minix/kernel/clock.c
Tomas Hruby 5efa92f754 NMI watchdog is an awesome feature for debugging locked up kernels.
There is not that much use for it on a single CPU, however, deadlock
between kernel and system task can be delected. Or a runaway loop.

If a kernel gets locked up the timer interrupts don't occure (as all
interrupts are disabled in kernel mode). The only chance is to
interrupt the kernel by a non-maskable interrupt.

This patch generates NMIs using performance counters. It uses the most
widely available performace counters. As the performance counters are 
highly model-specific this patch is not guaranteed to work on every
machine.  Unfortunately this is also true for KVM :-/ On the other
hand adding this feature for other models is not extremely difficult
and the framework makes it hopefully easy enough.

Depending on the frequency of the CPU an NMI is generated at most
about every 0.5s If the cpu's speed is less then 2Ghz it is generated
at most every 1s. In general an NMI is generated much less often as
the performance counter counts down only if the cpu is not idle.
Therefore the overhead of this feature is fairly minimal even if the
load is high.

Uppon detecting that the kernel is locked up the kernel dumps the 
state of the kernel registers and panics.

Local APIC must be enabled for the watchdog to work.

The code is _always_ compiled in, however, it is only enabled if  
watchdog=<non-zero> is set in the boot monitor.

One corner case is serial console debugging. As dumping a lot of stuff
to the serial link may take a lot of time, the watchdog does not 
detect lockups during this time!!! as it would result in too many
false positives. 10 nmi have to be handled before the lockup is
detected. This means something between ~5s to 10s.

Another corner case is that the watchdog is enabled only after the
paging is enabled as it would be pure madness to try to get it right.
2010-01-16 20:53:55 +00:00

