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/cpumask.h>
14 #include <linux/kexec.h>
15 #include <linux/smp.h>
16 #include <linux/thread_info.h>
17 #include <linux/errno.h>
20 #include <asm/current.h>
21 #include <asm/machdep.h>
22 #include <asm/cacheflush.h>
25 #include <asm/sections.h> /* _end */
29 int default_machine_kexec_prepare(struct kimage *image)
32 unsigned long begin, end; /* limits of segment */
33 unsigned long low, high; /* limits of blocked memory range */
34 struct device_node *node;
38 if (!ppc_md.hpte_clear_all)
42 * Since we use the kernel fault handlers and paging code to
43 * handle the virtual mode, we must make sure no destination
44 * overlaps kernel static data or bss.
46 for (i = 0; i < image->nr_segments; i++)
47 if (image->segment[i].mem < __pa(_end))
51 * For non-LPAR, we absolutely can not overwrite the mmu hash
52 * table, since we are still using the bolted entries in it to
53 * do the copy. Check that here.
55 * It is safe if the end is below the start of the blocked
56 * region (end <= low), or if the beginning is after the
57 * end of the blocked region (begin >= high). Use the
58 * boolean identity !(a || b) === (!a && !b).
61 low = __pa(htab_address);
62 high = low + htab_size_bytes;
64 for (i = 0; i < image->nr_segments; i++) {
65 begin = image->segment[i].mem;
66 end = begin + image->segment[i].memsz;
68 if ((begin < high) && (end > low))
73 /* We also should not overwrite the tce tables */
74 for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
75 node = of_find_node_by_type(node, "pci")) {
76 basep = (unsigned long *)get_property(node, "linux,tce-base",
78 sizep = (unsigned int *)get_property(node, "linux,tce-size",
80 if (basep == NULL || sizep == NULL)
84 high = low + (*sizep);
86 for (i = 0; i < image->nr_segments; i++) {
87 begin = image->segment[i].mem;
88 end = begin + image->segment[i].memsz;
90 if ((begin < high) && (end > low))
98 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
100 static void copy_segments(unsigned long ind)
108 * We rely on kexec_load to create a lists that properly
109 * initializes these pointers before they are used.
110 * We will still crash if the list is wrong, but at least
111 * the compiler will be quiet.
116 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
117 addr = __va(entry & PAGE_MASK);
119 switch (entry & IND_FLAGS) {
120 case IND_DESTINATION:
123 case IND_INDIRECTION:
127 copy_page(dest, addr);
133 void kexec_copy_flush(struct kimage *image)
135 long i, nr_segments = image->nr_segments;
136 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
138 /* save the ranges on the stack to efficiently flush the icache */
139 memcpy(ranges, image->segment, sizeof(ranges));
142 * After this call we may not use anything allocated in dynamic
143 * memory, including *image.
145 * Only globals and the stack are allowed.
147 copy_segments(image->head);
150 * we need to clear the icache for all dest pages sometime,
151 * including ones that were in place on the original copy
153 for (i = 0; i < nr_segments; i++)
154 flush_icache_range((unsigned long)__va(ranges[i].mem),
155 (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
160 /* FIXME: we should schedule this function to be called on all cpus based
161 * on calling the interrupts, but we would like to call it off irq level
162 * so that the interrupt controller is clean.
164 void kexec_smp_down(void *arg)
166 if (ppc_md.kexec_cpu_down)
167 ppc_md.kexec_cpu_down(0, 1);
174 static void kexec_prepare_cpus(void)
176 int my_cpu, i, notified=-1;
178 smp_call_function(kexec_smp_down, NULL, 0, /* wait */0);
181 /* check the others cpus are now down (via paca hw cpu id == -1) */
182 for (i=0; i < NR_CPUS; i++) {
186 while (paca[i].hw_cpu_id != -1) {
188 if (!cpu_possible(i)) {
189 printk("kexec: cpu %d hw_cpu_id %d is not"
190 " possible, ignoring\n",
191 i, paca[i].hw_cpu_id);
194 if (!cpu_online(i)) {
195 /* Fixme: this can be spinning in
196 * pSeries_secondary_wait with a paca
197 * waiting for it to go online.
