proc/meminfo.h \
proc/namespace.c \
proc/namespace.h \
+ proc/numa.c \
+ proc/numa.h \
proc/pids.c \
proc/pids.h \
proc/procps.h \
proc/devname.h \
proc/diskstats.h \
proc/escape.h \
+ proc/numa.h \
proc/pids.h \
proc/procps.h \
proc/pwcache.h \
--- /dev/null
+/*
+ * NUMA node support for <PIDS> & <STAT> interfaces
+ * Copyright 2017 by James C. Warmer
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef NUMA_DISABLE
+#include <dlfcn.h>
+#endif
+#include <stdlib.h>
+
+#include "numa.h"
+
+/*
+ * We're structured so that if numa_init() is NOT called or that ./configure |
+ * --disable-numa WAS specified, then calls to both of our primary functions |
+ * of numa_max_node() plus numa_node_of_cpu() would always return a negative |
+ * 1 which signifies that NUMA information isn't available. That ./configure |
+ * option might be required when libdl.so (necessary for dlopen) is missing. |
+ */
+
+
+/* ------------------------------------------------------------------------- +
+ a strictly development #define, existing specifically for the top program |
+ ( and it has no affect if ./configure --disable-numa has been specified ) | */
+//#define PRETEND_NUMA // pretend there are 3 'discontiguous' numa nodes |
+// ------------------------------------------------------------------------- +
+
+
+static int null_max_node (void) { return -1; }
+static int null_node_of_cpu (int n) { (void)n; return -1; }
+
+
+#ifndef NUMA_DISABLE
+ #ifdef PRETEND_NUMA
+static int fake_max_node (void) { return 3; }
+static int fake_node_of_cpu (int n) { return (1 == (n % 4)) ? 0 : (n % 4); }
+ #endif
+#endif
+
+
+#ifndef NUMA_DISABLE
+static void *libnuma_handle;
+#endif
+int (*numa_max_node) (void) = null_max_node;
+int (*numa_node_of_cpu) (int) = null_node_of_cpu;
+
+
+void numa_init (void) {
+ static int initialized;
+
+ if (initialized)
+ return;
+
+#ifndef NUMA_DISABLE
+ #ifndef PRETEND_NUMA
+ // we'll try for the most recent version, then a version we know works...
+ if ((libnuma_handle = dlopen("libnuma.so", RTLD_LAZY))
+ || (libnuma_handle = dlopen("libnuma.so.1", RTLD_LAZY))) {
+ numa_max_node = dlsym(libnuma_handle, "numa_max_node");
+ numa_node_of_cpu = dlsym(libnuma_handle, "numa_node_of_cpu");
+ if (numa_max_node == NULL
+ || (numa_node_of_cpu == NULL)) {
+ // this dlclose is safe - we've yet to call numa_node_of_cpu
+ // ( there's one other dlclose which has now been disabled )
+ dlclose(libnuma_handle);
+ libnuma_handle = NULL;
+ numa_max_node = null_max_node;
+ numa_node_of_cpu = null_node_of_cpu;
+ }
+ }
+ #else
+ libnuma_handle = (void *)-1;
+ numa_max_node = fake_max_node;
+ numa_node_of_cpu = fake_node_of_cpu;
+ #endif
+#endif
+ initialized = 1;
+} // end: numa_init
+
+
+void numa_uninit (void) {
+#ifndef PRETEND_NUMA
+ /* note: we'll skip a dlcose() to avoid the following libnuma memory
+ * leak which is triggered after a call to numa_node_of_cpu():
+ * ==1234== LEAK SUMMARY:
+ * ==1234== definitely lost: 512 bytes in 1 blocks
+ * ==1234== indirectly lost: 48 bytes in 2 blocks
+ * ==1234== ...
