Merge branch 'for-linus' of git://neil.brown.name/md
[linux-2.6] / arch / powerpc / platforms / pseries / phyp_dump.c
1 /*
2  * Hypervisor-assisted dump
3  *
4  * Linas Vepstas, Manish Ahuja 2008
5  * Copyright 2008 IBM Corp.
6  *
7  *      This program is free software; you can redistribute it and/or
8  *      modify it under the terms of the GNU General Public License
9  *      as published by the Free Software Foundation; either version
10  *      2 of the License, or (at your option) any later version.
11  *
12  */
13
14 #include <linux/init.h>
15 #include <linux/kobject.h>
16 #include <linux/mm.h>
17 #include <linux/of.h>
18 #include <linux/pfn.h>
19 #include <linux/swap.h>
20 #include <linux/sysfs.h>
21
22 #include <asm/page.h>
23 #include <asm/phyp_dump.h>
24 #include <asm/machdep.h>
25 #include <asm/prom.h>
26 #include <asm/rtas.h>
27
28 /* Variables, used to communicate data between early boot and late boot */
29 static struct phyp_dump phyp_dump_vars;
30 struct phyp_dump *phyp_dump_info = &phyp_dump_vars;
31
32 static int ibm_configure_kernel_dump;
33 /* ------------------------------------------------- */
34 /* RTAS interfaces to declare the dump regions */
35
36 struct dump_section {
37         u32 dump_flags;
38         u16 source_type;
39         u16 error_flags;
40         u64 source_address;
41         u64 source_length;
42         u64 length_copied;
43         u64 destination_address;
44 };
45
46 struct phyp_dump_header {
47         u32 version;
48         u16 num_of_sections;
49         u16 status;
50
51         u32 first_offset_section;
52         u32 dump_disk_section;
53         u64 block_num_dd;
54         u64 num_of_blocks_dd;
55         u32 offset_dd;
56         u32 maxtime_to_auto;
57         /* No dump disk path string used */
58
59         struct dump_section cpu_data;
60         struct dump_section hpte_data;
61         struct dump_section kernel_data;
62 };
63
64 /* The dump header *must be* in low memory, so .bss it */
65 static struct phyp_dump_header phdr;
66
67 #define NUM_DUMP_SECTIONS       3
68 #define DUMP_HEADER_VERSION     0x1
69 #define DUMP_REQUEST_FLAG       0x1
70 #define DUMP_SOURCE_CPU         0x0001
71 #define DUMP_SOURCE_HPTE        0x0002
72 #define DUMP_SOURCE_RMO         0x0011
73 #define DUMP_ERROR_FLAG         0x2000
74 #define DUMP_TRIGGERED          0x4000
75 #define DUMP_PERFORMED          0x8000
76
77
78 /**
79  * init_dump_header() - initialize the header declaring a dump
80  * Returns: length of dump save area.
81  *
82  * When the hypervisor saves crashed state, it needs to put
83  * it somewhere. The dump header tells the hypervisor where
84  * the data can be saved.
