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