]>
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> | |
b970b41e | 26 | #include <asm/firmware.h> |
fce0d574 S |
27 | #include <asm/paca.h> |
28 | #include <asm/mmu.h> | |
29 | #include <asm/sections.h> /* _end */ | |
30 | #include <asm/prom.h> | |
2249ca9d | 31 | #include <asm/smp.h> |
5aae8a53 | 32 | #include <asm/hw_breakpoint.h> |
42f5b4ca | 33 | #include <asm/asm-prototypes.h> |
fce0d574 | 34 | |
3d1229d6 | 35 | int default_machine_kexec_prepare(struct kimage *image) |
fce0d574 S |
36 | { |
37 | int i; | |
38 | unsigned long begin, end; /* limits of segment */ | |
39 | unsigned long low, high; /* limits of blocked memory range */ | |
40 | struct device_node *node; | |
a7f67bdf JK |
41 | const unsigned long *basep; |
42 | const unsigned int *sizep; | |
fce0d574 | 43 | |
fce0d574 S |
44 | /* |
45 | * Since we use the kernel fault handlers and paging code to | |
46 | * handle the virtual mode, we must make sure no destination | |
47 | * overlaps kernel static data or bss. | |
48 | */ | |
72414d3f | 49 | for (i = 0; i < image->nr_segments; i++) |
fce0d574 S |
50 | if (image->segment[i].mem < __pa(_end)) |
51 | return -ETXTBSY; | |
52 | ||
fce0d574 | 53 | /* We also should not overwrite the tce tables */ |
94db7c5e | 54 | for_each_node_by_type(node, "pci") { |
e2eb6392 SR |
55 | basep = of_get_property(node, "linux,tce-base", NULL); |
56 | sizep = of_get_property(node, "linux,tce-size", NULL); | |
fce0d574 S |
57 | if (basep == NULL || sizep == NULL) |
58 | continue; | |
59 | ||
60 | low = *basep; | |
61 | high = low + (*sizep); | |
62 | ||
72414d3f | 63 | for (i = 0; i < image->nr_segments; i++) { |
fce0d574 S |
64 | begin = image->segment[i].mem; |
65 | end = begin + image->segment[i].memsz; | |
66 | ||
67 | if ((begin < high) && (end > low)) | |
68 | return -ETXTBSY; | |
69 | } | |
70 | } | |
71 | ||
72 | return 0; | |
73 | } | |
74 | ||
fce0d574 S |
75 | static void copy_segments(unsigned long ind) |
76 | { | |
77 | unsigned long entry; | |
78 | unsigned long *ptr; | |
79 | void *dest; | |
80 | void *addr; | |
81 | ||
82 | /* | |
83 | * We rely on kexec_load to create a lists that properly | |
84 | * initializes these pointers before they are used. | |
85 | * We will still crash if the list is wrong, but at least | |
86 | * the compiler will be quiet. | |
87 | */ | |
88 | ptr = NULL; | |
89 | dest = NULL; | |
90 | ||
91 | for (entry = ind; !(entry & IND_DONE); entry = *ptr++) { | |
92 | addr = __va(entry & PAGE_MASK); | |
93 | ||
94 | switch (entry & IND_FLAGS) { | |
95 | case IND_DESTINATION: | |
96 | dest = addr; | |
97 | break; | |
98 | case IND_INDIRECTION: | |
99 | ptr = addr; | |
100 | break; | |
101 | case IND_SOURCE: | |
102 | copy_page(dest, addr); | |
103 | dest += PAGE_SIZE; | |
104 | } | |
105 | } | |
106 | } | |
107 | ||
108 | void kexec_copy_flush(struct kimage *image) | |
109 | { | |
110 | long i, nr_segments = image->nr_segments; | |
111 | struct kexec_segment ranges[KEXEC_SEGMENT_MAX]; | |
112 | ||
113 | /* save the ranges on the stack to efficiently flush the icache */ | |
114 | memcpy(ranges, image->segment, sizeof(ranges)); | |
115 | ||
116 | /* | |
117 | * After this call we may not use anything allocated in dynamic | |
118 | * memory, including *image. | |
119 | * | |
120 | * Only globals and the stack are allowed. | |
121 | */ | |
122 | copy_segments(image->head); | |
123 | ||
124 | /* | |
125 | * we need to clear the icache for all dest pages sometime, | |
126 | * including ones that were in place on the original copy | |
127 | */ | |
128 | for (i = 0; i < nr_segments; i++) | |
b5666f70 ME |
129 | flush_icache_range((unsigned long)__va(ranges[i].mem), |
130 | (unsigned long)__va(ranges[i].mem + ranges[i].memsz)); | |
fce0d574 S |
131 | } |
132 | ||
133 | #ifdef CONFIG_SMP | |
134 | ||
1fc711f7 MN |
135 | static int kexec_all_irq_disabled = 0; |
