Commit | Line | Data |
---|---|---|
eb39c880 MS |
1 | /* |
2 | * Firmware Assisted dump: A robust mechanism to get reliable kernel crash | |
3 | * dump with assistance from firmware. This approach does not use kexec, | |
4 | * instead firmware assists in booting the kdump kernel while preserving | |
5 | * memory contents. The most of the code implementation has been adapted | |
6 | * from phyp assisted dump implementation written by Linas Vepstas and | |
7 | * Manish Ahuja | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or modify | |
10 | * it under the terms of the GNU General Public License as published by | |
11 | * the Free Software Foundation; either version 2 of the License, or | |
12 | * (at your option) any later version. | |
13 | * | |
14 | * This program is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
17 | * GNU General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU General Public License | |
20 | * along with this program; if not, write to the Free Software | |
21 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
22 | * | |
23 | * Copyright 2011 IBM Corporation | |
24 | * Author: Mahesh Salgaonkar <mahesh@linux.vnet.ibm.com> | |
25 | */ | |
26 | ||
27 | #undef DEBUG | |
28 | #define pr_fmt(fmt) "fadump: " fmt | |
29 | ||
30 | #include <linux/string.h> | |
31 | #include <linux/memblock.h> | |
3ccc00a7 | 32 | #include <linux/delay.h> |
3ccc00a7 | 33 | #include <linux/seq_file.h> |
2df173d9 | 34 | #include <linux/crash_dump.h> |
b500afff MS |
35 | #include <linux/kobject.h> |
36 | #include <linux/sysfs.h> | |
eb39c880 | 37 | |
7644d581 | 38 | #include <asm/debugfs.h> |
eb39c880 MS |
39 | #include <asm/page.h> |
40 | #include <asm/prom.h> | |
41 | #include <asm/rtas.h> | |
42 | #include <asm/fadump.h> | |
cad3c834 | 43 | #include <asm/setup.h> |
eb39c880 MS |
44 | |
45 | static struct fw_dump fw_dump; | |
3ccc00a7 MS |
46 | static struct fadump_mem_struct fdm; |
47 | static const struct fadump_mem_struct *fdm_active; | |
48 | ||
49 | static DEFINE_MUTEX(fadump_mutex); | |
2df173d9 MS |
50 | struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES]; |
51 | int crash_mem_ranges; | |
eb39c880 MS |
52 | |
53 | /* Scan the Firmware Assisted dump configuration details. */ | |
54 | int __init early_init_dt_scan_fw_dump(unsigned long node, | |
55 | const char *uname, int depth, void *data) | |
56 | { | |
9d0c4dfe | 57 | const __be32 *sections; |
eb39c880 | 58 | int i, num_sections; |
9d0c4dfe | 59 | int size; |
408cddd9 | 60 | const __be32 *token; |
eb39c880 MS |
61 | |
62 | if (depth != 1 || strcmp(uname, "rtas") != 0) | |
63 | return 0; | |
64 | ||
65 | /* | |
66 | * Check if Firmware Assisted dump is supported. if yes, check | |
67 | * if dump has been initiated on last reboot. | |
68 | */ | |
69 | token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL); | |
70 | if (!token) | |
a7d04317 | 71 | return 1; |
eb39c880 MS |
72 | |
73 | fw_dump.fadump_supported = 1; | |
408cddd9 | 74 | fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token); |
eb39c880 MS |
75 | |
76 | /* | |
77 | * The 'ibm,kernel-dump' rtas node is present only if there is | |
78 | * dump data waiting for us. | |
79 | */ | |
3ccc00a7 MS |
80 | fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL); |
81 | if (fdm_active) | |
eb39c880 MS |
82 | fw_dump.dump_active = 1; |
83 | ||
84 | /* Get the sizes required to store dump data for the firmware provided | |
85 | * dump sections. | |
86 | * For each dump section type supported, a 32bit cell which defines | |
87 | * the ID of a supported section followed by two 32 bit cells which | |
88 | * gives teh size of the section in bytes. | |
89 | */ | |
90 | sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes", | |
91 | &size); | |
92 | ||
93 | if (!sections) | |
a7d04317 | 94 | return 1; |
eb39c880 MS |
95 | |
96 | num_sections = size / (3 * sizeof(u32)); | |
97 | ||
98 | for (i = 0; i < num_sections; i++, sections += 3) { | |
99 | u32 type = (u32)of_read_number(sections, 1); | |
100 | ||
101 | switch (type) { | |
102 | case FADUMP_CPU_STATE_DATA: | |
103 | fw_dump.cpu_state_data_size = | |
104 | of_read_ulong(§ions[1], 2); | |
105 | break; | |
106 | case FADUMP_HPTE_REGION: | |
107 | fw_dump.hpte_region_size = | |
108 | of_read_ulong(§ions[1], 2); | |
109 | break; | |
110 | } | |
111 | } | |
a7d04317 | 112 | |
eb39c880 MS |
113 | return 1; |
114 | } | |
115 | ||
eae0dfcc HB |
116 | /* |
117 | * If fadump is registered, check if the memory provided | |
118 | * falls within boot memory area. | |
119 | */ | |
120 | int is_fadump_boot_memory_area(u64 addr, ulong size) | |
121 | { | |
122 | if (!fw_dump.dump_registered) | |
123 | return 0; | |
124 | ||
125 | return (addr + size) > RMA_START && addr <= fw_dump.boot_memory_size; | |
126 | } | |
127 | ||
6fcd6baa NP |
128 | int should_fadump_crash(void) |
129 | { | |
130 | if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr) | |
131 | return 0; | |
132 | return 1; | |
133 | } | |
134 | ||
3ccc00a7 MS |
135 | int is_fadump_active(void) |
136 | { | |
137 | return fw_dump.dump_active; | |
138 | } | |
139 | ||
a5a05b91 HB |
140 | /* |
141 | * Returns 1, if there are no holes in boot memory area, | |
142 | * 0 otherwise. | |
143 | */ | |
144 | static int is_boot_memory_area_contiguous(void) | |
145 | { | |
146 | struct memblock_region *reg; | |
147 | unsigned long tstart, tend; | |
148 | unsigned long start_pfn = PHYS_PFN(RMA_START); | |
149 | unsigned long end_pfn = PHYS_PFN(RMA_START + fw_dump.boot_memory_size); | |
150 | unsigned int ret = 0; | |
151 | ||
152 | for_each_memblock(memory, reg) { | |
153 | tstart = max(start_pfn, memblock_region_memory_base_pfn(reg)); | |
154 | tend = min(end_pfn, memblock_region_memory_end_pfn(reg)); | |
155 | if (tstart < tend) { | |
156 | /* Memory hole from start_pfn to tstart */ | |
157 | if (tstart > start_pfn) | |
158 | break; | |
159 | ||
160 | if (tend == end_pfn) { | |
161 | ret = 1; | |
162 | break; | |
163 | } | |
164 | ||
165 | start_pfn = tend + 1; | |
166 | } | |
167 | } | |
168 | ||
169 | return ret; | |
170 | } | |
171 | ||
3ccc00a7 MS |
172 | /* Print firmware assisted dump configurations for debugging purpose. */ |
173 | static void fadump_show_config(void) | |
174 | { | |
175 | pr_debug("Support for firmware-assisted dump (fadump): %s\n", | |
176 | (fw_dump.fadump_supported ? "present" : "no support")); | |
177 | ||
178 | if (!fw_dump.fadump_supported) | |
179 | return; | |
180 | ||
181 | pr_debug("Fadump enabled : %s\n", | |
182 | (fw_dump.fadump_enabled ? "yes" : "no")); | |
183 | pr_debug("Dump Active : %s\n", | |
184 | (fw_dump.dump_active ? "yes" : "no")); | |
185 | pr_debug("Dump section sizes:\n"); | |
186 | pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size); | |
187 | pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size); | |
188 | pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size); | |
189 | } | |
190 | ||
191 | static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm, | |
192 | unsigned long addr) | |
193 | { | |
194 | if (!fdm) | |
195 | return 0; | |
196 | ||
197 | memset(fdm, 0, sizeof(struct fadump_mem_struct)); | |
