PageRenderTime 64ms CodeModel.GetById 20ms app.highlight 38ms RepoModel.GetById 2ms app.codeStats 0ms

/arch/powerpc/kernel/crash.c

http://github.com/mirrors/linux
C | 372 lines | 212 code | 62 blank | 98 comment | 44 complexity | 213ce2627e79844964bb77b8cac78463 MD5 | raw file
  1/*
  2 * Architecture specific (PPC64) functions for kexec based crash dumps.
  3 *
  4 * Copyright (C) 2005, IBM Corp.
  5 *
  6 * Created by: Haren Myneni
  7 *
  8 * This source code is licensed under the GNU General Public License,
  9 * Version 2.  See the file COPYING for more details.
 10 *
 11 */
 12
 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>
 22
 23#include <asm/processor.h>
 24#include <asm/machdep.h>
 25#include <asm/kexec.h>
 26#include <asm/kdump.h>
 27#include <asm/prom.h>
 28#include <asm/smp.h>
 29#include <asm/setjmp.h>
 30#include <asm/debug.h>
 31
 32/*
 33 * The primary CPU waits a while for all secondary CPUs to enter. This is to
 34 * avoid sending an IPI if the secondary CPUs are entering
 35 * crash_kexec_secondary on their own (eg via a system reset).
 36 *
 37 * The secondary timeout has to be longer than the primary. Both timeouts are
 38 * in milliseconds.
 39 */
 40#define PRIMARY_TIMEOUT		500
 41#define SECONDARY_TIMEOUT	1000
 42
 43#define IPI_TIMEOUT		10000
 44#define REAL_MODE_TIMEOUT	10000
 45
 46/* This keeps a track of which one is the crashing cpu. */
 47int crashing_cpu = -1;
 48static int time_to_dump;
 49
 50#define CRASH_HANDLER_MAX 3
 51/* List of shutdown handles */
 52static crash_shutdown_t crash_shutdown_handles[CRASH_HANDLER_MAX];
 53static DEFINE_SPINLOCK(crash_handlers_lock);
 54
 55static unsigned long crash_shutdown_buf[JMP_BUF_LEN];
 56static int crash_shutdown_cpu = -1;
 57
 58static int handle_fault(struct pt_regs *regs)
 59{
 60	if (crash_shutdown_cpu == smp_processor_id())
 61		longjmp(crash_shutdown_buf, 1);
 62	return 0;
 63}
 64
 65#ifdef CONFIG_SMP
 66
 67static atomic_t cpus_in_crash;
 68static void crash_ipi_callback(struct pt_regs *regs)
 69{
 70	static cpumask_t cpus_state_saved = CPU_MASK_NONE;
 71
 72	int cpu = smp_processor_id();
 73
 74	if (!cpu_online(cpu))
 75		return;
 76
 77	hard_irq_disable();
 78	if (!cpumask_test_cpu(cpu, &cpus_state_saved)) {
 79		crash_save_cpu(regs, cpu);
 80		cpumask_set_cpu(cpu, &cpus_state_saved);
 81	}
 82
 83	atomic_inc(&cpus_in_crash);
 84	smp_mb__after_atomic();
 85
 86	/*
 87	 * Starting the kdump boot.
 88	 * This barrier is needed to make sure that all CPUs are stopped.
 89	 */
 90	while (!time_to_dump)
 91		cpu_relax();
 92
 93	if (ppc_md.kexec_cpu_down)
 94		ppc_md.kexec_cpu_down(1, 1);
 95
 96#ifdef CONFIG_PPC64
 97	kexec_smp_wait();
 98#else
 99	for (;;);	/* FIXME */
100#endif
101
102	/* NOTREACHED */
103}
104
105static void crash_kexec_prepare_cpus(int cpu)
106{
107	unsigned int msecs;
108	unsigned int ncpus = num_online_cpus() - 1;/* Excluding the panic cpu */
109	int tries = 0;
110	int (*old_handler)(struct pt_regs *regs);
111
112	printk(KERN_EMERG "Sending IPI to other CPUs\n");
113
114	crash_send_ipi(crash_ipi_callback);
115	smp_wmb();
116
117again:
118	/*
119	 * FIXME: Until we will have the way to stop other CPUs reliably,
120	 * the crash CPU will send an IPI and wait for other CPUs to
121	 * respond.
