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/memblock.h> 16 #include <linux/highmem.h> 17 #include <linux/suspend.h> 18 #include <linux/dma-direct.h> 19 #include <linux/execmem.h> 20 #include <linux/vmalloc.h> 21 22 #include <asm/swiotlb.h> 23 #include <asm/machdep.h> 24 #include <asm/rtas.h> 25 #include <asm/kasan.h> 26 #include <asm/svm.h> 27 #include <asm/mmzone.h> 28 #include <asm/ftrace.h> 29 #include <asm/text-patching.h> 30 #include <asm/setup.h> 31 #include <asm/fixmap.h> 32 33 #include <mm/mmu_decl.h> 34 35 unsigned long long memory_limit __initdata; 36 37 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)] __page_aligned_bss; 38 EXPORT_SYMBOL(empty_zero_page); 39 40 pgprot_t __phys_mem_access_prot(unsigned long pfn, unsigned long size, 41 pgprot_t vma_prot) 42 { 43 if (ppc_md.phys_mem_access_prot) 44 return ppc_md.phys_mem_access_prot(pfn, size, vma_prot); 45 46 if (!page_is_ram(pfn)) 47 vma_prot = pgprot_noncached(vma_prot); 48 49 return vma_prot; 50 } 51 EXPORT_SYMBOL(__phys_mem_access_prot); 52 53 #ifdef CONFIG_MEMORY_HOTPLUG 54 static DEFINE_MUTEX(linear_mapping_mutex); 55 56 #ifdef CONFIG_NUMA 57 int memory_add_physaddr_to_nid(u64 start) 58 { 59 return hot_add_scn_to_nid(start); 60 } 61 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid); 62 #endif 63 64 int __weak create_section_mapping(unsigned long start, unsigned long end, 65 int nid, pgprot_t prot) 66 { 67 return -ENODEV; 68 } 69 70 int __weak remove_section_mapping(unsigned long start, unsigned long end) 71 { 72 return -ENODEV; 73 } 74 75 int __ref arch_create_linear_mapping(int nid, u64 start, u64 size, 76 struct mhp_params *params) 77 { 78 int rc; 79 80 start = (unsigned long)__va(start); 81 mutex_lock(&linear_mapping_mutex); 82 rc = create_section_mapping(start, start + size, nid, 83 params->pgprot); 84 mutex_unlock(&linear_mapping_mutex); 85 if (rc) { 86 pr_warn("Unable to create linear mapping for 0x%llx..0x%llx: %d\n", 87 start, start + size, rc); 88 return -EFAULT; 89 } 90 return 0; 91 } 92 93 void __ref arch_remove_linear_mapping(u64 start, u64 size) 94 { 95 int ret; 96 97 /* Remove htab bolted mappings for this section of memory */ 98 start = (unsigned long)__va(start); 99 100 mutex_lock(&linear_mapping_mutex); 101 ret = remove_section_mapping(start, start + size); 102 mutex_unlock(&linear_mapping_mutex); 103 if (ret) 104 pr_warn("Unable to remove linear mapping for 0x%llx..0x%llx: %d\n", 105 start, start + size, ret); 106 107 /* Ensure all vmalloc mappings are flushed in case they also 108 * hit that section of memory 109 */ 110 vm_unmap_aliases(); 111 } 112 113 /* 114 * After memory hotplug the variables max_pfn, max_low_pfn and high_memory need 115 * updating. 116 */ 117 static void update_end_of_memory_vars(u64 start, u64 size) 118 { 119 unsigned long end_pfn = PFN_UP(start + size); 120 121 if (end_pfn > max_pfn) { 122 max_pfn = end_pfn; 123 max_low_pfn = end_pfn; 124 high_memory = (void *)__va(max_pfn * PAGE_SIZE - 1) + 1; 125 } 126 } 127 128 int __ref add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages, 129 struct mhp_params *params) 130 { 131 int ret; 132 133 ret = __add_pages(nid, start_pfn, nr_pages, params); 134 if (ret) 135 return ret; 136 137 /* update max_pfn, max_low_pfn and high_memory */ 138 update_end_of_memory_vars(start_pfn << PAGE_SHIFT, 139 nr_pages << PAGE_SHIFT); 140 141 return ret; 142 } 143 144 int __ref arch_add_memory(int nid, u64 start, u64 size, 145 struct mhp_params *params) 146 { 147 unsigned long start_pfn = start >> PAGE_SHIFT; 148 unsigned long nr_pages = size >> PAGE_SHIFT; 149 int rc; 150 151 rc = arch_create_linear_mapping(nid, start, size, params); 152 if (rc) 153 return rc; 154 rc = add_pages(nid, start_pfn, nr_pages, params); 155 if (rc) 156 arch_remove_linear_mapping(start, size); 157 return rc; 158 } 159 160 void __ref arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap) 161 { 162 unsigned long start_pfn = start >> PAGE_SHIFT; 163 unsigned long nr_pages = size >> PAGE_SHIFT; 164 165 __remove_pages(start_pfn, nr_pages, altmap); 166 arch_remove_linear_mapping(start, size); 167 } 168 #endif 169 170 #ifndef CONFIG_NUMA 171 void __init mem_topology_setup(void) 172 { 173 max_low_pfn = max_pfn = memblock_end_of_DRAM() >> PAGE_SHIFT; 174 min_low_pfn = MEMORY_START >> PAGE_SHIFT; 175 #ifdef CONFIG_HIGHMEM 176 max_low_pfn = lowmem_end_addr >> PAGE_SHIFT; 177 #endif 178 179 /* Place all memblock_regions in the same node and merge contiguous 180 * memblock_regions 181 */ 182 memblock_set_node(0, PHYS_ADDR_MAX, &memblock.memory, 0); 183 } 184 185 /* mark pages that don't exist as nosave */ 186 static int __init mark_nonram_nosave(void) 187 { 188 unsigned long spfn, epfn, prev = 0; 189 int i; 190 191 for_each_mem_pfn_range(i, MAX_NUMNODES, &spfn, &epfn, NULL) { 192 if (prev && prev < spfn) 193 register_nosave_region(prev, spfn); 194 195 prev = epfn; 196 } 197 198 return 0; 199 } 200 #else /* CONFIG_NUMA */ 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 zone_dma_limit to inform the 215 * generic DMA mapping code. 32-bit only devices (if not handled by an IOMMU 216 * anyway) will take a first dip into ZONE_NORMAL and get otherwise served by 217 * ZONE_DMA. 218 */ 219 void __init arch_zone_limits_init(unsigned long *max_zone_pfns) 220 { 221 #ifdef CONFIG_ZONE_DMA 222 max_zone_pfns[ZONE_DMA] = min((zone_dma_limit >> PAGE_SHIFT) + 1, max_low_pfn); 223 #endif 224 max_zone_pfns[ZONE_NORMAL] = max_low_pfn; 225 #ifdef CONFIG_HIGHMEM 226 max_zone_pfns[ZONE_HIGHMEM] = max_pfn; 227 #endif 228 } 229 230 /* 231 * paging_init() sets up the page tables - in fact we've already done this. 232 */ 233 void __init paging_init(void) 234 { 235 unsigned long long total_ram = memblock_phys_mem_size(); 236 phys_addr_t top_of_ram = memblock_end_of_DRAM(); 237 int zone_dma_bits; 238 239 #ifdef CONFIG_HIGHMEM 240 unsigned long v = __fix_to_virt(FIX_KMAP_END); 241 unsigned long end = __fix_to_virt(FIX_KMAP_BEGIN); 242 243 for (; v < end; v += PAGE_SIZE) 244 map_kernel_page(v, 0, __pgprot(0)); /* XXX gross */ 245 246 map_kernel_page(PKMAP_BASE, 0, __pgprot(0)); /* XXX gross */ 247 pkmap_page_table = virt_to_kpte(PKMAP_BASE); 248 #endif /* CONFIG_HIGHMEM */ 249 250 printk(KERN_DEBUG "Top of RAM: 0x%llx, Total RAM: 0x%llx\n", 251 (unsigned long long)top_of_ram, total_ram); 252 printk(KERN_DEBUG "Memory hole size: %ldMB\n", 253 (long int)((top_of_ram - total_ram) >> 20)); 254 255 /* 256 * Allow 30-bit DMA for very limited Broadcom wifi chips on many 257 * powerbooks. 258 */ 259 if (IS_ENABLED(CONFIG_PPC32)) 260 zone_dma_bits = 30; 261 else 262 zone_dma_bits = 31; 263 264 zone_dma_limit = DMA_BIT_MASK(zone_dma_bits); 265 266 mark_nonram_nosave(); 267 } 268 269 void __init arch_mm_preinit(void) 270 { 271 /* 272 * book3s is limited to 16 page sizes due to encoding this in 273 * a 4-bit field for slices. 274 */ 275 BUILD_BUG_ON(MMU_PAGE_COUNT > 16); 276 277 #ifdef CONFIG_SWIOTLB 278 /* 279 * Some platforms (e.g. 85xx) limit DMA-able memory way below 280 * 4G. We force memblock to bottom-up mode to ensure that the 281 * memory allocated in swiotlb_init() is DMA-able. 282 * As it's the last memblock allocation, no need to reset it 283 * back to to-down. 284 */ 285 memblock_set_bottom_up(true); 286 swiotlb_init(ppc_swiotlb_enable, ppc_swiotlb_flags); 287 #endif 288 289 kasan_late_init(); 290 291 #if defined(CONFIG_PPC_E500) && !defined(CONFIG_SMP) 292 /* 293 * If smp is enabled, next_tlbcam_idx is initialized in the cpu up 294 * functions.... do it here for the non-smp case. 295 */ 296 per_cpu(next_tlbcam_idx, smp_processor_id()) = 297 (mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY) - 1; 298 #endif 299 } 300 301 void free_initmem(void) 302 { 303 ppc_md.progress = ppc_printk_progress; 304 mark_initmem_nx(); 305 free_initmem_default(POISON_FREE_INITMEM); 306 ftrace_free_init_tramp(); 307 } 308 309 /* 310 * System memory should not be in /proc/iomem but various tools expect it 311 * (eg kdump). 