xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision ae213c44) 
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 <mm/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 __ref arch_add_memory(int nid, u64 start, u64 size,
113 			struct mhp_restrictions *restrictions)
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, restrictions);
131 }
132 
133 #ifdef CONFIG_MEMORY_HOTREMOVE
134 void __ref 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 	__remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
151 
152 	/* Remove htab bolted mappings for this section of memory */
153 	start = (unsigned long)__va(start);
154 	flush_inval_dcache_range(start, start + size);
155 	ret = remove_section_mapping(start, start + size);
156 	WARN_ON_ONCE(ret);
157 
158 	/* Ensure all vmalloc mappings are flushed in case they also
159 	 * hit that section of memory
160 	 */
161 	vm_unmap_aliases();
162 
163 	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
164 		pr_warn("Hash collision while resizing HPT\n");
165 }
166 #endif
167 #endif /* CONFIG_MEMORY_HOTPLUG */
168 
169 #ifndef CONFIG_NEED_MULTIPLE_NODES
170 void __init mem_topology_setup(void)
171 {
172 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
173 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
174 #ifdef CONFIG_HIGHMEM
175 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
176 #endif
177 
178 	/* Place all memblock_regions in the same node and merge contiguous
179 	 * memblock_regions
180 	 */
181 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
182 }
183 
184 void __init initmem_init(void)
185 {
186 	/* XXX need to clip this if using highmem? */
187 	sparse_memory_present_with_active_regions(0);
188 	sparse_init();
189 }
190 
191 /* mark pages that don't exist as nosave */
192 static int __init mark_nonram_nosave(void)
193 {
194 	struct memblock_region *reg, *prev = NULL;
195 
196 	for_each_memblock(memory, reg) {
197 		if (prev &&
198 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
199 			register_nosave_region(memblock_region_memory_end_pfn(prev),
200 					       memblock_region_memory_base_pfn(reg));
201 		prev = reg;
202 	}
203 	return 0;
204 }
205 #else /* CONFIG_NEED_MULTIPLE_NODES */
206 static int __init mark_nonram_nosave(void)
207 {
208 	return 0;
209 }
210 #endif
211 
212 /*
213  * Zones usage:
214  *
215  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
216  * everything else. GFP_DMA32 page allocations automatically fall back to
217  * ZONE_DMA.
218  *
219  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
220  * inform the generic DMA mapping code.  32-bit only devices (if not handled
221  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
222  * otherwise served by ZONE_DMA.
223  */
224 static unsigned long max_zone_pfns[MAX_NR_ZONES];
225 
226 /*
227  * paging_init() sets up the page tables - in fact we've already done this.
228  */
229 void __init paging_init(void)
230 {
231 	unsigned long long total_ram = memblock_phys_mem_size();
232 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
233 
234 #ifdef CONFIG_PPC32
235 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
236 	unsigned long end = __fix_to_virt(FIX_HOLE);
237 
238 	for (; v < end; v += PAGE_SIZE)
239 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
240 #endif
241 
242 #ifdef CONFIG_HIGHMEM
243 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
244 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
245 
246 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
247 	kmap_prot = PAGE_KERNEL;
248 #endif /* CONFIG_HIGHMEM */
249 
250 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
251 	       (unsigned long long)top_of_ram, total_ram);
252 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
253 	       (long int)((top_of_ram - total_ram) >> 20));
254 
255 #ifdef CONFIG_ZONE_DMA
256 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
257 #endif
258 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
259 #ifdef CONFIG_HIGHMEM
260 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
261 #endif
262 
263 	free_area_init_nodes(max_zone_pfns);
264 
265 	mark_nonram_nosave();
266 }
267 
268 void __init mem_init(void)
269 {
270 	/*
271 	 * book3s is limited to 16 page sizes due to encoding this in
272 	 * a 4-bit field for slices.
273 	 */
274 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
275 
276 #ifdef CONFIG_SWIOTLB
277 	swiotlb_init(0);
278 #endif
279 
280 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
281 	set_max_mapnr(max_pfn);
282 	memblock_free_all();
283 
284 #ifdef CONFIG_HIGHMEM
285 	{
286 		unsigned long pfn, highmem_mapnr;
287 
288 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
289 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
290 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
291 			struct page *page = pfn_to_page(pfn);
292 			if (!memblock_is_reserved(paddr))
293 				free_highmem_page(page);
294 		}
295 	}
296 #endif /* CONFIG_HIGHMEM */
297 
298 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
299 	/*
300 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
301 	 * functions.... do it here for the non-smp case.
302 	 */
303 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
304 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
305 #endif
306 
307 	mem_init_print_info(NULL);
308 #ifdef CONFIG_PPC32
309 	pr_info("Kernel virtual memory layout:\n");
310 #ifdef CONFIG_KASAN
311 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
312 		KASAN_SHADOW_START, KASAN_SHADOW_END);
313 #endif
314 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
315 #ifdef CONFIG_HIGHMEM
316 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
317 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
318 #endif /* CONFIG_HIGHMEM */
319 #ifdef CONFIG_NOT_COHERENT_CACHE
320 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
321 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
322 #endif /* CONFIG_NOT_COHERENT_CACHE */
323 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
324 		ioremap_bot, IOREMAP_TOP);
325 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
326 		VMALLOC_START, VMALLOC_END);
327 #endif /* CONFIG_PPC32 */
328 }
329 
330 void free_initmem(void)
331 {
332 	ppc_md.progress = ppc_printk_progress;
333 	mark_initmem_nx();
334 	init_mem_is_free = true;
335 	free_initmem_default(POISON_FREE_INITMEM);
336 }
337 
338 /*
339  * This is called when a page has been modified by the kernel.
