Commit 4327bed2 authored by Philip Cox's avatar Philip Cox Committed by Alex Deucher

drm/amdkfd: Add process eviction counters to sysfs

Add per-process eviction counters to sysfs to keep track of
how many eviction events have happened for each process.

v2: rename the stats dir, and track all evictions per process, per device.
v3: Simplify the stats kobject handling and cleanup.
v4: more code cleanup
Signed-off-by: default avatarPhilip Cox <Philip.Cox@amd.com>
Reviewed-by: default avatarFelix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent 8a491bb3
...@@ -653,6 +653,7 @@ static int evict_process_queues_nocpsch(struct device_queue_manager *dqm, ...@@ -653,6 +653,7 @@ static int evict_process_queues_nocpsch(struct device_queue_manager *dqm,
pr_info_ratelimited("Evicting PASID 0x%x queues\n", pr_info_ratelimited("Evicting PASID 0x%x queues\n",
pdd->process->pasid); pdd->process->pasid);
pdd->last_evict_timestamp = get_jiffies_64();
/* Mark all queues as evicted. Deactivate all active queues on /* Mark all queues as evicted. Deactivate all active queues on
* the qpd. * the qpd.
*/ */
...@@ -714,6 +715,7 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm, ...@@ -714,6 +715,7 @@ static int evict_process_queues_cpsch(struct device_queue_manager *dqm,
q->properties.is_active = false; q->properties.is_active = false;
decrement_queue_count(dqm, q->properties.type); decrement_queue_count(dqm, q->properties.type);
} }
pdd->last_evict_timestamp = get_jiffies_64();
retval = execute_queues_cpsch(dqm, retval = execute_queues_cpsch(dqm,
qpd->is_debug ? qpd->is_debug ?
KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES : KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES :
...@@ -732,6 +734,7 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm, ...@@ -732,6 +734,7 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm,
struct mqd_manager *mqd_mgr; struct mqd_manager *mqd_mgr;
struct kfd_process_device *pdd; struct kfd_process_device *pdd;
uint64_t pd_base; uint64_t pd_base;
uint64_t eviction_duration;
int retval, ret = 0; int retval, ret = 0;
pdd = qpd_to_pdd(qpd); pdd = qpd_to_pdd(qpd);
...@@ -799,6 +802,8 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm, ...@@ -799,6 +802,8 @@ static int restore_process_queues_nocpsch(struct device_queue_manager *dqm,
ret = retval; ret = retval;
} }
qpd->evicted = 0; qpd->evicted = 0;
eviction_duration = get_jiffies_64() - pdd->last_evict_timestamp;
atomic64_add(eviction_duration, &pdd->evict_duration_counter);
out: out:
if (mm) if (mm)
mmput(mm); mmput(mm);
...@@ -812,6 +817,7 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm, ...@@ -812,6 +817,7 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm,
struct queue *q; struct queue *q;
struct kfd_process_device *pdd; struct kfd_process_device *pdd;
uint64_t pd_base; uint64_t pd_base;
uint64_t eviction_duration;
int retval = 0; int retval = 0;
pdd = qpd_to_pdd(qpd); pdd = qpd_to_pdd(qpd);
...@@ -845,6 +851,8 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm, ...@@ -845,6 +851,8 @@ static int restore_process_queues_cpsch(struct device_queue_manager *dqm,
retval = execute_queues_cpsch(dqm, retval = execute_queues_cpsch(dqm,
KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0); KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES, 0);
qpd->evicted = 0; qpd->evicted = 0;
eviction_duration = get_jiffies_64() - pdd->last_evict_timestamp;
atomic64_add(eviction_duration, &pdd->evict_duration_counter);
out: out:
dqm_unlock(dqm); dqm_unlock(dqm);
return retval; return retval;
......
...@@ -631,7 +631,7 @@ enum kfd_pdd_bound { ...@@ -631,7 +631,7 @@ enum kfd_pdd_bound {
PDD_BOUND_SUSPENDED, PDD_BOUND_SUSPENDED,
}; };
#define MAX_SYSFS_FILENAME_LEN 11 #define MAX_SYSFS_FILENAME_LEN 15
/* /*
* SDMA counter runs at 100MHz frequency. * SDMA counter runs at 100MHz frequency.
...@@ -692,6 +692,13 @@ struct kfd_process_device { ...@@ -692,6 +692,13 @@ struct kfd_process_device {
uint64_t sdma_past_activity_counter; uint64_t sdma_past_activity_counter;
struct attribute attr_sdma; struct attribute attr_sdma;
char sdma_filename[MAX_SYSFS_FILENAME_LEN]; char sdma_filename[MAX_SYSFS_FILENAME_LEN];
/* Eviction activity tracking */
uint64_t last_evict_timestamp;
atomic64_t evict_duration_counter;
struct attribute attr_evict;
struct kobject *kobj_stats;
}; };
#define qpd_to_pdd(x) container_of(x, struct kfd_process_device, qpd) #define qpd_to_pdd(x) container_of(x, struct kfd_process_device, qpd)
......
...@@ -344,6 +344,26 @@ static ssize_t kfd_procfs_queue_show(struct kobject *kobj, ...@@ -344,6 +344,26 @@ static ssize_t kfd_procfs_queue_show(struct kobject *kobj,
return 0; return 0;
} }
static ssize_t kfd_procfs_stats_show(struct kobject *kobj,
