7:19 p.m.
Sure and thanks,
(v1.13 worked fine btw)
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1191.11
Features : swp half thumb fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1
Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000
------------------------------
On Sun, May 20, 2012 11:05 AM CEST Sergey Senozhatsky wrote:
On (05/20/12 01:41), Rui DaCosta wrote:
> I get the following segfault running 2.0 from unmodified source and
> default build options in Arch Linux ARM:
> Linux alarm 3.1.10-10-ARCH #1 PREEMPT Thu May 10 02:00:37 UTC 2012
> armv5tel GNU/Linux
> Cannot load from file /var/cache/powertop/saved_results.powertop
> Cannot load from file /var/cache/powertop/saved_parameters.powertop
> Segmentation fault
> trace is:
> #0 handle_one_cpu (model=0, family=0, vendor=0xbeffcf8c ",
> number=4294967295)
weird and overflowed input.
could you please provide your /proc/cpuinfo output?
-ss
> at cpu/cpu.cpp:210
> #1 enumerate_cpus () at cpu/cpu.cpp:276
> #2 0x000495d8 in main (argc=1, argv=0xbefffcc4) at main.cpp:382
> please let me know if there's anything else I can provide to assist.
> Regards,
> RuiDC
7:34 p.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 02:19), Rui DaCosta wrote:
Sure and thanks,
(v1.13 worked fine btw)
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1191.11
Features : swp half thumb fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1
Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000
Thanks,
Well, that's the problem. Current cpu info parser doesn't understand your
cpuinfo format. It awaits for sane values on special places. For example, word
"processor" should be followed by a number, not model name.
processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 37
bogomips : 4522.66
while cpuinfo on your system is totally different.
the following is untested patch (I'm a bit skeptical) plus I don't have
ARM device for testing.
---
src/cpu/cpu.cpp | 26 ++++++++++++++------------
1 file changed, 14 insertions(+), 12 deletions(-)
diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
index 09d4a2d..143e18c 100644
--- a/src/cpu/cpu.cpp
+++ b/src/cpu/cpu.cpp
@@ -225,7 +225,7 @@ void enumerate_cpus(void)
ifstream file;
char line[1024];
- int number = -1;
+ int number = 1;
char vendor[128];
int family = 0;
int model = 0;
@@ -236,7 +236,6 @@ void enumerate_cpus(void)
return;
while (file) {
-
file.getline(line, sizeof(line));
if (strncmp(line, "vendor_id\t",10) == 0) {
char *c;
@@ -247,42 +246,45 @@ void enumerate_cpus(void)
c++;
strncpy(vendor,c, 127);
}
- }
- if (strncmp(line, "processor\t",10) == 0) {
+ } else if (strncmp(line, "processor\t",10) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
number = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "cpu family\t",11) == 0) {
+ } else if (strncmp(line, "Processor\t",10) == 0) {
+ char *c;
+ c = strchr(line, ':');
+ if (c) {
+ c++;
+ if (*c == ' ')
+ c++;
+ strncpy(vendor, c, 127);
+ }
+ } else if (strncmp(line, "cpu family\t",11) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
family = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "model\t",6) == 0) {
+ } else if (strncmp(line, "model\t",6) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
model = strtoull(c, NULL, 10);
}
- }
- if (strncasecmp(line, "bogomips\t", 9) == 0) {
+ } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
handle_one_cpu(number, vendor, family, model);
set_max_cpu(number);
}
}
-
file.close();
perf_events = new perf_power_bundle();
-
if (!perf_events->add_event("power:cpu_idle")){
perf_events->add_event("power:power_start");
perf_events->add_event("power:power_end");
9:01 p.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
Thanks,
this has gotten past the issue. I now get:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
all these except CONFIG_PERF_COUNTERS are already set, so i'll need to see if i can
get a kernel built with that option on.
Will this patch make it into trunk?
Many thanks.
________________________________
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; powertop(a)lists.01.org
Sent: Sunday, 20 May 2012, 11:34
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 02:19), Rui DaCosta wrote:
Sure and thanks,
(v1.13 worked fine btw)
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1191.11
Features : swp half thumb fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1
Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000
Thanks,
Well, that's the problem. Current cpu info parser doesn't understand your
cpuinfo format. It awaits for sane values on special places. For example, word
"processor" should be followed by a number, not model name.
processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 37
bogomips : 4522.66
while cpuinfo on your system is totally different.
the following is untested patch (I'm a bit skeptical) plus I don't have
ARM device for testing.
