[EMCAL-1048] Consistent check fo RCU trailer and channel header integrity

mfasDa · davidrohr · commit f201da9718f7 · 2023-10-06T09:07:09.000+02:00
- RCU trailer must have a valid trailer marker, valid
  trailer version and the corresponding expected size
  of the trailer
- Channel header must have a valid branch ID (0-1),
  FEC ID (0-9) and ALTRO ID (0,2,3,4)
- Channel headers only accept if bit 30 is set and
  bit 31 is not set

Treating corruptions in case of bitflip of bit 31 leading to
an artificial interpretation of a RCU trailer word in case
the word is the last word on the page, even though the payload
continues on the next page.
diff --git a/Detectors/EMCAL/base/include/EMCALBase/RCUTrailer.h b/Detectors/EMCAL/base/include/EMCALBase/RCUTrailer.h
@@ -388,6 +388,11 @@ class RCUTrailer
   /// are assigned based on the trailer word marker.
   static RCUTrailer constructFromPayloadWords(const gsl::span<const uint32_t> payloadwords);
 
+  /// \brief Check whether the word is a valid last trailer word
+  /// \param trailerword Word to be checked
+  /// \return True if the word is a valid last trailer word, false if there are inconsistencies
+  static bool checkLastTrailerWord(uint32_t trailerword);
+
  private:
   /// \struct AltroConfig
   /// \brief Bit field configuration of the ALTRO config registers
diff --git a/Detectors/EMCAL/base/src/RCUTrailer.cxx b/Detectors/EMCAL/base/src/RCUTrailer.cxx
@@ -35,6 +35,29 @@ void RCUTrailer::reset()
   mIsInitialized = false;
 }
 
+bool RCUTrailer::checkLastTrailerWord(uint32_t trailerword)
+{
+  const int MIN_FWVERSION = 2;
+  const int MAX_FWVERSION = 2;
+  if ((trailerword >> 30) != 3) {
+    return false;
+  }
+  auto firmwarevesion = (trailerword >> 16) & 0xFF;
+  auto trailerSize = (trailerword & 0x7F);
+  if (firmwarevesion < MIN_FWVERSION || firmwarevesion > MAX_FWVERSION) {
+    return false;
+  }
+  if (trailerSize < 2) {
+    return false;
+  }
+  if (firmwarevesion == 2) {
+    if (trailerSize < 9) {
+      return false;
+    }
+  }
+  return true;
+}
+
 void RCUTrailer::constructFromRawPayload(const gsl::span<const uint32_t> payloadwords)
 {
   reset();
diff --git a/Detectors/EMCAL/reconstruction/include/EMCALReconstruction/AltroDecoder.h b/Detectors/EMCAL/reconstruction/include/EMCALReconstruction/AltroDecoder.h
@@ -156,6 +156,7 @@ class MinorAltroDecodingError
   /// \brief Error codes connected with the ALTRO decoding
   enum class ErrorType_t {
     BUNCH_HEADER_NULL,            ///< Bunch header is 0
+    CHANNEL_HEADER,               ///< Channel header corruption
     CHANNEL_END_PAYLOAD_UNEXPECT, ///< Unexpected end of payload (channel or trailer word in bunch words)
     CHANNEL_PAYLOAD_EXCEED,       ///< Exceeding channel payload block
     CHANNEL_ORDER,                ///< Channels not in increasing order
@@ -210,7 +211,7 @@ class MinorAltroDecodingError
 
   /// \brief Get the number of error types handled by the AltroDecoderError
   /// \return Number of error types
-  static constexpr int getNumberOfErrorTypes() noexcept { return 7; }
+  static constexpr int getNumberOfErrorTypes() noexcept { return 8; }
 
   /// \brief Get the name connected to the error type
   ///
@@ -352,6 +353,11 @@ class AltroDecoder
   /// In case of failure an exception is thrown.
   void checkRCUTrailer();
 
+  /// \brief Check hardware address in channel header for consistency
+  /// \param hwaddress Hardware address to check
+  /// \return True if the channel is consistent (branch, FEC and Altro in expected range) - false otherwise
+  bool checkChannelHWAddress(int hwaddress);
+
   RawReaderMemory& mRawReader;                               ///< underlying raw reader
   RCUTrailer mRCUTrailer;                                    ///< RCU trailer
   std::vector<Channel> mChannels;                            ///< vector of channels in the raw stream
diff --git a/Detectors/EMCAL/reconstruction/src/AltroDecoder.cxx b/Detectors/EMCAL/reconstruction/src/AltroDecoder.cxx
@@ -75,6 +75,13 @@ void AltroDecoder::readChannels()
     uint16_t payloadsize = (channelheader >> 16) & 0x3FF;
     bool badchannel = (channelheader >> 29) & 0x1;
 
