1/* 2 * PPC64 code to handle Linux booting another kernel. 3 * 4 * Copyright (C) 2004-2005, IBM Corp. 5 * 6 * Created by: Milton D Miller II 7 * 8 * This source code is licensed under the GNU General Public License, 9 * Version 2. See the file COPYING for more details. 10 */ 11 12 13#include <linux/kexec.h> 14#include <linux/smp.h> 15#include <linux/thread_info.h> 16#include <linux/init_task.h> 17#include <linux/errno.h> 18#include <linux/kernel.h> 19#include <linux/cpu.h> 20 21#include <asm/page.h> 22#include <asm/current.h> 23#include <asm/machdep.h> 24#include <asm/cacheflush.h> 25#include <asm/paca.h> 26#include <asm/mmu.h> 27#include <asm/sections.h> /* _end */ 28#include <asm/prom.h> 29#include <asm/smp.h> 30#include <asm/hw_breakpoint.h> 31 32int default_machine_kexec_prepare(struct kimage *image) 33{ 34 int i; 35 unsigned long begin, end; /* limits of segment */ 36 unsigned long low, high; /* limits of blocked memory range */ 37 struct device_node *node; 38 const unsigned long *basep; 39 const unsigned int *sizep; 40 41 if (!ppc_md.hpte_clear_all) 42 return -ENOENT; 43 44 /* 45 * Since we use the kernel fault handlers and paging code to 46 * handle the virtual mode, we must make sure no destination 47 * overlaps kernel static data or bss. 48 */ 49 for (i = 0; i < image->nr_segments; i++) 50 if (image->segment[i].mem < __pa(_end)) 51 return -ETXTBSY; 52 53 /* 54 * For non-LPAR, we absolutely can not overwrite the mmu hash 55 * table, since we are still using the bolted entries in it to 56 * do the copy. Check that here. 57 * 58 * It is safe if the end is below the start of the blocked 59 * region (end <= low), or if the beginning is after the 60 * end of the blocked region (begin >= high). Use the 61 * boolean identity !(a || b) === (!a && !b). 62 */ 63 if (htab_address) { 64 low = __pa(htab_address); 65 high = low + htab_size_bytes; 66 67 for (i = 0; i < image->nr_segments; i++) { 68 begin = image->segment[i].mem; 69 end = begin + image->segment[i].memsz; 70 71 if ((begin < high) && (end > low)) 72 return -ETXTBSY; 73 } 74 } 75 76 /* We also should not overwrite the tce tables */ 77 for (node = of_find_node_by_type(NULL, "pci"); node != NULL; 78 node = of_find_node_by_type(node, "pci")) { 79 basep = of_get_property(node, "linux,tce-base", NULL); 80 sizep = of_get_property(node, "linux,tce-size", NULL); 81 if (basep == NULL || sizep == NULL) 82 continue; 83 84 low = *basep; 85 high = low + (*sizep); 86 87 for (i = 0; i < image->nr_segments; i++) { 88 begin = image->segment[i].mem; 89 end = begin + image->segment[i].memsz; 90 91 if ((begin < high) && (end > low)) 92 return -ETXTBSY; 93 } 94 } 95 96 return 0; 97} 98 99#define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE) 100 101static void copy_segments(unsigned long ind) 102{ 103 unsigned long entry; 104 unsigned long *ptr; 105 void *dest; 106 void *addr; 107 108 /* 109 * We rely on kexec_load to create a lists that properly 110 * initializes these pointers before they are used. 111 * We will still crash if the list is wrong, but at least 112 * the compiler will be quiet. 113 */ 114 ptr = NULL; 115 dest = NULL; 116 117 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) { 118 addr = __va(entry & PAGE_MASK); 119 120 switch (entry & IND_FLAGS) { 121 case IND_DESTINATION: 122 dest = addr; 123 break; 124 case IND_INDIRECTION: 125 ptr = addr; 126 break; 127 case IND_SOURCE: 128 copy_page(dest, addr); 129 dest += PAGE_SIZE; 130 } 131 } 132} 133 134void kexec_copy_flush(struct kimage *image) 135{ 136 long i, nr_segments = image->nr_segments; 137 struct kexec_segment ranges[KEXEC_SEGMENT_MAX]; 138 139 /* save the ranges on the stack to efficiently flush the icache */ 140 memcpy(ranges, image->segment, sizeof(ranges)); 141 142 /* 143 * After this call we may not use anything allocated in dynamic 144 * memory, including *image. 