---
src/cpu/cpu.cpp | 26 ++++++++++++++------------
1 file changed, 14 insertions(+), 12 deletions(-)
diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
index 09d4a2d..143e18c 100644
--- a/src/cpu/cpu.cpp
+++ b/src/cpu/cpu.cpp
@@ -225,7 +225,7 @@ void enumerate_cpus(void)
ifstream file;
char line[1024];
- int number = -1;
+ int number = 1;
char vendor[128];
int family = 0;
int model = 0;
@@ -236,7 +236,6 @@ void enumerate_cpus(void)
return;
while (file) {
-
file.getline(line, sizeof(line));
if (strncmp(line, "vendor_id\t",10) == 0) {
char *c;
@@ -247,42 +246,45 @@ void enumerate_cpus(void)
c++;
strncpy(vendor,c, 127);
}
- }
- if (strncmp(line, "processor\t",10) == 0) {
+ } else if (strncmp(line, "processor\t",10) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
number = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "cpu family\t",11) == 0) {
+ } else if (strncmp(line, "Processor\t",10) == 0) {
+ char *c;
+ c = strchr(line, ':');
+ if (c) {
+ c++;
+ if (*c == ' ')
+ c++;
+ strncpy(vendor, c, 127);
+ }
+ } else if (strncmp(line, "cpu family\t",11) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
family = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "model\t",6) == 0) {
+ } else if (strncmp(line, "model\t",6) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
model = strtoull(c, NULL, 10);
}
- }
- if (strncasecmp(line, "bogomips\t", 9) == 0) {
+ } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
handle_one_cpu(number, vendor, family, model);
set_max_cpu(number);
}
}
-
file.close();
perf_events = new perf_power_bundle();
-
if (!perf_events->add_event("power:cpu_idle")){
perf_events->add_event("power:power_start");
perf_events->add_event("power:power_end");
5:32 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
it was pointed out to me that CONFIG_PERF_COUNTERS has been superseded by PERF_EVENTS
since 2.6.32 - see bottom section of
http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
so is CONFIG_PERF_COUNTERS really required?
________________________________
From: Rui DaCosta <ruidc(a)yahoo.com>
To: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; "powertop(a)lists.01.org"
<powertop(a)lists.01.org>
Sent: Sunday, 20 May 2012, 13:01
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
Thanks,
this has gotten past the issue. I now get:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
all these except CONFIG_PERF_COUNTERS are already set, so i'll need to see if i can
get a kernel built with that option on.
Will this patch make it into trunk?
Many thanks.
________________________________
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; powertop(a)lists.01.org
Sent: Sunday, 20 May 2012, 11:34
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 02:19), Rui DaCosta wrote:
Sure and thanks,
(v1.13 worked fine btw)
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1191.11
Features : swp half thumb
fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1
Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000
Thanks,
Well, that's the problem. Current cpu info parser doesn't understand your
cpuinfo format. It awaits for sane values on special places. For example, word
"processor" should be followed by a number, not model name.
processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 37
bogomips : 4522.66
while cpuinfo on your system is totally different.
the following is untested patch (I'm a bit skeptical) plus I don't have
ARM device for testing.
---
src/cpu/cpu.cpp | 26 ++++++++++++++------------
1 file changed, 14 insertions(+), 12 deletions(-)
diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
index 09d4a2d..143e18c 100644
--- a/src/cpu/cpu.cpp
+++ b/src/cpu/cpu.cpp
@@ -225,7 +225,7 @@ void enumerate_cpus(void)
ifstream file;
char line[1024];
- int number = -1;
+ int number = 1;
char vendor[128];
int family = 0;
int model = 0;
@@ -236,7 +236,6 @@ void enumerate_cpus(void)
return;
while (file) {
-
file.getline(line, sizeof(line));
if (strncmp(line, "vendor_id\t",10) == 0) {
char *c;
@@ -247,42 +246,45 @@ void enumerate_cpus(void)
c++;
strncpy(vendor,c, 127);
}
- }
- if (strncmp(line, "processor\t",10) == 0) {
+ } else if (strncmp(line, "processor\t",10) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
number = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "cpu family\t",11) == 0) {
+ } else if (strncmp(line, "Processor\t",10) == 0) {
+ char *c;
+ c = strchr(line, ':');
+ if (c) {
+ c++;
+ if (*c == ' ')
+
c++;
+ strncpy(vendor, c, 127);
+ }
+ } else if (strncmp(line, "cpu family\t",11) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
family = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "model\t",6) == 0) {
+
} else if (strncmp(line, "model\t",6) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
model = strtoull(c, NULL, 10);
}
- }
- if (strncasecmp(line, "bogomips\t", 9) == 0) {
+ } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
handle_one_cpu(number, vendor, family, model);
set_max_cpu(number);
}
}
-
file.close();
perf_events = new perf_power_bundle();
-
if (!perf_events->add_event("power:cpu_idle")){
perf_events->add_event("power:power_start");
perf_events->add_event("power:power_end");
7:57 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 04:01), Rui DaCosta wrote:
Thanks,
this has gotten past the issue. I now get:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
all these except CONFIG_PERF_COUNTERS are already set, so i'll need to see
if i can get a kernel built with that option on.
