2 * machine_kexec.c - handle transition of 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 #define HASH_GROUP_SIZE 0x80 /* size of each hash group, asm/mmu.h */
31 /* Have this around till we move it into crash specific file */
32 note_buf_t crash_notes[NR_CPUS];
34 /* Dummy for now. Not sure if we need to have a crash shutdown in here
35 * and if what it will achieve. Letting it be now to compile the code
36 * in generic kexec environment
38 void machine_crash_shutdown(struct pt_regs *regs)
40 /* do nothing right now */
41 /* smp_relase_cpus() if we want smp on panic kernel */
42 /* cpu_irq_down to isolate us until we are ready */
45 int machine_kexec_prepare(struct kimage *image)
48 unsigned long begin, end; /* limits of segment */
49 unsigned long low, high; /* limits of blocked memory range */
50 struct device_node *node;
54 if (!ppc_md.hpte_clear_all)
58 * Since we use the kernel fault handlers and paging code to
59 * handle the virtual mode, we must make sure no destination
60 * overlaps kernel static data or bss.
62 for (i = 0; i < image->nr_segments; i++)
63 if (image->segment[i].mem < __pa(_end))
67 * For non-LPAR, we absolutely can not overwrite the mmu hash
68 * table, since we are still using the bolted entries in it to
69 * do the copy. Check that here.
71 * It is safe if the end is below the start of the blocked
72 * region (end <= low), or if the beginning is after the
73 * end of the blocked region (begin >= high). Use the
74 * boolean identity !(a || b) === (!a && !b).
77 low = __pa(htab_address);
78 high = low + (htab_hash_mask + 1) * HASH_GROUP_SIZE;
80 for (i = 0; i < image->nr_segments; i++) {
81 begin = image->segment[i].mem;
82 end = begin + image->segment[i].memsz;
84 if ((begin < high) && (end > low))
89 /* We also should not overwrite the tce tables */
90 for (node = of_find_node_by_type(NULL, "pci"); node != NULL;
91 node = of_find_node_by_type(node, "pci")) {
92 basep = (unsigned long *)get_property(node, "linux,tce-base",
94 sizep = (unsigned int *)get_property(node, "linux,tce-size",
96 if (basep == NULL || sizep == NULL)
100 high = low + (*sizep);
102 for (i = 0; i < image->nr_segments; i++) {
103 begin = image->segment[i].mem;
104 end = begin + image->segment[i].memsz;
106 if ((begin < high) && (end > low))
114 void machine_kexec_cleanup(struct kimage *image)
116 /* we do nothing in prepare that needs to be undone */
119 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
121 static void copy_segments(unsigned long ind)
129 * We rely on kexec_load to create a lists that properly
130 * initializes these pointers before they are used.
131 * We will still crash if the list is wrong, but at least
132 * the compiler will be quiet.
137 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
138 addr = __va(entry & PAGE_MASK);
140 switch (entry & IND_FLAGS) {
141 case IND_DESTINATION:
144 case IND_INDIRECTION:
148 copy_page(dest, addr);
154 void kexec_copy_flush(struct kimage *image)
156 long i, nr_segments = image->nr_segments;
157 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
159 /* save the ranges on the stack to efficiently flush the icache */
160 memcpy(ranges, image->segment, sizeof(ranges));
163 * After this call we may not use anything allocated in dynamic
164 * memory, including *image.
166 * Only globals and the stack are allowed.
168 copy_segments(image->head);
171 * we need to clear the icache for all dest pages sometime,
172 * including ones that were in place on the original copy
174 for (i = 0; i < nr_segments; i++)
175 flush_icache_range(ranges[i].mem + KERNELBASE,
176 ranges[i].mem + KERNELBASE +
182 /* FIXME: we should schedule this function to be called on all cpus based
183 * on calling the interrupts, but we would like to call it off irq level
184 * so that the interrupt controller is clean.
