Por si falta algun detallito

app_profiler.py

@@ -101,13 +101,16 @@ def generate_gold(gdb_exec, benchmark_binary, benchmark_args,device):
 
 
 def main():
-    os.system("rm -f {}".format(cp.KERNEL_INFO_DIR))
+    
     parser = argparse.ArgumentParser()
     parser.add_argument('-c', '--conf', dest="config_file", help='Configuration file', required=True)
     parser.add_argument('-d', '--device', dest="device", help="The GPU to perform FI."
                                                               " Default is 0.", required=False, default=0, type=int)
     args = parser.parse_args()
-
+    if (args.device>0):
+      cp.LOGS_PATH=cp.LOGS_PATH+"-"+str(args.device)    
+    cp.rewrite_path()
+    os.system("rm -f {}".format(cp.KERNEL_INFO_DIR))  
     # Read the configuration file with data for all the apps that will be executed
     conf = cf.load_config_file(args.config_file)
 

classes/BitFlip.py

@@ -38,30 +38,22 @@ class BitFlip:
     def numreg (kernel):
       try:
         disassemble_array = cf.execute_command(gdb=gdb, to_execute="disassemble {}".format(kernel))
-        print ("kenel={}".format(kernel))
+        #print ("kenel={}".format(kernel))
       except:
-        print("SIN NOmbre")
         disassemble_array = cf.execute_command(gdb=gdb, to_execute="disassemble")
       listareg=set()
       listaregdst=set()
       listaregcond=set()
       for i in range(0, len(disassemble_array) - 1):
         line = disassemble_array[i]
-        m=re.match(r".*:\t(\S+) .*",line)
+        #m=re.match(r".*:\t(\S+) .*",line)#Para que es m
         todacadena="ASSM_LINE:{}".format(line)
-        #print(todacadena)
         lista=re.findall(r"R(\d+)", line)
         if (len(lista) > 0):   	
            listareg.update(lista)
-           #print (listareg)
-           #listaregdst.add (lista[0])
-           #print (listaregdst)
         lista=re.findall(r" P(\d+)", line)
         if (len(lista) > 0): 
-         #print (listaregcond)
            listaregcond.update(lista)
-  #       if m and m.group(1) != '0':
-  #       Print ("Encontrado"
 
       print ("Registros Visibles ({})".format(len(listareg)))
       print (listareg)
@@ -75,13 +67,16 @@ class BitFlip:
     """
     def asmline(self):
       #Leo la linea de ejecucion
-      return cf.execute_command(gdb=gdb, to_execute="x/1i $pc")
+      linea=cf.execute_command(gdb=gdb, to_execute="x/1i $pc")
+      self.__logging.info("ASSM_LINE:{}".format(linea[0]))
+      return linea
+      
+      
     def reg_asmline( self):
       #Obtengo los registros de la instruccion
       linea= self.asmline()
-      self.__assmline="ASSM_LINE:{}".format(linea[0])
-      #print (linea)#+"-"+str(len(linea))+"-"+linea[0])
-      #print ("============")
+      
+     
       lista=re.findall(r"R(\d+)", linea[0])
       
       #Ahora son todos los registros.
@@ -89,8 +84,11 @@ class BitFlip:
       setlista=set()
       setlista.update(lista)
       
