The inclusion of the Initialize function is not needed by most of the classes - I had included it in the LTC Xen CIM provider more for future flexibility should the need ever arise, hence most of the initialize routines do little more than print the input params to tracing. The ctx param is passed in via the addition of the (optional) provider initialize routine in the CM*MIStub macro, which in turn is defined (with ctx) in your cmpimacs.h:

#ifdef DOC_ONLY /** This macro generates the function table and initialization stub for an instance provider. The initialization routine <pn>Create_InstanceMI is called when this provider module is loaded by the broker. This macro is for CMPI providers written in plain C. @param pfx The prefix for all mandatory association provider functions. This is a character string without quotes. Mandatory functions are: <pfx>Cleanup, <pfx>EnumInstanceNames, <pfx>EnumInstances, <pfx>GetInstance, <pfx>CreateInstance, <pfx>SetInstance, <pfx>DeleteInstance and <pfx>ExecQuery. @param pn The provider name under which this provider is registered. This is a character string without quotes. @param broker The name of the broker variable used by this macro to store the CMPIBroker pointer @param hook A statement that is executed within <pn>Create_InstanceMI routine. This enables you to perform additional initialization functions and is normally a function call like furtherInit(broker) or CMNoHook. Use CMNoHook if no further intialization is required. @return The function table of this instance provider. */ CMPIInstanceMI* CMInstanceMIStub(chars pfx, chars pn, CMPIBroker *broker, statement hook); #else

#if defined(CMPI_VER_100) #define CMInstanceMIStubChange(pfx) pfx##ModifyInstance #else #define CMInstanceMIStubChange(pfx) pfx##SetInstance #endif

#define CMInstanceMIStub(pfx,pn,broker,hook) \ static CMPIInstanceMIFT instMIFT__={ \ CMPICurrentVersion, \ CMPICurrentVersion, \ "instance" #pn, \ pfx##Cleanup, \ pfx##EnumInstanceNames, \ pfx##EnumInstances, \ pfx##GetInstance, \ pfx##CreateInstance, \ CMInstanceMIStubChange(pfx), \ pfx##DeleteInstance, \ pfx##ExecQuery, \ }; \ CMPI_EXTERN_C \ CMPIInstanceMI* pn##_Create_InstanceMI(const CMPIBroker* brkr,const CMPIContext *ctx, CMPIStatus *rc) { \ static CMPIInstanceMI mi={ \ NULL, \ &instMIFT__, \ }; \ broker=brkr; \ hook; \ return &mi; \ }

#endif

Note - the CMGetCharPtr() macro is actually redefined in cmpiutil.h to better handle the case when CMPIString objects are null, which always used to cause a crash - are you sure you are picking up the redefined one in cmpiutil.h?

- Gareth

Dr. Gareth S. Bestor IBM Linux Technology Center M/S DES2-01 15300 SW Koll Parkway, Beaverton, OR 97006 503-578-3186, T/L 775-3186, Fax 503-578-3186

|---------+-----------------------------------> | | Jim Fehlig | | | <jfeh...@novell.com> | | | Sent by: | | | xen-...@lists.xe| | | nsource.com | | | | | | | | | 04/24/06 05:01 PM | |---------+-----------------------------------> >------------------------------------------------------------------------------------------------------------------| |
| | To: xen-...@lists.xensource.com
| | cc:
| | Subject: [Xen-cim] Provider Initialize functions
| >------------------------------------------------------------------------------------------------------------------|

Most of the providers have an initialize function that is called when for example the <ProviderName>_Create_InstanceMI function is executed. The initialize function is provided in calls to the various CM*MIStub macros. The second parameter of the initialize function (ctx) is not defined anywhere that I can find but is referenced during execution of the initialize function and crashes the cimom. Only one of the providers does anything useful in the initialization function - Xen_CompterSystemIndication. All of them however print some debug info and in doing so reference the bogus context and crash. I simply commented all of the offending

_SBLIM_TRACE(2, ("--- context=\"%s\"", CMGetCharPtr(CDToString(_BROKER, context, NULL))));

statements in the various initialized routines to avoid the crash.

Do we anticipate needing the initialization routines? If not, I can remove them. Xen_ComputerSystemIndication apparently needs one although I doubt it works given the bogus context parameter.

BTW, I now have all of the providers working on SLES/openwbem. The "get it working on suse/openwbem" patch will be committed to the repository after some cleanup, e.g. like the issue described above.

Regards, Jim