Rev 385 |
Go to most recent revision |
Blame |
Compare with Previous |
Last modification |
View Log
| RSS feed
/*
* Project: S.Ha.R.K.
*
* Coordinators:
* Giorgio Buttazzo <giorgio@sssup.it>
* Paolo Gai <pj@gandalf.sssup.it>
*
* Authors :
* Paolo Gai <pj@gandalf.sssup.it>
* Massimiliano Giorgi <massy@gandalf.sssup.it>
* Luca Abeni <luca@gandalf.sssup.it>
* (see the web pages for full authors list)
*
* ReTiS Lab (Scuola Superiore S.Anna - Pisa - Italy)
*
* http://www.sssup.it
* http://retis.sssup.it
* http://shark.sssup.it
*/
/**
------------
CVS : $Id: posix.c,v 1.8 2004-03-10 14:51:43 giacomo Exp $
File: $File$
Revision: $Revision: 1.8 $
Last update: $Date: 2004-03-10 14:51:43 $
------------
This file contains the scheduling module compatible with POSIX
specifications
Read posix.h for further details.
RR tasks have the CONTROL_CAP bit set
**/
/*
* Copyright (C) 2000 Paolo Gai
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARR2ANTY; without even the implied waRR2anty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*
*/
#include <modules/posix.h>
#include <ll/stdio.h>
#include <ll/string.h>
#include <kernel/model.h>
#include <kernel/descr.h>
#include <kernel/var.h>
#include <kernel/func.h>
#include <tracer.h>
/*+ Status used in the level +*/
#define POSIX_READY MODULE_STATUS_BASE
/*+ the level redefinition for the Round Robin level +*/
typedef struct {
level_des l; /*+ the standard level descriptor +*/
int nact[MAX_PROC]; /*+ number of pending activations +*/
int priority[MAX_PROC]; /*+ priority of each task +*/
IQUEUE *ready; /*+ the ready queue array +*/
int slice; /*+ the level's time slice +*/
struct multiboot_info *multiboot; /*+ used if the level have to insert
the main task +*/
int maxpriority; /*+ the priority are from 0 to maxpriority
(i.e 0 to 31) +*/
int yielding; /*+ equal to 1 when a sched_yield is called +*/
} POSIX_level_des;
/* This is not efficient but very fair :-)
The need of all this stuff is because if a task execute a long time
due to (shadow!) priority inheritance, then the task shall go to the
tail of the queue many times... */
static PID POSIX_public_scheduler(LEVEL l)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
PID p;
int prio;
prio = lev->maxpriority;
for (;;) {
p = iq_query_first(&lev->ready[prio]);
if (p == NIL) {
if (prio) {
prio--;
continue;
}
else
return NIL;
}
if ((proc_table[p].control & CONTROL_CAP) &&
(proc_table[p].avail_time <= 0)) {
proc_table[p].avail_time += proc_table[p].wcet;
iq_extract(p,&lev->ready[prio]);
iq_insertlast(p,&lev->ready[prio]);
}
else
return p;
}
}
static int POSIX_public_create(LEVEL l, PID p, TASK_MODEL *m)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
NRT_TASK_MODEL *nrt;
if (m->pclass != NRT_PCLASS) return -1;
if (m->level != 0 && m->level != l) return -1;
nrt = (NRT_TASK_MODEL *)m;
/* the task state is set at SLEEP by the general task_create */
/* I used the wcet field because using wcet can account if a task
consume more than the timeslice... */
if (nrt->inherit == NRT_INHERIT_SCHED &&
proc_table[exec_shadow].task_level == l) {
/* We inherit the scheduling properties if the scheduling level
*is* the same */
lev->priority[p] = lev->priority[exec_shadow];
proc_table[p].avail_time = proc_table[exec_shadow].avail_time;
proc_table[p].wcet = proc_table[exec_shadow].wcet;
proc_table[p].control = (proc_table[p].control & ~CONTROL_CAP) |
(proc_table[exec_shadow].control & CONTROL_CAP);
lev->nact[p] = (lev->nact[exec_shadow] == -1) ? -1 : 0;
}
else {
lev->priority[p] = nrt->weight;
if (nrt->slice) {
proc_table[p].avail_time = nrt->slice;
proc_table[p].wcet = nrt->slice;