198 | addr = addr & PAGE_MASK; | |
199 | ||
408cddd9 HB |
200 | fdm->header.dump_format_version = cpu_to_be32(0x00000001); |
201 | fdm->header.dump_num_sections = cpu_to_be16(3); | |
3ccc00a7 MS |
202 | fdm->header.dump_status_flag = 0; |
203 | fdm->header.offset_first_dump_section = | |
408cddd9 | 204 | cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data)); |
3ccc00a7 MS |
205 | |
206 | /* | |
207 | * Fields for disk dump option. | |
208 | * We are not using disk dump option, hence set these fields to 0. | |
209 | */ | |
210 | fdm->header.dd_block_size = 0; | |
211 | fdm->header.dd_block_offset = 0; | |
212 | fdm->header.dd_num_blocks = 0; | |
213 | fdm->header.dd_offset_disk_path = 0; | |
214 | ||
215 | /* set 0 to disable an automatic dump-reboot. */ | |
216 | fdm->header.max_time_auto = 0; | |
217 | ||
218 | /* Kernel dump sections */ | |
219 | /* cpu state data section. */ | |
408cddd9 HB |
220 | fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG); |
221 | fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA); | |
3ccc00a7 | 222 | fdm->cpu_state_data.source_address = 0; |
408cddd9 HB |
223 | fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size); |
224 | fdm->cpu_state_data.destination_address = cpu_to_be64(addr); | |
3ccc00a7 MS |
225 | addr += fw_dump.cpu_state_data_size; |
226 | ||
227 | /* hpte region section */ | |
408cddd9 HB |
228 | fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG); |
229 | fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION); | |
3ccc00a7 | 230 | fdm->hpte_region.source_address = 0; |
408cddd9 HB |
231 | fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size); |
232 | fdm->hpte_region.destination_address = cpu_to_be64(addr); | |
3ccc00a7 MS |
233 | addr += fw_dump.hpte_region_size; |
234 | ||
235 | /* RMA region section */ | |
408cddd9 HB |
236 | fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG); |
237 | fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION); | |
238 | fdm->rmr_region.source_address = cpu_to_be64(RMA_START); | |
239 | fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size); | |
240 | fdm->rmr_region.destination_address = cpu_to_be64(addr); | |
3ccc00a7 MS |
241 | addr += fw_dump.boot_memory_size; |
242 | ||
243 | return addr; | |
244 | } | |
245 | ||
eb39c880 MS |
246 | /** |
247 | * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM | |
248 | * | |
249 | * Function to find the largest memory size we need to reserve during early | |
250 | * boot process. This will be the size of the memory that is required for a | |
251 | * kernel to boot successfully. | |
252 | * | |
253 | * This function has been taken from phyp-assisted dump feature implementation. | |
254 | * | |
255 | * returns larger of 256MB or 5% rounded down to multiples of 256MB. | |
256 | * | |
257 | * TODO: Come up with better approach to find out more accurate memory size | |
258 | * that is required for a kernel to boot successfully. | |
259 | * | |
260 | */ | |
261 | static inline unsigned long fadump_calculate_reserve_size(void) | |
262 | { | |
11550dc0 HB |
263 | int ret; |
264 | unsigned long long base, size; | |
eb39c880 | 265 | |
81d9eca5 HB |
266 | if (fw_dump.reserve_bootvar) |
267 | pr_warn("'fadump_reserve_mem=' parameter is deprecated in favor of 'crashkernel=' parameter.\n"); | |
268 | ||
eb39c880 | 269 | /* |
11550dc0 | 270 | * Check if the size is specified through crashkernel= cmdline |
e7467dc6 HB |
271 | * option. If yes, then use that but ignore base as fadump reserves |
272 | * memory at a predefined offset. | |
eb39c880 | 273 | */ |
11550dc0 HB |
274 | ret = parse_crashkernel(boot_command_line, memblock_phys_mem_size(), |
275 | &size, &base); | |
276 | if (ret == 0 && size > 0) { | |
48a316e3 HB |
277 | unsigned long max_size; |
278 | ||
81d9eca5 HB |
279 | if (fw_dump.reserve_bootvar) |
280 | pr_info("Using 'crashkernel=' parameter for memory reservation.\n"); | |
281 | ||
11550dc0 | 282 | fw_dump.reserve_bootvar = (unsigned long)size; |
48a316e3 HB |
283 | |
284 | /* | |
285 | * Adjust if the boot memory size specified is above | |
286 | * the upper limit. | |
287 | */ | |
288 | max_size = memblock_phys_mem_size() / MAX_BOOT_MEM_RATIO; | |
289 | if (fw_dump.reserve_bootvar > max_size) { | |
290 | fw_dump.reserve_bootvar = max_size; | |
291 | pr_info("Adjusted boot memory size to %luMB\n", | |
292 | (fw_dump.reserve_bootvar >> 20)); | |
293 | } | |
294 | ||
eb39c880 | 295 | return fw_dump.reserve_bootvar; |
81d9eca5 HB |
296 | } else if (fw_dump.reserve_bootvar) { |
297 | /* | |
298 | * 'fadump_reserve_mem=' is being used to reserve memory | |
299 | * for firmware-assisted dump. | |
300 | */ | |
301 | return fw_dump.reserve_bootvar; | |
11550dc0 | 302 | } |
eb39c880 MS |
303 | |
304 | /* divide by 20 to get 5% of value */ | |
48a316e3 | 305 | size = memblock_phys_mem_size() / 20; |
eb39c880 MS |
306 | |
307 | /* round it down in multiples of 256 */ | |
308 | size = size & ~0x0FFFFFFFUL; | |
309 | ||
310 | /* Truncate to memory_limit. We don't want to over reserve the memory.*/ | |
311 | if (memory_limit && size > memory_limit) | |
312 | size = memory_limit; | |
313 | ||
314 | return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM); | |
315 | } | |
316 | ||
317 | /* | |
318 | * Calculate the total memory size required to be reserved for | |
319 | * firmware-assisted dump registration. | |
320 | */ | |
321 | static unsigned long get_fadump_area_size(void) | |
322 | { | |
323 | unsigned long size = 0; | |
324 | ||
325 | size += fw_dump.cpu_state_data_size; | |
326 | size += fw_dump.hpte_region_size; | |
327 | size += fw_dump.boot_memory_size; | |
2df173d9 MS |
328 | size += sizeof(struct fadump_crash_info_header); |
329 | size += sizeof(struct elfhdr); /* ELF core header.*/ | |
ebaeb5ae | 330 | size += sizeof(struct elf_phdr); /* place holder for cpu notes */ |
2df173d9 MS |
331 | /* Program headers for crash memory regions. */ |
332 | size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2); | |
eb39c880 MS |
333 | |
334 | size = PAGE_ALIGN(size); | |
335 | return size; | |
336 | } | |
337 | ||
b71a693d MS |
338 | static void __init fadump_reserve_crash_area(unsigned long base, |
339 | unsigned long size) | |
340 | { | |
341 | struct memblock_region *reg; | |
342 | unsigned long mstart, mend, msize; | |
343 | ||
344 | for_each_memblock(memory, reg) { | |
345 | mstart = max_t(unsigned long, base, reg->base); | |
346 | mend = reg->base + reg->size; | |
347 | mend = min(base + size, mend); | |
348 | ||
349 | if (mstart < mend) { | |
350 | msize = mend - mstart; | |
351 | memblock_reserve(mstart, msize); | |
352 | pr_info("Reserved %ldMB of memory at %#016lx for saving crash dump\n", | |
353 | (msize >> 20), mstart); | |
354 | } | |
355 | } | |
356 | } | |
357 | ||
eb39c880 MS |
358 | int __init fadump_reserve_mem(void) |
359 | { | |
360 | unsigned long base, size, memory_boundary; | |
361 | ||
362 | if (!fw_dump.fadump_enabled) | |
363 | return 0; | |
364 | ||
365 | if (!fw_dump.fadump_supported) { | |
366 | printk(KERN_INFO "Firmware-assisted dump is not supported on" | |
367 | " this hardware\n"); | |
