xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision 7f6964c5)
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_inval_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;
135 	int ret;
136 
137 	/*
138 	 * If we have an altmap then we need to skip over any reserved PFNs
139 	 * when querying the zone.
140 	 */
141 	page = pfn_to_page(start_pfn);
142 	if (altmap)
143 		page += vmem_altmap_offset(altmap);
144 
145 	__remove_pages(page_zone(page), start_pfn, nr_pages, altmap);
146 
147 	/* Remove htab bolted mappings for this section of memory */
148 	start = (unsigned long)__va(start);
149 	flush_inval_dcache_range(start, start + size);
150 	ret = remove_section_mapping(start, start + size);
151 	WARN_ON_ONCE(ret);
152 
153 	/* Ensure all vmalloc mappings are flushed in case they also
154 	 * hit that section of memory
155 	 */
156 	vm_unmap_aliases();
157 
158 	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
159 		pr_warn("Hash collision while resizing HPT\n");
160 }
161 #endif
162 #endif /* CONFIG_MEMORY_HOTPLUG */
163 
164 #ifndef CONFIG_NEED_MULTIPLE_NODES
165 void __init mem_topology_setup(void)
166 {
167 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
168 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
169 #ifdef CONFIG_HIGHMEM
170 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
171 #endif
172 
173 	/* Place all memblock_regions in the same node and merge contiguous
174 	 * memblock_regions
175 	 */
176 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
177 }
178 
179 void __init initmem_init(void)
180 {
181 	/* XXX need to clip this if using highmem? */
182 	sparse_memory_present_with_active_regions(0);
183 	sparse_init();
184 }
185 
186 /* mark pages that don't exist as nosave */
187 static int __init mark_nonram_nosave(void)
188 {
189 	struct memblock_region *reg, *prev = NULL;
190 
191 	for_each_memblock(memory, reg) {
192 		if (prev &&
193 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
194 			register_nosave_region(memblock_region_memory_end_pfn(prev),
195 					       memblock_region_memory_base_pfn(reg));
196 		prev = reg;
197 	}
198 	return 0;
199 }
200 #else /* CONFIG_NEED_MULTIPLE_NODES */
201 static int __init mark_nonram_nosave(void)
202 {
203 	return 0;
204 }
205 #endif
206 
207 /*
208  * Zones usage:
209  *
210  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
211  * everything else. GFP_DMA32 page allocations automatically fall back to
212  * ZONE_DMA.
213  *
214  * By using 31-bit unconditionally, we can exploit ARCH_ZONE_DMA_BITS to
215  * inform the generic DMA mapping code.  32-bit only devices (if not handled
216  * by an IOMMU anyway) will take a first dip into ZONE_NORMAL and get
217  * otherwise served by ZONE_DMA.
218  */
219 static unsigned long max_zone_pfns[MAX_NR_ZONES];
220 
221 /*
222  * paging_init() sets up the page tables - in fact we've already done this.
223  */
224 void __init paging_init(void)
225 {
226 	unsigned long long total_ram = memblock_phys_mem_size();
227 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
228 
229 #ifdef CONFIG_PPC32
230 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
231 	unsigned long end = __fix_to_virt(FIX_HOLE);
232 
233 	for (; v < end; v += PAGE_SIZE)
234 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
235 #endif
236 
237 #ifdef CONFIG_HIGHMEM
238 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
239 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
240 
241 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
242 	kmap_prot = PAGE_KERNEL;
243 #endif /* CONFIG_HIGHMEM */
244 
245 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
246 	       (unsigned long long)top_of_ram, total_ram);
247 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
248 	       (long int)((top_of_ram - total_ram) >> 20));
249 
250 #ifdef CONFIG_ZONE_DMA
251 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn, 0x7fffffffUL >> PAGE_SHIFT);
252 #endif
253 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
254 #ifdef CONFIG_HIGHMEM
255 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
256 #endif
257 
258 	free_area_init_nodes(max_zone_pfns);
259 
260 	mark_nonram_nosave();
261 }
262 
263 void __init mem_init(void)
264 {
265 	/*
266 	 * book3s is limited to 16 page sizes due to encoding this in
267 	 * a 4-bit field for slices.
