* [PATCH] numactl: Support dumping nodes of pages in shared memory segments
@ 2009-06-05 9:33 Andi Kleen
2009-06-05 13:54 ` Cliff Wickman
0 siblings, 1 reply; 3+ messages in thread
From: Andi Kleen @ 2009-06-05 9:33 UTC (permalink / raw)
To: cpw, linux-numa
I got a request to be able to dump the nodes individually of a shared memory
segment. This can be very verbose for interleaving on large segments, but is still useful
for some analysis.
This patch implements that with a new --dump-nodes option
Against numactl 2.0.3-rc3
-Andi
---
numactl.8 | 5 +++++
numactl.c | 13 ++++++++++---
shm.c | 31 +++++++++++++++++++++++++++++++
shm.h | 1 +
4 files changed, 47 insertions(+), 3 deletions(-)
Index: numactl-2.0.3-rc3/numactl.c
===================================================================
--- numactl-2.0.3-rc3.orig/numactl.c
+++ numactl-2.0.3-rc3/numactl.c
@@ -49,6 +49,7 @@ struct option opts[] = {
{"strict", 0, 0, 't'},
{"shmmode", 1, 0, 'M'},
{"dump", 0, 0, 'd'},
+ {"dump-nodes", 0, 0, 'D'},
{"shmid", 1, 0, 'I'},
{"huge", 0, 0, 'u'},
{"touch", 0, 0, 'T'},
@@ -67,8 +68,8 @@ void usage(void)
" numactl [--length length] [--offset offset] [--shmmode shmmode]\n"
" [--strict]\n"
" [--shmid id] --shm shmkeyfile | --file tmpfsfile\n"
- " [--huge] [--touch]\n"
- " memory policy\n"
+ " [--huge] [--touch] \n"
+ " memory policy | --dump | --dump-nodes\n"
"\n"
"memory policy is --interleave, --preferred, --membind, --localalloc\n"
"nodes is a comma delimited list of node numbers or A-B ranges or all.\n"
@@ -466,6 +467,12 @@ int main(int ac, char **av)
dump_shm();
do_dump = 1;
break;
+ case 'D': /* --dump */
+ if (shmfd < 0)
+ complain("Cannot do --dump-nodes without shared memory.\n");
+ dump_shm_nodes();
+ do_dump = 1;
+ break;
case 't': /* --strict */
did_strict = 1;
numa_set_strict(1);
@@ -520,7 +527,7 @@ int main(int ac, char **av)
fprintf(stderr,"numactl: warning. Strict flag for process ignored.\n");
if (do_dump)
- usage_msg("cannot do --dump for process");
+ usage_msg("cannot do --dump|--dump-shm for process");
if (shmoption)
usage_msg("shm related option %s for process", shmoption);
Index: numactl-2.0.3-rc3/shm.c
===================================================================
--- numactl-2.0.3-rc3.orig/shm.c
+++ numactl-2.0.3-rc3/shm.c
@@ -202,6 +202,37 @@ void dump_shm(void)
dumppol(start, c, prevpol, prevnodes);
}
+static void dumpnode(unsigned long long start, unsigned long long end, int node)
+{
+ printf("%016Lx-%016Lx: %d\n", shmoffset+start, shmoffset+end, node);
+}
+
+/* Dump nodes in a shared memory segment. */
+void dump_shm_nodes(void)
+{
+ int prevnode = -1, node;
+ unsigned long long c, start;
+
+ start = 0;
+ if (shmlen == 0) {
+ printf("nothing to dump\n");
+ return;
+ }
+
+ for (c = 0; c < shmlen; c += shm_pagesize) {
+ if (get_mempolicy(&node, NULL, 0, c+shmptr,
+ MPOL_F_ADDR|MPOL_F_NODE) < 0)
+ err("get_mempolicy on shm");
+ if (node == prevnode)
+ continue;
+ if (prevnode != -1)
+ dumpnode(start, c, prevnode);
+ prevnode = node;
+ start = c;
+ }
+ dumpnode(start, c, prevnode);
+}
+
static void vwarn(char *ptr, char *fmt, ...)
