arch/powerpc/kernel/machine_kexec_64.c

   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 #include <linux/hardirq.h>
  21
  22 #include <asm/page.h>
  23 #include <asm/current.h>
  24 #include <asm/machdep.h>
  25 #include <asm/cacheflush.h>
  26 #include <asm/paca.h>
  27 #include <asm/mmu.h>
  28 #include <asm/sections.h>       /* _end */
  29 #include <asm/prom.h>
  30 #include <asm/smp.h>
  31 #include <asm/hw_breakpoint.h>
  32
  33 int default_machine_kexec_prepare(struct kimage *image)
  34 {
  35         int i;
  36         unsigned long begin, end;       /* limits of segment */
  37         unsigned long low, high;        /* limits of blocked memory range */
  38         struct device_node *node;
  39         const unsigned long *basep;
  40         const unsigned int *sizep;
  41
  42         if (!ppc_md.hpte_clear_all)
  43                 return -ENOENT;
  44
  45         /*
  46          * Since we use the kernel fault handlers and paging code to
  47          * handle the virtual mode, we must make sure no destination
  48          * overlaps kernel static data or bss.
  49          */
  50         for (i = 0; i < image->nr_segments; i++)
  51                 if (image->segment[i].mem < __pa(_end))
  52                         return -ETXTBSY;
  53
  54         /*
  55          * For non-LPAR, we absolutely can not overwrite the mmu hash
  56          * table, since we are still using the bolted entries in it to
  57          * do the copy.  Check that here.
  58          *
  59          * It is safe if the end is below the start of the blocked
  60          * region (end <= low), or if the beginning is after the
  61          * end of the blocked region (begin >= high).  Use the
  62          * boolean identity !(a || b)  === (!a && !b).
  63          */
  64         if (htab_address) {
  65                 low = __pa(htab_address);
  66                 high = low + htab_size_bytes;
  67
  68                 for (i = 0; i < image->nr_segments; i++) {
  69                         begin = image->segment[i].mem;
  70                         end = begin + image->segment[i].memsz;
  71
  72                         if ((begin < high) && (end > low))
  73                                 return -ETXTBSY;
  74                 }
  75         }
  76
  77         /* We also should not overwrite the tce tables */
  78         for_each_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 static void copy_segments(unsigned long ind)
 100 {
 101         unsigned long entry;
 102         unsigned long *ptr;
 103         void *dest;
 104         void *addr;
 105
 106         /*
 107          * We rely on kexec_load to create a lists that properly
 108          * initializes these pointers before they are used.
 109          * We will still crash if the list is wrong, but at least
 110          * the compiler will be quiet.
 111          */
 112         ptr = NULL;
 113         dest = NULL;
 114
 115         for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
 116                 addr = __va(entry & PAGE_MASK);
 117
 118                 switch (entry & IND_FLAGS) {
 119                 case IND_DESTINATION:
 120                         dest = addr;
 121                         break;
 122                 case IND_INDIRECTION:
 123                         ptr = addr;
 124                         break;
 125                 case IND_SOURCE:
 126                         copy_page(dest, addr);
 127                         dest += PAGE_SIZE;
 128                 }
 129         }
 130 }
 131
 132 void kexec_copy_flush(struct kimage *image)
 133 {
 134         long i, nr_segments = image->nr_segments;
 135         struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
 136
 137         /* save the ranges on the stack to efficiently flush the icache */
 138         memcpy(ranges, image->segment, sizeof(ranges));
 139
 140         /*
 141          * After this call we may not use anything allocated in dynamic
 142          * memory, including *image.
 143          *
 144          * Only globals and the stack are allowed.
 145          */
 146         copy_segments(image->head);
 147
 148         /*
 149          * we need to clear the icache for all dest pages sometime,
 150          * including ones that were in place on the original copy
 151          */
 152         for (i = 0; i < nr_segments; i++)
 153                 flush_icache_range((unsigned long)__va(ranges[i].mem),
 154                         (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
 155 }
 156
 157 #ifdef CONFIG_SMP
 158
 159 static int kexec_all_irq_disabled = 0;
 160
 161 static void kexec_smp_down(void *arg)
 162 {
 163         local_irq_disable();
 164         hard_irq_disable();
 165
 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
 183 static 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         /* Make sure each CPU has at least made it to the state we need.
