2 * PPC64 code to handle Linux booting another kernel.
4 * Copyright (C) 2004-2005, IBM Corp.
6 * Created by: Milton D Miller II
8 * This source code is licensed under the GNU General Public License,
9 * Version 2. See the file COPYING for more details.
13 #include <linux/kexec.h>
14 #include <linux/smp.h>
15 #include <linux/thread_info.h>
16 #include <linux/init_task.h>
17 #include <linux/errno.h>
18 #include <linux/kernel.h>
19 #include <linux/cpu.h>
20 #include <linux/hardirq.h>
23 #include <asm/current.h>
24 #include <asm/machdep.h>
25 #include <asm/cacheflush.h>
28 #include <asm/sections.h> /* _end */
31 #include <asm/hw_breakpoint.h>
32 #include <asm/asm-prototypes.h>
34 #ifdef CONFIG_PPC_BOOK3E
35 int default_machine_kexec_prepare(struct kimage *image)
39 * Since we use the kernel fault handlers and paging code to
40 * handle the virtual mode, we must make sure no destination
41 * overlaps kernel static data or bss.
43 for (i = 0; i < image->nr_segments; i++)
44 if (image->segment[i].mem < __pa(_end))
49 int default_machine_kexec_prepare(struct kimage *image)
52 unsigned long begin, end; /* limits of segment */
53 unsigned long low, high; /* limits of blocked memory range */
54 struct device_node *node;
55 const unsigned long *basep;
56 const unsigned int *sizep;
58 if (!mmu_hash_ops.hpte_clear_all)
62 * Since we use the kernel fault handlers and paging code to
63 * handle the virtual mode, we must make sure no destination
64 * overlaps kernel static data or bss.
66 for (i = 0; i < image->nr_segments; i++)
67 if (image->segment[i].mem < __pa(_end))
71 * For non-LPAR, we absolutely can not overwrite the mmu hash
72 * table, since we are still using the bolted entries in it to
73 * do the copy. Check that here.
75 * It is safe if the end is below the start of the blocked
76 * region (end <= low), or if the beginning is after the
77 * end of the blocked region (begin >= high). Use the
78 * boolean identity !(a || b) === (!a && !b).
80 #ifdef CONFIG_PPC_STD_MMU_64
82 low = __pa(htab_address);
83 high = low + htab_size_bytes;
85 for (i = 0; i < image->nr_segments; i++) {
86 begin = image->segment[i].mem;
87 end = begin + image->segment[i].memsz;
89 if ((begin < high) && (end > low))
93 #endif /* CONFIG_PPC_STD_MMU_64 */
95 /* We also should not overwrite the tce tables */
96 for_each_node_by_type(node, "pci") {
97 basep = of_get_property(node, "linux,tce-base", NULL);
98 sizep = of_get_property(node, "linux,tce-size", NULL);
99 if (basep == NULL || sizep == NULL)
103 high = low + (*sizep);
105 for (i = 0; i < image->nr_segments; i++) {
106 begin = image->segment[i].mem;
107 end = begin + image->segment[i].memsz;
109 if ((begin < high) && (end > low))
116 #endif /* !CONFIG_PPC_BOOK3E */
118 static void copy_segments(unsigned long ind)
126 * We rely on kexec_load to create a lists that properly
127 * initializes these pointers before they are used.
128 * We will still crash if the list is wrong, but at least
129 * the compiler will be quiet.
134 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
135 addr = __va(entry & PAGE_MASK);
137 switch (entry & IND_FLAGS) {
138 case IND_DESTINATION:
141 case IND_INDIRECTION:
145 copy_page(dest, addr);
151 void kexec_copy_flush(struct kimage *image)
153 long i, nr_segments = image->nr_segments;
154 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
156 /* save the ranges on the stack to efficiently flush the icache */
157 memcpy(ranges, image->segment, sizeof(ranges));
160 * After this call we may not use anything allocated in dynamic
161 * memory, including *image.
163 * Only globals and the stack are allowed.
165 copy_segments(image->head);
168 * we need to clear the icache for all dest pages sometime,
169 * including ones that were in place on the original copy
171 for (i = 0; i < nr_segments; i++)
172 flush_icache_range((unsigned long)__va(ranges[i].mem),
173 (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
178 static int kexec_all_irq_disabled = 0;
180 static void kexec_smp_down(void *arg)
185 mb(); /* make sure our irqs are disabled before we say they are */
186 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
187 while(kexec_all_irq_disabled == 0)
189 mb(); /* make sure all irqs are disabled before this */
190 hw_breakpoint_disable();
192 * Now every CPU has IRQs off, we can clear out any pending
193 * IPIs and be sure that no more will come in after this.
195 if (ppc_md.kexec_cpu_down)
196 ppc_md.kexec_cpu_down(0, 1);
202 static void kexec_prepare_cpus_wait(int wait_state)
204 int my_cpu, i, notified=-1;
206 hw_breakpoint_disable();
208 /* Make sure each CPU has at least made it to the state we need.
210 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
211 * are correctly onlined. If somehow we start a CPU on boot with RTAS
212 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
213 * time, the boot CPU will timeout. If it does eventually execute
214 * stuff, the secondary will start up (paca[].cpu_start was written) and
215 * get into a peculiar state. If the platform supports
216 * smp_ops->take_timebase(), the secondary CPU will probably be spinning
217 * in there. If not (i.e. pseries), the secondary will continue on and
218 * try to online itself/idle/etc. If it survives that, we need to find
219 * these possible-but-not-online-but-should-be CPUs and chaperone them
220 * into kexec_smp_wait().