136 | ||
1c21a293 | 137 | static void kexec_smp_down(void *arg) |
fce0d574 | 138 | { |
1fc711f7 | 139 | local_irq_disable(); |
8520e443 PF |
140 | hard_irq_disable(); |
141 | ||
1fc711f7 MN |
142 | mb(); /* make sure our irqs are disabled before we say they are */ |
143 | get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF; | |
144 | while(kexec_all_irq_disabled == 0) | |
145 | cpu_relax(); | |
146 | mb(); /* make sure all irqs are disabled before this */ | |
5aae8a53 | 147 | hw_breakpoint_disable(); |
1fc711f7 MN |
148 | /* |
149 | * Now every CPU has IRQs off, we can clear out any pending | |
150 | * IPIs and be sure that no more will come in after this. | |
151 | */ | |
c5e24354 ME |
152 | if (ppc_md.kexec_cpu_down) |
153 | ppc_md.kexec_cpu_down(0, 1); | |
fce0d574 | 154 | |
fce0d574 S |
155 | kexec_smp_wait(); |
156 | /* NOTREACHED */ | |
157 | } | |
158 | ||
1fc711f7 | 159 | static void kexec_prepare_cpus_wait(int wait_state) |
fce0d574 S |
160 | { |
161 | int my_cpu, i, notified=-1; | |
162 | ||
5aae8a53 | 163 | hw_breakpoint_disable(); |
fce0d574 | 164 | my_cpu = get_cpu(); |
e2f7f737 ME |
165 | /* Make sure each CPU has at least made it to the state we need. |
166 | * | |
167 | * FIXME: There is a (slim) chance of a problem if not all of the CPUs | |
168 | * are correctly onlined. If somehow we start a CPU on boot with RTAS | |
169 | * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in | |
170 | * time, the boot CPU will timeout. If it does eventually execute | |
171 | * stuff, the secondary will start up (paca[].cpu_start was written) and | |
172 | * get into a peculiar state. If the platform supports | |
173 | * smp_ops->take_timebase(), the secondary CPU will probably be spinning | |
174 | * in there. If not (i.e. pseries), the secondary will continue on and | |
175 | * try to online itself/idle/etc. If it survives that, we need to find | |
176 | * these possible-but-not-online-but-should-be CPUs and chaperone them | |
177 | * into kexec_smp_wait(). | |
178 | */ | |
b636f137 | 179 | for_each_online_cpu(i) { |
fce0d574 S |
180 | if (i == my_cpu) |
181 | continue; | |
182 | ||
1fc711f7 | 183 | while (paca[i].kexec_state < wait_state) { |
b3ca8093 | 184 | barrier(); |
fce0d574 | 185 | if (i != notified) { |
e2f7f737 ME |
186 | printk(KERN_INFO "kexec: waiting for cpu %d " |
187 | "(physical %d) to enter %i state\n", | |
188 | i, paca[i].hw_cpu_id, wait_state); | |
fce0d574 S |
189 | notified = i; |
190 | } | |
191 | } | |
192 | } | |
1fc711f7 MN |
193 | mb(); |
194 | } | |
195 | ||
e8e5c215 ME |
196 | /* |
197 | * We need to make sure each present CPU is online. The next kernel will scan | |
198 | * the device tree and assume primary threads are online and query secondary | |
199 | * threads via RTAS to online them if required. If we don't online primary | |
200 | * threads, they will be stuck. However, we also online secondary threads as we | |
201 | * may be using 'cede offline'. In this case RTAS doesn't see the secondary | |
202 | * threads as offline -- and again, these CPUs will be stuck. | |
203 | * | |
204 | * So, we online all CPUs that should be running, including secondary threads. | |
205 | */ | |
206 | static void wake_offline_cpus(void) | |
1fc711f7 | 207 | { |
e8e5c215 ME |
208 | int cpu = 0; |
209 | ||
210 | for_each_present_cpu(cpu) { | |
211 | if (!cpu_online(cpu)) { | |
212 | printk(KERN_INFO "kexec: Waking offline cpu %d.\n", | |
213 | cpu); | |
011e4b02 | 214 | WARN_ON(cpu_up(cpu)); |
e8e5c215 ME |
215 | } |
216 | } | |
217 | } | |
1fc711f7 | 218 | |
e8e5c215 ME |
219 | static void kexec_prepare_cpus(void) |
220 | { | |
221 | wake_offline_cpus(); | |
1fc711f7 MN |
222 | smp_call_function(kexec_smp_down, NULL, /* wait */0); |
223 | local_irq_disable(); | |
8520e443 PF |
224 | hard_irq_disable(); |
225 | ||
