xref: /linux/arch/powerpc/mm/mem.c (revision 2dbc0838bcf24ca59cabc3130cf3b1d6809cdcd4)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *  PowerPC version
4  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
5  *
6  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
7  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
8  *    Copyright (C) 1996 Paul Mackerras
9  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
10  *
11  *  Derived from "arch/i386/mm/init.c"
12  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
13  */
14 
15 #include <linux/export.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/errno.h>
19 #include <linux/string.h>
20 #include <linux/gfp.h>
21 #include <linux/types.h>
22 #include <linux/mm.h>
23 #include <linux/stddef.h>
24 #include <linux/init.h>
25 #include <linux/memblock.h>
26 #include <linux/highmem.h>
27 #include <linux/initrd.h>
28 #include <linux/pagemap.h>
29 #include <linux/suspend.h>
30 #include <linux/hugetlb.h>
31 #include <linux/slab.h>
32 #include <linux/vmalloc.h>
33 #include <linux/memremap.h>
34 
35 #include <asm/pgalloc.h>
36 #include <asm/prom.h>
37 #include <asm/io.h>
38 #include <asm/mmu_context.h>
39 #include <asm/pgtable.h>
40 #include <asm/mmu.h>
41 #include <asm/smp.h>
42 #include <asm/machdep.h>
43 #include <asm/btext.h>
44 #include <asm/tlb.h>
45 #include <asm/sections.h>
46 #include <asm/sparsemem.h>
47 #include <asm/vdso.h>
48 #include <asm/fixmap.h>
49 #include <asm/swiotlb.h>
50 #include <asm/rtas.h>
51 
52 #include <mm/mmu_decl.h>
53 
54 #ifndef CPU_FTR_COHERENT_ICACHE
55 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
56 #define CPU_FTR_NOEXECUTE	0
57 #endif
58 
59 unsigned long long memory_limit;
60 bool init_mem_is_free;
61 
62 #ifdef CONFIG_HIGHMEM
63 pte_t *kmap_pte;
64 EXPORT_SYMBOL(kmap_pte);
65 pgprot_t kmap_prot;
66 EXPORT_SYMBOL(kmap_prot);
67 
68 static inline pte_t *virt_to_kpte(unsigned long vaddr)
69 {
70 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
71 			vaddr), vaddr), vaddr);
72 }
73 #endif
74 
75 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
76 			      unsigned long size, pgprot_t vma_prot)
77 {
78 	if (ppc_md.phys_mem_access_prot)
79 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
80 
81 	if (!page_is_ram(pfn))
82 		vma_prot = pgprot_noncached(vma_prot);
83 
84 	return vma_prot;
85 }
86 EXPORT_SYMBOL(phys_mem_access_prot);
87 
88 #ifdef CONFIG_MEMORY_HOTPLUG
89 
90 #ifdef CONFIG_NUMA
91 int memory_add_physaddr_to_nid(u64 start)
92 {
93 	return hot_add_scn_to_nid(start);
94 }
95 #endif
96 
97 int __weak create_section_mapping(unsigned long start, unsigned long end, int nid)
98 {
99 	return -ENODEV;
100 }
101 
102 int __weak remove_section_mapping(unsigned long start, unsigned long end)
103 {
104 	return -ENODEV;
105 }
106 
107 int __ref arch_add_memory(int nid, u64 start, u64 size,
108 			struct mhp_restrictions *restrictions)
109 {
110 	unsigned long start_pfn = start >> PAGE_SHIFT;
111 	unsigned long nr_pages = size >> PAGE_SHIFT;
112 	int rc;
113 
114 	resize_hpt_for_hotplug(memblock_phys_mem_size());
115 
116 	start = (unsigned long)__va(start);
117 	rc = create_section_mapping(start, start + size, nid);
118 	if (rc) {
119 		pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
120 			start, start + size, rc);
121 		return -EFAULT;
122 	}
123 	flush_dcache_range(start, start + size);
124 
125 	return __add_pages(nid, start_pfn, nr_pages, restrictions);
126 }
127 
128 #ifdef CONFIG_MEMORY_HOTREMOVE
129 void __ref arch_remove_memory(int nid, u64 start, u64 size,
130 			     struct vmem_altmap *altmap)
131 {
132 	unsigned long start_pfn = start >> PAGE_SHIFT;
133 	unsigned long nr_pages = size >> PAGE_SHIFT;
134 	struct page *page = pfn_to_page(start_pfn) + vmem_altmap_offset(altmap);
135 	int ret;
136 
137 	__remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
