xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 95777591) 
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/memblock.h>
31 #include <linux/highmem.h>
32 #include <linux/initrd.h>
33 #include <linux/pagemap.h>
34 #include <linux/suspend.h>
35 #include <linux/hugetlb.h>
36 #include <linux/slab.h>
37 #include <linux/vmalloc.h>
38 #include <linux/memremap.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 bool init_mem_is_free;
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(int nid, u64 start, u64 size,
143 					struct vmem_altmap *altmap)
144 {
145 	unsigned long start_pfn = start >> PAGE_SHIFT;
146 	unsigned long nr_pages = size >> PAGE_SHIFT;
147 	struct page *page;
148 	int ret;
149 
150 	/*
151 	 * If we have an altmap then we need to skip over any reserved PFNs
152 	 * when querying the zone.
153 	 */
154 	page = pfn_to_page(start_pfn);
155 	if (altmap)
156 		page += vmem_altmap_offset(altmap);
157 
158 	ret = __remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
159 	if (ret)
160 		return ret;
161 
162 	/* Remove htab bolted mappings for this section of memory */
163 	start = (unsigned long)__va(start);
164 	flush_inval_dcache_range(start, start + size);
165 	ret = remove_section_mapping(start, start + size);
166 
167 	/* Ensure all vmalloc mappings are flushed in case they also
168 	 * hit that section of memory
169 	 */
170 	vm_unmap_aliases();
171 
172 	resize_hpt_for_hotplug(memblock_phys_mem_size());
173 
174 	return ret;
175 }
176 #endif
177 #endif /* CONFIG_MEMORY_HOTPLUG */
178 
179 /*
180  * walk_memory_resource() needs to make sure there is no holes in a given
181  * memory range.  PPC64 does not maintain the memory layout in /proc/iomem.
182  * Instead it maintains it in memblock.memory structures.  Walk through the
183  * memory regions, find holes and callback for contiguous regions.
184  */
185 int
186 walk_system_ram_range(unsigned long start_pfn, unsigned long nr_pages,
187 		void *arg, int (*func)(unsigned long, unsigned long, void *))
188 {
189 	struct memblock_region *reg;
190 	unsigned long end_pfn = start_pfn + nr_pages;
191 	unsigned long tstart, tend;
192 	int ret = -1;
193 
194 	for_each_memblock(memory, reg) {
195 		tstart = max(start_pfn, memblock_region_memory_base_pfn(reg));
196 		tend = min(end_pfn, memblock_region_memory_end_pfn(reg));
197 		if (tstart >= tend)
198 			continue;
199 		ret = (*func)(tstart, tend - tstart, arg);
200 		if (ret)
201 			break;
202 	}
203 	return ret;
204 }
205 EXPORT_SYMBOL_GPL(walk_system_ram_range);
206 
207 #ifndef CONFIG_NEED_MULTIPLE_NODES
208 void __init mem_topology_setup(void)
209 {
210 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
211 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
212 #ifdef CONFIG_HIGHMEM
213 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
214 #endif
215 
216 	/* Place all memblock_regions in the same node and merge contiguous
217 	 * memblock_regions
218 	 */
219 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
220 }
221 
222 void __init initmem_init(void)
223 {
224 	/* XXX need to clip this if using highmem? */
225 	sparse_memory_present_with_active_regions(0);
226 	sparse_init();
227 }
228 
229 /* mark pages that don't exist as nosave */
230 static int __init mark_nonram_nosave(void)
231 {
232 	struct memblock_region *reg, *prev = NULL;
233 
234 	for_each_memblock(memory, reg) {
235 		if (prev &&
236 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
237 			register_nosave_region(memblock_region_memory_end_pfn(prev),
238 					       memblock_region_memory_base_pfn(reg));
239 		prev = reg;
240 	}
241 	return 0;
242 }
243 #else /* CONFIG_NEED_MULTIPLE_NODES */
244 static int __init mark_nonram_nosave(void)
245 {
246 	return 0;
247 }
248 #endif
249 
250 /*
251  * Zones usage:
252  *
253  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
254  * everything else. GFP_DMA32 page allocations automatically fall back to
255  * ZONE_DMA.
