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/cpu_mf.h> 7 #include <asm/setup.h> 8 #include <asm/kasan.h> 9 #include <asm/kexec.h> 10 #include <asm/sclp.h> 11 #include <asm/diag.h> 12 #include <asm/uv.h> 13 #include "compressed/decompressor.h" 14 #include "boot.h" 15 #include "uv.h" 16 17 unsigned long __bootdata_preserved(__kaslr_offset); 18 unsigned long __bootdata(__amode31_base); 19 unsigned long __bootdata_preserved(VMALLOC_START); 20 unsigned long __bootdata_preserved(VMALLOC_END); 21 struct page *__bootdata_preserved(vmemmap); 22 unsigned long __bootdata_preserved(vmemmap_size); 23 unsigned long __bootdata_preserved(MODULES_VADDR); 24 unsigned long __bootdata_preserved(MODULES_END); 25 unsigned long __bootdata(ident_map_size); 26 int __bootdata(is_full_image) = 1; 27 struct initrd_data __bootdata(initrd_data); 28 29 u64 __bootdata_preserved(stfle_fac_list[16]); 30 u64 __bootdata_preserved(alt_stfle_fac_list[16]); 31 struct oldmem_data __bootdata_preserved(oldmem_data); 32 33 void error(char *x) 34 { 35 sclp_early_printk("\n\n"); 36 sclp_early_printk(x); 37 sclp_early_printk("\n\n -- System halted"); 38 39 disabled_wait(); 40 } 41 42 static void setup_lpp(void) 43 { 44 S390_lowcore.current_pid = 0; 45 S390_lowcore.lpp = LPP_MAGIC; 46 if (test_facility(40)) 47 lpp(&S390_lowcore.lpp); 48 } 49 50 #ifdef CONFIG_KERNEL_UNCOMPRESSED 51 unsigned long mem_safe_offset(void) 52 { 53 return vmlinux.default_lma + vmlinux.image_size + vmlinux.bss_size; 54 } 55 #endif 56 57 static void rescue_initrd(unsigned long addr) 58 { 59 if (!IS_ENABLED(CONFIG_BLK_DEV_INITRD)) 60 return; 61 if (!initrd_data.start || !initrd_data.size) 62 return; 63 if (addr <= initrd_data.start) 64 return; 65 memmove((void *)addr, (void *)initrd_data.start, initrd_data.size); 66 initrd_data.start = addr; 67 } 68 69 static void copy_bootdata(void) 70 { 71 if (__boot_data_end - __boot_data_start != vmlinux.bootdata_size) 72 error(".boot.data section size mismatch"); 73 memcpy((void *)vmlinux.bootdata_off, __boot_data_start, vmlinux.bootdata_size); 74 if (__boot_data_preserved_end - __boot_data_preserved_start != vmlinux.bootdata_preserved_size) 75 error(".boot.preserved.data section size mismatch"); 76 memcpy((void *)vmlinux.bootdata_preserved_off, __boot_data_preserved_start, vmlinux.bootdata_preserved_size); 77 } 78 79 static void handle_relocs(unsigned long offset) 80 { 81 Elf64_Rela *rela_start, *rela_end, *rela; 82 int r_type, r_sym, rc; 83 Elf64_Addr loc, val; 84 Elf64_Sym *dynsym; 85 86 rela_start = (Elf64_Rela *) vmlinux.rela_dyn_start; 87 rela_end = (Elf64_Rela *) vmlinux.rela_dyn_end; 88 dynsym = (Elf64_Sym *) vmlinux.dynsym_start; 89 for (rela = rela_start; rela < rela_end; rela++) { 90 loc = rela->r_offset + offset; 91 val = rela->r_addend; 92 r_sym = ELF64_R_SYM(rela->r_info); 93 if (r_sym) { 94 if (dynsym[r_sym].st_shndx != SHN_UNDEF) 95 val += dynsym[r_sym].st_value + offset; 96 } else { 97 /* 98 * 0 == undefined symbol table index (STN_UNDEF), 99 * used for R_390_RELATIVE, only add KASLR offset 100 */ 101 val += offset; 102 } 103 r_type = ELF64_R_TYPE(rela->r_info); 104 rc = arch_kexec_do_relocs(r_type, (void *) loc, val, 0); 105 if (rc) 106 error("Unknown relocation type"); 107 } 108 } 109 110 /* 111 * Merge information from several sources into a single ident_map_size value. 112 * "ident_map_size" represents the upper limit of physical memory we may ever 113 * reach. It might not be all online memory, but also include standby (offline) 114 * memory. "ident_map_size" could be lower then actual standby or even online 115 * memory present, due to limiting factors. We should never go above this limit. 