1 // SPDX-License-Identifier: GPL-2.0-only
3 * Architecture specific (PPC64) functions for kexec based crash dumps.
5 * Copyright (C) 2005, IBM Corp.
7 * Created by: Haren Myneni
10 #include <linux/kernel.h>
11 #include <linux/smp.h>
12 #include <linux/reboot.h>
13 #include <linux/kexec.h>
14 #include <linux/export.h>
15 #include <linux/crash_dump.h>
16 #include <linux/delay.h>
17 #include <linux/irq.h>
18 #include <linux/types.h>
20 #include <asm/processor.h>
21 #include <asm/machdep.h>
22 #include <asm/kexec.h>
25 #include <asm/setjmp.h>
26 #include <asm/debug.h>
27 #include <asm/interrupt.h>
30 * The primary CPU waits a while for all secondary CPUs to enter. This is to
31 * avoid sending an IPI if the secondary CPUs are entering
32 * crash_kexec_secondary on their own (eg via a system reset).
34 * The secondary timeout has to be longer than the primary. Both timeouts are
37 #define PRIMARY_TIMEOUT 500
38 #define SECONDARY_TIMEOUT 1000
40 #define IPI_TIMEOUT 10000
41 #define REAL_MODE_TIMEOUT 10000
43 static int time_to_dump;
45 * crash_wake_offline should be set to 1 by platforms that intend to wake
46 * up offline cpus prior to jumping to a kdump kernel. Currently powernv
47 * sets it to 1, since we want to avoid things from happening when an
48 * offline CPU wakes up due to something like an HMI (malfunction error),
49 * which propagates to all threads.
51 int crash_wake_offline;
53 #define CRASH_HANDLER_MAX 3
54 /* List of shutdown handles */
55 static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
56 static DEFINE_SPINLOCK(crash_handlers_lock);
58 static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
59 static int crash_shutdown_cpu = -1;
61 static int handle_fault(struct pt_regs *regs)
63 if (crash_shutdown_cpu == smp_processor_id())
64 longjmp(crash_shutdown_buf, 1);
70 static atomic_t cpus_in_crash;
71 void crash_ipi_callback(struct pt_regs *regs)
73 static cpumask_t cpus_state_saved = CPU_MASK_NONE;
75 int cpu = smp_processor_id();
78 if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
79 crash_save_cpu(regs, cpu);
80 cpumask_set_cpu(cpu, &cpus_state_saved);
83 atomic_inc(&cpus_in_crash);
84 smp_mb__after_atomic();
87 * Starting the kdump boot.
88 * This barrier is needed to make sure that all CPUs are stopped.
93 if (ppc_md.kexec_cpu_down)
94 ppc_md.kexec_cpu_down(1, 1);
105 static void crash_kexec_prepare_cpus(int cpu)
108 volatile unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
109 volatile int tries = 0;
110 int (*old_handler)(struct pt_regs *regs);
112 printk(KERN_EMERG "Sending IPI to other CPUs\n");
114 if (crash_wake_offline)
115 ncpus = num_present_cpus() - 1;
117 crash_send_ipi(crash_ipi_callback);
122 * FIXME: Until we will have the way to stop other CPUs reliably,
123 * the crash CPU will send an IPI and wait for other CPUs to
127 while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
130 /* Would it be better to replace the trap vector here? */
132 if (atomic_read(&cpus_in_crash) >= ncpus) {
133 printk(KERN_EMERG "IPI complete\n");
137 printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
138 ncpus - atomic_read(&cpus_in_crash));
141 * If we have a panic timeout set then we can't wait indefinitely
142 * for someone to activate system reset. We also give up on the
143 * second time through if system reset fail to work.
145 if ((panic_timeout > 0) || (tries > 0))
149 * A system reset will cause all CPUs to take an 0x100 exception.
150 * The primary CPU returns here via setjmp, and the secondary
151 * CPUs reexecute the crash_kexec_secondary path.
153 old_handler = __debugger;
154 __debugger = handle_fault;
155 crash_shutdown_cpu = smp_processor_id();
157 if (setjmp(crash_shutdown_buf) == 0) {
158 printk(KERN_EMERG "Activate system reset (dumprestart) "
159 "to stop other cpu(s)\n");
162 * A system reset will force all CPUs to execute the
163 * crash code again. We need to reset cpus_in_crash so we
164 * wait for everyone to do this.
166 atomic_set(&cpus_in_crash, 0);
169 while (atomic_read(&cpus_in_crash) < ncpus)
173 crash_shutdown_cpu = -1;
174 __debugger = old_handler;
181 * This function will be called by secondary cpus.
