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>
33 int default_machine_kexec_prepare(struct kimage *image)
36 unsigned long begin, end; /* limits of segment */
37 unsigned long low, high; /* limits of blocked memory range */
38 struct device_node *node;
39 const unsigned long *basep;
40 const unsigned int *sizep;
42 if (!ppc_md.hpte_clear_all)
46 * Since we use the kernel fault handlers and paging code to
47 * handle the virtual mode, we must make sure no destination
48 * overlaps kernel static data or bss.
50 for (i = 0; i < image->nr_segments; i++)
51 if (image->segment[i].mem < __pa(_end))
55 * For non-LPAR, we absolutely can not overwrite the mmu hash
56 * table, since we are still using the bolted entries in it to
57 * do the copy. Check that here.
59 * It is safe if the end is below the start of the blocked
60 * region (end <= low), or if the beginning is after the
61 * end of the blocked region (begin >= high). Use the
62 * boolean identity !(a || b) === (!a && !b).
65 low = __pa(htab_address);
66 high = low + htab_size_bytes;
68 for (i = 0; i < image->nr_segments; i++) {
69 begin = image->segment[i].mem;
70 end = begin + image->segment[i].memsz;
72 if ((begin < high) && (end > low))
77 /* We also should not overwrite the tce tables */
78 for_each_node_by_type(node, "pci") {
79 basep = of_get_property(node, "linux,tce-base", NULL);
80 sizep = of_get_property(node, "linux,tce-size", NULL);
81 if (basep == NULL || sizep == NULL)
85 high = low + (*sizep);
87 for (i = 0; i < image->nr_segments; i++) {
88 begin = image->segment[i].mem;
89 end = begin + image->segment[i].memsz;
91 if ((begin < high) && (end > low))
99 #define IND_FLAGS (IND_DESTINATION | IND_INDIRECTION | IND_DONE | IND_SOURCE)
101 static void copy_segments(unsigned long ind)
109 * We rely on kexec_load to create a lists that properly
110 * initializes these pointers before they are used.
111 * We will still crash if the list is wrong, but at least
112 * the compiler will be quiet.
117 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
118 addr = __va(entry & PAGE_MASK);
120 switch (entry & IND_FLAGS) {
121 case IND_DESTINATION:
124 case IND_INDIRECTION:
128 copy_page(dest, addr);
134 void kexec_copy_flush(struct kimage *image)
136 long i, nr_segments = image->nr_segments;
137 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
139 /* save the ranges on the stack to efficiently flush the icache */
140 memcpy(ranges, image->segment, sizeof(ranges));
143 * After this call we may not use anything allocated in dynamic
144 * memory, including *image.
146 * Only globals and the stack are allowed.
148 copy_segments(image->head);
151 * we need to clear the icache for all dest pages sometime,
152 * including ones that were in place on the original copy
154 for (i = 0; i < nr_segments; i++)
155 flush_icache_range((unsigned long)__va(ranges[i].mem),
156 (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
161 static int kexec_all_irq_disabled = 0;
163 static void kexec_smp_down(void *arg)
168 mb(); /* make sure our irqs are disabled before we say they are */
169 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
170 while(kexec_all_irq_disabled == 0)
172 mb(); /* make sure all irqs are disabled before this */
173 hw_breakpoint_disable();
175 * Now every CPU has IRQs off, we can clear out any pending
176 * IPIs and be sure that no more will come in after this.
178 if (ppc_md.kexec_cpu_down)
179 ppc_md.kexec_cpu_down(0, 1);
185 static void kexec_prepare_cpus_wait(int wait_state)
187 int my_cpu, i, notified=-1;
189 hw_breakpoint_disable();
191 /* Make sure each CPU has at least made it to the state we need.
193 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
194 * are correctly onlined. If somehow we start a CPU on boot with RTAS
195 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
196 * time, the boot CPU will timeout. If it does eventually execute
197 * stuff, the secondary will start up (paca[].cpu_start was written) and
198 * get into a peculiar state. If the platform supports
199 * smp_ops->take_timebase(), the secondary CPU will probably be spinning
200 * in there. If not (i.e. pseries), the secondary will continue on and
201 * try to online itself/idle/etc. If it survives that, we need to find
202 * these possible-but-not-online-but-should-be CPUs and chaperone them
203 * into kexec_smp_wait().
205 for_each_online_cpu(i) {
209 while (paca[i].kexec_state < wait_state) {
212 printk(KERN_INFO "kexec: waiting for cpu %d "
213 "(physical %d) to enter %i state\n",
214 i, paca[i].hw_cpu_id, wait_state);
223 * We need to make sure each present CPU is online. The next kernel will scan
224 * the device tree and assume primary threads are online and query secondary
225 * threads via RTAS to online them if required. If we don't online primary
226 * threads, they will be stuck. However, we also online secondary threads as we
227 * may be using 'cede offline'. In this case RTAS doesn't see the secondary
228 * threads as offline -- and again, these CPUs will be stuck.
