xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision b830f94f)
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 void __ref arch_remove_memory(int nid, u64 start, u64 size,
129 			     struct vmem_altmap *altmap)
130 {
131 	unsigned long start_pfn = start >> PAGE_SHIFT;
132 	unsigned long nr_pages = size >> PAGE_SHIFT;
133 	struct page *page = pfn_to_page(start_pfn) + vmem_altmap_offset(altmap);
134 	int ret;
135 
136 	__remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
137 
138 	/* Remove htab bolted mappings for this section of memory */
139 	start = (unsigned long)__va(start);
140 	flush_dcache_range(start, start + size);
141 	ret = remove_section_mapping(start, start + size);
142 	WARN_ON_ONCE(ret);
143 
144 	/* Ensure all vmalloc mappings are flushed in case they also
145 	 * hit that section of memory
146 	 */
147 	vm_unmap_aliases();
148 
149 	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
150 		pr_warn("Hash collision while resizing HPT\n");
151 }
152 #endif
153 
154 #ifndef CONFIG_NEED_MULTIPLE_NODES
155 void __init mem_topology_setup(void)
156 {
157 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
158 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
159 #ifdef CONFIG_HIGHMEM
160 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
161 #endif
162 
163 	/* Place all memblock_regions in the same node and merge contiguous
164 	 * memblock_regions
165 	 */
166 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
167 }
168 
169 void __init initmem_init(void)
170 {
171 	/* XXX need to clip this if using highmem? */
172 	sparse_memory_present_with_active_regions(0);
173 	sparse_init();
174 }
175 
176 /* mark pages that don't exist as nosave */
177 static int __init mark_nonram_nosave(void)
178 {
179 	struct memblock_region *reg, *prev = NULL;
180 
181 	for_each_memblock(memory, reg) {
182 		if (prev &&
183 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
184 			register_nosave_region(memblock_region_memory_end_pfn(prev),
185 					       memblock_region_memory_base_pfn(reg));
186 		prev = reg;
187 	}
188 	return 0;
189 }
190 #else /* CONFIG_NEED_MULTIPLE_NODES */
191 static int __init mark_nonram_nosave(void)
192 {
193 	return 0;
194 }
195 #endif
196 
197 /*
198  * Zones usage:
199  *
200  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
201  * everything else. GFP_DMA32 page allocations automatically fall back to
202  * ZONE_DMA.
203  *
204  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
205  * inform the generic DMA mapping code.  32-bit only devices (if not handled
206  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
207  * otherwise served by ZONE_DMA.
208  */
209 static unsigned long max_zone_pfns[MAX_NR_ZONES];
210 
211 /*
212  * paging_init() sets up the page tables - in fact we've already done this.
213  */
214 void __init paging_init(void)
215 {
216 	unsigned long long total_ram = memblock_phys_mem_size();
217 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
218 
219 #ifdef CONFIG_PPC32
220 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
221 	unsigned long end = __fix_to_virt(FIX_HOLE);
222 
223 	for (; v < end; v += PAGE_SIZE)
224 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
225 #endif
226 
227 #ifdef CONFIG_HIGHMEM
228 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
229 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
230 
231 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
232 	kmap_prot = PAGE_KERNEL;
233 #endif /* CONFIG_HIGHMEM */
234 
235 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
236 	       (unsigned long long)top_of_ram, total_ram);
237 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
238 	       (long int)((top_of_ram - total_ram) >> 20));
239 
240 #ifdef CONFIG_ZONE_DMA
241 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
242 			((1UL << ARCH_ZONE_DMA_BITS) - 1) >> PAGE_SHIFT);
243 #endif
244 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
245 #ifdef CONFIG_HIGHMEM
246 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
247 #endif
248 
249 	free_area_init_nodes(max_zone_pfns);
250 
251 	mark_nonram_nosave();
252 }
253 
254 void __init mem_init(void)
255 {
256 	/*
257 	 * book3s is limited to 16 page sizes due to encoding this in
258 	 * a 4-bit field for slices.