183 void crash_kexec_secondary(struct pt_regs *regs)
186 int msecs = SECONDARY_TIMEOUT;
188 local_irq_save(flags);
190 /* Wait for the primary crash CPU to signal its progress */
191 while (crashing_cpu < 0) {
193 /* No response, kdump image may not have been loaded */
194 local_irq_restore(flags);
201 crash_ipi_callback(regs);
204 #else /* ! CONFIG_SMP */
206 static void crash_kexec_prepare_cpus(int cpu)
209 * move the secondaries to us so that we can copy
210 * the new kernel 0-0x100 safely
212 * do this if kexec in setup.c ?
221 void crash_kexec_secondary(struct pt_regs *regs)
224 #endif /* CONFIG_SMP */
226 /* wait for all the CPUs to hit real mode but timeout if they don't come in */
227 #if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
228 static void __maybe_unused crash_kexec_wait_realmode(int cpu)
233 msecs = REAL_MODE_TIMEOUT;
234 for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
238 while (paca_ptrs[i]->kexec_state < KEXEC_STATE_REAL_MODE) {
240 if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
249 static inline void crash_kexec_wait_realmode(int cpu) {}
250 #endif /* CONFIG_SMP && CONFIG_PPC64 */
253 * Register a function to be called on shutdown. Only use this if you
254 * can't reset your device in the second kernel.
256 int crash_shutdown_register(crash_shutdown_t handler)
260 spin_lock(&crash_handlers_lock);
261 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
262 if (!crash_shutdown_handles[i]) {
263 /* Insert handle at first empty entry */
264 crash_shutdown_handles[i] = handler;
269 if (i == CRASH_HANDLER_MAX) {
270 printk(KERN_ERR "Crash shutdown handles full, "
271 "not registered.\n");
275 spin_unlock(&crash_handlers_lock);
278 EXPORT_SYMBOL(crash_shutdown_register);
280 int crash_shutdown_unregister(crash_shutdown_t handler)
284 spin_lock(&crash_handlers_lock);
285 for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
286 if (crash_shutdown_handles[i] == handler)
289 if (i == CRASH_HANDLER_MAX) {
290 printk(KERN_ERR "Crash shutdown handle not found\n");
293 /* Shift handles down */
294 for (; i < (CRASH_HANDLER_MAX - 1); i++)
295 crash_shutdown_handles[i] =
296 crash_shutdown_handles[i+1];
298 * Reset last entry to NULL now that it has been shifted down,
299 * this will allow new handles to be added here.
301 crash_shutdown_handles[i] = NULL;
305 spin_unlock(&crash_handlers_lock);
308 EXPORT_SYMBOL(crash_shutdown_unregister);
310 void default_machine_crash_shutdown(struct pt_regs *regs)
313 int (*old_handler)(struct pt_regs *regs);
315 /* Avoid hardlocking with irresponsive CPU holding logbuf_lock */
316 printk_deferred_enter();
319 * This function is only called after the system
320 * has panicked or is otherwise in a critical state.
321 * The minimum amount of code to allow a kexec'd kernel
322 * to run successfully needs to happen here.
324 * In practice this means stopping other cpus in
326 * The kernel is broken so disable interrupts.
331 * Make a note of crashing cpu. Will be used in machine_kexec
332 * such that another IPI will not be sent.
334 crashing_cpu = smp_processor_id();
337 * If we came in via system reset, wait a while for the secondary
340 if (TRAP(regs) == INTERRUPT_SYSTEM_RESET)
341 mdelay(PRIMARY_TIMEOUT);
343 crash_kexec_prepare_cpus(crashing_cpu);
345 crash_save_cpu(regs, crashing_cpu);
349 crash_kexec_wait_realmode(crashing_cpu);
351 machine_kexec_mask_interrupts();
354 * Call registered shutdown routines safely. Swap out
355 * __debugger_fault_handler, and replace on exit.
357 old_handler = __debugger_fault_handler;
358 __debugger_fault_handler = handle_fault;
359 crash_shutdown_cpu = smp_processor_id();
360 for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
361 if (setjmp(crash_shutdown_buf) == 0) {
363 * Insert syncs and delay to ensure
364 * instructions in the dangerous region don't
365 * leak away from this protected region.
367 asm volatile("sync; isync");
368 /* dangerous region */
369 crash_shutdown_handles[i]();
370 asm volatile("sync; isync");
373 crash_shutdown_cpu = -1;
374 __debugger_fault_handler = old_handler;
376 if (ppc_md.kexec_cpu_down)
377 ppc_md.kexec_cpu_down(1, 0);