xref: /openbmc/linux/arch/powerpc/mm/mem.c (revision c24c57a4)
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 #include <linux/dma-direct.h>
35 
36 #include <asm/pgalloc.h>
37 #include <asm/prom.h>
38 #include <asm/io.h>
39 #include <asm/mmu_context.h>
40 #include <asm/pgtable.h>
41 #include <asm/mmu.h>
42 #include <asm/smp.h>
43 #include <asm/machdep.h>
44 #include <asm/btext.h>
45 #include <asm/tlb.h>
46 #include <asm/sections.h>
47 #include <asm/sparsemem.h>
48 #include <asm/vdso.h>
49 #include <asm/fixmap.h>
50 #include <asm/swiotlb.h>
51 #include <asm/rtas.h>
52 
53 #include <mm/mmu_decl.h>
54 
55 #ifndef CPU_FTR_COHERENT_ICACHE
56 #define CPU_FTR_COHERENT_ICACHE	0	/* XXX for now */
57 #define CPU_FTR_NOEXECUTE	0
58 #endif
59 
60 unsigned long long memory_limit;
61 bool init_mem_is_free;
62 
63 #ifdef CONFIG_HIGHMEM
64 pte_t *kmap_pte;
65 EXPORT_SYMBOL(kmap_pte);
66 pgprot_t kmap_prot;
67 EXPORT_SYMBOL(kmap_prot);
68 
69 static inline pte_t *virt_to_kpte(unsigned long vaddr)
70 {
71 	return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
72 			vaddr), vaddr), vaddr);
73 }
74 #endif
75 
76 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
77 			      unsigned long size, pgprot_t vma_prot)
78 {
79 	if (ppc_md.phys_mem_access_prot)
80 		return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
81 
82 	if (!page_is_ram(pfn))
83 		vma_prot = pgprot_noncached(vma_prot);
84 
85 	return vma_prot;
86 }
87 EXPORT_SYMBOL(phys_mem_access_prot);
88 
89 #ifdef CONFIG_MEMORY_HOTPLUG
90 
91 #ifdef CONFIG_NUMA
92 int memory_add_physaddr_to_nid(u64 start)
93 {
94 	return hot_add_scn_to_nid(start);
95 }
96 #endif
97 
98 int __weak create_section_mapping(unsigned long start, unsigned long end, int nid)
99 {
100 	return -ENODEV;
101 }
102 
103 int __weak remove_section_mapping(unsigned long start, unsigned long end)
104 {
105 	return -ENODEV;
106 }
107 
108 int __ref arch_add_memory(int nid, u64 start, u64 size,
109 			struct mhp_restrictions *restrictions)
110 {
111 	unsigned long start_pfn = start >> PAGE_SHIFT;
112 	unsigned long nr_pages = size >> PAGE_SHIFT;
113 	int rc;
114 
115 	resize_hpt_for_hotplug(memblock_phys_mem_size());
116 
117 	start = (unsigned long)__va(start);
118 	rc = create_section_mapping(start, start + size, nid);
119 	if (rc) {
120 		pr_warn("Unable to create mapping for hot added memory 0x%llx..0x%llx: %d\n",
121 			start, start + size, rc);
122 		return -EFAULT;
123 	}
124 	flush_dcache_range(start, start + size);
125 
126 	return __add_pages(nid, start_pfn, nr_pages, restrictions);
127 }
128 
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 
155 #ifndef CONFIG_NEED_MULTIPLE_NODES
156 void __init mem_topology_setup(void)
157 {
158 	max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT;
159 	min_low_pfn = MEMORY_START >> PAGE_SHIFT;
160 #ifdef CONFIG_HIGHMEM
161 	max_low_pfn = lowmem_end_addr >> PAGE_SHIFT;
162 #endif
163 
164 	/* Place all memblock_regions in the same node and merge contiguous
165 	 * memblock_regions
166 	 */
167 	memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0);
168 }
169 
170 void __init initmem_init(void)
171 {
172 	/* XXX need to clip this if using highmem? */
173 	sparse_memory_present_with_active_regions(0);
174 	sparse_init();
175 }
176 
177 /* mark pages that don't exist as nosave */
178 static int __init mark_nonram_nosave(void)
179 {
180 	struct memblock_region *reg, *prev = NULL;
181 
182 	for_each_memblock(memory, reg) {
183 		if (prev &&
184 		    memblock_region_memory_end_pfn(prev) < memblock_region_memory_base_pfn(reg))
185 			register_nosave_region(memblock_region_memory_end_pfn(prev),
186 					       memblock_region_memory_base_pfn(reg));
187 		prev = reg;
188 	}
189 	return 0;
190 }
191 #else /* CONFIG_NEED_MULTIPLE_NODES */
192 static int __init mark_nonram_nosave(void)
193 {
194 	return 0;
195 }
196 #endif
197 
198 /*
199  * Zones usage:
200  *
201  * We setup ZONE_DMA to be 31-bits on all platforms and ZONE_NORMAL to be
202  * everything else. GFP_DMA32 page allocations automatically fall back to
203  * ZONE_DMA.