{
va_list ap;
Index: numactl-2.0.3-rc3/shm.h
===================================================================
--- numactl-2.0.3-rc3.orig/shm.h
+++ numactl-2.0.3-rc3/shm.h
@@ -8,6 +8,7 @@ extern unsigned long long shmoffset;
extern int shmflags;
extern void dump_shm(void);
+extern void dump_shm_nodes(void);
extern void attach_shared(char *);
extern void attach_sysvshm(char *);
extern void verify_shm(int policy, struct bitmask *);
Index: numactl-2.0.3-rc3/numactl.8
===================================================================
--- numactl-2.0.3-rc3.orig/numactl.8
+++ numactl-2.0.3-rc3/numactl.8
@@ -64,6 +64,8 @@ numactl \- Control NUMA policy for proce
.B \-\-touch
] [
.B \-\-dump
+] [
+.B \-\-dump-nodes
]
memory policy
.SH DESCRIPTION
@@ -225,6 +227,9 @@ is applied when an applications maps and
.B \-\-dump
Dump policy in the specified range.
.TP
+.B \-\-dump-nodes
+Dump all nodes of the specific range (very verbose!)
+.TP
Valid node specifiers
.TS
tab(:);
--
ak@linux.intel.com -- Speaking for myself only.
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] numactl: Support dumping nodes of pages in shared memory segments
2009-06-05 9:33 [PATCH] numactl: Support dumping nodes of pages in shared memory segments Andi Kleen
@ 2009-06-05 13:54 ` Cliff Wickman
2009-06-05 18:37 ` Andi Kleen
0 siblings, 1 reply; 3+ messages in thread
From: Cliff Wickman @ 2009-06-05 13:54 UTC (permalink / raw)
To: Andi Kleen; +Cc: linux-numa
Thanks Andi. I tested and applied your patch.
It's part of numactl-2.0.3-rc4.tar.gz
(ftp://oss.sgi.com/www/projects/libnuma/download/)
It's been a year(!) since the 2.0.2 release.
I think it's about time to release the current libnuma/numactl as 2.0.3.
Does anyone have other fixes or enhancements forthcoming?
-Cliff
On Fri, Jun 05, 2009 at 11:33:24AM +0200, Andi Kleen wrote:
>
> I got a request to be able to dump the nodes individually of a shared memory
> segment. This can be very verbose for interleaving on large segments, but is still useful
> for some analysis.
>
> This patch implements that with a new --dump-nodes option
>
> Against numactl 2.0.3-rc3
>
> -Andi
>
> ---
> numactl.8 | 5 +++++
> numactl.c | 13 ++++++++++---
> shm.c | 31 +++++++++++++++++++++++++++++++
> shm.h | 1 +
> 4 files changed, 47 insertions(+), 3 deletions(-)
>
> Index: numactl-2.0.3-rc3/numactl.c
> ===================================================================
> --- numactl-2.0.3-rc3.orig/numactl.c
> +++ numactl-2.0.3-rc3/numactl.c
> @@ -49,6 +49,7 @@ struct option opts[] = {
> {"strict", 0, 0, 't'},
> {"shmmode", 1, 0, 'M'},
> {"dump", 0, 0, 'd'},
> + {"dump-nodes", 0, 0, 'D'},
> {"shmid", 1, 0, 'I'},
> {"huge", 0, 0, 'u'},
> {"touch", 0, 0, 'T'},
> @@ -67,8 +68,8 @@ void usage(void)
> " numactl [--length length] [--offset offset] [--shmmode shmmode]\n"
> " [--strict]\n"
> " [--shmid id] --shm shmkeyfile | --file tmpfsfile\n"
> - " [--huge] [--touch]\n"
> - " memory policy\n"
> + " [--huge] [--touch] \n"
> + " memory policy | --dump | --dump-nodes\n"
> "\n"
> "memory policy is --interleave, --preferred, --membind, --localalloc\n"
> "nodes is a comma delimited list of node numbers or A-B ranges or all.\n"
> @@ -466,6 +467,12 @@ int main(int ac, char **av)
> dump_shm();
> do_dump = 1;
> break;
> + case 'D': /* --dump */
> + if (shmfd < 0)
> + complain("Cannot do --dump-nodes without shared memory.\n");
> + dump_shm_nodes();
> + do_dump = 1;
> + break;
> case 't': /* --strict */
> did_strict = 1;
> numa_set_strict(1);
> @@ -520,7 +527,7 @@ int main(int ac, char **av)
> fprintf(stderr,"numactl: warning. Strict flag for process ignored.\n");
>
> if (do_dump)
> - usage_msg("cannot do --dump for process");