Will this patch make it into trunk?
Well, it depends. The patch itself is quite innocent -- assuming default
number of processors being 1 instead of -1 will not do any harm. Of course,
such default value could be considered as debugging friendly, yet segfault
is still no good.
We'll see what project owners think about that.
-ss
Many thanks.
──────────────────────────────────────────────────────────────────────────
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; powertop(a)lists.01.org
Sent: Sunday, 20 May 2012, 11:34
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 02:19), Rui DaCosta wrote:
> Sure and thanks,
> (v1.13 worked fine btw)
> Processor : Feroceon 88FR131 rev 1 (v5l)
> BogoMIPS : 1191.11
> Features : swp half thumb fastmult edsp
> CPU implementer : 0x56
> CPU architecture: 5TE
> CPU variant : 0x2
> CPU part : 0x131
> CPU revision : 1
>
> Hardware : Marvell SheevaPlug Reference Board
> Revision : 0000
> Serial : 0000000000000000
>
>
Thanks,
Well, that's the problem. Current cpu info parser doesn't understand your
cpuinfo format. It awaits for sane values on special places. For example,
word
"processor" should be followed by a number, not model name.
processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 37
bogomips : 4522.66
while cpuinfo on your system is totally different.
the following is untested patch (I'm a bit skeptical) plus I don't have
ARM device for testing.
---
src/cpu/cpu.cpp | 26 ++++++++++++++------------
1 file changed, 14 insertions(+), 12 deletions(-)
diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
index 09d4a2d..143e18c 100644
--- a/src/cpu/cpu.cpp
+++ b/src/cpu/cpu.cpp
@@ -225,7 +225,7 @@ void enumerate_cpus(void)
ifstream file;
char line[1024];
- int number = -1;
+ int number = 1;
char vendor[128];
int family = 0;
int model = 0;
@@ -236,7 +236,6 @@ void enumerate_cpus(void)
return;
while (file) {
-
file.getline(line, sizeof(line));
if (strncmp(line, "vendor_id\t",10) == 0) {
char *c;
@@ -247,42 +246,45 @@ void enumerate_cpus(void)
c++;
strncpy(vendor,c, 127);
}
- }
- if (strncmp(line, "processor\t",10) == 0) {
+ } else if (strncmp(line, "processor\t",10) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
number = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "cpu family\t",11) == 0) {
+ } else if (strncmp(line, "Processor\t",10) == 0) {
+ char *c;
+ c = strchr(line, ':');
+ if (c) {
+ c++;
+ if (*c == ' ')
+ c++;
+ strncpy(vendor, c, 127);
+ }
+ } else if (strncmp(line, "cpu family\t",11) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
family = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "model\t",6) == 0) {
+ } else if (strncmp(line, "model\t",6) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
model = strtoull(c, NULL, 10);
}
- }
- if (strncasecmp(line, "bogomips\t", 9) == 0) {
+ } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
handle_one_cpu(number, vendor, family, model);
set_max_cpu(number);
}
}
-
file.close();
perf_events = new perf_power_bundle();
-
if (!perf_events->add_event("power:cpu_idle")){
perf_events->add_event("power:power_start");
perf_events->add_event("power:power_end");
8:05 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 12:32), Rui DaCosta wrote:
it was pointed out to me that CONFIG_PERF_COUNTERS has been superseded by
PERF_EVENTS since 2.6.32 - see bottom section of
http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
so is CONFIG_PERF_COUNTERS really required?
Does perf tool work for you?
-ss
──────────────────────────────────────────────────────────────────────────
From: Rui DaCosta <ruidc(a)yahoo.com>
To: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; "powertop(a)lists.01.org"
<powertop(a)lists.01.org>
Sent: Sunday, 20 May 2012, 13:01
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
Thanks,
this has gotten past the issue. I now get:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
all these except CONFIG_PERF_COUNTERS are already set, so i'll need to see
if i can get a kernel built with that option on.
Will this patch make it into trunk?
Many thanks.
