2 * Architecture specific (i386) functions for kexec based crash dumps.
4 * Created by: Hariprasad Nellitheertha (hari@in.ibm.com)
6 * Copyright (C) IBM Corporation, 2004. All rights reserved.
10 #include <linux/init.h>
11 #include <linux/types.h>
12 #include <linux/kernel.h>
13 #include <linux/smp.h>
14 #include <linux/irq.h>
15 #include <linux/reboot.h>
16 #include <linux/kexec.h>
17 #include <linux/irq.h>
18 #include <linux/delay.h>
19 #include <linux/elf.h>
20 #include <linux/elfcore.h>
22 #include <asm/processor.h>
23 #include <asm/hardirq.h>
25 #include <asm/hw_irq.h>
30 note_buf_t crash_notes
[NR_CPUS
];
32 static u32
*append_elf_note(u32
*buf
,
33 char *name
, unsigned type
, void *data
, size_t data_len
)
36 note
.n_namesz
= strlen(name
) + 1;
37 note
.n_descsz
= data_len
;
39 memcpy(buf
, ¬e
, sizeof(note
));
40 buf
+= (sizeof(note
) +3)/4;
41 memcpy(buf
, name
, note
.n_namesz
);
42 buf
+= (note
.n_namesz
+ 3)/4;
43 memcpy(buf
, data
, note
.n_descsz
);
44 buf
+= (note
.n_descsz
+ 3)/4;
48 static void final_note(u32
*buf
)
54 memcpy(buf
, ¬e
, sizeof(note
));
58 static void crash_save_this_cpu(struct pt_regs
*regs
, int cpu
)
60 struct elf_prstatus prstatus
;
62 if ((cpu
< 0) || (cpu
>= NR_CPUS
)) {
65 /* Using ELF notes here is opportunistic.
66 * I need a well defined structure format
67 * for the data I pass, and I need tags
68 * on the data to indicate what information I have
69 * squirrelled away. ELF notes happen to provide
70 * all of that that no need to invent something new.
72 buf
= &crash_notes
[cpu
][0];
73 memset(&prstatus
, 0, sizeof(prstatus
));
74 prstatus
.pr_pid
= current
->pid
;
75 elf_core_copy_regs(&prstatus
.pr_reg
, regs
);
76 buf
= append_elf_note(buf
, "CORE", NT_PRSTATUS
,
77 &prstatus
, sizeof(prstatus
));
82 static void crash_get_current_regs(struct pt_regs
*regs
)
84 __asm__
__volatile__("movl %%ebx,%0" : "=m"(regs
->ebx
));
85 __asm__
__volatile__("movl %%ecx,%0" : "=m"(regs
->ecx
));
86 __asm__
__volatile__("movl %%edx,%0" : "=m"(regs
->edx
));
87 __asm__
__volatile__("movl %%esi,%0" : "=m"(regs
->esi
));
88 __asm__
__volatile__("movl %%edi,%0" : "=m"(regs
->edi
));
89 __asm__
__volatile__("movl %%ebp,%0" : "=m"(regs
->ebp
));
90 __asm__
__volatile__("movl %%eax,%0" : "=m"(regs
->eax
));
91 __asm__
__volatile__("movl %%esp,%0" : "=m"(regs
->esp
));
92 __asm__
__volatile__("movw %%ss, %%ax;" :"=a"(regs
->xss
));
93 __asm__
__volatile__("movw %%cs, %%ax;" :"=a"(regs
->xcs
));
94 __asm__
__volatile__("movw %%ds, %%ax;" :"=a"(regs
->xds
));
95 __asm__
__volatile__("movw %%es, %%ax;" :"=a"(regs
->xes
));
96 __asm__
__volatile__("pushfl; popl %0" :"=m"(regs
->eflags
));
98 regs
->eip
= (unsigned long)current_text_addr();
101 static void crash_save_self(void)
105 cpu
= smp_processor_id();
106 crash_get_current_regs(®s
);
107 crash_save_this_cpu(®s
, cpu
);
111 static atomic_t waiting_for_crash_ipi
;
113 static int crash_nmi_callback(struct pt_regs
*regs
, int cpu
)
115 struct pt_regs fixed_regs
;
118 /* CPU does not save ss and esp on stack if execution is already
119 * running in kernel mode at the time of NMI occurrence. This code
122 if (!user_mode(regs
)) {
123 memcpy(&fixed_regs
, regs
, sizeof(*regs
));
124 fixed_regs
.esp
= (unsigned long)&(regs
->esp
);
125 __asm__
__volatile__("xorl %eax, %eax;");
126 __asm__
__volatile__ ("movw %%ss, %%ax;" :"=a"(fixed_regs
.xss
));
129 crash_save_this_cpu(regs
, cpu
);
130 disable_local_APIC();
131 atomic_dec(&waiting_for_crash_ipi
);
132 /* Assume hlt works */
139 * By using the NMI code instead of a vector we just sneak thru the
140 * word generator coming out with just what we want. AND it does
141 * not matter if clustered_apic_mode is set or not.
143 static void smp_send_nmi_allbutself(void)
145 send_IPI_allbutself(APIC_DM_NMI
);
148 static void nmi_shootdown_cpus(void)
151 atomic_set(&waiting_for_crash_ipi
, num_online_cpus() - 1);
153 /* Would it be better to replace the trap vector here? */
154 set_nmi_callback(crash_nmi_callback
);
155 /* Ensure the new callback function is set before sending
160 smp_send_nmi_allbutself();
162 msecs
= 1000; /* Wait at most a second for the other cpus to stop */
163 while ((atomic_read(&waiting_for_crash_ipi
) > 0) && msecs
) {
168 /* Leave the nmi callback set */
169 disable_local_APIC();
172 static void nmi_shootdown_cpus(void)
174 /* There are no cpus to shootdown */
178 void machine_crash_shutdown(void)
180 /* This function is only called after the system
181 * has paniced or is otherwise in a critical state.
182 * The minimum amount of code to allow a kexec'd kernel
183 * to run successfully needs to happen here.
185 * In practice this means shooting down the other cpus in
188 /* The kernel is broken so disable interrupts */
190 nmi_shootdown_cpus();
192 #if defined(CONFIG_X86_IO_APIC)