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/errno.h>
  17
  18 #include <asm/page.h>
  19 #include <asm/current.h>
  20 #include <asm/machdep.h>
  21 #include <asm/cacheflush.h>
  22 #include <asm/paca.h>
  23 #include <asm/mmu.h>
  24 #include <asm/sections.h>       /* _end */
  25 #include <asm/prom.h>
  26 #include <asm/smp.h>
  27
  28 int default_machine_kexec_prepare(struct kimage *image)
  29 {
  30         int i;
  31         unsigned long begin, end;       /* limits of segment */
  32         unsigned long low, high;        /* limits of blocked memory range */
  33         struct device_node *node;
  34         unsigned long *basep;
  35         unsigned int *sizep;
  36
  37         if (!ppc_md.hpte_clear_all)
  38                 return -ENOENT;
  39
  40         /*
  41          * Since we use the kernel fault handlers and paging code to
  42          * handle the virtual mode, we must make sure no destination
  43          * overlaps kernel static data or bss.
  44          */
  45         for (i = 0; i < image->nr_segments; i++)
  46                 if (image->segment[i].mem < __pa(_end))
  47                         return -ETXTBSY;
  48
  49         /*
  50          * For non-LPAR, we absolutely can not overwrite the mmu hash
  51          * table, since we are still using the bolted entries in it to
  52          * do the copy.  Check that here.
  53          *
  54          * It is safe if the end is below the start of the blocked
  55          * region (end <= low), or if the beginning is after the
  56          * end of the blocked region (begin >= high).  Use the
  57          * boolean identity !(a || b)  === (!a && !b).
  58          */
  59         if (htab_address) {
  60                 low = __pa(htab_address);
  61                 high = low + htab_size_bytes;
  62
  63                 for (i = 0; i < image->nr_segments; i++) {
  64                         begin = image->segment[i].mem;
  65                         end = begin + image->segment[i].memsz;
  66
  67                         if ((begin < high) && (end > low))
  68                                 return -ETXTBSY;
  69                 }
  70         }
  71
  72         /* We also should not overwrite the tce tables */
  73         for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
  74                         node = of_find_node_by_type(node, "pci")) {
  75                 basep = (unsigned long *)get_property(node, "linux,tce-base",
  76                                                         NULL);
  77                 sizep = (unsigned int *)get_property(node, "linux,tce-size",
  78                                                         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 /* FIXME: we should schedule this function to be called on all cpus based
 160  * on calling the interrupts, but we would like to call it off irq level
 161  * so that the interrupt controller is clean.
 162  */
 163 void kexec_smp_down(void *arg)
 164 {
 165         if (ppc_md.kexec_cpu_down)
 166                 ppc_md.kexec_cpu_down(0, 1);
 167
 168         local_irq_disable();
 169         kexec_smp_wait();
 170         /* NOTREACHED */
 171 }
 172
 173 static void kexec_prepare_cpus(void)
 174 {
 175         int my_cpu, i, notified=-1;
 176
 177         smp_call_function(kexec_smp_down, NULL, 0, /* wait */0);
 178         my_cpu = get_cpu();
 179
 180         /* check the others cpus are now down (via paca hw cpu id == -1) */
 181         for (i=0; i < NR_CPUS; i++) {
 182                 if (i == my_cpu)
 183                         continue;
 184
 185                 while (paca[i].hw_cpu_id != -1) {
 186                         barrier();
 187                         if (!cpu_possible(i)) {
 188                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 189                                                 " possible, ignoring\n",
 190                                                 i, paca[i].hw_cpu_id);
 191                                 break;
 192                         }
 193                         if (!cpu_online(i)) {
 194                                 /* Fixme: this can be spinning in
 195                                  * pSeries_secondary_wait with a paca
 196                                  * waiting for it to go online.
 197                                  */
 198                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 199                                                 " online, ignoring\n",
 200                                                 i, paca[i].hw_cpu_id);
 201                                 break;
 202                         }
 203                         if (i != notified) {
 204                                 printk( "kexec: waiting for cpu %d (physical"
 205                                                 " %d) to go down\n",
 206                                                 i, paca[i].hw_cpu_id);
 207                                 notified = i;
 208                         }
 209                 }
 210         }
 211
 212         /* after we tell the others to go down */
 213         if (ppc_md.kexec_cpu_down)
 214                 ppc_md.kexec_cpu_down(0, 0);
 215
 216         put_cpu();
 217
 218         local_irq_disable();
 219 }
 220
 221 #else /* ! SMP */
 222
 223 static void kexec_prepare_cpus(void)
 224 {
 225         /*
 226          * move the secondarys to us so that we can copy
 227          * the new kernel 0-0x100 safely
 228          *
 229          * do this if kexec in setup.c ?
 230          *
 231          * We need to release the cpus if we are ever going from an
 232          * UP to an SMP kernel.
 233          */
 234         smp_release_cpus();
 235         if (ppc_md.kexec_cpu_down)
 236                 ppc_md.kexec_cpu_down(0, 0);
 237         local_irq_disable();
 238 }
 239
 240 #endif /* SMP */
 241
 242 /*
 243  * kexec thread structure and stack.