──────────────────────────────────────────────────────────────────────────
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; powertop(a)lists.01.org
Sent: Sunday, 20 May 2012, 11:34
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 02:19), Rui DaCosta wrote:
> Sure and thanks,
> (v1.13 worked fine btw)
> Processor : Feroceon 88FR131 rev 1 (v5l)
> BogoMIPS : 1191.11
> Features : swp half thumb fastmult edsp
> CPU implementer : 0x56
> CPU architecture: 5TE
> CPU variant : 0x2
> CPU part : 0x131
> CPU revision : 1
>
> Hardware : Marvell SheevaPlug Reference Board
> Revision : 0000
> Serial : 0000000000000000
>
>
Thanks,
Well, that's the problem. Current cpu info parser doesn't understand your
cpuinfo format. It awaits for sane values on special places. For example,
word
"processor" should be followed by a number, not model name.
processor : 2
vendor_id : GenuineIntel
cpu family : 6
model : 37
bogomips : 4522.66
while cpuinfo on your system is totally different.
the following is untested patch (I'm a bit skeptical) plus I don't have
ARM device for testing.
---
src/cpu/cpu.cpp | 26 ++++++++++++++------------
1 file changed, 14 insertions(+), 12 deletions(-)
diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
index 09d4a2d..143e18c 100644
--- a/src/cpu/cpu.cpp
+++ b/src/cpu/cpu.cpp
@@ -225,7 +225,7 @@ void enumerate_cpus(void)
ifstream file;
char line[1024];
- int number = -1;
+ int number = 1;
char vendor[128];
int family = 0;
int model = 0;
@@ -236,7 +236,6 @@ void enumerate_cpus(void)
return;
while (file) {
-
file.getline(line, sizeof(line));
if (strncmp(line, "vendor_id\t",10) == 0) {
char *c;
@@ -247,42 +246,45 @@ void enumerate_cpus(void)
c++;
strncpy(vendor,c, 127);
}
- }
- if (strncmp(line, "processor\t",10) == 0) {
+ } else if (strncmp(line, "processor\t",10) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
number = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "cpu family\t",11) == 0) {
+ } else if (strncmp(line, "Processor\t",10) == 0) {
+ char *c;
+ c = strchr(line, ':');
+ if (c) {
+ c++;
+ if (*c == ' ')
+ c++;
+ strncpy(vendor, c, 127);
+ }
+ } else if (strncmp(line, "cpu family\t",11) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
family = strtoull(c, NULL, 10);
}
- }
- if (strncmp(line, "model\t",6) == 0) {
+ } else if (strncmp(line, "model\t",6) == 0) {
char *c;
c = strchr(line, ':');
if (c) {
c++;
model = strtoull(c, NULL, 10);
}
- }
- if (strncasecmp(line, "bogomips\t", 9) == 0) {
+ } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
handle_one_cpu(number, vendor, family, model);
set_max_cpu(number);
}
}
-
file.close();
perf_events = new perf_power_bundle();
-
if (!perf_events->add_event("power:cpu_idle")){
perf_events->add_event("power:power_start");
perf_events->add_event("power:power_end");
8:17 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 12:32), Rui DaCosta wrote:
it was pointed out to me that CONFIG_PERF_COUNTERS has been superseded by
PERF_EVENTS since 2.6.32 - see bottom section of
http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
so is CONFIG_PERF_COUNTERS really required?
Could you please try the following patch?
---
src/perf/perf.cpp | 3 +++
1 file changed, 3 insertions(+)
diff --git a/src/perf/perf.cpp b/src/perf/perf.cpp
index ce9ae6a..09ca390 100644
--- a/src/perf/perf.cpp
+++ b/src/perf/perf.cpp
@@ -48,6 +48,9 @@
#ifdef __alpha__
#include <asm-generic/unistd.h>
#endif
+#ifdef __arm__
+#include <asm-generic/unistd.h>
+#endif
#include "perf.h"
#include "../lib.h"
9:09 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On 05/20/2012 02:57 PM, Sergey Senozhatsky wrote:
On (05/20/12 04:01), Rui DaCosta wrote:
> Thanks,
> this has gotten past the issue. I now get:
> PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
> as well as support for trace points in the kernel:
>
> CONFIG_PERF_EVENTS=y
> CONFIG_PERF_COUNTERS=y
> CONFIG_TRACEPOINTS=y
> CONFIG_TRACING=y
>
> all these except CONFIG_PERF_COUNTERS are already set, so i'll need to see
> if i can get a kernel built with that option on.
>
> Will this patch make it into trunk?
>
Well, it depends. The patch itself is quite innocent -- assuming default
number of processors being 1 instead of -1 will not do any harm. Of course,
such default value could be considered as debugging friendly, yet segfault
is still no good.
We'll see what project owners think about that.
*In GENERAL*
Well that is an interesting question. I personally will not be spending
any time on arm.
That said, i am also not opposed to somewhat blindly accepting patches
for ARM, especially from trusted and active members.