+ * [ thanks very much libnuma for all the pains you have caused us ]
+ */
+// if (libnuma_handle)
+// dlclose(libnuma_handle);
+#endif
+} // end: numa_uninit
+
+
+#if defined(PRETEND_NUMA) && defined(NUMA_DISABLE)
+# warning 'PRETEND_NUMA' ignored, 'NUMA_DISABLE' is active
+#endif
--- /dev/null
+/*
+ * NUMA node support for <PIDS> & <STAT> interfaces
+ * Copyright 2017 by James C. Warmer
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef PROCPS_NUMA_H
+#define PROCPS_NUMA_H
+
+#include <features.h>
+
+__BEGIN_DECLS
+
+void numa_init (void);
+void numa_uninit (void);
+
+extern int (*numa_max_node) (void);
+extern int (*numa_node_of_cpu) (int);
+
+__END_DECLS
+
+#endif
#include <sys/types.h>
#include <proc/devname.h>
+#include <proc/numa.h>
#include <proc/readproc.h>
#include <proc/sysinfo.h>
#include <proc/uptime.h>
REG_set(OOM_SCORE, s_int, oom_score)
REG_set(PRIORITY, s_int, priority)
REG_set(PROCESSOR, u_int, processor)
+setDECL(PROCESSOR_NODE) { (void)I; R->result.s_int = numa_node_of_cpu(P->processor); }
REG_set(RSS, ul_int, rss)
REG_set(RSS_RLIM, ul_int, rss_rlim)
REG_set(RTPRIO, s_int, rtprio)
{ RS(OOM_SCORE), f_oom, NULL, QS(s_int), 0, TS(s_int) },
{ RS(PRIORITY), f_stat, NULL, QS(s_int), 0, TS(s_int) },
{ RS(PROCESSOR), f_stat, NULL, QS(u_int), 0, TS(u_int) },
+ { RS(PROCESSOR_NODE), f_stat, NULL, QS(s_int), 0, TS(s_int) },
{ RS(RSS), f_stat, NULL, QS(ul_int), 0, TS(ul_int) },
{ RS(RSS_RLIM), f_stat, NULL, QS(ul_int), 0, TS(ul_int) },
{ RS(RTPRIO), f_stat, NULL, QS(s_int), 0, TS(s_int) },
procps_uptime(&uptime_secs, NULL);
p->boot_seconds = uptime_secs;
+ numa_init();
+
p->fetch.results.counts = &p->fetch.counts;
p->refcount = 1;
free((*info)->hist);
}
+ numa_uninit();
+
free(*info);
*info = NULL;
return 0;
PIDS_OOM_SCORE, // s_int
PIDS_PRIORITY, // s_int
PIDS_PROCESSOR, // u_int
+ PIDS_PROCESSOR_NODE, // s_int
PIDS_RSS, // ul_int
PIDS_RSS_RLIM, // ul_int
PIDS_RTPRIO, // s_int
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#ifndef NUMA_DISABLE
-#include <dlfcn.h>
-#endif
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <sys/stat.h>
#include <sys/types.h>
+#include <proc/numa.h>
#include <proc/sysinfo.h>
#include <proc/procps-private.h>
#define STACKS_INCR 32 // amount reap stack allocations grow
#define NEWOLD_INCR 32 // amount jiffs hist allocations grow
-/* ------------------------------------------------------------------------- +
- a strictly development #define, existing specifically for the top program |
- ( and it has no affect if ./configure --disable-numa has been specified ) | */
-//#define PRETEND_NUMA // pretend there are 3 'discontiguous' numa nodes |
-// ------------------------------------------------------------------------- +
-
/* ------------------------------------------------------------------------- +
because 'reap' would be forced to duplicate the global SYS stuff in every |
TIC type results stack, the following #define can be used to enforce that |
struct ext_support cpu_summary; // supports /proc/stat line #1 results
struct ext_support select; // support for 'procps_stat_select()'
struct stat_reaped results; // for return to caller after a reap
-#ifndef NUMA_DISABLE
- void *libnuma_handle; // if dlopen() for libnuma succeessful
- int (*our_max_node)(void); // a libnuma function call via dlsym()
- int (*our_node_of_cpu)(int); // a libnuma function call via dlsym()
-#endif
struct stat_result get_this; // for return to caller after a get
struct item_support reap_items; // items used for reap (shared among 3)
struct item_support select_items; // items unique to select
// ___ Private Functions ||||||||||||||||||||||||||||||||||||||||||||||||||||||
-#ifndef NUMA_DISABLE
- #ifdef PRETEND_NUMA
-static int fake_max_node (void) { return 3; }
-static int fake_node_of_cpu (int n) { return (1 == (n % 4)) ? 0 : (n % 4); }
- #endif
-#endif
-
-
static inline void stat_assign_results (
struct stat_stack *stack,
struct hist_sys *sys_hist,
static int stat_make_numa_hist (
struct stat_info *info)
{
-#ifndef NUMA_DISABLE
struct hist_tic *cpu_ptr, *nod_ptr;
int i, node;
- if (info->libnuma_handle == NULL)
- return 0;
-
/* are numa nodes dynamic like online cpus can be?