1fc711f7 MN |
226 | mb(); /* make sure IRQs are disabled before we say they are */ |
227 | get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF; | |
228 | ||
229 | kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF); | |
230 | /* we are sure every CPU has IRQs off at this point */ | |
231 | kexec_all_irq_disabled = 1; | |
fce0d574 S |
232 | |
233 | /* after we tell the others to go down */ | |
c5e24354 ME |
234 | if (ppc_md.kexec_cpu_down) |
235 | ppc_md.kexec_cpu_down(0, 0); | |
fce0d574 | 236 | |
e2f7f737 ME |
237 | /* |
238 | * Before removing MMU mappings make sure all CPUs have entered real | |
239 | * mode: | |
240 | */ | |
1fc711f7 | 241 | kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE); |
fce0d574 | 242 | |
1fc711f7 | 243 | put_cpu(); |
fce0d574 S |
244 | } |
245 | ||
246 | #else /* ! SMP */ | |
247 | ||
248 | static void kexec_prepare_cpus(void) | |
249 | { | |
250 | /* | |
251 | * move the secondarys to us so that we can copy | |
252 | * the new kernel 0-0x100 safely | |
253 | * | |
254 | * do this if kexec in setup.c ? | |
75eedfed OJ |
255 | * |
256 | * We need to release the cpus if we are ever going from an | |
257 | * UP to an SMP kernel. | |
fce0d574 | 258 | */ |
75eedfed | 259 | smp_release_cpus(); |
c5e24354 ME |
260 | if (ppc_md.kexec_cpu_down) |
261 | ppc_md.kexec_cpu_down(0, 0); | |
fce0d574 | 262 | local_irq_disable(); |
8520e443 | 263 | hard_irq_disable(); |
fce0d574 S |
264 | } |
265 | ||
266 | #endif /* SMP */ | |
267 | ||
268 | /* | |
269 | * kexec thread structure and stack. | |
270 | * | |
271 | * We need to make sure that this is 16384-byte aligned due to the | |
272 | * way process stacks are handled. It also must be statically allocated | |
273 | * or allocated as part of the kimage, because everything else may be | |
274 | * overwritten when we copy the kexec image. We piggyback on the | |
275 | * "init_task" linker section here to statically allocate a stack. | |
276 | * | |
277 | * We could use a smaller stack if we don't care about anything using | |
278 | * current, but that audit has not been performed. | |
279 | */ | |
d200c922 JP |
280 | static union thread_union kexec_stack __init_task_data = |
281 | { }; | |
fce0d574 | 282 | |
fc53b420 ME |
283 | /* |
284 | * For similar reasons to the stack above, the kexecing CPU needs to be on a | |
285 | * static PACA; we switch to kexec_paca. | |
286 | */ | |
287 | struct paca_struct kexec_paca; | |
288 | ||
07fb41a7 | 289 | /* Our assembly helper, in misc_64.S */ |
9402c95f JP |
290 | extern void kexec_sequence(void *newstack, unsigned long start, |
291 | void *image, void *control, | |
b970b41e BH |
292 | void (*clear_all)(void), |
293 | bool copy_with_mmu_off) __noreturn; | |
fce0d574 S |
294 | |
295 | /* too late to fail here */ | |
3d1229d6 | 296 | void default_machine_kexec(struct kimage *image) |
fce0d574 | 297 | { |
b970b41e BH |
298 | bool copy_with_mmu_off; |
299 | ||
fce0d574 S |
300 | /* prepare control code if any */ |
301 | ||
cc532915 ME |
302 | /* |
303 | * If the kexec boot is the normal one, need to shutdown other cpus | |
304 | * into our wait loop and quiesce interrupts. | |
305 | * Otherwise, in the case of crashed mode (crashing_cpu >= 0), | |
306 | * stopping other CPUs and collecting their pt_regs is done before | |
307 | * using debugger IPI. | |
308 | */ | |
309 | ||
c1caae3d | 310 | if (!kdump_in_progress()) |
54622f10 | 311 | kexec_prepare_cpus(); |
fce0d574 | 312 | |
0d976313 | 313 | printk("kexec: Starting switchover sequence.\n"); |
e2f7f737 | 314 | |
fce0d574 | 315 | /* switch to a staticly allocated stack. Based on irq stack code. |
79c66ce8 | 316 | * We setup preempt_count to avoid using VMX in memcpy. |
fce0d574 S |
317 | * XXX: the task struct will likely be invalid once we do the copy! |
318 | */ | |
