]>
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 MS |
32 | #include <linux/delay.h> |
33 | #include <linux/debugfs.h> | |
34 | #include <linux/seq_file.h> | |
2df173d9 | 35 | #include <linux/crash_dump.h> |
b500afff MS |
36 | #include <linux/kobject.h> |
37 | #include <linux/sysfs.h> | |
eb39c880 MS |
38 | |
39 | #include <asm/page.h> | |
40 | #include <asm/prom.h> | |
41 | #include <asm/rtas.h> | |
42 | #include <asm/fadump.h> | |
cad3c834 SR |
43 | #include <asm/debug.h> |
44 | #include <asm/setup.h> | |
eb39c880 MS |
45 | |
46 | static struct fw_dump fw_dump; | |
3ccc00a7 MS |
47 | static struct fadump_mem_struct fdm; |
48 | static const struct fadump_mem_struct *fdm_active; | |
49 | ||
50 | static DEFINE_MUTEX(fadump_mutex); | |
2df173d9 MS |
51 | struct fad_crash_memory_ranges crash_memory_ranges[INIT_CRASHMEM_RANGES]; |
52 | int crash_mem_ranges; | |
eb39c880 MS |
53 | |
54 | /* Scan the Firmware Assisted dump configuration details. */ | |
55 | int __init early_init_dt_scan_fw_dump(unsigned long node, | |
56 | const char *uname, int depth, void *data) | |
57 | { | |
9d0c4dfe | 58 | const __be32 *sections; |
eb39c880 | 59 | int i, num_sections; |
9d0c4dfe | 60 | int size; |
408cddd9 | 61 | const __be32 *token; |
eb39c880 MS |
62 | |
63 | if (depth != 1 || strcmp(uname, "rtas") != 0) | |
64 | return 0; | |
65 | ||
66 | /* | |
67 | * Check if Firmware Assisted dump is supported. if yes, check | |
68 | * if dump has been initiated on last reboot. | |
69 | */ | |
70 | token = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL); | |
71 | if (!token) | |
a7d04317 | 72 | return 1; |
eb39c880 MS |
73 | |
74 | fw_dump.fadump_supported = 1; | |
408cddd9 | 75 | fw_dump.ibm_configure_kernel_dump = be32_to_cpu(*token); |
eb39c880 MS |
76 | |
77 | /* | |
78 | * The 'ibm,kernel-dump' rtas node is present only if there is | |
79 | * dump data waiting for us. | |
80 | */ | |
3ccc00a7 MS |
81 | fdm_active = of_get_flat_dt_prop(node, "ibm,kernel-dump", NULL); |
82 | if (fdm_active) | |
eb39c880 MS |
83 | fw_dump.dump_active = 1; |
84 | ||
85 | /* Get the sizes required to store dump data for the firmware provided | |
86 | * dump sections. | |
87 | * For each dump section type supported, a 32bit cell which defines | |
88 | * the ID of a supported section followed by two 32 bit cells which | |
89 | * gives teh size of the section in bytes. | |
90 | */ | |
91 | sections = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes", | |
92 | &size); | |
93 | ||
94 | if (!sections) | |
a7d04317 | 95 | return 1; |
eb39c880 MS |
96 | |
97 | num_sections = size / (3 * sizeof(u32)); | |
98 | ||
99 | for (i = 0; i < num_sections; i++, sections += 3) { | |
100 | u32 type = (u32)of_read_number(sections, 1); | |
101 | ||
102 | switch (type) { | |
103 | case FADUMP_CPU_STATE_DATA: | |
104 | fw_dump.cpu_state_data_size = | |
105 | of_read_ulong(§ions[1], 2); | |
106 | break; | |
107 | case FADUMP_HPTE_REGION: | |
108 | fw_dump.hpte_region_size = | |
109 | of_read_ulong(§ions[1], 2); | |
110 | break; | |
111 | } | |
112 | } | |
a7d04317 | 113 | |
eb39c880 MS |
114 | return 1; |
115 | } | |
116 | ||
3ccc00a7 MS |
117 | int is_fadump_active(void) |
118 | { | |
119 | return fw_dump.dump_active; | |
120 | } | |
121 | ||
122 | /* Print firmware assisted dump configurations for debugging purpose. */ | |
123 | static void fadump_show_config(void) | |
124 | { | |
125 | pr_debug("Support for firmware-assisted dump (fadump): %s\n", | |
126 | (fw_dump.fadump_supported ? "present" : "no support")); | |
127 | ||
128 | if (!fw_dump.fadump_supported) | |
129 | return; | |
130 | ||
131 | pr_debug("Fadump enabled : %s\n", | |
132 | (fw_dump.fadump_enabled ? "yes" : "no")); | |
133 | pr_debug("Dump Active : %s\n", | |
134 | (fw_dump.dump_active ? "yes" : "no")); | |
135 | pr_debug("Dump section sizes:\n"); | |
136 | pr_debug(" CPU state data size: %lx\n", fw_dump.cpu_state_data_size); | |
137 | pr_debug(" HPTE region size : %lx\n", fw_dump.hpte_region_size); | |
138 | pr_debug("Boot memory size : %lx\n", fw_dump.boot_memory_size); | |
139 | } | |
140 | ||
141 | static unsigned long init_fadump_mem_struct(struct fadump_mem_struct *fdm, | |
142 | unsigned long addr) | |
143 | { | |
144 | if (!fdm) | |
145 | return 0; | |
146 | ||
147 | memset(fdm, 0, sizeof(struct fadump_mem_struct)); | |
148 | addr = addr & PAGE_MASK; | |
149 | ||
408cddd9 HB |
150 | fdm->header.dump_format_version = cpu_to_be32(0x00000001); |
151 | fdm->header.dump_num_sections = cpu_to_be16(3); | |
3ccc00a7 MS |
152 | fdm->header.dump_status_flag = 0; |
153 | fdm->header.offset_first_dump_section = | |
408cddd9 | 154 | cpu_to_be32((u32)offsetof(struct fadump_mem_struct, cpu_state_data)); |
3ccc00a7 MS |
155 | |
156 | /* | |
157 | * Fields for disk dump option. | |
158 | * We are not using disk dump option, hence set these fields to 0. | |
159 | */ | |
160 | fdm->header.dd_block_size = 0; | |
161 | fdm->header.dd_block_offset = 0; | |
162 | fdm->header.dd_num_blocks = 0; | |
163 | fdm->header.dd_offset_disk_path = 0; | |
164 | ||
165 | /* set 0 to disable an automatic dump-reboot. */ | |
166 | fdm->header.max_time_auto = 0; | |
167 | ||
168 | /* Kernel dump sections */ | |
169 | /* cpu state data section. */ | |
408cddd9 HB |
170 | fdm->cpu_state_data.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG); |
171 | fdm->cpu_state_data.source_data_type = cpu_to_be16(FADUMP_CPU_STATE_DATA); | |
3ccc00a7 | 172 | fdm->cpu_state_data.source_address = 0; |
408cddd9 HB |
173 | fdm->cpu_state_data.source_len = cpu_to_be64(fw_dump.cpu_state_data_size); |
174 | fdm->cpu_state_data.destination_address = cpu_to_be64(addr); | |
3ccc00a7 MS |
175 | addr += fw_dump.cpu_state_data_size; |
176 | ||
177 | /* hpte region section */ | |
408cddd9 HB |
178 | fdm->hpte_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG); |
179 | fdm->hpte_region.source_data_type = cpu_to_be16(FADUMP_HPTE_REGION); | |
3ccc00a7 | 180 | fdm->hpte_region.source_address = 0; |
408cddd9 HB |
181 | fdm->hpte_region.source_len = cpu_to_be64(fw_dump.hpte_region_size); |
