xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision ff4a7481c3898ffc3cc271d6aca431d190c37247)
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(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_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 /*
250  * Zones usage:
251  *
252  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
253  * everything else. GFP_DMA32 page allocations automatically fall back to
254  * ZONE_DMA.
255  *
256  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
257  * inform the generic DMA mapping code.  32-bit only devices (if not handled
258  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
259  * otherwise served by ZONE_DMA.
260  */
261 static unsigned long max_zone_pfns[MAX_NR_ZONES];
262 
263 /*
264  * Find the least restrictive zone that is entirely below the
265  * specified pfn limit.  Returns < 0 if no suitable zone is found.
266  *
267  * pfn_limit must be u64 because it can exceed 32 bits even on 32-bit
268  * systems -- the DMA limit can be higher than any possible real pfn.
269  */
270 int dma_pfn_limit_to_zone(u64 pfn_limit)
271 {
272 	int i;
273 
274 	for (i = TOP_ZONE; i >= 0; i--) {
275 		if (max_zone_pfns[i] <= pfn_limit)
276 			return i;
277 	}
278 
279 	return -EPERM;
280 }
281 
282 /*
283  * paging_init() sets up the page tables - in fact we've already done this.
284  */
285 void __init paging_init(void)
286 {
287 	unsigned long long total_ram = memblock_phys_mem_size();
288 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
289 
290 #ifdef CONFIG_PPC32
291 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
292 	unsigned long end = __fix_to_virt(FIX_HOLE);
293 
294 	for (; v < end; v += PAGE_SIZE)
295 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
296 #endif
297 
298 #ifdef CONFIG_HIGHMEM
299 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
300 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
301 
302 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
303 	kmap_prot = PAGE_KERNEL;
304 #endif /* CONFIG_HIGHMEM */
305 
306 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
307 	       (unsigned long long)top_of_ram, total_ram);
308 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
309 	       (long int)((top_of_ram - total_ram) >> 20));
310 
311 #ifdef CONFIG_ZONE_DMA
312 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
313 #endif
314 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
315 #ifdef CONFIG_HIGHMEM
316 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
317 #endif
318 
319 	free_area_init_nodes(max_zone_pfns);
320 
321 	mark_nonram_nosave();
322 }
323 
324 void __init mem_init(void)
325 {
326 	/*
327 	 * book3s is limited to 16 page sizes due to encoding this in
328 	 * a 4-bit field for slices.
329 	 */
330 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
331 
332 #ifdef CONFIG_SWIOTLB
333 	swiotlb_init(0);
334 #endif
335 
336 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
337 	set_max_mapnr(max_pfn);
338 	memblock_free_all();
339 
340 #ifdef CONFIG_HIGHMEM
341 	{
342 		unsigned long pfn, highmem_mapnr;
343 
344 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
345 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
346 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
347 			struct page *page = pfn_to_page(pfn);
348 			if (!memblock_is_reserved(paddr))
349 				free_highmem_page(page);
350 		}
351 	}
352 #endif /* CONFIG_HIGHMEM */
353 
354 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
355 	/*
356 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
357 	 * functions.... do it here for the non-smp case.
358 	 */
359 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
360 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
361 #endif
362 
363 	mem_init_print_info(NULL);
364 #ifdef CONFIG_PPC32
365 	pr_info("Kernel virtual memory layout:\n");
366 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
367 #ifdef CONFIG_HIGHMEM
368 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
369 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
370 #endif /* CONFIG_HIGHMEM */
371 #ifdef CONFIG_NOT_COHERENT_CACHE
372 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
373 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
374 #endif /* CONFIG_NOT_COHERENT_CACHE */
375 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
376 		ioremap_bot, IOREMAP_TOP);
377 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
378 		VMALLOC_START, VMALLOC_END);
379 #endif /* CONFIG_PPC32 */
380 }
381 
382 void free_initmem(void)
383 {
384 	ppc_md.progress = ppc_printk_progress;
385 	mark_initmem_nx();
386 	init_mem_is_free = true;
387 	free_initmem_default(POISON_FREE_INITMEM);
388 }
389 
390 #ifdef CONFIG_BLK_DEV_INITRD
391 void __init free_initrd_mem(unsigned long start, unsigned long end)
392 {
393 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
394 }
395 #endif
396 
397 /*
398  * This is called when a page has been modified by the kernel.
399  * It just marks the page as not i-cache clean.  We do the i-cache
400  * flush later when the page is given to a user process, if necessary.