368 | fw_dump.fadump_enabled = 0; | |
369 | return 0; | |
370 | } | |
3ccc00a7 MS |
371 | /* |
372 | * Initialize boot memory size | |
373 | * If dump is active then we have already calculated the size during | |
374 | * first kernel. | |
375 | */ | |
376 | if (fdm_active) | |
408cddd9 | 377 | fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len); |
3ccc00a7 MS |
378 | else |
379 | fw_dump.boot_memory_size = fadump_calculate_reserve_size(); | |
eb39c880 MS |
380 | |
381 | /* | |
382 | * Calculate the memory boundary. | |
383 | * If memory_limit is less than actual memory boundary then reserve | |
384 | * the memory for fadump beyond the memory_limit and adjust the | |
385 | * memory_limit accordingly, so that the running kernel can run with | |
386 | * specified memory_limit. | |
387 | */ | |
388 | if (memory_limit && memory_limit < memblock_end_of_DRAM()) { | |
389 | size = get_fadump_area_size(); | |
390 | if ((memory_limit + size) < memblock_end_of_DRAM()) | |
391 | memory_limit += size; | |
392 | else | |
393 | memory_limit = memblock_end_of_DRAM(); | |
394 | printk(KERN_INFO "Adjusted memory_limit for firmware-assisted" | |
a84fcd46 | 395 | " dump, now %#016llx\n", memory_limit); |
eb39c880 MS |
396 | } |
397 | if (memory_limit) | |
398 | memory_boundary = memory_limit; | |
399 | else | |
400 | memory_boundary = memblock_end_of_DRAM(); | |
401 | ||
402 | if (fw_dump.dump_active) { | |
b71a693d MS |
403 | pr_info("Firmware-assisted dump is active.\n"); |
404 | ||
eb39c880 MS |
405 | /* |
406 | * If last boot has crashed then reserve all the memory | |
407 | * above boot_memory_size so that we don't touch it until | |
408 | * dump is written to disk by userspace tool. This memory | |
409 | * will be released for general use once the dump is saved. | |
410 | */ | |
411 | base = fw_dump.boot_memory_size; | |
412 | size = memory_boundary - base; | |
b71a693d | 413 | fadump_reserve_crash_area(base, size); |
2df173d9 MS |
414 | |
415 | fw_dump.fadumphdr_addr = | |
408cddd9 HB |
416 | be64_to_cpu(fdm_active->rmr_region.destination_address) + |
417 | be64_to_cpu(fdm_active->rmr_region.source_len); | |
2df173d9 MS |
418 | pr_debug("fadumphdr_addr = %p\n", |
419 | (void *) fw_dump.fadumphdr_addr); | |
eb39c880 | 420 | } else { |
eb39c880 | 421 | size = get_fadump_area_size(); |
f6e6bedb HB |
422 | |
423 | /* | |
424 | * Reserve memory at an offset closer to bottom of the RAM to | |
425 | * minimize the impact of memory hot-remove operation. We can't | |
426 | * use memblock_find_in_range() here since it doesn't allocate | |
427 | * from bottom to top. | |
428 | */ | |
429 | for (base = fw_dump.boot_memory_size; | |
430 | base <= (memory_boundary - size); | |
431 | base += size) { | |
432 | if (memblock_is_region_memory(base, size) && | |
433 | !memblock_is_region_reserved(base, size)) | |
434 | break; | |
435 | } | |
436 | if ((base > (memory_boundary - size)) || | |
437 | memblock_reserve(base, size)) { | |
438 | pr_err("Failed to reserve memory\n"); | |
439 | return 0; | |
440 | } | |
441 | ||
442 | pr_info("Reserved %ldMB of memory at %ldMB for firmware-" | |
443 | "assisted dump (System RAM: %ldMB)\n", | |
444 | (unsigned long)(size >> 20), | |
445 | (unsigned long)(base >> 20), | |
446 | (unsigned long)(memblock_phys_mem_size() >> 20)); | |
eb39c880 | 447 | } |
f6e6bedb | 448 | |
eb39c880 MS |
449 | fw_dump.reserve_dump_area_start = base; |
450 | fw_dump.reserve_dump_area_size = size; | |
451 | return 1; | |
452 | } | |
453 | ||
1e76609c SD |
454 | unsigned long __init arch_reserved_kernel_pages(void) |
455 | { | |
456 | return memblock_reserved_size() / PAGE_SIZE; | |
457 | } | |
458 | ||
eb39c880 MS |
459 | /* Look for fadump= cmdline option. */ |
460 | static int __init early_fadump_param(char *p) | |
461 | { | |
462 | if (!p) | |
463 | return 1; | |
464 | ||
465 | if (strncmp(p, "on", 2) == 0) | |
466 | fw_dump.fadump_enabled = 1; | |
467 | else if (strncmp(p, "off", 3) == 0) | |
468 | fw_dump.fadump_enabled = 0; | |
469 | ||
470 | return 0; | |
471 | } | |
472 | early_param("fadump", early_fadump_param); | |
473 | ||
81d9eca5 HB |
474 | /* |
475 | * Look for fadump_reserve_mem= cmdline option | |
476 | * TODO: Remove references to 'fadump_reserve_mem=' parameter, | |
477 | * the sooner 'crashkernel=' parameter is accustomed to. | |
478 | */ | |
479 | static int __init early_fadump_reserve_mem(char *p) | |
480 | { | |
481 | if (p) | |
482 | fw_dump.reserve_bootvar = memparse(p, &p); | |
483 | return 0; | |
484 | } | |
485 | early_param("fadump_reserve_mem", early_fadump_reserve_mem); | |
486 | ||
98b8cd7f | 487 | static int register_fw_dump(struct fadump_mem_struct *fdm) |
3ccc00a7 | 488 | { |
98b8cd7f | 489 | int rc, err; |
3ccc00a7 MS |
490 | unsigned int wait_time; |
491 | ||
492 | pr_debug("Registering for firmware-assisted kernel dump...\n"); | |
493 | ||
494 | /* TODO: Add upper time limit for the delay */ | |
495 | do { | |
496 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, | |
497 | FADUMP_REGISTER, fdm, | |
498 | sizeof(struct fadump_mem_struct)); | |
499 | ||
500 | wait_time = rtas_busy_delay_time(rc); | |
501 | if (wait_time) | |
502 | mdelay(wait_time); | |
503 | ||
504 | } while (wait_time); | |
505 | ||
98b8cd7f | 506 | err = -EIO; |
3ccc00a7 | 507 | switch (rc) { |
98b8cd7f MS |
508 | default: |
509 | pr_err("Failed to register. Unknown Error(%d).\n", rc); | |
510 | break; | |
3ccc00a7 MS |
511 | case -1: |
512 | printk(KERN_ERR "Failed to register firmware-assisted kernel" | |
513 | " dump. Hardware Error(%d).\n", rc); | |
514 | break; | |
515 | case -3: | |
a5a05b91 HB |
516 | if (!is_boot_memory_area_contiguous()) |
517 | pr_err("Can't have holes in boot memory area while " | |
518 | "registering fadump\n"); | |
519 | ||
3ccc00a7 MS |
520 | printk(KERN_ERR "Failed to register firmware-assisted kernel" |
521 | " dump. Parameter Error(%d).\n", rc); | |
98b8cd7f | 522 | err = -EINVAL; |
3ccc00a7 MS |
523 | break; |
524 | case -9: | |
525 | printk(KERN_ERR "firmware-assisted kernel dump is already " | |
526 | " registered."); | |
527 | fw_dump.dump_registered = 1; | |
98b8cd7f | 528 | err = -EEXIST; |
3ccc00a7 MS |
529 | break; |
530 | case 0: | |
531 | printk(KERN_INFO "firmware-assisted kernel dump registration" | |
532 | " is successful\n"); | |
533 | fw_dump.dump_registered = 1; | |
98b8cd7f | 534 | err = 0; |
3ccc00a7 MS |
535 | break; |
536 | } | |
98b8cd7f | 537 | return err; |
3ccc00a7 MS |
538 | } |
539 | ||
ebaeb5ae MS |
540 | void crash_fadump(struct pt_regs *regs, const char *str) |
541 | { | |
542 | struct fadump_crash_info_header *fdh = NULL; | |
f2a5e8f0 | 543 | int old_cpu, this_cpu; |
ebaeb5ae | 544 | |
6fcd6baa | 545 | if (!should_fadump_crash()) |
ebaeb5ae MS |
546 | return; |
547 | ||
f2a5e8f0 MS |
548 | /* |
549 | * old_cpu == -1 means this is the first CPU which has come here, | |
550 | * go ahead and trigger fadump. | |
551 | * | |
552 | * old_cpu != -1 means some other CPU has already on it's way | |
553 | * to trigger fadump, just keep looping here. | |
554 | */ | |
555 | this_cpu = smp_processor_id(); | |
556 | old_cpu = cmpxchg(&crashing_cpu, -1, this_cpu); | |