struct attribute *attr, char *buffer)
{
if (strcmp(attr->name, "evicted_ms") == 0) {
struct kfd_process_device *pdd = container_of(attr,
struct kfd_process_device,
attr_evict);
uint64_t evict_jiffies;
evict_jiffies = atomic64_read(&pdd->evict_duration_counter);
return snprintf(buffer,
PAGE_SIZE,
"%llu\n",
jiffies64_to_msecs(evict_jiffies));
} else
pr_err("Invalid attribute");
return 0;
}
static struct attribute attr_queue_size = { static struct attribute attr_queue_size = {
.name = "size", .name = "size",
...@@ -376,6 +396,19 @@ static struct kobj_type procfs_queue_type = { ...@@ -376,6 +396,19 @@ static struct kobj_type procfs_queue_type = {
.default_attrs = procfs_queue_attrs, .default_attrs = procfs_queue_attrs,
}; };
static const struct sysfs_ops procfs_stats_ops = {
.show = kfd_procfs_stats_show,
};
static struct attribute *procfs_stats_attrs[] = {
NULL
};
static struct kobj_type procfs_stats_type = {
.sysfs_ops = &procfs_stats_ops,
.default_attrs = procfs_stats_attrs,
};
int kfd_procfs_add_queue(struct queue *q) int kfd_procfs_add_queue(struct queue *q)
{ {
struct kfd_process *proc; struct kfd_process *proc;
...@@ -417,6 +450,58 @@ static int kfd_sysfs_create_file(struct kfd_process *p, struct attribute *attr, ...@@ -417,6 +450,58 @@ static int kfd_sysfs_create_file(struct kfd_process *p, struct attribute *attr,
return ret; return ret;
} }
static int kfd_procfs_add_sysfs_stats(struct kfd_process *p)
{
int ret = 0;
struct kfd_process_device *pdd;
char stats_dir_filename[MAX_SYSFS_FILENAME_LEN];
if (!p)
return -EINVAL;
if (!p->kobj)
return -EFAULT;
/*
* Create sysfs files for each GPU:
* - proc/<pid>/stats_<gpuid>/
* - proc/<pid>/stats_<gpuid>/evicted_ms
*/
list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
struct kobject *kobj_stats;
snprintf(stats_dir_filename, MAX_SYSFS_FILENAME_LEN,
"stats_%u", pdd->dev->id);
kobj_stats = kfd_alloc_struct(kobj_stats);
if (!kobj_stats)
return -ENOMEM;
ret = kobject_init_and_add(kobj_stats,
&procfs_stats_type,
p->kobj,
stats_dir_filename);
if (ret) {
pr_warn("Creating KFD proc/stats_%s folder failed",
stats_dir_filename);
kobject_put(kobj_stats);
goto err;
}
pdd->kobj_stats = kobj_stats;
pdd->attr_evict.name = "evicted_ms";
pdd->attr_evict.mode = KFD_SYSFS_FILE_MODE;
sysfs_attr_init(&pdd->attr_evict);
ret = sysfs_create_file(kobj_stats, &pdd->attr_evict);
if (ret)
pr_warn("Creating eviction stats for gpuid %d failed",
(int)pdd->dev->id);
}
err:
return ret;
}
static int kfd_procfs_add_sysfs_files(struct kfd_process *p) static int kfd_procfs_add_sysfs_files(struct kfd_process *p)
{ {
int ret = 0; int ret = 0;
...@@ -660,6 +745,16 @@ struct kfd_process *kfd_create_process(struct file *filep) ...@@ -660,6 +745,16 @@ struct kfd_process *kfd_create_process(struct file *filep)
if (!process->kobj_queues) if (!process->kobj_queues)
pr_warn("Creating KFD proc/queues folder failed"); pr_warn("Creating KFD proc/queues folder failed");
ret = kfd_procfs_add_sysfs_stats(process);
if (ret)
pr_warn("Creating sysfs stats dir for pid %d failed",
(int)process->lead_thread->pid);
ret = kfd_procfs_add_sysfs_stats(process);
if (ret)
pr_warn("Creating sysfs stats dir for pid %d failed",
(int)process->lead_thread->pid);
ret = kfd_procfs_add_sysfs_files(process); ret = kfd_procfs_add_sysfs_files(process);
if (ret) if (ret)
pr_warn("Creating sysfs usage file for pid %d failed", pr_warn("Creating sysfs usage file for pid %d failed",
...@@ -816,6 +911,10 @@ static void kfd_process_wq_release(struct work_struct *work) ...@@ -816,6 +911,10 @@ static void kfd_process_wq_release(struct work_struct *work)
list_for_each_entry(pdd, &p->per_device_data, per_device_list) { list_for_each_entry(pdd, &p->per_device_data, per_device_list) {
sysfs_remove_file(p->kobj, &pdd->attr_vram); sysfs_remove_file(p->kobj, &pdd->attr_vram);
sysfs_remove_file(p->kobj, &pdd->attr_sdma); sysfs_remove_file(p->kobj, &pdd->attr_sdma);
sysfs_remove_file(p->kobj, &pdd->attr_evict);
kobject_del(pdd->kobj_stats);
kobject_put(pdd->kobj_stats);
pdd->kobj_stats = NULL;
} }
kobject_del(p->kobj); kobject_del(p->kobj);
...@@ -1125,6 +1224,7 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev, ...@@ -1125,6 +1224,7 @@ struct kfd_process_device *kfd_create_process_device_data(struct kfd_dev *dev,
pdd->runtime_inuse = false; pdd->runtime_inuse = false;
pdd->vram_usage = 0; pdd->vram_usage = 0;
pdd->sdma_past_activity_counter = 0; pdd->sdma_past_activity_counter = 0;
atomic64_set(&pdd->evict_duration_counter, 0);
list_add(&pdd->per_device_list, &p->per_device_data); list_add(&pdd->per_device_list, &p->per_device_data);
/* Init idr used for memory handle translation */ /* Init idr used for memory handle translation */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment