1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Load ELF vmlinux file for the kexec_file_load syscall.
4 *
5 * Copyright (C) 2021 Huawei Technologies Co, Ltd.
6 *
7 * Author: Liao Chang (liaochang1@huawei.com)
8 *
9 * Based on kexec-tools' kexec-elf-riscv.c, heavily modified
10 * for kernel.
11 */
12
13 #define pr_fmt(fmt) "kexec_image: " fmt
14
15 #include <linux/elf.h>
16 #include <linux/kexec.h>
17 #include <linux/slab.h>
18 #include <linux/of.h>
19 #include <linux/libfdt.h>
20 #include <linux/types.h>
21 #include <linux/memblock.h>
22 #include <asm/setup.h>
23
riscv_kexec_elf_load(struct kimage * image,struct elfhdr * ehdr,struct kexec_elf_info * elf_info,unsigned long old_pbase,unsigned long new_pbase)24 static int riscv_kexec_elf_load(struct kimage *image, struct elfhdr *ehdr,
25 struct kexec_elf_info *elf_info, unsigned long old_pbase,
26 unsigned long new_pbase)
27 {
28 int i;
29 int ret = 0;
30 size_t size;
31 struct kexec_buf kbuf;
32 const struct elf_phdr *phdr;
33
34 kbuf.image = image;
35
36 for (i = 0; i < ehdr->e_phnum; i++) {
37 phdr = &elf_info->proghdrs[i];
38 if (phdr->p_type != PT_LOAD)
39 continue;
40
41 size = phdr->p_filesz;
42 if (size > phdr->p_memsz)
43 size = phdr->p_memsz;
44
45 kbuf.buffer = (void *) elf_info->buffer + phdr->p_offset;
46 kbuf.bufsz = size;
47 kbuf.buf_align = phdr->p_align;
48 kbuf.mem = phdr->p_paddr - old_pbase + new_pbase;
49 kbuf.memsz = phdr->p_memsz;
50 kbuf.top_down = false;
51 ret = kexec_add_buffer(&kbuf);
52 if (ret)
53 break;
54 }
55
56 return ret;
57 }
58
59 /*
60 * Go through the available phsyical memory regions and find one that hold
61 * an image of the specified size.
62 */
elf_find_pbase(struct kimage * image,unsigned long kernel_len,struct elfhdr * ehdr,struct kexec_elf_info * elf_info,unsigned long * old_pbase,unsigned long * new_pbase)63 static int elf_find_pbase(struct kimage *image, unsigned long kernel_len,
64 struct elfhdr *ehdr, struct kexec_elf_info *elf_info,
65 unsigned long *old_pbase, unsigned long *new_pbase)
66 {
67 int i;
68 int ret;
69 struct kexec_buf kbuf;
70 const struct elf_phdr *phdr;
71 unsigned long lowest_paddr = ULONG_MAX;
72 unsigned long lowest_vaddr = ULONG_MAX;
73
74 for (i = 0; i < ehdr->e_phnum; i++) {
75 phdr = &elf_info->proghdrs[i];
76 if (phdr->p_type != PT_LOAD)
77 continue;
78
79 if (lowest_paddr > phdr->p_paddr)
80 lowest_paddr = phdr->p_paddr;
81
82 if (lowest_vaddr > phdr->p_vaddr)
83 lowest_vaddr = phdr->p_vaddr;
84 }
85
86 kbuf.image = image;
87 kbuf.buf_min = lowest_paddr;
88 kbuf.buf_max = ULONG_MAX;
89 kbuf.buf_align = PAGE_SIZE;
90 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
91 kbuf.memsz = ALIGN(kernel_len, PAGE_SIZE);
92 kbuf.top_down = false;
93 ret = arch_kexec_locate_mem_hole(&kbuf);
94 if (!ret) {
