Carcosa Element

src/sst/elements/carcosa/Makefile.am

@@ -0,0 +1,66 @@
+# -*- Makefile -*-
+#
+#
+
+AM_CPPFLAGS += \
+	$(MPI_CPPFLAGS) \
+	-I$(top_srcdir)/src
+
+compdir = $(pkglibdir)
+comp_LTLIBRARIES = libcarcosa.la
+libcarcosa_la_SOURCES =                     \
+    injectors/faultInjectorBase.cc          \
+    injectors/faultInjectorBase.h           \
+    injectors/stuckAtFaultInjector.cc       \
+    injectors/stuckAtFaultInjector.h        \
+    injectors/corruptMemFaultInjector.cc    \
+    injectors/corruptMemFaultInjector.h     \
+    injectors/randomDropFaultInjector.cc    \
+    injectors/randomDropFaultInjector.h     \
+    injectors/randomFlipFaultInjector.cc    \
+    injectors/randomFlipFaultInjector.h     \
+    injectors/dropFlipFaultInjector.cc        \
+    injectors/dropFlipFaultInjector.h         \
+    faultlogic/faultBase.cc                 \
+    faultlogic/faultBase.h                  \
+    faultlogic/stuckAtFault.cc              \
+    faultlogic/stuckAtFault.h               \
+    faultlogic/corruptMemFault.cc           \
+    faultlogic/corruptMemFault.h            \
+    faultlogic/randomDropFault.cc           \
+    faultlogic/randomDropFault.h            \
+    faultlogic/randomFlipFault.cc           \
+    faultlogic/randomFlipFault.h
+
+EXTRA_DIST = \
+    tests/testCorruptMemBasic.py                  \
+    tests/testCorruptMemDouble.py                 \
+    tests/testCorruptMemDoubleOverlap.py          \
+    tests/testRandomDrop.py                       \
+    tests/testRandomFlip.py                       \
+    tests/testStuckAtBasic.py                     \
+    tests/testStuckAtMultiple.py                  \
+    tests/testStuckAtOverlap.py                   \
+    tests/testStuckAtSameByte.py
+
+sstdir = $(includedir)/sst/elements/carcosa
+nobase_sst_HEADERS =                        \
+    injectors/faultInjectorBase.h           \
+    injectors/stuckAtFaultInjector.h        \
+    injectors/corruptMemFaultInjector.h     \
+    injectors/randomDropFaultInjector.h     \
+    injectors/randomFlipFaultInjector.h     \
+    injectors/dropFlipFaultInjector.h       \
+    faultlogic/faultBase.h                  \
+    faultlogic/stuckAtFault.h               \
+    faultlogic/corruptMemFault.h            \
+    faultlogic/randomDropFault.h            \
+    faultlogic/randomFlipFault.h
+
+libcarcosa_la_LDFLAGS = -module -avoid-version
+libcarcosa_la_LIBADD =
+
+AM_CPPFLAGS += $(HMC_FLAG)
+install-exec-hook:
+	$(SST_REGISTER_TOOL) SST_ELEMENT_SOURCE     carcosa=$(abs_srcdir)
+	$(SST_REGISTER_TOOL) SST_ELEMENT_TESTS      carcosa=$(abs_srcdir)/tests

src/sst/elements/carcosa/configure.m4

@@ -0,0 +1,9 @@
+dnl -*- Autoconf -*-
+dnl vim:ft=config
+dnl
+
+AC_DEFUN([SST_carcosa_CONFIG], [
+	carcosa_happy="yes"
+
+  AS_IF([test "$carcosa_happy" = "yes"], [$1], [$2])
+])

