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 * 10 * Derived from "arch/i386/mm/init.c" 11 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds 12 * 13 * Dave Engebretsen <engebret@us.ibm.com> 14 * Rework for PPC64 port. 15 */ 16 17 #undef DEBUG 18 19 #include <linux/signal.h> 20 #include <linux/sched.h> 21 #include <linux/kernel.h> 22 #include <linux/errno.h> 23 #include <linux/string.h> 24 #include <linux/types.h> 25 #include <linux/mman.h> 26 #include <linux/mm.h> 27 #include <linux/swap.h> 28 #include <linux/stddef.h> 29 #include <linux/vmalloc.h> 30 #include <linux/init.h> 31 #include <linux/delay.h> 32 #include <linux/highmem.h> 33 #include <linux/idr.h> 34 #include <linux/nodemask.h> 35 #include <linux/module.h> 36 #include <linux/poison.h> 37 #include <linux/memblock.h> 38 #include <linux/hugetlb.h> 39 #include <linux/slab.h> 40 #include <linux/of_fdt.h> 41 #include <linux/libfdt.h> 42 #include <linux/memremap.h> 43 44 #include <asm/pgalloc.h> 45 #include <asm/page.h> 46 #include <asm/prom.h> 47 #include <asm/rtas.h> 48 #include <asm/io.h> 49 #include <asm/mmu_context.h> 50 #include <asm/pgtable.h> 51 #include <asm/mmu.h> 52 #include <linux/uaccess.h> 53 #include <asm/smp.h> 54 #include <asm/machdep.h> 55 #include <asm/tlb.h> 56 #include <asm/eeh.h> 57 #include <asm/processor.h> 58 #include <asm/mmzone.h> 59 #include <asm/cputable.h> 60 #include <asm/sections.h> 61 #include <asm/iommu.h> 62 #include <asm/vdso.h> 63 64 #include <mm/mmu_decl.h> 65 66 #ifdef CONFIG_SPARSEMEM_VMEMMAP 67 /* 68 * Given an address within the vmemmap, determine the page that 69 * represents the start of the subsection it is within. Note that we have to 70 * do this by hand as the proffered address may not be correctly aligned. 71 * Subtraction of non-aligned pointers produces undefined results. 72 */ 73 static struct page * __meminit vmemmap_subsection_start(unsigned long vmemmap_addr) 74 { 75 unsigned long start_pfn; 76 unsigned long offset = vmemmap_addr - ((unsigned long)(vmemmap)); 77 78 /* Return the pfn of the start of the section. */ 79 start_pfn = (offset / sizeof(struct page)) & PAGE_SUBSECTION_MASK; 80 return pfn_to_page(start_pfn); 81 } 82 83 /* 84 * Since memory is added in sub-section chunks, before creating a new vmemmap 85 * mapping, the kernel should check whether there is an existing memmap mapping 86 * covering the new subsection added. This is needed because kernel can map 87 * vmemmap area using 16MB pages which will cover a memory range of 16G. Such 88 * a range covers multiple subsections (2M) 89 * 90 * If any subsection in the 16G range mapped by vmemmap is valid we consider the 91 * vmemmap populated (There is a page table entry already present). We can't do 92 * a page table lookup here because with the hash translation we don't keep 93 * vmemmap details in linux page table. 94 */ 95 static int __meminit vmemmap_populated(unsigned long vmemmap_addr, int vmemmap_map_size) 96 { 97 struct page *start; 98 unsigned long vmemmap_end = vmemmap_addr + vmemmap_map_size; 99 start = vmemmap_subsection_start(vmemmap_addr); 100 101 for (; (unsigned long)start < vmemmap_end; start += PAGES_PER_SUBSECTION) 102 /* 103 * pfn valid check here is intended to really check 104 * whether we have any subsection already initialized 105 * in this range. 106 */ 107 if (pfn_valid(page_to_pfn(start))) 108 return 1; 109 110 return 0; 111 } 112 113 /* 114 * vmemmap virtual address space management does not have a traditonal page 115 * table to track which virtual struct pages are backed by physical mapping. 116 * The virtual to physical mappings are tracked in a simple linked list 117 * format. 'vmemmap_list' maintains the entire vmemmap physical mapping at 118 * all times where as the 'next' list maintains the available 119 * vmemmap_backing structures which have been deleted from the 120 * 'vmemmap_global' list during system runtime (memory hotplug remove 121 * operation). The freed 'vmemmap_backing' structures are reused later when 122 * new requests come in without allocating fresh memory. This pointer also 123 * tracks the allocated 'vmemmap_backing' structures as we allocate one 124 * full page memory at a time when we dont have any. 125 */ 126 struct vmemmap_backing *vmemmap_list; 127 static struct vmemmap_backing *next; 128 129 /* 130 * The same pointer 'next' tracks individual chunks inside the allocated 131 * full page during the boot time and again tracks the freeed nodes during 132 * runtime. It is racy but it does not happen as they are separated by the 133 * boot process. Will create problem if some how we have memory hotplug 134 * operation during boot !! 