> + usage_msg("cannot do --dump|--dump-shm for process");
>
> if (shmoption)
> usage_msg("shm related option %s for process", shmoption);
> Index: numactl-2.0.3-rc3/shm.c
> ===================================================================
> --- numactl-2.0.3-rc3.orig/shm.c
> +++ numactl-2.0.3-rc3/shm.c
> @@ -202,6 +202,37 @@ void dump_shm(void)
> dumppol(start, c, prevpol, prevnodes);
> }
>
> +static void dumpnode(unsigned long long start, unsigned long long end, int node)
> +{
> + printf("%016Lx-%016Lx: %d\n", shmoffset+start, shmoffset+end, node);
> +}
> +
> +/* Dump nodes in a shared memory segment. */
> +void dump_shm_nodes(void)
> +{
> + int prevnode = -1, node;
> + unsigned long long c, start;
> +
> + start = 0;
> + if (shmlen == 0) {
> + printf("nothing to dump\n");
> + return;
> + }
> +
> + for (c = 0; c < shmlen; c += shm_pagesize) {
> + if (get_mempolicy(&node, NULL, 0, c+shmptr,
> + MPOL_F_ADDR|MPOL_F_NODE) < 0)
> + err("get_mempolicy on shm");
> + if (node == prevnode)
> + continue;
> + if (prevnode != -1)
> + dumpnode(start, c, prevnode);
> + prevnode = node;
> + start = c;
> + }
> + dumpnode(start, c, prevnode);
> +}
> +
> static void vwarn(char *ptr, char *fmt, ...)
> {
> va_list ap;
> Index: numactl-2.0.3-rc3/shm.h
> ===================================================================
> --- numactl-2.0.3-rc3.orig/shm.h
> +++ numactl-2.0.3-rc3/shm.h
> @@ -8,6 +8,7 @@ extern unsigned long long shmoffset;
> extern int shmflags;
>
> extern void dump_shm(void);
> +extern void dump_shm_nodes(void);
> extern void attach_shared(char *);
> extern void attach_sysvshm(char *);
> extern void verify_shm(int policy, struct bitmask *);
> Index: numactl-2.0.3-rc3/numactl.8
> ===================================================================
> --- numactl-2.0.3-rc3.orig/numactl.8
> +++ numactl-2.0.3-rc3/numactl.8
> @@ -64,6 +64,8 @@ numactl \- Control NUMA policy for proce
> .B \-\-touch
> ] [
> .B \-\-dump
> +] [
> +.B \-\-dump-nodes
> ]
> memory policy
> .SH DESCRIPTION
> @@ -225,6 +227,9 @@ is applied when an applications maps and
> .B \-\-dump
> Dump policy in the specified range.
> .TP
> +.B \-\-dump-nodes
> +Dump all nodes of the specific range (very verbose!)
> +.TP
> Valid node specifiers
> .TS
> tab(:);
>
> --
> ak@linux.intel.com -- Speaking for myself only.
--
Cliff Wickman
SGI
cpw@sgi.com
(651) 683-3824
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] numactl: Support dumping nodes of pages in shared memory segments
2009-06-05 13:54 ` Cliff Wickman
@ 2009-06-05 18:37 ` Andi Kleen
0 siblings, 0 replies; 3+ messages in thread
From: Andi Kleen @ 2009-06-05 18:37 UTC (permalink / raw)
To: Cliff Wickman; +Cc: Andi Kleen, linux-numa
On Fri, Jun 05, 2009 at 08:54:25AM -0500, Cliff Wickman wrote:
>
> Thanks Andi. I tested and applied your patch.
> It's part of numactl-2.0.3-rc4.tar.gz
> (ftp://oss.sgi.com/www/projects/libnuma/download/)
>
>
>
> It's been a year(!) since the 2.0.2 release.
>
> I think it's about time to release the current libnuma/numactl as 2.0.3.
> Does anyone have other fixes or enhancements forthcoming?
I got a request to display large pages in --hardware. Haven't had time
to code that up yet. I'll be also gone for the next two weeks.
You probably don't want to delay a release for that long, i can submit it after
I'm back.
-Andi
--
ak@linux.intel.com -- Speaking for myself only.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2009-06-05 18:37 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-06-05 9:33 [PATCH] numactl: Support dumping nodes of pages in shared memory segments Andi Kleen
2009-06-05 13:54 ` Cliff Wickman
2009-06-05 18:37 ` Andi Kleen
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).