Commit | Line | Data |
---|---|---|
fce0d574 | 1 | /* |
3d1229d6 | 2 | * PPC64 code to handle Linux booting another kernel. |
fce0d574 S |
3 | * |
4 | * Copyright (C) 2004-2005, IBM Corp. | |
5 | * | |
6 | * Created by: Milton D Miller II | |
7 | * | |
8 | * This source code is licensed under the GNU General Public License, | |
9 | * Version 2. See the file COPYING for more details. | |
10 | */ | |
11 | ||
12 | ||
fce0d574 S |
13 | #include <linux/kexec.h> |
14 | #include <linux/smp.h> | |
15 | #include <linux/thread_info.h> | |
d200c922 | 16 | #include <linux/init_task.h> |
fce0d574 | 17 | #include <linux/errno.h> |
e2f7f737 | 18 | #include <linux/kernel.h> |
e8e5c215 | 19 | #include <linux/cpu.h> |
79c66ce8 | 20 | #include <linux/hardirq.h> |
fce0d574 S |
21 | |
22 | #include <asm/page.h> | |
23 | #include <asm/current.h> | |
24 | #include <asm/machdep.h> | |
25 | #include <asm/cacheflush.h> | |
26 | #include <asm/paca.h> | |
27 | #include <asm/mmu.h> | |
28 | #include <asm/sections.h> /* _end */ | |
29 | #include <asm/prom.h> | |
2249ca9d | 30 | #include <asm/smp.h> |
5aae8a53 | 31 | #include <asm/hw_breakpoint.h> |
fce0d574 | 32 | |
3d1229d6 | 33 | int default_machine_kexec_prepare(struct kimage *image) |
fce0d574 S |
34 | { |
35 | int i; | |
36 | unsigned long begin, end; /* limits of segment */ | |
37 | unsigned long low, high; /* limits of blocked memory range */ | |
38 | struct device_node *node; | |
a7f67bdf JK |
39 | const unsigned long *basep; |
40 | const unsigned int *sizep; | |
fce0d574 S |
41 | |
42 | if (!ppc_md.hpte_clear_all) | |
43 | return -ENOENT; | |
44 | ||
45 | /* | |
46 | * Since we use the kernel fault handlers and paging code to | |
47 | * handle the virtual mode, we must make sure no destination | |
48 | * overlaps kernel static data or bss. | |
49 | */ | |
72414d3f | 50 | for (i = 0; i < image->nr_segments; i++) |
fce0d574 S |
51 | if (image->segment[i].mem < __pa(_end)) |
52 | return -ETXTBSY; | |
53 | ||
54 | /* | |
55 | * For non-LPAR, we absolutely can not overwrite the mmu hash | |
56 | * table, since we are still using the bolted entries in it to | |
57 | * do the copy. Check that here. | |
58 | * | |
59 | * It is safe if the end is below the start of the blocked | |
60 | * region (end <= low), or if the beginning is after the | |
61 | * end of the blocked region (begin >= high). Use the | |
62 | * boolean identity !(a || b) === (!a && !b). | |
63 | */ | |
64 | if (htab_address) { | |
65 | low = __pa(htab_address); | |
337a7128 | 66 | high = low + htab_size_bytes; |
fce0d574 | 67 | |
72414d3f | 68 | for (i = 0; i < image->nr_segments; i++) { |
fce0d574 S |
69 | begin = image->segment[i].mem; |
70 | end = begin + image->segment[i].memsz; | |
71 | ||
72 | if ((begin < high) && (end > low)) | |
73 | return -ETXTBSY; | |
74 | } | |
75 | } | |
76 | ||
77 | /* We also should not overwrite the tce tables */ | |
94db7c5e | 78 | for_each_node_by_type(node, "pci") { |
e2eb6392 SR |
79 | basep = of_get_property(node, "linux,tce-base", NULL); |
80 | sizep = of_get_property(node, "linux,tce-size", NULL); | |
fce0d574 S |
81 | if (basep == NULL || sizep == NULL) |
82 | continue; | |
83 | ||
84 | low = *basep; | |
85 | high = low + (*sizep); | |
86 | ||