145 * 146 * Only globals and the stack are allowed. 147 */ 148 copy_segments(image->head); 149 150 /* 151 * we need to clear the icache for all dest pages sometime, 152 * including ones that were in place on the original copy 153 */ 154 for (i = 0; i < nr_segments; i++) 155 flush_icache_range((unsigned long)__va(ranges[i].mem), 156 (unsigned long)__va(ranges[i].mem + ranges[i].memsz)); 157} 158 159#ifdef CONFIG_SMP 160 161static int kexec_all_irq_disabled = 0; 162 163static void kexec_smp_down(void *arg) 164{ 165 local_irq_disable(); 166 mb(); /* make sure our irqs are disabled before we say they are */ 167 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF; 168 while(kexec_all_irq_disabled == 0) 169 cpu_relax(); 170 mb(); /* make sure all irqs are disabled before this */ 171 hw_breakpoint_disable(); 172 /* 173 * Now every CPU has IRQs off, we can clear out any pending 174 * IPIs and be sure that no more will come in after this. 175 */ 176 if (ppc_md.kexec_cpu_down) 177 ppc_md.kexec_cpu_down(0, 1); 178 179 kexec_smp_wait(); 180 /* NOTREACHED */ 181} 182 183static void kexec_prepare_cpus_wait(int wait_state) 184{ 185 int my_cpu, i, notified=-1; 186 187 hw_breakpoint_disable(); 188 my_cpu = get_cpu(); 189 for_each_online_cpu(i) { 190 if (i == my_cpu) 191 continue; 192 193 while (paca[i].kexec_state < wait_state) { 194 barrier(); 195 if (i != notified) { 196 printk(KERN_INFO "kexec: waiting for cpu %d " 197 "(physical %d) to enter %i state\n", 198 i, paca[i].hw_cpu_id, wait_state); 199 notified = i; 200 } 201 } 202 } 203 mb(); 204} 205 206/* 207 * We need to make sure each present CPU is online. The next kernel will scan 208 * the device tree and assume primary threads are online and query secondary 209 * threads via RTAS to online them if required. If we don't online primary 210 * threads, they will be stuck. However, we also online secondary threads as we 211 * may be using 'cede offline'. In this case RTAS doesn't see the secondary 212 * threads as offline -- and again, these CPUs will be stuck. 213 * 214 * So, we online all CPUs that should be running, including secondary threads. 215 */ 216static void wake_offline_cpus(void) 217{ 218 int cpu = 0; 219 220 for_each_present_cpu(cpu) { 221 if (!cpu_online(cpu)) { 222 printk(KERN_INFO "kexec: Waking offline cpu %d.\n", 223 cpu); 224 cpu_up(cpu); 225 } 226 } 227} 228 229static void kexec_prepare_cpus(void) 230{ 231 wake_offline_cpus(); 232 smp_call_function(kexec_smp_down, NULL, /* wait */0); 233 local_irq_disable(); 234 mb(); /* make sure IRQs are disabled before we say they are */ 235 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF; 236 237 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF); 238 /* we are sure every CPU has IRQs off at this point */ 239 kexec_all_irq_disabled = 1; 240 241 /* after we tell the others to go down */ 242 if (ppc_md.kexec_cpu_down) 243 ppc_md.kexec_cpu_down(0, 0); 244 245 /* 246 * Before removing MMU mappings make sure all CPUs have entered real 247 * mode: 248 */ 249 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE); 250 251 put_cpu(); 252} 253 254#else /* ! SMP */ 255 256static void kexec_prepare_cpus(void) 257{ 258 /* 259 * move the secondarys to us so that we can copy 260 * the new kernel 0-0x100 safely 261 * 262 * do this if kexec in setup.c ? 263 * 264 * We need to release the cpus if we are ever going from an 265 * UP to an SMP kernel. 