 190          *
 191          * FIXME: There is a (slim) chance of a problem if not all of the CPUs
 192          * are correctly onlined.  If somehow we start a CPU on boot with RTAS
 193          * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
 194          * time, the boot CPU will timeout.  If it does eventually execute
 195          * stuff, the secondary will start up (paca[].cpu_start was written) and
 196          * get into a peculiar state.  If the platform supports
 197          * smp_ops->take_timebase(), the secondary CPU will probably be spinning
 198          * in there.  If not (i.e. pseries), the secondary will continue on and
 199          * try to online itself/idle/etc. If it survives that, we need to find
 200          * these possible-but-not-online-but-should-be CPUs and chaperone them
 201          * into kexec_smp_wait().
 202          */
 203         for_each_online_cpu(i) {
 204                 if (i == my_cpu)
 205                         continue;
 206
 207                 while (paca[i].kexec_state < wait_state) {
 208                         barrier();
 209                         if (i != notified) {
 210                                 printk(KERN_INFO "kexec: waiting for cpu %d "
 211                                        "(physical %d) to enter %i state\n",
 212                                        i, paca[i].hw_cpu_id, wait_state);
 213                                 notified = i;
 214                         }
 215                 }
 216         }
 217         mb();
 218 }
 219
 220 /*
 221  * We need to make sure each present CPU is online.  The next kernel will scan
 222  * the device tree and assume primary threads are online and query secondary
 223  * threads via RTAS to online them if required.  If we don't online primary
 224  * threads, they will be stuck.  However, we also online secondary threads as we
 225  * may be using 'cede offline'.  In this case RTAS doesn't see the secondary
 226  * threads as offline -- and again, these CPUs will be stuck.
 227  *
 228  * So, we online all CPUs that should be running, including secondary threads.
 229  */
 230 static void wake_offline_cpus(void)
 231 {
 232         int cpu = 0;
 233
 234         for_each_present_cpu(cpu) {
 235                 if (!cpu_online(cpu)) {
 236                         printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
 237                                cpu);
 238                         WARN_ON(cpu_up(cpu));
 239                 }
 240         }
 241 }
 242
 243 static void kexec_prepare_cpus(void)
 244 {
 245         wake_offline_cpus();
 246         smp_call_function(kexec_smp_down, NULL, /* wait */0);
 247         local_irq_disable();
 248         hard_irq_disable();
 249
 250         mb(); /* make sure IRQs are disabled before we say they are */
 251         get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
 252
 253         kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
 254         /* we are sure every CPU has IRQs off at this point */
 255         kexec_all_irq_disabled = 1;
 256
 257         /* after we tell the others to go down */
 258         if (ppc_md.kexec_cpu_down)
 259                 ppc_md.kexec_cpu_down(0, 0);
 260
 261         /*
 262          * Before removing MMU mappings make sure all CPUs have entered real
 263          * mode:
 264          */
 265         kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
 266
 267         put_cpu();
 268 }
 269
 270 #else /* ! SMP */
 271
 272 static void kexec_prepare_cpus(void)
 273 {
 274         /*
 275          * move the secondarys to us so that we can copy
 276          * the new kernel 0-0x100 safely
 277          *
 278          * do this if kexec in setup.c ?
 279          *
 280          * We need to release the cpus if we are ever going from an
 281          * UP to an SMP kernel.
 282          */
 283         smp_release_cpus();
 284         if (ppc_md.kexec_cpu_down)
 285                 ppc_md.kexec_cpu_down(0, 0);
 286         local_irq_disable();
 287         hard_irq_disable();
 288 }
 289
 290 #endif /* SMP */
 291
 292 /*
 293  * kexec thread structure and stack.
 294  *
 295  * We need to make sure that this is 16384-byte aligned due to the
 296  * way process stacks are handled.  It also must be statically allocated
 297  * or allocated as part of the kimage, because everything else may be
 298  * overwritten when we copy the kexec image.  We piggyback on the
 299  * "init_task" linker section here to statically allocate a stack.
