]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/x86/kernel/kexec-bzimage64.c
x86/process: Allow runtime control of Speculative Store Bypass
[mirror_ubuntu-artful-kernel.git] / arch / x86 / kernel / kexec-bzimage64.c
1 /*
2 * Kexec bzImage loader
3 *
4 * Copyright (C) 2014 Red Hat Inc.
5 * Authors:
6 * Vivek Goyal <vgoyal@redhat.com>
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 #define pr_fmt(fmt) "kexec-bzImage64: " fmt
13
14 #include <linux/string.h>
15 #include <linux/printk.h>
16 #include <linux/errno.h>
17 #include <linux/slab.h>
18 #include <linux/kexec.h>
19 #include <linux/kernel.h>
20 #include <linux/mm.h>
21 #include <linux/efi.h>
22 #include <linux/verification.h>
23
24 #include <asm/bootparam.h>
25 #include <asm/setup.h>
26 #include <asm/crash.h>
27 #include <asm/efi.h>
28 #include <asm/e820/api.h>
29 #include <asm/kexec-bzimage64.h>
30
31 #define MAX_ELFCOREHDR_STR_LEN 30 /* elfcorehdr=0x<64bit-value> */
32
33 /*
34 * Defines lowest physical address for various segments. Not sure where
35 * exactly these limits came from. Current bzimage64 loader in kexec-tools
36 * uses these so I am retaining it. It can be changed over time as we gain
37 * more insight.
38 */
39 #define MIN_PURGATORY_ADDR 0x3000
40 #define MIN_BOOTPARAM_ADDR 0x3000
41 #define MIN_KERNEL_LOAD_ADDR 0x100000
42 #define MIN_INITRD_LOAD_ADDR 0x1000000
43
44 /*
45 * This is a place holder for all boot loader specific data structure which
46 * gets allocated in one call but gets freed much later during cleanup
47 * time. Right now there is only one field but it can grow as need be.
48 */
49 struct bzimage64_data {
50 /*
51 * Temporary buffer to hold bootparams buffer. This should be
52 * freed once the bootparam segment has been loaded.
53 */
54 void *bootparams_buf;
55 };
56
57 static int setup_initrd(struct boot_params *params,
58 unsigned long initrd_load_addr, unsigned long initrd_len)
59 {
60 params->hdr.ramdisk_image = initrd_load_addr & 0xffffffffUL;
61 params->hdr.ramdisk_size = initrd_len & 0xffffffffUL;
62
63 params->ext_ramdisk_image = initrd_load_addr >> 32;
64 params->ext_ramdisk_size = initrd_len >> 32;
65
66 return 0;
67 }
68
69 static int setup_cmdline(struct kimage *image, struct boot_params *params,
70 unsigned long bootparams_load_addr,
71 unsigned long cmdline_offset, char *cmdline,
72 unsigned long cmdline_len)
73 {
74 char *cmdline_ptr = ((char *)params) + cmdline_offset;
75 unsigned long cmdline_ptr_phys, len = 0;
76 uint32_t cmdline_low_32, cmdline_ext_32;
77
78 if (image->type == KEXEC_TYPE_CRASH) {
79 len = sprintf(cmdline_ptr,
80 "elfcorehdr=0x%lx ", image->arch.elf_load_addr);
81 }
82 memcpy(cmdline_ptr + len, cmdline, cmdline_len);
83 cmdline_len += len;
84
85 cmdline_ptr[cmdline_len - 1] = '\0';
86
87 pr_debug("Final command line is: %s\n", cmdline_ptr);
88 cmdline_ptr_phys = bootparams_load_addr + cmdline_offset;
89 cmdline_low_32 = cmdline_ptr_phys & 0xffffffffUL;
90 cmdline_ext_32 = cmdline_ptr_phys >> 32;
91
92 params->hdr.cmd_line_ptr = cmdline_low_32;
93 if (cmdline_ext_32)
94 params->ext_cmd_line_ptr = cmdline_ext_32;
95
96 return 0;
97 }
98
99 static int setup_e820_entries(struct boot_params *params)
100 {
101 unsigned int nr_e820_entries;
102
103 nr_e820_entries = e820_table_kexec->nr_entries;
104
105 /* TODO: Pass entries more than E820_MAX_ENTRIES_ZEROPAGE in bootparams setup data */
106 if (nr_e820_entries > E820_MAX_ENTRIES_ZEROPAGE)
