Skip to contentSkip to conversation

cyclictest overflow instance tracking -- irc continued

DISCUSSION John Kacur 2 participants 4 messages Oct 16, 2012 1 message not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

John Kacur

author
#1 lore raw

On Mon, 15 Oct 2012, Frank Rowand wrote:

Thanks for your further review Frank. Bhavesh - I think the right thing to do here, is to ask you to respin your patch. Can you make sure you are applying and testing against the latest. (0.84). Also, please put any white space clean-ups in a separate patch to make it easier to read the important patch. Oh, and send these in a new thread. If people could preface their mail with [PATCH RT-TESTS] that would also help.

Thank You.

John Kacur

Quoted 53 lines

16:13 jkacur frowand_bot: would you like a Reviewed-by: credit for the cyclictest histogram overflow instance tracking patch?
16:13 jkacur can I just add that or is the reviewer supposed to request it?
17:02 frowand jkacur: sure, you can add my reviewed-by (at least for the first version, i didn't really look at the second version in detail)

For the first version, it was a drive-by review, just obvious stuff.

Now that I look at the second version a little closer (but still not in detail), it looks like he did some white space clean up, which should probably be a separate patch.

It also looks like he created his patch against an old version of cyclictest, because he has:

(quoted diff) +1 -0
@@ -154,6 +155,7 @@ struct thread_stat {
 	long redmax;
 	long cycleofmax;
 	long hist_overflow;
+        long num_outliers;
 };
 
 

but struct thread_stat currently has another field:

        long redmax;
        long cycleofmax;
        long hist_overflow;
        unsigned long now_after_next;
};

and also the line numbers in the patch are off.

Quoted 6 lines

from #linux-rt:

17:07 jkacur frowand: hmm, the patch doesn't appear to be working for me
17:07 jkacur I'll look at it more tomorrow
17:08 jkacur Thread 0: 18446744073709551615 00000 00001 00002 00003 00004 00005 00006 00007 00008 # 02841 others
17:08 jkacur is an example of broken output

18446744073709551615 is 0xffffffffffffffff (or -1LL)

It looks like in timerthread(), the line:

(quoted diff) +1 -0
+					stat->outliers[stat->num_outliers++] = stat->cycles - 1;

is picking up stat->cycles == zero. Looking at the current version of cyclictest, for the first event cycles should already be one at this point. But given that the line numbers in the patch are not consistent with the current cyclictest, I don't know where that chunk actually got inserted.

-Frank

#2 lore raw

Sorry about that John, Frank! Don't know how that happened…

Please look for two separate patches against the right ToT: one for whitespace cleanup, and the other with this feature, in the next few hours…

Thanks

- Bhavesh

Quoted 65 lines · John Kacur

On Mon, 15 Oct 2012, Frank Rowand wrote:

Thanks for your further review Frank. Bhavesh - I think the right thing to do here, is to ask you to respin your patch. Can you make sure you are applying and testing against the latest. (0.84). Also, please put any white space clean-ups in a separate patch to make it easier to read the important patch. Oh, and send these in a new thread. If people could preface their mail with [PATCH RT-TESTS] that would also help.

Thank You.

John Kacur

Quoted 50 lines

16:13 jkacur frowand_bot: would you like a Reviewed-by: credit for the cyclictest histogram overflow instance tracking patch?
16:13 jkacur can I just add that or is the reviewer supposed to request it?
17:02 frowand jkacur: sure, you can add my reviewed-by (at least for the first version, i didn't really look at the second version in detail)

For the first version, it was a drive-by review, just obvious stuff.

Now that I look at the second version a little closer (but still not in detail), it looks like he did some white space clean up, which should probably be a separate patch.

It also looks like he created his patch against an old version of cyclictest, because he has:

(quoted diff) +1 -0
@@ -154,6 +155,7 @@ struct thread_stat {
 	long redmax;
 	long cycleofmax;
 	long hist_overflow;
+        long num_outliers;
 };
 
 

but struct thread_stat currently has another field:

       long redmax;
       long cycleofmax;
       long hist_overflow;
       unsigned long now_after_next;
};

and also the line numbers in the patch are off.

