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
  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
  30 int default_machine_kexec_prepare(struct kimage *image)
  31 {
  32         int i;
  33         unsigned long begin, end;       /* limits of segment */
  34         unsigned long low, high;        /* limits of blocked memory range */
  35         struct device_node *node;
  36         const unsigned long *basep;
  37         const unsigned int *sizep;
  38
  39         if (!ppc_md.hpte_clear_all)
  40                 return -ENOENT;
  41
  42         /*
  43          * Since we use the kernel fault handlers and paging code to
  44          * handle the virtual mode, we must make sure no destination
  45          * overlaps kernel static data or bss.
  46          */
  47         for (i = 0; i < image->nr_segments; i++)
  48                 if (image->segment[i].mem < __pa(_end))
  49                         return -ETXTBSY;
  50
  51         /*
  52          * For non-LPAR, we absolutely can not overwrite the mmu hash
  53          * table, since we are still using the bolted entries in it to
  54          * do the copy.  Check that here.
  55          *
  56          * It is safe if the end is below the start of the blocked
  57          * region (end <= low), or if the beginning is after the
  58          * end of the blocked region (begin >= high).  Use the
  59          * boolean identity !(a || b)  === (!a && !b).
  60          */
  61         if (htab_address) {
  62                 low = __pa(htab_address);
  63                 high = low + htab_size_bytes;
  64
  65                 for (i = 0; i < image->nr_segments; i++) {
  66                         begin = image->segment[i].mem;
  67                         end = begin + image->segment[i].memsz;
  68
  69                         if ((begin < high) && (end > low))
  70                                 return -ETXTBSY;
  71                 }
  72         }
  73
  74         /* We also should not overwrite the tce tables */
  75         for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
  76                         node = of_find_node_by_type(node, "pci")) {
  77                 basep = of_get_property(node, "linux,tce-base", NULL);
  78                 sizep = of_get_property(node, "linux,tce-size", NULL);
  79                 if (basep == NULL || sizep == NULL)
  80                         continue;
  81
  82                 low = *basep;
  83                 high = low + (*sizep);
  84
  85                 for (i = 0; i < image->nr_segments; i++) {
  86                         begin = image->segment[i].mem;
  87                         end = begin + image->segment[i].memsz;
  88
  89                         if ((begin < high) && (end > low))
  90                                 return -ETXTBSY;
  91                 }
  92         }
  93
  94         return 0;
  95 }
  96
  97 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
  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         mb(); /* make sure our irqs are disabled before we say they are */
 165         get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
 166         while(kexec_all_irq_disabled == 0)
 167                 cpu_relax();
 168         mb(); /* make sure all irqs are disabled before this */
 169         hw_breakpoint_disable();
 170         /*
 171          * Now every CPU has IRQs off, we can clear out any pending
 172          * IPIs and be sure that no more will come in after this.
 173          */
 174         if (ppc_md.kexec_cpu_down)
 175                 ppc_md.kexec_cpu_down(0, 1);
 176
 177         kexec_smp_wait();
 178         /* NOTREACHED */
 179 }
 180
 181 static void kexec_prepare_cpus_wait(int wait_state)
 182 {
 183         int my_cpu, i, notified=-1;
 184
 185         hw_breakpoint_disable();
 186         my_cpu = get_cpu();
 187         /* Make sure each CPU has atleast made it to the state we need */
 188         for_each_online_cpu(i) {
 189                 if (i == my_cpu)
 190                         continue;
 191
 192                 while (paca[i].kexec_state < wait_state) {
 193                         barrier();
 194                         if (i != notified) {
 195                                 printk( "kexec: waiting for cpu %d (physical"
 196                                                 " %d) to enter %i state\n",
 197                                         i, paca[i].hw_cpu_id, wait_state);
 198                                 notified = i;
 199                         }
 200                 }
 201         }
 202         mb();
 203 }
 204
 205 static void kexec_prepare_cpus(void)
 206 {
 207
 208         smp_call_function(kexec_smp_down, NULL, /* wait */0);
 209         local_irq_disable();
 210         mb(); /* make sure IRQs are disabled before we say they are */
 211         get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
 212
 213         kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
 214         /* we are sure every CPU has IRQs off at this point */
 215         kexec_all_irq_disabled = 1;
 216
 217         /* after we tell the others to go down */
 218         if (ppc_md.kexec_cpu_down)
 219                 ppc_md.kexec_cpu_down(0, 0);
 220
 221         /* Before removing MMU mapings make sure all CPUs have entered real mode */
 222         kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
 223
 224         put_cpu();
 225 }
 226
 227 #else /* ! SMP */
 228
 229 static void kexec_prepare_cpus(void)
 230 {
 231         /*
 232          * move the secondarys to us so that we can copy
 233          * the new kernel 0-0x100 safely
 234          *
 235          * do this if kexec in setup.c ?
