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/irq.h>
21 #include <linux/types.h>
23 #include <asm/processor.h>
24 #include <asm/machdep.h>
25 #include <asm/kexec.h>
28 #include <asm/setjmp.h>
29 #include <asm/debug.h>
32 * The primary CPU waits a while for all secondary CPUs to enter. This is to
33 * avoid sending an IPI if the secondary CPUs are entering
34 * crash_kexec_secondary on their own (eg via a system reset).
36 * The secondary timeout has to be longer than the primary. Both timeouts are
39 #define PRIMARY_TIMEOUT 500
40 #define SECONDARY_TIMEOUT 1000
42 #define IPI_TIMEOUT 10000
43 #define REAL_MODE_TIMEOUT 10000
45 static int time_to_dump;
47 * crash_wake_offline should be set to 1 by platforms that intend to wake
48 * up offline cpus prior to jumping to a kdump kernel. Currently powernv
49 * sets it to 1, since we want to avoid things from happening when an
50 * offline CPU wakes up due to something like an HMI (malfunction error),
51 * which propagates to all threads.
53 int crash_wake_offline;
55 #define CRASH_HANDLER_MAX 3
56 /* List of shutdown handles */
57 static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
58 static DEFINE_SPINLOCK(crash_handlers_lock);
60 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
61 static int crash_shutdown_cpu = -1;
63 static int handle_fault(struct pt_regs *regs)
65 if (crash_shutdown_cpu == smp_processor_id())
66 longjmp(crash_shutdown_buf, 1);
72 static atomic_t cpus_in_crash;
73 void crash_ipi_callback(struct pt_regs *regs)
75 static cpumask_t cpus_state_saved = CPU_MASK_NONE;
77 int cpu = smp_processor_id();
80 if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
81 crash_save_cpu(regs, cpu);
82 cpumask_set_cpu(cpu, &cpus_state_saved);
85 atomic_inc(&cpus_in_crash);
86 smp_mb__after_atomic();
89 * Starting the kdump boot.
90 * This barrier is needed to make sure that all CPUs are stopped.
95 if (ppc_md.kexec_cpu_down)
96 ppc_md.kexec_cpu_down(1, 1);
101 for (;;); /* FIXME */
107 static void crash_kexec_prepare_cpus(int cpu)
110 unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
112 int (*old_handler)(struct pt_regs *regs);
114 printk(KERN_EMERG "Sending IPI to other CPUs\n");
116 if (crash_wake_offline)
117 ncpus = num_present_cpus() - 1;
119 crash_send_ipi(crash_ipi_callback);
124 * FIXME: Until we will have the way to stop other CPUs reliably,
125 * the crash CPU will send an IPI and wait for other CPUs to
129 while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
132 /* Would it be better to replace the trap vector here? */
134 if (atomic_read(&cpus_in_crash) >= ncpus) {
135 printk(KERN_EMERG "IPI complete\n");
139 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
140 ncpus - atomic_read(&cpus_in_crash));
143 * If we have a panic timeout set then we can't wait indefinitely
144 * for someone to activate system reset. We also give up on the
145 * second time through if system reset fail to work.
147 if ((panic_timeout > 0) || (tries > 0))
151 * A system reset will cause all CPUs to take an 0x100 exception.
152 * The primary CPU returns here via setjmp, and the secondary
153 * CPUs reexecute the crash_kexec_secondary path.
155 old_handler = __debugger;
156 __debugger = handle_fault;
157 crash_shutdown_cpu = smp_processor_id();
159 if (setjmp(crash_shutdown_buf) == 0) {
160 printk(KERN_EMERG "Activate system reset (dumprestart) "
161 "to stop other cpu(s)\n");
164 * A system reset will force all CPUs to execute the
165 * crash code again. We need to reset cpus_in_crash so we
166 * wait for everyone to do this.
168 atomic_set(&cpus_in_crash, 0);
171 while (atomic_read(&cpus_in_crash) < ncpus)
175 crash_shutdown_cpu = -1;
176 __debugger = old_handler;
183 * This function will be called by secondary cpus.