107 nr_e820_entries = E820_MAX_ENTRIES_ZEROPAGE;
108
109 params->e820_entries = nr_e820_entries;
110 memcpy(&params->e820_table, &e820_table_kexec->entries, nr_e820_entries*sizeof(struct e820_entry));
111
112 return 0;
113 }
114
115 #ifdef CONFIG_EFI
116 static int setup_efi_info_memmap(struct boot_params *params,
117 unsigned long params_load_addr,
118 unsigned int efi_map_offset,
119 unsigned int efi_map_sz)
120 {
121 void *efi_map = (void *)params + efi_map_offset;
122 unsigned long efi_map_phys_addr = params_load_addr + efi_map_offset;
123 struct efi_info *ei = &params->efi_info;
124
125 if (!efi_map_sz)
126 return 0;
127
128 efi_runtime_map_copy(efi_map, efi_map_sz);
129
130 ei->efi_memmap = efi_map_phys_addr & 0xffffffff;
131 ei->efi_memmap_hi = efi_map_phys_addr >> 32;
132 ei->efi_memmap_size = efi_map_sz;
133
134 return 0;
135 }
136
137 static int
138 prepare_add_efi_setup_data(struct boot_params *params,
139 unsigned long params_load_addr,
140 unsigned int efi_setup_data_offset)
141 {
142 unsigned long setup_data_phys;
143 struct setup_data *sd = (void *)params + efi_setup_data_offset;
144 struct efi_setup_data *esd = (void *)sd + sizeof(struct setup_data);
145
146 esd->fw_vendor = efi.fw_vendor;
147 esd->runtime = efi.runtime;
148 esd->tables = efi.config_table;
149 esd->smbios = efi.smbios;
150
151 sd->type = SETUP_EFI;
152 sd->len = sizeof(struct efi_setup_data);
153
154 /* Add setup data */
155 setup_data_phys = params_load_addr + efi_setup_data_offset;
156 sd->next = params->hdr.setup_data;
157 params->hdr.setup_data = setup_data_phys;
158
159 return 0;
160 }
161
162 static int
163 setup_efi_state(struct boot_params *params, unsigned long params_load_addr,
164 unsigned int efi_map_offset, unsigned int efi_map_sz,
165 unsigned int efi_setup_data_offset)
166 {
167 struct efi_info *current_ei = &boot_params.efi_info;
168 struct efi_info *ei = &params->efi_info;
169
170 if (!current_ei->efi_memmap_size)
171 return 0;
172
173 /*
174 * If 1:1 mapping is not enabled, second kernel can not setup EFI
175 * and use EFI run time services. User space will have to pass
176 * acpi_rsdp=<addr> on kernel command line to make second kernel boot
177 * without efi.
178 */
179 if (efi_enabled(EFI_OLD_MEMMAP))
180 return 0;
181
182 params->secure_boot = boot_params.secure_boot;
183 ei->efi_loader_signature = current_ei->efi_loader_signature;
184 ei->efi_systab = current_ei->efi_systab;
185 ei->efi_systab_hi = current_ei->efi_systab_hi;
186
187 ei->efi_memdesc_version = current_ei->efi_memdesc_version;
188 ei->efi_memdesc_size = efi_get_runtime_map_desc_size();
189
190 setup_efi_info_memmap(params, params_load_addr, efi_map_offset,
191 efi_map_sz);
192 prepare_add_efi_setup_data(params, params_load_addr,
193 efi_setup_data_offset);
194 return 0;
195 }
196 #endif /* CONFIG_EFI */
197
198 static int
199 setup_boot_parameters(struct kimage *image, struct boot_params *params,
200 unsigned long params_load_addr,
201 unsigned int efi_map_offset, unsigned int efi_map_sz,
202 unsigned int efi_setup_data_offset)
203 {
204 unsigned int nr_e820_entries;
205 unsigned long long mem_k, start, end;
206 int i, ret = 0;
207
208 /* Get subarch from existing bootparams */
209 params->hdr.hardware_subarch = boot_params.hdr.hardware_subarch;
210
211 /* Copying screen_info will do? */
212 memcpy(&params->screen_info, &boot_params.screen_info,