316 lines
9.9 KiB
C

/* This file contains the clock task, which handles time related functions.
* Important events that are handled by the CLOCK include setting and
* monitoring alarm timers and deciding when to (re)schedule processes.
* The CLOCK offers a direct interface to kernel processes. System services
* can access its services through system calls, such as sys_setalarm(). The
* CLOCK task thus is hidden from the outside world.
*
* Changes:
* Aug 18, 2006 removed direct hardware access etc, MinixPPC (Ingmar Alting)
* Oct 08, 2005 reordering and comment editing (A. S. Woodhull)
* Mar 18, 2004 clock interface moved to SYSTEM task (Jorrit N. Herder)
* Sep 30, 2004 source code documentation updated (Jorrit N. Herder)
* Sep 24, 2004 redesigned alarm timers (Jorrit N. Herder)
*
* Clock task is notified by the clock's interrupt handler when a timer
* has expired.
*
* In addition to the main clock_task() entry point, which starts the main
* loop, there are several other minor entry points:
* clock_stop: called just before MINIX shutdown
* get_uptime: get realtime since boot in clock ticks
* set_timer: set a watchdog timer (+)
* reset_timer: reset a watchdog timer (+)
* read_clock: read the counter of channel 0 of the 8253A timer
*
* (+) The CLOCK task keeps tracks of watchdog timers for the entire kernel.
* It is crucial that watchdog functions not block, or the CLOCK task may
* be blocked. Do not send() a message when the receiver is not expecting it.
* Instead, notify(), which always returns, should be used.
*/
#include "kernel.h"
#include "proc.h"
#include <signal.h>
#include <minix/com.h>
#include <minix/endpoint.h>
#include <minix/portio.h>
#include "clock.h"
#ifdef CONFIG_WATCHDOG
#include "watchdog.h"
#endif
/* Function prototype for PRIVATE functions.
*/
FORWARD _PROTOTYPE( void init_clock, (void) );
FORWARD _PROTOTYPE( void load_update, (void));
/* The CLOCK's timers queue. The functions in <timers.h> operate on this.
* Each system process possesses a single synchronous alarm timer. If other
* kernel parts want to use additional timers, they must declare their own
* persistent (static) timer structure, which can be passed to the clock
* via (re)set_timer().
* When a timer expires its watchdog function is run by the CLOCK task.
*/
PRIVATE timer_t *clock_timers; /* queue of CLOCK timers */
PRIVATE clock_t next_timeout; /* realtime that next timer expires */
/* The time is incremented by the interrupt handler on each clock tick.
*/
PRIVATE clock_t realtime = 0; /* real time clock */
/*===========================================================================*
* clock_task *
*===========================================================================*/
PUBLIC void clock_task()
{
/* Main program of clock task. If the call is not HARD_INT it is an error.
*/
message m; /* message buffer for both input and output */
int result; /* result returned by the handler */
init_clock(); /* initialize clock task */
/* Main loop of the clock task. Get work, process it. Never reply. */
while(TRUE) {
/* Go get a message. */
result = receive(ANY, &m);
if(result != OK)
minix_panic("receive() failed", result);
/* Handle the request. Only clock ticks are expected. */
if (is_notify(m.m_type)) {
switch (_ENDPOINT_P(m.m_source)) {
case HARDWARE:
tmrs_exptimers(&clock_timers, realtime, NULL);
next_timeout = (clock_timers == NULL) ?
TMR_NEVER : clock_timers->tmr_exp_time;
break;
default: /* illegal request type */
kprintf("CLOCK: illegal notify %d from %d.\n",
m.m_type, m.m_source);
}
}
else {
/* illegal request type */
kprintf("CLOCK: illegal request %d from %d.\n",
m.m_type, m.m_source);
}
}
}
/*===========================================================================*
* init_clock *
*===========================================================================*/
PRIVATE void init_clock()
{
/* Set a watchdog timer to periodically balance the scheduling queues.
Side-effect sets new timer */
balance_queues(NULL);
}
/*
* The boot processor timer interrupt handler. In addition to non-boot cpus it
* keeps real time and notifies the clock task if need be
*/
PUBLIC int bsp_timer_int_handler(void)
{
unsigned ticks;
IDLE_STOP;
if(minix_panicing)
return 0;
/* Get number of ticks and update realtime. */
ticks = lost_ticks + 1;
lost_ticks = 0;
realtime += ticks;
ap_timer_int_handler();
/* if a timer expired, notify the clock task */
if ((next_timeout <= realtime)) {
mini_notify(proc_addr(HARDWARE), CLOCK); /* send notification */
}
if (do_serial_debug)
do_ser_debug();
return(1); /* reenable interrupts */
}
/*===========================================================================*
* get_uptime *
*===========================================================================*/
PUBLIC clock_t get_uptime(void)
{
/* Get and return the current clock uptime in ticks. */
return(realtime);
}