557 | ||
558 | if (old_cpu != -1) { | |
559 | /* | |
560 | * We can't loop here indefinitely. Wait as long as fadump | |
561 | * is in force. If we race with fadump un-registration this | |
562 | * loop will break and then we go down to normal panic path | |
563 | * and reboot. If fadump is in force the first crashing | |
564 | * cpu will definitely trigger fadump. | |
565 | */ | |
566 | while (fw_dump.dump_registered) | |
567 | cpu_relax(); | |
568 | return; | |
569 | } | |
570 | ||
ebaeb5ae | 571 | fdh = __va(fw_dump.fadumphdr_addr); |
ebaeb5ae MS |
572 | fdh->crashing_cpu = crashing_cpu; |
573 | crash_save_vmcoreinfo(); | |
574 | ||
575 | if (regs) | |
576 | fdh->regs = *regs; | |
577 | else | |
578 | ppc_save_regs(&fdh->regs); | |
579 | ||
a0512164 | 580 | fdh->online_mask = *cpu_online_mask; |
ebaeb5ae MS |
581 | |
582 | /* Call ibm,os-term rtas call to trigger firmware assisted dump */ | |
583 | rtas_os_term((char *)str); | |
584 | } | |
585 | ||
586 | #define GPR_MASK 0xffffff0000000000 | |
587 | static inline int fadump_gpr_index(u64 id) | |
588 | { | |
589 | int i = -1; | |
590 | char str[3]; | |
591 | ||
592 | if ((id & GPR_MASK) == REG_ID("GPR")) { | |
593 | /* get the digits at the end */ | |
594 | id &= ~GPR_MASK; | |
595 | id >>= 24; | |
596 | str[2] = '\0'; | |
597 | str[1] = id & 0xff; | |
598 | str[0] = (id >> 8) & 0xff; | |
599 | sscanf(str, "%d", &i); | |
600 | if (i > 31) | |
601 | i = -1; | |
602 | } | |
603 | return i; | |
604 | } | |
605 | ||
606 | static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id, | |
607 | u64 reg_val) | |
608 | { | |
609 | int i; | |
610 | ||
611 | i = fadump_gpr_index(reg_id); | |
612 | if (i >= 0) | |
613 | regs->gpr[i] = (unsigned long)reg_val; | |
614 | else if (reg_id == REG_ID("NIA")) | |
615 | regs->nip = (unsigned long)reg_val; | |
616 | else if (reg_id == REG_ID("MSR")) | |
617 | regs->msr = (unsigned long)reg_val; | |
618 | else if (reg_id == REG_ID("CTR")) | |
619 | regs->ctr = (unsigned long)reg_val; | |
620 | else if (reg_id == REG_ID("LR")) | |
621 | regs->link = (unsigned long)reg_val; | |
622 | else if (reg_id == REG_ID("XER")) | |
623 | regs->xer = (unsigned long)reg_val; | |
624 | else if (reg_id == REG_ID("CR")) | |
625 | regs->ccr = (unsigned long)reg_val; | |
626 | else if (reg_id == REG_ID("DAR")) | |
627 | regs->dar = (unsigned long)reg_val; | |
628 | else if (reg_id == REG_ID("DSISR")) | |
629 | regs->dsisr = (unsigned long)reg_val; | |
630 | } | |
631 | ||
632 | static struct fadump_reg_entry* | |
633 | fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs) | |
634 | { | |
635 | memset(regs, 0, sizeof(struct pt_regs)); | |
636 | ||
408cddd9 HB |
637 | while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) { |
638 | fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id), | |
639 | be64_to_cpu(reg_entry->reg_value)); | |
ebaeb5ae MS |
640 | reg_entry++; |
641 | } | |
642 | reg_entry++; | |
643 | return reg_entry; | |
644 | } | |
645 | ||
ebaeb5ae MS |
646 | static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs) |
647 | { | |
648 | struct elf_prstatus prstatus; | |
649 | ||
650 | memset(&prstatus, 0, sizeof(prstatus)); | |
651 | /* | |
652 | * FIXME: How do i get PID? Do I really need it? | |
653 | * prstatus.pr_pid = ???? | |
654 | */ | |
655 | elf_core_copy_kernel_regs(&prstatus.pr_reg, regs); | |
22bd0177 HB |
656 | buf = append_elf_note(buf, CRASH_CORE_NOTE_NAME, NT_PRSTATUS, |
657 | &prstatus, sizeof(prstatus)); | |
ebaeb5ae MS |
658 | return buf; |
659 | } | |
660 | ||
661 | static void fadump_update_elfcore_header(char *bufp) | |
662 | { | |
663 | struct elfhdr *elf; | |
664 | struct elf_phdr *phdr; | |
665 | ||
666 | elf = (struct elfhdr *)bufp; | |
667 | bufp += sizeof(struct elfhdr); | |
668 | ||
669 | /* First note is a place holder for cpu notes info. */ | |
670 | phdr = (struct elf_phdr *)bufp; | |
671 | ||
672 | if (phdr->p_type == PT_NOTE) { | |
673 | phdr->p_paddr = fw_dump.cpu_notes_buf; | |
674 | phdr->p_offset = phdr->p_paddr; | |
675 | phdr->p_filesz = fw_dump.cpu_notes_buf_size; | |
676 | phdr->p_memsz = fw_dump.cpu_notes_buf_size; | |
677 | } | |
678 | return; | |
679 | } | |
680 | ||
681 | static void *fadump_cpu_notes_buf_alloc(unsigned long size) | |
682 | { | |
683 | void *vaddr; | |
684 | struct page *page; | |
685 | unsigned long order, count, i; | |
686 | ||
687 | order = get_order(size); | |
688 | vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order); | |
689 | if (!vaddr) | |
690 | return NULL; | |
691 | ||
692 | count = 1 << order; | |
693 | page = virt_to_page(vaddr); | |
694 | for (i = 0; i < count; i++) | |
695 | SetPageReserved(page + i); | |
696 | return vaddr; | |
697 | } | |
698 | ||
699 | static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size) | |
700 | { | |
701 | struct page *page; | |
702 | unsigned long order, count, i; | |
703 | ||
704 | order = get_order(size); | |
705 | count = 1 << order; | |
706 | page = virt_to_page(vaddr); | |
707 | for (i = 0; i < count; i++) | |
708 | ClearPageReserved(page + i); | |
709 | __free_pages(page, order); | |
710 | } | |
711 | ||
712 | /* | |
713 | * Read CPU state dump data and convert it into ELF notes. | |
714 | * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be | |
715 | * used to access the data to allow for additional fields to be added without | |
716 | * affecting compatibility. Each list of registers for a CPU starts with | |
717 | * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes, | |
718 | * 8 Byte ASCII identifier and 8 Byte register value. The register entry | |
719 | * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part | |
720 | * of register value. For more details refer to PAPR document. | |
721 | * | |
722 | * Only for the crashing cpu we ignore the CPU dump data and get exact | |
723 | * state from fadump crash info structure populated by first kernel at the | |
724 | * time of crash. | |
725 | */ | |
726 | static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm) | |
727 | { | |
728 | struct fadump_reg_save_area_header *reg_header; | |
729 | struct fadump_reg_entry *reg_entry; | |
730 | struct fadump_crash_info_header *fdh = NULL; | |
731 | void *vaddr; | |
732 | unsigned long addr; | |
733 | u32 num_cpus, *note_buf; | |
734 | struct pt_regs regs; | |
735 | int i, rc = 0, cpu = 0; | |
736 | ||
737 | if (!fdm->cpu_state_data.bytes_dumped) | |
738 | return -EINVAL; | |
739 | ||
408cddd9 | 740 | addr = be64_to_cpu(fdm->cpu_state_data.destination_address); |
ebaeb5ae MS |
741 | vaddr = __va(addr); |
742 | ||
743 | reg_header = vaddr; | |
408cddd9 | 744 | if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) { |
ebaeb5ae MS |
745 | printk(KERN_ERR "Unable to read register save area.\n"); |
746 | return -ENOENT; | |
747 | } | |
748 | pr_debug("--------CPU State Data------------\n"); | |
408cddd9 HB |
749 | pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number)); |
750 | pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset)); | |
ebaeb5ae | 751 | |
408cddd9 HB |
752 | vaddr += be32_to_cpu(reg_header->num_cpu_offset); |
753 | num_cpus = be32_to_cpu(*((__be32 *)(vaddr))); | |
