gh0s7/mckernel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

mcexec: detect mismatch of mcexec -n and mpirun -ppn

Browse Source 
Change-Id: I0c42e3119143da40ea2e69cd9ec99bde78a0ad2a
Refs: #929
This commit is contained in:
Ken Sato
2020-01-08 12:06:43 +09:00
committed by Masamichi Takagi
parent 1cfc5ca71f
commit 1d135492c3
12 changed files with 451 additions and 29 deletions
Download Patch File Download Diff File Expand all files Collapse all files

82
executer/kernel/mcctrl/control.c
View File

@ -587,13 +587,14 @@ extern int mckernel_cpu_2_linux_cpu(struct mcctrl_usrdata *udp, int cpu_id);
static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
{
struct mcctrl_usrdata *udp = ihk_host_os_get_usrdata(os);
struct mcctrl_part_exec *pe;
struct mcctrl_part_exec *pe = NULL, *pe_itr;
struct get_cpu_set_arg req;
struct mcctrl_cpu_topology *cpu_top, *cpu_top_i;
struct cache_topology *cache_top;
int cpu, cpus_assigned, cpus_to_assign, cpu_prev;
int ret = 0;
int mcexec_linux_numa;
int pe_list_len = 0;
cpumask_t *mcexec_cpu_set = NULL;
cpumask_t *cpus_used = NULL;
cpumask_t *cpus_to_use = NULL;
@ -613,24 +614,54 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
return -EINVAL;
}
pe = &udp->part_exec;
mutex_lock(&pe->lock);
if (copy_from_user(&req, (void *)arg, sizeof(req))) {
printk("%s: error copying user request\n", __FUNCTION__);
pr_err("%s: error copying user request\n", __func__);
ret = -EINVAL;
goto put_and_unlock_out;
goto put_out;
}
/* First process to enter CPU partitioning */
if (pe->nr_processes == -1) {
mutex_lock(&udp->part_exec_lock);
/* Find part_exec having same node_proxy */
list_for_each_entry_reverse(pe_itr, &udp->part_exec_list, chain) {
pe_list_len++;
if (pe_itr->node_proxy_pid == req.ppid) {
pe = pe_itr;
break;
}
}
if (!pe) {
/* First process to enter CPU partitioning */
pr_debug("%s: pe_list_len:%d\n", __func__, pe_list_len);
if (pe_list_len >= PE_LIST_MAXLEN) {
/* delete head entry of pe_list */
pe_itr = list_first_entry(&udp->part_exec_list,
struct mcctrl_part_exec, chain);
list_del(&pe_itr->chain);
kfree(pe_itr);
}
pe = kzalloc(sizeof(struct mcctrl_part_exec), GFP_KERNEL);
if (!pe) {
mutex_unlock(&udp->part_exec_lock);
ret = -ENOMEM;
goto put_out;
}
/* Init part_exec */
mutex_init(&pe->lock);
INIT_LIST_HEAD(&pe->pli_list);
pe->nr_processes = req.nr_processes;
pe->nr_processes_left = req.nr_processes;
pe->nr_processes_joined = 0;
pe->node_proxy_pid = req.ppid;
list_add_tail(&pe->chain, &udp->part_exec_list);
dprintk("%s: nr_processes: %d (partitioned exec starts)\n",
__FUNCTION__,
pe->nr_processes);
__func__, pe->nr_processes);
}
mutex_unlock(&udp->part_exec_lock);
mutex_lock(&pe->lock);
if (pe->nr_processes != req.nr_processes) {
printk("%s: error: requested number of processes"
@ -640,7 +671,15 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
goto put_and_unlock_out;
}
if (pe->nr_processes_joined >= pe->nr_processes) {
printk("%s: too many processes have joined to the group of %d\n",
__func__, req.ppid);
ret = -EINVAL;
goto put_and_unlock_out;
}
--pe->nr_processes_left;
++pe->nr_processes_joined;
dprintk("%s: nr_processes: %d, nr_processes_left: %d\n",
__FUNCTION__,
pe->nr_processes,
@ -726,8 +765,6 @@ static long mcexec_get_cpuset(ihk_os_t os, unsigned long arg)
wake_up_interruptible(&pli_next->pli_wq);
}
/* Reset process counter to start state */
pe->nr_processes = -1;
ret = -ETIMEDOUT;
goto put_and_unlock_out;
}
@ -975,16 +1012,8 @@ next_cpu:
/* Commit used cores to OS structure */
memcpy(&pe->cpus_used, cpus_used, sizeof(*cpus_used));
/* Reset if last process */
if (pe->nr_processes_left == 0) {
dprintk("%s: nr_processes: %d (partitioned exec ends)\n",
__FUNCTION__,
pe->nr_processes);
pe->nr_processes = -1;
memset(&pe->cpus_used, 0, sizeof(pe->cpus_used));
}
/* Otherwise wake up next process in list */
else {
/* If not last process, wake up next process in list */
if (pe->nr_processes_left != 0) {
++pe->process_rank;
pli_next = list_first_entry(&pe->pli_list,
struct process_list_item, list);
@ -997,11 +1026,14 @@ next_cpu:
ret = 0;
put_and_unlock_out:
mutex_unlock(&pe->lock);
put_out:
mcctrl_put_per_proc_data(ppd);
kfree(cpus_to_use);
kfree(cpus_used);
kfree(mcexec_cpu_set);
mcctrl_put_per_proc_data(ppd);
mutex_unlock(&pe->lock);
return ret;
}