proc: add amdgpu support

qa/1222

@@ -58,6 +58,8 @@ _filter()
 	-e '/^proc\.id\.container: Missing metric value(s)/d' \
 	-e '/^proc\.psinfo\.labels: No value(s) available/d' \
 	-e '/^proc\.psinfo\.ngid: Metric not supported by this version/d' \
+	-e '/^proc\.fdinfo\..*: Metric not supported/d' \
+	-e '/^proc\.fdinfo\..*: No value(s) available/d' \
 	-e '/pmdaFetch: Fetch callback error from metric PMID 3\.11\.0\[.*]: No data available/d' \
 	-e '/proc\.psinfo\.tty_pgrp: No value(s) available/d' \
 	-e '/ acct: existing pacct file did not grow /d' \
@@ -105,6 +107,7 @@ NF == 0 && seen == 1		{ if (numval == 1) print metric ": 1 value"
 	-e '/^proc\.psinfo\.ngid:/d' \
 	-e '/^proc\.psinfo\.tty_pgrp:/d' \
 	-e '/^proc\.smaps\./d' \
+	-e '/^proc\.fdinfo\./d' \
 	-e '/^Command: /s/,proc_init .*/,proc_init ... metrics .../' \
 	-e '/ERROR SUMMARY/q' \
     # end

qa/364

@@ -288,6 +288,7 @@ _filter_linux()
 {
     # pcp-atop uses metrics not supported on some kernels
     # proc.smaps.* metrics are not present for older kernels
+    # proc.fdinfo.* metrics neither
     #
     if [ $PCP_PLATFORM != linux ]
     then
@@ -310,6 +311,7 @@ _filter_linux()
 	    -e '/^proc\.psinfo\.cgroups -12351 Missing metric value(s)/d' \
 	    -e '/^proc\.namespaces\.envid -12350 Metric not supported/d' \
 	    -e '/^hotproc\.namespaces\.envid -12350 Metric not supported/d' \
+	    -e '/^proc\.fdinfo\..* Metric not supported/d' \
 	# linux
     fi
 }

src/pmdas/linux_proc/clusters.h

@@ -70,8 +70,10 @@
 #define CLUSTER_PID_AUTOGROUP		74 /* /proc/<pid>/autogroup */
 #define CLUSTER_HOTPROC_PID_AUTOGROUP	75 /* /proc/<pid>/autogroup */
 #define CLUSTER_CGROUP2_IRQ_PRESSURE	76
+#define CLUSTER_PID_FDINFO		77 /* /proc/<pid>/fdinfo */
+#define CLUSTER_HOTPROC_PID_FDINFO	78 /* /proc/<pid>/fdinfo */
 
 #define MIN_CLUSTER  8		/* first cluster number we use here */
-#define MAX_CLUSTER 77		/* one more than highest cluster number used */
+#define MAX_CLUSTER 79		/* one more than highest cluster number used */
 
 #endif /* _CLUSTERS_H */

src/pmdas/linux_proc/help

@@ -591,3 +591,4 @@ of accounting information:
   0 inactive (no information available)
   1 system (system level accounting from whatever file accton(8) is using)
   2 private (accounting records from $PCP_VAR_DIR/pmcd/pacct)
+

