machine_kexec_64.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413
  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. #include <linux/kexec.h>
  12. #include <linux/smp.h>
  13. #include <linux/thread_info.h>
  14. #include <linux/init_task.h>
  15. #include <linux/errno.h>
  16. #include <linux/kernel.h>
  17. #include <linux/cpu.h>
  18. #include <linux/hardirq.h>
  19. #include <asm/page.h>
  20. #include <asm/current.h>
  21. #include <asm/machdep.h>
  22. #include <asm/cacheflush.h>
  23. #include <asm/paca.h>
  24. #include <asm/mmu.h>
  25. #include <asm/sections.h> /* _end */
  26. #include <asm/prom.h>
  27. #include <asm/smp.h>
  28. #include <asm/hw_breakpoint.h>
  29. int default_machine_kexec_prepare(struct kimage *image)
  30. {
  31. int i;
  32. unsigned long begin, end; /* limits of segment */
  33. unsigned long low, high; /* limits of blocked memory range */
  34. struct device_node *node;
  35. const unsigned long *basep;
  36. const unsigned int *sizep;
  37. if (!ppc_md.hpte_clear_all)
  38. return -ENOENT;
  39. /*
  40. * Since we use the kernel fault handlers and paging code to
  41. * handle the virtual mode, we must make sure no destination
  42. * overlaps kernel static data or bss.
  43. */
  44. for (i = 0; i < image->nr_segments; i++)
  45. if (image->segment[i].mem < __pa(_end))
  46. return -ETXTBSY;
  47. /*
  48. * For non-LPAR, we absolutely can not overwrite the mmu hash
  49. * table, since we are still using the bolted entries in it to
  50. * do the copy. Check that here.
  51. *
  52. * It is safe if the end is below the start of the blocked
  53. * region (end <= low), or if the beginning is after the
  54. * end of the blocked region (begin >= high). Use the
  55. * boolean identity !(a || b) === (!a && !b).
  56. */
  57. if (htab_address) {
  58. low = __pa(htab_address);
  59. high = low + htab_size_bytes;
  60. for (i = 0; i < image->nr_segments; i++) {
  61. begin = image->segment[i].mem;
  62. end = begin + image->segment[i].memsz;
  63. if ((begin < high) && (end > low))
  64. return -ETXTBSY;
  65. }
  66. }
  67. /* We also should not overwrite the tce tables */
  68. for_each_node_by_type(node, "pci") {
  69. basep = of_get_property(node, "linux,tce-base", NULL);
  70. sizep = of_get_property(node, "linux,tce-size", NULL);
  71. if (basep == NULL || sizep == NULL)
  72. continue;
  73. low = *basep;
  74. high = low + (*sizep);
  75. for (i = 0; i < image->nr_segments; i++) {
  76. begin = image->segment[i].mem;
  77. end = begin + image->segment[i].memsz;
  78. if ((begin < high) && (end > low))
  79. return -ETXTBSY;
  80. }
  81. }
  82. return 0;
  83. }
  84. #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
  85. static void copy_segments(unsigned long ind)
  86. {
  87. unsigned long entry;
  88. unsigned long *ptr;
  89. void *dest;
  90. void *addr;
  91. /*
  92. * We rely on kexec_load to create a lists that properly
  93. * initializes these pointers before they are used.
  94. * We will still crash if the list is wrong, but at least
  95. * the compiler will be quiet.
  96. */
  97. ptr = NULL;
  98. dest = NULL;
  99. for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
  100. addr = __va(entry & PAGE_MASK);
  101. switch (entry & IND_FLAGS) {
  102. case IND_DESTINATION:
  103. dest = addr;
  104. break;
  105. case IND_INDIRECTION:
  106. ptr = addr;
  107. break;
  108. case IND_SOURCE:
  109. copy_page(dest, addr);
  110. dest += PAGE_SIZE;
  111. }
  112. }
  113. }
  114. void kexec_copy_flush(struct kimage *image)
  115. {
  116. long i, nr_segments = image->nr_segments;
  117. struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
  118. /* save the ranges on the stack to efficiently flush the icache */
  119. memcpy(ranges, image->segment, sizeof(ranges));
  120. /*
  121. * After this call we may not use anything allocated in dynamic
  122. * memory, including *image.
  123. *
  124. * Only globals and the stack are allowed.