122	 */
123	msecs = IPI_TIMEOUT;
124	while ((atomic_read(&cpus_in_crash) < ncpus) && (--msecs > 0))
125		mdelay(1);
126
127	/* Would it be better to replace the trap vector here? */
128
129	if (atomic_read(&cpus_in_crash) >= ncpus) {
130		printk(KERN_EMERG "IPI complete\n");
131		return;
132	}
133
134	printk(KERN_EMERG "ERROR: %d cpu(s) not responding\n",
135		ncpus - atomic_read(&cpus_in_crash));
136
137	/*
138	 * If we have a panic timeout set then we can't wait indefinitely
139	 * for someone to activate system reset. We also give up on the
140	 * second time through if system reset fail to work.
141	 */
142	if ((panic_timeout > 0) || (tries > 0))
143		return;
144
145	/*
146	 * A system reset will cause all CPUs to take an 0x100 exception.
147	 * The primary CPU returns here via setjmp, and the secondary
148	 * CPUs reexecute the crash_kexec_secondary path.
149	 */
150	old_handler = __debugger;
151	__debugger = handle_fault;
152	crash_shutdown_cpu = smp_processor_id();
153
154	if (setjmp(crash_shutdown_buf) == 0) {
155		printk(KERN_EMERG "Activate system reset (dumprestart) "
156				  "to stop other cpu(s)\n");
157
158		/*
159		 * A system reset will force all CPUs to execute the
160		 * crash code again. We need to reset cpus_in_crash so we
161		 * wait for everyone to do this.
162		 */
163		atomic_set(&cpus_in_crash, 0);
164		smp_mb();
165
166		while (atomic_read(&cpus_in_crash) < ncpus)
167			cpu_relax();
168	}
169
170	crash_shutdown_cpu = -1;
171	__debugger = old_handler;
172
173	tries++;
174	goto again;
175}
176
177/*
178 * This function will be called by secondary cpus.
179 */
180void crash_kexec_secondary(struct pt_regs *regs)
181{
182	unsigned long flags;
183	int msecs = SECONDARY_TIMEOUT;
184
185	local_irq_save(flags);
186
187	/* Wait for the primary crash CPU to signal its progress */
188	while (crashing_cpu < 0) {
189		if (--msecs < 0) {
190			/* No response, kdump image may not have been loaded */
191			local_irq_restore(flags);
192			return;
193		}
194
195		mdelay(1);
196	}
197
198	crash_ipi_callback(regs);
199}
200
201#else	/* ! CONFIG_SMP */
202
203static void crash_kexec_prepare_cpus(int cpu)
204{
205	/*
206	 * move the secondaries to us so that we can copy
207	 * the new kernel 0-0x100 safely
208	 *
209	 * do this if kexec in setup.c ?
210	 */
211#ifdef CONFIG_PPC64
212	smp_release_cpus();
213#else
214	/* FIXME */
215#endif
216}
217
218void crash_kexec_secondary(struct pt_regs *regs)
219{
220}
221#endif	/* CONFIG_SMP */
222
223/* wait for all the CPUs to hit real mode but timeout if they don't come in */
224#if defined(CONFIG_SMP) && defined(CONFIG_PPC64)
225static void __maybe_unused crash_kexec_wait_realmode(int cpu)
226{
227	unsigned int msecs;
228	int i;
229
230	msecs = REAL_MODE_TIMEOUT;
231	for (i=0; i < nr_cpu_ids && msecs > 0; i++) {
232		if (i == cpu)
233			continue;
234
235		while (paca[i].kexec_state < KEXEC_STATE_REAL_MODE) {
236			barrier();
237			if (!cpu_possible(i) || !cpu_online(i) || (msecs <= 0))
238				break;
239			msecs--;
240			mdelay(1);
241		}
242	}
243	mb();
244}
245#else
246static inline void crash_kexec_wait_realmode(int cpu) {}
247#endif	/* CONFIG_SMP && CONFIG_PPC64 */
248
249/*
250 * Register a function to be called on shutdown.  Only use this if you
251 * can't reset your device in the second kernel.