182 | fdm->hpte_region.destination_address = cpu_to_be64(addr); | |
3ccc00a7 MS |
183 | addr += fw_dump.hpte_region_size; |
184 | ||
185 | /* RMA region section */ | |
408cddd9 HB |
186 | fdm->rmr_region.request_flag = cpu_to_be32(FADUMP_REQUEST_FLAG); |
187 | fdm->rmr_region.source_data_type = cpu_to_be16(FADUMP_REAL_MODE_REGION); | |
188 | fdm->rmr_region.source_address = cpu_to_be64(RMA_START); | |
189 | fdm->rmr_region.source_len = cpu_to_be64(fw_dump.boot_memory_size); | |
190 | fdm->rmr_region.destination_address = cpu_to_be64(addr); | |
3ccc00a7 MS |
191 | addr += fw_dump.boot_memory_size; |
192 | ||
193 | return addr; | |
194 | } | |
195 | ||
eb39c880 MS |
196 | /** |
197 | * fadump_calculate_reserve_size(): reserve variable boot area 5% of System RAM | |
198 | * | |
199 | * Function to find the largest memory size we need to reserve during early | |
200 | * boot process. This will be the size of the memory that is required for a | |
201 | * kernel to boot successfully. | |
202 | * | |
203 | * This function has been taken from phyp-assisted dump feature implementation. | |
204 | * | |
205 | * returns larger of 256MB or 5% rounded down to multiples of 256MB. | |
206 | * | |
207 | * TODO: Come up with better approach to find out more accurate memory size | |
208 | * that is required for a kernel to boot successfully. | |
209 | * | |
210 | */ | |
211 | static inline unsigned long fadump_calculate_reserve_size(void) | |
212 | { | |
213 | unsigned long size; | |
214 | ||
215 | /* | |
216 | * Check if the size is specified through fadump_reserve_mem= cmdline | |
217 | * option. If yes, then use that. | |
218 | */ | |
219 | if (fw_dump.reserve_bootvar) | |
220 | return fw_dump.reserve_bootvar; | |
221 | ||
222 | /* divide by 20 to get 5% of value */ | |
223 | size = memblock_end_of_DRAM() / 20; | |
224 | ||
225 | /* round it down in multiples of 256 */ | |
226 | size = size & ~0x0FFFFFFFUL; | |
227 | ||
228 | /* Truncate to memory_limit. We don't want to over reserve the memory.*/ | |
229 | if (memory_limit && size > memory_limit) | |
230 | size = memory_limit; | |
231 | ||
232 | return (size > MIN_BOOT_MEM ? size : MIN_BOOT_MEM); | |
233 | } | |
234 | ||
235 | /* | |
236 | * Calculate the total memory size required to be reserved for | |
237 | * firmware-assisted dump registration. | |
238 | */ | |
239 | static unsigned long get_fadump_area_size(void) | |
240 | { | |
241 | unsigned long size = 0; | |
242 | ||
243 | size += fw_dump.cpu_state_data_size; | |
244 | size += fw_dump.hpte_region_size; | |
245 | size += fw_dump.boot_memory_size; | |
2df173d9 MS |
246 | size += sizeof(struct fadump_crash_info_header); |
247 | size += sizeof(struct elfhdr); /* ELF core header.*/ | |
ebaeb5ae | 248 | size += sizeof(struct elf_phdr); /* place holder for cpu notes */ |
2df173d9 MS |
249 | /* Program headers for crash memory regions. */ |
250 | size += sizeof(struct elf_phdr) * (memblock_num_regions(memory) + 2); | |
eb39c880 MS |
251 | |
252 | size = PAGE_ALIGN(size); | |
253 | return size; | |
254 | } | |
255 | ||
256 | int __init fadump_reserve_mem(void) | |
257 | { | |
258 | unsigned long base, size, memory_boundary; | |
259 | ||
260 | if (!fw_dump.fadump_enabled) | |
261 | return 0; | |
262 | ||
263 | if (!fw_dump.fadump_supported) { | |
264 | printk(KERN_INFO "Firmware-assisted dump is not supported on" | |
265 | " this hardware\n"); | |
266 | fw_dump.fadump_enabled = 0; | |
267 | return 0; | |
268 | } | |
3ccc00a7 MS |
269 | /* |
270 | * Initialize boot memory size | |
271 | * If dump is active then we have already calculated the size during | |
272 | * first kernel. | |
273 | */ | |
274 | if (fdm_active) | |
408cddd9 | 275 | fw_dump.boot_memory_size = be64_to_cpu(fdm_active->rmr_region.source_len); |
3ccc00a7 MS |
276 | else |
277 | fw_dump.boot_memory_size = fadump_calculate_reserve_size(); | |
eb39c880 MS |
278 | |
279 | /* | |
280 | * Calculate the memory boundary. | |
281 | * If memory_limit is less than actual memory boundary then reserve | |
282 | * the memory for fadump beyond the memory_limit and adjust the | |
283 | * memory_limit accordingly, so that the running kernel can run with | |
284 | * specified memory_limit. | |
285 | */ | |
286 | if (memory_limit && memory_limit < memblock_end_of_DRAM()) { | |
287 | size = get_fadump_area_size(); | |
288 | if ((memory_limit + size) < memblock_end_of_DRAM()) | |
289 | memory_limit += size; | |
290 | else | |
291 | memory_limit = memblock_end_of_DRAM(); | |
292 | printk(KERN_INFO "Adjusted memory_limit for firmware-assisted" | |
a84fcd46 | 293 | " dump, now %#016llx\n", memory_limit); |
eb39c880 MS |
294 | } |
295 | if (memory_limit) | |
296 | memory_boundary = memory_limit; | |
297 | else | |
298 | memory_boundary = memblock_end_of_DRAM(); | |
299 | ||
300 | if (fw_dump.dump_active) { | |
301 | printk(KERN_INFO "Firmware-assisted dump is active.\n"); | |
302 | /* | |
303 | * If last boot has crashed then reserve all the memory | |
304 | * above boot_memory_size so that we don't touch it until | |
305 | * dump is written to disk by userspace tool. This memory | |
306 | * will be released for general use once the dump is saved. | |
307 | */ | |
308 | base = fw_dump.boot_memory_size; | |
309 | size = memory_boundary - base; | |
310 | memblock_reserve(base, size); | |
311 | printk(KERN_INFO "Reserved %ldMB of memory at %ldMB " | |
312 | "for saving crash dump\n", | |
313 | (unsigned long)(size >> 20), | |
314 | (unsigned long)(base >> 20)); | |
2df173d9 MS |
315 | |
316 | fw_dump.fadumphdr_addr = | |
408cddd9 HB |
317 | be64_to_cpu(fdm_active->rmr_region.destination_address) + |
318 | be64_to_cpu(fdm_active->rmr_region.source_len); | |
2df173d9 MS |
319 | pr_debug("fadumphdr_addr = %p\n", |
320 | (void *) fw_dump.fadumphdr_addr); | |
eb39c880 MS |
321 | } else { |
322 | /* Reserve the memory at the top of memory. */ | |
323 | size = get_fadump_area_size(); | |
324 | base = memory_boundary - size; | |
325 | memblock_reserve(base, size); | |
326 | printk(KERN_INFO "Reserved %ldMB of memory at %ldMB " | |
327 | "for firmware-assisted dump\n", | |