  125. */
  126. copy_segments(image->head);
  127. /*
  128. * we need to clear the icache for all dest pages sometime,
  129. * including ones that were in place on the original copy
  130. */
  131. for (i = 0; i < nr_segments; i++)
  132. flush_icache_range((unsigned long)__va(ranges[i].mem),
  133. (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
  134. }
  135. #ifdef CONFIG_SMP
  136. static int kexec_all_irq_disabled = 0;
  137. static void kexec_smp_down(void *arg)
  138. {
  139. local_irq_disable();
  140. hard_irq_disable();
  141. mb(); /* make sure our irqs are disabled before we say they are */
  142. get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
  143. while(kexec_all_irq_disabled == 0)
  144. cpu_relax();
  145. mb(); /* make sure all irqs are disabled before this */
  146. hw_breakpoint_disable();
  147. /*
  148. * Now every CPU has IRQs off, we can clear out any pending
  149. * IPIs and be sure that no more will come in after this.
  150. */
  151. if (ppc_md.kexec_cpu_down)
  152. ppc_md.kexec_cpu_down(0, 1);
  153. kexec_smp_wait();
  154. /* NOTREACHED */
  155. }
  156. static void kexec_prepare_cpus_wait(int wait_state)
  157. {
  158. int my_cpu, i, notified=-1;
  159. hw_breakpoint_disable();
  160. my_cpu = get_cpu();
  161. /* Make sure each CPU has at least made it to the state we need.
  162. *
  163. * FIXME: There is a (slim) chance of a problem if not all of the CPUs
  164. * are correctly onlined. If somehow we start a CPU on boot with RTAS
  165. * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
  166. * time, the boot CPU will timeout. If it does eventually execute
  167. * stuff, the secondary will start up (paca[].cpu_start was written) and
  168. * get into a peculiar state. If the platform supports
  169. * smp_ops->take_timebase(), the secondary CPU will probably be spinning
  170. * in there. If not (i.e. pseries), the secondary will continue on and
  171. * try to online itself/idle/etc. If it survives that, we need to find
  172. * these possible-but-not-online-but-should-be CPUs and chaperone them
  173. * into kexec_smp_wait().
  174. */
  175. for_each_online_cpu(i) {
  176. if (i == my_cpu)
  177. continue;
  178. while (paca[i].kexec_state < wait_state) {
  179. barrier();
  180. if (i != notified) {
  181. printk(KERN_INFO "kexec: waiting for cpu %d "
  182. "(physical %d) to enter %i state\n",
  183. i, paca[i].hw_cpu_id, wait_state);
  184. notified = i;
  185. }
  186. }
  187. }
  188. mb();
  189. }
  190. /*
  191. * We need to make sure each present CPU is online. The next kernel will scan
  192. * the device tree and assume primary threads are online and query secondary
  193. * threads via RTAS to online them if required. If we don't online primary
  194. * threads, they will be stuck. However, we also online secondary threads as we
  195. * may be using 'cede offline'. In this case RTAS doesn't see the secondary
  196. * threads as offline -- and again, these CPUs will be stuck.
  197. *
  198. * So, we online all CPUs that should be running, including secondary threads.
  199. */
  200. static void wake_offline_cpus(void)
  201. {
  202. int cpu = 0;
  203. for_each_present_cpu(cpu) {
  204. if (!cpu_online(cpu)) {
  205. printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
  206. cpu);
  207. cpu_up(cpu);
  208. }
  209. }
  210. }
  211. static void kexec_prepare_cpus(void)
  212. {
  213. wake_offline_cpus();
  214. smp_call_function(kexec_smp_down, NULL, /* wait */0);
  215. local_irq_disable();
  216. hard_irq_disable();
  217. mb(); /* make sure IRQs are disabled before we say they are */
  218. get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
  219. kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
  220. /* we are sure every CPU has IRQs off at this point */
  221. kexec_all_irq_disabled = 1;
  222. /* after we tell the others to go down */
  223. if (ppc_md.kexec_cpu_down)
  224. ppc_md.kexec_cpu_down(0, 0);
  225. /*
  226. * Before removing MMU mappings make sure all CPUs have entered real
  227. * mode:
  228. */
  229. kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
  230. put_cpu();
  231. }
  232. #else /* ! SMP */
  233. static void kexec_prepare_cpus(void)
  234. {
  235. /*
  236. * move the secondarys to us so that we can copy
  237. * the new kernel 0-0x100 safely
  238. *
  239. * do this if kexec in setup.c ?
  240. *
  241. * We need to release the cpus if we are ever going from an
  242. * UP to an SMP kernel.