138 
139 	/* Remove htab bolted mappings for this section of memory */
140 	start = (unsigned long)__va(start);
141 	flush_dcache_range(start, start + size);
142 	ret = remove_section_mapping(start, start + size);
143 	WARN_ON_ONCE(ret);
144 
145 	/* Ensure all vmalloc mappings are flushed in case they also
146 	 * hit that section of memory
147 	 */
148 	vm_unmap_aliases();
149 
150 	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
151 		pr_warn("Hash collision while resizing HPT\n");
152 }
153 #endif
154 #endif /* CONFIG_MEMORY_HOTPLUG */
155 
156 #ifndef CONFIG_NEED_MULTIPLE_NODES
157 void __init mem_topology_setup(void)
158 {
159 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
160 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
161 #ifdef CONFIG_HIGHMEM
162 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
163 #endif
164 
165 	/* Place all memblock_regions in the same node and merge contiguous
166 	 * memblock_regions
167 	 */
168 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
169 }
170 
171 void __init initmem_init(void)
172 {
173 	/* XXX need to clip this if using highmem? */
174 	sparse_memory_present_with_active_regions(0);
175 	sparse_init();
176 }
177 
178 /* mark pages that don't exist as nosave */
179 static int __init mark_nonram_nosave(void)
180 {
181 	struct memblock_region *reg, *prev = NULL;
182 
183 	for_each_memblock(memory, reg) {
184 		if (prev &&
185 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
186 			register_nosave_region(memblock_region_memory_end_pfn(prev),
187 					       memblock_region_memory_base_pfn(reg));
188 		prev = reg;
189 	}
190 	return 0;
191 }
192 #else /* CONFIG_NEED_MULTIPLE_NODES */
193 static int __init mark_nonram_nosave(void)
194 {
195 	return 0;
196 }
197 #endif
198 
199 /*
200  * Zones usage:
201  *
202  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
203  * everything else. GFP_DMA32 page allocations automatically fall back to
204  * ZONE_DMA.
205  *
206  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
207  * inform the generic DMA mapping code.  32-bit only devices (if not handled
208  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
209  * otherwise served by ZONE_DMA.
210  */
211 static unsigned long max_zone_pfns[MAX_NR_ZONES];
212 
213 /*
214  * paging_init() sets up the page tables - in fact we've already done this.
215  */
216 void __init paging_init(void)
217 {
218 	unsigned long long total_ram = memblock_phys_mem_size();
219 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
220 
221 #ifdef CONFIG_PPC32
222 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
223 	unsigned long end = __fix_to_virt(FIX_HOLE);
224 
225 	for (; v < end; v += PAGE_SIZE)
226 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
227 #endif
228 
229 #ifdef CONFIG_HIGHMEM
230 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
231 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
232 
233 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
234 	kmap_prot = PAGE_KERNEL;
235 #endif /* CONFIG_HIGHMEM */
236 
237 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
238 	       (unsigned long long)top_of_ram, total_ram);
239 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
240 	       (long int)((top_of_ram - total_ram) >> 20));
241 
242 #ifdef CONFIG_ZONE_DMA
243 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
244 			((1UL << ARCH_ZONE_DMA_BITS) - 1) >> PAGE_SHIFT);
245 #endif
246 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
247 #ifdef CONFIG_HIGHMEM
248 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
249 #endif
250 
251 	free_area_init_nodes(max_zone_pfns);
252 
253 	mark_nonram_nosave();
254 }
255 
256 void __init mem_init(void)
257 {
258 	/*
259 	 * book3s is limited to 16 page sizes due to encoding this in
260 	 * a 4-bit field for slices.