95 *old_pbase = lowest_paddr;
96 *new_pbase = kbuf.mem;
97 image->start = ehdr->e_entry - lowest_vaddr + kbuf.mem;
98 }
99 return ret;
100 }
101
get_nr_ram_ranges_callback(struct resource * res,void * arg)102 static int get_nr_ram_ranges_callback(struct resource *res, void *arg)
103 {
104 unsigned int *nr_ranges = arg;
105
106 (*nr_ranges)++;
107 return 0;
108 }
109
prepare_elf64_ram_headers_callback(struct resource * res,void * arg)110 static int prepare_elf64_ram_headers_callback(struct resource *res, void *arg)
111 {
112 struct crash_mem *cmem = arg;
113
114 cmem->ranges[cmem->nr_ranges].start = res->start;
115 cmem->ranges[cmem->nr_ranges].end = res->end;
116 cmem->nr_ranges++;
117
118 return 0;
119 }
120
prepare_elf_headers(void ** addr,unsigned long * sz)121 static int prepare_elf_headers(void **addr, unsigned long *sz)
122 {
123 struct crash_mem *cmem;
124 unsigned int nr_ranges;
125 int ret;
126
127 nr_ranges = 1; /* For exclusion of crashkernel region */
128 walk_system_ram_res(0, -1, &nr_ranges, get_nr_ram_ranges_callback);
129
130 cmem = kmalloc(struct_size(cmem, ranges, nr_ranges), GFP_KERNEL);
131 if (!cmem)
132 return -ENOMEM;
133
134 cmem->max_nr_ranges = nr_ranges;
135 cmem->nr_ranges = 0;
136 ret = walk_system_ram_res(0, -1, cmem, prepare_elf64_ram_headers_callback);
137 if (ret)
138 goto out;
139
140 /* Exclude crashkernel region */
141 ret = crash_exclude_mem_range(cmem, crashk_res.start, crashk_res.end);
142 if (!ret)
143 ret = crash_prepare_elf64_headers(cmem, true, addr, sz);
144
145 out:
146 kfree(cmem);
147 return ret;
148 }
149
setup_kdump_cmdline(struct kimage * image,char * cmdline,unsigned long cmdline_len)150 static char *setup_kdump_cmdline(struct kimage *image, char *cmdline,
151 unsigned long cmdline_len)
152 {
153 int elfcorehdr_strlen;
154 char *cmdline_ptr;
155
156 cmdline_ptr = kzalloc(COMMAND_LINE_SIZE, GFP_KERNEL);
157 if (!cmdline_ptr)
158 return NULL;
159
160 elfcorehdr_strlen = sprintf(cmdline_ptr, "elfcorehdr=0x%lx ",
161 image->elf_load_addr);
162
163 if (elfcorehdr_strlen + cmdline_len > COMMAND_LINE_SIZE) {
164 pr_err("Appending elfcorehdr=<addr> exceeds cmdline size\n");
165 kfree(cmdline_ptr);
166 return NULL;
167 }
168
169 memcpy(cmdline_ptr + elfcorehdr_strlen, cmdline, cmdline_len);
170 /* Ensure it's nul terminated */
171 cmdline_ptr[COMMAND_LINE_SIZE - 1] = '\0';
172 return cmdline_ptr;
173 }
174
elf_kexec_load(struct kimage * image,char * kernel_buf,unsigned long kernel_len,char * initrd,unsigned long initrd_len,char * cmdline,unsigned long cmdline_len)175 static void *elf_kexec_load(struct kimage *image, char *kernel_buf,
176 unsigned long kernel_len, char *initrd,
177 unsigned long initrd_len, char *cmdline,
178 unsigned long cmdline_len)
179 {
180 int ret;
181 unsigned long old_kernel_pbase = ULONG_MAX;