  243. */
  244. smp_release_cpus();
  245. if (ppc_md.kexec_cpu_down)
  246. ppc_md.kexec_cpu_down(0, 0);
  247. local_irq_disable();
  248. hard_irq_disable();
  249. }
  250. #endif /* SMP */
  251. /*
  252. * kexec thread structure and stack.
  253. *
  254. * We need to make sure that this is 16384-byte aligned due to the
  255. * way process stacks are handled. It also must be statically allocated
  256. * or allocated as part of the kimage, because everything else may be
  257. * overwritten when we copy the kexec image. We piggyback on the
  258. * "init_task" linker section here to statically allocate a stack.
  259. *
  260. * We could use a smaller stack if we don't care about anything using
  261. * current, but that audit has not been performed.
  262. */
  263. static union thread_union kexec_stack __init_task_data =
  264. { };
  265. /*
  266. * For similar reasons to the stack above, the kexecing CPU needs to be on a
  267. * static PACA; we switch to kexec_paca.
  268. */
  269. struct paca_struct kexec_paca;
  270. /* Our assembly helper, in kexec_stub.S */
  271. extern void kexec_sequence(void *newstack, unsigned long start,
  272. void *image, void *control,
  273. void (*clear_all)(void)) __noreturn;
  274. /* too late to fail here */
  275. void default_machine_kexec(struct kimage *image)
  276. {
  277. /* prepare control code if any */
  278. /*
  279. * If the kexec boot is the normal one, need to shutdown other cpus
  280. * into our wait loop and quiesce interrupts.
  281. * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
  282. * stopping other CPUs and collecting their pt_regs is done before
  283. * using debugger IPI.
  284. */
  285. if (crashing_cpu == -1)
  286. kexec_prepare_cpus();
  287. pr_debug("kexec: Starting switchover sequence.\n");
  288. /* switch to a staticly allocated stack. Based on irq stack code.
  289. * We setup preempt_count to avoid using VMX in memcpy.
  290. * XXX: the task struct will likely be invalid once we do the copy!
  291. */
  292. kexec_stack.thread_info.task = current_thread_info()->task;
  293. kexec_stack.thread_info.flags = 0;
  294. kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET;
  295. kexec_stack.thread_info.cpu = current_thread_info()->cpu;
  296. /* We need a static PACA, too; copy this CPU's PACA over and switch to
  297. * it. Also poison per_cpu_offset to catch anyone using non-static
  298. * data.
  299. */
  300. memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
  301. kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
  302. paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
  303. kexec_paca.paca_index;
  304. setup_paca(&kexec_paca);
  305. /* XXX: If anyone does 'dynamic lppacas' this will also need to be
  306. * switched to a static version!
  307. */
  308. /* Some things are best done in assembly. Finding globals with
  309. * a toc is easier in C, so pass in what we can.
  310. */
  311. kexec_sequence(&kexec_stack, image->start, image,
  312. page_address(image->control_code_page),
  313. ppc_md.hpte_clear_all);
  314. /* NOTREACHED */
  315. }
  316. /* Values we need to export to the second kernel via the device tree. */
  317. static unsigned long htab_base;
  318. static struct property htab_base_prop = {
  319. .name = "linux,htab-base",
  320. .length = sizeof(unsigned long),
  321. .value = &htab_base,
  322. };
  323. static struct property htab_size_prop = {
  324. .name = "linux,htab-size",
  325. .length = sizeof(unsigned long),
  326. .value = &htab_size_bytes,
  327. };
  328. static int __init export_htab_values(void)
  329. {
  330. struct device_node *node;
  331. struct property *prop;
  332. /* On machines with no htab htab_address is NULL */
  333. if (!htab_address)
  334. return -ENODEV;
  335. node = of_find_node_by_path("/chosen");
  336. if (!node)
  337. return -ENODEV;
  338. /* remove any stale propertys so ours can be found */
  339. prop = of_find_property(node, htab_base_prop.name, NULL);
  340. if (prop)
  341. of_remove_property(node, prop);
  342. prop = of_find_property(node, htab_size_prop.name, NULL);
  343. if (prop)
  344. of_remove_property(node, prop);
  345. htab_base = __pa(htab_address);
  346. of_add_property(node, &htab_base_prop);
  347. of_add_property(node, &htab_size_prop);
  348. of_node_put(node);
  349. return 0;
  350. }
  351. late_initcall(export_htab_values);