/*
* Validate the cpus and select the frequency to set
* Called from task cgroup cpuset code with string containing
* the list of cpus to be used by this step
*/
void
cpu_freq_cgroup_validate(stepd_step_rec_t *job, char *step_alloc_cores)
{
uint16_t start = USHRT_MAX;
uint16_t end = USHRT_MAX;
uint16_t cpuidx = 0;
char *core_range;
debug2("cpu_freq_cgroup_validate: request value = %12d %8x",
job->cpu_freq, job->cpu_freq);
debug2(" jobid=%u, stepid=%u, tasks=%u cpu/task=%u, cpus=%u",
job->jobid,job->stepid,job->node_tasks,
job->cpus_per_task,job->cpus);
debug2(" cpu_bind_type=%4x, cpu_bind map=%s",
job->cpu_bind_type, job->cpu_bind);
debug2(" step logical cores = %s, step physical cores = %s",
job->step_alloc_cores, step_alloc_cores);
if (!cpu_freq_count)
return;
/* set entries in cpu frequency table for this step's cpus */
core_range = step_alloc_cores;
while ( (cpuidx = _cpu_freq_next_cpu(&core_range, &cpuidx,
&start, &end)) != USHRT_MAX) {
if (cpuidx >= cpu_freq_count) {
error("cpu_freq_validate: index %u exceeds cpu count %u",
cpuidx, cpu_freq_count);
return;
}
_cpu_freq_find_valid(job->cpu_freq, cpuidx);
}
cpu_freq_set(job);
return;
}
static int
_handle_resume(int fd, stepd_step_rec_t *job, uid_t uid)
{
int rc = SLURM_SUCCESS;
int errnum = 0;
debug("_handle_resume for job %u.%u",
job->jobid, job->stepid);
debug3(" uid = %d", uid);
if (!_slurm_authorized_user(uid)) {
debug("job step resume request from uid %ld for job %u.%u ",
(long)uid, job->jobid, job->stepid);
rc = -1;
errnum = EPERM;
goto done;
}
if (job->cont_id == 0) {
debug ("step %u.%u invalid container [cont_id:%"PRIu64"]",
job->jobid, job->stepid, job->cont_id);
rc = -1;
errnum = ESLURMD_JOB_NOTRUNNING;
goto done;
}
acct_gather_resume_poll();
/*
* Signal the container
*/
pthread_mutex_lock(&suspend_mutex);
if (!suspended) {
rc = -1;
errnum = ESLURMD_STEP_NOTSUSPENDED;
pthread_mutex_unlock(&suspend_mutex);
goto done;
} else {
if (proctrack_g_signal(job->cont_id, SIGCONT) < 0) {
verbose("Error resuming %u.%u: %m",
job->jobid, job->stepid);
} else {
verbose("Resumed %u.%u", job->jobid, job->stepid);
}
suspended = false;
}
/* set the cpu frequencies if cpu_freq option used */
if (job->cpu_freq != NO_VAL)
cpu_freq_set(job);
pthread_mutex_unlock(&suspend_mutex);
done:
/* Send the return code and errno */
safe_write(fd, &rc, sizeof(int));
safe_write(fd, &errnum, sizeof(int));
return SLURM_SUCCESS;
rwfail:
return SLURM_FAILURE;
}
/*
* Validate the cpus and select the frequency to set
* Called from task cgroup cpuset code with string containing
* the list of cpus to be used by this step
*/
extern void
cpu_freq_cgroup_validate(stepd_step_rec_t *job, char *step_alloc_cores)
{
uint16_t start = USHRT_MAX;
uint16_t end = USHRT_MAX;
uint16_t cpuidx = 0;
char *core_range;
debug_flags = slurm_get_debug_flags(); /* init for slurmstepd */
if (debug_flags & DEBUG_FLAG_CPU_FREQ) {
info("cpu_freq_cgroup_validate: request: min=(%12d %8x) "
"max=(%12d %8x) governor=%8x",
job->cpu_freq_min, job->cpu_freq_min,
job->cpu_freq_max, job->cpu_freq_max,
job->cpu_freq_gov);
info(" jobid=%u, stepid=%u, tasks=%u cpu/task=%u, cpus=%u",
job->jobid,job->stepid,job->node_tasks,
job->cpus_per_task,job->cpus);
info(" cpu_bind_type=%4x, cpu_bind map=%s",
job->cpu_bind_type, job->cpu_bind);
info(" step logical cores = %s, step physical cores = %s",
job->step_alloc_cores, step_alloc_cores);
}
if (!cpu_freq_count)
return;
for (cpuidx = 0; cpuidx < cpu_freq_count; cpuidx++) {
_cpu_freq_init_data(cpuidx);
}
/* set entries in cpu frequency table for this step's cpus */
core_range = step_alloc_cores;
while ( (cpuidx = _cpu_freq_next_cpu(&core_range, &cpuidx,
&start, &end)) != USHRT_MAX) {
if (cpuidx >= cpu_freq_count) {
error("cpu_freq_validate: index %u exceeds cpu count %u",
cpuidx, cpu_freq_count);
return;
}
_cpu_freq_setup_data(job, cpuidx);
}
cpu_freq_set(job);
return;
}
/*
* Validate the cpus and select the frequency to set
* Called from task cpuset code with task launch request containing
