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         const unsigned long *basep;
  35         const 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 = of_get_property(node, "linux,tce-base", NULL);
  76                 sizep = of_get_property(node, "linux,tce-size", NULL);
  77                 if (basep == NULL || sizep == NULL)
  78                         continue;
  79
  80                 low = *basep;
  81                 high = low + (*sizep);
  82
  83                 for (i = 0; i < image->nr_segments; i++) {
  84                         begin = image->segment[i].mem;
  85                         end = begin + image->segment[i].memsz;
  86
  87                         if ((begin < high) && (end > low))
  88                                 return -ETXTBSY;
  89                 }
  90         }
  91
  92         return 0;
  93 }
  94
  95 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
  96
  97 static void copy_segments(unsigned long ind)
  98 {
  99         unsigned long entry;
 100         unsigned long *ptr;
 101         void *dest;
 102         void *addr;
 103
 104         /*
 105          * We rely on kexec_load to create a lists that properly
 106          * initializes these pointers before they are used.
 107          * We will still crash if the list is wrong, but at least
 108          * the compiler will be quiet.
 109          */
 110         ptr = NULL;
 111         dest = NULL;
 112
 113         for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
 114                 addr = __va(entry & PAGE_MASK);
 115
 116                 switch (entry & IND_FLAGS) {
 117                 case IND_DESTINATION:
 118                         dest = addr;
 119                         break;
 120                 case IND_INDIRECTION:
 121                         ptr = addr;
 122                         break;
 123                 case IND_SOURCE:
 124                         copy_page(dest, addr);
 125                         dest += PAGE_SIZE;
 126                 }
 127         }
 128 }
 129
 130 void kexec_copy_flush(struct kimage *image)
 131 {
 132         long i, nr_segments = image->nr_segments;
 133         struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
 134
 135         /* save the ranges on the stack to efficiently flush the icache */
 136         memcpy(ranges, image->segment, sizeof(ranges));
 137
 138         /*
 139          * After this call we may not use anything allocated in dynamic
 140          * memory, including *image.
 141          *
 142          * Only globals and the stack are allowed.
 143          */
 144         copy_segments(image->head);
 145
 146         /*
 147          * we need to clear the icache for all dest pages sometime,
 148          * including ones that were in place on the original copy
 149          */
 150         for (i = 0; i < nr_segments; i++)
 151                 flush_icache_range((unsigned long)__va(ranges[i].mem),
 152                         (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
 153 }
 154
 155 #ifdef CONFIG_SMP
 156
 157 /* FIXME: we should schedule this function to be called on all cpus based
 158  * on calling the interrupts, but we would like to call it off irq level
 159  * so that the interrupt controller is clean.
 160  */
 161 static void kexec_smp_down(void *arg)
 162 {
 163         if (ppc_md.kexec_cpu_down)
 164                 ppc_md.kexec_cpu_down(0, 1);
 165
 166         local_irq_disable();
 167         kexec_smp_wait();
 168         /* NOTREACHED */
 169 }
 170
 171 static void kexec_prepare_cpus(void)
 172 {
 173         int my_cpu, i, notified=-1;
 174
 175         smp_call_function(kexec_smp_down, NULL, /* wait */0);
 176         my_cpu = get_cpu();
 177
 178         /* check the others cpus are now down (via paca hw cpu id == -1) */
 179         for (i=0; i < NR_CPUS; i++) {
 180                 if (i == my_cpu)
 181                         continue;
 182
 183                 while (paca[i].hw_cpu_id != -1) {
 184                         barrier();
 185                         if (!cpu_possible(i)) {
 186                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 187                                                 " possible, ignoring\n",
 188                                                 i, paca[i].hw_cpu_id);
 189                                 break;
 190                         }
 191                         if (!cpu_online(i)) {
 192                                 /* Fixme: this can be spinning in
 193                                  * pSeries_secondary_wait with a paca
 194                                  * waiting for it to go online.
 195                                  */
 196                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 197                                                 " online, ignoring\n",
 198                                                 i, paca[i].hw_cpu_id);
 199                                 break;
 200                         }
 201                         if (i != notified) {
 202                                 printk( "kexec: waiting for cpu %d (physical"
 203                                                 " %d) to go down\n",
 204                                                 i, paca[i].hw_cpu_id);
 205                                 notified = i;
 206                         }
 207                 }
 208         }
 209
 210         /* after we tell the others to go down */
 211         if (ppc_md.kexec_cpu_down)
 212                 ppc_md.kexec_cpu_down(0, 0);
 213
 214         put_cpu();
 215
 216         local_irq_disable();
 217 }
 218
 219 #else /* ! SMP */
 220
 221 static void kexec_prepare_cpus(void)
 222 {
 223         /*
 224          * move the secondarys to us so that we can copy
 225          * the new kernel 0-0x100 safely
 226          *
 227          * do this if kexec in setup.c ?
 228          *
 229          * We need to release the cpus if we are ever going from an
 230          * UP to an SMP kernel.