204  *
205  * By using 31-bit unconditionally, we can exploit zone_dma_bits to inform the
206  * generic DMA mapping code.  32-bit only devices (if not handled by an IOMMU
207  * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by
208  * ZONE_DMA.
209  */
210 static unsigned long max_zone_pfns[MAX_NR_ZONES];
211 
212 /*
213  * paging_init() sets up the page tables - in fact we've already done this.
214  */
215 void __init paging_init(void)
216 {
217 	unsigned long long total_ram = memblock_phys_mem_size();
218 	phys_addr_t top_of_ram = memblock_end_of_DRAM();
219 
220 #ifdef CONFIG_PPC32
221 	unsigned long v = __fix_to_virt(__end_of_fixed_addresses - 1);
222 	unsigned long end = __fix_to_virt(FIX_HOLE);
223 
224 	for (; v < end; v += PAGE_SIZE)
225 		map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */
226 #endif
227 
228 #ifdef CONFIG_HIGHMEM
229 	map_kernel_page(PKMAP_BASE, 0, __pgprot(0));	/* XXX gross */
230 	pkmap_page_table = virt_to_kpte(PKMAP_BASE);
231 
232 	kmap_pte = virt_to_kpte(__fix_to_virt(FIX_KMAP_BEGIN));
233 	kmap_prot = PAGE_KERNEL;
234 #endif /* CONFIG_HIGHMEM */
235 
236 	printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n",
237 	       (unsigned long long)top_of_ram, total_ram);
238 	printk(KERN_DEBUG "Memory hole size: %ldMB\n",
239 	       (long int)((top_of_ram - total_ram) >> 20));
240 
241 	/*
242 	 * Allow 30-bit DMA for very limited Broadcom wifi chips on many
243 	 * powerbooks.
244 	 */
245 	if (IS_ENABLED(CONFIG_PPC32))
246 		zone_dma_bits = 30;
247 	else
248 		zone_dma_bits = 31;
249 
250 #ifdef CONFIG_ZONE_DMA
251 	max_zone_pfns[ZONE_DMA]	= min(max_low_pfn,
252 				      1UL << (zone_dma_bits - PAGE_SHIFT));
253 #endif
254 	max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
255 #ifdef CONFIG_HIGHMEM
256 	max_zone_pfns[ZONE_HIGHMEM] = max_pfn;
257 #endif
258 
259 	free_area_init_nodes(max_zone_pfns);
260 
261 	mark_nonram_nosave();
262 }
263 
264 void __init mem_init(void)
265 {
266 	/*
267 	 * book3s is limited to 16 page sizes due to encoding this in
268 	 * a 4-bit field for slices.
269 	 */
270 	BUILD_BUG_ON(MMU_PAGE_COUNT > 16);
271 
272 #ifdef CONFIG_SWIOTLB
273 	swiotlb_init(0);
274 #endif
275 
276 	high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
277 	set_max_mapnr(max_pfn);
278 	memblock_free_all();
279 
280 #ifdef CONFIG_HIGHMEM
281 	{
282 		unsigned long pfn, highmem_mapnr;
283 
284 		highmem_mapnr = lowmem_end_addr >> PAGE_SHIFT;
285 		for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
286 			phys_addr_t paddr = (phys_addr_t)pfn << PAGE_SHIFT;
287 			struct page *page = pfn_to_page(pfn);
288 			if (!memblock_is_reserved(paddr))
289 				free_highmem_page(page);
290 		}
291 	}
292 #endif /* CONFIG_HIGHMEM */
293 
294 #if defined(CONFIG_PPC_FSL_BOOK3E) && !defined(CONFIG_SMP)
295 	/*
296 	 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up
297 	 * functions.... do it here for the non-smp case.
298 	 */
299 	per_cpu(next_tlbcam_idx, smp_processor_id()) =
300 		(mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1;
301 #endif
302 
303 	mem_init_print_info(NULL);
304 #ifdef CONFIG_PPC32
305 	pr_info("Kernel virtual memory layout:\n");
306 #ifdef CONFIG_KASAN
307 	pr_info("  * 0x%08lx..0x%08lx  : kasan shadow mem\n",
308 		KASAN_SHADOW_START, KASAN_SHADOW_END);
309 #endif
310 	pr_info("  * 0x%08lx..0x%08lx  : fixmap\n", FIXADDR_START, FIXADDR_TOP);
311 #ifdef CONFIG_HIGHMEM
312 	pr_info("  * 0x%08lx..0x%08lx  : highmem PTEs\n",
313 		PKMAP_BASE, PKMAP_ADDR(LAST_PKMAP));
314 #endif /* CONFIG_HIGHMEM */
315 	if (ioremap_bot != IOREMAP_TOP)
316 		pr_info("  * 0x%08lx..0x%08lx  : early ioremap\n",
317 			ioremap_bot, IOREMAP_TOP);
318 	pr_info("  * 0x%08lx..0x%08lx  : vmalloc & ioremap\n",
319 		VMALLOC_START, VMALLOC_END);
320 #endif /* CONFIG_PPC32 */
321 }
322 
323 void free_initmem(void)
324 {
325 	ppc_md.progress = ppc_printk_progress;
326 	mark_initmem_nx();
327 	init_mem_is_free = true;
328 	free_initmem_default(POISON_FREE_INITMEM);
329 }
330 
331 /*
332  * This is called when a page has been modified by the kernel.
