1 // SPDX-License-Identifier: GPL-2.0 2 #include <linux/string.h> 3 #include <linux/elf.h> 4 #include <asm/boot_data.h> 5 #include <asm/sections.h> 6 #include <asm/maccess.h> 7 #include <asm/cpu_mf.h> 8 #include <asm/setup.h> 9 #include <asm/kasan.h> 10 #include <asm/kexec.h> 11 #include <asm/sclp.h> 12 #include <asm/diag.h> 13 #include <asm/uv.h> 14 #include <asm/abs_lowcore.h> 15 #include <asm/physmem_info.h> 16 #include "decompressor.h" 17 #include "boot.h" 18 #include "uv.h" 19 20 unsigned long __bootdata_preserved(__kaslr_offset); 21 unsigned long __bootdata_preserved(__abs_lowcore); 22 unsigned long __bootdata_preserved(__memcpy_real_area); 23 pte_t *__bootdata_preserved(memcpy_real_ptep); 24 unsigned long __bootdata_preserved(VMALLOC_START); 25 unsigned long __bootdata_preserved(VMALLOC_END); 26 struct page *__bootdata_preserved(vmemmap); 27 unsigned long __bootdata_preserved(vmemmap_size); 28 unsigned long __bootdata_preserved(MODULES_VADDR); 29 unsigned long __bootdata_preserved(MODULES_END); 30 unsigned long __bootdata_preserved(max_mappable); 31 unsigned long __bootdata(ident_map_size); 32 33 u64 __bootdata_preserved(stfle_fac_list[16]); 34 u64 __bootdata_preserved(alt_stfle_fac_list[16]); 35 struct oldmem_data __bootdata_preserved(oldmem_data); 36 37 struct machine_info machine; 38 39 void error(char *x) 40 { 41 sclp_early_printk("\n\n"); 42 sclp_early_printk(x); 43 sclp_early_printk("\n\n -- System halted"); 44 45 disabled_wait(); 46 } 47 48 static void detect_facilities(void) 49 { 50 if (test_facility(8)) { 51 machine.has_edat1 = 1; 52 __ctl_set_bit(0, 23); 53 } 54 if (test_facility(78)) 55 machine.has_edat2 = 1; 56 if (!noexec_disabled && test_facility(130)) { 57 machine.has_nx = 1; 58 __ctl_set_bit(0, 20); 59 } 60 } 61 62 static void setup_lpp(void) 63 { 64 S390_lowcore.current_pid = 0; 65 S390_lowcore.lpp = LPP_MAGIC; 66 if (test_facility(40)) 67 lpp(&S390_lowcore.lpp); 68 } 69 70 #ifdef CONFIG_KERNEL_UNCOMPRESSED 71 unsigned long mem_safe_offset(void) 72 { 73 return vmlinux.default_lma + vmlinux.image_size + vmlinux.bss_size; 74 } 75 #endif 76 77 static void rescue_initrd(unsigned long min, unsigned long max) 78 { 79 unsigned long old_addr, addr, size; 80 81 if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD)) 82 return; 83 if (!get_physmem_reserved(RR_INITRD, &addr, &size)) 84 return; 85 if (addr >= min && addr + size <= max) 86 return; 87 old_addr = addr; 88 physmem_free(RR_INITRD); 89 addr = physmem_alloc_top_down(RR_INITRD, size, 0); 90 memmove((void *)addr, (void *)old_addr, size); 91 } 92 93 static void copy_bootdata(void) 94 { 95 if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size) 96 error(".boot.data section size mismatch"); 97 memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size); 98 if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size) 99 error(".boot.preserved.data section size mismatch"); 100 memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size); 101 } 102 103 static void handle_relocs(unsigned long offset) 104 { 105 Elf64_Rela *rela_start, *rela_end, *rela; 106 int r_type, r_sym, rc; 107 Elf64_Addr loc, val; 108 Elf64_Sym *dynsym; 109 110 rela_start = (Elf64_Rela *) vmlinux.rela_dyn_start; 111 rela_end = (Elf64_Rela *) vmlinux.rela_dyn_end; 112 dynsym = (Elf64_Sym *) vmlinux.dynsym_start; 113 for (rela = rela_start; rela < rela_end; rela++) { 114 loc = rela->r_offset + offset; 115 val = rela->r_addend; 116 r_sym = ELF64_R_SYM(rela->r_info); 117 if (r_sym) { 118 if (dynsym[r_sym].st_shndx != SHN_UNDEF) 119 val += dynsym[r_sym].st_value + offset; 120 } else { 121 /* 122 * 0 == undefined symbol table index (STN_UNDEF), 123 * used for R_390_RELATIVE, only add KASLR offset 124 */ 125 val += offset; 126 } 127 r_type = ELF64_R_TYPE(rela->r_info); 128 rc = arch_kexec_do_relocs(r_type, (void *) loc, val, 0); 129 if (rc) 130 error("Unknown relocation type"); 131 } 132 } 133 134 /* 135 * Merge information from several sources into a single ident_map_size value. 136 * "ident_map_size" represents the upper limit of physical memory we may ever 137 * reach. It might not be all online memory, but also include standby (offline) 138 * memory. "ident_map_size" could be lower then actual standby or even online 139 * memory present, due to limiting factors. We should never go above this limit. 140 * It is the size of our identity mapping. 141 * 142 * Consider the following factors: 143 * 1. max_physmem_end - end of physical memory online or standby. 144 * Always >= end of the last online memory range (get_physmem_online_end()). 145 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the 146 * kernel is able to support. 147 * 3. "mem=" kernel command line option which limits physical memory usage. 148 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as 149 * crash kernel. 150 * 5. "hsa" size which is a memory limit when the kernel is executed during 151 * zfcp/nvme dump. 152 */ 153 static void setup_ident_map_size(unsigned long max_physmem_end) 154 { 155 unsigned long hsa_size; 156 157 ident_map_size = max_physmem_end; 158 if (memory_limit) 159 ident_map_size = min(ident_map_size, memory_limit); 160 ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS); 161 162 #ifdef CONFIG_CRASH_DUMP 163 if (oldmem_data.start) { 164 __kaslr_enabled = 0; 165 ident_map_size = min(ident_map_size, oldmem_data.size); 166 } else if (ipl_block_valid && is_ipl_block_dump()) { 167 __kaslr_enabled = 0; 168 if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size) 169 ident_map_size = min(ident_map_size, hsa_size); 170 } 171 #endif 172 } 173 174 static unsigned long setup_kernel_memory_layout(void) 175 { 176 unsigned long vmemmap_start; 177 unsigned long asce_limit; 178 unsigned long rte_size; 179 unsigned long pages; 180 unsigned long vmax; 181 182 pages = ident_map_size / PAGE_SIZE; 183 /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */ 184 vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page); 185 186 /* choose kernel address space layout: 4 or 3 levels. */ 187 vmemmap_start = round_up(ident_map_size, _REGION3_SIZE); 188 if (IS_ENABLED(CONFIG_KASAN) || 189 vmalloc_size > _REGION2_SIZE || 190 vmemmap_start + vmemmap_size + vmalloc_size + MODULES_LEN > 191 _REGION2_SIZE) { 192 asce_limit = _REGION1_SIZE; 193 rte_size = _REGION2_SIZE; 194 } else { 195 asce_limit = _REGION2_SIZE; 196 rte_size = _REGION3_SIZE; 197 } 198 /* 199 * forcing modules and vmalloc area under the ultravisor 200 * secure storage limit, so that any vmalloc allocation 201 * we do could be used to back secure guest storage. 