 231          */
 232         smp_release_cpus();
 233         if (ppc_md.kexec_cpu_down)
 234                 ppc_md.kexec_cpu_down(0, 0);
 235         local_irq_disable();
 236 }
 237
 238 #endif /* SMP */
 239
 240 /*
 241  * kexec thread structure and stack.
 242  *
 243  * We need to make sure that this is 16384-byte aligned due to the
 244  * way process stacks are handled.  It also must be statically allocated
 245  * or allocated as part of the kimage, because everything else may be
 246  * overwritten when we copy the kexec image.  We piggyback on the
 247  * "init_task" linker section here to statically allocate a stack.
 248  *
 249  * We could use a smaller stack if we don't care about anything using
 250  * current, but that audit has not been performed.
 251  */
 252 static union thread_union kexec_stack
 253         __attribute__((__section__(".data.init_task"))) = { };
 254
 255 /* Our assembly helper, in kexec_stub.S */
 256 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
 257                                         void *image, void *control,
 258                                         void (*clear_all)(void)) ATTRIB_NORET;
 259
 260 /* too late to fail here */
 261 void default_machine_kexec(struct kimage *image)
 262 {
 263         /* prepare control code if any */
 264
 265         /*
 266         * If the kexec boot is the normal one, need to shutdown other cpus
 267         * into our wait loop and quiesce interrupts.
 268         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
 269         * stopping other CPUs and collecting their pt_regs is done before
 270         * using debugger IPI.
 271         */
 272
 273         if (crashing_cpu == -1)
 274                 kexec_prepare_cpus();
 275
 276         /* switch to a staticly allocated stack.  Based on irq stack code.
 277          * XXX: the task struct will likely be invalid once we do the copy!
 278          */
 279         kexec_stack.thread_info.task = current_thread_info()->task;
 280         kexec_stack.thread_info.flags = 0;
 281
 282         /* Some things are best done in assembly.  Finding globals with
 283          * a toc is easier in C, so pass in what we can.
 284          */
 285         kexec_sequence(&kexec_stack, image->start, image,
 286                         page_address(image->control_code_page),
 287                         ppc_md.hpte_clear_all);
 288         /* NOTREACHED */
 289 }
 290
 291 /* Values we need to export to the second kernel via the device tree. */
 292 static unsigned long htab_base;
 293
 294 static struct property htab_base_prop = {
 295         .name = "linux,htab-base",
 296         .length = sizeof(unsigned long),
 297         .value = &htab_base,
 298 };
 299
 300 static struct property htab_size_prop = {
 301         .name = "linux,htab-size",
 302         .length = sizeof(unsigned long),
 303         .value = &htab_size_bytes,
 304 };
 305
 306 static void __init export_htab_values(void)
 307 {
 308         struct device_node *node;
 309         struct property *prop;
 310
 311         /* On machines with no htab htab_address is NULL */
 312         if (!htab_address)
 313                 return;
 314
 315         node = of_find_node_by_path("/chosen");
 316         if (!node)
 317                 return;
 318
 319         /* remove any stale propertys so ours can be found */
 320         prop = of_find_property(node, htab_base_prop.name, NULL);
 321         if (prop)
 322                 prom_remove_property(node, prop);
 323         prop = of_find_property(node, htab_size_prop.name, NULL);
 324         if (prop)
 325                 prom_remove_property(node, prop);
 326
 327         htab_base = __pa(htab_address);
 328         prom_add_property(node, &htab_base_prop);
 329         prom_add_property(node, &htab_size_prop);
 330
 331         of_node_put(node);
 332 }
 333
 334 static struct property crashk_base_prop = {
 335         .name = "linux,crashkernel-base",
 336         .length = sizeof(unsigned long),
 337         .value = &crashk_res.start,
 338 };
 339
 340 static unsigned long crashk_size;
 341
 342 static struct property crashk_size_prop = {
 343         .name = "linux,crashkernel-size",
 344         .length = sizeof(unsigned long),
 345         .value = &crashk_size,
 346 };
 347
 348 static void __init export_crashk_values(void)
 349 {
 350         struct device_node *node;
 351         struct property *prop;
 352
 353         node = of_find_node_by_path("/chosen");
 354         if (!node)
 355                 return;
 356
 357         /* There might be existing crash kernel properties, but we can't
 358          * be sure what's in them, so remove them. */
 359         prop = of_find_property(node, "linux,crashkernel-base", NULL);
 360         if (prop)
 361                 prom_remove_property(node, prop);
 362
 363         prop = of_find_property(node, "linux,crashkernel-size", NULL);
 364         if (prop)
 365                 prom_remove_property(node, prop);
 366
 367         if (crashk_res.start != 0) {
 368                 prom_add_property(node, &crashk_base_prop);
 369                 crashk_size = crashk_res.end - crashk_res.start + 1;
 370                 prom_add_property(node, &crashk_size_prop);
 371         }
 372
 373         of_node_put(node);
 374 }
 375
 376 static int __init kexec_setup(void)
 377 {
 378         export_htab_values();
 379         export_crashk_values();
 380         return 0;
 381 }
 382 __initcall(kexec_setup);