1 // SPDX-License-Identifier: GPL-2.0-only
3 * PPC64 code to handle Linux booting another kernel.
5 * Copyright (C) 2004-2005, IBM Corp.
7 * Created by: Milton D Miller II
11 #include <linux/kexec.h>
12 #include <linux/smp.h>
13 #include <linux/thread_info.h>
14 #include <linux/init_task.h>
15 #include <linux/errno.h>
16 #include <linux/kernel.h>
17 #include <linux/cpu.h>
18 #include <linux/hardirq.h>
21 #include <asm/current.h>
22 #include <asm/machdep.h>
23 #include <asm/cacheflush.h>
24 #include <asm/firmware.h>
27 #include <asm/sections.h> /* _end */
30 #include <asm/hw_breakpoint.h>
31 #include <asm/asm-prototypes.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;
43 * Since we use the kernel fault handlers and paging code to
44 * handle the virtual mode, we must make sure no destination
45 * overlaps kernel static data or bss.
47 for (i = 0; i < image->nr_segments; i++)
48 if (image->segment[i].mem < __pa(_end))
51 /* We also should not overwrite the tce tables */
52 for_each_node_by_type(node, "pci") {
53 basep = of_get_property(node, "linux,tce-base", NULL);
54 sizep = of_get_property(node, "linux,tce-size", NULL);
55 if (basep == NULL || sizep == NULL)
59 high = low + (*sizep);
61 for (i = 0; i < image->nr_segments; i++) {
62 begin = image->segment[i].mem;
63 end = begin + image->segment[i].memsz;
65 if ((begin < high) && (end > low))
73 static void copy_segments(unsigned long ind)
81 * We rely on kexec_load to create a lists that properly
82 * initializes these pointers before they are used.
83 * We will still crash if the list is wrong, but at least
84 * the compiler will be quiet.
89 for (entry = ind; !(entry & IND_DONE); entry = *ptr++) {
90 addr = __va(entry & PAGE_MASK);
92 switch (entry & IND_FLAGS) {
100 copy_page(dest, addr);
106 void kexec_copy_flush(struct kimage *image)
108 long i, nr_segments = image->nr_segments;
109 struct kexec_segment ranges[KEXEC_SEGMENT_MAX];
111 /* save the ranges on the stack to efficiently flush the icache */
112 memcpy(ranges, image->segment, sizeof(ranges));
115 * After this call we may not use anything allocated in dynamic
116 * memory, including *image.
118 * Only globals and the stack are allowed.
120 copy_segments(image->head);
123 * we need to clear the icache for all dest pages sometime,
124 * including ones that were in place on the original copy
126 for (i = 0; i < nr_segments; i++)
127 flush_icache_range((unsigned long)__va(ranges[i].mem),
128 (unsigned long)__va(ranges[i].mem + ranges[i].memsz));
133 static int kexec_all_irq_disabled = 0;
135 static void kexec_smp_down(void *arg)
140 mb(); /* make sure our irqs are disabled before we say they are */
141 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
142 while(kexec_all_irq_disabled == 0)
144 mb(); /* make sure all irqs are disabled before this */
145 hw_breakpoint_disable();
147 * Now every CPU has IRQs off, we can clear out any pending
148 * IPIs and be sure that no more will come in after this.
150 if (ppc_md.kexec_cpu_down)
151 ppc_md.kexec_cpu_down(0, 1);
157 static void kexec_prepare_cpus_wait(int wait_state)
159 int my_cpu, i, notified=-1;
161 hw_breakpoint_disable();
163 /* Make sure each CPU has at least made it to the state we need.
165 * FIXME: There is a (slim) chance of a problem if not all of the CPUs
166 * are correctly onlined. If somehow we start a CPU on boot with RTAS
167 * start-cpu, but somehow that CPU doesn't write callin_cpu_map[] in
168 * time, the boot CPU will timeout. If it does eventually execute
169 * stuff, the secondary will start up (paca_ptrs[]->cpu_start was
170 * written) and get into a peculiar state.
171 * If the platform supports smp_ops->take_timebase(), the secondary CPU
172 * will probably be spinning in there. If not (i.e. pseries), the
173 * secondary will continue on and try to online itself/idle/etc. If it
174 * survives that, we need to find these
175 * possible-but-not-online-but-should-be CPUs and chaperone them into
178 for_each_online_cpu(i) {
182 while (paca_ptrs[i]->kexec_state < wait_state) {
185 printk(KERN_INFO "kexec: waiting for cpu %d "
186 "(physical %d) to enter %i state\n",
187 i, paca_ptrs[i]->hw_cpu_id, wait_state);
196 * We need to make sure each present CPU is online. The next kernel will scan
197 * the device tree and assume primary threads are online and query secondary
198 * threads via RTAS to online them if required. If we don't online primary
199 * threads, they will be stuck. However, we also online secondary threads as we
200 * may be using 'cede offline'. In this case RTAS doesn't see the secondary
201 * threads as offline -- and again, these CPUs will be stuck.
203 * So, we online all CPUs that should be running, including secondary threads.