328 | (unsigned long)(size >> 20), | |
329 | (unsigned long)(base >> 20)); | |
330 | } | |
331 | fw_dump.reserve_dump_area_start = base; | |
332 | fw_dump.reserve_dump_area_size = size; | |
333 | return 1; | |
334 | } | |
335 | ||
1e76609c SD |
336 | unsigned long __init arch_reserved_kernel_pages(void) |
337 | { | |
338 | return memblock_reserved_size() / PAGE_SIZE; | |
339 | } | |
340 | ||
eb39c880 MS |
341 | /* Look for fadump= cmdline option. */ |
342 | static int __init early_fadump_param(char *p) | |
343 | { | |
344 | if (!p) | |
345 | return 1; | |
346 | ||
347 | if (strncmp(p, "on", 2) == 0) | |
348 | fw_dump.fadump_enabled = 1; | |
349 | else if (strncmp(p, "off", 3) == 0) | |
350 | fw_dump.fadump_enabled = 0; | |
351 | ||
352 | return 0; | |
353 | } | |
354 | early_param("fadump", early_fadump_param); | |
355 | ||
356 | /* Look for fadump_reserve_mem= cmdline option */ | |
357 | static int __init early_fadump_reserve_mem(char *p) | |
358 | { | |
359 | if (p) | |
360 | fw_dump.reserve_bootvar = memparse(p, &p); | |
361 | return 0; | |
362 | } | |
363 | early_param("fadump_reserve_mem", early_fadump_reserve_mem); | |
3ccc00a7 MS |
364 | |
365 | static void register_fw_dump(struct fadump_mem_struct *fdm) | |
366 | { | |
367 | int rc; | |
368 | unsigned int wait_time; | |
369 | ||
370 | pr_debug("Registering for firmware-assisted kernel dump...\n"); | |
371 | ||
372 | /* TODO: Add upper time limit for the delay */ | |
373 | do { | |
374 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, | |
375 | FADUMP_REGISTER, fdm, | |
376 | sizeof(struct fadump_mem_struct)); | |
377 | ||
378 | wait_time = rtas_busy_delay_time(rc); | |
379 | if (wait_time) | |
380 | mdelay(wait_time); | |
381 | ||
382 | } while (wait_time); | |
383 | ||
384 | switch (rc) { | |
385 | case -1: | |
386 | printk(KERN_ERR "Failed to register firmware-assisted kernel" | |
387 | " dump. Hardware Error(%d).\n", rc); | |
388 | break; | |
389 | case -3: | |
390 | printk(KERN_ERR "Failed to register firmware-assisted kernel" | |
391 | " dump. Parameter Error(%d).\n", rc); | |
392 | break; | |
393 | case -9: | |
394 | printk(KERN_ERR "firmware-assisted kernel dump is already " | |
395 | " registered."); | |
396 | fw_dump.dump_registered = 1; | |
397 | break; | |
398 | case 0: | |
399 | printk(KERN_INFO "firmware-assisted kernel dump registration" | |
400 | " is successful\n"); | |
401 | fw_dump.dump_registered = 1; | |
402 | break; | |
403 | } | |
404 | } | |
405 | ||
ebaeb5ae MS |
406 | void crash_fadump(struct pt_regs *regs, const char *str) |
407 | { | |
408 | struct fadump_crash_info_header *fdh = NULL; | |
409 | ||
410 | if (!fw_dump.dump_registered || !fw_dump.fadumphdr_addr) | |
411 | return; | |
412 | ||
413 | fdh = __va(fw_dump.fadumphdr_addr); | |
414 | crashing_cpu = smp_processor_id(); | |
415 | fdh->crashing_cpu = crashing_cpu; | |
416 | crash_save_vmcoreinfo(); | |
417 | ||
418 | if (regs) | |
419 | fdh->regs = *regs; | |
420 | else | |
421 | ppc_save_regs(&fdh->regs); | |
422 | ||
a0512164 | 423 | fdh->online_mask = *cpu_online_mask; |
ebaeb5ae MS |
424 | |
425 | /* Call ibm,os-term rtas call to trigger firmware assisted dump */ | |
426 | rtas_os_term((char *)str); | |
427 | } | |
428 | ||
429 | #define GPR_MASK 0xffffff0000000000 | |
430 | static inline int fadump_gpr_index(u64 id) | |
431 | { | |
432 | int i = -1; | |
433 | char str[3]; | |
434 | ||
435 | if ((id & GPR_MASK) == REG_ID("GPR")) { | |
436 | /* get the digits at the end */ | |
437 | id &= ~GPR_MASK; | |
438 | id >>= 24; | |
439 | str[2] = '\0'; | |
440 | str[1] = id & 0xff; | |
441 | str[0] = (id >> 8) & 0xff; | |
442 | sscanf(str, "%d", &i); | |
443 | if (i > 31) | |
444 | i = -1; | |
445 | } | |
446 | return i; | |
447 | } | |
448 | ||
449 | static inline void fadump_set_regval(struct pt_regs *regs, u64 reg_id, | |
450 | u64 reg_val) | |
451 | { | |
452 | int i; | |
453 | ||
454 | i = fadump_gpr_index(reg_id); | |
455 | if (i >= 0) | |
456 | regs->gpr[i] = (unsigned long)reg_val; | |
457 | else if (reg_id == REG_ID("NIA")) | |
458 | regs->nip = (unsigned long)reg_val; | |
459 | else if (reg_id == REG_ID("MSR")) | |
460 | regs->msr = (unsigned long)reg_val; | |
461 | else if (reg_id == REG_ID("CTR")) | |
462 | regs->ctr = (unsigned long)reg_val; | |
463 | else if (reg_id == REG_ID("LR")) | |
464 | regs->link = (unsigned long)reg_val; | |
465 | else if (reg_id == REG_ID("XER")) | |
466 | regs->xer = (unsigned long)reg_val; | |
467 | else if (reg_id == REG_ID("CR")) | |
468 | regs->ccr = (unsigned long)reg_val; | |
469 | else if (reg_id == REG_ID("DAR")) | |
470 | regs->dar = (unsigned long)reg_val; | |
471 | else if (reg_id == REG_ID("DSISR")) | |
472 | regs->dsisr = (unsigned long)reg_val; | |
473 | } | |
474 | ||
475 | static struct fadump_reg_entry* | |
476 | fadump_read_registers(struct fadump_reg_entry *reg_entry, struct pt_regs *regs) | |
477 | { | |
478 | memset(regs, 0, sizeof(struct pt_regs)); | |
479 | ||
408cddd9 HB |
480 | while (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUEND")) { |
481 | fadump_set_regval(regs, be64_to_cpu(reg_entry->reg_id), | |
482 | be64_to_cpu(reg_entry->reg_value)); | |
ebaeb5ae MS |
483 | reg_entry++; |
484 | } | |
485 | reg_entry++; | |
486 | return reg_entry; | |
487 | } | |
488 | ||
489 | static u32 *fadump_append_elf_note(u32 *buf, char *name, unsigned type, | |
490 | void *data, size_t data_len) | |
491 | { | |
492 | struct elf_note note; | |
493 | ||
494 | note.n_namesz = strlen(name) + 1; | |
495 | note.n_descsz = data_len; | |
496 | note.n_type = type; | |
497 | memcpy(buf, ¬e, sizeof(note)); | |
498 | buf += (sizeof(note) + 3)/4; | |
499 | memcpy(buf, name, note.n_namesz); | |
500 | buf += (note.n_namesz + 3)/4; | |
501 | memcpy(buf, data, note.n_descsz); | |
502 | buf += (note.n_descsz + 3)/4; | |
503 | ||
504 | return buf; | |
505 | } | |
506 | ||
507 | static void fadump_final_note(u32 *buf) | |
508 | { | |
509 | struct elf_note note; | |
510 | ||
511 | note.n_namesz = 0; | |
512 | note.n_descsz = 0; | |
513 | note.n_type = 0; | |
514 | memcpy(buf, ¬e, sizeof(note)); | |
515 | } | |
516 | ||
