2 * Architecture specific (PPC64) functions for kexec based crash dumps.
4 * Copyright (C) 2005, IBM Corp.
6 * Created by: Haren Myneni
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/kernel.h>
14 #include <linux/smp.h>
15 #include <linux/reboot.h>
16 #include <linux/kexec.h>
17 #include <linux/export.h>
18 #include <linux/crash_dump.h>
19 #include <linux/delay.h>
20 #include <linux/init.h>
21 #include <linux/irq.h>
22 #include <linux/types.h>
24 #include <asm/processor.h>
25 #include <asm/machdep.h>
26 #include <asm/kexec.h>
27 #include <asm/kdump.h>
30 #include <asm/system.h>
31 #include <asm/setjmp.h>
34 * The primary CPU waits a while for all secondary CPUs to enter. This is to
35 * avoid sending an IPI if the secondary CPUs are entering
36 * crash_kexec_secondary on their own (eg via a system reset).
38 * The secondary timeout has to be longer than the primary. Both timeouts are
41 #define PRIMARY_TIMEOUT 500
42 #define SECONDARY_TIMEOUT 1000
44 #define IPI_TIMEOUT 10000
45 #define REAL_MODE_TIMEOUT 10000
47 /* This keeps a track of which one is the crashing cpu. */
48 int crashing_cpu = -1;
49 static cpumask_t cpus_in_crash = CPU_MASK_NONE;
51 #define CRASH_HANDLER_MAX 3
52 /* NULL terminated list of shutdown handles */
53 static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX+1];
54 static DEFINE_SPINLOCK(crash_handlers_lock);
56 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
57 static int crash_shutdown_cpu = -1;
59 static int handle_fault(struct pt_regs *regs)
61 if (crash_shutdown_cpu == smp_processor_id())
62 longjmp(crash_shutdown_buf, 1);
68 void crash_ipi_callback(struct pt_regs *regs)
70 int cpu = smp_processor_id();
76 if (!cpumask_test_cpu(cpu, &cpus_in_crash))
77 crash_save_cpu(regs, cpu);
78 cpumask_set_cpu(cpu, &cpus_in_crash);
81 * Starting the kdump boot.
82 * This barrier is needed to make sure that all CPUs are stopped.
84 while (!cpumask_test_cpu(crashing_cpu, &cpus_in_crash))
87 if (ppc_md.kexec_cpu_down)
88 ppc_md.kexec_cpu_down(1, 1);
99 static void crash_kexec_prepare_cpus(int cpu)
102 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
104 int (*old_handler)(struct pt_regs *regs);
106 printk(KERN_EMERG "Sending IPI to other CPUs\n");
108 crash_send_ipi(crash_ipi_callback);
113 * FIXME: Until we will have the way to stop other CPUs reliably,
114 * the crash CPU will send an IPI and wait for other CPUs to
118 while ((cpumask_weight(&cpus_in_crash) < ncpus) && (--msecs > 0)) {
122 /* Would it be better to replace the trap vector here? */
124 if (cpumask_weight(&cpus_in_crash) >= ncpus) {
125 printk(KERN_EMERG "IPI complete\n");
129 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
130 ncpus - cpumask_weight(&cpus_in_crash));
133 * If we have a panic timeout set then we can't wait indefinitely
134 * for someone to activate system reset. We also give up on the
135 * second time through if system reset fail to work.
137 if ((panic_timeout > 0) || (tries > 0))
141 * A system reset will cause all CPUs to take an 0x100 exception.
142 * The primary CPU returns here via setjmp, and the secondary
143 * CPUs reexecute the crash_kexec_secondary path.
145 old_handler = __debugger;
146 __debugger = handle_fault;
147 crash_shutdown_cpu = smp_processor_id();
149 if (setjmp(crash_shutdown_buf) == 0) {
150 printk(KERN_EMERG "Activate system reset (dumprestart) "
151 "to stop other cpu(s)\n");
154 * A system reset will force all CPUs to execute the
155 * crash code again. We need to reset cpus_in_crash so we
156 * wait for everyone to do this.
158 cpus_in_crash = CPU_MASK_NONE;
161 while (cpumask_weight(&cpus_in_crash) < ncpus)
165 crash_shutdown_cpu = -1;
166 __debugger = old_handler;
173 * This function will be called by secondary cpus.
