quoted

quoted

On Wed, Jun 4, 2008 at  5:09 PM, in message

[ref], Peter Zijlstra
[off-list ref] wrote: 

On Wed, 2008-06-04 at 15:04 -0400, Gregory Haskins wrote:

quoted

A rounding error was pointed out by Peter Zijlstra which would result
in the structure holding priorities to be off by one.

Signed-off-by: Gregory Haskins <redacted>
CC: Peter Zijlstra <peterz@infradead.org>
---

 kernel/sched_cpupri.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/sched_cpupri.h b/kernel/sched_cpupri.h
index 16d29b9..817c55c 100644
--- a/kernel/sched_cpupri.h
+++ b/kernel/sched_cpupri.h

@@ -4,7 +4,7 @@
 #include <linux/sched.h>
 
 #define CPUPRI_NR_PRIORITIES 2+MAX_RT_PRIO
-#define CPUPRI_NR_PRI_WORDS CPUPRI_NR_PRIORITIES/BITS_PER_LONG
+#define CPUPRI_NR_PRI_WORDS (CPUPRI_NR_PRIORITIES + 

BITS_PER_LONG/2)/BITS_PER_LONG

(33 + 16) / 32 = 49 / 32 = 1

So its still wrong ;-)

Please use DECLARE_BITMAP and or BITS_TO_LONGS to avoid these issues.

Indeed.  Here is a new version:

(Thanks, Peter!)

------

    sched: fix cpupri priocount

    A rounding error was pointed out by Peter Zijlstra which would result
    in the structure holding priorities to be off by one.

    Signed-off-by: Gregory Haskins [off-list ref]
    CC: Peter Zijlstra [off-list ref]

diff --git a/kernel/sched_cpupri.h b/kernel/sched_cpupri.h
index 16d29b9..51e6553 100644
--- a/kernel/sched_cpupri.h
+++ b/kernel/sched_cpupri.h

@@ -4,7 +4,7 @@
 #include <linux/sched.h>

 #define CPUPRI_NR_PRIORITIES 2+MAX_RT_PRIO
-#define CPUPRI_NR_PRI_WORDS CPUPRI_NR_PRIORITIES/BITS_PER_LONG
+#define CPUPRI_NR_PRI_WORDS BITS_TO_LONGS(CPUPRI_NR_PRIORITIES)

 #define CPUPRI_INVALID -1
 #define CPUPRI_IDLE     0