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/cpumask.h>
  14 #include <linux/kexec.h>
  15 #include <linux/smp.h>
  16 #include <linux/thread_info.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
  29 #define HASH_GROUP_SIZE 0x80    /* size of each hash group, asm/mmu.h */
  30
  31 int default_machine_kexec_prepare(struct kimage *image)
  32 {
  33         int i;
  34         unsigned long begin, end;       /* limits of segment */
  35         unsigned long low, high;        /* limits of blocked memory range */
  36         struct device_node *node;
  37         unsigned long *basep;
  38         unsigned int *sizep;
  39
  40         if (!ppc_md.hpte_clear_all)
  41                 return -ENOENT;
  42
  43         /*
  44          * Since we use the kernel fault handlers and paging code to
  45          * handle the virtual mode, we must make sure no destination
  46          * overlaps kernel static data or bss.
  47          */
  48         for (i = 0; i < image->nr_segments; i++)
  49                 if (image->segment[i].mem < __pa(_end))
  50                         return -ETXTBSY;
  51
  52         /*
  53          * For non-LPAR, we absolutely can not overwrite the mmu hash
  54          * table, since we are still using the bolted entries in it to
  55          * do the copy.  Check that here.
  56          *
  57          * It is safe if the end is below the start of the blocked
  58          * region (end <= low), or if the beginning is after the
  59          * end of the blocked region (begin >= high).  Use the
  60          * boolean identity !(a || b)  === (!a && !b).
  61          */
  62         if (htab_address) {
  63                 low = __pa(htab_address);
  64                 high = low + (htab_hash_mask + 1) * HASH_GROUP_SIZE;
  65
  66                 for (i = 0; i < image->nr_segments; i++) {
  67                         begin = image->segment[i].mem;
  68                         end = begin + image->segment[i].memsz;
  69
  70                         if ((begin < high) && (end > low))
  71                                 return -ETXTBSY;
  72                 }
  73         }
  74
  75         /* We also should not overwrite the tce tables */
  76         for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
  77                         node = of_find_node_by_type(node, "pci")) {
  78                 basep = (unsigned long *)get_property(node, "linux,tce-base",
  79                                                         NULL);
  80                 sizep = (unsigned int *)get_property(node, "linux,tce-size",
  81                                                         NULL);
  82                 if (basep == NULL || sizep == NULL)
  83                         continue;
  84
  85                 low = *basep;
  86                 high = low + (*sizep);
  87
  88                 for (i = 0; i < image->nr_segments; i++) {
  89                         begin = image->segment[i].mem;
  90                         end = begin + image->segment[i].memsz;
  91
  92                         if ((begin < high) && (end > low))
  93                                 return -ETXTBSY;
  94                 }
  95         }
  96
  97         return 0;
  98 }
  99
 100 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
 101
 102 static void copy_segments(unsigned long ind)
 103 {
 104         unsigned long entry;
 105         unsigned long *ptr;
 106         void *dest;
 107         void *addr;
 108
 109         /*
 110          * We rely on kexec_load to create a lists that properly
 111          * initializes these pointers before they are used.
 112          * We will still crash if the list is wrong, but at least
 113          * the compiler will be quiet.
 114          */
 115         ptr = NULL;
 116         dest = NULL;
 117
 118         for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
 119                 addr = __va(entry & PAGE_MASK);
 120
 121                 switch (entry & IND_FLAGS) {
 122                 case IND_DESTINATION:
 123                         dest = addr;
 124                         break;
 125                 case IND_INDIRECTION:
 126                         ptr = addr;
 127                         break;
 128                 case IND_SOURCE:
 129                         copy_page(dest, addr);
 130                         dest += PAGE_SIZE;
 131                 }
 132         }
 133 }
 134
 135 void kexec_copy_flush(struct kimage *image)
 136 {
 137         long i, nr_segments = image->nr_segments;
 138         struct  kexec_segment ranges[KEXEC_SEGMENT_MAX];
 139
 140         /* save the ranges on the stack to efficiently flush the icache */
 141         memcpy(ranges, image->segment, sizeof(ranges));
 142
 143         /*
 144          * After this call we may not use anything allocated in dynamic
 145          * memory, including *image.
 146          *
 147          * Only globals and the stack are allowed.
 148          */
 149         copy_segments(image->head);
 150
 151         /*
 152          * we need to clear the icache for all dest pages sometime,
 153          * including ones that were in place on the original copy
 154          */
 155         for (i = 0; i < nr_segments; i++)
 156                 flush_icache_range((unsigned long)__va(ranges[i].mem),
 157                         (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
 158 }
 159
 160 #ifdef CONFIG_SMP
 161
 162 /* FIXME: we should schedule this function to be called on all cpus based
 163  * on calling the interrupts, but we would like to call it off irq level
 164  * so that the interrupt controller is clean.
 165  */
 166 void kexec_smp_down(void *arg)
 167 {
 168         if (ppc_md.kexec_cpu_down)
 169                 ppc_md.kexec_cpu_down(0, 1);
 170
 171         local_irq_disable();
 172         kexec_smp_wait();
 173         /* NOTREACHED */
 174 }
 175
 176 static void kexec_prepare_cpus(void)
 177 {
 178         int my_cpu, i, notified=-1;
 179
 180         smp_call_function(kexec_smp_down, NULL, 0, /* wait */0);
 181         my_cpu = get_cpu();
 182
 183         /* check the others cpus are now down (via paca hw cpu id == -1) */
 184         for (i=0; i < NR_CPUS; i++) {
 185                 if (i == my_cpu)
 186                         continue;
 187
 188                 while (paca[i].hw_cpu_id != -1) {
 189                         barrier();
 190                         if (!cpu_possible(i)) {
 191                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 192                                                 " possible, ignoring\n",
 193                                                 i, paca[i].hw_cpu_id);
 194                                 break;
 195                         }
 196                         if (!cpu_online(i)) {
 197                                 /* Fixme: this can be spinning in
 198                                  * pSeries_secondary_wait with a paca
 199                                  * waiting for it to go online.