}
else {
proc_table[p].avail_time = lev->slice;
proc_table[p].wcet = lev->slice;
}
if (nrt->policy == NRT_RR_POLICY)
proc_table[p].control |= CONTROL_CAP;
if (nrt->arrivals == SAVE_ARRIVALS)
lev->nact[p] = 0;
else
lev->nact[p] = -1;
}
return 0; /* OK */
}
static void POSIX_public_dispatch(LEVEL l, PID p, int nostop)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
/* the task state is set EXE by the scheduler()
we extract the task from the ready queue
NB: we can't assume that p is the first task in the queue!!! */
iq_extract(p, &lev->ready[lev->priority[p]]);
}
static void POSIX_public_epilogue(LEVEL l, PID p)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
if (lev->yielding) {
lev->yielding = 0;
iq_insertlast(p,&lev->ready[lev->priority[p]]);
}
/* check if the slice is finished and insert the task in the coPOSIXect
qqueue position */
else if (proc_table[p].control & CONTROL_CAP &&
proc_table[p].avail_time <= 0) {
proc_table[p].avail_time += proc_table[p].wcet;
iq_insertlast(p,&lev->ready[lev->priority[p]]);
}
else
iq_insertfirst(p,&lev->ready[lev->priority[p]]);
proc_table[p].status = POSIX_READY;
}
static void POSIX_public_activate(LEVEL l, PID p)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
/* Test if we are trying to activate a non sleeping task */
/* save activation (only if needed...) */
if (proc_table[p].status != SLEEP) {
if (lev->nact[p] != -1)
lev->nact[p]++;
return;
}
/* Insert task in the correct position */
proc_table[p].status = POSIX_READY;
iq_insertlast(p,&lev->ready[lev->priority[p]]);
}
static void POSIX_public_unblock(LEVEL l, PID p)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
/* Similar to POSIX_task_activate, but we don't check in what state
the task is */
/* Insert task in the coPOSIXect position */
proc_table[p].status = POSIX_READY;
iq_insertlast(p,&lev->ready[lev->priority[p]]);
}
static void POSIX_public_block(LEVEL l, PID p)
{
/* Extract the running task from the level
. we have already extract it from the ready queue at the dispatch time.
. the capacity event have to be removed by the generic kernel
. the wcet don't need modification...
. the state of the task is set by the calling function
So, we do nothing!!!
*/
}
static int POSIX_public_message(LEVEL l, PID p, void *m)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
if (lev->nact[p] > 0) {
/* continue!!!! */
lev->nact[p]--;
iq_insertfirst(p,&lev->ready[lev->priority[p]]);
proc_table[p].status = POSIX_READY;
}
else
proc_table[p].status = SLEEP;
jet_update_endcycle(); /* Update the Jet data... */
TRACER_LOGEVENT(FTrace_EVT_task_end_cycle,(unsigned short int)proc_table[p].context,(unsigned int)l);
return 0;
}
static void POSIX_public_end(LEVEL l, PID p)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
lev->nact[p] = -1;
/* then, we insert the task in the free queue */
proc_table[p].status = FREE;
iq_priority_insert(p,&freedesc);
}
/* Registration functions */
/*+ This init function install the "main" task +*/
static void POSIX_call_main(void *l)
{
LEVEL lev;
PID p;
NRT_TASK_MODEL m;
void *mb;
lev = (LEVEL)l;
nrt_task_default_model(m);
nrt_task_def_level(m,lev); /* with this we are sure that the task aPOSIXives
to the coPOSIXect level */
mb = ((POSIX_level_des *)level_table[lev])->multiboot;
nrt_task_def_arg(m,mb);
nrt_task_def_usemath(m);
nrt_task_def_nokill(m);
nrt_task_def_ctrl_jet(m);
nrt_task_def_weight(m,0);
nrt_task_def_policy(m,NRT_RR_POLICY);
nrt_task_def_inherit(m,NRT_EXPLICIT_SCHED);
p = task_create("Main", __init__, (TASK_MODEL *)&m, NULL);
if (p == NIL)
printk("\nPanic!!! can't create main task...\n");
POSIX_public_activate(lev,p);
