Skip to content

Commit e5622fb

Browse files
cvswenwghoffa
cvswen
authored and
wghoffa
committed
Insufficient Resources check needs to handle NVDIMM logic
Check added to platCheckMinimumHardware for Hostboot running on a proc with only NVDIMMs and no regular dimms. Change-Id: Ia8d413182209070b96d4d1cd61a3b4f3b713a596 RTC:201194 Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/75110 Reviewed-by: Christian R. Geddes <crgeddes@us.ibm.com> Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com> Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com> Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com> Reviewed-by: Matt Derksen <mderkse1@us.ibm.com> Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com> Reviewed-by: William G. Hoffa <wghoffa@us.ibm.com>
1 parent 16378af commit e5622fb

File tree

3 files changed

+196
-20
lines changed

3 files changed

+196
-20
lines changed

‎src/include/usr/hwas/hwasPlat.H

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -63,6 +63,12 @@ void markTargetChanged(TARGETING::TargetHandle_t i_target)
6363
{};
6464
#endif
6565

66+
/**
67+
* @brief Helper function to check if HB is running
68+
* from an NVDIMM and not a regular DIMM
69+
*/
70+
errlHndl_t checkForHbOnNvdimm();
71+
6672
// default unknown EC level
6773
constexpr TARGETING::ATTR_EC_type INVALID__ATTR_EC = 0xFF;
6874

‎src/include/usr/hwas/hwasPlatReasonCodes.H

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -38,6 +38,7 @@ namespace HWAS
3838
MOD_PLAT_DECONFIG_GARD = 0x81,
3939
MOD_PLAT_READIDEC = 0x82,
4040
MOD_PLAT_READLX = 0x83,
41+
MOD_CHECK_HB_NVDIMM = 0x84,
4142
};
4243

4344
enum HwasPlatReasonCode
@@ -52,6 +53,7 @@ namespace HWAS
5253
RC_BAD_MCA = HWAS_COMP_ID | 0x85,
5354
RC_RT_NULL_FIRMWARE_REQUEST_PTR = HWAS_COMP_ID | 0x86,
5455
RC_RT_NULL_FIRMWARE_MSG_PTR = HWAS_COMP_ID | 0x87,
56+
RC_HB_PROC_ONLY_NVDIMM = HWAS_COMP_ID | 0x88,
5557
};
5658

5759
};

‎src/usr/hwas/hwasPlat.C

Lines changed: 188 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -40,6 +40,7 @@
4040
#include <vpd/mvpdenums.H>
4141
#include <stdio.h>
4242
#include <sys/mm.h>
43+
#include <sys/misc.h>
4344