( and be careful, this libnuma call returns the highest node id in use, )
( NOT an actual number of nodes - some of those 'slots' might be unused ) */
- info->nodes.total = info->our_max_node() + 1;
+ if (!(info->nodes.total = numa_max_node() + 1))
+ return 0;
if (info->nodes.hist.n_alloc == 0
|| (info->nodes.total >= info->nodes.hist.n_alloc)) {
// spin thru each cpu and value the jiffs for it's numa node
for (i = 0; i < info->cpus.hist.n_inuse; i++) {
cpu_ptr = info->cpus.hist.tics + i;
- if (-1 < (node = info->our_node_of_cpu(cpu_ptr->id))) {
+ if (-1 < (node = numa_node_of_cpu(cpu_ptr->id))) {
nod_ptr = info->nodes.hist.tics + node;
nod_ptr->new.user += cpu_ptr->new.user; nod_ptr->old.user += cpu_ptr->old.user;
nod_ptr->new.nice += cpu_ptr->new.nice; nod_ptr->old.nice += cpu_ptr->old.nice;
}
info->nodes.hist.n_inuse = info->nodes.total;
return info->nodes.hist.n_inuse;
-#else
- return 0;
-#endif
} // end: stat_make_numa_hist
} // end: stat_update_single_stack
-#if defined(PRETEND_NUMA) && defined(NUMA_DISABLE)
-# warning 'PRETEND_NUMA' ignored, 'NUMA_DISABLE' is active
-#endif
-
// ___ Public Functions |||||||||||||||||||||||||||||||||||||||||||||||||||||||
// the select guy has its own set of items
p->select.items = &p->select_items;
-#ifndef NUMA_DISABLE
- #ifndef PRETEND_NUMA
- // we'll try for the most recent version, then a version we know works...
- if ((p->libnuma_handle = dlopen("libnuma.so", RTLD_LAZY))
- || (p->libnuma_handle = dlopen("libnuma.so.1", RTLD_LAZY))) {
- p->our_max_node = dlsym(p->libnuma_handle, "numa_max_node");
- p->our_node_of_cpu = dlsym(p->libnuma_handle, "numa_node_of_cpu");
- if (p->our_max_node == NULL
- || (p->our_node_of_cpu == NULL)) {
- // this dlclose is safe - we've yet to call numa_node_of_cpu
- // ( there's one other dlclose which has now been disabled )
- dlclose(p->libnuma_handle);
- p->libnuma_handle = NULL;
- }
- }
- #else
- p->libnuma_handle = (void *)-1;
- p->our_max_node = fake_max_node;
- p->our_node_of_cpu = fake_node_of_cpu;
- #endif
-#endif
+ numa_init();
/* do a priming read here for the following potential benefits: |
1) ensure there will be no problems with subsequent access |
if ((*info)->select_items.enums)
free((*info)->select_items.enums);
-#ifndef NUMA_DISABLE
- #ifndef PRETEND_NUMA
- /* note: we'll skip a dlcose() to avoid the following libnuma memory
- * leak which is triggered after a call to numa_node_of_cpu():
- * ==1234== LEAK SUMMARY:
- * ==1234== definitely lost: 512 bytes in 1 blocks
- * ==1234== indirectly lost: 48 bytes in 2 blocks
- * ==1234== ...
- * [ thanks very much libnuma, for all the pain you've caused ]
- */
-// if ((*info)->libnuma_handle)
-// dlclose((*info)->libnuma_handle);
- #endif
-#endif
+ numa_uninit();
+
free(*info);
*info = NULL;
return 0;
return NULL;
break;
case STAT_REAP_CPUS_AND_NODES:
-#ifndef NUMA_DISABLE
/* note: if we're doing numa at all, we must do this numa history |
before we build (fetch) the cpu stacks since the read_stat guy |
will have marked (temporarily) all the cpu node ids as invalid | */
return NULL;
// tolerate an unexpected absence of libnuma.so ...
stat_stacks_fetch(info, &info->nodes);
-#endif
if (!stat_stacks_fetch(info, &info->cpus))
return NULL;
break;
projects / procps-ng / commitdiff