src/pmdas/linux_proc/help_text.h

@@ -155,4 +155,18 @@ help_text_t  help_text[] = {
 { .name = "autogroup.enabled",    .shorthelp = "Scheduling autogroup feature for CFS is enabled in the kernel",   .longhelp = "Contents of /proc/sys/kernel/sched_autogroup_enabled as described in sched(7)." },
 { .name = "autogroup.id",    .shorthelp = "Process autogroup identifier from /proc/<pid>/autogroup",   .longhelp = "Process scheduling autogroup identifier as described in sched(7)." },
 { .name = "autogroup.nice",    .shorthelp = "Process autogroup nice level from /proc/<pid>/autogroup",   .longhelp = "Process scheduling autogroup nice level as described in sched(7)." },
+
+{ .name = "fdinfo.drm_memory_cpu",    .shorthelp = "Accumulation of the drm-memory-cpu field from /proc/<pid>/fdinfo/* file descriptors",   .longhelp = "CPU memory which can be used by the GPU to store buffer objects." },
+{ .name = "fdinfo.drm_memory_gtt",    .shorthelp = "Accumulation of the drm-memory-gtt field from /proc/<pid>/fdinfo/* file descriptors",   .longhelp = "GTT memory which can be used by the GPU to store buffer objects." },
+{ .name = "fdinfo.drm_memory_vram",    .shorthelp = "Accumulation of the drm-memory-vram field from /proc/<pid>/fdinfo/* file descriptors",   .longhelp = "VRAM memory which can be used by the GPU to store buffer objects." },
+{ .name = "fdinfo.drm_shared_cpu",    .shorthelp = "Accumulation of the drm-shared-cpu field from /proc/<pid>/fdinfo/* file descriptors",   .longhelp = "CPU memory which can be used by the GPU to store buffer objects, and is shared with another file." },
+{ .name = "fdinfo.drm_shared_gtt",    .shorthelp = "Accumulation of the drm-shared-gtt field from /proc/<pid>/fdinfo/* file descriptors",   .longhelp = "GTT memory which can be used by the GPU to store buffer objects, and is shared with another file." },
+{ .name = "fdinfo.drm_shared_vram",    .shorthelp = "Accumulation of the drm-shared-vram field from /proc/<pid>/fdinfo/* file descriptors",   .longhelp = "VRAM memory which can be used by the GPU to store buffer objects, and is shared with another file." },
+
+{ .name = "fdinfo.amd_evicted_visible_vram",  .shorthelp = "Accumulation of the amd-evicted-visible-vram field from /proc/<pid>/fdinfo/* file descriptors",  .longhelp = "Sum of evicted buffers due to CPU access." },
+{ .name = "fdinfo.amd_evicted_vram",  .shorthelp = "Accumulation of the amd-evicted-vram field from /proc/<pid>/fdinfo/* file descriptors",  .longhelp = "Sum of evicted buffers, includes visible VRAM" },
+{ .name = "fdinfo.amd_memory_visible_vram",  .shorthelp = "Accumulation of the amd-memory-visible-vram field from /proc/<pid>/fdinfo/* file descriptors",  .longhelp = "Current visible VRAM usage" },
+{ .name = "fdinfo.amd_requested_gtt",  .shorthelp = "Accumulation of the amd-requested-gtt field from /proc/<pid>/fdinfo/* file descriptors",  .longhelp = "How much GTT memory userspace asked for" },
+{ .name = "fdinfo.amd_requested_visible_vram",  .shorthelp = "Accumulation of the amd-requested-visible-vram field from /proc/<pid>/fdinfo/* file descriptors",  .longhelp = "How much visible VRAM userspace asked for" },
+{ .name = "fdinfo.amd_requested_vram",  .shorthelp = "Accumulation of the amd-requested-vram field from /proc/<pid>/fdinfo/* file descriptors",  .longhelp = "How much VRAM userspace asked for, includes visible VRAM" },
 };

src/pmdas/linux_proc/pmda.c

@@ -33,6 +33,7 @@
 #include "../linux/convert.h"
 
 #include <ctype.h>
+#include <sys/syslog.h>
 #include <unistd.h>
 #include <sys/vfs.h>
 #include <sys/stat.h>
@@ -1348,6 +1349,52 @@ static pmdaMetric metrictab[] = {
 /* acct.control.state */
   { NULL, { PMDA_PMID(CLUSTER_ACCT,CONTROL_ACCT_STATE), PM_TYPE_32, PM_INDOM_NULL,
     PM_SEM_DISCRETE, PMDA_PMUNITS(0,0,0,0,0,0) }, },
+
+/*
+ * Fdinfo cluster
+ */
+
+/* proc.fdinfo.drm_memory */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,0), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.drm.memory_cpu */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,1), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.drm.memory_gtt */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,2), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.drm.memory_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,3), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.drm.shared_cpu */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,4), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.drm.shared_gtt */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,5), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.drm.shared_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,6), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+
+/* proc.fdinfo.amd.evicted_visible_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,7), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.amd.evicted_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,8), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.amd.memory_visible_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,9), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.amd.requested_gtt */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,10), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.amd.requested_visible_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,11), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+/* proc.fdinfo.amd.requested_vram */
+  { NULL, { PMDA_PMID(CLUSTER_PID_FDINFO,12), PM_TYPE_U64, PROC_INDOM,
+    PM_SEM_INSTANT, PMDA_PMUNITS(1,0,0,PM_SPACE_KBYTE,0,0)}},
+
 };
 