185 void crash_kexec_secondary(struct pt_regs *regs)
188 int msecs = SECONDARY_TIMEOUT;
190 local_irq_save(flags);
192 /* Wait for the primary crash CPU to signal its progress */
193 while (crashing_cpu < 0) {
195 /* No response, kdump image may not have been loaded */
196 local_irq_restore(flags);
203 crash_ipi_callback(regs);
206 #else /* ! CONFIG_SMP */
208 static void crash_kexec_prepare_cpus(int cpu)
211 * move the secondaries to us so that we can copy
212 * the new kernel 0-0x100 safely
214 * do this if kexec in setup.c ?
223 void crash_kexec_secondary(struct pt_regs *regs)
226 #endif /* CONFIG_SMP */
228 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
229 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
230 static void __maybe_unused crash_kexec_wait_realmode(int cpu)
235 msecs = REAL_MODE_TIMEOUT;
236 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
240 while (paca_ptrs[i]->kexec_state < KEXEC_STATE_REAL_MODE) {
242 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
251 static inline void crash_kexec_wait_realmode(int cpu) {}
252 #endif /* CONFIG_SMP && CONFIG_PPC64 */
255 * Register a function to be called on shutdown. Only use this if you
256 * can't reset your device in the second kernel.
258 int crash_shutdown_register(crash_shutdown_t handler)
262 spin_lock(&crash_handlers_lock);
263 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
264 if (!crash_shutdown_handles[i]) {
265 /* Insert handle at first empty entry */
266 crash_shutdown_handles[i] = handler;
271 if (i == CRASH_HANDLER_MAX) {
272 printk(KERN_ERR "Crash shutdown handles full, "
273 "not registered.\n");
277 spin_unlock(&crash_handlers_lock);
280 EXPORT_SYMBOL(crash_shutdown_register);
282 int crash_shutdown_unregister(crash_shutdown_t handler)
286 spin_lock(&crash_handlers_lock);
287 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
288 if (crash_shutdown_handles[i] == handler)
291 if (i == CRASH_HANDLER_MAX) {
292 printk(KERN_ERR "Crash shutdown handle not found\n");
295 /* Shift handles down */
296 for (; i < (CRASH_HANDLER_MAX - 1); i++)
297 crash_shutdown_handles[i] =
298 crash_shutdown_handles[i+1];
300 * Reset last entry to NULL now that it has been shifted down,
301 * this will allow new handles to be added here.
303 crash_shutdown_handles[i] = NULL;
307 spin_unlock(&crash_handlers_lock);
310 EXPORT_SYMBOL(crash_shutdown_unregister);
312 void default_machine_crash_shutdown(struct pt_regs *regs)
315 int (*old_handler)(struct pt_regs *regs);
318 * This function is only called after the system
319 * has panicked or is otherwise in a critical state.
320 * The minimum amount of code to allow a kexec'd kernel
321 * to run successfully needs to happen here.
323 * In practice this means stopping other cpus in
325 * The kernel is broken so disable interrupts.
330 * Make a note of crashing cpu. Will be used in machine_kexec
331 * such that another IPI will not be sent.
333 crashing_cpu = smp_processor_id();
336 * If we came in via system reset, wait a while for the secondary
339 if (TRAP(regs) == 0x100)
340 mdelay(PRIMARY_TIMEOUT);
342 crash_kexec_prepare_cpus(crashing_cpu);
344 crash_save_cpu(regs, crashing_cpu);
348 crash_kexec_wait_realmode(crashing_cpu);
350 machine_kexec_mask_interrupts();
353 * Call registered shutdown routines safely. Swap out
354 * __debugger_fault_handler, and replace on exit.
356 old_handler = __debugger_fault_handler;
357 __debugger_fault_handler = handle_fault;
358 crash_shutdown_cpu = smp_processor_id();
359 for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
360 if (setjmp(crash_shutdown_buf) == 0) {
362 * Insert syncs and delay to ensure
363 * instructions in the dangerous region don't
364 * leak away from this protected region.
366 asm volatile("sync; isync");
367 /* dangerous region */
368 crash_shutdown_handles[i]();
369 asm volatile("sync; isync");
372 crash_shutdown_cpu = -1;
373 __debugger_fault_handler = old_handler;
375 if (ppc_md.kexec_cpu_down)
376 ppc_md.kexec_cpu_down(1, 0);