319 | kexec_stack.thread_info.task = current_thread_info()->task; | |
320 | kexec_stack.thread_info.flags = 0; | |
79c66ce8 AB |
321 | kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET; |
322 | kexec_stack.thread_info.cpu = current_thread_info()->cpu; | |
fce0d574 | 323 | |
fc53b420 ME |
324 | /* We need a static PACA, too; copy this CPU's PACA over and switch to |
325 | * it. Also poison per_cpu_offset to catch anyone using non-static | |
326 | * data. | |
327 | */ | |
328 | memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct)); | |
329 | kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL; | |
330 | paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) - | |
331 | kexec_paca.paca_index; | |
332 | setup_paca(&kexec_paca); | |
333 | ||
334 | /* XXX: If anyone does 'dynamic lppacas' this will also need to be | |
335 | * switched to a static version! | |
336 | */ | |
b970b41e BH |
337 | /* |
338 | * On Book3S, the copy must happen with the MMU off if we are either | |
339 | * using Radix page tables or we are not in an LPAR since we can | |
340 | * overwrite the page tables while copying. | |
341 | * | |
342 | * In an LPAR, we keep the MMU on otherwise we can't access beyond | |
343 | * the RMA. On BookE there is no real MMU off mode, so we have to | |
344 | * keep it enabled as well (but then we have bolted TLB entries). | |
345 | */ | |
346 | #ifdef CONFIG_PPC_BOOK3E | |
347 | copy_with_mmu_off = false; | |
348 | #else | |
349 | copy_with_mmu_off = radix_enabled() || | |
350 | !(firmware_has_feature(FW_FEATURE_LPAR) || | |
351 | firmware_has_feature(FW_FEATURE_PS3_LV1)); | |
352 | #endif | |
fc53b420 | 353 | |
fce0d574 S |
354 | /* Some things are best done in assembly. Finding globals with |
355 | * a toc is easier in C, so pass in what we can. | |
356 | */ | |
357 | kexec_sequence(&kexec_stack, image->start, image, | |
fe036a06 | 358 | page_address(image->control_code_page), |
b970b41e | 359 | mmu_cleanup_all, copy_with_mmu_off); |
fce0d574 S |
360 | /* NOTREACHED */ |
361 | } | |
593e537b | 362 | |
4e003747 | 363 | #ifdef CONFIG_PPC_BOOK3S_64 |
593e537b | 364 | /* Values we need to export to the second kernel via the device tree. */ |
2e8e4f5b | 365 | static unsigned long htab_base; |
ea961a82 | 366 | static unsigned long htab_size; |
593e537b ME |
367 | |
368 | static struct property htab_base_prop = { | |
369 | .name = "linux,htab-base", | |
370 | .length = sizeof(unsigned long), | |
1a38147e | 371 | .value = &htab_base, |
593e537b ME |
372 | }; |
373 | ||
374 | static struct property htab_size_prop = { | |
375 | .name = "linux,htab-size", | |
376 | .length = sizeof(unsigned long), | |
ea961a82 | 377 | .value = &htab_size, |
593e537b ME |
378 | }; |
379 | ||
6f29c329 | 380 | static int __init export_htab_values(void) |
593e537b ME |
381 | { |
382 | struct device_node *node; | |
383 | ||
2e8e4f5b DF |
384 | /* On machines with no htab htab_address is NULL */ |
385 | if (!htab_address) | |
6f29c329 | 386 | return -ENODEV; |
2e8e4f5b | 387 | |
593e537b ME |
388 | node = of_find_node_by_path("/chosen"); |
389 | if (!node) | |
6f29c329 | 390 | return -ENODEV; |
593e537b | 391 | |
ed7b2144 | 392 | /* remove any stale propertys so ours can be found */ |
925e2d1d SJS |
393 | of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL)); |
394 | of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL)); | |
ed7b2144 | 395 | |
ea961a82 | 396 | htab_base = cpu_to_be64(__pa(htab_address)); |
79d1c712 | 397 | of_add_property(node, &htab_base_prop); |
ea961a82 | 398 | htab_size = cpu_to_be64(htab_size_bytes); |
79d1c712 | 399 | of_add_property(node, &htab_size_prop); |
593e537b | 400 | |
593e537b | 401 | of_node_put(node); |
aa98c50d | 402 | return 0; |
35dd5432 | 403 | } |
6f29c329 | 404 | late_initcall(export_htab_values); |
4e003747 | 405 | #endif /* CONFIG_PPC_BOOK3S_64 */ |