116 * It is the size of our identity mapping. 117 * 118 * Consider the following factors: 119 * 1. max_physmem_end - end of physical memory online or standby. 120 * Always <= end of the last online memory block (get_mem_detect_end()). 121 * 2. CONFIG_MAX_PHYSMEM_BITS - the maximum size of physical memory the 122 * kernel is able to support. 123 * 3. "mem=" kernel command line option which limits physical memory usage. 124 * 4. OLDMEM_BASE which is a kdump memory limit when the kernel is executed as 125 * crash kernel. 126 * 5. "hsa" size which is a memory limit when the kernel is executed during 127 * zfcp/nvme dump. 128 */ 129 static void setup_ident_map_size(unsigned long max_physmem_end) 130 { 131 unsigned long hsa_size; 132 133 ident_map_size = max_physmem_end; 134 if (memory_limit) 135 ident_map_size = min(ident_map_size, memory_limit); 136 ident_map_size = min(ident_map_size, 1UL << MAX_PHYSMEM_BITS); 137 138 #ifdef CONFIG_CRASH_DUMP 139 if (oldmem_data.start) { 140 kaslr_enabled = 0; 141 ident_map_size = min(ident_map_size, oldmem_data.size); 142 } else if (ipl_block_valid && is_ipl_block_dump()) { 143 kaslr_enabled = 0; 144 if (!sclp_early_get_hsa_size(&hsa_size) && hsa_size) 145 ident_map_size = min(ident_map_size, hsa_size); 146 } 147 #endif 148 } 149 150 static void setup_kernel_memory_layout(void) 151 { 152 bool vmalloc_size_verified = false; 153 unsigned long vmemmap_off; 154 unsigned long vspace_left; 155 unsigned long rte_size; 156 unsigned long pages; 157 unsigned long vmax; 158 159 pages = ident_map_size / PAGE_SIZE; 160 /* vmemmap contains a multiple of PAGES_PER_SECTION struct pages */ 161 vmemmap_size = SECTION_ALIGN_UP(pages) * sizeof(struct page); 162 163 /* choose kernel address space layout: 4 or 3 levels. */ 164 vmemmap_off = round_up(ident_map_size, _REGION3_SIZE); 165 if (IS_ENABLED(CONFIG_KASAN) || 166 vmalloc_size > _REGION2_SIZE || 167 vmemmap_off + vmemmap_size + vmalloc_size + MODULES_LEN > _REGION2_SIZE) 168 vmax = _REGION1_SIZE; 169 else 170 vmax = _REGION2_SIZE; 171 172 /* keep vmemmap_off aligned to a top level region table entry */ 173 rte_size = vmax == _REGION1_SIZE ? _REGION2_SIZE : _REGION3_SIZE; 174 MODULES_END = vmax; 175 if (is_prot_virt_host()) { 176 /* 177 * forcing modules and vmalloc area under the ultravisor 178 * secure storage limit, so that any vmalloc allocation 179 * we do could be used to back secure guest storage. 180 */ 181 adjust_to_uv_max(&MODULES_END); 182 } 183 184 #ifdef CONFIG_KASAN 185 if (MODULES_END < vmax) { 186 /* force vmalloc and modules below kasan shadow */ 187 MODULES_END = min(MODULES_END, KASAN_SHADOW_START); 188 } else { 189 /* 190 * leave vmalloc and modules above kasan shadow but make 191 * sure they don't overlap with it 192 */ 193 vmalloc_size = min(vmalloc_size, vmax - KASAN_SHADOW_END - MODULES_LEN); 194 vmalloc_size_verified = true; 195 vspace_left = KASAN_SHADOW_START; 196 } 197 #endif 198 MODULES_VADDR = MODULES_END - MODULES_LEN; 199 VMALLOC_END = MODULES_VADDR; 200 201 if (vmalloc_size_verified) { 202 VMALLOC_START = VMALLOC_END - vmalloc_size; 203 } else { 204 vmemmap_off = round_up(ident_map_size, rte_size); 205 206 if (vmemmap_off + vmemmap_size > VMALLOC_END || 207 vmalloc_size > VMALLOC_END - vmemmap_off - vmemmap_size) { 208 /* 209 * allow vmalloc area to occupy up to 1/2 of 210 * the rest virtual space left. 211 */ 212 vmalloc_size = min(vmalloc_size, VMALLOC_END / 2); 213 } 214 VMALLOC_START = VMALLOC_END - vmalloc_size; 215 vspace_left = VMALLOC_START; 216 } 217 218 pages = vspace_left / (PAGE_SIZE + sizeof(struct page)); 219 pages = SECTION_ALIGN_UP(pages); 220 vmemmap_off = round_up(vspace_left - pages * sizeof(struct page), rte_size); 221 /* keep vmemmap left most starting from a fresh region table entry */ 222 vmemmap_off = min(vmemmap_off, round_up(ident_map_size, rte_size)); 223 /* take care that identity map is lower then vmemmap */ 224 ident_map_size = min(ident_map_size, vmemmap_off); 225 vmemmap_size = SECTION_ALIGN_UP(ident_map_size / PAGE_SIZE) * sizeof(struct page); 226 VMALLOC_START = max(vmemmap_off + vmemmap_size, VMALLOC_START); 227 vmemmap = (struct page *)vmemmap_off; 228 } 229 230 /* 231 * This function clears the BSS section of the decompressed Linux kernel and NOT the decompressor's. 232 */ 233 static void clear_bss_section(void) 234 { 235 memset((void *)vmlinux.default_lma + vmlinux.image_size, 0, vmlinux.bss_size); 236 } 237 238 /* 239 * Set vmalloc area size to an 8th of (potential) physical memory 240 * size, unless size has been set by kernel command line parameter. 241 */ 242 static void setup_vmalloc_size(void) 243 { 244 unsigned long size; 245 246 if (vmalloc_size_set) 247 return; 248 size = round_up(ident_map_size / 8, _SEGMENT_SIZE); 249 vmalloc_size = max(size, vmalloc_size); 250 } 251 252 static void offset_vmlinux_info(unsigned long offset) 253 { 254 vmlinux.default_lma += offset; 255 *(unsigned long *)(&vmlinux.entry) += offset; 256 vmlinux.bootdata_off += offset; 257 vmlinux.bootdata_preserved_off += offset; 258 vmlinux.rela_dyn_start += offset; 259 vmlinux.rela_dyn_end += offset; 260 vmlinux.dynsym_start += offset; 261 } 262 263 static unsigned long reserve_amode31(unsigned long safe_addr) 264 { 265 __amode31_base = PAGE_ALIGN(safe_addr); 266 return safe_addr + vmlinux.amode31_size; 267 } 268 269 void startup_kernel(void) 270 { 271 unsigned long random_lma; 272 unsigned long safe_addr; 273 void *img; 274 275 initrd_data.start = parmarea.initrd_start; 276 initrd_data.size = parmarea.initrd_size; 277 oldmem_data.start = parmarea.oldmem_base; 278 oldmem_data.size = parmarea.oldmem_size; 279 280 setup_lpp(); 281 store_ipl_parmblock(); 282 safe_addr = mem_safe_offset(); 283 safe_addr = reserve_amode31(safe_addr); 284 safe_addr = read_ipl_report(safe_addr); 285 uv_query_info(); 286 rescue_initrd(safe_addr); 287 sclp_early_read_info(); 288 setup_boot_command_line(); 289 parse_boot_command_line(); 290 sanitize_prot_virt_host(); 291 setup_ident_map_size(detect_memory()); 292 setup_vmalloc_size(); 293 setup_kernel_memory_layout(); 294 295 if (IS_ENABLED(CONFIG_RANDOMIZE_BASE) && kaslr_enabled) { 296 random_lma = get_random_base(safe_addr); 297 if (random_lma) { 298 __kaslr_offset = random_lma - vmlinux.default_lma; 299 img = (void *)vmlinux.default_lma; 300 offset_vmlinux_info(__kaslr_offset); 301 } 302 } 303 304 if (!IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) { 305 img = decompress_kernel(); 306 memmove((void *)vmlinux.default_lma, img, vmlinux.image_size); 307 } else if (__kaslr_offset) 308 memcpy((void *)vmlinux.default_lma, img, vmlinux.image_size); 309 310 clear_bss_section(); 311 copy_bootdata(); 312 if (IS_ENABLED(CONFIG_RELOCATABLE)) 313 handle_relocs(__kaslr_offset); 314 315 if (__kaslr_offset) { 316 /* 317 * Save KASLR offset for early dumps, before vmcore_info is set. 318 * Mark as uneven to distinguish from real vmcore_info pointer. 319 */ 320 S390_lowcore.vmcore_info = __kaslr_offset | 0x1UL; 321 /* Clear non-relocated kernel */ 322 if (IS_ENABLED(CONFIG_KERNEL_UNCOMPRESSED)) 323 memset(img, 0, vmlinux.image_size); 324 } 325 vmlinux.entry(); 326 } 327