src/sst/elements/carcosa/faultlogic/corruptMemFault.cc

@@ -0,0 +1,125 @@
+// Copyright 2009-2025 NTESS. Under the terms
+// of Contract DE-NA0003525 with NTESS, the U.S.
+// Government retains certain rights in this software.
+//
+// Copyright (c) 2009-2025, NTESS
+// All rights reserved.
+//
+// This file is part of the SST software package. For license
+// information, see the LICENSE file in the top level directory of the
+// distribution.
+
+#include "sst/elements/carcosa/faultlogic/corruptMemFault.h"
+
+using namespace SST::Carcosa;
+
+CorruptMemFault::CorruptMemFault(Params& params, FaultInjectorBase* injector) : FaultBase(params, injector) {
+#ifdef __SST_DEBUG_OUTPUT__
+    getSimulationDebug()->debug(CALL_INFO_LONG, 1, 0, "Fault type: Corrupt Memory Region\n");
+#endif
+    // read in data regions
+    std::vector<std::string> regionVec;
+
+    // parameter format: {"regions": ["start_addr0, end_addr0", "start_addr1, end_addr1",...]}
+    params.find_array<std::string>("regions", regionVec);
+
+    // process entries into region
+    for (std::string region: regionVec) {
+        std::pair<uint64_t,uint64_t> region_pair = convertString(region);
+
+        // check validity
+        if (region_pair.first > region_pair.second) {
+            getSimulationOutput()->fatal(CALL_INFO_LONG, -1, "Invalid corruption region: [0x%zx, 0x%zx].\n",
+                                        region_pair.first, region_pair.second);
+        }
+
+        corruptionRegions_.push_back(region_pair);
+#ifdef __SST_DEBUG_OUTPUT__
+        getSimulationDebug()->debug(CALL_INFO_LONG, 1, 0, "Inserted corruption region: [0x%zx, 0x%zx]\n",
+                                    region_pair.first, region_pair.second);
+#endif
+    }
+}
+
+bool CorruptMemFault::faultLogic(Event*& ev) {
+    SST::MemHierarchy::MemEvent* mem_ev = convertMemEvent(ev);
+
+    Addr base_addr = mem_ev->getBaseAddr();
+    dataVec original_payload = mem_ev->getPayload();
+    dataVec new_payload(original_payload);
+    for (int r: regionsToUse_) {
+        auto& region = corruptionRegions_[r];
+        size_t payload_sz = mem_ev->getPayloadSize();
+        int32_t start = computeStartIndex(base_addr, payload_sz, region.first);
+        if (start < 0) {
+            getSimulationOutput()->fatal(CALL_INFO_LONG, -1, "No valid start index for corruption.\n");
+        }
+        int32_t end = computeEndIndex(base_addr, payload_sz, region.second);
+        if (end < 0) {
+            getSimulationOutput()->fatal(CALL_INFO_LONG, -1, "No valid start index for corruption.\n");
+        }
+        for (int i = start; i < end; i++) {
+            new_payload[i] = static_cast<uint8_t>(injector_->randUInt32(0,255));
+        }
+    }
+    setMemEventPayload(ev, new_payload);
+    return true;
+}
+
+std::pair<uint64_t,uint64_t> CorruptMemFault::convertString(std::string& region) {
+    std::stringstream ss(region);
+    uint64_t addr0, addr1;
+
+    ss >> std::hex >> addr0;
+    if (ss.peek() == ','){
+        ss.ignore();
+    }
+    ss >> std::hex >> addr1;
+
+#ifdef __SST_DEBUG_OUTPUT__
+    getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "Extracted region pair: [0x%zx, 0x%zx]\n",
+                                addr0, addr1);
+#endif
+    return make_pair(addr0, addr1);
+}
+
+int32_t CorruptMemFault::computeStartIndex(Addr base_addr, size_t payload_sz, Addr region_start) {
+    // start index is always the first byte of this payload in the corruption region
+    int payload_bytes = payload_sz / 8;
+    Addr addr = base_addr;
+    for (int i = 0; i < payload_bytes; i++, addr+=8) {
+        if (addr >= region_start) {
+            return addr - base_addr;
+        }
+    }
+    return -1;
+}
+
+int32_t CorruptMemFault::computeEndIndex(Addr base_addr, size_t payload_sz, Addr region_end) {
+    // end index is either the final addr's final byte, or the region end's addr's final byte
+    int payload_bytes = payload_sz / 8;
+    Addr addr = base_addr + ((payload_bytes - 1) * 8);
+    for (int i = payload_bytes; i >= 0; i--, addr-=8) {
+        if (addr <= region_end) {
+            return addr - base_addr + 8;
+        }
+    }
+    return -1;
+}
+
+std::vector<uint32_t>* CorruptMemFault::checkAddrUsage(Event*& ev) {
+    Addr base_addr = convertMemEvent(ev)->getBaseAddr();
+    for (int i = 0; i < corruptionRegions_.size(); i++) {
+        auto& region = corruptionRegions_[i];
+        // check if message contains ANY address in this region
+        int payload_bytes = convertMemEvent(ev)->getPayloadSize() / 8;
+        Addr addr = base_addr;
+        for (int j = 0; j < payload_bytes; addr+=8, j++) {
+            if ((addr >= region.first) && (addr <= region.second)) {
+                regionsToUse_.push_back(i);
+                break;
+            }
+        }
+    }
+    return &regionsToUse_;
+}