213 sizeof(struct screen_info));
214
215 /* Fill in memsize later */
216 params->screen_info.ext_mem_k = 0;
217 params->alt_mem_k = 0;
218
219 /* Default APM info */
220 memset(&params->apm_bios_info, 0, sizeof(params->apm_bios_info));
221
222 /* Default drive info */
223 memset(&params->hd0_info, 0, sizeof(params->hd0_info));
224 memset(&params->hd1_info, 0, sizeof(params->hd1_info));
225
226 if (image->type == KEXEC_TYPE_CRASH) {
227 ret = crash_setup_memmap_entries(image, params);
228 if (ret)
229 return ret;
230 } else
231 setup_e820_entries(params);
232
233 nr_e820_entries = params->e820_entries;
234
235 for (i = 0; i < nr_e820_entries; i++) {
236 if (params->e820_table[i].type != E820_TYPE_RAM)
237 continue;
238 start = params->e820_table[i].addr;
239 end = params->e820_table[i].addr + params->e820_table[i].size - 1;
240
241 if ((start <= 0x100000) && end > 0x100000) {
242 mem_k = (end >> 10) - (0x100000 >> 10);
243 params->screen_info.ext_mem_k = mem_k;
244 params->alt_mem_k = mem_k;
245 if (mem_k > 0xfc00)
246 params->screen_info.ext_mem_k = 0xfc00; /* 64M*/
247 if (mem_k > 0xffffffff)
248 params->alt_mem_k = 0xffffffff;
249 }
250 }
251
252 #ifdef CONFIG_EFI
253 /* Setup EFI state */
254 setup_efi_state(params, params_load_addr, efi_map_offset, efi_map_sz,
255 efi_setup_data_offset);
256 #endif
257
258 /* Setup EDD info */
259 memcpy(params->eddbuf, boot_params.eddbuf,
260 EDDMAXNR * sizeof(struct edd_info));
261 params->eddbuf_entries = boot_params.eddbuf_entries;
262
263 memcpy(params->edd_mbr_sig_buffer, boot_params.edd_mbr_sig_buffer,
264 EDD_MBR_SIG_MAX * sizeof(unsigned int));
265
266 return ret;
267 }
268
269 static int bzImage64_probe(const char *buf, unsigned long len)
270 {
271 int ret = -ENOEXEC;
272 struct setup_header *header;
273
274 /* kernel should be at least two sectors long */
275 if (len < 2 * 512) {
276 pr_err("File is too short to be a bzImage\n");
277 return ret;
278 }
279
280 header = (struct setup_header *)(buf + offsetof(struct boot_params, hdr));
281 if (memcmp((char *)&header->header, "HdrS", 4) != 0) {
282 pr_err("Not a bzImage\n");
283 return ret;
284 }
285
286 if (header->boot_flag != 0xAA55) {
287 pr_err("No x86 boot sector present\n");
288 return ret;
289 }
290
291 if (header->version < 0x020C) {
292 pr_err("Must be at least protocol version 2.12\n");
293 return ret;
294 }
295
296 if (!(header->loadflags & LOADED_HIGH)) {
297 pr_err("zImage not a bzImage\n");
298 return ret;
299 }
300
301 if (!(header->xloadflags & XLF_KERNEL_64)) {
302 pr_err("Not a bzImage64. XLF_KERNEL_64 is not set.\n");
303 return ret;
304 }
305
306 if (!(header->xloadflags & XLF_CAN_BE_LOADED_ABOVE_4G)) {
307 pr_err("XLF_CAN_BE_LOADED_ABOVE_4G is not set.\n");
308 return ret;
309 }
310
311 /*
312 * Can't handle 32bit EFI as it does not allow loading kernel
313 * above 4G. This should be handled by 32bit bzImage loader
314 */
315 if (efi_enabled(EFI_RUNTIME_SERVICES) && !efi_enabled(EFI_64BIT)) {
316 pr_debug("EFI is 32 bit. Can't load kernel above 4G.\n");
317 return ret;
318 }
319
320 /* I've got a bzImage */
321 pr_debug("It's a relocatable bzImage64\n");
322 ret = 0;
323
324 return ret;
325 }
326
327 static void *bzImage64_load(struct kimage *image, char *kernel,
328 unsigned long kernel_len, char *initrd,
329 unsigned long initrd_len, char *cmdline,
330 unsigned long cmdline_len)
331 {
332
333 struct setup_header *header;