266 */ 267 smp_release_cpus(); 268 if (ppc_md.kexec_cpu_down) 269 ppc_md.kexec_cpu_down(0, 0); 270 local_irq_disable(); 271} 272 273#endif /* SMP */ 274 275/* 276 * kexec thread structure and stack. 277 * 278 * We need to make sure that this is 16384-byte aligned due to the 279 * way process stacks are handled. It also must be statically allocated 280 * or allocated as part of the kimage, because everything else may be 281 * overwritten when we copy the kexec image. We piggyback on the 282 * "init_task" linker section here to statically allocate a stack. 283 * 284 * We could use a smaller stack if we don't care about anything using 285 * current, but that audit has not been performed. 286 */ 287static union thread_union kexec_stack __init_task_data = 288 { }; 289 290/* 291 * For similar reasons to the stack above, the kexecing CPU needs to be on a 292 * static PACA; we switch to kexec_paca. 293 */ 294struct paca_struct kexec_paca; 295 296/* Our assembly helper, in kexec_stub.S */ 297extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start, 298 void *image, void *control, 299 void (*clear_all)(void)) ATTRIB_NORET; 300 301/* too late to fail here */ 302void default_machine_kexec(struct kimage *image) 303{ 304 /* prepare control code if any */ 305 306 /* 307 * If the kexec boot is the normal one, need to shutdown other cpus 308 * into our wait loop and quiesce interrupts. 309 * Otherwise, in the case of crashed mode (crashing_cpu >= 0), 310 * stopping other CPUs and collecting their pt_regs is done before 311 * using debugger IPI. 312 */ 313 314 if (crashing_cpu == -1) 315 kexec_prepare_cpus(); 316 317 pr_debug("kexec: Starting switchover sequence.\n"); 318 319 kexec_stack.thread_info.task = current_thread_info()->task; 320 kexec_stack.thread_info.flags = 0; 321 322 /* We need a static PACA, too; copy this CPU's PACA over and switch to 323 * it. Also poison per_cpu_offset to catch anyone using non-static 324 * data. 325 */ 326 memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct)); 327 kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL; 328 paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) - 329 kexec_paca.paca_index; 330 setup_paca(&kexec_paca); 331 332 333 /* Some things are best done in assembly. Finding globals with 334 * a toc is easier in C, so pass in what we can. 335 */ 336 kexec_sequence(&kexec_stack, image->start, image, 337 page_address(image->control_code_page), 338 ppc_md.hpte_clear_all); 339 /* NOTREACHED */ 340} 341 342/* Values we need to export to the second kernel via the device tree. */ 343static unsigned long htab_base; 344 345static struct property htab_base_prop = { 346 .name = "linux,htab-base", 347 .length = sizeof(unsigned long), 348 .value = &htab_base, 349}; 350 351static struct property htab_size_prop = { 352 .name = "linux,htab-size", 353 .length = sizeof(unsigned long), 354 .value = &htab_size_bytes, 355}; 356 357static int __init export_htab_values(void) 358{ 359 struct device_node *node; 360 struct property *prop; 361 362 /* On machines with no htab htab_address is NULL */ 363 if (!htab_address) 364 return -ENODEV; 365 366 node = of_find_node_by_path("/chosen"); 367 if (!node) 368 return -ENODEV; 369 370 /* remove any stale propertys so ours can be found */ 371 prop = of_find_property(node, htab_base_prop.name, NULL); 372 if (prop) 373 prom_remove_property(node, prop); 374 prop = of_find_property(node, htab_size_prop.name, NULL); 375 if (prop) 376 prom_remove_property(node, prop); 377 378 htab_base = __pa(htab_address); 379 prom_add_property(node, &htab_base_prop); 380 prom_add_property(node, &htab_size_prop); 381 382 of_node_put(node); 383 return 0; 384} 385late_initcall(export_htab_values); 386