261 	 */
262 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
263 
264 #ifdef CONFIG_SWIOTLB
265 	swiotlb_init(0);
266 #endif
267 
268 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
269 	set_max_mapnr(max_pfn);
270 	memblock_free_all();
271 
272 #ifdef CONFIG_HIGHMEM
273 	{
274 		unsigned long pfn, highmem_mapnr;
275 
276 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
277 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
278 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
279 			struct page *page = pfn_to_page(pfn);
280 			if (!memblock_is_reserved(paddr))
281 				free_highmem_page(page);
282 		}
283 	}
284 #endif /* CONFIG_HIGHMEM */
285 
286 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
287 	/*
288 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
289 	 * functions.... do it here for the non-smp case.
290 	 */
291 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
292 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
293 #endif
294 
295 	mem_init_print_info(NULL);
296 #ifdef CONFIG_PPC32
297 	pr_info("Kernel virtual memory layout:\n");
298 #ifdef CONFIG_KASAN
299 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
300 		KASAN_SHADOW_START, KASAN_SHADOW_END);
301 #endif
302 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
303 #ifdef CONFIG_HIGHMEM
304 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
305 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
306 #endif /* CONFIG_HIGHMEM */
307 #ifdef CONFIG_NOT_COHERENT_CACHE
308 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
309 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
310 #endif /* CONFIG_NOT_COHERENT_CACHE */
311 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
312 		ioremap_bot, IOREMAP_TOP);
313 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
314 		VMALLOC_START, VMALLOC_END);
315 #endif /* CONFIG_PPC32 */
316 }
317 
318 void free_initmem(void)
319 {
320 	ppc_md.progress = ppc_printk_progress;
321 	mark_initmem_nx();
322 	init_mem_is_free = true;
323 	free_initmem_default(POISON_FREE_INITMEM);
324 }
325 
326 /*
327  * This is called when a page has been modified by the kernel.
328  * It just marks the page as not i-cache clean.  We do the i-cache
329  * flush later when the page is given to a user process, if necessary.
330  */
331 void flush_dcache_page(struct page *page)
332 {
333 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
334 		return;
335 	/* avoid an atomic op if possible */
336 	if (test_bit(PG_arch_1, &page->flags))
337 		clear_bit(PG_arch_1, &page->flags);
338 }
339 EXPORT_SYMBOL(flush_dcache_page);
340 
341 void flush_dcache_icache_page(struct page *page)
342 {
343 #ifdef CONFIG_HUGETLB_PAGE
344 	if (PageCompound(page)) {
345 		flush_dcache_icache_hugepage(page);
346 		return;
347 	}
348 #endif
349 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
350 	/* On 8xx there is no need to kmap since highmem is not supported */
351 	__flush_dcache_icache(page_address(page));
352 #else
353 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
354 		void *start = kmap_atomic(page);
355 		__flush_dcache_icache(start);
356 		kunmap_atomic(start);
357 	} else {
358 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
359 	}
360 #endif
361 }
362 EXPORT_SYMBOL(flush_dcache_icache_page);
363 
364 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
365 {
366 	clear_page(page);
367 
368 	/*
369 	 * We shouldn't have to do this, but some versions of glibc
370 	 * require it (ld.so assumes zero filled pages are icache clean)
371 	 * - Anton
372 	 */
373 	flush_dcache_page(pg);
374 }
375 EXPORT_SYMBOL(clear_user_page);
376 
377 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
378 		    struct page *pg)
379 {
380 	copy_page(vto, vfrom);
381 
382 	/*
383 	 * We should be able to use the following optimisation, however
384 	 * there are two problems.
385 	 * Firstly a bug in some versions of binutils meant PLT sections
386 	 * were not marked executable.