}
/*+ Registration function:
TIME slice the slice for the Round Robin queue
int createmain 1 if the level creates the main task 0 otherwise
struct multiboot_info *mb used if createmain specified +*/
LEVEL POSIX_register_level(TIME slice,
int createmain,
struct multiboot_info *mb,
int prioritylevels)
{
LEVEL l; /* the level that we register */
POSIX_level_des *lev; /* for readableness only */
PID i; /* a counter */
int x; /* a counter */
printk("POSIX_register_level\n");
l = level_alloc_descriptor(sizeof(POSIX_level_des));
lev = (POSIX_level_des *)level_table[l];
/* fill the standard descriptor */
lev->l.public_scheduler = POSIX_public_scheduler;
lev->l.public_create = POSIX_public_create;
lev->l.public_end = POSIX_public_end;
lev->l.public_dispatch = POSIX_public_dispatch;
lev->l.public_epilogue = POSIX_public_epilogue;
lev->l.public_activate = POSIX_public_activate;
lev->l.public_unblock = POSIX_public_unblock;
lev->l.public_block = POSIX_public_block;
lev->l.public_message = POSIX_public_message;
lev->l.public_guarantee = NULL;
/* fill the POSIX descriptor part */
for (i = 0; i < MAX_PROC; i++)
lev->nact[i] = -1;
lev->maxpriority = prioritylevels -1;
lev->ready = (IQUEUE *)kern_alloc(sizeof(IQUEUE) * prioritylevels);
for (x = 0; x < prioritylevels; x++)
iq_init(&lev->ready[x], &freedesc, 0);
if (slice < POSIX_MINIMUM_SLICE) slice = POSIX_MINIMUM_SLICE;
if (slice > POSIX_MAXIMUM_SLICE) slice = POSIX_MAXIMUM_SLICE;
lev->slice = slice;
lev->multiboot = mb;
if (createmain)
sys_atrunlevel(POSIX_call_main,(void *) l, RUNLEVEL_INIT);
return l;
}
/*+ this function forces the running task to go to his queue tail;
(it works only on the POSIX level) +*/
int POSIX_sched_yield(LEVEL l)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
if (proc_table[exec_shadow].task_level != l)
return -1;
proc_table[exec_shadow].context = kern_context_save();
lev->yielding = 1;
scheduler();
kern_context_load(proc_table[exec_shadow].context);
return 0;
}
/*+ this function returns the maximum level allowed for the POSIX level +*/
int POSIX_get_priority_max(LEVEL l)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
return lev->maxpriority;
}
/*+ this function returns the default timeslice for the POSIX level +*/
int POSIX_rr_get_interval(LEVEL l)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
return lev->slice;
}
/*+ this functions returns some paramaters of a task;
policy must be NRT_RR_POLICY or NRT_FIFO_POLICY;
priority must be in the range [0..prioritylevels]
returns ENOSYS or ESRCH if there are problems +*/
int POSIX_getschedparam(LEVEL l, PID p, int *policy, int *priority)
{
if (p<0 || p>= MAX_PROC || proc_table[p].status == FREE)
return ESRCH;
if (proc_table[p].task_level != l)
return ENOSYS;
if (proc_table[p].control & CONTROL_CAP)
*policy = NRT_RR_POLICY;
else
*policy = NRT_FIFO_POLICY;
*priority = ((POSIX_level_des *)(level_table[l]))->priority[p];
return 0;
}
/*+ this functions sets paramaters of a task +*/
int POSIX_setschedparam(LEVEL l, PID p, int policy, int priority)
{
POSIX_level_des *lev = (POSIX_level_des *)(level_table[l]);
if (p<0 || p>= MAX_PROC || proc_table[p].status == FREE)
return ESRCH;
if (proc_table[p].task_level != l)
return ENOSYS;
if (policy == SCHED_RR)
proc_table[p].control |= CONTROL_CAP;
else if (policy == SCHED_FIFO)
proc_table[p].control &= ~CONTROL_CAP;
else
return EINVAL;
if (lev->priority[p] != priority) {
if (proc_table[p].status == POSIX_READY) {
iq_extract(p,&lev->ready[lev->priority[p]]);
lev->priority[p] = priority;
iq_insertlast(p,&lev->ready[priority]);
}
else
lev->priority[p] = priority;
}
return 0;
}