+    // check hardware address for consistency
+    if (!checkChannelHWAddress(hwaddress)) {
+      // Inconsistent HW address information, channel header corrupted, payload must be skipped
+      mMinorDecodingErrors.emplace_back(MinorAltroDecodingError::ErrorType_t::CHANNEL_ORDER, channelheader, currentword);
+      continue;
+    }
+
     int currentfec = 10 * Channel::getBranchIndexFromHwAddress(hwaddress) + Channel::getFecIndexFromHwAddress(hwaddress);
     // std::cout << "Branch: " << Channel::getBranchIndexFromHwAddress(hwaddress) << ", FEC " << Channel::getFecIndexFromHwAddress(hwaddress) << " -> " << currentfec << " (channel " << Channel::getChannelIndexFromHwAddress(hwaddress) << " )" << std::endl;
     if (currentfec < lastFEC) {
@@ -160,6 +167,23 @@ void AltroDecoder::readChannels()
   mChannelsInitialized = true;
 }
 
+bool AltroDecoder::checkChannelHWAddress(int hwaddress)
+{
+  unsigned int branch = Channel::getBranchIndexFromHwAddress(hwaddress),
+               fec = Channel::getFecIndexFromHwAddress(hwaddress),
+               altro = Channel::getAltroIndexFromHwAddress(hwaddress);
+  if (branch > 1) {
+    return false;
+  }
+  if (fec > 9) {
+    return false;
+  }
+  if (!(altro == 0 || altro == 2 || altro == 3 || altro == 4)) {
+    return false;
+  }
+  return true;
+}
+
 const RCUTrailer& AltroDecoder::getRCUTrailer() const
 {
   if (!mRCUTrailer.isInitialized()) {
@@ -354,18 +378,21 @@ int MinorAltroDecodingError::errorTypeToInt(MinorAltroErrType errortype)
     case MinorAltroErrType::CHANNEL_ORDER:
       errorNumber = 2;
       break;
-    case MinorAltroErrType::BUNCH_HEADER_NULL:
+    case MinorAltroErrType::CHANNEL_HEADER:
       errorNumber = 3;
       break;
-    case MinorAltroErrType::BUNCH_LENGTH_EXCEED:
+    case MinorAltroErrType::BUNCH_HEADER_NULL:
       errorNumber = 4;
       break;
-    case MinorAltroErrType::BUNCH_LENGTH_ALLOW_EXCEED:
+    case MinorAltroErrType::BUNCH_LENGTH_EXCEED:
       errorNumber = 5;
       break;
-    case MinorAltroErrType::BUNCH_STARTTIME:
+    case MinorAltroErrType::BUNCH_LENGTH_ALLOW_EXCEED:
       errorNumber = 6;
       break;
+    case MinorAltroErrType::BUNCH_STARTTIME:
+      errorNumber = 7;
+      break;
   };
 