222 for_each_online_cpu(i) {
226 while (paca[i].kexec_state < wait_state) {
229 printk(KERN_INFO "kexec: waiting for cpu %d "
230 "(physical %d) to enter %i state\n",
231 i, paca[i].hw_cpu_id, wait_state);
240 * We need to make sure each present CPU is online. The next kernel will scan
241 * the device tree and assume primary threads are online and query secondary
242 * threads via RTAS to online them if required. If we don't online primary
243 * threads, they will be stuck. However, we also online secondary threads as we
244 * may be using 'cede offline'. In this case RTAS doesn't see the secondary
245 * threads as offline -- and again, these CPUs will be stuck.
247 * So, we online all CPUs that should be running, including secondary threads.
249 static void wake_offline_cpus(void)
253 for_each_present_cpu(cpu) {
254 if (!cpu_online(cpu)) {
255 printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
257 WARN_ON(cpu_up(cpu));
262 static void kexec_prepare_cpus(void)
265 smp_call_function(kexec_smp_down, NULL, /* wait */0);
269 mb(); /* make sure IRQs are disabled before we say they are */
270 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
272 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
273 /* we are sure every CPU has IRQs off at this point */
274 kexec_all_irq_disabled = 1;
276 /* after we tell the others to go down */
277 if (ppc_md.kexec_cpu_down)
278 ppc_md.kexec_cpu_down(0, 0);
281 * Before removing MMU mappings make sure all CPUs have entered real
284 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
291 static void kexec_prepare_cpus(void)
294 * move the secondarys to us so that we can copy
295 * the new kernel 0-0x100 safely
297 * do this if kexec in setup.c ?
299 * We need to release the cpus if we are ever going from an
300 * UP to an SMP kernel.
303 if (ppc_md.kexec_cpu_down)
304 ppc_md.kexec_cpu_down(0, 0);
312 * kexec thread structure and stack.
314 * We need to make sure that this is 16384-byte aligned due to the
315 * way process stacks are handled. It also must be statically allocated
316 * or allocated as part of the kimage, because everything else may be
317 * overwritten when we copy the kexec image. We piggyback on the
318 * "init_task" linker section here to statically allocate a stack.
320 * We could use a smaller stack if we don't care about anything using
321 * current, but that audit has not been performed.
323 static union thread_union kexec_stack __init_task_data =
327 * For similar reasons to the stack above, the kexecing CPU needs to be on a
328 * static PACA; we switch to kexec_paca.
330 struct paca_struct kexec_paca;
332 /* Our assembly helper, in misc_64.S */
333 extern void kexec_sequence(void *newstack, unsigned long start,
334 void *image, void *control,
335 void (*clear_all)(void)) __noreturn;
337 /* too late to fail here */
338 void default_machine_kexec(struct kimage *image)
340 /* prepare control code if any */
343 * If the kexec boot is the normal one, need to shutdown other cpus
344 * into our wait loop and quiesce interrupts.
345 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
346 * stopping other CPUs and collecting their pt_regs is done before
347 * using debugger IPI.
350 if (!kdump_in_progress())
351 kexec_prepare_cpus();
353 pr_debug("kexec: Starting switchover sequence.\n");
355 /* switch to a staticly allocated stack. Based on irq stack code.
356 * We setup preempt_count to avoid using VMX in memcpy.
357 * XXX: the task struct will likely be invalid once we do the copy!
359 kexec_stack.thread_info.task = current_thread_info()->task;
360 kexec_stack.thread_info.flags = 0;
361 kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET;
362 kexec_stack.thread_info.cpu = current_thread_info()->cpu;
364 /* We need a static PACA, too; copy this CPU's PACA over and switch to
365 * it. Also poison per_cpu_offset to catch anyone using non-static
368 memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
369 kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
370 paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
371 kexec_paca.paca_index;
372 setup_paca(&kexec_paca);
374 /* XXX: If anyone does 'dynamic lppacas' this will also need to be
375 * switched to a static version!
378 /* Some things are best done in assembly. Finding globals with
379 * a toc is easier in C, so pass in what we can.
381 kexec_sequence(&kexec_stack, image->start, image,
382 page_address(image->control_code_page),
383 #ifdef CONFIG_PPC_STD_MMU
384 mmu_hash_ops.hpte_clear_all
392 #ifdef CONFIG_PPC_STD_MMU_64
393 /* Values we need to export to the second kernel via the device tree. */
394 static unsigned long htab_base;
395 static unsigned long htab_size;
397 static struct property htab_base_prop = {
398 .name = "linux,htab-base",
399 .length = sizeof(unsigned long),
403 static struct property htab_size_prop = {
404 .name = "linux,htab-size",
405 .length = sizeof(unsigned long),
409 static int __init export_htab_values(void)
411 struct device_node *node;
413 /* On machines with no htab htab_address is NULL */
417 node = of_find_node_by_path("/chosen");
421 /* remove any stale propertys so ours can be found */
422 of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL));
423 of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL));
425 htab_base = cpu_to_be64(__pa(htab_address));
426 of_add_property(node, &htab_base_prop);
427 htab_size = cpu_to_be64(htab_size_bytes);
428 of_add_property(node, &htab_size_prop);
433 late_initcall(export_htab_values);
434 #endif /* CONFIG_PPC_STD_MMU_64 */