4445
#include <pnor/pnorif.H>
4546

@@ -75,6 +76,7 @@ RegisterHWASFunctions registerHWASFunctions;
7576

7677
using namespace TARGETING;
7778

79+
7880
//******************************************************************************
7981
// platReadIDEC function
8082
//******************************************************************************
@@ -696,28 +698,194 @@ void markTargetChanged(TARGETING::TargetHandle_t i_target)
696698
// platCheckMinimumHardware()
697699
//******************************************************************************
698700
void platCheckMinimumHardware(uint32_t & io_plid,
699-
const TARGETING::ConstTargetHandle_t i_node,
700-
bool *o_bootable)
701+
const TARGETING::ConstTargetHandle_t i_node,
702+
bool *o_bootable)
701703
{
702-
//errlHndl_t l_errl = NULL;
703-
704-
// nothing to do yet...
705-
706-
// if you add something here, don't forget to
707-
// add a procedure callout
708-
//l_errl->addProcedureCallout( EPUB_PRC_FIND_DECONFIGURED_PART,
709-
// SRCI_PRIORITY_HIGH);
710-
711-
// and update the common plid
712-
//if (io_plid != 0)
713-
//{
714-
// l_errl->plid(io_plid) ;
715-
//}
716-
//else
717-
//{
718-
// io_plid = l_errl->plid();
719-
//}
704+
errlHndl_t l_errl = NULL;
705+
706+
Target* l_pMasterProcChip = NULL;
707+
targetService().masterProcChipTargetHandle(l_pMasterProcChip);
708+
709+
// NVDIMM only supported on Nimbus
710+
ATTR_MODEL_type l_model = l_pMasterProcChip->getAttr<ATTR_MODEL>();
711+
if (l_model == MODEL_NIMBUS)
712+
{
713+
l_errl = checkForHbOnNvdimm();
714+
if (l_errl)
715+
{
716+
HWAS_ERR("platCheckMinimumHardware::checkForHbOnNvdimm() failed.");
717+
718+
if(o_bootable)
719+
{
720+
*o_bootable = false;
721+
}
722+
723+
// Add procedure callout, update common plid, commit
724+
hwasErrorAddProcedureCallout(l_errl,
725+
EPUB_PRC_FIND_DECONFIGURED_PART,
726+
SRCI_PRIORITY_HIGH);
727+
hwasErrorUpdatePlid(l_errl, io_plid);
728+
errlCommit(l_errl, HWAS_COMP_ID);
729+
}
730+
}
731+
720732
}
721733