72414d3f | 87 | for (i = 0; i < image->nr_segments; i++) { |
fce0d574 S |
88 | begin = image->segment[i].mem; |
89 | end = begin + image->segment[i].memsz; | |
90 | ||
91 | if ((begin < high) && (end > low)) | |
92 | return -ETXTBSY; | |
93 | } | |
94 | } | |
95 | ||
96 | return 0; | |
97 | } | |
98 | ||
fce0d574 S |
99 | #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE) |
100 | ||
101 | static void copy_segments(unsigned long ind) | |
102 | { | |
103 | unsigned long entry; | |
104 | unsigned long *ptr; | |
105 | void *dest; | |
106 | void *addr; | |
107 | ||
108 | /* | |
109 | * We rely on kexec_load to create a lists that properly | |
110 | * initializes these pointers before they are used. | |
111 | * We will still crash if the list is wrong, but at least | |
112 | * the compiler will be quiet. | |
113 | */ | |
114 | ptr = NULL; | |
115 | dest = NULL; | |
116 | ||
117 | for (entry = ind; !(entry & IND_DONE); entry = *ptr++) { | |
118 | addr = __va(entry & PAGE_MASK); | |
119 | ||
120 | switch (entry & IND_FLAGS) { | |
121 | case IND_DESTINATION: | |
122 | dest = addr; | |
123 | break; | |
124 | case IND_INDIRECTION: | |
125 | ptr = addr; | |
126 | break; | |
127 | case IND_SOURCE: | |
128 | copy_page(dest, addr); | |
129 | dest += PAGE_SIZE; | |
130 | } | |
131 | } | |
132 | } | |
133 | ||
134 | void kexec_copy_flush(struct kimage *image) | |
135 | { | |
136 | long i, nr_segments = image->nr_segments; | |
137 | struct kexec_segment ranges[KEXEC_SEGMENT_MAX]; | |
138 | ||
139 | /* save the ranges on the stack to efficiently flush the icache */ | |
140 | memcpy(ranges, image->segment, sizeof(ranges)); | |
141 | ||
142 | /* | |
143 | * After this call we may not use anything allocated in dynamic | |
144 | * memory, including *image. | |
145 | * | |
146 | * Only globals and the stack are allowed. | |
147 | */ | |
148 | copy_segments(image->head); | |
149 | ||
150 | /* | |
151 | * we need to clear the icache for all dest pages sometime, | |
152 | * including ones that were in place on the original copy | |
153 | */ | |
154 | for (i = 0; i < nr_segments; i++) | |
b5666f70 ME |
155 | flush_icache_range((unsigned long)__va(ranges[i].mem), |
156 | (unsigned long)__va(ranges[i].mem + ranges[i].memsz)); | |
fce0d574 S |
157 | } |
158 | ||
159 | #ifdef CONFIG_SMP | |
160 | ||
1fc711f7 MN |
161 | static int kexec_all_irq_disabled = 0; |
162 | ||
1c21a293 | 163 | static void kexec_smp_down(void *arg) |
fce0d574 | 164 | { |
1fc711f7 | 165 | local_irq_disable(); |
8520e443 PF |
166 | hard_irq_disable(); |
167 | ||
1fc711f7 MN |
168 | mb(); /* make sure our irqs are disabled before we say they are */ |
169 | get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF; | |
170 | while(kexec_all_irq_disabled == 0) | |
171 | cpu_relax(); | |
172 | mb(); /* make sure all irqs are disabled before this */ | |
5aae8a53 | 173 | hw_breakpoint_disable(); |
1fc711f7 MN |
174 | /* |
175 | * Now every CPU has IRQs off, we can clear out any pending | |
176 | * IPIs and be sure that no more will come in after this. | |
177 | */ | |
c5e24354 ME |
178 | if (ppc_md.kexec_cpu_down) |
179 | ppc_md.kexec_cpu_down(0, 1); | |
fce0d574 | 180 | |
fce0d574 S |
181 | kexec_smp_wait(); |
182 | /* NOTREACHED */ | |
183 | } | |
184 | ||