*As long as ARM work isn't effecting PowerTOP.*
Now if you find a bug, then that is a different story, as long as the
bug is not purely ARM specific it will get attention.
*this instance*
I will pull this patch and take a good look at it, since it looks to me
to be a bug as well. Thanks Sergey
-Chris
-ss
> Many thanks.
>
> ──────────────────────────────────────────────────────────────────────────
>
> From: Sergey Senozhatsky<sergey.senozhatsky(a)gmail.com>
> To: Rui DaCosta<ruidc(a)yahoo.com>
> Cc: Arjan van de Ven<arjan(a)linux.intel.com>; Chris Ferron
> <chris.e.ferron(a)linux.intel.com>; powertop(a)lists.01.org
> Sent: Sunday, 20 May 2012, 11:34
> Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
> On (05/20/12 02:19), Rui DaCosta wrote:
> > Sure and thanks,
> > (v1.13 worked fine btw)
> > Processor : Feroceon 88FR131 rev 1 (v5l)
> > BogoMIPS : 1191.11
> > Features : swp half thumb fastmult edsp
> > CPU implementer : 0x56
> > CPU architecture: 5TE
> > CPU variant : 0x2
> > CPU part : 0x131
> > CPU revision : 1
> >
> > Hardware : Marvell SheevaPlug Reference Board
> > Revision : 0000
> > Serial : 0000000000000000
> >
> >
>
> Thanks,
>
> Well, that's the problem. Current cpu info parser doesn't understand
your
> cpuinfo format. It awaits for sane values on special places. For example,
> word
> "processor" should be followed by a number, not model name.
>
> processor : 2
> vendor_id : GenuineIntel
> cpu family : 6
> model : 37
> bogomips : 4522.66
>
> while cpuinfo on your system is totally different.
>
> the following is untested patch (I'm a bit skeptical) plus I don't have
> ARM device for testing.
>
> ---
>
> src/cpu/cpu.cpp | 26 ++++++++++++++------------
> 1 file changed, 14 insertions(+), 12 deletions(-)
>
> diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
> index 09d4a2d..143e18c 100644
> --- a/src/cpu/cpu.cpp
> +++ b/src/cpu/cpu.cpp
> @@ -225,7 +225,7 @@ void enumerate_cpus(void)
> ifstream file;
> char line[1024];
>
> - int number = -1;
> + int number = 1;
> char vendor[128];
> int family = 0;
> int model = 0;
> @@ -236,7 +236,6 @@ void enumerate_cpus(void)
> return;
>
> while (file) {
> -
> file.getline(line, sizeof(line));
> if (strncmp(line, "vendor_id\t",10) == 0) {
> char *c;
> @@ -247,42 +246,45 @@ void enumerate_cpus(void)
> c++;
> strncpy(vendor,c, 127);
> }
> - }
> - if (strncmp(line, "processor\t",10) == 0) {
> + } else if (strncmp(line, "processor\t",10) == 0) {
> char *c;
> c = strchr(line, ':');
> if (c) {
> c++;
> number = strtoull(c, NULL, 10);
> }
> - }
> - if (strncmp(line, "cpu family\t",11) == 0) {
> + } else if (strncmp(line, "Processor\t",10) == 0) {
> + char *c;
> + c = strchr(line, ':');
> + if (c) {
> + c++;
> + if (*c == ' ')
> + c++;
> + strncpy(vendor, c, 127);
> + }
> + } else if (strncmp(line, "cpu family\t",11) == 0) {
> char *c;
> c = strchr(line, ':');
> if (c) {
> c++;
> family = strtoull(c, NULL, 10);
> }
> - }
> - if (strncmp(line, "model\t",6) == 0) {
> + } else if (strncmp(line, "model\t",6) == 0) {
> char *c;
> c = strchr(line, ':');
> if (c) {
> c++;
> model = strtoull(c, NULL, 10);
> }
> - }
> - if (strncasecmp(line, "bogomips\t", 9) == 0) {
> + } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
> handle_one_cpu(number, vendor, family, model);
> set_max_cpu(number);
> }
> }
>
> -
> file.close();
>
> perf_events = new perf_power_bundle();
> -
> if (!perf_events->add_event("power:cpu_idle")){
> perf_events->add_event("power:power_start");
> perf_events->add_event("power:power_end");
5:19 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
I'd love to help, but development in low-level languages is far from my
"forte", but I can try to submituser-level bugs and perform some testing, (
I'm merely a user of powertop and an enthusiast for low-power computing).
I understand ARM support in Linux is a mess, but nevertheless it is Linux's most
successful platform (taking mobile devices into account), and there's lots of effort
being put in to cleaning it up, though i can also understand a reluctance on Intel's
part to commit much in that direction ;)
I don't know enough about the issues here to comment on *where* the bug is, but am
happy to chase things upstream to get it fixed, if, after review, you think it is a bug
there.