135 */ 136 static int num_left; 137 static int num_freed; 138 139 static __meminit struct vmemmap_backing * vmemmap_list_alloc(int node) 140 { 141 struct vmemmap_backing *vmem_back; 142 /* get from freed entries first */ 143 if (num_freed) { 144 num_freed--; 145 vmem_back = next; 146 next = next->list; 147 148 return vmem_back; 149 } 150 151 /* allocate a page when required and hand out chunks */ 152 if (!num_left) { 153 next = vmemmap_alloc_block(PAGE_SIZE, node); 154 if (unlikely(!next)) { 155 WARN_ON(1); 156 return NULL; 157 } 158 num_left = PAGE_SIZE / sizeof(struct vmemmap_backing); 159 } 160 161 num_left--; 162 163 return next++; 164 } 165 166 static __meminit void vmemmap_list_populate(unsigned long phys, 167 unsigned long start, 168 int node) 169 { 170 struct vmemmap_backing *vmem_back; 171 172 vmem_back = vmemmap_list_alloc(node); 173 if (unlikely(!vmem_back)) { 174 WARN_ON(1); 175 return; 176 } 177 178 vmem_back->phys = phys; 179 vmem_back->virt_addr = start; 180 vmem_back->list = vmemmap_list; 181 182 vmemmap_list = vmem_back; 183 } 184 185 static bool altmap_cross_boundary(struct vmem_altmap *altmap, unsigned long start, 186 unsigned long page_size) 187 { 188 unsigned long nr_pfn = page_size / sizeof(struct page); 189 unsigned long start_pfn = page_to_pfn((struct page *)start); 190 191 if ((start_pfn + nr_pfn) > altmap->end_pfn) 192 return true; 193 194 if (start_pfn < altmap->base_pfn) 195 return true; 196 197 return false; 198 } 199 200 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, 201 struct vmem_altmap *altmap) 202 { 203 unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift; 204 205 /* Align to the page size of the linear mapping. */ 206 start = _ALIGN_DOWN(start, page_size); 207 208 pr_debug("vmemmap_populate %lx..%lx, node %d\n", start, end, node); 209 210 for (; start < end; start += page_size) { 211 void *p = NULL; 212 int rc; 213 214 /* 215 * This vmemmap range is backing different subsections. If any 216 * of that subsection is marked valid, that means we already 217 * have initialized a page table covering this range and hence 218 * the vmemmap range is populated. 219 */ 220 if (vmemmap_populated(start, page_size)) 221 continue; 222 223 /* 224 * Allocate from the altmap first if we have one. This may 225 * fail due to alignment issues when using 16MB hugepages, so 226 * fall back to system memory if the altmap allocation fail. 227 */ 228 if (altmap && !altmap_cross_boundary(altmap, start, page_size)) { 229 p = altmap_alloc_block_buf(page_size, altmap); 230 if (!p) 231 pr_debug("altmap block allocation failed, falling back to system memory"); 232 } 233 if (!p) 234 p = vmemmap_alloc_block_buf(page_size, node); 235 if (!p) 236 return -ENOMEM; 237 238 vmemmap_list_populate(__pa(p), start, node); 239 240 pr_debug(" * %016lx..%016lx allocated at %p\n", 241 start, start + page_size, p); 242 243 rc = vmemmap_create_mapping(start, page_size, __pa(p)); 244 if (rc < 0) { 245 pr_warn("%s: Unable to create vmemmap mapping: %d\n", 246 __func__, rc); 247 return -EFAULT; 248 } 249 } 250 251 return 0; 252 } 253 254 #ifdef CONFIG_MEMORY_HOTPLUG 255 static unsigned long vmemmap_list_free(unsigned long start) 256 { 257 struct vmemmap_backing *vmem_back, *vmem_back_prev; 258 259 vmem_back_prev = vmem_back = vmemmap_list; 260 261 /* look for it with prev pointer recorded */ 262 for (; vmem_back; vmem_back = vmem_back->list) { 263 if (vmem_back->virt_addr == start) 264 break; 265 vmem_back_prev = vmem_back; 266 } 267 268 if (unlikely(!vmem_back)) { 269 WARN_ON(1); 270 return 0; 271 } 272 273 /* remove it from vmemmap_list */ 274 if (vmem_back == vmemmap_list) /* remove head */ 275 vmemmap_list = vmem_back->list; 276 else 277 vmem_back_prev->list = vmem_back->list; 278 279 /* next point to this freed entry */ 280 vmem_back->list = next; 281 next = vmem_back; 282 num_freed++; 283 284 return vmem_back->phys; 285 } 286 287 void __ref vmemmap_free(unsigned long start, unsigned long end, 288 struct vmem_altmap *altmap) 289 { 290 unsigned long page_size = 1 << mmu_psize_defs[mmu_vmemmap_psize].shift; 291 unsigned long page_order = get_order(page_size); 292 unsigned long alt_start = ~0, alt_end = ~0; 293 unsigned long base_pfn; 294 295 start = _ALIGN_DOWN(start, page_size); 296 if (altmap) { 297 alt_start = altmap->base_pfn; 298 alt_end = altmap->base_pfn + altmap->reserve + 299 altmap->free + altmap->alloc + altmap->align; 300 } 301 302 pr_debug("vmemmap_free %lx...