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