199 printk("kexec: cpu %d hw_cpu_id %d is not"
200 " online, ignoring\n",
201 i, paca[i].hw_cpu_id);
205 printk( "kexec: waiting for cpu %d (physical"
207 i, paca[i].hw_cpu_id);
213 /* after we tell the others to go down */
214 if (ppc_md.kexec_cpu_down)
215 ppc_md.kexec_cpu_down(0, 0);
224 static void kexec_prepare_cpus(void)
227 * move the secondarys to us so that we can copy
228 * the new kernel 0-0x100 safely
230 * do this if kexec in setup.c ?
232 * We need to release the cpus if we are ever going from an
233 * UP to an SMP kernel.
236 if (ppc_md.kexec_cpu_down)
237 ppc_md.kexec_cpu_down(0, 0);
244 * kexec thread structure and stack.
246 * We need to make sure that this is 16384-byte aligned due to the
247 * way process stacks are handled. It also must be statically allocated
248 * or allocated as part of the kimage, because everything else may be
249 * overwritten when we copy the kexec image. We piggyback on the
250 * "init_task" linker section here to statically allocate a stack.
252 * We could use a smaller stack if we don't care about anything using
253 * current, but that audit has not been performed.
255 union thread_union kexec_stack
256 __attribute__((__section__(".data.init_task"))) = { };
258 /* Our assembly helper, in kexec_stub.S */
259 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
260 void *image, void *control,
261 void (*clear_all)(void)) ATTRIB_NORET;
263 /* too late to fail here */
264 void default_machine_kexec(struct kimage *image)
266 /* prepare control code if any */
269 * If the kexec boot is the normal one, need to shutdown other cpus
270 * into our wait loop and quiesce interrupts.
271 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
272 * stopping other CPUs and collecting their pt_regs is done before
273 * using debugger IPI.
276 if (crashing_cpu == -1)
277 kexec_prepare_cpus();
279 /* switch to a staticly allocated stack. Based on irq stack code.
280 * XXX: the task struct will likely be invalid once we do the copy!
282 kexec_stack.thread_info.task = current_thread_info()->task;
283 kexec_stack.thread_info.flags = 0;
285 /* Some things are best done in assembly. Finding globals with
286 * a toc is easier in C, so pass in what we can.
288 kexec_sequence(&kexec_stack, image->start, image,
289 page_address(image->control_code_page),
290 ppc_md.hpte_clear_all);
294 /* Values we need to export to the second kernel via the device tree. */
295 static unsigned long htab_base, kernel_end;
297 static struct property htab_base_prop = {
298 .name = "linux,htab-base",
299 .length = sizeof(unsigned long),
300 .value = (unsigned char *)&htab_base,
303 static struct property htab_size_prop = {
304 .name = "linux,htab-size",
305 .length = sizeof(unsigned long),
306 .value = (unsigned char *)&htab_size_bytes,
309 static struct property kernel_end_prop = {
310 .name = "linux,kernel-end",
311 .length = sizeof(unsigned long),
312 .value = (unsigned char *)&kernel_end,
315 static void __init export_htab_values(void)
317 struct device_node *node;
319 node = of_find_node_by_path("/chosen");
323 kernel_end = __pa(_end);
324 prom_add_property(node, &kernel_end_prop);
326 /* On machines with no htab htab_address is NULL */
327 if (NULL == htab_address)
330 htab_base = __pa(htab_address);
331 prom_add_property(node, &htab_base_prop);
332 prom_add_property(node, &htab_size_prop);
338 void __init kexec_setup(void)
340 export_htab_values();