734+
//******************************************************************************
735+
// checkForHbOnNvdimm()
736+
//******************************************************************************
737+
errlHndl_t checkForHbOnNvdimm(void)
738+
{
739+
errlHndl_t l_errl = nullptr;
740+
741+
do
742+
{
743+
HWAS_DBG("Check if HB is running on a proc with only NVDIMMs.");
744+
745+
// Get all functional proc chip targets
746+
TargetHandleList l_procList;
747+
getAllChips(l_procList, TARGETING::TYPE_PROC);
748+
assert(l_procList.size() != 0, "Empty proc list returned!");
749+
750+
// Use the hrmor to find which proc HB is running on
751+
const auto l_hbHrmor = cpu_spr_value(CPU_SPR_HRMOR);
752+
753+
// Use the MemBases and MemSizes to find which group
754+
// includes the hrmor
755+
ATTR_PROC_MEM_BASES_type l_memBases = {0};
756+
ATTR_PROC_MEM_SIZES_type l_memSizes = {0};
757+
size_t l_numGroups = sizeof(ATTR_PROC_MEM_SIZES_type)/sizeof(uint64_t);
758+
759+
// checkMinimumHardware may be called before the groups are set up
760+
// If the MemSizes are all zero the groups are not set up yet
761+
// Break out of the check
762+
bool l_memSizesAllZero = true;
763+
764+
// Save the HB proc target and group number
765+
Target *l_hbProc = nullptr;
766+
uint32_t l_hbGroup = 0;
767+
768+
for (auto l_pProc : l_procList)
769+
{
770+
// Get the memory group ranges under this proc
771+
assert(l_pProc->tryGetAttr<ATTR_PROC_MEM_BASES>(l_memBases),
772+
"Unable to get ATTR_PROC_MEM_BASES attribute");
773+
assert(l_pProc->tryGetAttr<ATTR_PROC_MEM_SIZES>(l_memSizes),
774+
"Unable to get ATTR_PROC_MEM_SIZES attribute");
775+
776+
777+
for (size_t l_grp=0; l_grp < l_numGroups; l_grp++)
778+
{
779+
// Non-zero size means that there is memory present
780+
if (l_memSizes[l_grp])
781+
{
782+
l_memSizesAllZero = false;
783+
// Check if hrmor is in this group's memory range
784+
if ( (l_hbHrmor >= l_memBases[l_grp]) &&
785+
(l_hbHrmor < (l_memBases[l_grp] +
786+
l_memSizes[l_grp])) )
787+
{
788+
l_hbProc = l_pProc;
789+
l_hbGroup = l_grp;
790+
break;
791+
}
792+
}
793+
}
794+
if (l_hbProc != nullptr)
795+
{
796+
break;
797+
}
798+
}
799+
800+
if (l_memSizesAllZero)
801+
{
802+
break;
803+
}
804+
805+
if (l_hbProc != nullptr)
806+
{
807+
// Found the proc/group HB is running in, now check the dimms
808+
// If we find a regular dimm then we assume that is where
809+
// HB is running
810+
bool l_foundNonNvdimm = false;
811+
812+
// Get the array of mcas/group from the attribute
813+
// The attr contains 8 8-bit entries, one entry per group
814+
// The bits specify which mcas are included in the group
815+
ATTR_MSS_MEM_MC_IN_GROUP_type l_memMcGroup = {0};
816+
assert(l_hbProc->tryGetAttr<ATTR_MSS_MEM_MC_IN_GROUP>(l_memMcGroup),
817+
"Unable to get ATTR_MSS_MEM_MC_IN_GROUP attribute");
818+
819+
// Get list of mcas under this proc
820+
TargetHandleList l_mcaList;
821+
getChildAffinityTargets( l_mcaList,
822+
l_hbProc,
823+
CLASS_UNIT,
824+
TYPE_MCA );
825+
826+
// Loop through the mcas on this proc
827+
for (const auto & l_mcaTarget : l_mcaList)
828+
{
829+
// Get the chip unit for this mca
830+
ATTR_CHIP_UNIT_type l_mcaUnit = 0;
831+
l_mcaUnit = l_mcaTarget->getAttr<ATTR_CHIP_UNIT>();
832+
833+
// Check if this mca is included in the hb memory group
834+
const uint8_t l_mcMask = 0x80;
835+
if (l_memMcGroup[l_hbGroup] & (l_mcMask >> l_mcaUnit))
836+
{
837+
// Get the list of dimms under this mca
838+
TargetHandleList l_dimmList;
839+
getChildAffinityTargets( l_dimmList,
840+
l_mcaTarget,
841+
CLASS_NA,
842+
TYPE_DIMM );
843+
for (const auto & l_dimmTarget : l_dimmList)
844+
{
845+
if (!isNVDIMM(l_dimmTarget))
846+
{
847+
// Found a regular dimm, exit
848+
l_foundNonNvdimm = true;
849+
break;
850+
}
851+
}
852+
}
853+
if (l_foundNonNvdimm)
854+
{
855+
break;
856+
}
857+
}
858+
859+
// If only nvdimms then error
860+
if (!l_foundNonNvdimm)
861+
{
862+
HWAS_ERR("checkForHbOnNvdimm: HB is running on a proc with only NVDIMMS.");
863+
/*@
864+
* @errortype ERRORLOG::ERRL_SEV_UNRECOVERABLE
865+
* @moduleid HWAS::MOD_CHECK_HB_NVDIMM
866+
* @reasoncode HWAS::RC_HB_PROC_ONLY_NVDIMM
867+
* @userdata1 Hostboot Proc Target HUID
868+
* @userdata2 Hostboot Memory Group
869+
* @devdesc Hostboot running on proc with only NVDIMMs
870+
* @custdesc Insufficient DIMM resources
871+
*/
872+
l_errl = new ERRORLOG::ErrlEntry(
873+
ERRORLOG::ERRL_SEV_UNRECOVERABLE,
874+
HWAS::MOD_CHECK_HB_NVDIMM,
875+
HWAS::RC_HB_PROC_ONLY_NVDIMM,
876+
get_huid(l_hbProc),
877+
l_hbGroup);
878+
}
879+
}
880+
else
881+
{
882+
// Should never get here, would be caught elsewhere
883+
HWAS_ERR("checkForHbOnNvdimm: HB execution proc not found.");
884+
}
885+
886+
} while(0);
887+
888+
return l_errl;
889+
}
722890

723891
} // namespace HWAS

0 commit comments

Comments
 (0)