ebaeb5ae MS |
754 | pr_debug("NumCpus : %u\n", num_cpus); |
755 | vaddr += sizeof(u32); | |
756 | reg_entry = (struct fadump_reg_entry *)vaddr; | |
757 | ||
758 | /* Allocate buffer to hold cpu crash notes. */ | |
759 | fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t); | |
760 | fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size); | |
761 | note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size); | |
762 | if (!note_buf) { | |
763 | printk(KERN_ERR "Failed to allocate 0x%lx bytes for " | |
764 | "cpu notes buffer\n", fw_dump.cpu_notes_buf_size); | |
765 | return -ENOMEM; | |
766 | } | |
767 | fw_dump.cpu_notes_buf = __pa(note_buf); | |
768 | ||
769 | pr_debug("Allocated buffer for cpu notes of size %ld at %p\n", | |
770 | (num_cpus * sizeof(note_buf_t)), note_buf); | |
771 | ||
772 | if (fw_dump.fadumphdr_addr) | |
773 | fdh = __va(fw_dump.fadumphdr_addr); | |
774 | ||
775 | for (i = 0; i < num_cpus; i++) { | |
408cddd9 | 776 | if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) { |
ebaeb5ae MS |
777 | printk(KERN_ERR "Unable to read CPU state data\n"); |
778 | rc = -ENOENT; | |
779 | goto error_out; | |
780 | } | |
781 | /* Lower 4 bytes of reg_value contains logical cpu id */ | |
408cddd9 | 782 | cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK; |
a0512164 | 783 | if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) { |
ebaeb5ae MS |
784 | SKIP_TO_NEXT_CPU(reg_entry); |
785 | continue; | |
786 | } | |
787 | pr_debug("Reading register data for cpu %d...\n", cpu); | |
788 | if (fdh && fdh->crashing_cpu == cpu) { | |
789 | regs = fdh->regs; | |
790 | note_buf = fadump_regs_to_elf_notes(note_buf, ®s); | |
791 | SKIP_TO_NEXT_CPU(reg_entry); | |
792 | } else { | |
793 | reg_entry++; | |
794 | reg_entry = fadump_read_registers(reg_entry, ®s); | |
795 | note_buf = fadump_regs_to_elf_notes(note_buf, ®s); | |
796 | } | |
797 | } | |
22bd0177 | 798 | final_note(note_buf); |
ebaeb5ae | 799 | |
b717d985 RS |
800 | if (fdh) { |
801 | pr_debug("Updating elfcore header (%llx) with cpu notes\n", | |
ebaeb5ae | 802 | fdh->elfcorehdr_addr); |
b717d985 RS |
803 | fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr)); |
804 | } | |
ebaeb5ae MS |
805 | return 0; |
806 | ||
807 | error_out: | |
808 | fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf), | |
809 | fw_dump.cpu_notes_buf_size); | |
810 | fw_dump.cpu_notes_buf = 0; | |
811 | fw_dump.cpu_notes_buf_size = 0; | |
812 | return rc; | |
813 | ||
814 | } | |
815 | ||
2df173d9 MS |
816 | /* |
817 | * Validate and process the dump data stored by firmware before exporting | |
818 | * it through '/proc/vmcore'. | |
819 | */ | |
820 | static int __init process_fadump(const struct fadump_mem_struct *fdm_active) | |
821 | { | |
822 | struct fadump_crash_info_header *fdh; | |
ebaeb5ae | 823 | int rc = 0; |
2df173d9 MS |
824 | |
825 | if (!fdm_active || !fw_dump.fadumphdr_addr) | |
826 | return -EINVAL; | |
827 | ||
828 | /* Check if the dump data is valid. */ | |
408cddd9 | 829 | if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) || |
ebaeb5ae | 830 | (fdm_active->cpu_state_data.error_flags != 0) || |
2df173d9 MS |
831 | (fdm_active->rmr_region.error_flags != 0)) { |
832 | printk(KERN_ERR "Dump taken by platform is not valid\n"); | |
833 | return -EINVAL; | |
834 | } | |
ebaeb5ae MS |
835 | if ((fdm_active->rmr_region.bytes_dumped != |
836 | fdm_active->rmr_region.source_len) || | |
837 | !fdm_active->cpu_state_data.bytes_dumped) { | |
2df173d9 MS |
838 | printk(KERN_ERR "Dump taken by platform is incomplete\n"); |
839 | return -EINVAL; | |
840 | } | |
841 | ||
842 | /* Validate the fadump crash info header */ | |
843 | fdh = __va(fw_dump.fadumphdr_addr); | |
844 | if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) { | |
845 | printk(KERN_ERR "Crash info header is not valid.\n"); | |
846 | return -EINVAL; | |
847 | } | |
848 | ||
ebaeb5ae MS |
849 | rc = fadump_build_cpu_notes(fdm_active); |
850 | if (rc) | |
851 | return rc; | |
852 | ||
2df173d9 MS |
853 | /* |
854 | * We are done validating dump info and elfcore header is now ready | |
855 | * to be exported. set elfcorehdr_addr so that vmcore module will | |
856 | * export the elfcore header through '/proc/vmcore'. | |
857 | */ | |
858 | elfcorehdr_addr = fdh->elfcorehdr_addr; | |
859 | ||
860 | return 0; | |
861 | } | |
862 | ||
863 | static inline void fadump_add_crash_memory(unsigned long long base, | |
864 | unsigned long long end) | |
865 | { | |
866 | if (base == end) | |
867 | return; | |
868 | ||
869 | pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n", | |
870 | crash_mem_ranges, base, end - 1, (end - base)); | |
871 | crash_memory_ranges[crash_mem_ranges].base = base; | |
872 | crash_memory_ranges[crash_mem_ranges].size = end - base; | |
873 | crash_mem_ranges++; | |
874 | } | |
875 | ||
876 | static void fadump_exclude_reserved_area(unsigned long long start, | |
877 | unsigned long long end) | |
878 | { | |
879 | unsigned long long ra_start, ra_end; | |
880 | ||
881 | ra_start = fw_dump.reserve_dump_area_start; | |
882 | ra_end = ra_start + fw_dump.reserve_dump_area_size; | |
883 | ||
884 | if ((ra_start < end) && (ra_end > start)) { | |
885 | if ((start < ra_start) && (end > ra_end)) { | |
886 | fadump_add_crash_memory(start, ra_start); | |
887 | fadump_add_crash_memory(ra_end, end); | |
888 | } else if (start < ra_start) { | |
889 | fadump_add_crash_memory(start, ra_start); | |
890 | } else if (ra_end < end) { | |
891 | fadump_add_crash_memory(ra_end, end); | |
892 | } | |
893 | } else | |
894 | fadump_add_crash_memory(start, end); | |
895 | } | |
896 | ||
897 | static int fadump_init_elfcore_header(char *bufp) | |
898 | { | |
899 | struct elfhdr *elf; | |
900 | ||
901 | elf = (struct elfhdr *) bufp; | |
902 | bufp += sizeof(struct elfhdr); | |
903 | memcpy(elf->e_ident, ELFMAG, SELFMAG); | |
904 | elf->e_ident[EI_CLASS] = ELF_CLASS; | |
905 | elf->e_ident[EI_DATA] = ELF_DATA; | |
906 | elf->e_ident[EI_VERSION] = EV_CURRENT; | |
907 | elf->e_ident[EI_OSABI] = ELF_OSABI; | |
908 | memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); | |
909 | elf->e_type = ET_CORE; | |
910 | elf->e_machine = ELF_ARCH; | |
911 | elf->e_version = EV_CURRENT; | |
912 | elf->e_entry = 0; | |
913 | elf->e_phoff = sizeof(struct elfhdr); | |
914 | elf->e_shoff = 0; | |
d8bced27 DA |
915 | #if defined(_CALL_ELF) |
916 | elf->e_flags = _CALL_ELF; | |
917 | #else | |
918 | elf->e_flags = 0; | |
919 | #endif | |
2df173d9 MS |
920 | elf->e_ehsize = sizeof(struct elfhdr); |
921 | elf->e_phentsize = sizeof(struct elf_phdr); | |
922 | elf->e_phnum = 0; | |
923 | elf->e_shentsize = 0; | |
924 | elf->e_shnum = 0; | |
925 | elf->e_shstrndx = 0; | |
926 | ||
927 | return 0; | |
928 | } | |
929 | ||
930 | /* | |
931 | * Traverse through memblock structure and setup crash memory ranges. These | |
932 | * ranges will be used create PT_LOAD program headers in elfcore header. | |
933 | */ | |
934 | static void fadump_setup_crash_memory_ranges(void) | |
935 | { | |
936 | struct memblock_region *reg; | |
937 | unsigned long long start, end; | |
938 | ||
