xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision e657c18a) 
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 
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 #endif
79 
80 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
81 			      unsigned long size, pgprot_t vma_prot)
82 {
83 	if (ppc_md.phys_mem_access_prot)
84 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
85 
86 	if (!page_is_ram(pfn))
87 		vma_prot = pgprot_noncached(vma_prot);
88 
89 	return vma_prot;
90 }
91 EXPORT_SYMBOL(phys_mem_access_prot);
92 
93 #ifdef CONFIG_MEMORY_HOTPLUG
94 
95 #ifdef CONFIG_NUMA
96 int memory_add_physaddr_to_nid(u64 start)
97 {
98 	return hot_add_scn_to_nid(start);
99 }
100 #endif
101 
102 int __weak create_section_mapping(unsigned long start, unsigned long end, int nid)
103 {
104 	return -ENODEV;
105 }
106 
107 int __weak remove_section_mapping(unsigned long start, unsigned long end)
108 {
109 	return -ENODEV;
110 }
111 
112 int __meminit arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
113 		bool want_memblock)
114 {
115 	unsigned long start_pfn = start >> PAGE_SHIFT;
116 	unsigned long nr_pages = size >> PAGE_SHIFT;
117 	int rc;
118 
119 	resize_hpt_for_hotplug(memblock_phys_mem_size());
120 
121 	start = (unsigned long)__va(start);
122 	rc = create_section_mapping(start, start + size, nid);
123 	if (rc) {
124 		pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
125 			start, start + size, rc);
126 		return -EFAULT;
127 	}
128 	flush_inval_dcache_range(start, start + size);
129 
130 	return __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
131 }
132 
133 #ifdef CONFIG_MEMORY_HOTREMOVE
134 int __meminit arch_remove_memory(int nid, u64 start, u64 size,
135 					struct vmem_altmap *altmap)
136 {
137 	unsigned long start_pfn = start >> PAGE_SHIFT;
138 	unsigned long nr_pages = size >> PAGE_SHIFT;
139 	struct page *page;
140 	int ret;
141 
142 	/*
143 	 * If we have an altmap then we need to skip over any reserved PFNs
144 	 * when querying the zone.
145 	 */
146 	page = pfn_to_page(start_pfn);
147 	if (altmap)
148 		page += vmem_altmap_offset(altmap);
149 
150 	ret = __remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
151 	if (ret)
152 		return ret;
153 
154 	/* Remove htab bolted mappings for this section of memory */
155 	start = (unsigned long)__va(start);
156 	flush_inval_dcache_range(start, start + size);
157 	ret = remove_section_mapping(start, start + size);
158 
159 	/* Ensure all vmalloc mappings are flushed in case they also
160 	 * hit that section of memory
161 	 */
162 	vm_unmap_aliases();
163 
164 	resize_hpt_for_hotplug(memblock_phys_mem_size());
165 
166 	return ret;
167 }
168 #endif
169 #endif /* CONFIG_MEMORY_HOTPLUG */
170 
171 #ifndef CONFIG_NEED_MULTIPLE_NODES
172 void __init mem_topology_setup(void)
173 {
174 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
175 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
176 #ifdef CONFIG_HIGHMEM
177 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
178 #endif
179 
180 	/* Place all memblock_regions in the same node and merge contiguous
181 	 * memblock_regions
182 	 */
183 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
184 }
185 
186 void __init initmem_init(void)
187 {
188 	/* XXX need to clip this if using highmem? */
189 	sparse_memory_present_with_active_regions(0);
190 	sparse_init();
191 }
192 
193 /* mark pages that don't exist as nosave */
194 static int __init mark_nonram_nosave(void)
195 {
196 	struct memblock_region *reg, *prev = NULL;
197 
198 	for_each_memblock(memory, reg) {
199 		if (prev &&
200 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
201 			register_nosave_region(memblock_region_memory_end_pfn(prev),
202 					       memblock_region_memory_base_pfn(reg));
203 		prev = reg;
204 	}
205 	return 0;
206 }
207 #else /* CONFIG_NEED_MULTIPLE_NODES */
208 static int __init mark_nonram_nosave(void)
209 {
210 	return 0;
211 }
212 #endif
213 
214 /*
215  * Zones usage:
216  *
217  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
218  * everything else. GFP_DMA32 page allocations automatically fall back to
219  * ZONE_DMA.
220  *
221  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
222  * inform the generic DMA mapping code.  32-bit only devices (if not handled
223  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
224  * otherwise served by ZONE_DMA.
