[PATCH] scripts/bootgraph.pl: ignore times until the clock is running

[PATCH] scripts/bootgraph.pl: ignore times until the clock is running

[Aaro Koskinen]

The clock used for printk.time=1 does not necessarily start running
from zero. On such platforms the bootgraph does not work properly as
the calculated boottime will be too long.

To correct this, the patch modifies the script to accept only non-zero
start times. This ensures the report contains entries only from the
period when the clock was running.

Signed-off-by: Aaro Koskinen <aaro.koskinen@nokia.com>
---
 scripts/bootgraph.pl |   18 +++++++++++-------
 1 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/scripts/bootgraph.pl b/scripts/bootgraph.pl
index 12caa82..ec57dd4 100644
--- a/scripts/bootgraph.pl
+++ b/scripts/bootgraph.pl
@@ -44,7 +44,7 @@ my %end;
 my %type;
 my $done = 0;
 my $maxtime = 0;
-my $firsttime = 100;
+my $firsttime = undef;
 my $count = 0;
 my %pids;
 my %pidctr;
@@ -54,10 +54,12 @@ while (<>) {
 	if ($line =~ /([0-9\.]+)\] calling  ([a-zA-Z0-9\_\.]+)\+/) {
 		my $func = $2;
 		if ($done == 0) {
-			$start{$func} = $1;
+			my $time = $1;
+			next unless $time =~ /[1-9]/;
+			$start{$func} = $time;
 			$type{$func} = 0;
-			if ($1 < $firsttime) {
-				$firsttime = $1;
+			if (not defined $firsttime) {
+				$firsttime = $time;
 			}
 		}
 		if ($line =~ /\@ ([0-9]+)/) {
@@ -77,10 +79,12 @@ while (<>) {
 			$func = "wait_" . $pid . "_" . $pidctr{$pid};
 		}
 		if ($done == 0) {
-			$start{$func} = $1;
+			my $time = $1;
+			next unless $time =~ /[1-9]/;
+			$start{$func} = $time;
 			$type{$func} = 1;
-			if ($1 < $firsttime) {
-				$firsttime = $1;
+			if (not defined $firsttime) {
+				$firsttime = $time;
 			}
 		}
 		$pids{$func} = $pid;
-- 
1.5.6.5

Re: [PATCH] scripts/bootgraph.pl: ignore times until the clock is running

[Arjan van de Ven]

On 3/18/2010 10:49, Aaro Koskinen wrote:
> The clock used for printk.time=1 does not necessarily start running
> from zero. On such platforms the bootgraph does not work properly as
> the calculated boottime will be too long.

can we fix the platform to be smarter instead ? I'd think that's needed anyway for various
timers etc to actually work well

>
> To correct this, the patch modifies the script to accept only non-zero
> start times. This ensures the report contains entries only from the
> period when the clock was running.
>
> Signed-off-by: Aaro Koskinen<aaro.koskinen@nokia.com>

nothing wrong with the patch itself... just sad that it's needed to work around such bugs ;(

Acked-by: Arjan van de Ven <arjan@linux.intel.com>

Re: [PATCH] scripts/bootgraph.pl: ignore times until the clock is running

[Aaro Koskinen]

Hi,

Arjan van de Ven wrote:
> On 3/18/2010 10:49, Aaro Koskinen wrote:
>> The clock used for printk.time=1 does not necessarily start running
>> from zero. On such platforms the bootgraph does not work properly as
>> the calculated boottime will be too long.
> 
> can we fix the platform to be smarter instead ? I'd think that's needed anyway for various
> timers etc to actually work well

Hmm, on OMAP I have gotten so used to this "feature" that I never
thought it was a bug. However, I don't think it will affect anything
else than printk time stamps.

Anyway, I sent a fix proposal to linux-omap list:
http://marc.info/?l=linux-omap&m=126900738712041&w=2

>> To correct this, the patch modifies the script to accept only non-zero
>> start times. This ensures the report contains entries only from the
>> period when the clock was running.
>>
>> Signed-off-by: Aaro Koskinen<aaro.koskinen@nokia.com>
> 
> nothing wrong with the patch itself... just sad that it's needed to work around such bugs ;(
> 
> Acked-by: Arjan van de Ven <arjan@linux.intel.com>

I guess we can forget this patch if we agree the platform behaviour is
not correct.

A.