939 | pr_debug("Setup crash memory ranges.\n"); | |
940 | crash_mem_ranges = 0; | |
941 | /* | |
942 | * add the first memory chunk (RMA_START through boot_memory_size) as | |
943 | * a separate memory chunk. The reason is, at the time crash firmware | |
944 | * will move the content of this memory chunk to different location | |
945 | * specified during fadump registration. We need to create a separate | |
946 | * program header for this chunk with the correct offset. | |
947 | */ | |
948 | fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size); | |
949 | ||
950 | for_each_memblock(memory, reg) { | |
951 | start = (unsigned long long)reg->base; | |
952 | end = start + (unsigned long long)reg->size; | |
a77af552 HB |
953 | |
954 | /* | |
955 | * skip the first memory chunk that is already added (RMA_START | |
956 | * through boot_memory_size). This logic needs a relook if and | |
957 | * when RMA_START changes to a non-zero value. | |
958 | */ | |
959 | BUILD_BUG_ON(RMA_START != 0); | |
960 | if (start < fw_dump.boot_memory_size) { | |
961 | if (end > fw_dump.boot_memory_size) | |
962 | start = fw_dump.boot_memory_size; | |
963 | else | |
964 | continue; | |
965 | } | |
2df173d9 MS |
966 | |
967 | /* add this range excluding the reserved dump area. */ | |
968 | fadump_exclude_reserved_area(start, end); | |
969 | } | |
970 | } | |
971 | ||
d34c5f26 MS |
972 | /* |
973 | * If the given physical address falls within the boot memory region then | |
974 | * return the relocated address that points to the dump region reserved | |
975 | * for saving initial boot memory contents. | |
976 | */ | |
977 | static inline unsigned long fadump_relocate(unsigned long paddr) | |
978 | { | |
979 | if (paddr > RMA_START && paddr < fw_dump.boot_memory_size) | |
408cddd9 | 980 | return be64_to_cpu(fdm.rmr_region.destination_address) + paddr; |
d34c5f26 MS |
981 | else |
982 | return paddr; | |
983 | } | |
984 | ||
2df173d9 MS |
985 | static int fadump_create_elfcore_headers(char *bufp) |
986 | { | |
987 | struct elfhdr *elf; | |
988 | struct elf_phdr *phdr; | |
989 | int i; | |
990 | ||
991 | fadump_init_elfcore_header(bufp); | |
992 | elf = (struct elfhdr *)bufp; | |
993 | bufp += sizeof(struct elfhdr); | |
994 | ||
ebaeb5ae MS |
995 | /* |
996 | * setup ELF PT_NOTE, place holder for cpu notes info. The notes info | |
997 | * will be populated during second kernel boot after crash. Hence | |
998 | * this PT_NOTE will always be the first elf note. | |
999 | * | |
1000 | * NOTE: Any new ELF note addition should be placed after this note. | |
1001 | */ | |
1002 | phdr = (struct elf_phdr *)bufp; | |
1003 | bufp += sizeof(struct elf_phdr); | |
1004 | phdr->p_type = PT_NOTE; | |
1005 | phdr->p_flags = 0; | |
1006 | phdr->p_vaddr = 0; | |
1007 | phdr->p_align = 0; | |
1008 | ||
1009 | phdr->p_offset = 0; | |
1010 | phdr->p_paddr = 0; | |
1011 | phdr->p_filesz = 0; | |
1012 | phdr->p_memsz = 0; | |
1013 | ||
1014 | (elf->e_phnum)++; | |
1015 | ||
d34c5f26 MS |
1016 | /* setup ELF PT_NOTE for vmcoreinfo */ |
1017 | phdr = (struct elf_phdr *)bufp; | |
1018 | bufp += sizeof(struct elf_phdr); | |
1019 | phdr->p_type = PT_NOTE; | |
1020 | phdr->p_flags = 0; | |
1021 | phdr->p_vaddr = 0; | |
1022 | phdr->p_align = 0; | |
1023 | ||
1024 | phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note()); | |
1025 | phdr->p_offset = phdr->p_paddr; | |
5203f499 | 1026 | phdr->p_memsz = phdr->p_filesz = VMCOREINFO_NOTE_SIZE; |
d34c5f26 MS |
1027 | |
1028 | /* Increment number of program headers. */ | |
1029 | (elf->e_phnum)++; | |
1030 | ||
2df173d9 MS |
1031 | /* setup PT_LOAD sections. */ |
1032 | ||
1033 | for (i = 0; i < crash_mem_ranges; i++) { | |
1034 | unsigned long long mbase, msize; | |
1035 | mbase = crash_memory_ranges[i].base; | |
1036 | msize = crash_memory_ranges[i].size; | |
1037 | ||
1038 | if (!msize) | |
1039 | continue; | |
1040 | ||
1041 | phdr = (struct elf_phdr *)bufp; | |
1042 | bufp += sizeof(struct elf_phdr); | |
1043 | phdr->p_type = PT_LOAD; | |
1044 | phdr->p_flags = PF_R|PF_W|PF_X; | |
1045 | phdr->p_offset = mbase; | |
1046 | ||
1047 | if (mbase == RMA_START) { | |
1048 | /* | |
1049 | * The entire RMA region will be moved by firmware | |
1050 | * to the specified destination_address. Hence set | |
1051 | * the correct offset. | |
1052 | */ | |
408cddd9 | 1053 | phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address); |
2df173d9 MS |
1054 | } |
1055 | ||
1056 | phdr->p_paddr = mbase; | |
1057 | phdr->p_vaddr = (unsigned long)__va(mbase); | |
1058 | phdr->p_filesz = msize; | |
1059 | phdr->p_memsz = msize; | |
1060 | phdr->p_align = 0; | |
1061 | ||
1062 | /* Increment number of program headers. */ | |
1063 | (elf->e_phnum)++; | |
1064 | } | |
1065 | return 0; | |
1066 | } | |
1067 | ||
1068 | static unsigned long init_fadump_header(unsigned long addr) | |
1069 | { | |
1070 | struct fadump_crash_info_header *fdh; | |
1071 | ||
1072 | if (!addr) | |
1073 | return 0; | |
1074 | ||
1075 | fw_dump.fadumphdr_addr = addr; | |
1076 | fdh = __va(addr); | |
1077 | addr += sizeof(struct fadump_crash_info_header); | |
1078 | ||
1079 | memset(fdh, 0, sizeof(struct fadump_crash_info_header)); | |
1080 | fdh->magic_number = FADUMP_CRASH_INFO_MAGIC; | |
1081 | fdh->elfcorehdr_addr = addr; | |
ebaeb5ae MS |
1082 | /* We will set the crashing cpu id in crash_fadump() during crash. */ |
1083 | fdh->crashing_cpu = CPU_UNKNOWN; | |
2df173d9 MS |
1084 | |
1085 | return addr; | |
1086 | } | |
1087 | ||
98b8cd7f | 1088 | static int register_fadump(void) |
3ccc00a7 | 1089 | { |
2df173d9 MS |
1090 | unsigned long addr; |
1091 | void *vaddr; | |
1092 | ||
3ccc00a7 MS |
1093 | /* |
1094 | * If no memory is reserved then we can not register for firmware- | |
1095 | * assisted dump. | |
1096 | */ | |
1097 | if (!fw_dump.reserve_dump_area_size) | |
98b8cd7f | 1098 | return -ENODEV; |
3ccc00a7 | 1099 | |
2df173d9 MS |
1100 | fadump_setup_crash_memory_ranges(); |
1101 | ||
408cddd9 | 1102 | addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len); |
2df173d9 MS |
1103 | /* Initialize fadump crash info header. */ |
1104 | addr = init_fadump_header(addr); | |
1105 | vaddr = __va(addr); | |
1106 | ||
1107 | pr_debug("Creating ELF core headers at %#016lx\n", addr); | |
1108 | fadump_create_elfcore_headers(vaddr); | |
1109 | ||
3ccc00a7 | 1110 | /* register the future kernel dump with firmware. */ |
98b8cd7f | 1111 | return register_fw_dump(&fdm); |
3ccc00a7 MS |
1112 | } |
1113 | ||
1114 | static int fadump_unregister_dump(struct fadump_mem_struct *fdm) | |
1115 | { | |
1116 | int rc = 0; | |
1117 | unsigned int wait_time; | |
1118 | ||
1119 | pr_debug("Un-register firmware-assisted dump\n"); | |
1120 | ||
1121 | /* TODO: Add upper time limit for the delay */ | |
1122 | do { | |
1123 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, | |
1124 | FADUMP_UNREGISTER, fdm, | |
1125 | sizeof(struct fadump_mem_struct)); | |
1126 | ||
1127 | wait_time = rtas_busy_delay_time(rc); | |
1128 | if (wait_time) | |
1129 | mdelay(wait_time); | |
1130 | } while (wait_time); | |
1131 | ||
1132 | if (rc) { | |
1133 | printk(KERN_ERR "Failed to un-register firmware-assisted dump." | |