225  */
226 static unsigned long max_zone_pfns[MAX_NR_ZONES];
227 
228 /*
229  * paging_init() sets up the page tables - in fact we've already done this.
230  */
231 void __init paging_init(void)
232 {
233 	unsigned long long total_ram = memblock_phys_mem_size();
234 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
235 
236 #ifdef CONFIG_PPC32
237 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
238 	unsigned long end = __fix_to_virt(FIX_HOLE);
239 
240 	for (; v < end; v += PAGE_SIZE)
241 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
242 #endif
243 
244 #ifdef CONFIG_HIGHMEM
245 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
246 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
247 
248 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
249 	kmap_prot = PAGE_KERNEL;
250 #endif /* CONFIG_HIGHMEM */
251 
252 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
253 	       (unsigned long long)top_of_ram, total_ram);
254 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
255 	       (long int)((top_of_ram - total_ram) >> 20));
256 
257 #ifdef CONFIG_ZONE_DMA
258 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
259 #endif
260 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
261 #ifdef CONFIG_HIGHMEM
262 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
263 #endif
264 
265 	free_area_init_nodes(max_zone_pfns);
266 
267 	mark_nonram_nosave();
268 }
269 
270 void __init mem_init(void)
271 {
272 	/*
273 	 * book3s is limited to 16 page sizes due to encoding this in
274 	 * a 4-bit field for slices.
275 	 */
276 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
277 
278 #ifdef CONFIG_SWIOTLB
279 	swiotlb_init(0);
280 #endif
281 
282 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
283 	set_max_mapnr(max_pfn);
284 	memblock_free_all();
285 
286 #ifdef CONFIG_HIGHMEM
287 	{
288 		unsigned long pfn, highmem_mapnr;
289 
290 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
291 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
292 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
293 			struct page *page = pfn_to_page(pfn);
294 			if (!memblock_is_reserved(paddr))
295 				free_highmem_page(page);
296 		}
297 	}
298 #endif /* CONFIG_HIGHMEM */
299 
300 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
301 	/*
302 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
303 	 * functions.... do it here for the non-smp case.
304 	 */
305 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
306 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
307 #endif
308 
309 	mem_init_print_info(NULL);
310 #ifdef CONFIG_PPC32
311 	pr_info("Kernel virtual memory layout:\n");
312 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
313 #ifdef CONFIG_HIGHMEM
314 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
315 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
316 #endif /* CONFIG_HIGHMEM */
317 #ifdef CONFIG_NOT_COHERENT_CACHE
318 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
319 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
320 #endif /* CONFIG_NOT_COHERENT_CACHE */
321 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
322 		ioremap_bot, IOREMAP_TOP);
323 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
324 		VMALLOC_START, VMALLOC_END);
325 #endif /* CONFIG_PPC32 */
326 }
327 
328 void free_initmem(void)
329 {
330 	ppc_md.progress = ppc_printk_progress;
331 	mark_initmem_nx();
332 	init_mem_is_free = true;
333 	free_initmem_default(POISON_FREE_INITMEM);
334 }
335 
336 #ifdef CONFIG_BLK_DEV_INITRD
337 void __init free_initrd_mem(unsigned long start, unsigned long end)
338 {
339 	free_reserved_area((void *)start, (void *)end, -1, "initrd");
340 }
341 #endif
342 
343 /*
344  * This is called when a page has been modified by the kernel.
345  * It just marks the page as not i-cache clean.  We do the i-cache
346  * flush later when the page is given to a user process, if necessary.
347  */
348 void flush_dcache_page(struct page *page)
349 {
350 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
351 		return;
352 	/* avoid an atomic op if possible */
353 	if (test_bit(PG_arch_1, &page->flags))
354 		clear_bit(PG_arch_1, &page->flags);
355 }
356 EXPORT_SYMBOL(flush_dcache_page);
357 
358 void flush_dcache_icache_page(struct page *page)
359 {
360 #ifdef CONFIG_HUGETLB_PAGE
361 	if (PageCompound(page)) {
362 		flush_dcache_icache_hugepage(page);
363 		return;
364 	}
365 #endif
366 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
367 	/* On 8xx there is no need to kmap since highmem is not supported */
368 	__flush_dcache_icache(page_address(page));
369 #else
370 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
371 		void *start = kmap_atomic(page);
372 		__flush_dcache_icache(start);
373 		kunmap_atomic(start);
374 	} else {
375 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
376 	}
377 #endif
378 }
379 EXPORT_SYMBOL(flush_dcache_icache_page);
380 
381 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
382 {
383 	clear_page(page);
384 
385 	/*
386 	 * We shouldn't have to do this, but some versions of glibc
387 	 * require it (ld.so assumes zero filled pages are icache clean)
388 	 * - Anton
389 	 */
390 	flush_dcache_page(pg);
391 }
392 EXPORT_SYMBOL(clear_user_page);
393 
394 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
395 		    struct page *pg)
396 {
397 	copy_page(vto, vfrom);
398 
399 	/*
400 	 * We should be able to use the following optimisation, however
401 	 * there are two problems.