From my understanding, this Kirkwood platform is also a reasonably popular ARM device,
particularly with the power-conscious users, and has been incorporated in many mass-market
devices, so i'd think patching until things are fixed upstream would be in many
people's interest - and particularly as this appears as a regression from powertop
1.13.
I could also perhaps, on arrangement, organize root access to this device for any devs
wishing to test support - so long as I don't have to go too wild with kernel changes
(I still rely upon this device as a home NAS).
But i look forward to your conclusion and respect your decision regardless.
... Great tool by the way - doing it's own massive bit to keep the world a greener
place.
Kind Regards,
Rui
________________________________
From: Chris Ferron <chris.e.ferron(a)linux.intel.com>
To: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
Cc: Rui DaCosta <ruidc(a)yahoo.com>; Arjan van de Ven <arjan(a)linux.intel.com>;
"powertop(a)lists.01.org" <powertop(a)lists.01.org>
Sent: Tuesday, 22 May 2012, 1:09
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On 05/20/2012 02:57 PM, Sergey Senozhatsky wrote:
On (05/20/12 04:01), Rui DaCosta wrote:
> Thanks,
> this has gotten past the issue. I now get:
> PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
> as well as support for trace points in the kernel:
>
> CONFIG_PERF_EVENTS=y
> CONFIG_PERF_COUNTERS=y
> CONFIG_TRACEPOINTS=y
> CONFIG_TRACING=y
>
> all these except CONFIG_PERF_COUNTERS are already set, so i'll need to see
> if i can get a kernel built with that option on.
>
> Will this patch make it into trunk?
>
Well, it depends. The patch itself is quite innocent -- assuming default
number of processors being 1 instead of -1 will not do any harm. Of course,
such default value could be considered as debugging friendly, yet segfault
is still no good.
We'll see what project owners think about that.
*In GENERAL*
Well that is an interesting question. I personally will not be spending
any time on arm.
That said, i am also not opposed to somewhat blindly accepting patches
for ARM, especially from trusted and active members.
*As long as ARM work isn't effecting PowerTOP.*
Now if you find a bug, then that is a different story, as long as the
bug is not purely ARM specific it will get attention.
*this instance*
I will pull this patch and take a good look at it, since it looks to me
to be a bug as well. Thanks Sergey
-Chris
-ss
> Many thanks.
>
> ──────────────────────────────────────────────────────────────────────────
>
> From: Sergey Senozhatsky<sergey.senozhatsky(a)gmail.com>
> To: Rui DaCosta<ruidc(a)yahoo.com>
> Cc: Arjan van de Ven<arjan(a)linux.intel.com>; Chris Ferron
> <chris.e.ferron(a)linux.intel.com>; powertop(a)lists.01.org
> Sent: Sunday, 20 May 2012, 11:34
> Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
> On (05/20/12 02:19), Rui DaCosta wrote:
> > Sure and thanks,
> > (v1.13 worked fine btw)
> > Processor : Feroceon 88FR131 rev 1 (v5l)
> > BogoMIPS : 1191.11
> > Features : swp half thumb fastmult edsp
> > CPU implementer : 0x56
> > CPU architecture: 5TE
> > CPU variant : 0x2
> > CPU part : 0x131
> > CPU revision : 1
> >
> > Hardware : Marvell SheevaPlug Reference Board
> > Revision : 0000
> > Serial : 0000000000000000
> >
> >
>
> Thanks,
>
> Well, that's the problem. Current cpu info parser doesn't understand
your
> cpuinfo format. It awaits for sane values on special places. For example,
> word
> "processor" should be followed by a number, not model name.
>
> processor : 2
> vendor_id : GenuineIntel
> cpu family : 6
> model : 37
> bogomips : 4522.66
>
> while cpuinfo on your system is totally different.
>
> the following is untested patch (I'm a bit skeptical) plus I don't have
> ARM device for testing.