256  *
257  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
258  * inform the generic DMA mapping code.  32-bit only devices (if not handled
259  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
260  * otherwise served by ZONE_DMA.
261  */
262 static unsigned long max_zone_pfns[MAX_NR_ZONES];
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_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
297 #endif
298 
299 #ifdef CONFIG_HIGHMEM
300 	map_kernel_page(PKMAP_BASE, 0, __pgprot(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_ZONE_DMA
313 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
314 #endif
315 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
316 #ifdef CONFIG_HIGHMEM
317 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
318 #endif
319 
320 	free_area_init_nodes(max_zone_pfns);
321 
322 	mark_nonram_nosave();
323 }
324 
325 void __init mem_init(void)
326 {
327 	/*
328 	 * book3s is limited to 16 page sizes due to encoding this in
329 	 * a 4-bit field for slices.
330 	 */
331 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
332 
333 #ifdef CONFIG_SWIOTLB
334 	swiotlb_init(0);
335 #endif
336 
337 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
338 	set_max_mapnr(max_pfn);
339 	memblock_free_all();
340 
341 #ifdef CONFIG_HIGHMEM
342 	{
343 		unsigned long pfn, highmem_mapnr;
344 
345 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
346 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
347 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
348 			struct page *page = pfn_to_page(pfn);
349 			if (!memblock_is_reserved(paddr))
350 				free_highmem_page(page);
351 		}
352 	}
353 #endif /* CONFIG_HIGHMEM */
354 
355 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
356 	/*
357 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
358 	 * functions.... do it here for the non-smp case.
359 	 */
360 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
361 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
362 #endif
363 
364 	mem_init_print_info(NULL);
365 #ifdef CONFIG_PPC32
366 	pr_info("Kernel virtual memory layout:\n");
367 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
368 #ifdef CONFIG_HIGHMEM
369 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
370 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
371 #endif /* CONFIG_HIGHMEM */
372 #ifdef CONFIG_NOT_COHERENT_CACHE
373 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
374 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
375 #endif /* CONFIG_NOT_COHERENT_CACHE */
376 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
377 		ioremap_bot, IOREMAP_TOP);
378 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
379 		VMALLOC_START, VMALLOC_END);
380 #endif /* CONFIG_PPC32 */
381 }
382 
383 void free_initmem(void)
384 {
385 	ppc_md.progress = ppc_printk_progress;
386 	mark_initmem_nx();
387 	init_mem_is_free = true;
388 	free_initmem_default(POISON_FREE_INITMEM);
389 }
390 
391 #ifdef CONFIG_BLK_DEV_INITRD
392 void __init free_initrd_mem(unsigned long start, unsigned long end)
393 {
394 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
395 }
396 #endif
397 
398 /*
399  * This is called when a page has been modified by the kernel.
400  * It just marks the page as not i-cache clean.  We do the i-cache
401  * flush later when the page is given to a user process, if necessary.
402  */
403 void flush_dcache_page(struct page *page)
404 {
405 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
406 		return;
407 	/* avoid an atomic op if possible */
408 	if (test_bit(PG_arch_1, &page->flags))
409 		clear_bit(PG_arch_1, &page->flags);
410 }
411 EXPORT_SYMBOL(flush_dcache_page);
412 
413 void flush_dcache_icache_page(struct page *page)
414 {
415 #ifdef CONFIG_HUGETLB_PAGE
416 	if (PageCompound(page)) {
417 		flush_dcache_icache_hugepage(page);
418 		return;
419 	}
420 #endif
421 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
422 	/* On 8xx there is no need to kmap since highmem is not supported */
423 	__flush_dcache_icache(page_address(page));
424 #else
425 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
426 		void *start = kmap_atomic(page);
427 		__flush_dcache_icache(start);
428 		kunmap_atomic(start);
429 	} else {
430 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
431 	}
432 #endif
433 }
434 EXPORT_SYMBOL(flush_dcache_icache_page);
435 
436 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
437 {
438 	clear_page(page);
439 
440 	/*
441 	 * We shouldn't have to do this, but some versions of glibc
442 	 * require it (ld.so assumes zero filled pages are icache clean)
443 	 * - Anton
444 	 */
445 	flush_dcache_page(pg);
446 }
447 EXPORT_SYMBOL(clear_user_page);
448 
449 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
450 		    struct page *pg)
451 {
452 	copy_page(vto, vfrom);
453 
454 	/*
455 	 * We should be able to use the following optimisation, however
456 	 * there are two problems.