252 */
253int crash_shutdown_register(crash_shutdown_t handler)
254{
255	unsigned int i, rc;
256
257	spin_lock(&crash_handlers_lock);
258	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
259		if (!crash_shutdown_handles[i]) {
260			/* Insert handle at first empty entry */
261			crash_shutdown_handles[i] = handler;
262			rc = 0;
263			break;
264		}
265
266	if (i == CRASH_HANDLER_MAX) {
267		printk(KERN_ERR "Crash shutdown handles full, "
268		       "not registered.\n");
269		rc = 1;
270	}
271
272	spin_unlock(&crash_handlers_lock);
273	return rc;
274}
275EXPORT_SYMBOL(crash_shutdown_register);
276
277int crash_shutdown_unregister(crash_shutdown_t handler)
278{
279	unsigned int i, rc;
280
281	spin_lock(&crash_handlers_lock);
282	for (i = 0 ; i < CRASH_HANDLER_MAX; i++)
283		if (crash_shutdown_handles[i] == handler)
284			break;
285
286	if (i == CRASH_HANDLER_MAX) {
287		printk(KERN_ERR "Crash shutdown handle not found\n");
288		rc = 1;
289	} else {
290		/* Shift handles down */
291		for (; i < (CRASH_HANDLER_MAX - 1); i++)
292			crash_shutdown_handles[i] =
293				crash_shutdown_handles[i+1];
294		/*
295		 * Reset last entry to NULL now that it has been shifted down,
296		 * this will allow new handles to be added here.
297		 */
298		crash_shutdown_handles[i] = NULL;
299		rc = 0;
300	}
301
302	spin_unlock(&crash_handlers_lock);
303	return rc;
304}
305EXPORT_SYMBOL(crash_shutdown_unregister);
306
307void default_machine_crash_shutdown(struct pt_regs *regs)
308{
309	unsigned int i;
310	int (*old_handler)(struct pt_regs *regs);
311
312	/*
313	 * This function is only called after the system
314	 * has panicked or is otherwise in a critical state.
315	 * The minimum amount of code to allow a kexec'd kernel
316	 * to run successfully needs to happen here.
317	 *
318	 * In practice this means stopping other cpus in
319	 * an SMP system.
320	 * The kernel is broken so disable interrupts.
321	 */
322	hard_irq_disable();
323
324	/*
325	 * Make a note of crashing cpu. Will be used in machine_kexec
326	 * such that another IPI will not be sent.
327	 */
328	crashing_cpu = smp_processor_id();
329
330	/*
331	 * If we came in via system reset, wait a while for the secondary
332	 * CPUs to enter.
333	 */
334	if (TRAP(regs) == 0x100)
335		mdelay(PRIMARY_TIMEOUT);
336
337	crash_kexec_prepare_cpus(crashing_cpu);
338
339	crash_save_cpu(regs, crashing_cpu);
340
341	time_to_dump = 1;
342
343	crash_kexec_wait_realmode(crashing_cpu);
344
345	machine_kexec_mask_interrupts();
346
347	/*
348	 * Call registered shutdown routines safely.  Swap out
349	 * __debugger_fault_handler, and replace on exit.
350	 */
351	old_handler = __debugger_fault_handler;
352	__debugger_fault_handler = handle_fault;
353	crash_shutdown_cpu = smp_processor_id();
354	for (i = 0; i < CRASH_HANDLER_MAX && crash_shutdown_handles[i]; i++) {
355		if (setjmp(crash_shutdown_buf) == 0) {
356			/*
357			 * Insert syncs and delay to ensure
358			 * instructions in the dangerous region don't
359			 * leak away from this protected region.
360			 */
361			asm volatile("sync; isync");
362			/* dangerous region */
363			crash_shutdown_handles[i]();
364			asm volatile("sync; isync");
365		}
366	}
367	crash_shutdown_cpu = -1;
368	__debugger_fault_handler = old_handler;
369
370	if (ppc_md.kexec_cpu_down)
371		ppc_md.kexec_cpu_down(1, 0);
372}