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/errno.h>
19 #include <asm/current.h>
20 #include <asm/machdep.h>
21 #include <asm/cacheflush.h>
24 #include <asm/sections.h> /* _end */
28 int default_machine_kexec_prepare(struct kimage *image)
31 unsigned long begin, end; /* limits of segment */
32 unsigned long low, high; /* limits of blocked memory range */
33 struct device_node *node;
34 const unsigned long *basep;
35 const unsigned int *sizep;
37 if (!ppc_md.hpte_clear_all)
41 * Since we use the kernel fault handlers and paging code to
42 * handle the virtual mode, we must make sure no destination
43 * overlaps kernel static data or bss.
45 for (i = 0; i < image->nr_segments; i++)
46 if (image->segment[i].mem < __pa(_end))
50 * For non-LPAR, we absolutely can not overwrite the mmu hash
51 * table, since we are still using the bolted entries in it to
52 * do the copy. Check that here.
54 * It is safe if the end is below the start of the blocked
55 * region (end <= low), or if the beginning is after the
56 * end of the blocked region (begin >= high). Use the
57 * boolean identity !(a || b) === (!a && !b).
60 low = __pa(htab_address);
61 high = low + htab_size_bytes;
63 for (i = 0; i < image->nr_segments; i++) {
64 begin = image->segment[i].mem;
65 end = begin + image->segment[i].memsz;
67 if ((begin < high) && (end > low))
72 /* We also should not overwrite the tce tables */
73 for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
74 node = of_find_node_by_type(node, "pci")) {
75 basep = of_get_property(node, "linux,tce-base", NULL);
76 sizep = of_get_property(node, "linux,tce-size", NULL);
77 if (basep == NULL || sizep == NULL)
81 high = low + (*sizep);
83 for (i = 0; i < image->nr_segments; i++) {
84 begin = image->segment[i].mem;
85 end = begin + image->segment[i].memsz;
87 if ((begin < high) && (end > low))
95 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
97 static void copy_segments(unsigned long ind)
105 * We rely on kexec_load to create a lists that properly
106 * initializes these pointers before they are used.
107 * We will still crash if the list is wrong, but at least
108 * the compiler will be quiet.
113 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
114 addr = __va(entry & PAGE_MASK);
116 switch (entry & IND_FLAGS) {
117 case IND_DESTINATION:
120 case IND_INDIRECTION:
124 copy_page(dest, addr);
130 void kexec_copy_flush(struct kimage *image)
132 long i, nr_segments = image->nr_segments;
133 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
135 /* save the ranges on the stack to efficiently flush the icache */
136 memcpy(ranges, image->segment, sizeof(ranges));
139 * After this call we may not use anything allocated in dynamic
140 * memory, including *image.
142 * Only globals and the stack are allowed.
144 copy_segments(image->head);
147 * we need to clear the icache for all dest pages sometime,
148 * including ones that were in place on the original copy
150 for (i = 0; i < nr_segments; i++)
151 flush_icache_range((unsigned long)__va(ranges[i].mem),
152 (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
157 /* FIXME: we should schedule this function to be called on all cpus based
158 * on calling the interrupts, but we would like to call it off irq level
159 * so that the interrupt controller is clean.
161 void kexec_smp_down(void *arg)
163 if (ppc_md.kexec_cpu_down)
164 ppc_md.kexec_cpu_down(0, 1);
171 static void kexec_prepare_cpus(void)
173 int my_cpu, i, notified=-1;
175 smp_call_function(kexec_smp_down, NULL, 0, /* wait */0);
178 /* check the others cpus are now down (via paca hw cpu id == -1) */
179 for (i=0; i < NR_CPUS; i++) {
183 while (paca[i].hw_cpu_id != -1) {
185 if (!cpu_possible(i)) {
186 printk("kexec: cpu %d hw_cpu_id %d is not"
187 " possible, ignoring\n",
188 i, paca[i].hw_cpu_id);
191 if (!cpu_online(i)) {
192 /* Fixme: this can be spinning in
193 * pSeries_secondary_wait with a paca
194 * waiting for it to go online.
196 printk("kexec: cpu %d hw_cpu_id %d is not"
197 " online, ignoring\n",
198 i, paca[i].hw_cpu_id);
202 printk( "kexec: waiting for cpu %d (physical"
204 i, paca[i].hw_cpu_id);
210 /* after we tell the others to go down */
211 if (ppc_md.kexec_cpu_down)
212 ppc_md.kexec_cpu_down(0, 0);
221 static void kexec_prepare_cpus(void)
224 * move the secondarys to us so that we can copy
225 * the new kernel 0-0x100 safely
227 * do this if kexec in setup.c ?
229 * We need to release the cpus if we are ever going from an
230 * UP to an SMP kernel.
233 if (ppc_md.kexec_cpu_down)
234 ppc_md.kexec_cpu_down(0, 0);
241 * kexec thread structure and stack.
243 * We need to make sure that this is 16384-byte aligned due to the
244 * way process stacks are handled. It also must be statically allocated
245 * or allocated as part of the kimage, because everything else may be
246 * overwritten when we copy the kexec image. We piggyback on the
247 * "init_task" linker section here to statically allocate a stack.
249 * We could use a smaller stack if we don't care about anything using
250 * current, but that audit has not been performed.
252 union thread_union kexec_stack
253 __attribute__((__section__(".data.init_task"))) = { };
255 /* Our assembly helper, in kexec_stub.S */
256 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
257 void *image, void *control,
258 void (*clear_all)(void)) ATTRIB_NORET;
260 /* too late to fail here */
261 void default_machine_kexec(struct kimage *image)
263 /* prepare control code if any */
266 * If the kexec boot is the normal one, need to shutdown other cpus
267 * into our wait loop and quiesce interrupts.
268 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
269 * stopping other CPUs and collecting their pt_regs is done before
270 * using debugger IPI.
273 if (crashing_cpu == -1)
274 kexec_prepare_cpus();
276 /* switch to a staticly allocated stack. Based on irq stack code.
277 * XXX: the task struct will likely be invalid once we do the copy!
279 kexec_stack.thread_info.task = current_thread_info()->task;
280 kexec_stack.thread_info.flags = 0;
282 /* Some things are best done in assembly. Finding globals with
283 * a toc is easier in C, so pass in what we can.
285 kexec_sequence(&kexec_stack, image->start, image,
286 page_address(image->control_code_page),
287 ppc_md.hpte_clear_all);
291 /* Values we need to export to the second kernel via the device tree. */
292 static unsigned long htab_base, kernel_end;
294 static struct property htab_base_prop = {
295 .name = "linux,htab-base",
296 .length = sizeof(unsigned long),
300 static struct property htab_size_prop = {
301 .name = "linux,htab-size",
302 .length = sizeof(unsigned long),
303 .value = &htab_size_bytes,
306 static struct property kernel_end_prop = {
307 .name = "linux,kernel-end",
308 .length = sizeof(unsigned long),
309 .value = &kernel_end,
312 static void __init export_htab_values(void)
314 struct device_node *node;
316 node = of_find_node_by_path("/chosen");
320 kernel_end = __pa(_end);
321 prom_add_property(node, &kernel_end_prop);
323 /* On machines with no htab htab_address is NULL */
324 if (NULL == htab_address)
327 htab_base = __pa(htab_address);
328 prom_add_property(node, &htab_base_prop);
329 prom_add_property(node, &htab_size_prop);
335 static struct property crashk_base_prop = {
336 .name = "linux,crashkernel-base",
337 .length = sizeof(unsigned long),
338 .value = &crashk_res.start,
341 static unsigned long crashk_size;
343 static struct property crashk_size_prop = {
344 .name = "linux,crashkernel-size",
345 .length = sizeof(unsigned long),
346 .value = &crashk_size,
349 static void __init export_crashk_values(void)
351 struct device_node *node;
352 struct property *prop;
354 node = of_find_node_by_path("/chosen");
358 /* There might be existing crash kernel properties, but we can't
359 * be sure what's in them, so remove them. */
360 prop = of_find_property(node, "linux,crashkernel-base", NULL);
362 prom_remove_property(node, prop);
364 prop = of_find_property(node, "linux,crashkernel-size", NULL);
366 prom_remove_property(node, prop);
368 if (crashk_res.start != 0) {
369 prom_add_property(node, &crashk_base_prop);
370 crashk_size = crashk_res.end - crashk_res.start + 1;
371 prom_add_property(node, &crashk_size_prop);
377 static int __init kexec_setup(void)
379 export_htab_values();
380 export_crashk_values();
383 __initcall(kexec_setup);