457 	 * Firstly a bug in some versions of binutils meant PLT sections
458 	 * were not marked executable.
459 	 * Secondly the first word in the GOT section is blrl, used
460 	 * to establish the GOT address. Until recently the GOT was
461 	 * not marked executable.
462 	 * - Anton
463 	 */
464 #if 0
465 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
466 		return;
467 #endif
468 
469 	flush_dcache_page(pg);
470 }
471 
472 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
473 			     unsigned long addr, int len)
474 {
475 	unsigned long maddr;
476 
477 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
478 	flush_icache_range(maddr, maddr + len);
479 	kunmap(page);
480 }
481 EXPORT_SYMBOL(flush_icache_user_range);
482 
483 /*
484  * This is called at the end of handling a user page fault, when the
485  * fault has been handled by updating a PTE in the linux page tables.
486  * We use it to preload an HPTE into the hash table corresponding to
487  * the updated linux PTE.
488  *
489  * This must always be called with the pte lock held.
490  */
491 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
492 		      pte_t *ptep)
493 {
494 #ifdef CONFIG_PPC_BOOK3S
495 	/*
496 	 * We don't need to worry about _PAGE_PRESENT here because we are
497 	 * called with either mm->page_table_lock held or ptl lock held
498 	 */
499 	unsigned long trap;
500 	bool is_exec;
501 
502 	if (radix_enabled()) {
503 		prefetch((void *)address);
504 		return;
505 	}
506 
507 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
508 	if (!pte_young(*ptep) || address >= TASK_SIZE)
509 		return;
510 
511 	/* We try to figure out if we are coming from an instruction
512 	 * access fault and pass that down to __hash_page so we avoid
513 	 * double-faulting on execution of fresh text. We have to test
514 	 * for regs NULL since init will get here first thing at boot
515 	 *
516 	 * We also avoid filling the hash if not coming from a fault
517 	 */
518 
519 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
520 	switch (trap) {
521 	case 0x300:
522 		is_exec = false;
523 		break;
524 	case 0x400:
525 		is_exec = true;
526 		break;
527 	default:
528 		return;
529 	}
530 
531 	hash_preload(vma->vm_mm, address, is_exec, trap);
532 #endif /* CONFIG_PPC_BOOK3S */
533 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
534 	&& defined(CONFIG_HUGETLB_PAGE)
535 	if (is_vm_hugetlb_page(vma))
536 		book3e_hugetlb_preload(vma, address, *ptep);
537 #endif
538 }
539 
540 /*
541  * System memory should not be in /proc/iomem but various tools expect it
542  * (eg kdump).
543  */
544 static int __init add_system_ram_resources(void)
545 {
546 	struct memblock_region *reg;
547 
548 	for_each_memblock(memory, reg) {
549 		struct resource *res;
550 		unsigned long base = reg->base;
551 		unsigned long size = reg->size;
552 
553 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
554 		WARN_ON(!res);
555 
556 		if (res) {
557 			res->name = "System RAM";
558 			res->start = base;
559 			res->end = base + size - 1;
560 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
561 			WARN_ON(request_resource(&iomem_resource, res) < 0);
562 		}
563 	}
564 
565 	return 0;
566 }
567 subsys_initcall(add_system_ram_resources);
568 
569 #ifdef CONFIG_STRICT_DEVMEM
570 /*
571  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
572  * is valid. The argument is a physical page number.
573  *
574  * Access has to be given to non-kernel-ram areas as well, these contain the
575  * PCI mmio resources as well as potential bios/acpi data regions.
576  */
577 int devmem_is_allowed(unsigned long pfn)
578 {
579 	if (page_is_rtas_user_buf(pfn))
580 		return 1;
581 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
582 		return 0;
583 	if (!page_is_ram(pfn))
584 		return 1;
585 	return 0;
586 }
587 #endif /* CONFIG_STRICT_DEVMEM */
588