();
226 }
227 
228 /* mark pages that don't exist as nosave */
229 static int __init mark_nonram_nosave(void)
230 {
231 	struct memblock_region *reg, *prev = NULL;
232 
233 	for_each_memblock(memory, reg) {
234 		if (prev &&
235 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
236 			register_nosave_region(memblock_region_memory_end_pfn(prev),
237 					       memblock_region_memory_base_pfn(reg));
238 		prev = reg;
239 	}
240 	return 0;
241 }
242 #else /* CONFIG_NEED_MULTIPLE_NODES */
243 static int __init mark_nonram_nosave(void)
244 {
245 	return 0;
246 }
247 #endif
248 
249 static bool zone_limits_final;
250 
251 /*
252  * The memory zones past TOP_ZONE are managed by generic mm code.
253  * These should be set to zero since that's what every other
254  * architecture does.
255  */
256 static unsigned long max_zone_pfns[MAX_NR_ZONES] = {
257 	[0            ... TOP_ZONE        ] = ~0UL,
258 	[TOP_ZONE + 1 ... MAX_NR_ZONES - 1] = 0
259 };
260 
261 /*
262  * Restrict the specified zone and all more restrictive zones
263  * to be below the specified pfn.  May not be called after
264  * paging_init().
265  */
266 void __init limit_zone_pfn(enum zone_type zone, unsigned long pfn_limit)
267 {
268 	int i;
269 
270 	if (WARN_ON(zone_limits_final))
271 		return;
272 
273 	for (i = zone; i >= 0; i--) {
274 		if (max_zone_pfns[i] > pfn_limit)
275 			max_zone_pfns[i] = pfn_limit;
276 	}
277 }
278 
279 /*
280  * Find the least restrictive zone that is entirely below the
281  * specified pfn limit.  Returns < 0 if no suitable zone is found.
282  *
283  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
284  * systems -- the DMA limit can be higher than any possible real pfn.
285  */
286 int dma_pfn_limit_to_zone(u64 pfn_limit)
287 {
288 	int i;
289 
290 	for (i = TOP_ZONE; i >= 0; i--) {
291 		if (max_zone_pfns[i] <= pfn_limit)
292 			return i;
293 	}
294 
295 	return -EPERM;
296 }
297 
298 /*
299  * paging_init() sets up the page tables - in fact we've already done this.
300  */
301 void __init paging_init(void)
302 {
303 	unsigned long long total_ram = memblock_phys_mem_size();
304 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
305 
306 #ifdef CONFIG_PPC32
307 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
308 	unsigned long end = __fix_to_virt(FIX_HOLE);
309 
310 	for (; v < end; v += PAGE_SIZE)
311 		map_kernel_page(v, 0, 0); /* XXX gross */
312 #endif
313 
314 #ifdef CONFIG_HIGHMEM
315 	map_kernel_page(PKMAP_BASE, 0, 0);	/* XXX gross */
316 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
317 
318 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
319 	kmap_prot = PAGE_KERNEL;
320 #endif /* CONFIG_HIGHMEM */
321 
322 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
323 	       (unsigned long long)top_of_ram, total_ram);
324 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
325 	       (long int)((top_of_ram - total_ram) >> 20));
326 
327 #ifdef CONFIG_HIGHMEM
328 	limit_zone_pfn(ZONE_NORMAL, lowmem_end_addr >> PAGE_SHIFT);
329 #endif
330 	limit_zone_pfn(TOP_ZONE, top_of_ram >> PAGE_SHIFT);
331 	zone_limits_final = true;
332 	free_area_init_nodes(max_zone_pfns);
333 
334 	mark_nonram_nosave();
335 }
336 
337 void __init mem_init(void)
338 {
339 	/*
340 	 * book3s is limited to 16 page sizes due to encoding this in
341 	 * a 4-bit field for slices.
342 	 */
343 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
344 
345 #ifdef CONFIG_SWIOTLB
346 	swiotlb_init(0);
347 #endif
348 
349 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
350 	set_max_mapnr(max_pfn);
351 	free_all_bootmem();
352 
353 #ifdef CONFIG_HIGHMEM
354 	{
355 		unsigned long pfn, highmem_mapnr;
356 
357 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
358 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
359 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
360 			struct page *page = pfn_to_page(pfn);
361 			if (!memblock_is_reserved(paddr))
362 				free_highmem_page(page);
363 		}
364 	}
365 #endif /* CONFIG_HIGHMEM */
366 
367 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
368 	/*
369 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
370 	 * functions.... do it here for the non-smp case.
371 	 */
372 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
373 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
374 #endif
375 
376 	mem_init_print_info(NULL);
377 #ifdef CONFIG_PPC32
378 	pr_info("Kernel virtual memory layout:\n");
379 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
380 #ifdef CONFIG_HIGHMEM
381 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
382 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
383 #endif /* CONFIG_HIGHMEM */
384 #ifdef CONFIG_NOT_COHERENT_CACHE
385 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
386 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
387 #endif /* CONFIG_NOT_COHERENT_CACHE */
388 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
389 		ioremap_bot, IOREMAP_TOP);
390 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
391 		VMALLOC_START, VMALLOC_END);
392 #endif /* CONFIG_PPC32 */
393 }
394 
395 void free_initmem(void)
396 {
397 	ppc_md.progress = ppc_printk_progress;
398 	mark_initmem_nx();
399 	free_initmem_default(POISON_FREE_INITMEM);
400 }
401 
402 #ifdef CONFIG_BLK_DEV_INITRD
403 void __init free_initrd_mem(unsigned long start, unsigned long end)
404 {
405 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
406 }
407 #endif
408 
409 /*
410  * This is called when a page has been modified by the kernel.