401  */
402 void flush_dcache_page(struct page *page)
403 {
404 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
405 		return;
406 	/* avoid an atomic op if possible */
407 	if (test_bit(PG_arch_1, &page->flags))
408 		clear_bit(PG_arch_1, &page->flags);
409 }
410 EXPORT_SYMBOL(flush_dcache_page);
411 
412 void flush_dcache_icache_page(struct page *page)
413 {
414 #ifdef CONFIG_HUGETLB_PAGE
415 	if (PageCompound(page)) {
416 		flush_dcache_icache_hugepage(page);
417 		return;
418 	}
419 #endif
420 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
421 	/* On 8xx there is no need to kmap since highmem is not supported */
422 	__flush_dcache_icache(page_address(page));
423 #else
424 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
425 		void *start = kmap_atomic(page);
426 		__flush_dcache_icache(start);
427 		kunmap_atomic(start);
428 	} else {
429 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
430 	}
431 #endif
432 }
433 EXPORT_SYMBOL(flush_dcache_icache_page);
434 
435 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
436 {
437 	clear_page(page);
438 
439 	/*
440 	 * We shouldn't have to do this, but some versions of glibc
441 	 * require it (ld.so assumes zero filled pages are icache clean)
442 	 * - Anton
443 	 */
444 	flush_dcache_page(pg);
445 }
446 EXPORT_SYMBOL(clear_user_page);
447 
448 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
449 		    struct page *pg)
450 {
451 	copy_page(vto, vfrom);
452 
453 	/*
454 	 * We should be able to use the following optimisation, however
455 	 * there are two problems.
456 	 * Firstly a bug in some versions of binutils meant PLT sections
457 	 * were not marked executable.
458 	 * Secondly the first word in the GOT section is blrl, used
459 	 * to establish the GOT address. Until recently the GOT was
460 	 * not marked executable.
461 	 * - Anton
462 	 */
463 #if 0
464 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
465 		return;
466 #endif
467 
468 	flush_dcache_page(pg);
469 }
470 
471 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
472 			     unsigned long addr, int len)
473 {
474 	unsigned long maddr;
475 
476 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
477 	flush_icache_range(maddr, maddr + len);
478 	kunmap(page);
479 }
480 EXPORT_SYMBOL(flush_icache_user_range);
481 
482 /*
483  * This is called at the end of handling a user page fault, when the
484  * fault has been handled by updating a PTE in the linux page tables.
485  * We use it to preload an HPTE into the hash table corresponding to
486  * the updated linux PTE.
487  *
488  * This must always be called with the pte lock held.
489  */
490 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
491 		      pte_t *ptep)
492 {
493 #ifdef CONFIG_PPC_BOOK3S
494 	/*
495 	 * We don't need to worry about _PAGE_PRESENT here because we are
496 	 * called with either mm->page_table_lock held or ptl lock held
497 	 */
498 	unsigned long trap;
499 	bool is_exec;
500 
501 	if (radix_enabled()) {
502 		prefetch((void *)address);
503 		return;
504 	}
505 
506 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
507 	if (!pte_young(*ptep) || address >= TASK_SIZE)
508 		return;
509 
510 	/* We try to figure out if we are coming from an instruction
511 	 * access fault and pass that down to __hash_page so we avoid
512 	 * double-faulting on execution of fresh text. We have to test
513 	 * for regs NULL since init will get here first thing at boot
514 	 *
515 	 * We also avoid filling the hash if not coming from a fault
516 	 */
517 
518 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
519 	switch (trap) {
520 	case 0x300:
521 		is_exec = false;
522 		break;
523 	case 0x400:
524 		is_exec = true;
525 		break;
526 	default:
527 		return;
528 	}
529 
530 	hash_preload(vma->vm_mm, address, is_exec, trap);
531 #endif /* CONFIG_PPC_BOOK3S */
532 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
533 	&& defined(CONFIG_HUGETLB_PAGE)
534 	if (is_vm_hugetlb_page(vma))
535 		book3e_hugetlb_preload(vma, address, *ptep);
536 #endif
537 }
538 
539 /*
540  * System memory should not be in /proc/iomem but various tools expect it
541  * (eg kdump).
542  */
543 static int __init add_system_ram_resources(void)
544 {
545 	struct memblock_region *reg;
546 
547 	for_each_memblock(memory, reg) {
548 		struct resource *res;
549 		unsigned long base = reg->base;
550 		unsigned long size = reg->size;
551 
552 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
553 		WARN_ON(!res);
554 
555 		if (res) {
556 			res->name = "System RAM";
557 			res->start = base;
558 			res->end = base + size - 1;
559 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
560 			WARN_ON(request_resource(&iomem_resource, res) < 0);
561 		}
562 	}
563 
564 	return 0;
565 }
566 subsys_initcall(add_system_ram_resources);
567 
568 #ifdef CONFIG_STRICT_DEVMEM
569 /*
570  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
571  * is valid. The argument is a physical page number.
572  *
573  * Access has to be given to non-kernel-ram areas as well, these contain the
574  * PCI mmio resources as well as potential bios/acpi data regions.
575  */
576 int devmem_is_allowed(unsigned long pfn)
577 {
578 	if (page_is_rtas_user_buf(pfn))
579 		return 1;
580 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
581 		return 0;
582 	if (!page_is_ram(pfn))
583 		return 1;
584 	return 0;
585 }
586 #endif /* CONFIG_STRICT_DEVMEM */
587