85  */
86 static unsigned long init_dump_header(struct phyp_dump_header *ph)
87 {
88         unsigned long addr_offset = 0;
89
90         /* Set up the dump header */
91         ph->version = DUMP_HEADER_VERSION;
92         ph->num_of_sections = NUM_DUMP_SECTIONS;
93         ph->status = 0;
94
95         ph->first_offset_section =
96                 (u32)offsetof(struct phyp_dump_header, cpu_data);
97         ph->dump_disk_section = 0;
98         ph->block_num_dd = 0;
99         ph->num_of_blocks_dd = 0;
100         ph->offset_dd = 0;
101
102         ph->maxtime_to_auto = 0; /* disabled */
103
104         /* The first two sections are mandatory */
105         ph->cpu_data.dump_flags = DUMP_REQUEST_FLAG;
106         ph->cpu_data.source_type = DUMP_SOURCE_CPU;
107         ph->cpu_data.source_address = 0;
108         ph->cpu_data.source_length = phyp_dump_info->cpu_state_size;
109         ph->cpu_data.destination_address = addr_offset;
110         addr_offset += phyp_dump_info->cpu_state_size;
111
112         ph->hpte_data.dump_flags = DUMP_REQUEST_FLAG;
113         ph->hpte_data.source_type = DUMP_SOURCE_HPTE;
114         ph->hpte_data.source_address = 0;
115         ph->hpte_data.source_length = phyp_dump_info->hpte_region_size;
116         ph->hpte_data.destination_address = addr_offset;
117         addr_offset += phyp_dump_info->hpte_region_size;
118
119         /* This section describes the low kernel region */
120         ph->kernel_data.dump_flags = DUMP_REQUEST_FLAG;
121         ph->kernel_data.source_type = DUMP_SOURCE_RMO;
122         ph->kernel_data.source_address = PHYP_DUMP_RMR_START;
123         ph->kernel_data.source_length = PHYP_DUMP_RMR_END;
124         ph->kernel_data.destination_address = addr_offset;
125         addr_offset += ph->kernel_data.source_length;
126
127         return addr_offset;
128 }
129
130 static void print_dump_header(const struct phyp_dump_header *ph)
131 {
132 #ifdef DEBUG
133         printk(KERN_INFO "dump header:\n");
134         /* setup some ph->sections required */
135         printk(KERN_INFO "version = %d\n", ph->version);
136         printk(KERN_INFO "Sections = %d\n", ph->num_of_sections);
137         printk(KERN_INFO "Status = 0x%x\n", ph->status);
138
139         /* No ph->disk, so all should be set to 0 */
140         printk(KERN_INFO "Offset to first section 0x%x\n",
141                 ph->first_offset_section);
142         printk(KERN_INFO "dump disk sections should be zero\n");
143         printk(KERN_INFO "dump disk section = %d\n", ph->dump_disk_section);
144         printk(KERN_INFO "block num = %ld\n", ph->block_num_dd);
145         printk(KERN_INFO "number of blocks = %ld\n", ph->num_of_blocks_dd);
146         printk(KERN_INFO "dump disk offset = %d\n", ph->offset_dd);
147         printk(KERN_INFO "Max auto time= %d\n", ph->maxtime_to_auto);
148
149         /*set cpu state and hpte states as well scratch pad area */
150         printk(KERN_INFO " CPU AREA \n");
151         printk(KERN_INFO "cpu dump_flags =%d\n", ph->cpu_data.dump_flags);
152         printk(KERN_INFO "cpu source_type =%d\n", ph->cpu_data.source_type);
153         printk(KERN_INFO "cpu error_flags =%d\n", ph->cpu_data.error_flags);
154         printk(KERN_INFO "cpu source_address =%lx\n",
155                 ph->cpu_data.source_address);
156         printk(KERN_INFO "cpu source_length =%lx\n",
157                 ph->cpu_data.source_length);
158         printk(KERN_INFO "cpu length_copied =%lx\n",
159                 ph->cpu_data.length_copied);
160
161         printk(KERN_INFO " HPTE AREA \n");
162         printk(KERN_INFO "HPTE dump_flags =%d\n", ph->hpte_data.dump_flags);
163         printk(KERN_INFO "HPTE source_type =%d\n", ph->hpte_data.source_type);
164         printk(KERN_INFO "HPTE error_flags =%d\n", ph->hpte_data.error_flags);
165         printk(KERN_INFO "HPTE source_address =%lx\n",
166                 ph->hpte_data.source_address);
167         printk(KERN_INFO "HPTE source_length =%lx\n",
168                 ph->hpte_data.source_length);
169         printk(KERN_INFO "HPTE length_copied =%lx\n",
170                 ph->hpte_data.length_copied);
171
172         printk(KERN_INFO " SRSD AREA \n");
173         printk(KERN_INFO "SRSD dump_flags =%d\n", ph->kernel_data.dump_flags);
