arch/powerpc/kernel/machine_kexec_file_64.c

   1 /*
   2  * ppc64 code to implement the kexec_file_load syscall
   3  *
   4  * Copyright (C) 2004  Adam Litke (agl@us.ibm.com)
   5  * Copyright (C) 2004  IBM Corp.
   6  * Copyright (C) 2004,2005  Milton D Miller II, IBM Corporation
   7  * Copyright (C) 2005  R Sharada (sharada@in.ibm.com)
   8  * Copyright (C) 2006  Mohan Kumar M (mohan@in.ibm.com)
   9  * Copyright (C) 2016  IBM Corporation
  10  *
  11  * Based on kexec-tools' kexec-elf-ppc64.c, fs2dt.c.
  12  * Heavily modified for the kernel by
  13  * Thiago Jung Bauermann <bauerman@linux.vnet.ibm.com>.
  14  *
  15  * This program is free software; you can redistribute it and/or modify
  16  * it under the terms of the GNU General Public License as published by
  17  * the Free Software Foundation (version 2 of the License).
  18  *
  19  * This program is distributed in the hope that it will be useful,
  20  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  * GNU General Public License for more details.
  23  */
  24
  25 #include <linux/slab.h>
  26 #include <linux/kexec.h>
  27 #include <linux/memblock.h>
  28 #include <linux/of_fdt.h>
  29 #include <linux/libfdt.h>
  30 #include <asm/ima.h>
  31
  32 #define SLAVE_CODE_SIZE         256
  33
  34 static struct kexec_file_ops *kexec_file_loaders[] = {
  35         &kexec_elf64_ops,
  36 };
  37
  38 int arch_kexec_kernel_image_probe(struct kimage *image, void *buf,
  39                                   unsigned long buf_len)
  40 {
  41         int i, ret = -ENOEXEC;
  42         struct kexec_file_ops *fops;
  43
  44         /* We don't support crash kernels yet. */
  45         if (image->type == KEXEC_TYPE_CRASH)
  46                 return -EOPNOTSUPP;
  47
  48         for (i = 0; i < ARRAY_SIZE(kexec_file_loaders); i++) {
  49                 fops = kexec_file_loaders[i];
  50                 if (!fops || !fops->probe)
  51                         continue;
  52
  53                 ret = fops->probe(buf, buf_len);
  54                 if (!ret) {
  55                         image->fops = fops;
  56                         return ret;
  57                 }
  58         }
  59
  60         return ret;
  61 }
  62
  63 void *arch_kexec_kernel_image_load(struct kimage *image)
  64 {
  65         if (!image->fops || !image->fops->load)
  66                 return ERR_PTR(-ENOEXEC);
  67
  68         return image->fops->load(image, image->kernel_buf,
  69                                  image->kernel_buf_len, image->initrd_buf,
  70                                  image->initrd_buf_len, image->cmdline_buf,
  71                                  image->cmdline_buf_len);
  72 }
  73
  74 int arch_kimage_file_post_load_cleanup(struct kimage *image)
  75 {
  76         if (!image->fops || !image->fops->cleanup)
  77                 return 0;
  78
  79         return image->fops->cleanup(image->image_loader_data);
  80 }
  81
  82 /**
  83  * arch_kexec_walk_mem - call func(data) for each unreserved memory block
  84  * @kbuf:       Context info for the search. Also passed to @func.
  85  * @func:       Function to call for each memory block.
  86  *
  87  * This function is used by kexec_add_buffer and kexec_locate_mem_hole
  88  * to find unreserved memory to load kexec segments into.
  89  *
  90  * Return: The memory walk will stop when func returns a non-zero value
  91  * and that value will be returned. If all free regions are visited without
  92  * func returning non-zero, then zero will be returned.
  93  */
  94 int arch_kexec_walk_mem(struct kexec_buf *kbuf, int (*func)(u64, u64, void *))
  95 {
  96         int ret = 0;
  97         u64 i;
  98         phys_addr_t mstart, mend;
  99
 100         if (kbuf->top_down) {
 101                 for_each_free_mem_range_reverse(i, NUMA_NO_NODE, 0,
 102                                                 &mstart, &mend, NULL) {
 103                         /*
 104                          * In memblock, end points to the first byte after the
 105                          * range while in kexec, end points to the last byte
 106                          * in the range.
 107                          */
 108                         ret = func(mstart, mend - 1, kbuf);
 109                         if (ret)
 110                                 break;
 111                 }
 112         } else {
 113                 for_each_free_mem_range(i, NUMA_NO_NODE, 0, &mstart, &mend,
 114                                         NULL) {
 115                         /*
 116                          * In memblock, end points to the first byte after the
 117                          * range while in kexec, end points to the last byte
 118                          * in the range.