205 static void wake_offline_cpus(void)
209 for_each_present_cpu(cpu) {
210 if (!cpu_online(cpu)) {
211 printk(KERN_INFO "kexec: Waking offline cpu %d.\n",
213 WARN_ON(cpu_up(cpu));
218 static void kexec_prepare_cpus(void)
221 smp_call_function(kexec_smp_down, NULL, /* wait */0);
225 mb(); /* make sure IRQs are disabled before we say they are */
226 get_paca()->kexec_state = KEXEC_STATE_IRQS_OFF;
228 kexec_prepare_cpus_wait(KEXEC_STATE_IRQS_OFF);
229 /* we are sure every CPU has IRQs off at this point */
230 kexec_all_irq_disabled = 1;
233 * Before removing MMU mappings make sure all CPUs have entered real
236 kexec_prepare_cpus_wait(KEXEC_STATE_REAL_MODE);
238 /* after we tell the others to go down */
239 if (ppc_md.kexec_cpu_down)
240 ppc_md.kexec_cpu_down(0, 0);
247 static void kexec_prepare_cpus(void)
250 * move the secondarys to us so that we can copy
251 * the new kernel 0-0x100 safely
253 * do this if kexec in setup.c ?
255 * We need to release the cpus if we are ever going from an
256 * UP to an SMP kernel.
259 if (ppc_md.kexec_cpu_down)
260 ppc_md.kexec_cpu_down(0, 0);
268 * kexec thread structure and stack.
270 * We need to make sure that this is 16384-byte aligned due to the
271 * way process stacks are handled. It also must be statically allocated
272 * or allocated as part of the kimage, because everything else may be
273 * overwritten when we copy the kexec image. We piggyback on the
274 * "init_task" linker section here to statically allocate a stack.
276 * We could use a smaller stack if we don't care about anything using
277 * current, but that audit has not been performed.
279 static union thread_union kexec_stack __init_task_data =
283 * For similar reasons to the stack above, the kexecing CPU needs to be on a
284 * static PACA; we switch to kexec_paca.
286 struct paca_struct kexec_paca;
288 /* Our assembly helper, in misc_64.S */
289 extern void kexec_sequence(void *newstack, unsigned long start,
290 void *image, void *control,
291 void (*clear_all)(void),
292 bool copy_with_mmu_off) __noreturn;
294 /* too late to fail here */
295 void default_machine_kexec(struct kimage *image)
297 bool copy_with_mmu_off;
299 /* prepare control code if any */
302 * If the kexec boot is the normal one, need to shutdown other cpus
303 * into our wait loop and quiesce interrupts.
304 * Otherwise, in the case of crashed mode (crashing_cpu >= 0),
305 * stopping other CPUs and collecting their pt_regs is done before
306 * using debugger IPI.
309 if (!kdump_in_progress())
310 kexec_prepare_cpus();
312 printk("kexec: Starting switchover sequence.\n");
314 /* switch to a staticly allocated stack. Based on irq stack code.
315 * We setup preempt_count to avoid using VMX in memcpy.
316 * XXX: the task struct will likely be invalid once we do the copy!
318 current_thread_info()->flags = 0;
319 current_thread_info()->preempt_count = HARDIRQ_OFFSET;
321 /* We need a static PACA, too; copy this CPU's PACA over and switch to
322 * it. Also poison per_cpu_offset and NULL lppaca to catch anyone using
325 memcpy(&kexec_paca, get_paca(), sizeof(struct paca_struct));
326 kexec_paca.data_offset = 0xedeaddeadeeeeeeeUL;
327 #ifdef CONFIG_PPC_PSERIES
328 kexec_paca.lppaca_ptr = NULL;
330 paca_ptrs[kexec_paca.paca_index] = &kexec_paca;
332 setup_paca(&kexec_paca);
335 * The lppaca should be unregistered at this point so the HV won't
336 * touch it. In the case of a crash, none of the lppacas are
337 * unregistered so there is not much we can do about it here.
341 * On Book3S, the copy must happen with the MMU off if we are either
342 * using Radix page tables or we are not in an LPAR since we can
343 * overwrite the page tables while copying.
345 * In an LPAR, we keep the MMU on otherwise we can't access beyond
346 * the RMA. On BookE there is no real MMU off mode, so we have to
347 * keep it enabled as well (but then we have bolted TLB entries).
349 #ifdef CONFIG_PPC_BOOK3E
350 copy_with_mmu_off = false;
352 copy_with_mmu_off = radix_enabled() ||
353 !(firmware_has_feature(FW_FEATURE_LPAR) ||
354 firmware_has_feature(FW_FEATURE_PS3_LV1));
357 /* Some things are best done in assembly. Finding globals with
358 * a toc is easier in C, so pass in what we can.
360 kexec_sequence(&kexec_stack, image->start, image,
361 page_address(image->control_code_page),
362 mmu_cleanup_all, copy_with_mmu_off);
366 #ifdef CONFIG_PPC_BOOK3S_64
367 /* Values we need to export to the second kernel via the device tree. */
368 static unsigned long htab_base;
369 static unsigned long htab_size;
371 static struct property htab_base_prop = {
372 .name = "linux,htab-base",
373 .length = sizeof(unsigned long),
377 static struct property htab_size_prop = {
378 .name = "linux,htab-size",
379 .length = sizeof(unsigned long),
383 static int __init export_htab_values(void)
385 struct device_node *node;
387 /* On machines with no htab htab_address is NULL */
391 node = of_find_node_by_path("/chosen");
395 /* remove any stale propertys so ours can be found */
396 of_remove_property(node, of_find_property(node, htab_base_prop.name, NULL));
397 of_remove_property(node, of_find_property(node, htab_size_prop.name, NULL));
399 htab_base = cpu_to_be64(__pa(htab_address));
400 of_add_property(node, &htab_base_prop);
401 htab_size = cpu_to_be64(htab_size_bytes);
402 of_add_property(node, &htab_size_prop);
407 late_initcall(export_htab_values);
408 #endif /* CONFIG_PPC_BOOK3S_64 */