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>
26 #include <asm/firmware.h>
29 #include <asm/sections.h> /* _end */
32 #include <asm/hw_breakpoint.h>
33 #include <asm/asm-prototypes.h>
35 #ifdef CONFIG_PPC_BOOK3E
36 int default_machine_kexec_prepare(struct kimage *image)
40 * Since we use the kernel fault handlers and paging code to
41 * handle the virtual mode, we must make sure no destination
42 * overlaps kernel static data or bss.
44 for (i = 0; i < image->nr_segments; i++)
45 if (image->segment[i].mem < __pa(_end))
50 int default_machine_kexec_prepare(struct kimage *image)
53 unsigned long begin, end; /* limits of segment */
54 unsigned long low, high; /* limits of blocked memory range */
55 struct device_node *node;
56 const unsigned long *basep;
57 const unsigned int *sizep;
60 * Since we use the kernel fault handlers and paging code to
61 * handle the virtual mode, we must make sure no destination
62 * overlaps kernel static data or bss.
64 for (i = 0; i < image->nr_segments; i++)
65 if (image->segment[i].mem < __pa(_end))
68 /* We also should not overwrite the tce tables */
69 for_each_node_by_type(node, "pci") {
70 basep = of_get_property(node, "linux,tce-base", NULL);
71 sizep = of_get_property(node, "linux,tce-size", NULL);
72 if (basep == NULL || sizep == NULL)
76 high = low + (*sizep);
78 for (i = 0; i < image->nr_segments; i++) {
79 begin = image->segment[i].mem;
80 end = begin + image->segment[i].memsz;
82 if ((begin < high) && (end > low))
89 #endif /* !CONFIG_PPC_BOOK3E */
91 static void copy_segments(unsigned long ind)
99 * We rely on kexec_load to create a lists that properly
100 * initializes these pointers before they are used.
101 * We will still crash if the list is wrong, but at least
102 * the compiler will be quiet.
107 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
108 addr = __va(entry & PAGE_MASK);
110 switch (entry & IND_FLAGS) {
111 case IND_DESTINATION:
114 case IND_INDIRECTION:
118 copy_page(dest, addr);
124 void kexec_copy_flush(struct kimage *image)
126 long i, nr_segments = image->nr_segments;
127 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
129 /* save the ranges on the stack to efficiently flush the icache */
130 memcpy(ranges, image->segment, sizeof(ranges));
133 * After this call we may not use anything allocated in dynamic
134 * memory, including *image.
136 * Only globals and the stack are allowed.
138 copy_segments(image->head);
141 * we need to clear the icache for all dest pages sometime,
142 * including ones that were in place on the original copy
144 for (i = 0; i < nr_segments; i++)
145 flush_icache_range((unsigned long)__va(ranges[i].mem),
146 (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
151 static int kexec_all_irq_disabled = 0;
153 static void kexec_smp_down(void *arg)
158 mb(); /* make sure our irqs are disabled before we say they are */
159 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
160 while(kexec_all_irq_disabled == 0)
162 mb(); /* make sure all irqs are disabled before this */
163 hw_breakpoint_disable();
165 * Now every CPU has IRQs off, we can clear out any pending
166 * IPIs and be sure that no more will come in after this.
168 if (ppc_md.kexec_cpu_down)
169 ppc_md.kexec_cpu_down(0, 1);
175 static void kexec_prepare_cpus_wait(int wait_state)
177 int my_cpu, i, notified=-1;
179 hw_breakpoint_disable();
181 /* Make sure each CPU has at least made it to the state we need.
183 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
184 * are correctly onlined. If somehow we start a CPU on boot with RTAS
185 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
186 * time, the boot CPU will timeout. If it does eventually execute
187 * stuff, the secondary will start up (paca[].cpu_start was written) and
188 * get into a peculiar state. If the platform supports
189 * smp_ops->take_timebase(), the secondary CPU will probably be spinning
190 * in there. If not (i.e. pseries), the secondary will continue on and
191 * try to online itself/idle/etc. If it survives that, we need to find
192 * these possible-but-not-online-but-should-be CPUs and chaperone them
193 * into kexec_smp_wait().
195 for_each_online_cpu(i) {
199 while (paca[i].kexec_state < wait_state) {
202 printk(KERN_INFO "kexec: waiting for cpu %d "
203 "(physical %d) to enter %i state\n",
204 i, paca[i].hw_cpu_id, wait_state);
213 * We need to make sure each present CPU is online. The next kernel will scan
214 * the device tree and assume primary threads are online and query secondary
215 * threads via RTAS to online them if required. If we don't online primary
216 * threads, they will be stuck. However, we also online secondary threads as we
217 * may be using 'cede offline'. In this case RTAS doesn't see the secondary
218 * threads as offline -- and again, these CPUs will be stuck.
220 * So, we online all CPUs that should be running, including secondary threads.