 300  *
 301  * We could use a smaller stack if we don't care about anything using
 302  * current, but that audit has not been performed.
 303  */
 304 static union thread_union kexec_stack __init_task_data =
 305         { };
 306
 307 /*
 308  * For similar reasons to the stack above, the kexecing CPU needs to be on a
 309  * static PACA; we switch to kexec_paca.
 310  */
 311 struct paca_struct kexec_paca;
 312
 313 /* Our assembly helper, in misc_64.S */
 314 extern void kexec_sequence(void *newstack, unsigned long start,
 315                            void *image, void *control,
 316                            void (*clear_all)(void)) __noreturn;
 317
 318 /* too late to fail here */
 319 void default_machine_kexec(struct kimage *image)
 320 {
 321         /* prepare control code if any */
 322
 323         /*
 324         * If the kexec boot is the normal one, need to shutdown other cpus
 325         * into our wait loop and quiesce interrupts.
 326         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
 327         * stopping other CPUs and collecting their pt_regs is done before
 328         * using debugger IPI.
 329         */
 330
 331         if (!kdump_in_progress())
 332                 kexec_prepare_cpus();
 333
 334         pr_debug("kexec: Starting switchover sequence.\n");
 335
 336         /* switch to a staticly allocated stack.  Based on irq stack code.
 337          * We setup preempt_count to avoid using VMX in memcpy.
 338          * XXX: the task struct will likely be invalid once we do the copy!
 339          */
 340         kexec_stack.thread_info.task = current_thread_info()->task;
 341         kexec_stack.thread_info.flags = 0;
 342         kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET;
 343         kexec_stack.thread_info.cpu = current_thread_info()->cpu;
 344
 345         /* We need a static PACA, too; copy this CPU's PACA over and switch to
 346          * it.  Also poison per_cpu_offset to catch anyone using non-static
 347          * data.
 348          */
 349         memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
 350         kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
 351         paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
 352                 kexec_paca.paca_index;
 353         setup_paca(&kexec_paca);
 354
 355         /* XXX: If anyone does 'dynamic lppacas' this will also need to be
 356          * switched to a static version!
 357          */
 358
 359         /* Some things are best done in assembly.  Finding globals with
 360          * a toc is easier in C, so pass in what we can.
 361          */
 362         kexec_sequence(&kexec_stack, image->start, image,
 363                         page_address(image->control_code_page),
 364                         ppc_md.hpte_clear_all);
 365         /* NOTREACHED */
 366 }
 367
 368 /* Values we need to export to the second kernel via the device tree. */
 369 static unsigned long htab_base;
 370 static unsigned long htab_size;
 371
 372 static struct property htab_base_prop = {
 373         .name = "linux,htab-base",
 374         .length = sizeof(unsigned long),
 375         .value = &htab_base,
 376 };
 377
 378 static struct property htab_size_prop = {
 379         .name = "linux,htab-size",
 380         .length = sizeof(unsigned long),
 381         .value = &htab_size,
 382 };
 383
 384 static int __init export_htab_values(void)
 385 {
 386         struct device_node *node;
 387         struct property *prop;
 388
 389         /* On machines with no htab htab_address is NULL */
 390         if (!htab_address)
 391                 return -ENODEV;
 392
 393         node = of_find_node_by_path("/chosen");
 394         if (!node)
 395                 return -ENODEV;
 396
 397         /* remove any stale propertys so ours can be found */
 398         prop = of_find_property(node, htab_base_prop.name, NULL);
 399         if (prop)
 400                 of_remove_property(node, prop);
 401         prop = of_find_property(node, htab_size_prop.name, NULL);
 402         if (prop)
 403                 of_remove_property(node, prop);
 404
 405         htab_base = cpu_to_be64(__pa(htab_address));
 406         of_add_property(node, &htab_base_prop);
 407         htab_size = cpu_to_be64(htab_size_bytes);
 408         of_add_property(node, &htab_size_prop);
 409
 410         of_node_put(node);
 411         return 0;
 412 }
 413 late_initcall(export_htab_values);