517 | static u32 *fadump_regs_to_elf_notes(u32 *buf, struct pt_regs *regs) | |
518 | { | |
519 | struct elf_prstatus prstatus; | |
520 | ||
521 | memset(&prstatus, 0, sizeof(prstatus)); | |
522 | /* | |
523 | * FIXME: How do i get PID? Do I really need it? | |
524 | * prstatus.pr_pid = ???? | |
525 | */ | |
526 | elf_core_copy_kernel_regs(&prstatus.pr_reg, regs); | |
527 | buf = fadump_append_elf_note(buf, KEXEC_CORE_NOTE_NAME, NT_PRSTATUS, | |
528 | &prstatus, sizeof(prstatus)); | |
529 | return buf; | |
530 | } | |
531 | ||
532 | static void fadump_update_elfcore_header(char *bufp) | |
533 | { | |
534 | struct elfhdr *elf; | |
535 | struct elf_phdr *phdr; | |
536 | ||
537 | elf = (struct elfhdr *)bufp; | |
538 | bufp += sizeof(struct elfhdr); | |
539 | ||
540 | /* First note is a place holder for cpu notes info. */ | |
541 | phdr = (struct elf_phdr *)bufp; | |
542 | ||
543 | if (phdr->p_type == PT_NOTE) { | |
544 | phdr->p_paddr = fw_dump.cpu_notes_buf; | |
545 | phdr->p_offset = phdr->p_paddr; | |
546 | phdr->p_filesz = fw_dump.cpu_notes_buf_size; | |
547 | phdr->p_memsz = fw_dump.cpu_notes_buf_size; | |
548 | } | |
549 | return; | |
550 | } | |
551 | ||
552 | static void *fadump_cpu_notes_buf_alloc(unsigned long size) | |
553 | { | |
554 | void *vaddr; | |
555 | struct page *page; | |
556 | unsigned long order, count, i; | |
557 | ||
558 | order = get_order(size); | |
559 | vaddr = (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, order); | |
560 | if (!vaddr) | |
561 | return NULL; | |
562 | ||
563 | count = 1 << order; | |
564 | page = virt_to_page(vaddr); | |
565 | for (i = 0; i < count; i++) | |
566 | SetPageReserved(page + i); | |
567 | return vaddr; | |
568 | } | |
569 | ||
570 | static void fadump_cpu_notes_buf_free(unsigned long vaddr, unsigned long size) | |
571 | { | |
572 | struct page *page; | |
573 | unsigned long order, count, i; | |
574 | ||
575 | order = get_order(size); | |
576 | count = 1 << order; | |
577 | page = virt_to_page(vaddr); | |
578 | for (i = 0; i < count; i++) | |
579 | ClearPageReserved(page + i); | |
580 | __free_pages(page, order); | |
581 | } | |
582 | ||
583 | /* | |
584 | * Read CPU state dump data and convert it into ELF notes. | |
585 | * The CPU dump starts with magic number "REGSAVE". NumCpusOffset should be | |
586 | * used to access the data to allow for additional fields to be added without | |
587 | * affecting compatibility. Each list of registers for a CPU starts with | |
588 | * "CPUSTRT" and ends with "CPUEND". Each register entry is of 16 bytes, | |
589 | * 8 Byte ASCII identifier and 8 Byte register value. The register entry | |
590 | * with identifier "CPUSTRT" and "CPUEND" contains 4 byte cpu id as part | |
591 | * of register value. For more details refer to PAPR document. | |
592 | * | |
593 | * Only for the crashing cpu we ignore the CPU dump data and get exact | |
594 | * state from fadump crash info structure populated by first kernel at the | |
595 | * time of crash. | |
596 | */ | |
597 | static int __init fadump_build_cpu_notes(const struct fadump_mem_struct *fdm) | |
598 | { | |
599 | struct fadump_reg_save_area_header *reg_header; | |
600 | struct fadump_reg_entry *reg_entry; | |
601 | struct fadump_crash_info_header *fdh = NULL; | |
602 | void *vaddr; | |
603 | unsigned long addr; | |
604 | u32 num_cpus, *note_buf; | |
605 | struct pt_regs regs; | |
606 | int i, rc = 0, cpu = 0; | |
607 | ||
608 | if (!fdm->cpu_state_data.bytes_dumped) | |
609 | return -EINVAL; | |
610 | ||
408cddd9 | 611 | addr = be64_to_cpu(fdm->cpu_state_data.destination_address); |
ebaeb5ae MS |
612 | vaddr = __va(addr); |
613 | ||
614 | reg_header = vaddr; | |
408cddd9 | 615 | if (be64_to_cpu(reg_header->magic_number) != REGSAVE_AREA_MAGIC) { |
ebaeb5ae MS |
616 | printk(KERN_ERR "Unable to read register save area.\n"); |
617 | return -ENOENT; | |
618 | } | |
619 | pr_debug("--------CPU State Data------------\n"); | |
408cddd9 HB |
620 | pr_debug("Magic Number: %llx\n", be64_to_cpu(reg_header->magic_number)); |
621 | pr_debug("NumCpuOffset: %x\n", be32_to_cpu(reg_header->num_cpu_offset)); | |
ebaeb5ae | 622 | |
408cddd9 HB |
623 | vaddr += be32_to_cpu(reg_header->num_cpu_offset); |
624 | num_cpus = be32_to_cpu(*((__be32 *)(vaddr))); | |
ebaeb5ae MS |
625 | pr_debug("NumCpus : %u\n", num_cpus); |
626 | vaddr += sizeof(u32); | |
627 | reg_entry = (struct fadump_reg_entry *)vaddr; | |
628 | ||
629 | /* Allocate buffer to hold cpu crash notes. */ | |
630 | fw_dump.cpu_notes_buf_size = num_cpus * sizeof(note_buf_t); | |
631 | fw_dump.cpu_notes_buf_size = PAGE_ALIGN(fw_dump.cpu_notes_buf_size); | |
632 | note_buf = fadump_cpu_notes_buf_alloc(fw_dump.cpu_notes_buf_size); | |
633 | if (!note_buf) { | |
634 | printk(KERN_ERR "Failed to allocate 0x%lx bytes for " | |
635 | "cpu notes buffer\n", fw_dump.cpu_notes_buf_size); | |
636 | return -ENOMEM; | |
637 | } | |
638 | fw_dump.cpu_notes_buf = __pa(note_buf); | |
639 | ||
640 | pr_debug("Allocated buffer for cpu notes of size %ld at %p\n", | |
641 | (num_cpus * sizeof(note_buf_t)), note_buf); | |
642 | ||
643 | if (fw_dump.fadumphdr_addr) | |
644 | fdh = __va(fw_dump.fadumphdr_addr); | |
645 | ||
646 | for (i = 0; i < num_cpus; i++) { | |
408cddd9 | 647 | if (be64_to_cpu(reg_entry->reg_id) != REG_ID("CPUSTRT")) { |
ebaeb5ae MS |
648 | printk(KERN_ERR "Unable to read CPU state data\n"); |
649 | rc = -ENOENT; | |
650 | goto error_out; | |
651 | } | |
652 | /* Lower 4 bytes of reg_value contains logical cpu id */ | |
408cddd9 | 653 | cpu = be64_to_cpu(reg_entry->reg_value) & FADUMP_CPU_ID_MASK; |
a0512164 | 654 | if (fdh && !cpumask_test_cpu(cpu, &fdh->online_mask)) { |
ebaeb5ae MS |
655 | SKIP_TO_NEXT_CPU(reg_entry); |
656 | continue; | |
657 | } | |
658 | pr_debug("Reading register data for cpu %d...\n", cpu); | |
659 | if (fdh && fdh->crashing_cpu == cpu) { | |
660 | regs = fdh->regs; | |
661 | note_buf = fadump_regs_to_elf_notes(note_buf, ®s); | |
662 | SKIP_TO_NEXT_CPU(reg_entry); | |
663 | } else { | |
664 | reg_entry++; | |
665 | reg_entry = fadump_read_registers(reg_entry, ®s); | |
