crash.c revision 8385a6a3acfbb4b68150c25cfe9084b6c4f501cf 
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#undef DEBUG
14
15#include <linux/kernel.h>
16#include <linux/smp.h>
17#include <linux/reboot.h>
18#include <linux/kexec.h>
19#include <linux/bootmem.h>
20#include <linux/crash_dump.h>
21#include <linux/delay.h>
22#include <linux/elf.h>
23#include <linux/elfcore.h>
24#include <linux/init.h>
25#include <linux/types.h>
26
27#include <asm/processor.h>
28#include <asm/machdep.h>
29#include <asm/kdump.h>
30#include <asm/lmb.h>
31#include <asm/firmware.h>
32#include <asm/smp.h>
33
34#ifdef DEBUG
35#include <asm/udbg.h>
36#define DBG(fmt...) udbg_printf(fmt)
37#else
38#define DBG(fmt...)
39#endif
40
41/* This keeps a track of which one is crashing cpu. */
42int crashing_cpu = -1;
43
44static u32 *append_elf_note(u32 *buf, char *name, unsigned type, void *data,
45							       size_t data_len)
46{
47	struct elf_note note;
48
49	note.n_namesz = strlen(name) + 1;
50	note.n_descsz = data_len;
51	note.n_type   = type;
52	memcpy(buf, &note, sizeof(note));
53	buf += (sizeof(note) +3)/4;
54	memcpy(buf, name, note.n_namesz);
55	buf += (note.n_namesz + 3)/4;
56	memcpy(buf, data, note.n_descsz);
57	buf += (note.n_descsz + 3)/4;
58
59	return buf;
60}
61
62static void final_note(u32 *buf)
63{
64	struct elf_note note;
65
66	note.n_namesz = 0;
67	note.n_descsz = 0;
68	note.n_type   = 0;
69	memcpy(buf, &note, sizeof(note));
70}
71
72static void crash_save_this_cpu(struct pt_regs *regs, int cpu)
73{
74	struct elf_prstatus prstatus;
75	u32 *buf;
76
77	if ((cpu < 0) || (cpu >= NR_CPUS))
78		return;
79
80	/* Using ELF notes here is opportunistic.
81	 * I need a well defined structure format
82	 * for the data I pass, and I need tags
83	 * on the data to indicate what information I have
84	 * squirrelled away.  ELF notes happen to provide
85	 * all of that that no need to invent something new.
86	 */
87	buf = (u32*)per_cpu_ptr(crash_notes, cpu);
88	if (!buf)
89		return;
90
91	memset(&prstatus, 0, sizeof(prstatus));
92	prstatus.pr_pid = current->pid;
93	elf_core_copy_regs(&prstatus.pr_reg, regs);
94	buf = append_elf_note(buf, "CORE", NT_PRSTATUS, &prstatus,
95			sizeof(prstatus));
96	final_note(buf);
97}
98
99#ifdef CONFIG_SMP
100static atomic_t waiting_for_crash_ipi;
101
102void crash_ipi_callback(struct pt_regs *regs)
103{
104	int cpu = smp_processor_id();
105
106	if (cpu == crashing_cpu)
107		return;
108
109	if (!cpu_online(cpu))
110		return;
111
112	if (ppc_md.kexec_cpu_down)
113		ppc_md.kexec_cpu_down(1, 1);
114
115	local_irq_disable();
116
117	crash_save_this_cpu(regs, cpu);
118	atomic_dec(&waiting_for_crash_ipi);
119	kexec_smp_wait();
120	/* NOTREACHED */
121}
122
123static void crash_kexec_prepare_cpus(void)
124{
125	unsigned int msecs;
126
127	atomic_set(&waiting_for_crash_ipi, num_online_cpus() - 1);
128
129	crash_send_ipi(crash_ipi_callback);
130	smp_wmb();
131
132	/*
133	 * FIXME: Until we will have the way to stop other CPUSs reliabally,
134	 * the crash CPU will send an IPI and wait for other CPUs to
135	 * respond. If not, proceed the kexec boot even though we failed to
136	 * capture other CPU states.
137	 */
138	msecs = 1000000;
139	while ((atomic_read(&waiting_for_crash_ipi) > 0) && (--msecs > 0)) {
140		barrier();
141		mdelay(1);
142	}
143
144	/* Would it be better to replace the trap vector here? */
145
146	/*
147	 * FIXME: In case if we do not get all CPUs, one possibility: ask the
148	 * user to do soft reset such that we get all.
149	 * IPI handler is already set by the panic cpu initially. Therefore,
150	 * all cpus could invoke this handler from die() and the panic CPU
151	 * will call machine_kexec() directly from this handler to do
152	 * kexec boot.
153	 */
154	if (atomic_read(&waiting_for_crash_ipi))
155		printk(KERN_ALERT "done waiting: %d cpus not responding\n",
156			atomic_read(&waiting_for_crash_ipi));
157	/* Leave the IPI callback set */
158}
159#else
160static void crash_kexec_prepare_cpus(void)
161{
162	/*
163	 * move the secondarys to us so that we can copy
164	 * the new kernel 0-0x100 safely
165	 *
166	 * do this if kexec in setup.c ?
167	 */
168	smp_release_cpus();
169}
170
171#endif
172
173void default_machine_crash_shutdown(struct pt_regs *regs)
174{
175	/*
176	 * This function is only called after the system
177	 * has paniced or is otherwise in a critical state.
178	 * The minimum amount of code to allow a kexec'd kernel
179	 * to run successfully needs to happen here.
180	 *
181	 * In practice this means stopping other cpus in
182	 * an SMP system.
183	 * The kernel is broken so disable interrupts.
184	 */
185	local_irq_disable();
186
187	if (ppc_md.kexec_cpu_down)
188		ppc_md.kexec_cpu_down(1, 0);
189
190	/*
191	 * Make a note of crashing cpu. Will be used in machine_kexec
192	 * such that another IPI will not be sent.
193	 */
194	crashing_cpu = smp_processor_id();
195	crash_kexec_prepare_cpus();
196	crash_save_this_cpu(regs, crashing_cpu);
197}
198