 pmInDom
@@ -1444,6 +1491,7 @@ proc_refresh(pmdaExt *pmda, int *need_refresh)
 	need_refresh[CLUSTER_PID_CWD] ||
 	need_refresh[CLUSTER_PID_EXE] ||
 	need_refresh[CLUSTER_PID_FD] ||
+	need_refresh[CLUSTER_PID_FDINFO] ||
 	need_refresh[CLUSTER_PROC_RUNQ]) {
 	refresh_proc_pid(&proc_pid,
 		need_refresh[CLUSTER_PROC_RUNQ]? &proc_runq : NULL,
@@ -1464,6 +1512,7 @@ proc_refresh(pmdaExt *pmda, int *need_refresh)
         need_refresh[CLUSTER_HOTPROC_PID_CWD] ||
         need_refresh[CLUSTER_HOTPROC_PID_EXE] ||
         need_refresh[CLUSTER_HOTPROC_PID_FD] ||
+        need_refresh[CLUSTER_HOTPROC_PID_FDINFO] ||
         need_refresh[CLUSTER_HOTPROC_GLOBAL] ||
         need_refresh[CLUSTER_HOTPROC_PRED]){
         refresh_hotproc_pid(&hotproc_pid,
@@ -1495,6 +1544,7 @@ proc_instance(pmInDom indom, int inst, char *name, pmInResult **result, pmdaExt
         need_refresh[CLUSTER_PID_CWD]++;
         need_refresh[CLUSTER_PID_IO]++;
         need_refresh[CLUSTER_PID_FD]++;
+        need_refresh[CLUSTER_PID_FDINFO]++;
 	break;
     case HOTPROC_INDOM:
         need_refresh[CLUSTER_HOTPROC_PID_STAT]++;
@@ -1511,6 +1561,7 @@ proc_instance(pmInDom indom, int inst, char *name, pmInResult **result, pmdaExt
         need_refresh[CLUSTER_HOTPROC_PID_FD]++;
         need_refresh[CLUSTER_HOTPROC_GLOBAL]++;
         need_refresh[CLUSTER_HOTPROC_PRED]++;
+        need_refresh[CLUSTER_HOTPROC_PID_FDINFO]++;
         break;
 
     case CGROUP_CPUSET_INDOM:
@@ -3430,6 +3481,59 @@ proc_fetchCallBack(pmdaMetric *mdesc, unsigned int inst, pmAtomValue *atom)
 	}
 	break;
 
+    case CLUSTER_HOTPROC_PID_FDINFO:
+	active_proc_pid = &hotproc_pid;
+	/*FALLTHROUGH*/
+    case CLUSTER_PID_FDINFO:
+	if (!have_access)
+	  return PM_ERR_PERMISSION;
+	if ((entry = fetch_proc_pid_fdinfo(inst, active_proc_pid, &sts)) == NULL)
+	  return sts;
+	if (!(entry->success & PROC_PID_FLAG_FDINFO))
+	  return 0;
+
+	switch (item) {
+	  case 0: /* proc.fdinfo.drm.memory_cpu */
+	    atom->ull = entry->fdinfo.drm_memory_cpu;
+	    break;
+	  case 1: /* proc.fdinfo.drm.memory_gtt */
+	    atom->ull = entry->fdinfo.drm_memory_gtt;
+	    break;
+	  case 2: /* proc.fdinfo.drm.memory_vram */
+	    atom->ull = entry->fdinfo.drm_memory_vram;
+	    break;
+	  case 3: /* proc.fdinfo.drm.shared_cpu */
+	    atom->ull = entry->fdinfo.drm_shared_cpu;
+	    break;
+	  case 4: /* proc.fdinfo.drm.shared_gtt */
+	    atom->ull = entry->fdinfo.drm_shared_gtt;
+	    break;
+	  case 5: /* proc.fdinfo.drm.shared_vram */
+	    atom->ull = entry->fdinfo.drm_shared_vram;
+	    break;
+
+	  case 6: /* proc.fdinfo.amd.evicted_visible_vram */
+	    atom->ull = entry->fdinfo.amd_evicted_visible_vram;
+	    break;
+	  case 7: /* proc.fdinfo.amd.evicted_vram */
+	    atom->ull = entry->fdinfo.amd_evicted_vram;
+	    break;
+	  case 8: /* proc.fdinfo.amd.memory_visible_vram */
+	    atom->ull = entry->fdinfo.amd_memory_visible_vram;
+	    break;
+	  case 9: /* proc.fdinfo.amd.requested_gtt */
+	    atom->ull = entry->fdinfo.amd_requested_gtt;
+	    break;
+	  case 10: /* proc.fdinfo.amd.requested_visible_vram */
+	    atom->ull = entry->fdinfo.amd_requested_visible_vram;
+	    break;
+	  case 11: /* proc.fdinfo.amd.requested_vram */
+	    atom->ull = entry->fdinfo.amd_requested_vram;
+	    break;
+	  default: /* unknown cluster */
+	    return PM_ERR_PMID;
+	}
+	break;
     default: /* unknown cluster */
 	return PM_ERR_PMID;
     }

src/pmdas/linux_proc/proc_dynamic.c

@@ -40,6 +40,7 @@ enum {
     DYNPROC_GROUP_NAMESPACE,
     DYNPROC_GROUP_SMAPS,
     DYNPROC_GROUP_AUTOGROUP,
+    DYNPROC_GROUP_FDINFO,
 