333  * It just marks the page as not i-cache clean.  We do the i-cache
334  * flush later when the page is given to a user process, if necessary.
335  */
336 void flush_dcache_page(struct page *page)
337 {
338 	if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
339 		return;
340 	/* avoid an atomic op if possible */
341 	if (test_bit(PG_arch_1, &page->flags))
342 		clear_bit(PG_arch_1, &page->flags);
343 }
344 EXPORT_SYMBOL(flush_dcache_page);
345 
346 void flush_dcache_icache_page(struct page *page)
347 {
348 #ifdef CONFIG_HUGETLB_PAGE
349 	if (PageCompound(page)) {
350 		flush_dcache_icache_hugepage(page);
351 		return;
352 	}
353 #endif
354 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC64)
355 	/* On 8xx there is no need to kmap since highmem is not supported */
356 	__flush_dcache_icache(page_address(page));
357 #else
358 	if (IS_ENABLED(CONFIG_BOOKE) || sizeof(phys_addr_t) > sizeof(void *)) {
359 		void *start = kmap_atomic(page);
360 		__flush_dcache_icache(start);
361 		kunmap_atomic(start);
362 	} else {
363 		__flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
364 	}
365 #endif
366 }
367 EXPORT_SYMBOL(flush_dcache_icache_page);
368 
369 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
370 {
371 	clear_page(page);
372 
373 	/*
374 	 * We shouldn't have to do this, but some versions of glibc
375 	 * require it (ld.so assumes zero filled pages are icache clean)
376 	 * - Anton
377 	 */
378 	flush_dcache_page(pg);
379 }
380 EXPORT_SYMBOL(clear_user_page);
381 
382 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
383 		    struct page *pg)
384 {
385 	copy_page(vto, vfrom);
386 
387 	/*
388 	 * We should be able to use the following optimisation, however
389 	 * there are two problems.
390 	 * Firstly a bug in some versions of binutils meant PLT sections
391 	 * were not marked executable.
392 	 * Secondly the first word in the GOT section is blrl, used
393 	 * to establish the GOT address. Until recently the GOT was
394 	 * not marked executable.
395 	 * - Anton
396 	 */
397 #if 0
398 	if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
399 		return;
400 #endif
401 
402 	flush_dcache_page(pg);
403 }
404 
405 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
406 			     unsigned long addr, int len)
407 {
408 	unsigned long maddr;
409 
410 	maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
411 	flush_icache_range(maddr, maddr + len);
412 	kunmap(page);
413 }
414 EXPORT_SYMBOL(flush_icache_user_range);
415 
416 /*
417  * System memory should not be in /proc/iomem but various tools expect it
418  * (eg kdump).
419  */
420 static int __init add_system_ram_resources(void)
421 {
422 	struct memblock_region *reg;
423 
424 	for_each_memblock(memory, reg) {
425 		struct resource *res;
426 		unsigned long base = reg->base;
427 		unsigned long size = reg->size;
428 
429 		res = kzalloc(sizeof(struct resource), GFP_KERNEL);
430 		WARN_ON(!res);
431 
432 		if (res) {
433 			res->name = "System RAM";
434 			res->start = base;
435 			res->end = base + size - 1;
436 			res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY;
437 			WARN_ON(request_resource(&iomem_resource, res) < 0);
438 		}
439 	}
440 
441 	return 0;
442 }
443 subsys_initcall(add_system_ram_resources);
444 
445 #ifdef CONFIG_STRICT_DEVMEM
446 /*
447  * devmem_is_allowed(): check to see if /dev/mem access to a certain address
448  * is valid. The argument is a physical page number.
449  *
450  * Access has to be given to non-kernel-ram areas as well, these contain the
451  * PCI mmio resources as well as potential bios/acpi data regions.
452  */
453 int devmem_is_allowed(unsigned long pfn)
454 {
455 	if (page_is_rtas_user_buf(pfn))
456 		return 1;
457 	if (iomem_is_exclusive(PFN_PHYS(pfn)))
458 		return 0;
459 	if (!page_is_ram(pfn))
460 		return 1;
461 	return 0;
462 }
463 #endif /* CONFIG_STRICT_DEVMEM */
464 
465 /*
466  * This is defined in kernel/resource.c but only powerpc needs to export it, for
467  * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed.
468  */
469 EXPORT_SYMBOL_GPL(walk_system_ram_range);
470