334 int setup_sects, kern16_size, ret = 0;
335 unsigned long setup_header_size, params_cmdline_sz;
336 struct boot_params *params;
337 unsigned long bootparam_load_addr, kernel_load_addr, initrd_load_addr;
338 unsigned long purgatory_load_addr;
339 struct bzimage64_data *ldata;
340 struct kexec_entry64_regs regs64;
341 void *stack;
342 unsigned int setup_hdr_offset = offsetof(struct boot_params, hdr);
343 unsigned int efi_map_offset, efi_map_sz, efi_setup_data_offset;
344 struct kexec_buf kbuf = { .image = image, .buf_max = ULONG_MAX,
345 .top_down = true };
346
347 header = (struct setup_header *)(kernel + setup_hdr_offset);
348 setup_sects = header->setup_sects;
349 if (setup_sects == 0)
350 setup_sects = 4;
351
352 kern16_size = (setup_sects + 1) * 512;
353 if (kernel_len < kern16_size) {
354 pr_err("bzImage truncated\n");
355 return ERR_PTR(-ENOEXEC);
356 }
357
358 if (cmdline_len > header->cmdline_size) {
359 pr_err("Kernel command line too long\n");
360 return ERR_PTR(-EINVAL);
361 }
362
363 /*
364 * In case of crash dump, we will append elfcorehdr=<addr> to
365 * command line. Make sure it does not overflow
366 */
367 if (cmdline_len + MAX_ELFCOREHDR_STR_LEN > header->cmdline_size) {
368 pr_debug("Appending elfcorehdr=<addr> to command line exceeds maximum allowed length\n");
369 return ERR_PTR(-EINVAL);
370 }
371
372 /* Allocate and load backup region */
373 if (image->type == KEXEC_TYPE_CRASH) {
374 ret = crash_load_segments(image);
375 if (ret)
376 return ERR_PTR(ret);
377 }
378
379 /*
380 * Load purgatory. For 64bit entry point, purgatory code can be
381 * anywhere.
382 */
383 ret = kexec_load_purgatory(image, MIN_PURGATORY_ADDR, ULONG_MAX, 1,
384 &purgatory_load_addr);
385 if (ret) {
386 pr_err("Loading purgatory failed\n");
387 return ERR_PTR(ret);
388 }
389
390 pr_debug("Loaded purgatory at 0x%lx\n", purgatory_load_addr);
391
392
393 /*
394 * Load Bootparams and cmdline and space for efi stuff.
395 *
396 * Allocate memory together for multiple data structures so
397 * that they all can go in single area/segment and we don't
398 * have to create separate segment for each. Keeps things
399 * little bit simple
400 */
401 efi_map_sz = efi_get_runtime_map_size();
402 efi_map_sz = ALIGN(efi_map_sz, 16);
403 params_cmdline_sz = sizeof(struct boot_params) + cmdline_len +
404 MAX_ELFCOREHDR_STR_LEN;
405 params_cmdline_sz = ALIGN(params_cmdline_sz, 16);
406 kbuf.bufsz = params_cmdline_sz + efi_map_sz +
407 sizeof(struct setup_data) +
408 sizeof(struct efi_setup_data);
409
410 params = kzalloc(kbuf.bufsz, GFP_KERNEL);
411 if (!params)
412 return ERR_PTR(-ENOMEM);
413 efi_map_offset = params_cmdline_sz;
414 efi_setup_data_offset = efi_map_offset + efi_map_sz;
415
416 /* Copy setup header onto bootparams. Documentation/x86/boot.txt */
417 setup_header_size = 0x0202 + kernel[0x0201] - setup_hdr_offset;
418
419 /* Is there a limit on setup header size? */
420 memcpy(&params->hdr, (kernel + setup_hdr_offset), setup_header_size);
421
422 kbuf.buffer = params;
423 kbuf.memsz = kbuf.bufsz;
424 kbuf.buf_align = 16;
425 kbuf.buf_min = MIN_BOOTPARAM_ADDR;
426 ret = kexec_add_buffer(&kbuf);
427 if (ret)
428 goto out_free_params;
429 bootparam_load_addr = kbuf.mem;
430 pr_debug("Loaded boot_param, command line and misc at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
431 bootparam_load_addr, kbuf.bufsz, kbuf.bufsz);
432
433 /* Load kernel */
