arch/tile/kernel/machine_kexec.c

   1 /*
   2  * Copyright 2010 Tilera Corporation. All Rights Reserved.
   3  *
   4  *   This program is free software; you can redistribute it and/or
   5  *   modify it under the terms of the GNU General Public License
   6  *   as published by the Free Software Foundation, version 2.
   7  *
   8  *   This program is distributed in the hope that it will be useful, but
   9  *   WITHOUT ANY WARRANTY; without even the implied warranty of
  10  *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  11  *   NON INFRINGEMENT.  See the GNU General Public License for
  12  *   more details.
  13  *
  14  * based on machine_kexec.c from other architectures in linux-2.6.18
  15  */
  16
  17 #include <linux/mm.h>
  18 #include <linux/kexec.h>
  19 #include <linux/delay.h>
  20 #include <linux/reboot.h>
  21 #include <linux/errno.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/cpumask.h>
  24 #include <linux/kernel.h>
  25 #include <linux/elf.h>
  26 #include <linux/highmem.h>
  27 #include <linux/mmu_context.h>
  28 #include <linux/io.h>
  29 #include <linux/timex.h>
  30 #include <asm/pgtable.h>
  31 #include <asm/pgalloc.h>
  32 #include <asm/cacheflush.h>
  33 #include <asm/checksum.h>
  34 #include <hv/hypervisor.h>
  35
  36
  37 /*
  38  * This stuff is not in elf.h and is not in any other kernel include.
  39  * This stuff is needed below in the little boot notes parser to
  40  * extract the command line so we can pass it to the hypervisor.
  41  */
  42 struct Elf32_Bhdr {
  43         Elf32_Word b_signature;
  44         Elf32_Word b_size;
  45         Elf32_Half b_checksum;
  46         Elf32_Half b_records;
  47 };
  48 #define ELF_BOOT_MAGIC          0x0E1FB007
  49 #define EBN_COMMAND_LINE        0x00000004
  50 #define roundupsz(X) (((X) + 3) & ~3)
  51
  52 /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
  53
  54
  55 void machine_shutdown(void)
  56 {
  57         /*
  58          * Normally we would stop all the other processors here, but
  59          * the check in machine_kexec_prepare below ensures we'll only
  60          * get this far if we've been booted with "nosmp" on the
  61          * command line or without CONFIG_SMP so there's nothing to do
  62          * here (for now).
  63          */
  64 }
  65
  66 void machine_crash_shutdown(struct pt_regs *regs)
  67 {
  68         /*
  69          * Cannot happen.  This type of kexec is disabled on this
  70          * architecture (and enforced in machine_kexec_prepare below).
  71          */
  72 }
  73
  74
  75 int machine_kexec_prepare(struct kimage *image)
  76 {
  77         if (num_online_cpus() > 1) {
  78                 pr_warning("%s: detected attempt to kexec "
  79                        "with num_online_cpus() > 1\n",
  80                        __func__);
  81                 return -ENOSYS;
  82         }
  83         if (image->type != KEXEC_TYPE_DEFAULT) {
  84                 pr_warning("%s: detected attempt to kexec "
  85                        "with unsupported type: %d\n",
  86                        __func__,
  87                        image->type);
  88                 return -ENOSYS;
  89         }
  90         return 0;
  91 }
  92
  93 void machine_kexec_cleanup(struct kimage *image)
  94 {
  95         /*
  96          * We did nothing in machine_kexec_prepare,
  97          * so we have nothing to do here.
  98          */
  99 }
 100
 101 /*
 102  * If we can find elf boot notes on this page, return the command
 103  * line.  Otherwise, silently return null.  Somewhat kludgy, but no
 104  * good way to do this without significantly rearchitecting the
 105  * architecture-independent kexec code.
 106  */
 107
 108 static unsigned char *kexec_bn2cl(void *pg)
 109 {
 110         struct Elf32_Bhdr *bhdrp;
 111         Elf32_Nhdr *nhdrp;
 112         unsigned char *desc;
 113         unsigned char *command_line;
 114         __sum16 csum;
 115
 116         bhdrp = (struct Elf32_Bhdr *) pg;
 117
 118         /*
 119          * This routine is invoked for every source page, so make
 120          * sure to quietly ignore every impossible page.
 121          */
 122         if (bhdrp->b_signature != ELF_BOOT_MAGIC ||
 123             bhdrp->b_size > PAGE_SIZE)
 124                 return 0;
 125
 126         /*
 127          * If we get a checksum mismatch, warn with the checksum
 128          * so we can diagnose better.
