2005-06-26 01:57:56 +04:00
/*
* Architecture specific ( i386 ) functions for kexec based crash dumps .
*
* Created by : Hariprasad Nellitheertha ( hari @ in . ibm . com )
*
* Copyright ( C ) IBM Corporation , 2004. All rights reserved .
*
*/
# include <linux/init.h>
# include <linux/types.h>
# include <linux/kernel.h>
# include <linux/smp.h>
# include <linux/reboot.h>
# include <linux/kexec.h>
# include <linux/delay.h>
# include <linux/elf.h>
# include <linux/elfcore.h>
# include <asm/processor.h>
# include <asm/hardirq.h>
# include <asm/nmi.h>
# include <asm/hw_irq.h>
2005-11-15 11:09:04 +03:00
# include <asm/apic.h>
2005-06-26 01:57:58 +04:00
# include <mach_ipi.h>
2005-06-26 01:57:56 +04:00
2005-06-26 01:58:14 +04:00
/* This keeps a track of which one is crashing cpu. */
static int crashing_cpu ;
2005-06-26 01:57:56 +04:00
2005-06-26 01:58:28 +04:00
static u32 * append_elf_note ( u32 * buf , char * name , unsigned type , void * data ,
size_t data_len )
2005-06-26 01:57:59 +04:00
{
struct elf_note note ;
2005-06-26 01:58:28 +04:00
2005-06-26 01:57:59 +04:00
note . n_namesz = strlen ( name ) + 1 ;
note . n_descsz = data_len ;
note . n_type = type ;
memcpy ( buf , & note , sizeof ( note ) ) ;
buf + = ( sizeof ( note ) + 3 ) / 4 ;
memcpy ( buf , name , note . n_namesz ) ;
buf + = ( note . n_namesz + 3 ) / 4 ;
memcpy ( buf , data , note . n_descsz ) ;
buf + = ( note . n_descsz + 3 ) / 4 ;
2005-06-26 01:58:28 +04:00
2005-06-26 01:57:59 +04:00
return buf ;
}
static void final_note ( u32 * buf )
{
struct elf_note note ;
2005-06-26 01:58:28 +04:00
2005-06-26 01:57:59 +04:00
note . n_namesz = 0 ;
note . n_descsz = 0 ;
note . n_type = 0 ;
memcpy ( buf , & note , sizeof ( note ) ) ;
}
static void crash_save_this_cpu ( struct pt_regs * regs , int cpu )
{
struct elf_prstatus prstatus ;
u32 * buf ;
2005-06-26 01:58:28 +04:00
if ( ( cpu < 0 ) | | ( cpu > = NR_CPUS ) )
2005-06-26 01:57:59 +04:00
return ;
2005-06-26 01:58:28 +04:00
2005-06-26 01:57:59 +04:00
/* Using ELF notes here is opportunistic.
* I need a well defined structure format
* for the data I pass , and I need tags
* on the data to indicate what information I have
* squirrelled away . ELF notes happen to provide
2006-04-01 03:39:17 +04:00
* all of that , so there is no need to invent something new .
2005-06-26 01:57:59 +04:00
*/
2006-01-10 07:51:41 +03:00
buf = ( u32 * ) per_cpu_ptr ( crash_notes , cpu ) ;
if ( ! buf )
return ;
2005-06-26 01:57:59 +04:00
memset ( & prstatus , 0 , sizeof ( prstatus ) ) ;
prstatus . pr_pid = current - > pid ;
elf_core_copy_regs ( & prstatus . pr_reg , regs ) ;
2005-06-26 01:58:28 +04:00
buf = append_elf_note ( buf , " CORE " , NT_PRSTATUS , & prstatus ,
sizeof ( prstatus ) ) ;
2005-06-26 01:57:59 +04:00
final_note ( buf ) ;
}
2006-01-10 07:51:44 +03:00
static void crash_save_self ( struct pt_regs * regs )
2005-06-26 01:57:59 +04:00
{
int cpu ;
2005-06-26 01:58:26 +04:00
2005-06-26 01:58:28 +04:00
cpu = smp_processor_id ( ) ;
2006-01-10 07:51:44 +03:00
crash_save_this_cpu ( regs , cpu ) ;
2005-06-26 01:57:59 +04:00
}
2005-06-26 01:57:58 +04:00
# ifdef CONFIG_SMP
static atomic_t waiting_for_crash_ipi ;
static int crash_nmi_callback ( struct pt_regs * regs , int cpu )
{
2005-06-26 01:58:13 +04:00
struct pt_regs fixed_regs ;
2005-06-26 01:58:14 +04:00
/* Don't do anything if this handler is invoked on crashing cpu.