>
> ---
>
> src/cpu/cpu.cpp | 26 ++++++++++++++------------
> 1 file changed, 14 insertions(+), 12 deletions(-)
>
> diff --git a/src/cpu/cpu.cpp b/src/cpu/cpu.cpp
> index 09d4a2d..143e18c 100644
> --- a/src/cpu/cpu.cpp
> +++ b/src/cpu/cpu.cpp
> @@ -225,7 +225,7 @@ void enumerate_cpus(void)
> ifstream file;
> char line[1024];
>
> - int number = -1;
> + int number = 1;
> char vendor[128];
> int family = 0;
> int model = 0;
> @@ -236,7 +236,6 @@ void enumerate_cpus(void)
> return;
>
> while (file) {
> -
> file.getline(line, sizeof(line));
> if (strncmp(line, "vendor_id\t",10) == 0) {
> char *c;
> @@ -247,42 +246,45 @@ void enumerate_cpus(void)
> c++;
> strncpy(vendor,c, 127);
> }
> - }
> - if (strncmp(line, "processor\t",10) == 0) {
> + } else if (strncmp(line, "processor\t",10) == 0) {
> char *c;
> c = strchr(line, ':');
> if (c) {
> c++;
> number = strtoull(c, NULL, 10);
> }
> - }
> - if (strncmp(line, "cpu family\t",11) == 0) {
> + } else if (strncmp(line, "Processor\t",10) == 0) {
> + char *c;
> + c = strchr(line, ':');
> + if (c) {
> + c++;
> + if (*c == ' ')
> + c++;
> + strncpy(vendor, c, 127);
> + }
> + } else if (strncmp(line, "cpu family\t",11) == 0) {
> char *c;
> c = strchr(line, ':');
> if (c) {
> c++;
> family = strtoull(c, NULL, 10);
> }
> - }
> - if (strncmp(line, "model\t",6) == 0) {
> + } else if (strncmp(line, "model\t",6) == 0) {
> char *c;
> c = strchr(line, ':');
> if (c) {
> c++;
> model = strtoull(c, NULL, 10);
> }
> - }
> - if (strncasecmp(line, "bogomips\t", 9) == 0) {
> + } else if (strncasecmp(line, "bogomips\t", 9) == 0) {
> handle_one_cpu(number, vendor, family, model);
> set_max_cpu(number);
> }
> }
>
> -
> file.close();
>
> perf_events = new perf_power_bundle();
> -
> if (!perf_events->add_event("power:cpu_idle")){
> perf_events->add_event("power:power_start");
> perf_events->add_event("power:power_end");
5:35 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
thanks, same result:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
Regards,
Rui
________________________________
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; "powertop(a)lists.01.org"
<powertop(a)lists.01.org>
Sent: Monday, 21 May 2012, 0:17
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 12:32), Rui DaCosta wrote:
it was pointed out to me that CONFIG_PERF_COUNTERS has been superseded by
PERF_EVENTS since 2.6.32 - see bottom section of
http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
so is CONFIG_PERF_COUNTERS really required?
Could you please try the following patch?
---
src/perf/perf.cpp | 3 +++
1 file changed, 3 insertions(+)
diff --git a/src/perf/perf.cpp b/src/perf/perf.cpp
index ce9ae6a..09ca390 100644
--- a/src/perf/perf.cpp
+++ b/src/perf/perf.cpp
@@ -48,6 +48,9 @@
#ifdef __alpha__
#include <asm-generic/unistd.h>
#endif
+#ifdef __arm__
+#include <asm-generic/unistd.h>
+#endif
#include "perf.h"
#include "../lib.h"
6:50 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On Sun, May 20, 2012 at 11:19 AM, Rui DaCosta <ruidc(a)yahoo.com> wrote:
Sure and thanks,
(v1.13 worked fine btw)
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1191.11
Features : swp half thumb fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1
Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000
Here perf is trying to open this file:
/sys/kernel/debug/tracing/events/power/cpu_frequency/id
with the old version it was:
/sys/kernel/debug/tracing/events/vfs/dirty_inode/id
(which doesn't exist here anymore)
Could you please look which files of those two exist on your arm device?
Greetings,
Tom
7:23 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/22/12 12:35), Rui DaCosta wrote:
thanks, same result:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
Hm...
Could you please tell the number of '__NR_perf_event_open' syscall in your
unistd.h?
-ss
Regards,
Rui
──────────────────────────────────────────────────────────────────────────
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; "powertop(a)lists.01.org"
<powertop(a)lists.01.org>
Sent: Monday, 21 May 2012, 0:17
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 12:32), Rui DaCosta wrote:
>
> it was pointed out to me that CONFIG_PERF_COUNTERS has been
superseded by
> PERF_EVENTS since 2.6.32 - see bottom section of
> [1]http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
> so is CONFIG_PERF_COUNTERS really required?
>
Could you please try the following patch?
---
src/perf/perf.cpp | 3 +++
1 file changed, 3 insertions(+)
diff --git a/src/perf/perf.cpp b/src/perf/perf.cpp
index ce9ae6a..09ca390 100644
--- a/src/perf/perf.cpp
+++ b/src/perf/perf.cpp
@@ -48,6 +48,9 @@
#ifdef __alpha__
#include <asm-generic/unistd.h>
#endif
+#ifdef __arm__
+#include <asm-generic/unistd.h>
+#endif
#include "perf.h"
#include "../lib.h"
References
Visible links
1. http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
7:58 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
nothing exists below /sys/kernel/debug
does the kernel need to be built with debug flags? or a switch flicked on somewhere?