 244  *
 245  * We need to make sure that this is 16384-byte aligned due to the
 246  * way process stacks are handled.  It also must be statically allocated
 247  * or allocated as part of the kimage, because everything else may be
 248  * overwritten when we copy the kexec image.  We piggyback on the
 249  * "init_task" linker section here to statically allocate a stack.
 250  *
 251  * We could use a smaller stack if we don't care about anything using
 252  * current, but that audit has not been performed.
 253  */
 254 union thread_union kexec_stack
 255         __attribute__((__section__(".data.init_task"))) = { };
 256
 257 /* Our assembly helper, in kexec_stub.S */
 258 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
 259                                         void *image, void *control,
 260                                         void (*clear_all)(void)) ATTRIB_NORET;
 261
 262 /* too late to fail here */
 263 void default_machine_kexec(struct kimage *image)
 264 {
 265         /* prepare control code if any */
 266
 267         /*
 268         * If the kexec boot is the normal one, need to shutdown other cpus
 269         * into our wait loop and quiesce interrupts.
 270         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
 271         * stopping other CPUs and collecting their pt_regs is done before
 272         * using debugger IPI.
 273         */
 274
 275        if (crashing_cpu == -1)
 276                kexec_prepare_cpus();
 277
 278         /* switch to a staticly allocated stack.  Based on irq stack code.
 279          * XXX: the task struct will likely be invalid once we do the copy!
 280          */
 281         kexec_stack.thread_info.task = current_thread_info()->task;
 282         kexec_stack.thread_info.flags = 0;
 283
 284         /* Some things are best done in assembly.  Finding globals with
 285          * a toc is easier in C, so pass in what we can.
 286          */
 287         kexec_sequence(&kexec_stack, image->start, image,
 288                         page_address(image->control_code_page),
 289                         ppc_md.hpte_clear_all);
 290         /* NOTREACHED */
 291 }
 292
 293 /* Values we need to export to the second kernel via the device tree. */
 294 static unsigned long htab_base, kernel_end;
 295
 296 static struct property htab_base_prop = {
 297         .name = "linux,htab-base",
 298         .length = sizeof(unsigned long),
 299         .value = (unsigned char *)&htab_base,
 300 };
 301
 302 static struct property htab_size_prop = {
 303         .name = "linux,htab-size",
 304         .length = sizeof(unsigned long),
 305         .value = (unsigned char *)&htab_size_bytes,
 306 };
 307
 308 static struct property kernel_end_prop = {
 309         .name = "linux,kernel-end",
 310         .length = sizeof(unsigned long),
 311         .value = (unsigned char *)&kernel_end,
 312 };
 313
 314 static void __init export_htab_values(void)
 315 {
 316         struct device_node *node;
 317
 318         node = of_find_node_by_path("/chosen");
 319         if (!node)
 320                 return;
 321
 322         kernel_end = __pa(_end);
 323         prom_add_property(node, &kernel_end_prop);
 324
 325         /* On machines with no htab htab_address is NULL */
 326         if (NULL == htab_address)
 327                 goto out;
 328
 329         htab_base = __pa(htab_address);
 330         prom_add_property(node, &htab_base_prop);
 331         prom_add_property(node, &htab_size_prop);
 332
 333  out:
 334         of_node_put(node);
 335 }
 336
 337 static struct property crashk_base_prop = {
 338         .name = "linux,crashkernel-base",
 339         .length = sizeof(unsigned long),
 340         .value = (unsigned char *)&crashk_res.start,
 341 };
 342
 343 static unsigned long crashk_size;
 344
 345 static struct property crashk_size_prop = {
 346         .name = "linux,crashkernel-size",
 347         .length = sizeof(unsigned long),
 348         .value = (unsigned char *)&crashk_size,
 349 };
 350
 351 static void __init export_crashk_values(void)
 352 {
 353         struct device_node *node;
 354         struct property *prop;
 355
 356         node = of_find_node_by_path("/chosen");
 357         if (!node)
 358                 return;
 359
 360         /* There might be existing crash kernel properties, but we can't
 361          * be sure what's in them, so remove them. */
 362         prop = of_find_property(node, "linux,crashkernel-base", NULL);
 363         if (prop)
 364                 prom_remove_property(node, prop);
 365
 366         prop = of_find_property(node, "linux,crashkernel-size", NULL);
 367         if (prop)
 368                 prom_remove_property(node, prop);
 369
 370         if (crashk_res.start != 0) {
 371                 prom_add_property(node, &crashk_base_prop);
 372                 crashk_size = crashk_res.end - crashk_res.start + 1;
 373                 prom_add_property(node, &crashk_size_prop);
 374         }
 375
 376         of_node_put(node);
 377 }
 378
 379 static int __init kexec_setup(void)
 380 {
 381         export_htab_values();
 382         export_crashk_values();
 383         return 0;
 384 }
 385 __initcall(kexec_setup);