Quoted 6 lines

from #linux-rt:

17:07 jkacur frowand: hmm, the patch doesn't appear to be working for me
17:07 jkacur I'll look at it more tomorrow
17:08 jkacur Thread 0: 18446744073709551615 00000 00001 00002 00003 00004 00005 00006 00007 00008 # 02841 others
17:08 jkacur is an example of broken output

18446744073709551615 is 0xffffffffffffffff (or -1LL)

It looks like in timerthread(), the line:

(quoted diff) +1 -0
+					stat->outliers[stat->num_outliers++] = stat->cycles - 1;

is picking up stat->cycles == zero. Looking at the current version of cyclictest, for the first event cycles should already be one at this point. But given that the line numbers in the patch are not consistent with the current cyclictest, I don't know where that chunk actually got inserted.

Quoted 3 lines
Quoted 3 lines

-Frank

--
Bhavesh Davda
[email protected]

--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#3 lore raw

Also, just to confirm: where is the GIT tree located?

I was sync'ed to https://github.com/clrkwllms/rt-tests.git

But I also see a tree at http://git.kernel.org/pub/scm/linux/kernel/git/clrkwllms/rt-tests.git

Thanks

Quoted 75 lines · Bhavesh Davda

Sorry about that John, Frank! Don't know how that happened…

Please look for two separate patches against the right ToT: one for whitespace cleanup, and the other with this feature, in the next few hours…

Thanks

- Bhavesh

Quoted 65 lines · John Kacur

On Mon, 15 Oct 2012, Frank Rowand wrote:

Thanks for your further review Frank. Bhavesh - I think the right thing to do here, is to ask you to respin your patch. Can you make sure you are applying and testing against the latest. (0.84). Also, please put any white space clean-ups in a separate patch to make it easier to read the important patch. Oh, and send these in a new thread. If people could preface their mail with [PATCH RT-TESTS] that would also help.

Thank You.

John Kacur

Quoted 50 lines

16:13 jkacur frowand_bot: would you like a Reviewed-by: credit for the cyclictest histogram overflow instance tracking patch?
16:13 jkacur can I just add that or is the reviewer supposed to request it?
17:02 frowand jkacur: sure, you can add my reviewed-by (at least for the first version, i didn't really look at the second version in detail)

For the first version, it was a drive-by review, just obvious stuff.

Now that I look at the second version a little closer (but still not in detail), it looks like he did some white space clean up, which should probably be a separate patch.

It also looks like he created his patch against an old version of cyclictest, because he has:

(quoted diff) +1 -0
@@ -154,6 +155,7 @@ struct thread_stat {
 	long redmax;
 	long cycleofmax;
 	long hist_overflow;
+        long num_outliers;
 };
 
 

but struct thread_stat currently has another field:

      long redmax;
      long cycleofmax;
      long hist_overflow;
      unsigned long now_after_next;
};

and also the line numbers in the patch are off.

Quoted 6 lines

from #linux-rt:

17:07 jkacur frowand: hmm, the patch doesn't appear to be working for me
17:07 jkacur I'll look at it more tomorrow
17:08 jkacur Thread 0: 18446744073709551615 00000 00001 00002 00003 00004 00005 00006 00007 00008 # 02841 others
17:08 jkacur is an example of broken output

18446744073709551615 is 0xffffffffffffffff (or -1LL)

It looks like in timerthread(), the line:

(quoted diff) +1 -0
+					stat->outliers[stat->num_outliers++] = stat->cycles - 1;

is picking up stat->cycles == zero. Looking at the current version of cyclictest, for the first event cycles should already be one at this point. But given that the line numbers in the patch are not consistent with the current cyclictest, I don't know where that chunk actually got inserted.

Quoted 13 lines
Quoted 3 lines
Quoted 3 lines

-Frank

--
Bhavesh Davda
[email protected]

--
Bhavesh Davda
[email protected]

--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html