387 	 * Secondly the first word in the GOT section is blrl, used
388 	 * to establish the GOT address. Until recently the GOT was
389 	 * not marked executable.
390 	 * - Anton
391 	 */
392 #if 0
393 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
394 		return;
395 #endif
396 
397 	flush_dcache_page(pg);
398 }
399 
400 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
401 			     unsigned long addr, int len)
402 {
403 	unsigned long maddr;
404 
405 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
406 	flush_icache_range(maddr, maddr + len);
407 	kunmap(page);
408 }
409 EXPORT_SYMBOL(flush_icache_user_range);
410 
411 /*
412  * This is called at the end of handling a user page fault, when the
413  * fault has been handled by updating a PTE in the linux page tables.
414  * We use it to preload an HPTE into the hash table corresponding to
415  * the updated linux PTE.
416  *
417  * This must always be called with the pte lock held.
418  */
419 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
420 		      pte_t *ptep)
421 {
422 #ifdef CONFIG_PPC_BOOK3S
423 	/*
424 	 * We don't need to worry about _PAGE_PRESENT here because we are
425 	 * called with either mm->page_table_lock held or ptl lock held
426 	 */
427 	unsigned long trap;
428 	bool is_exec;
429 
430 	if (radix_enabled()) {
431 		prefetch((void *)address);
432 		return;
433 	}
434 
435 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
436 	if (!pte_young(*ptep) || address >= TASK_SIZE)
437 		return;
438 
439 	/* We try to figure out if we are coming from an instruction
440 	 * access fault and pass that down to __hash_page so we avoid
441 	 * double-faulting on execution of fresh text. We have to test
442 	 * for regs NULL since init will get here first thing at boot
443 	 *
444 	 * We also avoid filling the hash if not coming from a fault
445 	 */
446 
447 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
448 	switch (trap) {
449 	case 0x300:
450 		is_exec = false;
451 		break;
452 	case 0x400:
453 		is_exec = true;
454 		break;
455 	default:
456 		return;
457 	}
458 
459 	hash_preload(vma->vm_mm, address, is_exec, trap);
460 #endif /* CONFIG_PPC_BOOK3S */
461 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
462 	&& defined(CONFIG_HUGETLB_PAGE)
463 	if (is_vm_hugetlb_page(vma))
464 		book3e_hugetlb_preload(vma, address, *ptep);
465 #endif
466 }
467 
468 /*
469  * System memory should not be in /proc/iomem but various tools expect it
470  * (eg kdump).
471  */
472 static int __init add_system_ram_resources(void)
473 {
474 	struct memblock_region *reg;
475 
476 	for_each_memblock(memory, reg) {
477 		struct resource *res;
478 		unsigned long base = reg->base;
479 		unsigned long size = reg->size;
480 
481 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
482 		WARN_ON(!res);
483 
484 		if (res) {
485 			res->name = "System RAM";
486 			res->start = base;
487 			res->end = base + size - 1;
488 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
489 			WARN_ON(request_resource(&iomem_resource, res) < 0);
490 		}
491 	}
492 
493 	return 0;
494 }
495 subsys_initcall(add_system_ram_resources);
496 
497 #ifdef CONFIG_STRICT_DEVMEM
498 /*
499  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
500  * is valid. The argument is a physical page number.
501  *
502  * Access has to be given to non-kernel-ram areas as well, these contain the
503  * PCI mmio resources as well as potential bios/acpi data regions.
504  */
505 int devmem_is_allowed(unsigned long pfn)
506 {
507 	if (page_is_rtas_user_buf(pfn))
508 		return 1;
509 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
510 		return 0;
511 	if (!page_is_ram(pfn))
512 		return 1;
513 	return 0;
514 }
515 #endif /* CONFIG_STRICT_DEVMEM */
516 
517 /*
518  * This is defined in kernel/resource.c but only powerpc needs to export it, for
519  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
520  */
521 EXPORT_SYMBOL_GPL(walk_system_ram_range);
522