666 | note_buf = fadump_regs_to_elf_notes(note_buf, ®s); | |
667 | } | |
668 | } | |
669 | fadump_final_note(note_buf); | |
670 | ||
b717d985 RS |
671 | if (fdh) { |
672 | pr_debug("Updating elfcore header (%llx) with cpu notes\n", | |
ebaeb5ae | 673 | fdh->elfcorehdr_addr); |
b717d985 RS |
674 | fadump_update_elfcore_header((char *)__va(fdh->elfcorehdr_addr)); |
675 | } | |
ebaeb5ae MS |
676 | return 0; |
677 | ||
678 | error_out: | |
679 | fadump_cpu_notes_buf_free((unsigned long)__va(fw_dump.cpu_notes_buf), | |
680 | fw_dump.cpu_notes_buf_size); | |
681 | fw_dump.cpu_notes_buf = 0; | |
682 | fw_dump.cpu_notes_buf_size = 0; | |
683 | return rc; | |
684 | ||
685 | } | |
686 | ||
2df173d9 MS |
687 | /* |
688 | * Validate and process the dump data stored by firmware before exporting | |
689 | * it through '/proc/vmcore'. | |
690 | */ | |
691 | static int __init process_fadump(const struct fadump_mem_struct *fdm_active) | |
692 | { | |
693 | struct fadump_crash_info_header *fdh; | |
ebaeb5ae | 694 | int rc = 0; |
2df173d9 MS |
695 | |
696 | if (!fdm_active || !fw_dump.fadumphdr_addr) | |
697 | return -EINVAL; | |
698 | ||
699 | /* Check if the dump data is valid. */ | |
408cddd9 | 700 | if ((be16_to_cpu(fdm_active->header.dump_status_flag) == FADUMP_ERROR_FLAG) || |
ebaeb5ae | 701 | (fdm_active->cpu_state_data.error_flags != 0) || |
2df173d9 MS |
702 | (fdm_active->rmr_region.error_flags != 0)) { |
703 | printk(KERN_ERR "Dump taken by platform is not valid\n"); | |
704 | return -EINVAL; | |
705 | } | |
ebaeb5ae MS |
706 | if ((fdm_active->rmr_region.bytes_dumped != |
707 | fdm_active->rmr_region.source_len) || | |
708 | !fdm_active->cpu_state_data.bytes_dumped) { | |
2df173d9 MS |
709 | printk(KERN_ERR "Dump taken by platform is incomplete\n"); |
710 | return -EINVAL; | |
711 | } | |
712 | ||
713 | /* Validate the fadump crash info header */ | |
714 | fdh = __va(fw_dump.fadumphdr_addr); | |
715 | if (fdh->magic_number != FADUMP_CRASH_INFO_MAGIC) { | |
716 | printk(KERN_ERR "Crash info header is not valid.\n"); | |
717 | return -EINVAL; | |
718 | } | |
719 | ||
ebaeb5ae MS |
720 | rc = fadump_build_cpu_notes(fdm_active); |
721 | if (rc) | |
722 | return rc; | |
723 | ||
2df173d9 MS |
724 | /* |
725 | * We are done validating dump info and elfcore header is now ready | |
726 | * to be exported. set elfcorehdr_addr so that vmcore module will | |
727 | * export the elfcore header through '/proc/vmcore'. | |
728 | */ | |
729 | elfcorehdr_addr = fdh->elfcorehdr_addr; | |
730 | ||
731 | return 0; | |
732 | } | |
733 | ||
734 | static inline void fadump_add_crash_memory(unsigned long long base, | |
735 | unsigned long long end) | |
736 | { | |
737 | if (base == end) | |
738 | return; | |
739 | ||
740 | pr_debug("crash_memory_range[%d] [%#016llx-%#016llx], %#llx bytes\n", | |
741 | crash_mem_ranges, base, end - 1, (end - base)); | |
742 | crash_memory_ranges[crash_mem_ranges].base = base; | |
743 | crash_memory_ranges[crash_mem_ranges].size = end - base; | |
744 | crash_mem_ranges++; | |
745 | } | |
746 | ||
747 | static void fadump_exclude_reserved_area(unsigned long long start, | |
748 | unsigned long long end) | |
749 | { | |
750 | unsigned long long ra_start, ra_end; | |
751 | ||
752 | ra_start = fw_dump.reserve_dump_area_start; | |
753 | ra_end = ra_start + fw_dump.reserve_dump_area_size; | |
754 | ||
755 | if ((ra_start < end) && (ra_end > start)) { | |
756 | if ((start < ra_start) && (end > ra_end)) { | |
757 | fadump_add_crash_memory(start, ra_start); | |
758 | fadump_add_crash_memory(ra_end, end); | |
759 | } else if (start < ra_start) { | |
760 | fadump_add_crash_memory(start, ra_start); | |
761 | } else if (ra_end < end) { | |
762 | fadump_add_crash_memory(ra_end, end); | |
763 | } | |
764 | } else | |
765 | fadump_add_crash_memory(start, end); | |
766 | } | |
767 | ||
768 | static int fadump_init_elfcore_header(char *bufp) | |
769 | { | |
770 | struct elfhdr *elf; | |
771 | ||
772 | elf = (struct elfhdr *) bufp; | |
773 | bufp += sizeof(struct elfhdr); | |
774 | memcpy(elf->e_ident, ELFMAG, SELFMAG); | |
775 | elf->e_ident[EI_CLASS] = ELF_CLASS; | |
776 | elf->e_ident[EI_DATA] = ELF_DATA; | |
777 | elf->e_ident[EI_VERSION] = EV_CURRENT; | |
778 | elf->e_ident[EI_OSABI] = ELF_OSABI; | |
779 | memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD); | |
780 | elf->e_type = ET_CORE; | |
781 | elf->e_machine = ELF_ARCH; | |
782 | elf->e_version = EV_CURRENT; | |
783 | elf->e_entry = 0; | |
784 | elf->e_phoff = sizeof(struct elfhdr); | |
785 | elf->e_shoff = 0; | |
d8bced27 DA |
786 | #if defined(_CALL_ELF) |
787 | elf->e_flags = _CALL_ELF; | |
788 | #else | |
789 | elf->e_flags = 0; | |
790 | #endif | |
2df173d9 MS |
791 | elf->e_ehsize = sizeof(struct elfhdr); |
792 | elf->e_phentsize = sizeof(struct elf_phdr); | |
793 | elf->e_phnum = 0; | |
794 | elf->e_shentsize = 0; | |
795 | elf->e_shnum = 0; | |
796 | elf->e_shstrndx = 0; | |
797 | ||
798 | return 0; | |
799 | } | |
800 | ||
801 | /* | |
802 | * Traverse through memblock structure and setup crash memory ranges. These | |
803 | * ranges will be used create PT_LOAD program headers in elfcore header. | |
804 | */ | |
805 | static void fadump_setup_crash_memory_ranges(void) | |
806 | { | |
807 | struct memblock_region *reg; | |
808 | unsigned long long start, end; | |
809 | ||
810 | pr_debug("Setup crash memory ranges.\n"); | |
811 | crash_mem_ranges = 0; | |
812 | /* | |
813 | * add the first memory chunk (RMA_START through boot_memory_size) as | |
814 | * a separate memory chunk. The reason is, at the time crash firmware | |
815 | * will move the content of this memory chunk to different location | |
816 | * specified during fadump registration. We need to create a separate | |
817 | * program header for this chunk with the correct offset. | |
818 | */ | |
819 | fadump_add_crash_memory(RMA_START, fw_dump.boot_memory_size); | |
820 | ||
821 | for_each_memblock(memory, reg) { | |
822 | start = (unsigned long long)reg->base; | |
823 | end = start + (unsigned long long)reg->size; | |
824 | if (start == RMA_START && end >= fw_dump.boot_memory_size) | |