259 	 */
260 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
261 
262 #ifdef CONFIG_SWIOTLB
263 	swiotlb_init(0);
264 #endif
265 
266 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
267 	set_max_mapnr(max_pfn);
268 	memblock_free_all();
269 
270 #ifdef CONFIG_HIGHMEM
271 	{
272 		unsigned long pfn, highmem_mapnr;
273 
274 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
275 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
276 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
277 			struct page *page = pfn_to_page(pfn);
278 			if (!memblock_is_reserved(paddr))
279 				free_highmem_page(page);
280 		}
281 	}
282 #endif /* CONFIG_HIGHMEM */
283 
284 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
285 	/*
286 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
287 	 * functions.... do it here for the non-smp case.
288 	 */
289 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
290 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
291 #endif
292 
293 	mem_init_print_info(NULL);
294 #ifdef CONFIG_PPC32
295 	pr_info("Kernel virtual memory layout:\n");
296 #ifdef CONFIG_KASAN
297 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
298 		KASAN_SHADOW_START, KASAN_SHADOW_END);
299 #endif
300 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
301 #ifdef CONFIG_HIGHMEM
302 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
303 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
304 #endif /* CONFIG_HIGHMEM */
305 #ifdef CONFIG_NOT_COHERENT_CACHE
306 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
307 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
308 #endif /* CONFIG_NOT_COHERENT_CACHE */
309 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
310 		ioremap_bot, IOREMAP_TOP);
311 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
312 		VMALLOC_START, VMALLOC_END);
313 #endif /* CONFIG_PPC32 */
314 }
315 
316 void free_initmem(void)
317 {
318 	ppc_md.progress = ppc_printk_progress;
319 	mark_initmem_nx();
320 	init_mem_is_free = true;
321 	free_initmem_default(POISON_FREE_INITMEM);
322 }
323 
324 /*
325  * This is called when a page has been modified by the kernel.
326  * It just marks the page as not i-cache clean.  We do the i-cache
327  * flush later when the page is given to a user process, if necessary.
328  */
329 void flush_dcache_page(struct page *page)
330 {
331 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
332 		return;
333 	/* avoid an atomic op if possible */
334 	if (test_bit(PG_arch_1, &page->flags))
335 		clear_bit(PG_arch_1, &page->flags);
336 }
337 EXPORT_SYMBOL(flush_dcache_page);
338 
339 void flush_dcache_icache_page(struct page *page)
340 {
341 #ifdef CONFIG_HUGETLB_PAGE
342 	if (PageCompound(page)) {
343 		flush_dcache_icache_hugepage(page);
344 		return;
345 	}
346 #endif
347 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
348 	/* On 8xx there is no need to kmap since highmem is not supported */
349 	__flush_dcache_icache(page_address(page));
350 #else
351 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
352 		void *start = kmap_atomic(page);
353 		__flush_dcache_icache(start);
354 		kunmap_atomic(start);
355 	} else {
356 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
357 	}
358 #endif
359 }
360 EXPORT_SYMBOL(flush_dcache_icache_page);
361 
362 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
363 {
364 	clear_page(page);
365 
366 	/*
367 	 * We shouldn't have to do this, but some versions of glibc
368 	 * require it (ld.so assumes zero filled pages are icache clean)
369 	 * - Anton
370 	 */
371 	flush_dcache_page(pg);
372 }
373 EXPORT_SYMBOL(clear_user_page);
374 
375 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
376 		    struct page *pg)
377 {
378 	copy_page(vto, vfrom);
379 
380 	/*
381 	 * We should be able to use the following optimisation, however
382 	 * there are two problems.
383 	 * Firstly a bug in some versions of binutils meant PLT sections
384 	 * were not marked executable.