434 kbuf.buffer = kernel + kern16_size;
435 kbuf.bufsz = kernel_len - kern16_size;
436 kbuf.memsz = PAGE_ALIGN(header->init_size);
437 kbuf.buf_align = header->kernel_alignment;
438 kbuf.buf_min = MIN_KERNEL_LOAD_ADDR;
439 ret = kexec_add_buffer(&kbuf);
440 if (ret)
441 goto out_free_params;
442 kernel_load_addr = kbuf.mem;
443
444 pr_debug("Loaded 64bit kernel at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
445 kernel_load_addr, kbuf.bufsz, kbuf.memsz);
446
447 /* Load initrd high */
448 if (initrd) {
449 kbuf.buffer = initrd;
450 kbuf.bufsz = kbuf.memsz = initrd_len;
451 kbuf.buf_align = PAGE_SIZE;
452 kbuf.buf_min = MIN_INITRD_LOAD_ADDR;
453 ret = kexec_add_buffer(&kbuf);
454 if (ret)
455 goto out_free_params;
456 initrd_load_addr = kbuf.mem;
457
458 pr_debug("Loaded initrd at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
459 initrd_load_addr, initrd_len, initrd_len);
460
461 setup_initrd(params, initrd_load_addr, initrd_len);
462 }
463
464 setup_cmdline(image, params, bootparam_load_addr,
465 sizeof(struct boot_params), cmdline, cmdline_len);
466
467 /* bootloader info. Do we need a separate ID for kexec kernel loader? */
468 params->hdr.type_of_loader = 0x0D << 4;
469 params->hdr.loadflags = 0;
470
471 /* Setup purgatory regs for entry */
472 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", &regs64,
473 sizeof(regs64), 1);
474 if (ret)
475 goto out_free_params;
476
477 regs64.rbx = 0; /* Bootstrap Processor */
478 regs64.rsi = bootparam_load_addr;
479 regs64.rip = kernel_load_addr + 0x200;
480 stack = kexec_purgatory_get_symbol_addr(image, "stack_end");
481 if (IS_ERR(stack)) {
482 pr_err("Could not find address of symbol stack_end\n");
483 ret = -EINVAL;
484 goto out_free_params;
485 }
486
487 regs64.rsp = (unsigned long)stack;
488 ret = kexec_purgatory_get_set_symbol(image, "entry64_regs", &regs64,
489 sizeof(regs64), 0);
490 if (ret)
491 goto out_free_params;
492
493 ret = setup_boot_parameters(image, params, bootparam_load_addr,
494 efi_map_offset, efi_map_sz,
495 efi_setup_data_offset);
496 if (ret)
497 goto out_free_params;
498
499 /* Allocate loader specific data */
500 ldata = kzalloc(sizeof(struct bzimage64_data), GFP_KERNEL);
501 if (!ldata) {
502 ret = -ENOMEM;
503 goto out_free_params;
504 }
505
506 /*
507 * Store pointer to params so that it could be freed after loading
508 * params segment has been loaded and contents have been copied
509 * somewhere else.
510 */
511 ldata->bootparams_buf = params;
512 return ldata;
513
514 out_free_params:
515 kfree(params);
516 return ERR_PTR(ret);
517 }
518
519 /* This cleanup function is called after various segments have been loaded */
520 static int bzImage64_cleanup(void *loader_data)
521 {
522 struct bzimage64_data *ldata = loader_data;
523
524 if (!ldata)
525 return 0;
526
527 kfree(ldata->bootparams_buf);
528 ldata->bootparams_buf = NULL;
529
530 return 0;
531 }
532
533 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG
534 static int bzImage64_verify_sig(const char *kernel, unsigned long kernel_len)
535 {
536 return verify_pefile_signature(kernel, kernel_len,
537 NULL,
538 VERIFYING_KEXEC_PE_SIGNATURE);
539 }
540 #endif
541
542 struct kexec_file_ops kexec_bzImage64_ops = {
543 .probe = bzImage64_probe,
544 .load = bzImage64_load,
545 .cleanup = bzImage64_cleanup,
546 #ifdef CONFIG_KEXEC_BZIMAGE_VERIFY_SIG
547 .verify_sig = bzImage64_verify_sig,
548 #endif
549 };