312 */ 313 static int __init add_system_ram_resources(void) 314 { 315 phys_addr_t start, end; 316 u64 i; 317 318 for_each_mem_range(i, &start, &end) { 319 struct resource *res; 320 321 res = kzalloc(sizeof(struct resource), GFP_KERNEL); 322 WARN_ON(!res); 323 324 if (res) { 325 res->name = "System RAM"; 326 res->start = start; 327 /* 328 * In memblock, end points to the first byte after 329 * the range while in resourses, end points to the 330 * last byte in the range. 331 */ 332 res->end = end - 1; 333 res->flags = IORESOURCE_SYSTEM_RAM | IORESOURCE_BUSY; 334 WARN_ON(insert_resource(&iomem_resource, res) < 0); 335 } 336 } 337 338 return 0; 339 } 340 subsys_initcall(add_system_ram_resources); 341 342 #ifdef CONFIG_STRICT_DEVMEM 343 /* 344 * devmem_is_allowed(): check to see if /dev/mem access to a certain address 345 * is valid. The argument is a physical page number. 346 * 347 * Access has to be given to non-kernel-ram areas as well, these contain the 348 * PCI mmio resources as well as potential bios/acpi data regions. 349 */ 350 int devmem_is_allowed(unsigned long pfn) 351 { 352 if (page_is_rtas_user_buf(pfn)) 353 return 1; 354 if (iomem_is_exclusive(PFN_PHYS(pfn))) 355 return 0; 356 if (!page_is_ram(pfn)) 357 return 1; 358 return 0; 359 } 360 #endif /* CONFIG_STRICT_DEVMEM */ 361 362 /* 363 * This is defined in kernel/resource.c but only powerpc needs to export it, for 364 * the EHEA driver. Drop this when drivers/net/ethernet/ibm/ehea is removed. 365 */ 366 EXPORT_SYMBOL_GPL(walk_system_ram_range); 367 368 #ifdef CONFIG_EXECMEM 369 static struct execmem_info execmem_info __ro_after_init; 370 371 #if defined(CONFIG_PPC_8xx) || defined(CONFIG_PPC_BOOK3S_603) 372 static void prealloc_execmem_pgtable(void) 373 { 374 unsigned long va; 375 376 for (va = ALIGN_DOWN(MODULES_VADDR, PGDIR_SIZE); va < MODULES_END; va += PGDIR_SIZE) 377 pte_alloc_kernel(pmd_off_k(va), va); 378 } 379 #else 380 static void prealloc_execmem_pgtable(void) { } 381 #endif 382 383 struct execmem_info __init *execmem_arch_setup(void) 384 { 385 pgprot_t kprobes_prot = strict_module_rwx_enabled() ? PAGE_KERNEL_ROX : PAGE_KERNEL_EXEC; 386 pgprot_t prot = strict_module_rwx_enabled() ? PAGE_KERNEL : PAGE_KERNEL_EXEC; 387 unsigned long fallback_start = 0, fallback_end = 0; 388 unsigned long start, end; 389 390 /* 391 * BOOK3S_32 and 8xx define MODULES_VADDR for text allocations and 392 * allow allocating data in the entire vmalloc space 393 */ 394 #ifdef MODULES_VADDR 395 unsigned long limit = (unsigned long)_etext - SZ_32M; 396 397 BUILD_BUG_ON(TASK_SIZE > MODULES_VADDR); 398 399 /* First try within 32M limit from _etext to avoid branch trampolines */ 400 if (MODULES_VADDR < PAGE_OFFSET && MODULES_END > limit) { 401 start = limit; 402 fallback_start = MODULES_VADDR; 403 fallback_end = MODULES_END; 404 } else { 405 start = MODULES_VADDR; 406 } 407 408 end = MODULES_END; 409 #else 410 start = VMALLOC_START; 411 end = VMALLOC_END; 412 #endif 413 414 prealloc_execmem_pgtable(); 415 416 execmem_info = (struct execmem_info){ 417 .ranges = { 418 [EXECMEM_DEFAULT] = { 419 .start = start, 420 .end = end, 421 .pgprot = prot, 422 .alignment = 1, 423 .fallback_start = fallback_start, 424 .fallback_end = fallback_end, 425 }, 426 [EXECMEM_KPROBES] = { 427 .start = VMALLOC_START, 428 .end = VMALLOC_END, 429 .pgprot = kprobes_prot, 430 .alignment = 1, 431 }, 432 [EXECMEM_MODULE_DATA] = { 433 .start = VMALLOC_START, 434 .end = VMALLOC_END, 435 .pgprot = PAGE_KERNEL, 436 .alignment = 1, 437 }, 438 }, 439 }; 440 441 return &execmem_info; 442 } 443 #endif /* CONFIG_EXECMEM */ 444