202 */ 203 vmax = adjust_to_uv_max(asce_limit); 204 #ifdef CONFIG_KASAN 205 /* force vmalloc and modules below kasan shadow */ 206 vmax = min(vmax, KASAN_SHADOW_START); 207 #endif 208 __memcpy_real_area = round_down(vmax - PAGE_SIZE, PAGE_SIZE); 209 __abs_lowcore = round_down(__memcpy_real_area - ABS_LOWCORE_MAP_SIZE, 210 sizeof(struct lowcore)); 211 MODULES_END = round_down(__abs_lowcore, _SEGMENT_SIZE); 212 MODULES_VADDR = MODULES_END - MODULES_LEN; 213 VMALLOC_END = MODULES_VADDR; 214 215 /* allow vmalloc area to occupy up to about 1/2 of the rest virtual space left */ 216 vmalloc_size = min(vmalloc_size, round_down(VMALLOC_END / 2, _REGION3_SIZE)); 217 VMALLOC_START = VMALLOC_END - vmalloc_size; 218 219 /* split remaining virtual space between 1:1 mapping & vmemmap array */ 220 pages = VMALLOC_START / (PAGE_SIZE + sizeof(struct page)); 221 pages = SECTION_ALIGN_UP(pages); 222 /* keep vmemmap_start aligned to a top level region table entry */ 223 vmemmap_start = round_down(VMALLOC_START - pages * sizeof(struct page), rte_size); 224 vmemmap_start = min(vmemmap_start, 1UL << MAX_PHYSMEM_BITS); 225 /* maximum mappable address as seen by arch_get_mappable_range() */ 226 max_mappable = vmemmap_start; 227 /* make sure identity map doesn't overlay with vmemmap */ 228 ident_map_size = min(ident_map_size, vmemmap_start); 229 vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page); 230 /* make sure vmemmap doesn't overlay with vmalloc area */ 231 VMALLOC_START = max(vmemmap_start + vmemmap_size, VMALLOC_START); 232 vmemmap = (struct page *)vmemmap_start; 233 234 return asce_limit; 235 } 236 237 /* 238 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's. 239 */ 240 static void clear_bss_section(unsigned long vmlinux_lma) 241 { 242 memset((void *)vmlinux_lma + vmlinux.image_size, 0, vmlinux.bss_size); 243 } 244 245 /* 246 * Set vmalloc area size to an 8th of (potential) physical memory 247 * size, unless size has been set by kernel command line parameter. 248 */ 249 static void setup_vmalloc_size(void) 250 { 251 unsigned long size; 252 253 if (vmalloc_size_set) 254 return; 255 size = round_up(ident_map_size / 8, _SEGMENT_SIZE); 256 vmalloc_size = max(size, vmalloc_size); 257 } 258 259 static void offset_vmlinux_info(unsigned long offset) 260 { 261 *(unsigned long *)(&vmlinux.entry) += offset; 262 vmlinux.bootdata_off += offset; 263 vmlinux.bootdata_preserved_off += offset; 264 vmlinux.rela_dyn_start += offset; 265 vmlinux.rela_dyn_end += offset; 266 vmlinux.dynsym_start += offset; 267 vmlinux.init_mm_off += offset; 268 vmlinux.swapper_pg_dir_off += offset; 269 vmlinux.invalid_pg_dir_off += offset; 270 #ifdef CONFIG_KASAN 271 vmlinux.kasan_early_shadow_page_off += offset; 272 vmlinux.kasan_early_shadow_pte_off += offset; 273 vmlinux.kasan_early_shadow_pmd_off += offset; 274 vmlinux.kasan_early_shadow_pud_off += offset; 275 vmlinux.kasan_early_shadow_p4d_off += offset; 276 #endif 277 } 278 279 void startup_kernel(void) 280 { 281 unsigned long max_physmem_end; 282 unsigned long vmlinux_lma = 0; 283 unsigned long amode31_lma = 0; 284 unsigned long asce_limit; 285 unsigned long safe_addr; 286 void *img; 287 psw_t psw; 288 289 setup_lpp(); 290 safe_addr = mem_safe_offset(); 291 /* 292 * reserve decompressor memory together with decompression heap, buffer and 293 * memory which might be occupied by uncompressed kernel at default 1Mb 294 * position (if KASLR is off or failed). 