230 * So, we online all CPUs that should be running, including secondary threads.
232 static void wake_offline_cpus(void)
236 for_each_present_cpu(cpu) {
237 if (!cpu_online(cpu)) {
238 printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
245 static void kexec_prepare_cpus(void)
248 smp_call_function(kexec_smp_down, NULL, /* wait */0);
252 mb(); /* make sure IRQs are disabled before we say they are */
253 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
255 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
256 /* we are sure every CPU has IRQs off at this point */
257 kexec_all_irq_disabled = 1;
259 /* after we tell the others to go down */
260 if (ppc_md.kexec_cpu_down)
261 ppc_md.kexec_cpu_down(0, 0);
264 * Before removing MMU mappings make sure all CPUs have entered real
267 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
274 static void kexec_prepare_cpus(void)
277 * move the secondarys to us so that we can copy
278 * the new kernel 0-0x100 safely
280 * do this if kexec in setup.c ?
282 * We need to release the cpus if we are ever going from an
283 * UP to an SMP kernel.
286 if (ppc_md.kexec_cpu_down)
287 ppc_md.kexec_cpu_down(0, 0);
295 * kexec thread structure and stack.
297 * We need to make sure that this is 16384-byte aligned due to the
298 * way process stacks are handled. It also must be statically allocated
299 * or allocated as part of the kimage, because everything else may be
300 * overwritten when we copy the kexec image. We piggyback on the
301 * "init_task" linker section here to statically allocate a stack.
303 * We could use a smaller stack if we don't care about anything using
304 * current, but that audit has not been performed.
306 static union thread_union kexec_stack __init_task_data =
310 * For similar reasons to the stack above, the kexecing CPU needs to be on a
311 * static PACA; we switch to kexec_paca.
313 struct paca_struct kexec_paca;
315 /* Our assembly helper, in misc_64.S */
316 extern void kexec_sequence(void *newstack, unsigned long start,
317 void *image, void *control,
318 void (*clear_all)(void)) __noreturn;
320 /* too late to fail here */
321 void default_machine_kexec(struct kimage *image)
323 /* prepare control code if any */
326 * If the kexec boot is the normal one, need to shutdown other cpus
327 * into our wait loop and quiesce interrupts.
328 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
329 * stopping other CPUs and collecting their pt_regs is done before
330 * using debugger IPI.
333 if (crashing_cpu == -1)
334 kexec_prepare_cpus();
336 pr_debug("kexec: Starting switchover sequence.\n");
338 /* switch to a staticly allocated stack. Based on irq stack code.
339 * We setup preempt_count to avoid using VMX in memcpy.
340 * XXX: the task struct will likely be invalid once we do the copy!
342 kexec_stack.thread_info.task = current_thread_info()->task;
343 kexec_stack.thread_info.flags = 0;
344 kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET;
345 kexec_stack.thread_info.cpu = current_thread_info()->cpu;
347 /* We need a static PACA, too; copy this CPU's PACA over and switch to
348 * it. Also poison per_cpu_offset to catch anyone using non-static
351 memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
352 kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
353 paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
354 kexec_paca.paca_index;
355 setup_paca(&kexec_paca);
357 /* XXX: If anyone does 'dynamic lppacas' this will also need to be
358 * switched to a static version!
361 /* Some things are best done in assembly. Finding globals with
362 * a toc is easier in C, so pass in what we can.
364 kexec_sequence(&kexec_stack, image->start, image,
365 page_address(image->control_code_page),
366 ppc_md.hpte_clear_all);
370 /* Values we need to export to the second kernel via the device tree. */
371 static unsigned long htab_base;
372 static unsigned long htab_size;
374 static struct property htab_base_prop = {
375 .name = "linux,htab-base",
376 .length = sizeof(unsigned long),
380 static struct property htab_size_prop = {
381 .name = "linux,htab-size",
382 .length = sizeof(unsigned long),
386 static int __init export_htab_values(void)
388 struct device_node *node;
389 struct property *prop;
391 /* On machines with no htab htab_address is NULL */
395 node = of_find_node_by_path("/chosen");
399 /* remove any stale propertys so ours can be found */
400 prop = of_find_property(node, htab_base_prop.name, NULL);
402 of_remove_property(node, prop);
403 prop = of_find_property(node, htab_size_prop.name, NULL);
405 of_remove_property(node, prop);
407 htab_base = cpu_to_be64(__pa(htab_address));
408 of_add_property(node, &htab_base_prop);
409 htab_size = cpu_to_be64(htab_size_bytes);
410 of_add_property(node, &htab_size_prop);
415 late_initcall(export_htab_values);