182 unsigned long new_kernel_pbase = 0UL;
183 unsigned long initrd_pbase = 0UL;
184 unsigned long headers_sz;
185 unsigned long kernel_start;
186 void *fdt, *headers;
187 struct elfhdr ehdr;
188 struct kexec_buf kbuf;
189 struct kexec_elf_info elf_info;
190 char *modified_cmdline = NULL;
191
192 ret = kexec_build_elf_info(kernel_buf, kernel_len, &ehdr, &elf_info);
193 if (ret)
194 return ERR_PTR(ret);
195
196 ret = elf_find_pbase(image, kernel_len, &ehdr, &elf_info,
197 &old_kernel_pbase, &new_kernel_pbase);
198 if (ret)
199 goto out;
200 kernel_start = image->start;
201 pr_notice("The entry point of kernel at 0x%lx\n", image->start);
202
203 /* Add the kernel binary to the image */
204 ret = riscv_kexec_elf_load(image, &ehdr, &elf_info,
205 old_kernel_pbase, new_kernel_pbase);
206 if (ret)
207 goto out;
208
209 kbuf.image = image;
210 kbuf.buf_min = new_kernel_pbase + kernel_len;
211 kbuf.buf_max = ULONG_MAX;
212
213 /* Add elfcorehdr */
214 if (image->type == KEXEC_TYPE_CRASH) {
215 ret = prepare_elf_headers(&headers, &headers_sz);
216 if (ret) {
217 pr_err("Preparing elf core header failed\n");
218 goto out;
219 }
220
221 kbuf.buffer = headers;
222 kbuf.bufsz = headers_sz;
223 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
224 kbuf.memsz = headers_sz;
225 kbuf.buf_align = ELF_CORE_HEADER_ALIGN;
226 kbuf.top_down = true;
227
228 ret = kexec_add_buffer(&kbuf);
229 if (ret) {
230 vfree(headers);
231 goto out;
232 }
233 image->elf_headers = headers;
234 image->elf_load_addr = kbuf.mem;
235 image->elf_headers_sz = headers_sz;
236
237 pr_debug("Loaded elf core header at 0x%lx bufsz=0x%lx memsz=0x%lx\n",
238 image->elf_load_addr, kbuf.bufsz, kbuf.memsz);
239
240 /* Setup cmdline for kdump kernel case */
241 modified_cmdline = setup_kdump_cmdline(image, cmdline,
242 cmdline_len);
243 if (!modified_cmdline) {
244 pr_err("Setting up cmdline for kdump kernel failed\n");
245 ret = -EINVAL;
246 goto out;
247 }
248 cmdline = modified_cmdline;
249 }
250
251 #ifdef CONFIG_ARCH_HAS_KEXEC_PURGATORY
252 /* Add purgatory to the image */
253 kbuf.top_down = true;
254 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
255 ret = kexec_load_purgatory(image, &kbuf);
256 if (ret) {
257 pr_err("Error loading purgatory ret=%d\n", ret);
258 goto out;
259 }
260 ret = kexec_purgatory_get_set_symbol(image, "riscv_kernel_entry",
261 &kernel_start,
262 sizeof(kernel_start), 0);
263 if (ret)
264 pr_err("Error update purgatory ret=%d\n", ret);
265 #endif /* CONFIG_ARCH_HAS_KEXEC_PURGATORY */
266
267 /* Add the initrd to the image */
268 if (initrd != NULL) {
269 kbuf.buffer = initrd;
270 kbuf.bufsz = kbuf.memsz = initrd_len;
271 kbuf.buf_align = PAGE_SIZE;
272 kbuf.top_down = false;
273 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