* a pointer to a hex map string of the cpus to be used by this step
*/
extern void
cpu_freq_cpuset_validate(stepd_step_rec_t *job)
{
int cpuidx, cpu_num;
bitstr_t *cpus_to_set;
bitstr_t *cpu_map;
char *cpu_bind;
char *cpu_str;
char *savestr = NULL;
debug_flags = slurm_get_debug_flags(); /* init for slurmstepd */
if (debug_flags & DEBUG_FLAG_CPU_FREQ) {
info("cpu_freq_cpuset_validate: request: min=(%12d %8x) "
"max=(%12d %8x) governor=%8x",
job->cpu_freq_min, job->cpu_freq_min,
job->cpu_freq_max, job->cpu_freq_max,
job->cpu_freq_gov);
info(" jobid=%u, stepid=%u, tasks=%u cpu/task=%u, cpus=%u",
job->jobid, job->stepid, job->node_tasks,
job->cpus_per_task, job->cpus);
info(" cpu_bind_type=%4x, cpu_bind map=%s",
job->cpu_bind_type, job->cpu_bind);
}
if (!cpu_freq_count)
return;
if (job->cpu_bind == NULL) {
error("cpu_freq_cpuset_validate: cpu_bind string is null");
return;
}
cpu_bind = xstrdup(job->cpu_bind);
if ( (cpu_str = strtok_r(cpu_bind, ",", &savestr) ) == NULL) {
error("cpu_freq_cpuset_validate: cpu_bind string invalid");
xfree(cpu_bind);
return;
}
cpu_map = (bitstr_t *) bit_alloc(cpu_freq_count);
cpus_to_set = (bitstr_t *) bit_alloc(cpu_freq_count);
do {
debug3(" cpu_str = %s", cpu_str);
if ((job->cpu_bind_type & CPU_BIND_MAP) == CPU_BIND_MAP) {
cpu_num = atoi(cpu_str);
if (cpu_num >= cpu_freq_count) {
error("cpu_freq_cpuset_validate: invalid cpu "
"number %d", cpu_num);
bit_free(cpu_map);
bit_free(cpus_to_set);
xfree(cpu_bind);
return;
}
bit_set(cpu_map, (bitoff_t)cpu_num);
} else {
if (bit_unfmt_hexmask(cpu_map, cpu_str) == -1) {
error("cpu_freq_cpuset_validate: invalid cpu "
"mask %s", cpu_bind);
bit_free(cpu_map);
bit_free(cpus_to_set);
xfree(cpu_bind);
return;
}
}
bit_or(cpus_to_set, cpu_map);
} while ( (cpu_str = strtok_r(NULL, ",", &savestr) ) != NULL);
for (cpuidx = 0; cpuidx < cpu_freq_count; cpuidx++) {
_cpu_freq_init_data(cpuidx);
if (bit_test(cpus_to_set, cpuidx)) {
_cpu_freq_setup_data(job, cpuidx);
}
}
cpu_freq_set(job);
bit_free(cpu_map);
bit_free(cpus_to_set);
xfree(cpu_bind);
return;
}
/*
* Validate the cpus and select the frequency to set
* Called from task cpuset code with task launch request containing
* a pointer to a hex map string of the cpus to be used by this step
*/
void
cpu_freq_cpuset_validate(stepd_step_rec_t *job)
{
int cpuidx, cpu_num;
bitstr_t *cpus_to_set;
bitstr_t *cpu_map;
char *cpu_bind;
char *cpu_str;
char *savestr = NULL;
debug2("cpu_freq_cpuset_validate: request = %12d %8x",
job->cpu_freq, job->cpu_freq);
debug2(" jobid=%u, stepid=%u, tasks=%u cpu/task=%u, cpus=%u",
job->jobid, job->stepid, job->node_tasks,
job->cpus_per_task,job->cpus);
debug2(" cpu_bind_type=%4x, cpu_bind map=%s",
job->cpu_bind_type, job->cpu_bind);
if (!cpu_freq_count)
return;
if (job->cpu_bind == NULL) {
error("cpu_freq_cpuset_validate: cpu_bind string is null");
return;
}
cpu_bind = xstrdup(job->cpu_bind);
if ( (cpu_str = strtok_r(cpu_bind, ",", &savestr) ) == NULL) {
error("cpu_freq_cpuset_validate: cpu_bind string invalid");
xfree(cpu_bind);
return;
}
cpu_map = (bitstr_t *) bit_alloc(cpu_freq_count);
cpus_to_set = (bitstr_t *) bit_alloc(cpu_freq_count);
do {
debug3(" cpu_str = %s", cpu_str);
if ((job->cpu_bind_type & CPU_BIND_MAP) == CPU_BIND_MAP) {
cpu_num = atoi(cpu_str);
if (cpu_num >= cpu_freq_count) {
error("cpu_freq_cpuset_validate: invalid cpu "
"number %d", cpu_num);
bit_free(cpu_map);
bit_free(cpus_to_set);
xfree(cpu_bind);
return;
}
bit_set(cpu_map, (bitoff_t)cpu_num);
} else {
if (bit_unfmt_hexmask(cpu_map, cpu_str) == -1) {
error("cpu_freq_cpuset_validate: invalid cpu "
"mask %s", cpu_bind);
bit_free(cpu_map);
bit_free(cpus_to_set);
xfree(cpu_bind);
return;
}
}
bit_or(cpus_to_set, cpu_map);
} while ( (cpu_str = strtok_r(NULL, ",", &savestr) ) != NULL);
for (cpuidx=0; cpuidx < cpu_freq_count; cpuidx++) {
if (bit_test(cpus_to_set, cpuidx)) {
_cpu_freq_find_valid(job->cpu_freq, cpuidx);
}
}
cpu_freq_set(job);
bit_free(cpu_map);
bit_free(cpus_to_set);
xfree(cpu_bind);
return;
}