411  * It just marks the page as not i-cache clean.  We do the i-cache
412  * flush later when the page is given to a user process, if necessary.
413  */
414 void flush_dcache_page(struct page *page)
415 {
416 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
417 		return;
418 	/* avoid an atomic op if possible */
419 	if (test_bit(PG_arch_1, &page->flags))
420 		clear_bit(PG_arch_1, &page->flags);
421 }
422 EXPORT_SYMBOL(flush_dcache_page);
423 
424 void flush_dcache_icache_page(struct page *page)
425 {
426 #ifdef CONFIG_HUGETLB_PAGE
427 	if (PageCompound(page)) {
428 		flush_dcache_icache_hugepage(page);
429 		return;
430 	}
431 #endif
432 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
433 	/* On 8xx there is no need to kmap since highmem is not supported */
434 	__flush_dcache_icache(page_address(page));
435 #else
436 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
437 		void *start = kmap_atomic(page);
438 		__flush_dcache_icache(start);
439 		kunmap_atomic(start);
440 	} else {
441 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
442 	}
443 #endif
444 }
445 EXPORT_SYMBOL(flush_dcache_icache_page);
446 
447 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
448 {
449 	clear_page(page);
450 
451 	/*
452 	 * We shouldn't have to do this, but some versions of glibc
453 	 * require it (ld.so assumes zero filled pages are icache clean)
454 	 * - Anton
455 	 */
456 	flush_dcache_page(pg);
457 }
458 EXPORT_SYMBOL(clear_user_page);
459 
460 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
461 		    struct page *pg)
462 {
463 	copy_page(vto, vfrom);
464 
465 	/*
466 	 * We should be able to use the following optimisation, however
467 	 * there are two problems.
468 	 * Firstly a bug in some versions of binutils meant PLT sections
469 	 * were not marked executable.
470 	 * Secondly the first word in the GOT section is blrl, used
471 	 * to establish the GOT address. Until recently the GOT was
472 	 * not marked executable.
473 	 * - Anton
474 	 */
475 #if 0
476 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
477 		return;
478 #endif
479 
480 	flush_dcache_page(pg);
481 }
482 
483 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
484 			     unsigned long addr, int len)
485 {
486 	unsigned long maddr;
487 
488 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
489 	flush_icache_range(maddr, maddr + len);
490 	kunmap(page);
491 }
492 EXPORT_SYMBOL(flush_icache_user_range);
493 
494 /*
495  * This is called at the end of handling a user page fault, when the
496  * fault has been handled by updating a PTE in the linux page tables.
497  * We use it to preload an HPTE into the hash table corresponding to
498  * the updated linux PTE.
499  *
500  * This must always be called with the pte lock held.
501  */
502 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
503 		      pte_t *ptep)
504 {
505 #ifdef CONFIG_PPC_STD_MMU
506 	/*
507 	 * We don't need to worry about _PAGE_PRESENT here because we are
508 	 * called with either mm->page_table_lock held or ptl lock held
509 	 */
510 	unsigned long access, trap;
511 
512 	if (radix_enabled())
513 		return;
514 
515 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
516 	if (!pte_young(*ptep) || address >= TASK_SIZE)
517 		return;
518 
519 	/* We try to figure out if we are coming from an instruction
520 	 * access fault and pass that down to __hash_page so we avoid
521 	 * double-faulting on execution of fresh text. We have to test
522 	 * for regs NULL since init will get here first thing at boot
523 	 *
524 	 * We also avoid filling the hash if not coming from a fault
525 	 */
526 
527 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
528 	switch (trap) {
529 	case 0x300:
530 		access = 0UL;
531 		break;
532 	case 0x400:
533 		access = _PAGE_EXEC;
534 		break;
535 	default:
536 		return;
537 	}
538 
539 	hash_preload(vma->vm_mm, address, access, trap);
540 #endif /* CONFIG_PPC_STD_MMU */
541 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
542 	&& defined(CONFIG_HUGETLB_PAGE)
543 	if (is_vm_hugetlb_page(vma))
544 		book3e_hugetlb_preload(vma, address, *ptep);
545 #endif
546 }
547 
548 /*
549  * System memory should not be in /proc/iomem but various tools expect it
550  * (eg kdump).
551  */
552 static int __init add_system_ram_resources(void)
553 {
554 	struct memblock_region *reg;
555 
556 	for_each_memblock(memory, reg) {
557 		struct resource *res;
558 		unsigned long base = reg->base;
559 		unsigned long size = reg->size;
560 
561 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
562 		WARN_ON(!res);
563 
564 		if (res) {
565 			res->name = "System RAM";
566 			res->start = base;
567 			res->end = base + size - 1;
568 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
569 			WARN_ON(request_resource(&iomem_resource, res) < 0);
570 		}
571 	}
572 
573 	return 0;
574 }
575 subsys_initcall(add_system_ram_resources);
576 
577 #ifdef CONFIG_STRICT_DEVMEM
578 /*
579  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
580  * is valid. The argument is a physical page number.
581  *
582  * Access has to be given to non-kernel-ram areas as well, these contain the
583  * PCI mmio resources as well as potential bios/acpi data regions.
584  */
585 int devmem_is_allowed(unsigned long pfn)
586 {
587 	if (page_is_rtas_user_buf(pfn))
588 		return 1;
589 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
590 		return 0;
591 	if (!page_is_ram(pfn))
592 		return 1;
593 	return 0;
594 }
595 #endif /* CONFIG_STRICT_DEVMEM */
596