[openib-general] Re: [PATCH] uDAPL build fix for OS vendor variations of IA64_FETCHADD

Arlin Davis ardavis at ichips.intel.com
Wed Sep 28 10:31:26 PDT 2005


James Lentini wrote:

>Hi Arlin, 
>
>A couple of questions:
>
>  
>
>>Index: dapl/udapl/linux/dapl_osd.h
>>===================================================================
>>--- dapl/udapl/linux/dapl_osd.h	(revision 3541)
>>+++ dapl/udapl/linux/dapl_osd.h	(working copy)
>>@@ -83,7 +83,6 @@
>> #include <asm/system.h>
>> #endif
>> 
>>-
>> /* Useful debug definitions */
>> #ifndef STATIC
>> #define STATIC static
>>@@ -156,13 +155,17 @@
>> #ifdef __ia64__
>>     DAT_COUNT	old_value;
>> 
>>-#if OS_VERSION >= LINUX_VERSION(2,6)
>>-    IA64_FETCHADD  (old_value,v,1,4,rel);
>>+#ifndef REDHAT_EL4
>>+#   if OS_RELEASE >= LINUX_VERSION(2,6)
>>+	IA64_FETCHADD(old_value,v,1,4,rel);
>>+#   else
>>+	IA64_FETCHADD(old_value,v,1,4);
>>+#   endif
>> #else
>>-    IA64_FETCHADD  (old_value,v,1,4);
>>+   IA64_FETCHADD(old_value,v,1,4);
>> #endif
>>    
>>
>
>Previously, if we were on Linux => 2.6, we used the 5 parameter 
>version, otherwise we used the 4 parameter version. 
>
>Why don't we continue to use the 5 parameter version if we are on 
>Linux => 2.6 and not REHHAT_EL4?
>  
>
good point. something like this ?

#if !defined(REDHAT_EL4) && (OS_RELEASE >= LINUX_VERSION(2,6))
        IA64_FETCHADD(old_value,v,-1,4,rel);
#   else
        IA64_FETCHADD(old_value,v,-1,4);
#endif


>  
>
>> 
>>-#else  /* !__ia64__ */
>>+#else
>>    
>>
>
>Why remove /* !__ia64__ */?
>
>  
>
no reason. comment should stay.

>>     __asm__ __volatile__ (
>> 	"lock;" "incl %0"
>> 	:"=m" (*v)
>>@@ -184,13 +187,17 @@
>> #ifdef __ia64__
>>     DAT_COUNT	old_value;
>> 
>>-#if OS_VERSION >= LINUX_VERSION(2,6)
>>-    IA64_FETCHADD  (old_value,v,-1,4,rel);
>>+#ifndef REDHAT_EL4
>>+#   if OS_RELEASE >= LINUX_VERSION(2,6)
>>+	IA64_FETCHADD(old_value,v,-1,4,rel);
>>+#   else
>>+	IA64_FETCHADD(old_value,v,-1,4);
>>+#   endif
>> #else
>>-    IA64_FETCHADD  (old_value,v,-1,4);
>>+   IA64_FETCHADD(old_value,v,-1,4);
>>    
>>
>
>Why not continue to use the 5 parameter version if we are on
>Linux => 2.6 and not REHHAT_EL4?
>  
>

same as above





More information about the general mailing list