1fc711f7 | 185 | static void kexec_prepare_cpus_wait(int wait_state) |
fce0d574 S |
186 | { |
187 | int my_cpu, i, notified=-1; | |
188 | ||
5aae8a53 | 189 | hw_breakpoint_disable(); |
fce0d574 | 190 | my_cpu = get_cpu(); |
e2f7f737 ME |
191 | /* Make sure each CPU has at least made it to the state we need. |
192 | * | |
193 | * FIXME: There is a (slim) chance of a problem if not all of the CPUs | |
194 | * are correctly onlined. If somehow we start a CPU on boot with RTAS | |
195 | * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in | |
196 | * time, the boot CPU will timeout. If it does eventually execute | |
197 | * stuff, the secondary will start up (paca[].cpu_start was written) and | |
198 | * get into a peculiar state. If the platform supports | |
199 | * smp_ops->take_timebase(), the secondary CPU will probably be spinning | |
200 | * in there. If not (i.e. pseries), the secondary will continue on and | |
201 | * try to online itself/idle/etc. If it survives that, we need to find | |
202 | * these possible-but-not-online-but-should-be CPUs and chaperone them | |
203 | * into kexec_smp_wait(). | |
204 | */ | |
b636f137 | 205 | for_each_online_cpu(i) { |
fce0d574 S |
206 | if (i == my_cpu) |
207 | continue; | |
208 | ||
1fc711f7 | 209 | while (paca[i].kexec_state < wait_state) { |
b3ca8093 | 210 | barrier(); |
fce0d574 | 211 | if (i != notified) { |
e2f7f737 ME |
212 | printk(KERN_INFO "kexec: waiting for cpu %d " |
213 | "(physical %d) to enter %i state\n", | |
214 | i, paca[i].hw_cpu_id, wait_state); | |
fce0d574 S |
215 | notified = i; |
216 | } | |
217 | } | |
218 | } | |
1fc711f7 MN |
219 | mb(); |
220 | } | |
221 | ||
e8e5c215 ME |
222 | /* |
223 | * We need to make sure each present CPU is online. The next kernel will scan | |
224 | * the device tree and assume primary threads are online and query secondary | |
225 | * threads via RTAS to online them if required. If we don't online primary | |
226 | * threads, they will be stuck. However, we also online secondary threads as we | |
227 | * may be using 'cede offline'. In this case RTAS doesn't see the secondary | |
228 | * threads as offline -- and again, these CPUs will be stuck. | |
229 | * | |
230 | * So, we online all CPUs that should be running, including secondary threads. | |
231 | */ | |
232 | static void wake_offline_cpus(void) | |
1fc711f7 | 233 | { |
e8e5c215 ME |
234 | int cpu = 0; |
235 | ||
236 | for_each_present_cpu(cpu) { | |
237 | if (!cpu_online(cpu)) { | |
238 | printk(KERN_INFO "kexec: Waking offline cpu %d.\n", | |
239 | cpu); | |
011e4b02 | 240 | WARN_ON(cpu_up(cpu)); |
e8e5c215 ME |
241 | } |
242 | } | |
243 | } | |
1fc711f7 | 244 | |
e8e5c215 ME |
245 | static void kexec_prepare_cpus(void) |
246 | { | |
247 | wake_offline_cpus(); | |
1fc711f7 MN |
248 | smp_call_function(kexec_smp_down, NULL, /* wait */0); |
249 | local_irq_disable(); | |
8520e443 PF |
250 | hard_irq_disable(); |
251 | ||
1fc711f7 MN |
252 | mb(); /* make sure IRQs are disabled before we say they are */ |
253 | get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF; | |
254 | ||
255 | kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF); | |
256 | /* we are sure every CPU has IRQs off at this point */ | |
257 | kexec_all_irq_disabled = 1; | |
fce0d574 S |
258 | |