 119                          */
 120                         ret = func(mstart, mend - 1, kbuf);
 121                         if (ret)
 122                                 break;
 123                 }
 124         }
 125
 126         return ret;
 127 }
 128
 129 /**
 130  * setup_purgatory - initialize the purgatory's global variables
 131  * @image:              kexec image.
 132  * @slave_code:         Slave code for the purgatory.
 133  * @fdt:                Flattened device tree for the next kernel.
 134  * @kernel_load_addr:   Address where the kernel is loaded.
 135  * @fdt_load_addr:      Address where the flattened device tree is loaded.
 136  *
 137  * Return: 0 on success, or negative errno on error.
 138  */
 139 int setup_purgatory(struct kimage *image, const void *slave_code,
 140                     const void *fdt, unsigned long kernel_load_addr,
 141                     unsigned long fdt_load_addr)
 142 {
 143         unsigned int *slave_code_buf, master_entry;
 144         int ret;
 145
 146         slave_code_buf = kmalloc(SLAVE_CODE_SIZE, GFP_KERNEL);
 147         if (!slave_code_buf)
 148                 return -ENOMEM;
 149
 150         /* Get the slave code from the new kernel and put it in purgatory. */
 151         ret = kexec_purgatory_get_set_symbol(image, "purgatory_start",
 152                                              slave_code_buf, SLAVE_CODE_SIZE,
 153                                              true);
 154         if (ret) {
 155                 kfree(slave_code_buf);
 156                 return ret;
 157         }
 158
 159         master_entry = slave_code_buf[0];
 160         memcpy(slave_code_buf, slave_code, SLAVE_CODE_SIZE);
 161         slave_code_buf[0] = master_entry;
 162         ret = kexec_purgatory_get_set_symbol(image, "purgatory_start",
 163                                              slave_code_buf, SLAVE_CODE_SIZE,
 164                                              false);
 165         kfree(slave_code_buf);
 166
 167         ret = kexec_purgatory_get_set_symbol(image, "kernel", &kernel_load_addr,
 168                                              sizeof(kernel_load_addr), false);
 169         if (ret)
 170                 return ret;
 171         ret = kexec_purgatory_get_set_symbol(image, "dt_offset", &fdt_load_addr,
 172                                              sizeof(fdt_load_addr), false);
 173         if (ret)
 174                 return ret;
 175
 176         return 0;
 177 }
 178
 179 /**
 180  * delete_fdt_mem_rsv - delete memory reservation with given address and size
 181  *
 182  * Return: 0 on success, or negative errno on error.
 183  */
 184 int delete_fdt_mem_rsv(void *fdt, unsigned long start, unsigned long size)
 185 {
 186         int i, ret, num_rsvs = fdt_num_mem_rsv(fdt);
 187
 188         for (i = 0; i < num_rsvs; i++) {
 189                 uint64_t rsv_start, rsv_size;
 190
 191                 ret = fdt_get_mem_rsv(fdt, i, &rsv_start, &rsv_size);
 192                 if (ret) {
 193                         pr_err("Malformed device tree.\n");
 194                         return -EINVAL;
 195                 }
 196
 197                 if (rsv_start == start && rsv_size == size) {
 198                         ret = fdt_del_mem_rsv(fdt, i);
 199                         if (ret) {
 200                                 pr_err("Error deleting device tree reservation.\n");
 201                                 return -EINVAL;
 202                         }
 203
 204                         return 0;
 205                 }
 206         }
 207
 208         return -ENOENT;
 209 }
 210
 211 /*
 212  * setup_new_fdt - modify /chosen and memory reservation for the next kernel
 213  * @image:              kexec image being loaded.
 214  * @fdt:                Flattened device tree for the next kernel.
 215  * @initrd_load_addr:   Address where the next initrd will be loaded.
 216  * @initrd_len:         Size of the next initrd, or 0 if there will be none.
 217  * @cmdline:            Command line for the next kernel, or NULL if there will
 218  *                      be none.
 219  *
 220  * Return: 0 on success, or negative errno on error.