825 | start = fw_dump.boot_memory_size; | |
826 | ||
827 | /* add this range excluding the reserved dump area. */ | |
828 | fadump_exclude_reserved_area(start, end); | |
829 | } | |
830 | } | |
831 | ||
d34c5f26 MS |
832 | /* |
833 | * If the given physical address falls within the boot memory region then | |
834 | * return the relocated address that points to the dump region reserved | |
835 | * for saving initial boot memory contents. | |
836 | */ | |
837 | static inline unsigned long fadump_relocate(unsigned long paddr) | |
838 | { | |
839 | if (paddr > RMA_START && paddr < fw_dump.boot_memory_size) | |
408cddd9 | 840 | return be64_to_cpu(fdm.rmr_region.destination_address) + paddr; |
d34c5f26 MS |
841 | else |
842 | return paddr; | |
843 | } | |
844 | ||
2df173d9 MS |
845 | static int fadump_create_elfcore_headers(char *bufp) |
846 | { | |
847 | struct elfhdr *elf; | |
848 | struct elf_phdr *phdr; | |
849 | int i; | |
850 | ||
851 | fadump_init_elfcore_header(bufp); | |
852 | elf = (struct elfhdr *)bufp; | |
853 | bufp += sizeof(struct elfhdr); | |
854 | ||
ebaeb5ae MS |
855 | /* |
856 | * setup ELF PT_NOTE, place holder for cpu notes info. The notes info | |
857 | * will be populated during second kernel boot after crash. Hence | |
858 | * this PT_NOTE will always be the first elf note. | |
859 | * | |
860 | * NOTE: Any new ELF note addition should be placed after this note. | |
861 | */ | |
862 | phdr = (struct elf_phdr *)bufp; | |
863 | bufp += sizeof(struct elf_phdr); | |
864 | phdr->p_type = PT_NOTE; | |
865 | phdr->p_flags = 0; | |
866 | phdr->p_vaddr = 0; | |
867 | phdr->p_align = 0; | |
868 | ||
869 | phdr->p_offset = 0; | |
870 | phdr->p_paddr = 0; | |
871 | phdr->p_filesz = 0; | |
872 | phdr->p_memsz = 0; | |
873 | ||
874 | (elf->e_phnum)++; | |
875 | ||
d34c5f26 MS |
876 | /* setup ELF PT_NOTE for vmcoreinfo */ |
877 | phdr = (struct elf_phdr *)bufp; | |
878 | bufp += sizeof(struct elf_phdr); | |
879 | phdr->p_type = PT_NOTE; | |
880 | phdr->p_flags = 0; | |
881 | phdr->p_vaddr = 0; | |
882 | phdr->p_align = 0; | |
883 | ||
884 | phdr->p_paddr = fadump_relocate(paddr_vmcoreinfo_note()); | |
885 | phdr->p_offset = phdr->p_paddr; | |
886 | phdr->p_memsz = vmcoreinfo_max_size; | |
887 | phdr->p_filesz = vmcoreinfo_max_size; | |
888 | ||
889 | /* Increment number of program headers. */ | |
890 | (elf->e_phnum)++; | |
891 | ||
2df173d9 MS |
892 | /* setup PT_LOAD sections. */ |
893 | ||
894 | for (i = 0; i < crash_mem_ranges; i++) { | |
895 | unsigned long long mbase, msize; | |
896 | mbase = crash_memory_ranges[i].base; | |
897 | msize = crash_memory_ranges[i].size; | |
898 | ||
899 | if (!msize) | |
900 | continue; | |
901 | ||
902 | phdr = (struct elf_phdr *)bufp; | |
903 | bufp += sizeof(struct elf_phdr); | |
904 | phdr->p_type = PT_LOAD; | |
905 | phdr->p_flags = PF_R|PF_W|PF_X; | |
906 | phdr->p_offset = mbase; | |
907 | ||
908 | if (mbase == RMA_START) { | |
909 | /* | |
910 | * The entire RMA region will be moved by firmware | |
911 | * to the specified destination_address. Hence set | |
912 | * the correct offset. | |
913 | */ | |
408cddd9 | 914 | phdr->p_offset = be64_to_cpu(fdm.rmr_region.destination_address); |
2df173d9 MS |
915 | } |
916 | ||
917 | phdr->p_paddr = mbase; | |
918 | phdr->p_vaddr = (unsigned long)__va(mbase); | |
919 | phdr->p_filesz = msize; | |
920 | phdr->p_memsz = msize; | |
921 | phdr->p_align = 0; | |
922 | ||
923 | /* Increment number of program headers. */ | |
924 | (elf->e_phnum)++; | |
925 | } | |
926 | return 0; | |
927 | } | |
928 | ||
929 | static unsigned long init_fadump_header(unsigned long addr) | |
930 | { | |
931 | struct fadump_crash_info_header *fdh; | |
932 | ||
933 | if (!addr) | |
934 | return 0; | |
935 | ||
936 | fw_dump.fadumphdr_addr = addr; | |
937 | fdh = __va(addr); | |
938 | addr += sizeof(struct fadump_crash_info_header); | |
939 | ||
940 | memset(fdh, 0, sizeof(struct fadump_crash_info_header)); | |
941 | fdh->magic_number = FADUMP_CRASH_INFO_MAGIC; | |
942 | fdh->elfcorehdr_addr = addr; | |
ebaeb5ae MS |
943 | /* We will set the crashing cpu id in crash_fadump() during crash. */ |
944 | fdh->crashing_cpu = CPU_UNKNOWN; | |
2df173d9 MS |
945 | |
946 | return addr; | |
947 | } | |
948 | ||
3ccc00a7 MS |
949 | static void register_fadump(void) |
950 | { | |
2df173d9 MS |
951 | unsigned long addr; |
952 | void *vaddr; | |
953 | ||
3ccc00a7 MS |
954 | /* |
955 | * If no memory is reserved then we can not register for firmware- | |
956 | * assisted dump. | |
957 | */ | |
958 | if (!fw_dump.reserve_dump_area_size) | |
959 | return; | |
960 | ||
2df173d9 MS |
961 | fadump_setup_crash_memory_ranges(); |
962 | ||
408cddd9 | 963 | addr = be64_to_cpu(fdm.rmr_region.destination_address) + be64_to_cpu(fdm.rmr_region.source_len); |
2df173d9 MS |
964 | /* Initialize fadump crash info header. */ |
965 | addr = init_fadump_header(addr); | |
966 | vaddr = __va(addr); | |
967 | ||
968 | pr_debug("Creating ELF core headers at %#016lx\n", addr); | |
969 | fadump_create_elfcore_headers(vaddr); | |
970 | ||
3ccc00a7 MS |
971 | /* register the future kernel dump with firmware. */ |
972 | register_fw_dump(&fdm); | |
973 | } | |
974 | ||
975 | static int fadump_unregister_dump(struct fadump_mem_struct *fdm) | |
976 | { | |
977 | int rc = 0; | |
978 | unsigned int wait_time; | |
979 | ||
980 | pr_debug("Un-register firmware-assisted dump\n"); | |
981 | ||
982 | /* TODO: Add upper time limit for the delay */ | |
983 | do { | |
984 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, | |
985 | FADUMP_UNREGISTER, fdm, | |
986 | sizeof(struct fadump_mem_struct)); | |
987 | ||
988 | wait_time = rtas_busy_delay_time(rc); | |
989 | if (wait_time) | |
990 | mdelay(wait_time); | |
991 | } while (wait_time); | |
992 | ||
993 | if (rc) { | |
994 | printk(KERN_ERR "Failed to un-register firmware-assisted dump." | |
995 | " unexpected error(%d).\n", rc); | |
996 | return rc; | |
997 | } | |
998 | fw_dump.dump_registered = 0; | |
999 | return 0; | |
1000 | } | |
1001 | ||
b500afff MS |
1002 | static int fadump_invalidate_dump(struct fadump_mem_struct *fdm) |
