From: Renier M. <re...@us...> - 2007-01-18 15:43:52
|
Comments below. ope...@li... wrote on 01/18/2007 09:56:43 AM: > Umm, why are you comparing two different machines? What you did proves > nothing. If you had used the ipmidirect plugin on the HP machine, maybe > that would mean something. Can you try the same scenario on the same > machine with the two different plugins? They should both work in both > scenarios (except that the ipmi plugin will not create a watchdog RDR > for LAN connections). It is possible that the HP server is not properly > advertising that it has a watchdog (or maybe it doesn't have one, but > that would be bizarre). > > It is my opinion, however, that in general you should not use the > watchdog interface of either plugin. They both talk directly to the > management controller and do not go through /dev/watchdog. This is bad > for two reasons: > > 1. If you are taking kernel coredumps, the watchdog will likely reset > the system before the coredump can be taken (or in kdump, before > you can get the next kernel up). If you use the in-kernel IPMI > watchdog driver, it will extend the timeout at panic time to allow > a coredump to be taken. > 2. If you have some need to use a different watchdog (like the > watchdogs available on most southbridges now), you can't use that > interface. > > Since in highly reliable systems you should be taking kernel coredumps, > reason 1 above should be enough to avoid using the current openhpi > watchdog interface. /dev/watchdog is very simple to use. Corey, you mean to say the watchdog interface through OpenHPI IPMI plugins. I clarify, because OpenHPI has a watchdog plugin that goes directly to /dev/watchdog. > > Also, creating a watchdog RDR for a remote device over LAN is, in my > opinion, a bad idea. A watchdog is only a resource for local use. > Advertising it for a remote user is asking for trouble. For better or for worse, HPI's spec says that a resource supporting watchdogs should have a watchdog RDR regardless of it being local or remote. So the plugin needs to create it to be HPI compliant. Another way about it is that the plugin could simply *not* advertise a resource's capability of having a watchdog when the resource is remote. In that case it wouldn't need to create the watchdog RDR in the HPI model. --Renier > > -Corey > > abhi raghav wrote: > > Hello, > > > > IPMI plugin is not able to create watchdog RDR for the resource > with watchdog capability. > > > > To validate this state, > > we have crosschecked the watchdog behaviour on another > hardware(ATCA) using IPMIdirect plugin. > > > > Scenario 1: HP Hardware #1 using IPMI plugin with "smi" interface > > Scenario 2: Hardware using IPMIdirect plugin with RMCP interface > > > > Method: > > SAF Verification Testcase: > > HPI SAFTestsuite code "saftest/HPI-B.01. > 01/src/watchdogtimer/saHpiWatchdogTimerGet/3.c" > > This file contains the positive testcase for > saHpiWatchdogTimerGet, this testcases checks for the following: > > 1) The presence of Watchdog capability with the resource > > 2) If the Resource is associated with the watchdog capabilitity, > it checks for the existence of watchdog RDR. > > 3) If the watchdog RDR is present, it retrieves the watchdog > number from the RDR (which is "0") > > 4) Using the watchdog number, saHpiWatchdogTimerGet API is called. > > 5) If the API returns successfully the watchdog structure, > Testcase status is set as SAF_TEST_PASS > > If the resource fails any of the above condition, > SAF_TEST_NOTSUPPORT is returned. > > > > Scenario 1. Watchdog functionality analysis on HP hardware #1 > using IPMI plugin: > > > -------------------------------------------------------------------------------------------------------------------------- > > Software/package used on this system: > > HPI = OpenHPI.2.6.3 > > IPMI = OpenIPMI.2.0.10 > > Driver = OpenIPMI Driver > > > > Listed below is one of the resource detected with