george anzinger wrote:
> 
> The patch to fix idr is attached.  Cleans up the int/long confusion
> and also rearranges a couple of structures to honor the sizes involved.
> 
I definitly agree with you George. However, in order to avoid hardcoding
the size of an int to 32 I used a constant called BITS_PER_INT (that's
really just for cleanness):

 #endif
 
 #define IDR_MASK ((1 << IDR_BITS)-1)
 
-/* Leave the possibility of an incomplete final layer */
-#define MAX_LEVEL (BITS_PER_LONG - RESERVED_ID_BITS + IDR_BITS - 1) /
IDR_BITS
-#define MAX_ID_SHIFT (BITS_PER_LONG - RESERVED_ID_BITS)
+/* Define the size of the id's */
+#define BITS_PER_INT (sizeof(int)*8)
+
+#define MAX_ID_SHIFT (BITS_PER_INT - RESERVED_ID_BITS)
 #define MAX_ID_BIT (1 << MAX_ID_SHIFT)
 #define MAX_ID_MASK (MAX_ID_BIT - 1)
 
+/* Leave the possibility of an incomplete final layer */
+#define MAX_LEVEL (MAX_ID_SHIFT + IDR_BITS - 1) / IDR_BITS
+
 /* Number of id_layer structs to leave in free list */
 #define IDR_FREE_MAX MAX_LEVEL + MAX_LEVEL


> > Sure, except I don't have a test-program. ;-)
> >
> That is why you should visit the High-res-timers web site (see URL
> below) and get the "support" patch.  It installs in you kernel tree at
> .../Documentation/high-res-timers/ and has test programs as well as
> man pages, readme files etc.
> 
I used the test programs and they are really good to test the interface.
To have them working on IA64 I had to slightly modify the file
syscall_timer.c in the lib. This is due to the fact only a function
syscall() is used under ia64 instead of _syscall{1,2,3,4}(). I've
attached a patch that does the trick, it's a bit "raw" but it works and
that's only until glibc is upgraded to support those syscalls :-)

Eric