274 ret = kexec_add_buffer(&kbuf);
275 if (ret)
276 goto out;
277 initrd_pbase = kbuf.mem;
278 pr_notice("Loaded initrd at 0x%lx\n", initrd_pbase);
279 }
280
281 /* Add the DTB to the image */
282 fdt = of_kexec_alloc_and_setup_fdt(image, initrd_pbase,
283 initrd_len, cmdline, 0);
284 if (!fdt) {
285 pr_err("Error setting up the new device tree.\n");
286 ret = -EINVAL;
287 goto out;
288 }
289
290 fdt_pack(fdt);
291 kbuf.buffer = fdt;
292 kbuf.bufsz = kbuf.memsz = fdt_totalsize(fdt);
293 kbuf.buf_align = PAGE_SIZE;
294 kbuf.mem = KEXEC_BUF_MEM_UNKNOWN;
295 kbuf.top_down = true;
296 ret = kexec_add_buffer(&kbuf);
297 if (ret) {
298 pr_err("Error add DTB kbuf ret=%d\n", ret);
299 goto out_free_fdt;
300 }
301 pr_notice("Loaded device tree at 0x%lx\n", kbuf.mem);
302 goto out;
303
304 out_free_fdt:
305 kvfree(fdt);
306 out:
307 kfree(modified_cmdline);
308 kexec_free_elf_info(&elf_info);
309 return ret ? ERR_PTR(ret) : NULL;
310 }
311
312 #define RV_X(x, s, n) (((x) >> (s)) & ((1 << (n)) - 1))
313 #define RISCV_IMM_BITS 12
314 #define RISCV_IMM_REACH (1LL << RISCV_IMM_BITS)
315 #define RISCV_CONST_HIGH_PART(x) \
316 (((x) + (RISCV_IMM_REACH >> 1)) & ~(RISCV_IMM_REACH - 1))
317 #define RISCV_CONST_LOW_PART(x) ((x) - RISCV_CONST_HIGH_PART(x))
318
319 #define ENCODE_ITYPE_IMM(x) \
320 (RV_X(x, 0, 12) << 20)
321 #define ENCODE_BTYPE_IMM(x) \
322 ((RV_X(x, 1, 4) << 8) | (RV_X(x, 5, 6) << 25) | \
323 (RV_X(x, 11, 1) << 7) | (RV_X(x, 12, 1) << 31))
324 #define ENCODE_UTYPE_IMM(x) \
325 (RV_X(x, 12, 20) << 12)
326 #define ENCODE_JTYPE_IMM(x) \
327 ((RV_X(x, 1, 10) << 21) | (RV_X(x, 11, 1) << 20) | \
328 (RV_X(x, 12, 8) << 12) | (RV_X(x, 20, 1) << 31))
329 #define ENCODE_CBTYPE_IMM(x) \
330 ((RV_X(x, 1, 2) << 3) | (RV_X(x, 3, 2) << 10) | (RV_X(x, 5, 1) << 2) | \
331 (RV_X(x, 6, 2) << 5) | (RV_X(x, 8, 1) << 12))
332 #define ENCODE_CJTYPE_IMM(x) \
333 ((RV_X(x, 1, 3) << 3) | (RV_X(x, 4, 1) << 11) | (RV_X(x, 5, 1) << 2) | \
334 (RV_X(x, 6, 1) << 7) | (RV_X(x, 7, 1) << 6) | (RV_X(x, 8, 2) << 9) | \
335 (RV_X(x, 10, 1) << 8) | (RV_X(x, 11, 1) << 12))
336 #define ENCODE_UJTYPE_IMM(x) \
337 (ENCODE_UTYPE_IMM(RISCV_CONST_HIGH_PART(x)) | \
338 (ENCODE_ITYPE_IMM(RISCV_CONST_LOW_PART(x)) << 32))
339 #define ENCODE_UITYPE_IMM(x) \
340 (ENCODE_UTYPE_IMM(x) | (ENCODE_ITYPE_IMM(x) << 32))
341
342 #define CLEAN_IMM(type, x) \
343 ((~ENCODE_##type##_IMM((uint64_t)(-1))) & (x))
344
arch_kexec_apply_relocations_add(struct purgatory_info * pi,Elf_Shdr * section,const Elf_Shdr * relsec,const Elf_Shdr * symtab)345 int arch_kexec_apply_relocations_add(struct purgatory_info *pi,
346 Elf_Shdr *section,
347 const Elf_Shdr *relsec,
348 const Elf_Shdr *symtab)
349 {
350 const char *strtab, *name, *shstrtab;
351 const Elf_Shdr *sechdrs;
352 Elf64_Rela *relas;
353 int i, r_type;