%lx\n", start, end); 303 304 for (; start < end; start += page_size) { 305 unsigned long nr_pages, addr; 306 struct page *page; 307 308 /* 309 * We have already marked the subsection we are trying to remove 310 * invalid. So if we want to remove the vmemmap range, we 311 * need to make sure there is no subsection marked valid 312 * in this range. 313 */ 314 if (vmemmap_populated(start, page_size)) 315 continue; 316 317 addr = vmemmap_list_free(start); 318 if (!addr) 319 continue; 320 321 page = pfn_to_page(addr >> PAGE_SHIFT); 322 nr_pages = 1 << page_order; 323 base_pfn = PHYS_PFN(addr); 324 325 if (base_pfn >= alt_start && base_pfn < alt_end) { 326 vmem_altmap_free(altmap, nr_pages); 327 } else if (PageReserved(page)) { 328 /* allocated from bootmem */ 329 if (page_size < PAGE_SIZE) { 330 /* 331 * this shouldn't happen, but if it is 332 * the case, leave the memory there 333 */ 334 WARN_ON_ONCE(1); 335 } else { 336 while (nr_pages--) 337 free_reserved_page(page++); 338 } 339 } else { 340 free_pages((unsigned long)(__va(addr)), page_order); 341 } 342 343 vmemmap_remove_mapping(start, page_size); 344 } 345 } 346 #endif 347 void register_page_bootmem_memmap(unsigned long section_nr, 348 struct page *start_page, unsigned long size) 349 { 350 } 351 352 #endif /* CONFIG_SPARSEMEM_VMEMMAP */ 353 354 #ifdef CONFIG_PPC_BOOK3S_64 355 static bool disable_radix = !IS_ENABLED(CONFIG_PPC_RADIX_MMU_DEFAULT); 356 357 static int __init parse_disable_radix(char *p) 358 { 359 bool val; 360 361 if (!p) 362 val = true; 363 else if (kstrtobool(p, &val)) 364 return -EINVAL; 365 366 disable_radix = val; 367 368 return 0; 369 } 370 early_param("disable_radix", parse_disable_radix); 371 372 /* 373 * If we're running under a hypervisor, we need to check the contents of 374 * /chosen/ibm,architecture-vec-5 to see if the hypervisor is willing to do 375 * radix. If not, we clear the radix feature bit so we fall back to hash. 376 */ 377 static void __init early_check_vec5(void) 378 { 379 unsigned long root, chosen; 380 int size; 381 const u8 *vec5; 382 u8 mmu_supported; 383 384 root = of_get_flat_dt_root(); 385 chosen = of_get_flat_dt_subnode_by_name(root, "chosen"); 386 if (chosen == -FDT_ERR_NOTFOUND) { 387 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; 388 return; 389 } 390 vec5 = of_get_flat_dt_prop(chosen, "ibm,architecture-vec-5", &size); 391 if (!vec5) { 392 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; 393 return; 394 } 395 if (size <= OV5_INDX(OV5_MMU_SUPPORT)) { 396 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; 397 return; 398 } 399 400 /* Check for supported configuration */ 401 mmu_supported = vec5[OV5_INDX(OV5_MMU_SUPPORT)] & 402 OV5_FEAT(OV5_MMU_SUPPORT); 403 if (mmu_supported == OV5_FEAT(OV5_MMU_RADIX)) { 404 /* Hypervisor only supports radix - check enabled && GTSE */ 405 if (!early_radix_enabled()) { 406 pr_warn("WARNING: Ignoring cmdline option disable_radix\n"); 407 } 408 if (!(vec5[OV5_INDX(OV5_RADIX_GTSE)] & 409 OV5_FEAT(OV5_RADIX_GTSE))) { 410 pr_warn("WARNING: Hypervisor doesn't support RADIX with GTSE\n"); 411 } 412 /* Do radix anyway - the hypervisor said we had to */ 413 cur_cpu_spec->mmu_features |= MMU_FTR_TYPE_RADIX; 414 } else if (mmu_supported == OV5_FEAT(OV5_MMU_HASH)) { 415 /* Hypervisor only supports hash - disable radix */ 416 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; 417 } 418 } 419 420 void __init mmu_early_init_devtree(void) 421 { 422 /* Disable radix mode based on kernel command line. */ 423 if (disable_radix) 424 cur_cpu_spec->mmu_features &= ~MMU_FTR_TYPE_RADIX; 425 426 /* 427 * Check /chosen/ibm,architecture-vec-5 if running as a guest. 428 * When running bare-metal, we can use radix if we like 429 * even though the ibm,architecture-vec-5 property created by 430 * skiboot doesn't have the necessary bits set. 431 */ 432 if (!(mfmsr() & MSR_HV)) 433 early_check_vec5(); 434 435 if (early_radix_enabled()) 436 radix__early_init_devtree(); 437 else 438 hash__early_init_devtree(); 439 } 440 #endif /* CONFIG_PPC_BOOK3S_64 */ 441