402 	 * Firstly a bug in some versions of binutils meant PLT sections
403 	 * were not marked executable.
404 	 * Secondly the first word in the GOT section is blrl, used
405 	 * to establish the GOT address. Until recently the GOT was
406 	 * not marked executable.
407 	 * - Anton
408 	 */
409 #if 0
410 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
411 		return;
412 #endif
413 
414 	flush_dcache_page(pg);
415 }
416 
417 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
418 			     unsigned long addr, int len)
419 {
420 	unsigned long maddr;
421 
422 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
423 	flush_icache_range(maddr, maddr + len);
424 	kunmap(page);
425 }
426 EXPORT_SYMBOL(flush_icache_user_range);
427 
428 /*
429  * This is called at the end of handling a user page fault, when the
430  * fault has been handled by updating a PTE in the linux page tables.
431  * We use it to preload an HPTE into the hash table corresponding to
432  * the updated linux PTE.
433  *
434  * This must always be called with the pte lock held.
435  */
436 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
437 		      pte_t *ptep)
438 {
439 #ifdef CONFIG_PPC_BOOK3S
440 	/*
441 	 * We don't need to worry about _PAGE_PRESENT here because we are
442 	 * called with either mm->page_table_lock held or ptl lock held
443 	 */
444 	unsigned long trap;
445 	bool is_exec;
446 
447 	if (radix_enabled()) {
448 		prefetch((void *)address);
449 		return;
450 	}
451 
452 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
453 	if (!pte_young(*ptep) || address >= TASK_SIZE)
454 		return;
455 
456 	/* We try to figure out if we are coming from an instruction
457 	 * access fault and pass that down to __hash_page so we avoid
458 	 * double-faulting on execution of fresh text. We have to test
459 	 * for regs NULL since init will get here first thing at boot
460 	 *
461 	 * We also avoid filling the hash if not coming from a fault
462 	 */
463 
464 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
465 	switch (trap) {
466 	case 0x300:
467 		is_exec = false;
468 		break;
469 	case 0x400:
470 		is_exec = true;
471 		break;
472 	default:
473 		return;
474 	}
475 
476 	hash_preload(vma->vm_mm, address, is_exec, trap);
477 #endif /* CONFIG_PPC_BOOK3S */
478 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
479 	&& defined(CONFIG_HUGETLB_PAGE)
480 	if (is_vm_hugetlb_page(vma))
481 		book3e_hugetlb_preload(vma, address, *ptep);
482 #endif
483 }
484 
485 /*
486  * System memory should not be in /proc/iomem but various tools expect it
487  * (eg kdump).
488  */
489 static int __init add_system_ram_resources(void)
490 {
491 	struct memblock_region *reg;
492 
493 	for_each_memblock(memory, reg) {
494 		struct resource *res;
495 		unsigned long base = reg->base;
496 		unsigned long size = reg->size;
497 
498 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
499 		WARN_ON(!res);
500 
501 		if (res) {
502 			res->name = "System RAM";
503 			res->start = base;
504 			res->end = base + size - 1;
505 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
506 			WARN_ON(request_resource(&iomem_resource, res) < 0);
507 		}
508 	}
509 
510 	return 0;
511 }
512 subsys_initcall(add_system_ram_resources);
513 
514 #ifdef CONFIG_STRICT_DEVMEM
515 /*
516  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
517  * is valid. The argument is a physical page number.
518  *
519  * Access has to be given to non-kernel-ram areas as well, these contain the
520  * PCI mmio resources as well as potential bios/acpi data regions.
521  */
522 int devmem_is_allowed(unsigned long pfn)
523 {
524 	if (page_is_rtas_user_buf(pfn))
525 		return 1;
526 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
527 		return 0;
528 	if (!page_is_ram(pfn))
529 		return 1;
530 	return 0;
531 }
532 #endif /* CONFIG_STRICT_DEVMEM */
533 
534 /*
535  * This is defined in kernel/resource.c but only powerpc needs to export it, for
536  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
537  */
538 EXPORT_SYMBOL_GPL(walk_system_ram_range);
539