354
355 /* String & section header string table */
356 sechdrs = (void *)pi->ehdr + pi->ehdr->e_shoff;
357 strtab = (char *)pi->ehdr + sechdrs[symtab->sh_link].sh_offset;
358 shstrtab = (char *)pi->ehdr + sechdrs[pi->ehdr->e_shstrndx].sh_offset;
359
360 relas = (void *)pi->ehdr + relsec->sh_offset;
361
362 for (i = 0; i < relsec->sh_size / sizeof(*relas); i++) {
363 const Elf_Sym *sym; /* symbol to relocate */
364 unsigned long addr; /* final location after relocation */
365 unsigned long val; /* relocated symbol value */
366 unsigned long sec_base; /* relocated symbol value */
367 void *loc; /* tmp location to modify */
368
369 sym = (void *)pi->ehdr + symtab->sh_offset;
370 sym += ELF64_R_SYM(relas[i].r_info);
371
372 if (sym->st_name)
373 name = strtab + sym->st_name;
374 else
375 name = shstrtab + sechdrs[sym->st_shndx].sh_name;
376
377 loc = pi->purgatory_buf;
378 loc += section->sh_offset;
379 loc += relas[i].r_offset;
380
381 if (sym->st_shndx == SHN_ABS)
382 sec_base = 0;
383 else if (sym->st_shndx >= pi->ehdr->e_shnum) {
384 pr_err("Invalid section %d for symbol %s\n",
385 sym->st_shndx, name);
386 return -ENOEXEC;
387 } else
388 sec_base = pi->sechdrs[sym->st_shndx].sh_addr;
389
390 val = sym->st_value;
391 val += sec_base;
392 val += relas[i].r_addend;
393
394 addr = section->sh_addr + relas[i].r_offset;
395
396 r_type = ELF64_R_TYPE(relas[i].r_info);
397
398 switch (r_type) {
399 case R_RISCV_BRANCH:
400 *(u32 *)loc = CLEAN_IMM(BTYPE, *(u32 *)loc) |
401 ENCODE_BTYPE_IMM(val - addr);
402 break;
403 case R_RISCV_JAL:
404 *(u32 *)loc = CLEAN_IMM(JTYPE, *(u32 *)loc) |
405 ENCODE_JTYPE_IMM(val - addr);
406 break;
407 /*
408 * With no R_RISCV_PCREL_LO12_S, R_RISCV_PCREL_LO12_I
409 * sym is expected to be next to R_RISCV_PCREL_HI20
410 * in purgatory relsec. Handle it like R_RISCV_CALL
411 * sym, instead of searching the whole relsec.
412 */
413 case R_RISCV_PCREL_HI20:
414 case R_RISCV_CALL:
415 *(u64 *)loc = CLEAN_IMM(UITYPE, *(u64 *)loc) |
416 ENCODE_UJTYPE_IMM(val - addr);
417 break;
418 case R_RISCV_RVC_BRANCH:
419 *(u32 *)loc = CLEAN_IMM(CBTYPE, *(u32 *)loc) |
420 ENCODE_CBTYPE_IMM(val - addr);
421 break;
422 case R_RISCV_RVC_JUMP:
423 *(u32 *)loc = CLEAN_IMM(CJTYPE, *(u32 *)loc) |
424 ENCODE_CJTYPE_IMM(val - addr);
425 break;
426 case R_RISCV_ADD32:
427 *(u32 *)loc += val;
428 break;
429 case R_RISCV_SUB32:
430 *(u32 *)loc -= val;
431 break;
432 /* It has been applied by R_RISCV_PCREL_HI20 sym */
433 case R_RISCV_PCREL_LO12_I:
434 case R_RISCV_ALIGN:
435 case R_RISCV_RELAX:
436 break;
437 default:
438 pr_err("Unknown rela relocation: %d\n", r_type);
439 return -ENOEXEC;
440 }
441 }
442 return 0;
443 }
444
445 const struct kexec_file_ops elf_kexec_ops = {
446 .probe = kexec_elf_probe,
447 .load = elf_kexec_load,
448 };
449