   return errorNumber;
@@ -387,15 +414,18 @@ MinorAltroErrType MinorAltroDecodingError::intToErrorType(int errornumber)
       errorType = MinorAltroErrType::CHANNEL_ORDER;
       break;
     case 3:
-      errorType = MinorAltroErrType::BUNCH_HEADER_NULL;
+      errorType = MinorAltroErrType::CHANNEL_HEADER;
       break;
     case 4:
-      errorType = MinorAltroErrType::BUNCH_LENGTH_EXCEED;
+      errorType = MinorAltroErrType::BUNCH_HEADER_NULL;
       break;
     case 5:
-      errorType = MinorAltroErrType::BUNCH_LENGTH_ALLOW_EXCEED;
+      errorType = MinorAltroErrType::BUNCH_LENGTH_EXCEED;
       break;
     case 6:
+      errorType = MinorAltroErrType::BUNCH_LENGTH_ALLOW_EXCEED;
+      break;
+    case 7:
       errorType = MinorAltroErrType::BUNCH_STARTTIME;
       break;
     default:
@@ -414,6 +444,8 @@ const char* MinorAltroDecodingError::getErrorTypeName(ErrorType_t errortype)
       return "ChannelPayloadExceed";
     case MinorAltroErrType::CHANNEL_ORDER:
       return "ChannelOrderError";
+    case MinorAltroErrType::CHANNEL_HEADER:
+      return "ChannelHeader";
     case MinorAltroErrType::BUNCH_HEADER_NULL:
       return "BunchHeaderNull";
     case MinorAltroErrType::BUNCH_LENGTH_EXCEED:
@@ -435,6 +467,8 @@ const char* MinorAltroDecodingError::getErrorTypeTitle(ErrorType_t errortype)
       return "Channel exceed";
     case MinorAltroErrType::CHANNEL_ORDER:
       return "FEC order";
+    case MinorAltroErrType::CHANNEL_HEADER:
+      return "Channel header invalid";
     case MinorAltroErrType::BUNCH_HEADER_NULL:
       return "Bunch header null";
     case MinorAltroErrType::BUNCH_LENGTH_EXCEED:
@@ -456,6 +490,8 @@ const char* MinorAltroDecodingError::getErrorTypeDescription(ErrorType_t errorty
       return "Trying to access out-of-bound payload!";
     case MinorAltroErrType::CHANNEL_ORDER:
       return "Invalid FEC order";
+    case MinorAltroErrType::CHANNEL_HEADER:
+      return "Invalid channel header";
     case MinorAltroErrType::BUNCH_HEADER_NULL:
       return "Bunch header 0 or not configured!";
     case MinorAltroErrType::BUNCH_LENGTH_EXCEED:
diff --git a/Detectors/EMCAL/reconstruction/src/RawReaderMemory.cxx b/Detectors/EMCAL/reconstruction/src/RawReaderMemory.cxx
@@ -173,7 +173,7 @@ void RawReaderMemory::nextPage(bool doResetPayload)
       // The trailer is only decoded if the DDL is in the range of SRU DDLs. STU pages are propagated
       // 1-1 without trailer parsing
       auto lastword = *(mRawBuffer.getDataWords().rbegin());
-      if (lastword >> 30 == 3) {
+      if (RCUTrailer::checkLastTrailerWord(lastword)) {
         // lastword is a trailer word
         // decode trailer and chop
         try {
diff --git a/Detectors/EMCAL/reconstruction/test/testMinorAltroDecodingError.cxx b/Detectors/EMCAL/reconstruction/test/testMinorAltroDecodingError.cxx
@@ -23,31 +23,35 @@ namespace emcal
 
 BOOST_AUTO_TEST_CASE(MinorAltroDecodingError_test)
 {
-  BOOST_CHECK_EQUAL(MinorAltroDecodingError::getNumberOfErrorTypes(), 7);
-  std::array<std::string, 7> errornames = {{"ChannelEndPayloadUnexpected",
+  BOOST_CHECK_EQUAL(MinorAltroDecodingError::getNumberOfErrorTypes(), 8);
+  std::array<std::string, 8> errornames = {{"ChannelEndPayloadUnexpected",
                                             "ChannelPayloadExceed",
                                             "ChannelOrderError",
+                                            "ChannelHeader",
                                             "BunchHeaderNull",
                                             "BunchLengthExceed",
                                             "BunchLengthAllowExceed",
                                             "BunchStarttimeExceed"}},
                              errortitles = {{"Channel end unexpected",
                                              "Channel exceed",
                                              "FEC order",
+                                             "Channel header invalid",
                                              "Bunch header null",
                                              "Bunch length exceed",
                                              "Bunch length impossible",
                                              "Bunch starttime exceed"}},
                              errordescriptions = {{"Unexpected end of payload in altro channel payload!",
                                                    "Trying to access out-of-bound payload!",
                                                    "Invalid FEC order",
+                                                   "Invalid channel header",
                                                    "Bunch header 0 or not configured!",
                                                    "Bunch length exceeding channel payload size!",
                                                    "Bunch length exceeding max. possible bunch size!",
                                                    "Bunch start time outside range!"}};
-  std::array<MinorAltroDecodingError::ErrorType_t, 7> errortypes = {{MinorAltroDecodingError::ErrorType_t::CHANNEL_END_PAYLOAD_UNEXPECT,
+  std::array<MinorAltroDecodingError::ErrorType_t, 8> errortypes = {{MinorAltroDecodingError::ErrorType_t::CHANNEL_END_PAYLOAD_UNEXPECT,
                                                                      MinorAltroDecodingError::ErrorType_t::CHANNEL_PAYLOAD_EXCEED,
                                                                      MinorAltroDecodingError::ErrorType_t::CHANNEL_ORDER,
+                                                                     MinorAltroDecodingError::ErrorType_t::CHANNEL_HEADER,
                                                                      MinorAltroDecodingError::ErrorType_t::BUNCH_HEADER_NULL,
                                                                      MinorAltroDecodingError::ErrorType_t::BUNCH_LENGTH_EXCEED,
                                                                      MinorAltroDecodingError::ErrorType_t::BUNCH_LENGTH_ALLOW_EXCEED,