259 | /* after we tell the others to go down */ | |
c5e24354 ME |
260 | if (ppc_md.kexec_cpu_down) |
261 | ppc_md.kexec_cpu_down(0, 0); | |
fce0d574 | 262 | |
e2f7f737 ME |
263 | /* |
264 | * Before removing MMU mappings make sure all CPUs have entered real | |
265 | * mode: | |
266 | */ | |
1fc711f7 | 267 | kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE); |
fce0d574 | 268 | |
1fc711f7 | 269 | put_cpu(); |
fce0d574 S |
270 | } |
271 | ||
272 | #else /* ! SMP */ | |
273 | ||
274 | static void kexec_prepare_cpus(void) | |
275 | { | |
276 | /* | |
277 | * move the secondarys to us so that we can copy | |
278 | * the new kernel 0-0x100 safely | |
279 | * | |
280 | * do this if kexec in setup.c ? | |
75eedfed OJ |
281 | * |
282 | * We need to release the cpus if we are ever going from an | |
283 | * UP to an SMP kernel. | |
fce0d574 | 284 | */ |
75eedfed | 285 | smp_release_cpus(); |
c5e24354 ME |
286 | if (ppc_md.kexec_cpu_down) |
287 | ppc_md.kexec_cpu_down(0, 0); | |
fce0d574 | 288 | local_irq_disable(); |
8520e443 | 289 | hard_irq_disable(); |
fce0d574 S |
290 | } |
291 | ||
292 | #endif /* SMP */ | |
293 | ||
294 | /* | |
295 | * kexec thread structure and stack. | |
296 | * | |
297 | * We need to make sure that this is 16384-byte aligned due to the | |
298 | * way process stacks are handled. It also must be statically allocated | |
299 | * or allocated as part of the kimage, because everything else may be | |
300 | * overwritten when we copy the kexec image. We piggyback on the | |
301 | * "init_task" linker section here to statically allocate a stack. | |
302 | * | |
303 | * We could use a smaller stack if we don't care about anything using | |
304 | * current, but that audit has not been performed. | |
305 | */ | |
d200c922 JP |
306 | static union thread_union kexec_stack __init_task_data = |
307 | { }; | |
fce0d574 | 308 | |
fc53b420 ME |
309 | /* |
310 | * For similar reasons to the stack above, the kexecing CPU needs to be on a | |
311 | * static PACA; we switch to kexec_paca. | |
312 | */ | |
313 | struct paca_struct kexec_paca; | |
314 | ||
07fb41a7 | 315 | /* Our assembly helper, in misc_64.S */ |
9402c95f JP |
316 | extern void kexec_sequence(void *newstack, unsigned long start, |
317 | void *image, void *control, | |
ff2d8b19 | 318 | void (*clear_all)(void)) __noreturn; |
fce0d574 S |
319 | |
320 | /* too late to fail here */ | |
3d1229d6 | 321 | void default_machine_kexec(struct kimage *image) |
fce0d574 | 322 | { |
fce0d574 S |
323 | /* prepare control code if any */ |
324 | ||
cc532915 ME |
325 | /* |
326 | * If the kexec boot is the normal one, need to shutdown other cpus | |
327 | * into our wait loop and quiesce interrupts. | |
328 | * Otherwise, in the case of crashed mode (crashing_cpu >= 0), | |
329 | * stopping other CPUs and collecting their pt_regs is done before | |
330 | * using debugger IPI. | |
331 | */ | |
332 | ||
c1caae3d | 333 | if (!kdump_in_progress()) |
54622f10 | 334 | kexec_prepare_cpus(); |
fce0d574 | 335 | |
e2f7f737 ME |
336 | pr_debug("kexec: Starting switchover sequence.\n"); |
337 | ||
fce0d574 | 338 | /* switch to a staticly allocated stack. Based on irq stack code. |
79c66ce8 | 339 | * We setup preempt_count to avoid using VMX in memcpy. |
fce0d574 S |