222 static void wake_offline_cpus(void)
226 for_each_present_cpu(cpu) {
227 if (!cpu_online(cpu)) {
228 printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
230 WARN_ON(cpu_up(cpu));
235 static void kexec_prepare_cpus(void)
238 smp_call_function(kexec_smp_down, NULL, /* wait */0);
242 mb(); /* make sure IRQs are disabled before we say they are */
243 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
245 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
246 /* we are sure every CPU has IRQs off at this point */
247 kexec_all_irq_disabled = 1;
249 /* after we tell the others to go down */
250 if (ppc_md.kexec_cpu_down)
251 ppc_md.kexec_cpu_down(0, 0);
254 * Before removing MMU mappings make sure all CPUs have entered real
257 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
264 static void kexec_prepare_cpus(void)
267 * move the secondarys to us so that we can copy
268 * the new kernel 0-0x100 safely
270 * do this if kexec in setup.c ?
272 * We need to release the cpus if we are ever going from an
273 * UP to an SMP kernel.
276 if (ppc_md.kexec_cpu_down)
277 ppc_md.kexec_cpu_down(0, 0);
285 * kexec thread structure and stack.
287 * We need to make sure that this is 16384-byte aligned due to the
288 * way process stacks are handled. It also must be statically allocated
289 * or allocated as part of the kimage, because everything else may be
290 * overwritten when we copy the kexec image. We piggyback on the
291 * "init_task" linker section here to statically allocate a stack.
293 * We could use a smaller stack if we don't care about anything using
294 * current, but that audit has not been performed.
296 static union thread_union kexec_stack __init_task_data =
300 * For similar reasons to the stack above, the kexecing CPU needs to be on a
301 * static PACA; we switch to kexec_paca.
303 struct paca_struct kexec_paca;
305 /* Our assembly helper, in misc_64.S */
306 extern void kexec_sequence(void *newstack, unsigned long start,
307 void *image, void *control,
308 void (*clear_all)(void),
309 bool copy_with_mmu_off) __noreturn;
311 /* too late to fail here */
312 void default_machine_kexec(struct kimage *image)
314 bool copy_with_mmu_off;
316 /* prepare control code if any */
319 * If the kexec boot is the normal one, need to shutdown other cpus
320 * into our wait loop and quiesce interrupts.
321 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
322 * stopping other CPUs and collecting their pt_regs is done before
323 * using debugger IPI.
326 if (!kdump_in_progress())
327 kexec_prepare_cpus();
329 pr_debug("kexec: Starting switchover sequence.\n");
331 /* switch to a staticly allocated stack. Based on irq stack code.
332 * We setup preempt_count to avoid using VMX in memcpy.
333 * XXX: the task struct will likely be invalid once we do the copy!
335 kexec_stack.thread_info.task = current_thread_info()->task;
336 kexec_stack.thread_info.flags = 0;
337 kexec_stack.thread_info.preempt_count = HARDIRQ_OFFSET;
338 kexec_stack.thread_info.cpu = current_thread_info()->cpu;
340 /* We need a static PACA, too; copy this CPU's PACA over and switch to
341 * it. Also poison per_cpu_offset to catch anyone using non-static
344 memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
345 kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
346 paca = (struct paca_struct *)RELOC_HIDE(&kexec_paca, 0) -
347 kexec_paca.paca_index;
348 setup_paca(&kexec_paca);
350 /* XXX: If anyone does 'dynamic lppacas' this will also need to be
351 * switched to a static version!
354 * On Book3S, the copy must happen with the MMU off if we are either
355 * using Radix page tables or we are not in an LPAR since we can
356 * overwrite the page tables while copying.
358 * In an LPAR, we keep the MMU on otherwise we can't access beyond
359 * the RMA. On BookE there is no real MMU off mode, so we have to
360 * keep it enabled as well (but then we have bolted TLB entries).
362 #ifdef CONFIG_PPC_BOOK3E
363 copy_with_mmu_off = false;
365 copy_with_mmu_off = radix_enabled() ||
366 !(firmware_has_feature(FW_FEATURE_LPAR) ||
367 firmware_has_feature(FW_FEATURE_PS3_LV1));
370 /* Some things are best done in assembly. Finding globals with
371 * a toc is easier in C, so pass in what we can.
373 kexec_sequence(&kexec_stack, image->start, image,
374 page_address(image->control_code_page),
375 mmu_cleanup_all, copy_with_mmu_off);
379 #ifdef CONFIG_PPC_STD_MMU_64
380 /* Values we need to export to the second kernel via the device tree. */
381 static unsigned long htab_base;
382 static unsigned long htab_size;
384 static struct property htab_base_prop = {
385 .name = "linux,htab-base",
386 .length = sizeof(unsigned long),
390 static struct property htab_size_prop = {
391 .name = "linux,htab-size",
392 .length = sizeof(unsigned long),
396 static int __init export_htab_values(void)
398 struct device_node *node;
400 /* On machines with no htab htab_address is NULL */
404 node = of_find_node_by_path("/chosen");
408 /* remove any stale propertys so ours can be found */
409 of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL));
410 of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL));
412 htab_base = cpu_to_be64(__pa(htab_address));
413 of_add_property(node, &htab_base_prop);
414 htab_size = cpu_to_be64(htab_size_bytes);
415 of_add_property(node, &htab_size_prop);
420 late_initcall(export_htab_values);
421 #endif /* CONFIG_PPC_STD_MMU_64 */