174         printk(KERN_INFO "SRSD source_type =%d\n", ph->kernel_data.source_type);
175         printk(KERN_INFO "SRSD error_flags =%d\n", ph->kernel_data.error_flags);
176         printk(KERN_INFO "SRSD source_address =%lx\n",
177                 ph->kernel_data.source_address);
178         printk(KERN_INFO "SRSD source_length =%lx\n",
179                 ph->kernel_data.source_length);
180         printk(KERN_INFO "SRSD length_copied =%lx\n",
181                 ph->kernel_data.length_copied);
182 #endif
183 }
184
185 static ssize_t show_phyp_dump_active(struct kobject *kobj,
186                         struct kobj_attribute *attr, char *buf)
187 {
188
189         /* create filesystem entry so kdump is phyp-dump aware */
190         return sprintf(buf, "%lx\n", phyp_dump_info->phyp_dump_at_boot);
191 }
192
193 static struct kobj_attribute pdl = __ATTR(phyp_dump_active, 0600,
194                                         show_phyp_dump_active,
195                                         NULL);
196
197 static void register_dump_area(struct phyp_dump_header *ph, unsigned long addr)
198 {
199         int rc;
200
201         /* Add addr value if not initialized before */
202         if (ph->cpu_data.destination_address == 0) {
203                 ph->cpu_data.destination_address += addr;
204                 ph->hpte_data.destination_address += addr;
205                 ph->kernel_data.destination_address += addr;
206         }
207
208         /* ToDo Invalidate kdump and free memory range. */
209
210         do {
211                 rc = rtas_call(ibm_configure_kernel_dump, 3, 1, NULL,
212                                 1, ph, sizeof(struct phyp_dump_header));
213         } while (rtas_busy_delay(rc));
214
215         if (rc) {
216                 printk(KERN_ERR "phyp-dump: unexpected error (%d) on "
217                                                 "register\n", rc);
218                 print_dump_header(ph);
219                 return;
220         }
221
222         rc = sysfs_create_file(kernel_kobj, &pdl.attr);
223         if (rc)
224                 printk(KERN_ERR "phyp-dump: unable to create sysfs"
225                                 " file (%d)\n", rc);
226 }
227
228 static
229 void invalidate_last_dump(struct phyp_dump_header *ph, unsigned long addr)
230 {
231         int rc;
232
233         /* Add addr value if not initialized before */
234         if (ph->cpu_data.destination_address == 0) {
235                 ph->cpu_data.destination_address += addr;
236                 ph->hpte_data.destination_address += addr;
237                 ph->kernel_data.destination_address += addr;
238         }
239
240         do {
241                 rc = rtas_call(ibm_configure_kernel_dump, 3, 1, NULL,
242                                 2, ph, sizeof(struct phyp_dump_header));
243         } while (rtas_busy_delay(rc));
244
245         if (rc) {
246                 printk(KERN_ERR "phyp-dump: unexpected error (%d) "
247                                                 "on invalidate\n", rc);
248                 print_dump_header(ph);
249         }
250 }
251
252 /* ------------------------------------------------- */
253 /**
254  * release_memory_range -- release memory previously lmb_reserved
255  * @start_pfn: starting physical frame number
256  * @nr_pages: number of pages to free.
257  *
258  * This routine will release memory that had been previously
259  * lmb_reserved in early boot. The released memory becomes
260  * available for genreal use.
261  */
262 static void release_memory_range(unsigned long start_pfn,
263                         unsigned long nr_pages)
264 {
265         struct page *rpage;
266         unsigned long end_pfn;
267         long i;
268
269         end_pfn = start_pfn + nr_pages;
270
271         for (i = start_pfn; i <= end_pfn; i++) {
272                 rpage = pfn_to_page(i);
273                 if (PageReserved(rpage)) {
274                         ClearPageReserved(rpage);
275                         init_page_count(rpage);
276                         __free_page(rpage);
277                         totalram_pages++;
278                 }
279         }
280 }
281
282 /**
283  * track_freed_range -- Counts the range being freed.
284  * Once the counter goes to zero, it re-registers dump for
285  * future use.