340  * It just marks the page as not i-cache clean.  We do the i-cache
341  * flush later when the page is given to a user process, if necessary.
342  */
343 void flush_dcache_page(struct page *page)
344 {
345 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
346 		return;
347 	/* avoid an atomic op if possible */
348 	if (test_bit(PG_arch_1, &page->flags))
349 		clear_bit(PG_arch_1, &page->flags);
350 }
351 EXPORT_SYMBOL(flush_dcache_page);
352 
353 void flush_dcache_icache_page(struct page *page)
354 {
355 #ifdef CONFIG_HUGETLB_PAGE
356 	if (PageCompound(page)) {
357 		flush_dcache_icache_hugepage(page);
358 		return;
359 	}
360 #endif
361 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
362 	/* On 8xx there is no need to kmap since highmem is not supported */
363 	__flush_dcache_icache(page_address(page));
364 #else
365 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
366 		void *start = kmap_atomic(page);
367 		__flush_dcache_icache(start);
368 		kunmap_atomic(start);
369 	} else {
370 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
371 	}
372 #endif
373 }
374 EXPORT_SYMBOL(flush_dcache_icache_page);
375 
376 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
377 {
378 	clear_page(page);
379 
380 	/*
381 	 * We shouldn't have to do this, but some versions of glibc
382 	 * require it (ld.so assumes zero filled pages are icache clean)
383 	 * - Anton
384 	 */
385 	flush_dcache_page(pg);
386 }
387 EXPORT_SYMBOL(clear_user_page);
388 
389 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
390 		    struct page *pg)
391 {
392 	copy_page(vto, vfrom);
393 
394 	/*
395 	 * We should be able to use the following optimisation, however
396 	 * there are two problems.
397 	 * Firstly a bug in some versions of binutils meant PLT sections
398 	 * were not marked executable.
399 	 * Secondly the first word in the GOT section is blrl, used
400 	 * to establish the GOT address. Until recently the GOT was
401 	 * not marked executable.
402 	 * - Anton
403 	 */
404 #if 0
405 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
406 		return;
407 #endif
408 
409 	flush_dcache_page(pg);
410 }
411 
412 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
413 			     unsigned long addr, int len)
414 {
415 	unsigned long maddr;
416 
417 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
418 	flush_icache_range(maddr, maddr + len);
419 	kunmap(page);
420 }
421 EXPORT_SYMBOL(flush_icache_user_range);
422 
423 /*
424  * This is called at the end of handling a user page fault, when the
425  * fault has been handled by updating a PTE in the linux page tables.
426  * We use it to preload an HPTE into the hash table corresponding to
427  * the updated linux PTE.
428  *
429  * This must always be called with the pte lock held.
430  */
431 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
432 		      pte_t *ptep)
433 {
434 #ifdef CONFIG_PPC_BOOK3S
435 	/*
436 	 * We don't need to worry about _PAGE_PRESENT here because we are
437 	 * called with either mm->page_table_lock held or ptl lock held
438 	 */
439 	unsigned long trap;
440 	bool is_exec;
441 
442 	if (radix_enabled()) {
443 		prefetch((void *)address);
444 		return;
445 	}
446 
447 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
448 	if (!pte_young(*ptep) || address >= TASK_SIZE)
449 		return;
450 
451 	/* We try to figure out if we are coming from an instruction
452 	 * access fault and pass that down to __hash_page so we avoid
453 	 * double-faulting on execution of fresh text. We have to test
454 	 * for regs NULL since init will get here first thing at boot
455 	 *
456 	 * We also avoid filling the hash if not coming from a fault
457 	 */
458 
459 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
460 	switch (trap) {
461 	case 0x300:
462 		is_exec = false;
463 		break;
464 	case 0x400:
465 		is_exec = true;
466 		break;
467 	default:
468 		return;
469 	}
470 
471 	hash_preload(vma->vm_mm, address, is_exec, trap);
472 #endif /* CONFIG_PPC_BOOK3S */
473 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
474 	&& defined(CONFIG_HUGETLB_PAGE)
475 	if (is_vm_hugetlb_page(vma))
476 		book3e_hugetlb_preload(vma, address, *ptep);
477 #endif
478 }
479 
480 /*
481  * System memory should not be in /proc/iomem but various tools expect it
482  * (eg kdump).
483  */
484 static int __init add_system_ram_resources(void)
485 {
486 	struct memblock_region *reg;
487 
488 	for_each_memblock(memory, reg) {
489 		struct resource *res;
490 		unsigned long base = reg->base;
491 		unsigned long size = reg->size;
492 
493 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
494 		WARN_ON(!res);
495 
496 		if (res) {
497 			res->name = "System RAM";
498 			res->start = base;
499 			res->end = base + size - 1;
500 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
501 			WARN_ON(request_resource(&iomem_resource, res) < 0);
502 		}
503 	}
504 
505 	return 0;
506 }
507 subsys_initcall(add_system_ram_resources);
508 
509 #ifdef CONFIG_STRICT_DEVMEM
510 /*
511  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
512  * is valid. The argument is a physical page number.
513  *
514  * Access has to be given to non-kernel-ram areas as well, these contain the
515  * PCI mmio resources as well as potential bios/acpi data regions.
516  */
517 int devmem_is_allowed(unsigned long pfn)
518 {
519 	if (page_is_rtas_user_buf(pfn))
520 		return 1;
521 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
522 		return 0;
523 	if (!page_is_ram(pfn))
524 		return 1;
525 	return 0;
526 }
527 #endif /* CONFIG_STRICT_DEVMEM */
528 
529 /*
530  * This is defined in kernel/resource.c but only powerpc needs to export it, for
531  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
532  */
533 EXPORT_SYMBOL_GPL(walk_system_ram_range);
534