1003 | { | |
1004 | int rc = 0; | |
1005 | unsigned int wait_time; | |
1006 | ||
1007 | pr_debug("Invalidating firmware-assisted dump registration\n"); | |
1008 | ||
1009 | /* TODO: Add upper time limit for the delay */ | |
1010 | do { | |
1011 | rc = rtas_call(fw_dump.ibm_configure_kernel_dump, 3, 1, NULL, | |
1012 | FADUMP_INVALIDATE, fdm, | |
1013 | sizeof(struct fadump_mem_struct)); | |
1014 | ||
1015 | wait_time = rtas_busy_delay_time(rc); | |
1016 | if (wait_time) | |
1017 | mdelay(wait_time); | |
1018 | } while (wait_time); | |
1019 | ||
1020 | if (rc) { | |
4a03749f | 1021 | pr_err("Failed to invalidate firmware-assisted dump registration. Unexpected error (%d).\n", rc); |
b5b1cfc5 | 1022 | return rc; |
b500afff MS |
1023 | } |
1024 | fw_dump.dump_active = 0; | |
1025 | fdm_active = NULL; | |
1026 | return 0; | |
1027 | } | |
1028 | ||
1029 | void fadump_cleanup(void) | |
1030 | { | |
1031 | /* Invalidate the registration only if dump is active. */ | |
1032 | if (fw_dump.dump_active) { | |
1033 | init_fadump_mem_struct(&fdm, | |
408cddd9 | 1034 | be64_to_cpu(fdm_active->cpu_state_data.destination_address)); |
b500afff MS |
1035 | fadump_invalidate_dump(&fdm); |
1036 | } | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Release the memory that was reserved in early boot to preserve the memory | |
1041 | * contents. The released memory will be available for general use. | |
1042 | */ | |
1043 | static void fadump_release_memory(unsigned long begin, unsigned long end) | |
1044 | { | |
1045 | unsigned long addr; | |
1046 | unsigned long ra_start, ra_end; | |
1047 | ||
1048 | ra_start = fw_dump.reserve_dump_area_start; | |
1049 | ra_end = ra_start + fw_dump.reserve_dump_area_size; | |
1050 | ||
1051 | for (addr = begin; addr < end; addr += PAGE_SIZE) { | |
1052 | /* | |
1053 | * exclude the dump reserve area. Will reuse it for next | |
1054 | * fadump registration. | |
1055 | */ | |
1056 | if (addr <= ra_end && ((addr + PAGE_SIZE) > ra_start)) | |
1057 | continue; | |
1058 | ||
5d585e5c | 1059 | free_reserved_page(pfn_to_page(addr >> PAGE_SHIFT)); |
b500afff MS |
1060 | } |
1061 | } | |
1062 | ||
1063 | static void fadump_invalidate_release_mem(void) | |
1064 | { | |
1065 | unsigned long reserved_area_start, reserved_area_end; | |
1066 | unsigned long destination_address; | |
1067 | ||
1068 | mutex_lock(&fadump_mutex); | |
1069 | if (!fw_dump.dump_active) { | |
1070 | mutex_unlock(&fadump_mutex); | |
1071 | return; | |
1072 | } | |
1073 | ||
408cddd9 | 1074 | destination_address = be64_to_cpu(fdm_active->cpu_state_data.destination_address); |
b500afff MS |
1075 | fadump_cleanup(); |
1076 | mutex_unlock(&fadump_mutex); | |
1077 | ||
1078 | /* | |
1079 | * Save the current reserved memory bounds we will require them | |
1080 | * later for releasing the memory for general use. | |
1081 | */ | |
1082 | reserved_area_start = fw_dump.reserve_dump_area_start; | |
1083 | reserved_area_end = reserved_area_start + | |
1084 | fw_dump.reserve_dump_area_size; | |
1085 | /* | |
1086 | * Setup reserve_dump_area_start and its size so that we can | |
1087 | * reuse this reserved memory for Re-registration. | |
1088 | */ | |
1089 | fw_dump.reserve_dump_area_start = destination_address; | |
1090 | fw_dump.reserve_dump_area_size = get_fadump_area_size(); | |
1091 | ||
1092 | fadump_release_memory(reserved_area_start, reserved_area_end); | |
1093 | if (fw_dump.cpu_notes_buf) { | |
1094 | fadump_cpu_notes_buf_free( | |
1095 | (unsigned long)__va(fw_dump.cpu_notes_buf), | |
1096 | fw_dump.cpu_notes_buf_size); | |
1097 | fw_dump.cpu_notes_buf = 0; | |
1098 | fw_dump.cpu_notes_buf_size = 0; | |
1099 | } | |
1100 | /* Initialize the kernel dump memory structure for FAD registration. */ | |
1101 | init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start); | |
1102 | } | |
1103 | ||
1104 | static ssize_t fadump_release_memory_store(struct kobject *kobj, | |
1105 | struct kobj_attribute *attr, | |
1106 | const char *buf, size_t count) | |
1107 | { | |
1108 | if (!fw_dump.dump_active) | |
1109 | return -EPERM; | |
1110 | ||
1111 | if (buf[0] == '1') { | |
1112 | /* | |
1113 | * Take away the '/proc/vmcore'. We are releasing the dump | |
1114 | * memory, hence it will not be valid anymore. | |
1115 | */ | |
2685f826 | 1116 | #ifdef CONFIG_PROC_VMCORE |
b500afff | 1117 | vmcore_cleanup(); |
2685f826 | 1118 | #endif |
b500afff MS |
1119 | fadump_invalidate_release_mem(); |
1120 | ||
1121 | } else | |
1122 | return -EINVAL; | |
1123 | return count; | |
1124 | } | |
1125 | ||
3ccc00a7 MS |
1126 | static ssize_t fadump_enabled_show(struct kobject *kobj, |
1127 | struct kobj_attribute *attr, | |
1128 | char *buf) | |
1129 | { | |
1130 | return sprintf(buf, "%d\n", fw_dump.fadump_enabled); | |
1131 | } | |
1132 | ||
1133 | static ssize_t fadump_register_show(struct kobject *kobj, | |
1134 | struct kobj_attribute *attr, | |
1135 | char *buf) | |
1136 | { | |
1137 | return sprintf(buf, "%d\n", fw_dump.dump_registered); | |
1138 | } | |
1139 | ||
1140 | static ssize_t fadump_register_store(struct kobject *kobj, | |
1141 | struct kobj_attribute *attr, | |
1142 | const char *buf, size_t count) | |
1143 | { | |
1144 | int ret = 0; | |
1145 | ||
1146 | if (!fw_dump.fadump_enabled || fdm_active) | |
1147 | return -EPERM; | |
1148 | ||
1149 | mutex_lock(&fadump_mutex); | |
1150 | ||
1151 | switch (buf[0]) { | |
1152 | case '0': | |
1153 | if (fw_dump.dump_registered == 0) { | |
1154 | ret = -EINVAL; | |
1155 | goto unlock_out; | |
1156 | } | |
1157 | /* Un-register Firmware-assisted dump */ | |
1158 | fadump_unregister_dump(&fdm); | |
1159 | break; | |
1160 | case '1': | |
1161 | if (fw_dump.dump_registered == 1) { | |
1162 | ret = -EINVAL; | |
1163 | goto unlock_out; | |
1164 | } | |
1165 | /* Register Firmware-assisted dump */ | |
1166 | register_fadump(); | |
1167 | break; | |
1168 | default: | |
1169 | ret = -EINVAL; | |
1170 | break; | |
1171 | } | |
1172 | ||
1173 | unlock_out: | |
1174 | mutex_unlock(&fadump_mutex); | |
1175 | return ret < 0 ? ret : count; | |
1176 | } | |
1177 | ||
