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