/*===========================================================================*
* set_timer *
*===========================================================================*/
PUBLIC void set_timer(tp, exp_time, watchdog)
struct timer *tp; /* pointer to timer structure */
clock_t exp_time; /* expiration realtime */
tmr_func_t watchdog; /* watchdog to be called */
{
/* Insert the new timer in the active timers list. Always update the
* next timeout time by setting it to the front of the active list.
*/
tmrs_settimer(&clock_timers, tp, exp_time, watchdog, NULL);
next_timeout = clock_timers->tmr_exp_time;
}
/*===========================================================================*
* reset_timer *
*===========================================================================*/
PUBLIC void reset_timer(tp)
struct timer *tp; /* pointer to timer structure */
{
/* The timer pointed to by 'tp' is no longer needed. Remove it from both the
* active and expired lists. Always update the next timeout time by setting
* it to the front of the active list.
*/
tmrs_clrtimer(&clock_timers, tp, NULL);
next_timeout = (clock_timers == NULL) ?
TMR_NEVER : clock_timers->tmr_exp_time;
}
/*===========================================================================*
* load_update *
*===========================================================================*/
PRIVATE void load_update(void)
{
u16_t slot;
int enqueued = 0, q;
struct proc *p;
/* Load average data is stored as a list of numbers in a circular
* buffer. Each slot accumulates _LOAD_UNIT_SECS of samples of
* the number of runnable processes. Computations can then
* be made of the load average over variable periods, in the
* user library (see getloadavg(3)).
*/
slot = (realtime / system_hz / _LOAD_UNIT_SECS) % _LOAD_HISTORY;
if(slot != kloadinfo.proc_last_slot) {
kloadinfo.proc_load_history[slot] = 0;
kloadinfo.proc_last_slot = slot;
}
/* Cumulation. How many processes are ready now? */
for(q = 0; q < NR_SCHED_QUEUES; q++)
for(p = rdy_head[q]; p != NIL_PROC; p = p->p_nextready)
enqueued++;
kloadinfo.proc_load_history[slot] += enqueued;
/* Up-to-dateness. */
kloadinfo.last_clock = realtime;
}
/*
* Timer interupt handler. This is the only think executed on non boot
* processors. It is called by bsp_timer_int_handler() on the boot processor
*/
PUBLIC int ap_timer_int_handler(void)
{
/* Update user and system accounting times. Charge the current process
* for user time. If the current process is not billable, that is, if a
* non-user process is running, charge the billable process for system
* time as well. Thus the unbillable process' user time is the billable
* user's system time.
*/
unsigned ticks = 1;
int expired = 0;
struct proc * p, * billp;
IDLE_STOP;
#ifdef CONFIG_WATCHDOG
/*
* we need to know whether local timer ticks are happening or whether
* the kernel is locked up. We don't care about overflows as we only
* need to know that it's still ticking or not
*/
watchdog_local_timer_ticks++;
#endif
/* Update user and system accounting times. Charge the current process
* for user time. If the current process is not billable, that is, if a
* non-user process is running, charge the billable process for system
* time as well. Thus the unbillable process' user time is the billable
* user's system time.
*/
/* FIXME prepared for get_cpu_local_var() */
p = proc_ptr;
billp = bill_ptr;
p->p_user_time += ticks;
if (priv(p)->s_flags & PREEMPTIBLE) {
p->p_ticks_left -= ticks;
}
if (! (priv(p)->s_flags & BILLABLE)) {
billp->p_sys_time += ticks;
billp->p_ticks_left -= ticks;
}
/* Decrement virtual timers, if applicable. We decrement both the
* virtual and the profile timer of the current process, and if the
* current process is not billable, the timer of the billed process as
* well. If any of the timers expire, do_clocktick() will send out
* signals.
*/
if ((p->p_misc_flags & MF_VIRT_TIMER) &&
(p->p_virt_left -= ticks) <= 0) expired = 1;
if ((p->p_misc_flags & MF_PROF_TIMER) &&
(p->p_prof_left -= ticks) <= 0) expired = 1;
if (! (priv(p)->s_flags & BILLABLE) &&
(billp->p_misc_flags & MF_PROF_TIMER) &&
(billp->p_prof_left -= ticks) <= 0) expired = 1;
/*
* Check if a process-virtual timer expired. Check current process, but
* also bill_ptr - one process's user time is another's system time, and
* the profile timer decreases for both!
*/
vtimer_check(p);
if (p != billp)
vtimer_check(billp);
/* Update load average. */
load_update();
/* check if the process is still runnable after checking the vtimer */
if (p->p_rts_flags == 0 && p->p_ticks_left <= 0 &&
priv(p)->s_flags & PREEMPTIBLE) {
/* this dequeues the process */
RTS_SET(p, RTS_NO_QUANTUM);
}
return 1;
}
PUBLIC int boot_cpu_init_timer(unsigned freq)
{
if (arch_init_local_timer(freq))
return -1;
if (arch_register_local_timer_handler(
(irq_handler_t) bsp_timer_int_handler))
return -1;
return 0;
}