1178 | static int fadump_region_show(struct seq_file *m, void *private) | |
1179 | { | |
1180 | const struct fadump_mem_struct *fdm_ptr; | |
1181 | ||
1182 | if (!fw_dump.fadump_enabled) | |
1183 | return 0; | |
1184 | ||
b500afff | 1185 | mutex_lock(&fadump_mutex); |
3ccc00a7 MS |
1186 | if (fdm_active) |
1187 | fdm_ptr = fdm_active; | |
b500afff MS |
1188 | else { |
1189 | mutex_unlock(&fadump_mutex); | |
3ccc00a7 | 1190 | fdm_ptr = &fdm; |
b500afff | 1191 | } |
3ccc00a7 MS |
1192 | |
1193 | seq_printf(m, | |
1194 | "CPU : [%#016llx-%#016llx] %#llx bytes, " | |
1195 | "Dumped: %#llx\n", | |
408cddd9 HB |
1196 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address), |
1197 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) + | |
1198 | be64_to_cpu(fdm_ptr->cpu_state_data.source_len) - 1, | |
1199 | be64_to_cpu(fdm_ptr->cpu_state_data.source_len), | |
1200 | be64_to_cpu(fdm_ptr->cpu_state_data.bytes_dumped)); | |
3ccc00a7 MS |
1201 | seq_printf(m, |
1202 | "HPTE: [%#016llx-%#016llx] %#llx bytes, " | |
1203 | "Dumped: %#llx\n", | |
408cddd9 HB |
1204 | be64_to_cpu(fdm_ptr->hpte_region.destination_address), |
1205 | be64_to_cpu(fdm_ptr->hpte_region.destination_address) + | |
1206 | be64_to_cpu(fdm_ptr->hpte_region.source_len) - 1, | |
1207 | be64_to_cpu(fdm_ptr->hpte_region.source_len), | |
1208 | be64_to_cpu(fdm_ptr->hpte_region.bytes_dumped)); | |
3ccc00a7 MS |
1209 | seq_printf(m, |
1210 | "DUMP: [%#016llx-%#016llx] %#llx bytes, " | |
1211 | "Dumped: %#llx\n", | |
408cddd9 HB |
1212 | be64_to_cpu(fdm_ptr->rmr_region.destination_address), |
1213 | be64_to_cpu(fdm_ptr->rmr_region.destination_address) + | |
1214 | be64_to_cpu(fdm_ptr->rmr_region.source_len) - 1, | |
1215 | be64_to_cpu(fdm_ptr->rmr_region.source_len), | |
1216 | be64_to_cpu(fdm_ptr->rmr_region.bytes_dumped)); | |
3ccc00a7 MS |
1217 | |
1218 | if (!fdm_active || | |
1219 | (fw_dump.reserve_dump_area_start == | |
408cddd9 | 1220 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address))) |
b500afff | 1221 | goto out; |
3ccc00a7 MS |
1222 | |
1223 | /* Dump is active. Show reserved memory region. */ | |
1224 | seq_printf(m, | |
1225 | " : [%#016llx-%#016llx] %#llx bytes, " | |
1226 | "Dumped: %#llx\n", | |
1227 | (unsigned long long)fw_dump.reserve_dump_area_start, | |
408cddd9 HB |
1228 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - 1, |
1229 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - | |
3ccc00a7 | 1230 | fw_dump.reserve_dump_area_start, |
408cddd9 | 1231 | be64_to_cpu(fdm_ptr->cpu_state_data.destination_address) - |
3ccc00a7 | 1232 | fw_dump.reserve_dump_area_start); |
b500afff MS |
1233 | out: |
1234 | if (fdm_active) | |
1235 | mutex_unlock(&fadump_mutex); | |
3ccc00a7 MS |
1236 | return 0; |
1237 | } | |
1238 | ||
b500afff MS |
1239 | static struct kobj_attribute fadump_release_attr = __ATTR(fadump_release_mem, |
1240 | 0200, NULL, | |
1241 | fadump_release_memory_store); | |
3ccc00a7 MS |
1242 | static struct kobj_attribute fadump_attr = __ATTR(fadump_enabled, |
1243 | 0444, fadump_enabled_show, | |
1244 | NULL); | |
1245 | static struct kobj_attribute fadump_register_attr = __ATTR(fadump_registered, | |
1246 | 0644, fadump_register_show, | |
1247 | fadump_register_store); | |
1248 | ||
1249 | static int fadump_region_open(struct inode *inode, struct file *file) | |
1250 | { | |
1251 | return single_open(file, fadump_region_show, inode->i_private); | |
1252 | } | |
1253 | ||
1254 | static const struct file_operations fadump_region_fops = { | |
1255 | .open = fadump_region_open, | |
1256 | .read = seq_read, | |
1257 | .llseek = seq_lseek, | |
1258 | .release = single_release, | |
1259 | }; | |
1260 | ||
1261 | static void fadump_init_files(void) | |
1262 | { | |
1263 | struct dentry *debugfs_file; | |
1264 | int rc = 0; | |
1265 | ||
1266 | rc = sysfs_create_file(kernel_kobj, &fadump_attr.attr); | |
1267 | if (rc) | |
1268 | printk(KERN_ERR "fadump: unable to create sysfs file" | |
1269 | " fadump_enabled (%d)\n", rc); | |
1270 | ||
1271 | rc = sysfs_create_file(kernel_kobj, &fadump_register_attr.attr); | |
1272 | if (rc) | |
1273 | printk(KERN_ERR "fadump: unable to create sysfs file" | |
1274 | " fadump_registered (%d)\n", rc); | |
1275 | ||
1276 | debugfs_file = debugfs_create_file("fadump_region", 0444, | |
1277 | powerpc_debugfs_root, NULL, | |
1278 | &fadump_region_fops); | |
1279 | if (!debugfs_file) | |
1280 | printk(KERN_ERR "fadump: unable to create debugfs file" | |
1281 | " fadump_region\n"); | |
b500afff MS |
1282 | |
1283 | if (fw_dump.dump_active) { | |
1284 | rc = sysfs_create_file(kernel_kobj, &fadump_release_attr.attr); | |
1285 | if (rc) | |
1286 | printk(KERN_ERR "fadump: unable to create sysfs file" | |
1287 | " fadump_release_mem (%d)\n", rc); | |
1288 | } | |
3ccc00a7 MS |
1289 | return; |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Prepare for firmware-assisted dump. | |
1294 | */ | |
1295 | int __init setup_fadump(void) | |
1296 | { | |
1297 | if (!fw_dump.fadump_enabled) | |
1298 | return 0; | |
1299 | ||
1300 | if (!fw_dump.fadump_supported) { | |
1301 | printk(KERN_ERR "Firmware-assisted dump is not supported on" | |
1302 | " this hardware\n"); | |
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | fadump_show_config(); | |
2df173d9 MS |
1307 | /* |
1308 | * If dump data is available then see if it is valid and prepare for | |
1309 | * saving it to the disk. | |
1310 | */ | |
b500afff MS |
1311 | if (fw_dump.dump_active) { |
1312 | /* | |
1313 | * if dump process fails then invalidate the registration | |
1314 | * and release memory before proceeding for re-registration. | |
1315 | */ | |
1316 | if (process_fadump(fdm_active) < 0) | |
1317 | fadump_invalidate_release_mem(); | |
1318 | } | |
3ccc00a7 | 1319 | /* Initialize the kernel dump memory structure for FAD registration. */ |
2df173d9 | 1320 | else if (fw_dump.reserve_dump_area_size) |
3ccc00a7 MS |
1321 | init_fadump_mem_struct(&fdm, fw_dump.reserve_dump_area_start); |
1322 | fadump_init_files(); | |
1323 | ||
1324 | return 1; | |
1325 | } | |
1326 | subsys_initcall(setup_fadump); |