340 | * XXX: the task struct will likely be invalid once we do the copy! |
341 | */ | |
342 | kexec_stack.thread_info.task = current_thread_info()->task; | |
343 | kexec_stack.thread_info.flags = 0; | |
79c66ce8 AB |
344 | kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET; |
345 | kexec_stack.thread_info.cpu = current_thread_info()->cpu; | |
fce0d574 | 346 | |
fc53b420 ME |
347 | /* We need a static PACA, too; copy this CPU's PACA over and switch to |
348 | * it. Also poison per_cpu_offset to catch anyone using non-static | |
349 | * data. | |
350 | */ | |
351 | memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct)); | |
352 | kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL; | |
353 | paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) - | |
354 | kexec_paca.paca_index; | |
355 | setup_paca(&kexec_paca); | |
356 | ||
357 | /* XXX: If anyone does 'dynamic lppacas' this will also need to be | |
358 | * switched to a static version! | |
359 | */ | |
360 | ||
fce0d574 S |
361 | /* Some things are best done in assembly. Finding globals with |
362 | * a toc is easier in C, so pass in what we can. | |
363 | */ | |
364 | kexec_sequence(&kexec_stack, image->start, image, | |
365 | page_address(image->control_code_page), | |
1767c8f3 | 366 | ppc_md.hpte_clear_all); |
fce0d574 S |
367 | /* NOTREACHED */ |
368 | } | |
593e537b ME |
369 | |
370 | /* Values we need to export to the second kernel via the device tree. */ | |
2e8e4f5b | 371 | static unsigned long htab_base; |
ea961a82 | 372 | static unsigned long htab_size; |
593e537b ME |
373 | |
374 | static struct property htab_base_prop = { | |
375 | .name = "linux,htab-base", | |
376 | .length = sizeof(unsigned long), | |
1a38147e | 377 | .value = &htab_base, |
593e537b ME |
378 | }; |
379 | ||
380 | static struct property htab_size_prop = { | |
381 | .name = "linux,htab-size", | |
382 | .length = sizeof(unsigned long), | |
ea961a82 | 383 | .value = &htab_size, |
593e537b ME |
384 | }; |
385 | ||
6f29c329 | 386 | static int __init export_htab_values(void) |
593e537b ME |
387 | { |
388 | struct device_node *node; | |
ed7b2144 | 389 | struct property *prop; |
593e537b | 390 | |
2e8e4f5b DF |
391 | /* On machines with no htab htab_address is NULL */ |
392 | if (!htab_address) | |
6f29c329 | 393 | return -ENODEV; |
2e8e4f5b | 394 | |
593e537b ME |
395 | node = of_find_node_by_path("/chosen"); |
396 | if (!node) | |
6f29c329 | 397 | return -ENODEV; |
593e537b | 398 | |
ed7b2144 | 399 | /* remove any stale propertys so ours can be found */ |
ed7b2144 MM |
400 | prop = of_find_property(node, htab_base_prop.name, NULL); |
401 | if (prop) | |
79d1c712 | 402 | of_remove_property(node, prop); |
ed7b2144 MM |
403 | prop = of_find_property(node, htab_size_prop.name, NULL); |
404 | if (prop) | |
79d1c712 | 405 | of_remove_property(node, prop); |
ed7b2144 | 406 | |
ea961a82 | 407 | htab_base = cpu_to_be64(__pa(htab_address)); |
79d1c712 | 408 | of_add_property(node, &htab_base_prop); |
ea961a82 | 409 | htab_size = cpu_to_be64(htab_size_bytes); |
79d1c712 | 410 | of_add_property(node, &htab_size_prop); |
593e537b | 411 | |
593e537b | 412 | of_node_put(node); |
aa98c50d | 413 | return 0; |
35dd5432 | 414 | } |
6f29c329 | 415 | late_initcall(export_htab_values); |