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 static int kexec_all_irq_disabled = 0;
 159
 160 static void kexec_smp_down(void *arg)
 161 {
 162         local_irq_disable();
 163         mb(); /* make sure our irqs are disabled before we say they are */
 164         get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
 165         while(kexec_all_irq_disabled == 0)
 166                 cpu_relax();
 167         mb(); /* make sure all irqs are disabled before this */
 168         /*
 169          * Now every CPU has IRQs off, we can clear out any pending
 170          * IPIs and be sure that no more will come in after this.
 171          */
 172         if (ppc_md.kexec_cpu_down)
 173                 ppc_md.kexec_cpu_down(0, 1);
 174
 175         kexec_smp_wait();
 176         /* NOTREACHED */
 177 }
 178
 179 static void kexec_prepare_cpus_wait(int wait_state)
 180 {
 181         int my_cpu, i, notified=-1;
 182
 183         my_cpu = get_cpu();
 184         /* Make sure each CPU has atleast made it to the state we need */
 185         for (i=0; i < NR_CPUS; i++) {
 186                 if (i == my_cpu)
 187                         continue;
 188
 189                 while (paca[i].kexec_state < wait_state) {
 190                         barrier();
 191                         if (!cpu_possible(i)) {
 192                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 193                                                 " possible, ignoring\n",
 194                                                 i, paca[i].hw_cpu_id);
 195                                 break;
 196                         }
 197                         if (!cpu_online(i)) {
 198                                 /* Fixme: this can be spinning in
 199                                  * pSeries_secondary_wait with a paca
 200                                  * waiting for it to go online.
 201                                  */
 202                                 printk("kexec: cpu %d hw_cpu_id %d is not"
 203                                                 " online, ignoring\n",
 204                                                 i, paca[i].hw_cpu_id);
 205                                 break;
 206                         }
 207                         if (i != notified) {
 208                                 printk( "kexec: waiting for cpu %d (physical"
 209                                                 " %d) to enter %i state\n",
 210                                         i, paca[i].hw_cpu_id, wait_state);
 211                                 notified = i;
 212                         }
 213                 }
 214         }
 215         mb();
 216 }
 217
 218 static void kexec_prepare_cpus(void)
 219 {
 220
 221         smp_call_function(kexec_smp_down, NULL, /* wait */0);
 222         local_irq_disable();
 223         mb(); /* make sure IRQs are disabled before we say they are */
 224         get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
 225
 226         kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
 227         /* we are sure every CPU has IRQs off at this point */
 228         kexec_all_irq_disabled = 1;
 229
 230         /* after we tell the others to go down */
 231         if (ppc_md.kexec_cpu_down)
 232                 ppc_md.kexec_cpu_down(0, 0);
 233
 234         /* Before removing MMU mapings make sure all CPUs have entered real mode */
 235         kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
 236
 237         put_cpu();
 238 }
 239
 240 #else /* ! SMP */
 241
 242 static void kexec_prepare_cpus(void)
 243 {
 244         /*
 245          * move the secondarys to us so that we can copy
 246          * the new kernel 0-0x100 safely
 247          *
 248          * do this if kexec in setup.c ?
 249          *
 250          * We need to release the cpus if we are ever going from an
 251          * UP to an SMP kernel.
 252          */
 253         smp_release_cpus();
 254         if (ppc_md.kexec_cpu_down)
 255                 ppc_md.kexec_cpu_down(0, 0);
 256         local_irq_disable();
 257 }
 258
 259 #endif /* SMP */
 260
 261 /*
 262  * kexec thread structure and stack.
 263  *
 264  * We need to make sure that this is 16384-byte aligned due to the
 265  * way process stacks are handled.  It also must be statically allocated
 266  * or allocated as part of the kimage, because everything else may be
 267  * overwritten when we copy the kexec image.  We piggyback on the
 268  * "init_task" linker section here to statically allocate a stack.
 269  *
 270  * We could use a smaller stack if we don't care about anything using
 271  * current, but that audit has not been performed.
 272  */
 273 static union thread_union kexec_stack __init_task_data =
 274         { };
 275
 276 /* Our assembly helper, in kexec_stub.S */
 277 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
 278                                         void *image, void *control,
 279                                         void (*clear_all)(void)) ATTRIB_NORET;
 280
 281 /* too late to fail here */
 282 void default_machine_kexec(struct kimage *image)
 283 {
 284         /* prepare control code if any */
 285
 286         /*
 287         * If the kexec boot is the normal one, need to shutdown other cpus
 288         * into our wait loop and quiesce interrupts.
 289         * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
 290         * stopping other CPUs and collecting their pt_regs is done before
 291         * using debugger IPI.
 292         */
 293
 294         if (crashing_cpu == -1)
 295                 kexec_prepare_cpus();
 296
 297         /* switch to a staticly allocated stack.  Based on irq stack code.
 298          * XXX: the task struct will likely be invalid once we do the copy!
 299          */
 300         kexec_stack.thread_info.task = current_thread_info()->task;
 301         kexec_stack.thread_info.flags = 0;
 302
 303         /* Some things are best done in assembly.  Finding globals with
 304          * a toc is easier in C, so pass in what we can.
 305          */
 306         kexec_sequence(&kexec_stack, image->start, image,
 307                         page_address(image->control_code_page),
 308                         ppc_md.hpte_clear_all);
 309         /* NOTREACHED */
 310 }
 311
 312 /* Values we need to export to the second kernel via the device tree. */
 313 static unsigned long htab_base;
 314
 315 static struct property htab_base_prop = {
 316         .name = "linux,htab-base",
 317         .length = sizeof(unsigned long),
 318         .value = &htab_base,
 319 };
 320
 321 static struct property htab_size_prop = {
 322         .name = "linux,htab-size",
 323         .length = sizeof(unsigned long),
 324         .value = &htab_size_bytes,
 325 };
 326
 327 static int __init export_htab_values(void)
 328 {
 329         struct device_node *node;
 330         struct property *prop;
 331
 332         /* On machines with no htab htab_address is NULL */
 333         if (!htab_address)
 334                 return -ENODEV;
 335
 336         node = of_find_node_by_path("/chosen");
 337         if (!node)
 338                 return -ENODEV;
 339
 340         /* remove any stale propertys so ours can be found */
 341         prop = of_find_property(node, htab_base_prop.name, NULL);
 342         if (prop)
 343                 prom_remove_property(node, prop);
 344         prop = of_find_property(node, htab_size_prop.name, NULL);
 345         if (prop)
 346                 prom_remove_property(node, prop);
 347
 348         htab_base = __pa(htab_address);
 349         prom_add_property(node, &htab_base_prop);
 350         prom_add_property(node, &htab_size_prop);
 351
 352         of_node_put(node);
 353         return 0;
 354 }
 355 late_initcall(export_htab_values);