________________________________
From: Thomas Spura <thomas.spura(a)googlemail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: powertop(a)lists.01.org
Sent: Tuesday, 22 May 2012, 22:50
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On Sun, May 20, 2012 at 11:19 AM, Rui DaCosta <ruidc(a)yahoo.com> wrote:
Sure and thanks,
(v1.13 worked fine btw)
Processor : Feroceon 88FR131 rev 1 (v5l)
BogoMIPS : 1191.11
Features : swp half thumb fastmult edsp
CPU implementer : 0x56
CPU architecture: 5TE
CPU variant : 0x2
CPU part : 0x131
CPU revision : 1
Hardware : Marvell SheevaPlug Reference Board
Revision : 0000
Serial : 0000000000000000
Here perf is trying to open this file:
/sys/kernel/debug/tracing/events/power/cpu_frequency/id
with the old version it was:
/sys/kernel/debug/tracing/events/vfs/dirty_inode/id
(which doesn't exist here anymore)
Could you please look which files of those two exist on your arm device?
Greetings,
Tom
8:08 a.m.
New subject: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
running:
find / -name unistd.h | xargs /bin/grep __NR_perf_event_open
resulted in output of:
/usr/include/asm/unistd.h:#define __NR_perf_event_open (__NR_SYSCALL_BASE+364)
/usr/include/asm-generic/unistd.h:#define __NR_perf_event_open 241
/usr/include/asm-generic/unistd.h:__SYSCALL(__NR_perf_event_open, sys_perf_event_open)
/usr/src/linux-3.1.10-11-ARCH/include/asm-generic/unistd.h:#define __NR_perf_event_open
241
/usr/src/linux-3.1.10-11-ARCH/include/asm-generic/unistd.h:__SYSCALL(__NR_perf_event_open,
sys_perf_event_open)
/usr/src/linux-3.1.10-11-ARCH/arch/arm/include/asm/unistd.h:#define __NR_perf_event_open
(__NR_SYSCALL_BASE+364)
HTH
Rui
________________________________
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; "powertop(a)lists.01.org"
<powertop(a)lists.01.org>
Sent: Tuesday, 22 May 2012, 23:23
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/22/12 12:35), Rui DaCosta wrote:
thanks, same result:
PowerTOP v2.0 needs the kernel to support the 'perf' subsystem
as well as support for trace points in the kernel:
CONFIG_PERF_EVENTS=y
CONFIG_PERF_COUNTERS=y
CONFIG_TRACEPOINTS=y
CONFIG_TRACING=y
Hm...
Could you please tell the number of '__NR_perf_event_open' syscall in your
unistd.h?
-ss
Regards,
Rui
──────────────────────────────────────────────────────────────────────────
From: Sergey Senozhatsky <sergey.senozhatsky(a)gmail.com>
To: Rui DaCosta <ruidc(a)yahoo.com>
Cc: Arjan van de Ven <arjan(a)linux.intel.com>; Chris Ferron
<chris.e.ferron(a)linux.intel.com>; "powertop(a)lists.01.org"
<powertop(a)lists.01.org>
Sent: Monday, 21 May 2012, 0:17
Subject: Re: [Powertop] segfault on Sheevaplug (ARM Kirkwood)
On (05/20/12 12:32), Rui DaCosta wrote:
>
> it was pointed out to me that CONFIG_PERF_COUNTERS has been
superseded by
> PERF_EVENTS since 2.6.32 - see bottom section of
> [1]http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
> so is CONFIG_PERF_COUNTERS really required?
>
Could you please try the following patch?
---
src/perf/perf.cpp | 3 +++
1 file changed, 3 insertions(+)
diff --git a/src/perf/perf.cpp b/src/perf/perf.cpp
index ce9ae6a..09ca390 100644
--- a/src/perf/perf.cpp
+++ b/src/perf/perf.cpp
@@ -48,6 +48,9 @@
#ifdef __alpha__
#include <asm-generic/unistd.h>
#endif
+#ifdef __arm__
+#include <asm-generic/unistd.h>
+#endif
#include "perf.h"
#include "../lib.h"
References
Visible links
1. http://cateee.net/lkddb/web-lkddb/PERF_COUNTERS.html
3771
days inactive
3773
days old
13 comments
4 participants
participants (4)
-
Chris Ferron -
Rui DaCosta -
Sergey Senozhatsky -
Thomas Spura