 129          */
 130         csum = ip_compute_csum(pg, bhdrp->b_size);
 131         if (csum != 0) {
 132                 pr_warning("%s: bad checksum %#x (size %d)\n",
 133                            __func__, csum, bhdrp->b_size);
 134                 return 0;
 135         }
 136
 137         nhdrp = (Elf32_Nhdr *) (bhdrp + 1);
 138
 139         while (nhdrp->n_type != EBN_COMMAND_LINE) {
 140
 141                 desc = (unsigned char *) (nhdrp + 1);
 142                 desc += roundupsz(nhdrp->n_descsz);
 143
 144                 nhdrp = (Elf32_Nhdr *) desc;
 145
 146                 /* still in bounds? */
 147                 if ((unsigned char *) (nhdrp + 1) >
 148                     ((unsigned char *) pg) + bhdrp->b_size) {
 149
 150                         pr_info("%s: out of bounds\n", __func__);
 151                         return 0;
 152                 }
 153         }
 154
 155         command_line = (unsigned char *) (nhdrp + 1);
 156         desc = command_line;
 157
 158         while (*desc != '\0') {
 159                 desc++;
 160                 if (((unsigned long)desc & PAGE_MASK) != (unsigned long)pg) {
 161                         pr_info("%s: ran off end of page\n",
 162                                __func__);
 163                         return 0;
 164                 }
 165         }
 166
 167         return command_line;
 168 }
 169
 170 static void kexec_find_and_set_command_line(struct kimage *image)
 171 {
 172         kimage_entry_t *ptr, entry;
 173
 174         unsigned char *command_line = 0;
 175         unsigned char *r;
 176         HV_Errno hverr;
 177
 178         for (ptr = &image->head;
 179              (entry = *ptr) && !(entry & IND_DONE);
 180              ptr = (entry & IND_INDIRECTION) ?
 181                      phys_to_virt((entry & PAGE_MASK)) : ptr + 1) {
 182
 183                 if ((entry & IND_SOURCE)) {
 184                         void *va =
 185                                 kmap_atomic_pfn(entry >> PAGE_SHIFT);
 186                         r = kexec_bn2cl(va);
 187                         if (r) {
 188                                 command_line = r;
 189                                 break;
 190                         }
 191                         kunmap_atomic(va);
 192                 }
 193         }
 194
 195         if (command_line != 0) {
 196                 pr_info("setting new command line to \"%s\"\n",
 197                        command_line);
 198
 199                 hverr = hv_set_command_line(
 200                         (HV_VirtAddr) command_line, strlen(command_line));
 201                 kunmap_atomic(command_line);
 202         } else {
 203                 pr_info("%s: no command line found; making empty\n",
 204                        __func__);
 205                 hverr = hv_set_command_line((HV_VirtAddr) command_line, 0);
 206         }
 207         if (hverr)
 208                 pr_warning("%s: hv_set_command_line returned error: %d\n",
 209                            __func__, hverr);
 210 }
 211
 212 /*
 213  * The kexec code range-checks all its PAs, so to avoid having it run
 214  * amok and allocate memory and then sequester it from every other
 215  * controller, we force it to come from controller zero.  We also
 216  * disable the oom-killer since if we do end up running out of memory,
 217  * that almost certainly won't help.
 218  */
 219 struct page *kimage_alloc_pages_arch(gfp_t gfp_mask, unsigned int order)
 220 {
 221         gfp_mask |= __GFP_THISNODE | __GFP_NORETRY;
 222         return alloc_pages_node(0, gfp_mask, order);
 223 }
 224
 225 static void setup_quasi_va_is_pa(void)
 226 {
 227         HV_PTE *pgtable;
 228         HV_PTE pte;
 229         int i;
 230
 231         /*
 232          * Flush our TLB to prevent conflicts between the previous contents
 233          * and the new stuff we're about to add.
 234          */
 235         local_flush_tlb_all();
 236
 237         /* setup VA is PA, at least up to PAGE_OFFSET */
 238
 239         pgtable = (HV_PTE *)current->mm->pgd;
 240         pte = hv_pte(_PAGE_KERNEL | _PAGE_HUGE_PAGE);
 241         pte = hv_pte_set_mode(pte, HV_PTE_MODE_CACHE_NO_L3);
 242
 243         for (i = 0; i < pgd_index(PAGE_OFFSET); i++) {
 244                 unsigned long pfn = i << (HPAGE_SHIFT - PAGE_SHIFT);
 245                 if (pfn_valid(pfn))
 246                         __set_pte(&pgtable[i], pfn_pte(pfn, pte));
 247         }
 248 }
 249
 250
 251 NORET_TYPE void machine_kexec(struct kimage *image)
 252 {
 253         void *reboot_code_buffer;
 254         NORET_TYPE void (*rnk)(unsigned long, void *, unsigned long)
 255                 ATTRIB_NORET;
 256
 257         /* Mask all interrupts before starting to reboot. */
 258         interrupt_mask_set_mask(~0ULL);
 259
 260         kexec_find_and_set_command_line(image);
 261
 262         /*
 263          * Adjust the home caching of the control page to be cached on
 264          * this cpu, and copy the assembly helper into the control
 265          * code page, which we map in the vmalloc area.
 266          */
 267         homecache_change_page_home(image->control_code_page, 0,
 268                                    smp_processor_id());
 269         reboot_code_buffer = vmap(&image->control_code_page, 1, 0,
 270                                   __pgprot(_PAGE_KERNEL | _PAGE_EXECUTABLE));
 271         memcpy(reboot_code_buffer, relocate_new_kernel,
 272                relocate_new_kernel_size);
 273         __flush_icache_range(
 274                 (unsigned long) reboot_code_buffer,
 275                 (unsigned long) reboot_code_buffer + relocate_new_kernel_size);
 276
 277         setup_quasi_va_is_pa();
 278
 279         /* now call it */
 280         rnk = reboot_code_buffer;
 281         (*rnk)(image->head, reboot_code_buffer, image->start);
 282 }