268 	 */
269 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
270 
271 #ifdef CONFIG_SWIOTLB
272 	swiotlb_init(0);
273 #endif
274 
275 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
276 	set_max_mapnr(max_pfn);
277 	memblock_free_all();
278 
279 #ifdef CONFIG_HIGHMEM
280 	{
281 		unsigned long pfn, highmem_mapnr;
282 
283 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
284 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
285 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
286 			struct page *page = pfn_to_page(pfn);
287 			if (!memblock_is_reserved(paddr))
288 				free_highmem_page(page);
289 		}
290 	}
291 #endif /* CONFIG_HIGHMEM */
292 
293 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
294 	/*
295 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
296 	 * functions.... do it here for the non-smp case.
297 	 */
298 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
299 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
300 #endif
301 
302 	mem_init_print_info(NULL);
303 #ifdef CONFIG_PPC32
304 	pr_info("Kernel virtual memory layout:\n");
305 #ifdef CONFIG_KASAN
306 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
307 		KASAN_SHADOW_START, KASAN_SHADOW_END);
308 #endif
309 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
310 #ifdef CONFIG_HIGHMEM
311 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
312 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
313 #endif /* CONFIG_HIGHMEM */
314 #ifdef CONFIG_NOT_COHERENT_CACHE
315 	pr_info("  * 0x%08lx..0x%08lx  : consistent mem\n",
316 		IOREMAP_TOP, IOREMAP_TOP + CONFIG_CONSISTENT_SIZE);
317 #endif /* CONFIG_NOT_COHERENT_CACHE */
318 	pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
319 		ioremap_bot, IOREMAP_TOP);
320 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
321 		VMALLOC_START, VMALLOC_END);
322 #endif /* CONFIG_PPC32 */
323 }
324 
325 void free_initmem(void)
326 {
327 	ppc_md.progress = ppc_printk_progress;
328 	mark_initmem_nx();
329 	init_mem_is_free = true;
330 	free_initmem_default(POISON_FREE_INITMEM);
331 }
332 
333 /*
334  * This is called when a page has been modified by the kernel.
335  * It just marks the page as not i-cache clean.  We do the i-cache
336  * flush later when the page is given to a user process, if necessary.
337  */
338 void flush_dcache_page(struct page *page)
339 {
340 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
341 		return;
342 	/* avoid an atomic op if possible */
343 	if (test_bit(PG_arch_1, &page->flags))
344 		clear_bit(PG_arch_1, &page->flags);
345 }
346 EXPORT_SYMBOL(flush_dcache_page);
347 
348 void flush_dcache_icache_page(struct page *page)
349 {
350 #ifdef CONFIG_HUGETLB_PAGE
351 	if (PageCompound(page)) {
352 		flush_dcache_icache_hugepage(page);
353 		return;
354 	}
355 #endif
356 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
357 	/* On 8xx there is no need to kmap since highmem is not supported */
358 	__flush_dcache_icache(page_address(page));
359 #else
360 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
361 		void *start = kmap_atomic(page);
362 		__flush_dcache_icache(start);
363 		kunmap_atomic(start);
364 	} else {
365 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
366 	}
367 #endif
368 }
369 EXPORT_SYMBOL(flush_dcache_icache_page);
370 
371 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
372 {
373 	clear_page(page);
374 
375 	/*
376 	 * We shouldn't have to do this, but some versions of glibc
377 	 * require it (ld.so assumes zero filled pages are icache clean)
378 	 * - Anton
379 	 */
380 	flush_dcache_page(pg);
381 }
382 EXPORT_SYMBOL(clear_user_page);
383 
384 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
385 		    struct page *pg)
386 {
387 	copy_page(vto, vfrom);
388 
389 	/*
390 	 * We should be able to use the following optimisation, however
391 	 * there are two problems.
392 	 * Firstly a bug in some versions of binutils meant PLT sections
393 	 * were not marked executable.