286  */
287 static void
288 track_freed_range(unsigned long addr, unsigned long length)
289 {
290         static unsigned long scratch_area_size, reserved_area_size;
291
292         if (addr < phyp_dump_info->init_reserve_start)
293                 return;
294
295         if ((addr >= phyp_dump_info->init_reserve_start) &&
296             (addr <= phyp_dump_info->init_reserve_start +
297              phyp_dump_info->init_reserve_size))
298                 reserved_area_size += length;
299
300         if ((addr >= phyp_dump_info->reserved_scratch_addr) &&
301             (addr <= phyp_dump_info->reserved_scratch_addr +
302              phyp_dump_info->reserved_scratch_size))
303                 scratch_area_size += length;
304
305         if ((reserved_area_size == phyp_dump_info->init_reserve_size) &&
306             (scratch_area_size == phyp_dump_info->reserved_scratch_size)) {
307
308                 invalidate_last_dump(&phdr,
309                                 phyp_dump_info->reserved_scratch_addr);
310                 register_dump_area(&phdr,
311                                 phyp_dump_info->reserved_scratch_addr);
312         }
313 }
314
315 /* ------------------------------------------------- */
316 /**
317  * sysfs_release_region -- sysfs interface to release memory range.
318  *
319  * Usage:
320  *   "echo <start addr> <length> > /sys/kernel/release_region"
321  *
322  * Example:
323  *   "echo 0x40000000 0x10000000 > /sys/kernel/release_region"
324  *
325  * will release 256MB starting at 1GB.
326  */
327 static ssize_t store_release_region(struct kobject *kobj,
328                                 struct kobj_attribute *attr,
329                                 const char *buf, size_t count)
330 {
331         unsigned long start_addr, length, end_addr;
332         unsigned long start_pfn, nr_pages;
333         ssize_t ret;
334
335         ret = sscanf(buf, "%lx %lx", &start_addr, &length);
336         if (ret != 2)
337                 return -EINVAL;
338
339         track_freed_range(start_addr, length);
340
341         /* Range-check - don't free any reserved memory that
342          * wasn't reserved for phyp-dump */
343         if (start_addr < phyp_dump_info->init_reserve_start)
344                 start_addr = phyp_dump_info->init_reserve_start;
345
346         end_addr = phyp_dump_info->init_reserve_start +
347                         phyp_dump_info->init_reserve_size;
348         if (start_addr+length > end_addr)
349                 length = end_addr - start_addr;
350
351         /* Release the region of memory assed in by user */
352         start_pfn = PFN_DOWN(start_addr);
353         nr_pages = PFN_DOWN(length);
354         release_memory_range(start_pfn, nr_pages);
355
356         return count;
357 }
358
359 static ssize_t show_release_region(struct kobject *kobj,
360                         struct kobj_attribute *attr, char *buf)
361 {
362         u64 second_addr_range;
363
364         /* total reserved size - start of scratch area */
365         second_addr_range = phyp_dump_info->init_reserve_size -
366                                 phyp_dump_info->reserved_scratch_size;
367         return sprintf(buf, "CPU:0x%lx-0x%lx: HPTE:0x%lx-0x%lx:"
368                             " DUMP:0x%lx-0x%lx, 0x%lx-0x%lx:\n",
369                 phdr.cpu_data.destination_address,
370                 phdr.cpu_data.length_copied,
371                 phdr.hpte_data.destination_address,
372                 phdr.hpte_data.length_copied,
373                 phdr.kernel_data.destination_address,
374                 phdr.kernel_data.length_copied,
375                 phyp_dump_info->init_reserve_start,
376                 second_addr_range);
377 }
378
379 static struct kobj_attribute rr = __ATTR(release_region, 0600,
380                                         show_release_region,
381                                         store_release_region);
382
383 static int __init phyp_dump_setup(void)
384 {
385         struct device_node *rtas;
386         const struct phyp_dump_header *dump_header = NULL;
387         unsigned long dump_area_start;
388         unsigned long dump_area_length;
389         int header_len = 0;
390         int rc;
391
392         /* If no memory was reserved in early boot, there is nothing to do */
393         if (phyp_dump_info->init_reserve_size == 0)
394                 return 0;
395
396         /* Return if phyp dump not supported */
397         if (!phyp_dump_info->phyp_dump_configured)
398                 return -ENOSYS;
399
400         /* Is there dump data waiting for us? If there isn't,
401          * then register a new dump area, and release all of
402          * the rest of the reserved ram.