* Otherwise , system will completely hang . Crashing cpu can get
* an NMI if system was initially booted with nmi_watchdog parameter .
*/
if ( cpu = = crashing_cpu )
return 1 ;
2005-06-26 01:57:58 +04:00
local_irq_disable ( ) ;
2005-06-26 01:58:13 +04:00
2006-03-23 13:59:46 +03:00
if ( ! user_mode_vm ( regs ) ) {
2006-01-10 07:51:44 +03:00
crash_fixup_ss_esp ( & fixed_regs , regs ) ;
2005-06-26 01:58:13 +04:00
regs = & fixed_regs ;
}
2005-06-26 01:57:59 +04:00
crash_save_this_cpu ( regs , cpu ) ;
2005-11-15 11:09:04 +03:00
disable_local_APIC ( ) ;
2005-06-26 01:57:58 +04:00
atomic_dec ( & waiting_for_crash_ipi ) ;
/* Assume hlt works */
2005-09-04 02:56:42 +04:00
halt ( ) ;
2006-06-25 16:46:53 +04:00
for ( ; ; )
cpu_relax ( ) ;
2005-06-26 01:58:28 +04:00
2005-06-26 01:57:58 +04:00
return 1 ;
}
/*
* By using the NMI code instead of a vector we just sneak thru the
* word generator coming out with just what we want . AND it does
* not matter if clustered_apic_mode is set or not .
*/
static void smp_send_nmi_allbutself ( void )
{
send_IPI_allbutself ( APIC_DM_NMI ) ;
}
static void nmi_shootdown_cpus ( void )
{
unsigned long msecs ;
2005-06-26 01:58:28 +04:00
atomic_set ( & waiting_for_crash_ipi , num_online_cpus ( ) - 1 ) ;
2005-06-26 01:57:58 +04:00
/* Would it be better to replace the trap vector here? */
set_nmi_callback ( crash_nmi_callback ) ;
/* Ensure the new callback function is set before sending
* out the NMI
*/
wmb ( ) ;
smp_send_nmi_allbutself ( ) ;
msecs = 1000 ; /* Wait at most a second for the other cpus to stop */
while ( ( atomic_read ( & waiting_for_crash_ipi ) > 0 ) & & msecs ) {
mdelay ( 1 ) ;
msecs - - ;
}
/* Leave the nmi callback set */
2005-11-15 11:09:04 +03:00
disable_local_APIC ( ) ;
2005-06-26 01:57:58 +04:00
}
# else
static void nmi_shootdown_cpus ( void )
{
/* There are no cpus to shootdown */
}
# endif
2005-06-26 01:58:26 +04:00
void machine_crash_shutdown ( struct pt_regs * regs )
2005-06-26 01:57:56 +04:00
{
/* This function is only called after the system
* has paniced or is otherwise in a critical state .
* The minimum amount of code to allow a kexec ' d kernel
* to run successfully needs to happen here .
*
* In practice this means shooting down the other cpus in
* an SMP system .
*/
2005-06-26 01:57:58 +04:00
/* The kernel is broken so disable interrupts */
local_irq_disable ( ) ;
2005-06-26 01:58:14 +04:00
/* Make a note of crashing cpu. Will be used in NMI callback.*/
crashing_cpu = smp_processor_id ( ) ;
2005-06-26 01:57:58 +04:00
nmi_shootdown_cpus ( ) ;
2005-11-15 11:09:04 +03:00
lapic_shutdown ( ) ;
# if defined(CONFIG_X86_IO_APIC)
disable_IO_APIC ( ) ;
# endif
2005-06-26 01:58:26 +04:00
crash_save_self ( regs ) ;
2005-06-26 01:57:56 +04:00
}