175 void crash_kexec_secondary(struct pt_regs *regs)
178 int msecs = SECONDARY_TIMEOUT;
180 local_irq_save(flags);
182 /* Wait for the primary crash CPU to signal its progress */
183 while (crashing_cpu < 0) {
185 /* No response, kdump image may not have been loaded */
186 local_irq_restore(flags);
193 crash_ipi_callback(regs);
196 #else /* ! CONFIG_SMP */
198 static void crash_kexec_prepare_cpus(int cpu)
201 * move the secondaries to us so that we can copy
202 * the new kernel 0-0x100 safely
204 * do this if kexec in setup.c ?
213 void crash_kexec_secondary(struct pt_regs *regs)
216 #endif /* CONFIG_SMP */
218 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
219 #if defined(CONFIG_SMP) && defined(CONFIG_PPC_STD_MMU_64)
220 static void crash_kexec_wait_realmode(int cpu)
225 msecs = REAL_MODE_TIMEOUT;
226 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
230 while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
232 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
241 static inline void crash_kexec_wait_realmode(int cpu) {}
242 #endif /* CONFIG_SMP && CONFIG_PPC_STD_MMU_64 */
245 * Register a function to be called on shutdown. Only use this if you
246 * can't reset your device in the second kernel.
248 int crash_shutdown_register(crash_shutdown_t handler)
252 spin_lock(&crash_handlers_lock);
253 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
254 if (!crash_shutdown_handles[i]) {
255 /* Insert handle at first empty entry */
256 crash_shutdown_handles[i] = handler;
261 if (i == CRASH_HANDLER_MAX) {
262 printk(KERN_ERR "Crash shutdown handles full, "
263 "not registered.\n");
267 spin_unlock(&crash_handlers_lock);
270 EXPORT_SYMBOL(crash_shutdown_register);
272 int crash_shutdown_unregister(crash_shutdown_t handler)
276 spin_lock(&crash_handlers_lock);
277 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
278 if (crash_shutdown_handles[i] == handler)
281 if (i == CRASH_HANDLER_MAX) {
282 printk(KERN_ERR "Crash shutdown handle not found\n");
285 /* Shift handles down */
286 for (; crash_shutdown_handles[i]; i++)
287 crash_shutdown_handles[i] =
288 crash_shutdown_handles[i+1];
292 spin_unlock(&crash_handlers_lock);
295 EXPORT_SYMBOL(crash_shutdown_unregister);
297 void default_machine_crash_shutdown(struct pt_regs *regs)
300 int (*old_handler)(struct pt_regs *regs);
303 * This function is only called after the system
304 * has panicked or is otherwise in a critical state.
305 * The minimum amount of code to allow a kexec'd kernel
306 * to run successfully needs to happen here.
308 * In practice this means stopping other cpus in
310 * The kernel is broken so disable interrupts.
315 * Make a note of crashing cpu. Will be used in machine_kexec
316 * such that another IPI will not be sent.
318 crashing_cpu = smp_processor_id();
319 crash_save_cpu(regs, crashing_cpu);
322 * If we came in via system reset, wait a while for the secondary
325 if (TRAP(regs) == 0x100)
326 mdelay(PRIMARY_TIMEOUT);
328 crash_kexec_prepare_cpus(crashing_cpu);
329 cpumask_set_cpu(crashing_cpu, &cpus_in_crash);
330 crash_kexec_wait_realmode(crashing_cpu);
332 machine_kexec_mask_interrupts();
335 * Call registered shutdown routines safely. Swap out
336 * __debugger_fault_handler, and replace on exit.
338 old_handler = __debugger_fault_handler;
339 __debugger_fault_handler = handle_fault;
340 crash_shutdown_cpu = smp_processor_id();
341 for (i = 0; crash_shutdown_handles[i]; i++) {
342 if (setjmp(crash_shutdown_buf) == 0) {
344 * Insert syncs and delay to ensure
345 * instructions in the dangerous region don't
346 * leak away from this protected region.
348 asm volatile("sync; isync");
349 /* dangerous region */
350 crash_shutdown_handles[i]();
351 asm volatile("sync; isync");
354 crash_shutdown_cpu = -1;
355 __debugger_fault_handler = old_handler;
357 if (ppc_md.kexec_cpu_down)
358 ppc_md.kexec_cpu_down(1, 0);