watchdog capability: > > OpenHPI> showrpt > > (017):system_board:{S|RDR|WTD|RES} > > > > RPT ID ==> 017 > > EntryId: 17 > > ResourceId: 17 > > ResourceInfo: > > ResourceRev: 0 > > SpecificVer: 0 > > DeviceSupport: 0 > > ManufacturerId: 0 > > ProductId: 0 > > FirmwareMajorRev: 0 > > FirmwareMinorRev: 0 > > AuxFirmwareRev: 0 > > Guid: > > ResourceEntity: {RACK,3}{SYSTEM_CHASSIS,1}{PHYSICAL_SLOT,1}{SYSTEM_BOARD,2} > > Capabilities: RDR | RESOURCE | SENSOR | WATCHDOG > > HotSwapCapabilities: None > > ResourceSeverity: MAJOR > > ResourceFailed: FALSE > > Tag: TEXT: ENGLISH: system_board (len=12) > > OpenHPI> rdr 017 w > > No rdrs for rpt: 17 > > Command failed: > > rdr 017 w > > OpenHPI> rdr 017 all > > (131086):SENSOR_RDR NUM=014 Ctrl=1 EvtCtrl=RM Tag=Virtual Fan > > > > when the above mentioned SAFTestsuite verification testcase is run > on this hardware, the output is as follows: > > > > *****************Resource begin****************** > > rdr_type=2, rdr_name=Virtual Fan, return=SAF_TEST_NOTSUPPORT > > resource_id=2, tag=system_board, return=SAF_TEST_NOTSUPPORT > > *****************Resource end******************** > > > > Debugging this SAFTestsuite testcase using gdb, > > it is observed that only one RDR is associated with systemboard > resource. And this RDR is of type SAHPI_SENSOR_RDR. > > > > Scenario 2: Watchdog functionality analysis on ATCA hardware using > IPMIDirect plugin: > > > --------------------------------------------------------------------------------------------------------------------------------- > > Software/package on this ATCA server blade: > > HPI = OpenHPI.2.6.3 > > IPMI = OpenIPMI.2.0.10 > > Driver = OpenIPMI Driver > > > > Listed below is one of the resource detected with watchdog capability: > > OpenHPI> showrpt > > (030):MPCBL0030:{S|RDR|INV|RST|FRU|WTD|HS|RES} > > > > RPT ID ==> 30 > > EntryId: 30 > > ResourceId: 30 > > ResourceInfo: > > ResourceRev: 0 > > SpecificVer: 0 > > DeviceSupport: 0 > > ManufacturerId: 0 > > ProductId: 0 > > FirmwareMajorRev: 0 > > FirmwareMinorRev: 0 > > AuxFirmwareRev: 0 > > Guid: > > ResourceEntity: {SYSTEM_CHASSIS,1}{PHYSICAL_SLOT,2}{160,0} > > Capabilities: FRU | INVENTORY_DATA | MANAGED_HOTSWAP | RDR | RESET > | RESOURCE | SENSOR | WATCHDOG > > HotSwapCapabilities: INDICATOR_SUPPORTED > > ResourceSeverity: OK > > ResourceFailed: FALSE > > Tag: TEXT: ENGLISH: MPCBL0030 (len=9) > > OpenHPI> rdr 30 w > > (262144):WATCHDOG_RDR NUM=000 Tag=Watchdog > > RDR NUM ==> 000 > > RecordId: 262144 > > RdrType: WATCHDOG_RDR > > EntityPath: {SYSTEM_CHASSIS,1}{PHYSICAL_SLOT,2}{160,0} > > IsFru: FALSE > > Watchdog: > > Num: 0 > > Oem: 0 > > IdString: TEXT: ENGLISH: Watchdog (len=8) > > > > When the above SAFTestsuite verification testcase is run, the > testcase passes for this resource: > > > > *****************Resource begin****************** > > > > rdr_type=3, rdr_name=MPCBL0030, return=SAF_TEST_NOTSUPPORT > > rdr_type=4, rdr_name=Watchdog, return=SAF_TEST_PASS > > resource_id=30, tag=MPCBL0030, return=SAF_TEST_PASS > > *****************Resource end******************** > > Associated with resource is a watchdog RDR of rdr_type = 4 (This > was missing in IPMI plugin case). Hence this testcase passes > > > > With this finding, I would like to conclude that watchdog RDR is > not created by IPMI plugin for the resource with watchdog capability. > > > > Thanks in advance for your responses. > > > > Regards, > > Raghavendra > > > > > > > > > > > ------------------------------------------------------------------------- > Take Surveys. Earn Cash. Influence the Future of IT > Join SourceForge.net's Techsay panel and you'll get the chance to share your > opinions on IT & business topics through brief surveys - and earn cash > http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV > _______________________________________________ > Openhpi-devel mailing list > Ope...@li... > https://lists.sourceforge.net/lists/listinfo/openhpi-devel |