 221  */
 222 int setup_new_fdt(const struct kimage *image, void *fdt,
 223                   unsigned long initrd_load_addr, unsigned long initrd_len,
 224                   const char *cmdline)
 225 {
 226         int ret, chosen_node;
 227         const void *prop;
 228
 229         /* Remove memory reservation for the current device tree. */
 230         ret = delete_fdt_mem_rsv(fdt, __pa(initial_boot_params),
 231                                  fdt_totalsize(initial_boot_params));
 232         if (ret == 0)
 233                 pr_debug("Removed old device tree reservation.\n");
 234         else if (ret != -ENOENT)
 235                 return ret;
 236
 237         chosen_node = fdt_path_offset(fdt, "/chosen");
 238         if (chosen_node == -FDT_ERR_NOTFOUND) {
 239                 chosen_node = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
 240                                               "chosen");
 241                 if (chosen_node < 0) {
 242                         pr_err("Error creating /chosen.\n");
 243                         return -EINVAL;
 244                 }
 245         } else if (chosen_node < 0) {
 246                 pr_err("Malformed device tree: error reading /chosen.\n");
 247                 return -EINVAL;
 248         }
 249
 250         /* Did we boot using an initrd? */
 251         prop = fdt_getprop(fdt, chosen_node, "linux,initrd-start", NULL);
 252         if (prop) {
 253                 uint64_t tmp_start, tmp_end, tmp_size;
 254
 255                 tmp_start = fdt64_to_cpu(*((const fdt64_t *) prop));
 256
 257                 prop = fdt_getprop(fdt, chosen_node, "linux,initrd-end", NULL);
 258                 if (!prop) {
 259                         pr_err("Malformed device tree.\n");
 260                         return -EINVAL;
 261                 }
 262                 tmp_end = fdt64_to_cpu(*((const fdt64_t *) prop));
 263
 264                 /*
 265                  * kexec reserves exact initrd size, while firmware may
 266                  * reserve a multiple of PAGE_SIZE, so check for both.
 267                  */
 268                 tmp_size = tmp_end - tmp_start;
 269                 ret = delete_fdt_mem_rsv(fdt, tmp_start, tmp_size);
 270                 if (ret == -ENOENT)
 271                         ret = delete_fdt_mem_rsv(fdt, tmp_start,
 272                                                  round_up(tmp_size, PAGE_SIZE));
 273                 if (ret == 0)
 274                         pr_debug("Removed old initrd reservation.\n");
 275                 else if (ret != -ENOENT)
 276                         return ret;
 277
 278                 /* If there's no new initrd, delete the old initrd's info. */
 279                 if (initrd_len == 0) {
 280                         ret = fdt_delprop(fdt, chosen_node,
 281                                           "linux,initrd-start");
 282                         if (ret) {
 283                                 pr_err("Error deleting linux,initrd-start.\n");
 284                                 return -EINVAL;
 285                         }
 286
 287                         ret = fdt_delprop(fdt, chosen_node, "linux,initrd-end");
 288                         if (ret) {
 289                                 pr_err("Error deleting linux,initrd-end.\n");
 290                                 return -EINVAL;
 291                         }
 292                 }
 293         }
 294
 295         if (initrd_len) {
 296                 ret = fdt_setprop_u64(fdt, chosen_node,
 297                                       "linux,initrd-start",
 298                                       initrd_load_addr);
 299                 if (ret < 0) {
 300                         pr_err("Error setting up the new device tree.\n");
 301                         return -EINVAL;
 302                 }
 303
 304                 /* initrd-end is the first address after the initrd image. */
 305                 ret = fdt_setprop_u64(fdt, chosen_node, "linux,initrd-end",
 306                                       initrd_load_addr + initrd_len);
 307                 if (ret < 0) {
 308                         pr_err("Error setting up the new device tree.\n");
 309                         return -EINVAL;
 310                 }
 311
 312                 ret = fdt_add_mem_rsv(fdt, initrd_load_addr, initrd_len);
 313                 if (ret) {
 314                         pr_err("Error reserving initrd memory: %s\n",
 315                                fdt_strerror(ret));
 316                         return -EINVAL;
 317                 }
 318         }
 319
 320         if (cmdline != NULL) {
 321                 ret = fdt_setprop_string(fdt, chosen_node, "bootargs", cmdline);
 322                 if (ret < 0) {
 323                         pr_err("Error setting up the new device tree.\n");
 324                         return -EINVAL;
 325                 }
 326         } else {
 327                 ret = fdt_delprop(fdt, chosen_node, "bootargs");
 328                 if (ret && ret != -FDT_ERR_NOTFOUND) {
 329                         pr_err("Error deleting bootargs.\n");
 330                         return -EINVAL;
 331                 }
 332         }
 333
 334         ret = setup_ima_buffer(image, fdt, chosen_node);
 335         if (ret) {
 336                 pr_err("Error setting up the new device tree.\n");
 337                 return ret;
 338         }
 339
 340         ret = fdt_setprop(fdt, chosen_node, "linux,booted-from-kexec", NULL, 0);
 341         if (ret) {
 342                 pr_err("Error setting up the new device tree.\n");
 343                 return -EINVAL;
 344         }
 345
 346         return 0;
 347 }