 236          *
 237          * We need to release the cpus if we are ever going from an
 238          * UP to an SMP kernel.
 239          */
 240         smp_release_cpus();
 241         if (ppc_md.kexec_cpu_down)
 242                 ppc_md.kexec_cpu_down(0, 0);
 243         local_irq_disable();
 244 }
 245
 246 #endif /* SMP */
 247
 248 /*
 249  * kexec thread structure and stack.
 250  *
 251  * We need to make sure that this is 16384-byte aligned due to the
 252  * way process stacks are handled.  It also must be statically allocated
 253  * or allocated as part of the kimage, because everything else may be
 254  * overwritten when we copy the kexec image.  We piggyback on the
 255  * "init_task" linker section here to statically allocate a stack.
 256  *
 257  * We could use a smaller stack if we don't care about anything using
 258  * current, but that audit has not been performed.
 259  */
 260 static union thread_union kexec_stack __init_task_data =
 261         { };
 262
 263 /*
 264  * For similar reasons to the stack above, the kexecing CPU needs to be on a
 265  * static PACA; we switch to kexec_paca.
 266  */
 267 struct paca_struct kexec_paca;
 268
 269 /* Our assembly helper, in kexec_stub.S */
 270 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
 271                                         void *image, void *control,
 272                                         void (*clear_all)(void)) ATTRIB_NORET;
 273
 274 /* too late to fail here */
 275 void default_machine_kexec(struct kimage *image)
 276 {
 277         /* prepare control code if any */
 278
 279         /*
 280         * If the kexec boot is the normal one, need to shutdown other cpus
 281         * into our wait loop and quiesce interrupts.
 282         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
 283         * stopping other CPUs and collecting their pt_regs is done before
 284         * using debugger IPI.
 285         */
 286
 287         if (crashing_cpu == -1)
 288                 kexec_prepare_cpus();
 289
 290         /* switch to a staticly allocated stack.  Based on irq stack code.
 291          * XXX: the task struct will likely be invalid once we do the copy!
 292          */
 293         kexec_stack.thread_info.task = current_thread_info()->task;
 294         kexec_stack.thread_info.flags = 0;
 295
 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
 306         /* XXX: If anyone does 'dynamic lppacas' this will also need to be
 307          * switched to a static version!
 308          */
 309
 310         /* Some things are best done in assembly.  Finding globals with
 311          * a toc is easier in C, so pass in what we can.
 312          */
 313         kexec_sequence(&kexec_stack, image->start, image,
 314                         page_address(image->control_code_page),
 315                         ppc_md.hpte_clear_all);
 316         /* NOTREACHED */
 317 }
 318
 319 /* Values we need to export to the second kernel via the device tree. */
 320 static unsigned long htab_base;
 321
 322 static struct property htab_base_prop = {
 323         .name = "linux,htab-base",
 324         .length = sizeof(unsigned long),
 325         .value = &htab_base,
 326 };
 327
 328 static struct property htab_size_prop = {
 329         .name = "linux,htab-size",
 330         .length = sizeof(unsigned long),
 331         .value = &htab_size_bytes,
 332 };
 333
 334 static int __init export_htab_values(void)
 335 {
 336         struct device_node *node;
 337         struct property *prop;
 338
 339         /* On machines with no htab htab_address is NULL */
 340         if (!htab_address)
 341                 return -ENODEV;
 342
 343         node = of_find_node_by_path("/chosen");
 344         if (!node)
 345                 return -ENODEV;
 346
 347         /* remove any stale propertys so ours can be found */
 348         prop = of_find_property(node, htab_base_prop.name, NULL);
 349         if (prop)
 350                 prom_remove_property(node, prop);
 351         prop = of_find_property(node, htab_size_prop.name, NULL);
 352         if (prop)
 353                 prom_remove_property(node, prop);
 354
 355         htab_base = __pa(htab_address);
 356         prom_add_property(node, &htab_base_prop);
 357         prom_add_property(node, &htab_size_prop);
 358
 359         of_node_put(node);
 360         return 0;
 361 }
 362 late_initcall(export_htab_values);