1134 | " unexpected error(%d).\n", rc); | |
1135 | return rc; | |
1136 | } | |
1137 | fw_dump.dump_registered = 0; | |
1138 | return 0; | |
1139 | } | |
1140 | ||
b500afff MS |
1141 | static int fadump_invalidate_dump(struct fadump_mem_struct *fdm) |
1142 | { | |
1143 | int rc = 0; | |
1144 | unsigned int wait_time; | |
1145 | ||
1146 | pr_debug("Invalidating firmware-assisted dump registration\n"); | |
1147 | ||
1148 | /* TODO: Add upper time limit for the delay */ | |
1149 | do { | |
1150 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, | |
1151 | FADUMP_INVALIDATE, fdm, | |
1152 | sizeof(struct fadump_mem_struct)); | |
1153 | ||
1154 | wait_time = rtas_busy_delay_time(rc); | |
1155 | if (wait_time) | |
1156 | mdelay(wait_time); | |
1157 | } while (wait_time); | |
1158 | ||
1159 | if (rc) { | |
4a03749f | 1160 | pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc); |
b5b1cfc5 | 1161 | return rc; |
b500afff MS |
1162 | } |
1163 | fw_dump.dump_active = 0; | |
1164 | fdm_active = NULL; | |
1165 | return 0; | |
1166 | } | |
1167 | ||
1168 | void fadump_cleanup(void) | |
1169 | { | |
1170 | /* Invalidate the registration only if dump is active. */ | |
1171 | if (fw_dump.dump_active) { | |
1172 | init_fadump_mem_struct(&fdm, | |
408cddd9 | 1173 | be64_to_cpu(fdm_active->cpu_state_data.destination_address)); |
b500afff MS |
1174 | fadump_invalidate_dump(&fdm); |
1175 | } | |
1176 | } | |
1177 | ||
68fa6478 HB |
1178 | static void fadump_free_reserved_memory(unsigned long start_pfn, |
1179 | unsigned long end_pfn) | |
1180 | { | |
1181 | unsigned long pfn; | |
1182 | unsigned long time_limit = jiffies + HZ; | |
1183 | ||
1184 | pr_info("freeing reserved memory (0x%llx - 0x%llx)\n", | |
1185 | PFN_PHYS(start_pfn), PFN_PHYS(end_pfn)); | |
1186 | ||
1187 | for (pfn = start_pfn; pfn < end_pfn; pfn++) { | |
1188 | free_reserved_page(pfn_to_page(pfn)); | |
1189 | ||
1190 | if (time_after(jiffies, time_limit)) { | |
1191 | cond_resched(); | |
1192 | time_limit = jiffies + HZ; | |
1193 | } | |
1194 | } | |
1195 | } | |
1196 | ||
1197 | /* | |
1198 | * Skip memory holes and free memory that was actually reserved. | |
1199 | */ | |
1200 | static void fadump_release_reserved_area(unsigned long start, unsigned long end) | |
1201 | { | |
1202 | struct memblock_region *reg; | |
1203 | unsigned long tstart, tend; | |
1204 | unsigned long start_pfn = PHYS_PFN(start); | |
1205 | unsigned long end_pfn = PHYS_PFN(end); | |
1206 | ||
1207 | for_each_memblock(memory, reg) { | |
1208 | tstart = max(start_pfn, memblock_region_memory_base_pfn(reg)); | |
1209 | tend = min(end_pfn, memblock_region_memory_end_pfn(reg)); | |
1210 | if (tstart < tend) { | |
1211 | fadump_free_reserved_memory(tstart, tend); | |
1212 | ||
1213 | if (tend == end_pfn) | |
1214 | break; | |
1215 | ||
1216 | start_pfn = tend + 1; | |
1217 | } | |
1218 | } | |
1219 | } | |
1220 | ||
b500afff MS |
1221 | /* |
1222 | * Release the memory that was reserved in early boot to preserve the memory | |
1223 | * contents. The released memory will be available for general use. | |
1224 | */ | |
1225 | static void fadump_release_memory(unsigned long begin, unsigned long end) | |
1226 | { | |
b500afff MS |
1227 | unsigned long ra_start, ra_end; |
1228 | ||
1229 | ra_start = fw_dump.reserve_dump_area_start; | |
1230 | ra_end = ra_start + fw_dump.reserve_dump_area_size; | |
1231 | ||
68fa6478 HB |
1232 | /* |
1233 | * exclude the dump reserve area. Will reuse it for next | |
1234 | * fadump registration. | |
1235 | */ | |
1236 | if (begin < ra_end && end > ra_start) { | |
1237 | if (begin < ra_start) | |
1238 | fadump_release_reserved_area(begin, ra_start); | |
1239 | if (end > ra_end) | |
1240 | fadump_release_reserved_area(ra_end, end); | |
1241 | } else | |
1242 | fadump_release_reserved_area(begin, end); | |
b500afff MS |
1243 | } |
1244 | ||
1245 | static void fadump_invalidate_release_mem(void) | |
1246 | { | |
1247 | unsigned long reserved_area_start, reserved_area_end; | |
1248 | unsigned long destination_address; | |
1249 | ||
1250 | mutex_lock(&fadump_mutex); | |
1251 | if (!fw_dump.dump_active) { | |
1252 | mutex_unlock(&fadump_mutex); | |
1253 | return; | |
1254 | } | |
1255 | ||
408cddd9 | 1256 | destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address); |
b500afff MS |
1257 | fadump_cleanup(); |
1258 | mutex_unlock(&fadump_mutex); | |
1259 | ||
1260 | /* | |
1261 | * Save the current reserved memory bounds we will require them | |
1262 | * later for releasing the memory for general use. | |
1263 | */ | |
1264 | reserved_area_start = fw_dump.reserve_dump_area_start; | |
1265 | reserved_area_end = reserved_area_start + | |
1266 | fw_dump.reserve_dump_area_size; | |
1267 | /* | |
1268 | * Setup reserve_dump_area_start and its size so that we can | |
1269 | * reuse this reserved memory for Re-registration. | |
1270 | */ | |
1271 | fw_dump.reserve_dump_area_start = destination_address; | |
1272 | fw_dump.reserve_dump_area_size = get_fadump_area_size(); | |
1273 | ||
1274 | fadump_release_memory(reserved_area_start, reserved_area_end); | |
1275 | if (fw_dump.cpu_notes_buf) { | |
1276 | fadump_cpu_notes_buf_free( | |
1277 | (unsigned long)__va(fw_dump.cpu_notes_buf), | |
1278 | fw_dump.cpu_notes_buf_size); | |
1279 | fw_dump.cpu_notes_buf = 0; | |
1280 | fw_dump.cpu_notes_buf_size = 0; | |
1281 | } | |
1282 | /* Initialize the kernel dump memory structure for FAD registration. */ | |
1283 | init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start); | |
1284 | } | |
1285 | ||
1286 | static ssize_t fadump_release_memory_store(struct kobject *kobj, | |
1287 | struct kobj_attribute *attr, | |
1288 | const char *buf, size_t count) | |
1289 | { | |
dcdc4679 MS |
1290 | int input = -1; |
1291 | ||
b500afff MS |
1292 | if (!fw_dump.dump_active) |
1293 | return -EPERM; | |
1294 | ||
dcdc4679 MS |
1295 | if (kstrtoint(buf, 0, &input)) |
1296 | return -EINVAL; | |
1297 | ||
1298 | if (input == 1) { | |
b500afff MS |
1299 | /* |
1300 | * Take away the '/proc/vmcore'. We are releasing the dump | |
1301 | * memory, hence it will not be valid anymore. | |
1302 | */ | |
2685f826 | 1303 | #ifdef CONFIG_PROC_VMCORE |
b500afff | 1304 | vmcore_cleanup(); |
2685f826 | 1305 | #endif |
b500afff MS |
1306 | fadump_invalidate_release_mem(); |
1307 | ||
1308 | } else | |
1309 | return -EINVAL; | |
1310 | return count; | |
1311 | } | |
1312 | ||
3ccc00a7 MS |
1313 | static ssize_t fadump_enabled_show(struct kobject *kobj, |
1314 | struct kobj_attribute *attr, | |
1315 | char *buf) | |
1316 | { | |
1317 | return sprintf(buf, "%d\n", fw_dump.fadump_enabled); | |
1318 | } | |
1319 | ||
1320 | static ssize_t fadump_register_show(struct kobject *kobj, | |
1321 | struct kobj_attribute *attr, | |
1322 | char *buf) | |
1323 | { | |
1324 | return sprintf(buf, "%d\n", fw_dump.dump_registered); | |
1325 | } | |
1326 | ||
1327 | static ssize_t fadump_register_store(struct kobject *kobj, | |
1328 | struct kobj_attribute *attr, | |
1329 | const char *buf, size_t count) | |
1330 | { | |
1331 | int ret = 0; | |
dcdc4679 | 1332 | int input = -1; |
3ccc00a7 MS |