 200                                  */
 201                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 202                                                 " online, ignoring\n",
 203                                                 i, paca[i].hw_cpu_id);
 204                                 break;
 205                         }
 206                         if (i != notified) {
 207                                 printk( "kexec: waiting for cpu %d (physical"
 208                                                 " %d) to go down\n",
 209                                                 i, paca[i].hw_cpu_id);
 210                                 notified = i;
 211                         }
 212                 }
 213         }
 214
 215         /* after we tell the others to go down */
 216         if (ppc_md.kexec_cpu_down)
 217                 ppc_md.kexec_cpu_down(0, 0);
 218
 219         put_cpu();
 220
 221         local_irq_disable();
 222 }
 223
 224 #else /* ! SMP */
 225
 226 static void kexec_prepare_cpus(void)
 227 {
 228         /*
 229          * move the secondarys to us so that we can copy
 230          * the new kernel 0-0x100 safely
 231          *
 232          * do this if kexec in setup.c ?
 233          *
 234          * We need to release the cpus if we are ever going from an
 235          * UP to an SMP kernel.
 236          */
 237         smp_release_cpus();
 238         if (ppc_md.kexec_cpu_down)
 239                 ppc_md.kexec_cpu_down(0, 0);
 240         local_irq_disable();
 241 }
 242
 243 #endif /* SMP */
 244
 245 /*
 246  * kexec thread structure and stack.
 247  *
 248  * We need to make sure that this is 16384-byte aligned due to the
 249  * way process stacks are handled.  It also must be statically allocated
 250  * or allocated as part of the kimage, because everything else may be
 251  * overwritten when we copy the kexec image.  We piggyback on the
 252  * "init_task" linker section here to statically allocate a stack.
 253  *
 254  * We could use a smaller stack if we don't care about anything using
 255  * current, but that audit has not been performed.
 256  */
 257 union thread_union kexec_stack
 258         __attribute__((__section__(".data.init_task"))) = { };
 259
 260 /* Our assembly helper, in kexec_stub.S */
 261 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
 262                                         void *image, void *control,
 263                                         void (*clear_all)(void)) ATTRIB_NORET;
 264
 265 /* too late to fail here */
 266 void default_machine_kexec(struct kimage *image)
 267 {
 268         /* prepare control code if any */
 269
 270         /*
 271         * If the kexec boot is the normal one, need to shutdown other cpus
 272         * into our wait loop and quiesce interrupts.
 273         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
 274         * stopping other CPUs and collecting their pt_regs is done before
 275         * using debugger IPI.
 276         */
 277
 278        if (crashing_cpu == -1)
 279                kexec_prepare_cpus();
 280
 281         /* switch to a staticly allocated stack.  Based on irq stack code.
 282          * XXX: the task struct will likely be invalid once we do the copy!
 283          */
 284         kexec_stack.thread_info.task = current_thread_info()->task;
 285         kexec_stack.thread_info.flags = 0;
 286
 287         /* Some things are best done in assembly.  Finding globals with
 288          * a toc is easier in C, so pass in what we can.
 289          */
 290         kexec_sequence(&kexec_stack, image->start, image,
 291                         page_address(image->control_code_page),
 292                         ppc_md.hpte_clear_all);
 293         /* NOTREACHED */
 294 }
 295
 296 /* Values we need to export to the second kernel via the device tree. */
 297 static unsigned long htab_base, htab_size, kernel_end;
 298
 299 static struct property htab_base_prop = {
 300         .name = "linux,htab-base",
 301         .length = sizeof(unsigned long),
 302         .value = (unsigned char *)&htab_base,
 303 };
 304
 305 static struct property htab_size_prop = {
 306         .name = "linux,htab-size",
 307         .length = sizeof(unsigned long),
 308         .value = (unsigned char *)&htab_size,
 309 };
 310
 311 static struct property kernel_end_prop = {
 312         .name = "linux,kernel-end",
 313         .length = sizeof(unsigned long),
 314         .value = (unsigned char *)&kernel_end,
 315 };
 316
 317 static void __init export_htab_values(void)
 318 {
 319         struct device_node *node;
 320
 321         node = of_find_node_by_path("/chosen");
 322         if (!node)
 323                 return;
 324
 325         kernel_end = __pa(_end);
 326         prom_add_property(node, &kernel_end_prop);
 327
 328         /* On machines with no htab htab_address is NULL */
 329         if (NULL == htab_address)
 330                 goto out;
 331
 332         htab_base = __pa(htab_address);
 333         prom_add_property(node, &htab_base_prop);
 334
 335         htab_size = 1UL << ppc64_pft_size;
 336         prom_add_property(node, &htab_size_prop);
 337
 338  out:
 339         of_node_put(node);
 340 }
 341
 342 void __init kexec_setup(void)
 343 {
 344         export_htab_values();
 345 }