394 	 * Secondly the first word in the GOT section is blrl, used
395 	 * to establish the GOT address. Until recently the GOT was
396 	 * not marked executable.
397 	 * - Anton
398 	 */
399 #if 0
400 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
401 		return;
402 #endif
403 
404 	flush_dcache_page(pg);
405 }
406 
407 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
408 			     unsigned long addr, int len)
409 {
410 	unsigned long maddr;
411 
412 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
413 	flush_icache_range(maddr, maddr + len);
414 	kunmap(page);
415 }
416 EXPORT_SYMBOL(flush_icache_user_range);
417 
418 /*
419  * This is called at the end of handling a user page fault, when the
420  * fault has been handled by updating a PTE in the linux page tables.
421  * We use it to preload an HPTE into the hash table corresponding to
422  * the updated linux PTE.
423  *
424  * This must always be called with the pte lock held.
425  */
426 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
427 		      pte_t *ptep)
428 {
429 #ifdef CONFIG_PPC_BOOK3S
430 	/*
431 	 * We don't need to worry about _PAGE_PRESENT here because we are
432 	 * called with either mm->page_table_lock held or ptl lock held
433 	 */
434 	unsigned long trap;
435 	bool is_exec;
436 
437 	if (radix_enabled()) {
438 		prefetch((void *)address);
439 		return;
440 	}
441 
442 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
443 	if (!pte_young(*ptep) || address >= TASK_SIZE)
444 		return;
445 
446 	/* We try to figure out if we are coming from an instruction
447 	 * access fault and pass that down to __hash_page so we avoid
448 	 * double-faulting on execution of fresh text. We have to test
449 	 * for regs NULL since init will get here first thing at boot
450 	 *
451 	 * We also avoid filling the hash if not coming from a fault
452 	 */
453 
454 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
455 	switch (trap) {
456 	case 0x300:
457 		is_exec = false;
458 		break;
459 	case 0x400:
460 		is_exec = true;
461 		break;
462 	default:
463 		return;
464 	}
465 
466 	hash_preload(vma->vm_mm, address, is_exec, trap);
467 #endif /* CONFIG_PPC_BOOK3S */
468 #if (defined(CONFIG_PPC_BOOK3E_64) || defined(CONFIG_PPC_FSL_BOOK3E)) \
469 	&& defined(CONFIG_HUGETLB_PAGE)
470 	if (is_vm_hugetlb_page(vma))
471 		book3e_hugetlb_preload(vma, address, *ptep);
472 #endif
473 }
474 
475 /*
476  * System memory should not be in /proc/iomem but various tools expect it
477  * (eg kdump).
478  */
479 static int __init add_system_ram_resources(void)
480 {
481 	struct memblock_region *reg;
482 
483 	for_each_memblock(memory, reg) {
484 		struct resource *res;
485 		unsigned long base = reg->base;
486 		unsigned long size = reg->size;
487 
488 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
489 		WARN_ON(!res);
490 
491 		if (res) {
492 			res->name = "System RAM";
493 			res->start = base;
494 			res->end = base + size - 1;
495 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
496 			WARN_ON(request_resource(&iomem_resource, res) < 0);
497 		}
498 	}
499 
500 	return 0;
501 }
502 subsys_initcall(add_system_ram_resources);
503 
504 #ifdef CONFIG_STRICT_DEVMEM
505 /*
506  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
507  * is valid. The argument is a physical page number.
508  *
509  * Access has to be given to non-kernel-ram areas as well, these contain the
510  * PCI mmio resources as well as potential bios/acpi data regions.
511  */
512 int devmem_is_allowed(unsigned long pfn)
513 {
514 	if (page_is_rtas_user_buf(pfn))
515 		return 1;
516 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
517 		return 0;
518 	if (!page_is_ram(pfn))
519 		return 1;
520 	return 0;
521 }
522 #endif /* CONFIG_STRICT_DEVMEM */
523 
524 /*
525  * This is defined in kernel/resource.c but only powerpc needs to export it, for
526  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
527  */
528 EXPORT_SYMBOL_GPL(walk_system_ram_range);
529