     NUM_DYNPROC_GROUPS
 };
@@ -69,6 +70,7 @@ static int proc_hotproc_cluster_list[][2] = {
 	{ CLUSTER_PID_EXE,	    CLUSTER_HOTPROC_PID_EXE },
 	{ CLUSTER_PID_CWD,	    CLUSTER_HOTPROC_PID_CWD },
 	{ CLUSTER_PID_AUTOGROUP,    CLUSTER_HOTPROC_PID_AUTOGROUP },
+	{ CLUSTER_PID_FDINFO,	    CLUSTER_HOTPROC_PID_FDINFO },
 };
 
 
@@ -257,6 +259,21 @@ static dynproc_metric_t smaps_metrics[] = {
 	{ .name = "pss_dirty",       .cluster = CLUSTER_PID_SMAPS,  .item=20 },
 };
 
+static dynproc_metric_t fdinfo_metrics[] = {
+	{ .name = "drm_memory_cpu",             .cluster = CLUSTER_PID_FDINFO, .item=0 },
+	{ .name = "drm_memory_gtt",             .cluster = CLUSTER_PID_FDINFO, .item=1 },
+	{ .name = "drm_memory_vram",            .cluster = CLUSTER_PID_FDINFO, .item=2 },
+	{ .name = "drm_shared_cpu",             .cluster = CLUSTER_PID_FDINFO, .item=3 },
+	{ .name = "drm_shared_gtt",             .cluster = CLUSTER_PID_FDINFO, .item=4 },
+	{ .name = "drm_shared_vram",            .cluster = CLUSTER_PID_FDINFO, .item=5 },
+	{ .name = "amd_evicted_visible_vram",   .cluster = CLUSTER_PID_FDINFO, .item=6 },
+	{ .name = "amd_evicted_vram",           .cluster = CLUSTER_PID_FDINFO, .item=7 },
+	{ .name = "amd_memory_visible_vram",    .cluster = CLUSTER_PID_FDINFO, .item=8 },
+	{ .name = "amd_requested_gtt",          .cluster = CLUSTER_PID_FDINFO, .item=9 },
+	{ .name = "amd_requested_visible_vram", .cluster = CLUSTER_PID_FDINFO, .item=10 },
+	{ .name = "amd_requested_vram",         .cluster = CLUSTER_PID_FDINFO, .item=11 },
+};
+
 static dynproc_group_t dynproc_groups[] = {
 	[DYNPROC_GROUP_PSINFO]    = { .name = "psinfo",	    .metrics = psinfo_metrics,	    .nmetrics = sizeof(psinfo_metrics)/sizeof(dynproc_metric_t)},
 	[DYNPROC_GROUP_ID]	  = { .name = "id",	    .metrics = id_metrics,	    .nmetrics = sizeof(id_metrics)/sizeof(dynproc_metric_t)},
@@ -267,6 +284,7 @@ static dynproc_group_t dynproc_groups[] = {
 	[DYNPROC_GROUP_NAMESPACE] = { .name = "namespaces", .metrics = namespace_metrics,   .nmetrics = sizeof(namespace_metrics)/sizeof(dynproc_metric_t) },
 	[DYNPROC_GROUP_SMAPS]     = { .name = "smaps",	    .metrics = smaps_metrics,	    .nmetrics = sizeof(smaps_metrics)/sizeof(dynproc_metric_t)},
 	[DYNPROC_GROUP_AUTOGROUP] = { .name = "autogroup", .metrics = autogroup_metrics,   .nmetrics = sizeof(autogroup_metrics)/sizeof(dynproc_metric_t) },
+	[DYNPROC_GROUP_FDINFO]    = { .name = "fdinfo",	    .metrics = fdinfo_metrics,	    .nmetrics = sizeof(fdinfo_metrics)/sizeof(dynproc_metric_t) },
 };
 
 /*