186 void kexec_smp_down(void *arg)
188 if (ppc_md.kexec_cpu_down)
189 ppc_md.kexec_cpu_down(0, 1);
196 static void kexec_prepare_cpus(void)
198 int my_cpu, i, notified=-1;
200 smp_call_function(kexec_smp_down, NULL, 0, /* wait */0);
203 /* check the others cpus are now down (via paca hw cpu id == -1) */
204 for (i=0; i < NR_CPUS; i++) {
208 while (paca[i].hw_cpu_id != -1) {
210 if (!cpu_possible(i)) {
211 printk("kexec: cpu %d hw_cpu_id %d is not"
212 " possible, ignoring\n",
213 i, paca[i].hw_cpu_id);
216 if (!cpu_online(i)) {
217 /* Fixme: this can be spinning in
218 * pSeries_secondary_wait with a paca
219 * waiting for it to go online.
221 printk("kexec: cpu %d hw_cpu_id %d is not"
222 " online, ignoring\n",
223 i, paca[i].hw_cpu_id);
227 printk( "kexec: waiting for cpu %d (physical"
229 i, paca[i].hw_cpu_id);
235 /* after we tell the others to go down */
236 if (ppc_md.kexec_cpu_down)
237 ppc_md.kexec_cpu_down(0, 0);
246 static void kexec_prepare_cpus(void)
249 * move the secondarys to us so that we can copy
250 * the new kernel 0-0x100 safely
252 * do this if kexec in setup.c ?
254 * We need to release the cpus if we are ever going from an
255 * UP to an SMP kernel.
258 if (ppc_md.kexec_cpu_down)
259 ppc_md.kexec_cpu_down(0, 0);
266 * kexec thread structure and stack.
268 * We need to make sure that this is 16384-byte aligned due to the
269 * way process stacks are handled. It also must be statically allocated
270 * or allocated as part of the kimage, because everything else may be
271 * overwritten when we copy the kexec image. We piggyback on the
272 * "init_task" linker section here to statically allocate a stack.
274 * We could use a smaller stack if we don't care about anything using
275 * current, but that audit has not been performed.
277 union thread_union kexec_stack
278 __attribute__((__section__(".data.init_task"))) = { };
280 /* Our assembly helper, in kexec_stub.S */
281 extern NORET_TYPE void kexec_sequence(void *newstack, unsigned long start,
282 void *image, void *control,
283 void (*clear_all)(void)) ATTRIB_NORET;
285 /* too late to fail here */
286 void machine_kexec(struct kimage *image)
289 /* prepare control code if any */
291 /* shutdown other cpus into our wait loop and quiesce interrupts */
292 kexec_prepare_cpus();
294 /* switch to a staticly allocated stack. Based on irq stack code.
295 * XXX: the task struct will likely be invalid once we do the copy!
297 kexec_stack.thread_info.task = current_thread_info()->task;
298 kexec_stack.thread_info.flags = 0;
300 /* Some things are best done in assembly. Finding globals with
301 * a toc is easier in C, so pass in what we can.
303 kexec_sequence(&kexec_stack, image->start, image,
304 page_address(image->control_code_page),
305 ppc_md.hpte_clear_all);
309 /* Values we need to export to the second kernel via the device tree. */
310 static unsigned long htab_base, htab_size, kernel_end;
312 static struct property htab_base_prop = {
313 .name = "linux,htab-base",
314 .length = sizeof(unsigned long),
315 .value = (unsigned char *)&htab_base,
318 static struct property htab_size_prop = {
319 .name = "linux,htab-size",
320 .length = sizeof(unsigned long),
321 .value = (unsigned char *)&htab_size,
324 static struct property kernel_end_prop = {
325 .name = "linux,kernel-end",
326 .length = sizeof(unsigned long),
327 .value = (unsigned char *)&kernel_end,
330 static void __init export_htab_values(void)
332 struct device_node *node;
334 node = of_find_node_by_path("/chosen");
338 kernel_end = __pa(_end);
339 prom_add_property(node, &kernel_end_prop);
341 /* On machines with no htab htab_address is NULL */
342 if (NULL == htab_address)
345 htab_base = __pa(htab_address);
346 prom_add_property(node, &htab_base_prop);
348 htab_size = 1UL << ppc64_pft_size;
349 prom_add_property(node, &htab_size_prop);
355 void __init kexec_setup(void)
357 export_htab_values();