385 	 * Secondly the first word in the GOT section is blrl, used
386 	 * to establish the GOT address. Until recently the GOT was
387 	 * not marked executable.
388 	 * - Anton
389 	 */
390 #if 0
391 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
392 		return;
393 #endif
394 
395 	flush_dcache_page(pg);
396 }
397 
398 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
399 			     unsigned long addr, int len)
400 {
401 	unsigned long maddr;
402 
403 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
404 	flush_icache_range(maddr, maddr + len);
405 	kunmap(page);
406 }
407 EXPORT_SYMBOL(flush_icache_user_range);
408 
409 /*
410  * This is called at the end of handling a user page fault, when the
411  * fault has been handled by updating a PTE in the linux page tables.
412  * We use it to preload an HPTE into the hash table corresponding to
413  * the updated linux PTE.
414  *
415  * This must always be called with the pte lock held.
416  */
417 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
418 		      pte_t *ptep)
419 {
420 #ifdef CONFIG_PPC_BOOK3S
421 	/*
422 	 * We don't need to worry about _PAGE_PRESENT here because we are
423 	 * called with either mm->page_table_lock held or ptl lock held
424 	 */
425 	unsigned long trap;
426 	bool is_exec;
427 
428 	if (radix_enabled()) {
429 		prefetch((void *)address);
430 		return;
431 	}
432 
433 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
434 	if (!pte_young(*ptep) || address >= TASK_SIZE)
435 		return;
436 
437 	/* We try to figure out if we are coming from an instruction
438 	 * access fault and pass that down to __hash_page so we avoid
439 	 * double-faulting on execution of fresh text. We have to test
440 	 * for regs NULL since init will get here first thing at boot
441 	 *
442 	 * We also avoid filling the hash if not coming from a fault
443 	 */
444 
445 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
446 	switch (trap) {
447 	case 0x300:
448 		is_exec = false;
449 		break;
450 	case 0x400:
451 		is_exec = true;
452 		break;
453 	default:
454 		return;
455 	}
456 
457 	hash_preload(vma->vm_mm, address, is_exec, trap);
458 #endif /* CONFIG_PPC_BOOK3S */
459 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
460 	&& defined(CONFIG_HUGETLB_PAGE)
461 	if (is_vm_hugetlb_page(vma))
462 		book3e_hugetlb_preload(vma, address, *ptep);
463 #endif
464 }
465 
466 /*
467  * System memory should not be in /proc/iomem but various tools expect it
468  * (eg kdump).
469  */
470 static int __init add_system_ram_resources(void)
471 {
472 	struct memblock_region *reg;
473 
474 	for_each_memblock(memory, reg) {
475 		struct resource *res;
476 		unsigned long base = reg->base;
477 		unsigned long size = reg->size;
478 
479 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
480 		WARN_ON(!res);
481 
482 		if (res) {
483 			res->name = "System RAM";
484 			res->start = base;
485 			res->end = base + size - 1;
486 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
487 			WARN_ON(request_resource(&iomem_resource, res) < 0);
488 		}
489 	}
490 
491 	return 0;
492 }
493 subsys_initcall(add_system_ram_resources);
494 
495 #ifdef CONFIG_STRICT_DEVMEM
496 /*
497  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
498  * is valid. The argument is a physical page number.
499  *
500  * Access has to be given to non-kernel-ram areas as well, these contain the
501  * PCI mmio resources as well as potential bios/acpi data regions.
502  */
503 int devmem_is_allowed(unsigned long pfn)
504 {
505 	if (page_is_rtas_user_buf(pfn))
506 		return 1;
507 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
508 		return 0;
509 	if (!page_is_ram(pfn))
510 		return 1;
511 	return 0;
512 }
513 #endif /* CONFIG_STRICT_DEVMEM */
514 
515 /*
516  * This is defined in kernel/resource.c but only powerpc needs to export it, for
517  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
518  */
519 EXPORT_SYMBOL_GPL(walk_system_ram_range);
520