1333 | |
1334 | if (!fw_dump.fadump_enabled || fdm_active) | |
1335 | return -EPERM; | |
1336 | ||
dcdc4679 MS |
1337 | if (kstrtoint(buf, 0, &input)) |
1338 | return -EINVAL; | |
1339 | ||
3ccc00a7 MS |
1340 | mutex_lock(&fadump_mutex); |
1341 | ||
dcdc4679 MS |
1342 | switch (input) { |
1343 | case 0: | |
3ccc00a7 | 1344 | if (fw_dump.dump_registered == 0) { |
3ccc00a7 MS |
1345 | goto unlock_out; |
1346 | } | |
1347 | /* Un-register Firmware-assisted dump */ | |
1348 | fadump_unregister_dump(&fdm); | |
1349 | break; | |
dcdc4679 | 1350 | case 1: |
3ccc00a7 | 1351 | if (fw_dump.dump_registered == 1) { |
98b8cd7f | 1352 | ret = -EEXIST; |
3ccc00a7 MS |
1353 | goto unlock_out; |
1354 | } | |
1355 | /* Register Firmware-assisted dump */ | |
98b8cd7f | 1356 | ret = register_fadump(); |
3ccc00a7 MS |
1357 | break; |
1358 | default: | |
1359 | ret = -EINVAL; | |
1360 | break; | |
1361 | } | |
1362 | ||
1363 | unlock_out: | |
1364 | mutex_unlock(&fadump_mutex); | |
1365 | return ret < 0 ? ret : count; | |
1366 | } | |
1367 | ||
1368 | static int fadump_region_show(struct seq_file *m, void *private) | |
1369 | { | |
1370 | const struct fadump_mem_struct *fdm_ptr; | |
1371 | ||
1372 | if (!fw_dump.fadump_enabled) | |
1373 | return 0; | |
1374 | ||
b500afff | 1375 | mutex_lock(&fadump_mutex); |
3ccc00a7 MS |
1376 | if (fdm_active) |
1377 | fdm_ptr = fdm_active; | |
b500afff MS |
1378 | else { |
1379 | mutex_unlock(&fadump_mutex); | |
3ccc00a7 | 1380 | fdm_ptr = &fdm; |
b500afff | 1381 | } |
3ccc00a7 MS |
1382 | |
1383 | seq_printf(m, | |
1384 | "CPU : [%#016llx-%#016llx] %#llx bytes, " | |
1385 | "Dumped: %#llx\n", | |
408cddd9 HB |
1386 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address), |
1387 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) + | |
1388 | be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1, | |
1389 | be64_to_cpu(fdm_ptr->cpu_state_data.source_len), | |
1390 | be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped)); | |
3ccc00a7 MS |
1391 | seq_printf(m, |
1392 | "HPTE: [%#016llx-%#016llx] %#llx bytes, " | |
1393 | "Dumped: %#llx\n", | |
408cddd9 HB |
1394 | be64_to_cpu(fdm_ptr->hpte_region.destination_address), |
1395 | be64_to_cpu(fdm_ptr->hpte_region.destination_address) + | |
1396 | be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1, | |
1397 | be64_to_cpu(fdm_ptr->hpte_region.source_len), | |
1398 | be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped)); | |
3ccc00a7 MS |
1399 | seq_printf(m, |
1400 | "DUMP: [%#016llx-%#016llx] %#llx bytes, " | |
1401 | "Dumped: %#llx\n", | |
408cddd9 HB |
1402 | be64_to_cpu(fdm_ptr->rmr_region.destination_address), |
1403 | be64_to_cpu(fdm_ptr->rmr_region.destination_address) + | |
1404 | be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1, | |
1405 | be64_to_cpu(fdm_ptr->rmr_region.source_len), | |
1406 | be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped)); | |
3ccc00a7 MS |
1407 | |
1408 | if (!fdm_active || | |
1409 | (fw_dump.reserve_dump_area_start == | |
408cddd9 | 1410 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address))) |
b500afff | 1411 | goto out; |
3ccc00a7 MS |
1412 | |
1413 | /* Dump is active. Show reserved memory region. */ | |
1414 | seq_printf(m, | |
1415 | " : [%#016llx-%#016llx] %#llx bytes, " | |
1416 | "Dumped: %#llx\n", | |
1417 | (unsigned long long)fw_dump.reserve_dump_area_start, | |
408cddd9 HB |
1418 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1, |
1419 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - | |
3ccc00a7 | 1420 | fw_dump.reserve_dump_area_start, |
408cddd9 | 1421 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - |
3ccc00a7 | 1422 | fw_dump.reserve_dump_area_start); |
b500afff MS |
1423 | out: |
1424 | if (fdm_active) | |
1425 | mutex_unlock(&fadump_mutex); | |
3ccc00a7 MS |
1426 | return 0; |
1427 | } | |
1428 | ||
b500afff MS |
1429 | static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem, |
1430 | 0200, NULL, | |
1431 | fadump_release_memory_store); | |
3ccc00a7 MS |
1432 | static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled, |
1433 | 0444, fadump_enabled_show, | |
1434 | NULL); | |
1435 | static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered, | |
1436 | 0644, fadump_register_show, | |
1437 | fadump_register_store); | |
1438 | ||
1439 | static int fadump_region_open(struct inode *inode, struct file *file) | |
1440 | { | |
1441 | return single_open(file, fadump_region_show, inode->i_private); | |
1442 | } | |
1443 | ||
1444 | static const struct file_operations fadump_region_fops = { | |
1445 | .open = fadump_region_open, | |
1446 | .read = seq_read, | |
1447 | .llseek = seq_lseek, | |
1448 | .release = single_release, | |
1449 | }; | |
1450 | ||
1451 | static void fadump_init_files(void) | |
1452 | { | |
1453 | struct dentry *debugfs_file; | |
1454 | int rc = 0; | |
1455 | ||
1456 | rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr); | |
1457 | if (rc) | |
1458 | printk(KERN_ERR "fadump: unable to create sysfs file" | |
1459 | " fadump_enabled (%d)\n", rc); | |
1460 | ||
1461 | rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr); | |
1462 | if (rc) | |
1463 | printk(KERN_ERR "fadump: unable to create sysfs file" | |
1464 | " fadump_registered (%d)\n", rc); | |
1465 | ||
1466 | debugfs_file = debugfs_create_file("fadump_region", 0444, | |
1467 | powerpc_debugfs_root, NULL, | |
1468 | &fadump_region_fops); | |
1469 | if (!debugfs_file) | |
1470 | printk(KERN_ERR "fadump: unable to create debugfs file" | |
1471 | " fadump_region\n"); | |
b500afff MS |
1472 | |
1473 | if (fw_dump.dump_active) { | |
1474 | rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr); | |
1475 | if (rc) | |
1476 | printk(KERN_ERR "fadump: unable to create sysfs file" | |
1477 | " fadump_release_mem (%d)\n", rc); | |
1478 | } | |
3ccc00a7 MS |
1479 | return; |
1480 | } | |
1481 | ||
1482 | /* | |
1483 | * Prepare for firmware-assisted dump. | |
1484 | */ | |
1485 | int __init setup_fadump(void) | |
1486 | { | |
1487 | if (!fw_dump.fadump_enabled) | |
1488 | return 0; | |
1489 | ||
1490 | if (!fw_dump.fadump_supported) { | |
1491 | printk(KERN_ERR "Firmware-assisted dump is not supported on" | |
1492 | " this hardware\n"); | |
1493 | return 0; | |
1494 | } | |
1495 | ||
1496 | fadump_show_config(); | |
2df173d9 MS |
1497 | /* |
1498 | * If dump data is available then see if it is valid and prepare for | |
1499 | * saving it to the disk. | |
1500 | */ | |
b500afff MS |
1501 | if (fw_dump.dump_active) { |
1502 | /* | |
1503 | * if dump process fails then invalidate the registration | |
1504 | * and release memory before proceeding for re-registration. | |
1505 | */ | |
1506 | if (process_fadump(fdm_active) < 0) | |
1507 | fadump_invalidate_release_mem(); | |
1508 | } | |
3ccc00a7 | 1509 | /* Initialize the kernel dump memory structure for FAD registration. */ |
2df173d9 | 1510 | else if (fw_dump.reserve_dump_area_size) |
3ccc00a7 MS |
1511 | init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start); |
1512 | fadump_init_files(); | |
1513 | ||
1514 | return 1; | |
1515 | } | |
1516 | subsys_initcall(setup_fadump); |