403          *
404          * The /rtas/ibm,kernel-dump rtas node is present only
405          * if there is dump data waiting for us.
406          */
407         rtas = of_find_node_by_path("/rtas");
408         if (rtas) {
409                 dump_header = of_get_property(rtas, "ibm,kernel-dump",
410                                                 &header_len);
411                 of_node_put(rtas);
412         }
413
414         print_dump_header(dump_header);
415         dump_area_length = init_dump_header(&phdr);
416         /* align down */
417         dump_area_start = phyp_dump_info->init_reserve_start & PAGE_MASK;
418
419         if (dump_header == NULL) {
420                 register_dump_area(&phdr, dump_area_start);
421                 return 0;
422         }
423
424         /* re-register the dump area, if old dump was invalid */
425         if ((dump_header) && (dump_header->status & DUMP_ERROR_FLAG)) {
426                 invalidate_last_dump(&phdr, dump_area_start);
427                 register_dump_area(&phdr, dump_area_start);
428                 return 0;
429         }
430
431         if (dump_header) {
432                 phyp_dump_info->reserved_scratch_addr =
433                                 dump_header->cpu_data.destination_address;
434                 phyp_dump_info->reserved_scratch_size =
435                                 dump_header->cpu_data.source_length +
436                                 dump_header->hpte_data.source_length +
437                                 dump_header->kernel_data.source_length;
438         }
439
440         /* Should we create a dump_subsys, analogous to s390/ipl.c ? */
441         rc = sysfs_create_file(kernel_kobj, &rr.attr);
442         if (rc)
443                 printk(KERN_ERR "phyp-dump: unable to create sysfs file (%d)\n",
444                                                                         rc);
445
446         /* ToDo: re-register the dump area, for next time. */
447         return 0;
448 }
449 machine_subsys_initcall(pseries, phyp_dump_setup);
450
451 int __init early_init_dt_scan_phyp_dump(unsigned long node,
452                 const char *uname, int depth, void *data)
453 {
454         const unsigned int *sizes;
455
456         phyp_dump_info->phyp_dump_configured = 0;
457         phyp_dump_info->phyp_dump_is_active = 0;
458
459         if (depth != 1 || strcmp(uname, "rtas") != 0)
460                 return 0;
461
462         if (of_get_flat_dt_prop(node, "ibm,configure-kernel-dump", NULL))
463                 phyp_dump_info->phyp_dump_configured++;
464
465         if (of_get_flat_dt_prop(node, "ibm,dump-kernel", NULL))
466                 phyp_dump_info->phyp_dump_is_active++;
467
468         sizes = of_get_flat_dt_prop(node, "ibm,configure-kernel-dump-sizes",
469                                     NULL);
470         if (!sizes)
471                 return 0;
472
473         if (sizes[0] == 1)
474                 phyp_dump_info->cpu_state_size = *((unsigned long *)&sizes[1]);
475
476         if (sizes[3] == 2)
477                 phyp_dump_info->hpte_region_size =
478                                                 *((unsigned long *)&sizes[4]);
479         return 1;
480 }
481
482 /* Look for phyp_dump= cmdline option */
483 static int __init early_phyp_dump_enabled(char *p)
484 {
485         phyp_dump_info->phyp_dump_at_boot = 1;
486
487         if (!p)
488                 return 0;
489
490         if (strncmp(p, "1", 1) == 0)
491                 phyp_dump_info->phyp_dump_at_boot = 1;
492         else if (strncmp(p, "0", 1) == 0)
493                 phyp_dump_info->phyp_dump_at_boot = 0;
494
495         return 0;
496 }
497 early_param("phyp_dump", early_phyp_dump_enabled);
498
499 /* Look for phyp_dump_reserve_size= cmdline option */
500 static int __init early_phyp_dump_reserve_size(char *p)
501 {
502         if (p)
503                 phyp_dump_info->reserve_bootvar = memparse(p, &p);
504
505         return 0;
506 }
507 early_param("phyp_dump_reserve_size", early_phyp_dump_reserve_size);