src/sst/elements/carcosa/faultlogic/corruptMemFault.h

@@ -0,0 +1,72 @@
+// Copyright 2009-2025 NTESS. Under the terms
+// of Contract DE-NA0003525 with NTESS, the U.S.
+// Government retains certain rights in this software.
+//
+// Copyright (c) 2009-2025, NTESS
+// All rights reserved.
+//
+// This file is part of the SST software package. For license
+// information, see the LICENSE file in the top level directory of the
+// distribution.
+
+#ifndef SST_ELEMENTS_CARCOSA_CORRUPTMEMFAULT_H
+#define SST_ELEMENTS_CARCOSA_CORRUPTMEMFAULT_H
+
+#include "sst/elements/carcosa/faultlogic/faultBase.h"
+#include "sst/core/rng/mersenne.h"
+#include <vector>
+#include <utility>
+#include <string>
+#include <sstream>
+
+namespace SST::Carcosa {
+
+typedef std::vector<uint8_t> dataVec;
+typedef SST::MemHierarchy::Addr Addr;
+
+/**
+ * This fault is intended to be placed on the input/output ports
+ * of memory components such as DRAM or HBM. Events that pass through
+ * it, and whose data addresses fall within the ranges set in this
+ * module's parameters, will have their payloads randomly altered
+ * to simulate corruption in the affected region of memory.
+ */
+class CorruptMemFault : public FaultBase
+{
+public:
+
+    CorruptMemFault(Params& params, FaultInjectorBase* injector);
+
+    CorruptMemFault() = default;
+    ~CorruptMemFault() {}
+
+    /**
+     * 1. Read in event
+     * 2. Test if event is in specified region
+     * 3. Corrupt event payload if necessary
+     * 4. Replace payload
+     */
+    bool faultLogic(Event*& ev) override;
+
+    std::vector<uint32_t>* checkAddrUsage(Event*& ev);
+protected:
+
+    std::vector<std::pair<uint64_t, uint64_t>> corruptionRegions_;
+
+    std::vector<uint32_t> regionsToUse_;
+
+    std::pair<uint64_t,uint64_t> convertString(std::string& region);
+
+    int32_t computeStartIndex(Addr base_addr, size_t payload_sz, Addr region_start);
+    int32_t computeEndIndex(Addr base_addr, size_t payload_sz, Addr region_end);
+
+    void serialize_order(SST::Core::Serialization::serializer& ser) override {
+        FaultBase::serialize_order(ser);
+        SST_SER(corruptionRegions_);
+        SST_SER(regionsToUse_);
+    }
+    ImplementVirtualSerializable(CorruptMemFault)
+}; // CorruptMemFault
+} // namespace SST::Carcosa
+
+#endif // SST_ELEMENTS_CARCOSA_CORRUPTMEMFAULT_H

src/sst/elements/carcosa/faultlogic/faultBase.cc

@@ -0,0 +1,77 @@
+// Copyright 2009-2025 NTESS. Under the terms
+// of Contract DE-NA0003525 with NTESS, the U.S.
+// Government retains certain rights in this software.
+//
+// Copyright (c) 2009-2025, NTESS
+// All rights reserved.
+//
+// This file is part of the SST software package. For license
+// information, see the LICENSE file in the top level directory of the
+// distribution.
+
+#include "sst/elements/carcosa/faultlogic/faultBase.h"
+
+using namespace SST::Carcosa;
+
+FaultBase::FaultBase(Params& params, FaultInjectorBase* injector) : injector_(injector) {
+    //
+}
+
+bool FaultBase::faultLogic(Event*& ev) {
+    return true;
+}
+
+SST::Output* FaultBase::getSimulationOutput() {
+    return injector_->getOutput();
+}
+
+SST::Output* FaultBase::getSimulationDebug() {
+    return injector_->getDebug();
+}
+
+SST::MemHierarchy::MemEvent* FaultBase::convertMemEvent(Event*& ev) {
+    SST::MemHierarchy::MemEvent* mem_ev = dynamic_cast<SST::MemHierarchy::MemEvent*>(ev);
+
+    if (mem_ev == nullptr) {
+        getSimulationOutput()->fatal(CALL_INFO_LONG, -1, "Attempting to inject mem fault on a non-MemEvent type.\n");
+    }
+
+#ifdef __SST_DEBUG_OUTPUT__
+    getSimulationDebug()->debug(CALL_INFO_LONG, 3, 0, "Intercepted event %zu/%d\n", mem_ev->getID().first, mem_ev->getID().second);
+#endif
+    return mem_ev;
+}
+
+dataVec& FaultBase::getMemEventPayload(Event*& ev) {
+    return convertMemEvent(ev)->getPayload();
+}
+
+void FaultBase::setMemEventPayload(Event*& ev, dataVec newPayload) {
+#ifdef __SST_DEBUG_OUTPUT__
+    getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "Payload before replacement:\n");
+    SST::MemHierarchy::MemEvent* mem_ev = convertMemEvent(ev);
+    dataVec payload = getMemEventPayload(ev);
+    for (int i = 0; i < payload.size(); i+=8) {
+        getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "\n0x%zx: [\t", mem_ev->getBaseAddr() + i);
+        for (int j = i; j < (i+8); j++) {
+            getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "%d\t", payload[j]);
+        }
+        getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "]\n");
+    }
+#endif
+    convertMemEvent(ev)->setPayload(newPayload);
+
+#ifdef __SST_DEBUG_OUTPUT__
+    getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "Payload after replacement:\n");
+    mem_ev = convertMemEvent(ev);
+    payload = getMemEventPayload(ev);
+    for (int i = 0; i < payload.size(); i+=8) {
+        getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "\n0x%zx: [\t", mem_ev->getBaseAddr() + i);
+        for (int j = i; j < (i+8); j++) {
+            getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "%d\t", payload[j]);
+        }
+        getSimulationDebug()->debug(CALL_INFO_LONG, 2, 0, "]\n");
+    }
+#endif
+}
+