295 */ 296 physmem_reserve(RR_DECOMPRESSOR, 0, safe_addr); 297 if (IS_ENABLED(CONFIG_BLK_DEV_INITRD) && parmarea.initrd_size) 298 physmem_reserve(RR_INITRD, parmarea.initrd_start, parmarea.initrd_size); 299 oldmem_data.start = parmarea.oldmem_base; 300 oldmem_data.size = parmarea.oldmem_size; 301 302 store_ipl_parmblock(); 303 read_ipl_report(); 304 uv_query_info(); 305 sclp_early_read_info(); 306 setup_boot_command_line(); 307 parse_boot_command_line(); 308 detect_facilities(); 309 sanitize_prot_virt_host(); 310 max_physmem_end = detect_max_physmem_end(); 311 setup_ident_map_size(max_physmem_end); 312 setup_vmalloc_size(); 313 asce_limit = setup_kernel_memory_layout(); 314 /* got final ident_map_size, physmem allocations could be performed now */ 315 physmem_set_usable_limit(ident_map_size); 316 detect_physmem_online_ranges(max_physmem_end); 317 save_ipl_cert_comp_list(); 318 rescue_initrd(safe_addr, ident_map_size); 319 320 if (kaslr_enabled()) { 321 vmlinux_lma = randomize_within_range(vmlinux.image_size + vmlinux.bss_size, 322 THREAD_SIZE, vmlinux.default_lma, 323 ident_map_size); 324 if (vmlinux_lma) { 325 __kaslr_offset = vmlinux_lma - vmlinux.default_lma; 326 offset_vmlinux_info(__kaslr_offset); 327 } 328 } 329 vmlinux_lma = vmlinux_lma ?: vmlinux.default_lma; 330 physmem_reserve(RR_VMLINUX, vmlinux_lma, vmlinux.image_size + vmlinux.bss_size); 331 332 if (!IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) { 333 img = decompress_kernel(); 334 memmove((void *)vmlinux_lma, img, vmlinux.image_size); 335 } else if (__kaslr_offset) { 336 img = (void *)vmlinux.default_lma; 337 memmove((void *)vmlinux_lma, img, vmlinux.image_size); 338 memset(img, 0, vmlinux.image_size); 339 } 340 341 /* vmlinux decompression is done, shrink reserved low memory */ 342 physmem_reserve(RR_DECOMPRESSOR, 0, (unsigned long)_decompressor_end); 343 if (kaslr_enabled()) 344 amode31_lma = randomize_within_range(vmlinux.amode31_size, PAGE_SIZE, 0, SZ_2G); 345 amode31_lma = amode31_lma ?: vmlinux.default_lma - vmlinux.amode31_size; 346 physmem_reserve(RR_AMODE31, amode31_lma, vmlinux.amode31_size); 347 348 /* 349 * The order of the following operations is important: 350 * 351 * - handle_relocs() must follow clear_bss_section() to establish static 352 * memory references to data in .bss to be used by setup_vmem() 353 * (i.e init_mm.pgd) 354 * 355 * - setup_vmem() must follow handle_relocs() to be able using 356 * static memory references to data in .bss (i.e init_mm.pgd) 357 * 358 * - copy_bootdata() must follow setup_vmem() to propagate changes to 359 * bootdata made by setup_vmem() 360 */ 361 clear_bss_section(vmlinux_lma); 362 handle_relocs(__kaslr_offset); 363 setup_vmem(asce_limit); 364 copy_bootdata(); 365 366 /* 367 * Save KASLR offset for early dumps, before vmcore_info is set. 368 * Mark as uneven to distinguish from real vmcore_info pointer. 369 */ 370 S390_lowcore.vmcore_info = __kaslr_offset ? __kaslr_offset | 0x1UL : 0; 371 372 /* 373 * Jump to the decompressed kernel entry point and switch DAT mode on. 374 */ 375 psw.addr = vmlinux.entry; 376 psw.mask = PSW_KERNEL_BITS; 377 __load_psw(psw); 378 } 379