-      #self.__logging.info(str(len(setlista)))
+   
       return setlista
+      
+      
+      
     def regmod (self):
       lista=self.reg_asmline()
       while len(lista) == 0: #Habria que poner un limite. 
@@ -102,9 +100,9 @@ class BitFlip:
       strlistareg="info registers " 
       for x in listareg:
         strlistareg+=x;
-      #print (strlistareg)  
+      
       self.__stringregs=strlistareg
-      #self.__logging.info("REFERED REGISTERS :"+strlistareg)
+
       #Obtengo el valor de los registro referenciados
       valores= cf.execute_command(gdb=gdb, to_execute=strlistareg)
       return valores
@@ -112,11 +110,12 @@ class BitFlip:
        #Obtengo el valor de los registro referenciados por la primera instruccipn
        #self.__logging.info("===============================")
        gdb.execute("nexti")
-       #self.__logging.info("==============================="+self.__stringregs)
+      
        linea= self.asmline()
-       #self.__logging.info("ASSM_LINE:{}".format(linea))
+     
        valores= cf.execute_command(gdb=gdb, to_execute=self.__stringregs)  
        return valores
+       
     def dictreg(self,valores):
       #self.__logging.info("Execute dictreg")
       #Almaceno en un dictionario los valores de los registros
@@ -126,18 +125,11 @@ class BitFlip:
         m = re.match(r".*R(\d+).*0x([0-9a-fA-F]+).*", x)
         if m:
           regs[m.group(1)]=m.group(2)
-          #print (str(m.group(1))+":= "+str(m.group(2)))   
-      #print(regs)
-      #print ("========DR")
+          
       return regs    
     def cmpregdst (self,valores,regs):
-      #Comparo los regsitros, para saber cuales he modificado.
-      #self.__logging.info("Execute cmpregdst------------")#+str(type(valores)))
+      
       regdst=set()
-      #print("==cmpregdst")
-      #print("Arg;..valores...")
-      #print(type(valores))
-      #print (valores)
      
       
       for x in valores:
@@ -150,14 +142,13 @@ class BitFlip:
       #self.__logging.info(str(len(regdst)))
       return regdst
     def  __inst_generic_injector(self):
-    #def miinst_out():
-    	#self.__logging.info("Comienzo")
+  
         regs=self.dictreg(self.regmod())
-        #self.__logging.info("Reg1")
+       
         valores=self.nextinstr()
-        #self.__logging.info("Ejecutado sfte instuccion")
+      
         r=self.cmpregdst(valores,regs)
-        #self.__logging.info("Se han modificado {}".format(len(r)))
+       
         while (len(r) ==0):
              self.__logging.info("INSTRUCTION WITHOUT OPERANDS")
              gdb.execute("nexti")
@@ -167,7 +158,7 @@ class BitFlip:
                     
         self.__register="R{}".format(r.pop())
         self.__logging.info("SELECTED_REGISTER:{}".format(self.__register))
-        self.__logging.info("ASSM_LINE:{}".format(self.__assmline))
+        
         
                 # __rf_generic_injector will set fault_injected attribute
         self.__rf_generic_injector()
@@ -265,6 +256,7 @@ class BitFlip:
     def __rf_generic_injector(self):
         try:
             # get register content
+            self.asmline()
             reg_cmd = cf.execute_command(gdb, "p/t ${}".format(self.__register))
             m = re.match(r'\$(\d+)[ ]*=[ ]*(\S+).*', reg_cmd[0])
 

codes/hotspot/hotspot.conf

@@ -26,20 +26,20 @@ maxWaitTimes = 5
 
 # binary file of the application
 # Must be full path
-benchmarkBinary = /home/badia/rodinia_3.1/cuda/hotspot/hotspot
+benchmarkBinary = /home/badia/mycarol-fi/codes/hotspot/hotspot
 
 #benchmarkBinary_noverificar = /home/badia/rodinia_3.1/cuda/lavaMD/lavaMD 
 
 # Commands to set the session inside GDB environment
-benchmarkArgs =  1024  2 10000 /home/badia/rodinia_3.1/data/hotspot/temp_1024 /home/badia/rodinia_3.1//data/hotspot/power_1024 output.out 
+benchmarkArgs = 2 10000  /home/badia/mycarol-fi/data/hotspot/temp_1024  /home/badia/mycarol-fi/data/hotspot/power_1024 output.out
 #benchmarkArgs = 16384  2 2 /home/badia/rodinia_3.1/data/hotspot/temp_16384 /home/badia/rodinia_3.1/data/hotspot/power_16384 output.out
-benchmarkArgs_noverificar =  1024  2 10000 /home/badia/rodinia_3.1/data/hotspot/temp_1024 /home/badia/rodinia_3.1/data/hotspot/power_1024 output.out
+benchmarkArgs_noverificar =  1024  2 10000  /home/badia/mycarol-fi/data/hotspot/temp_1024  /home/badia/mycarol-fi/data/hotspot/power_1024 output.out
 #benchmarkArgs_noverificar = 16384  2 2 /home/badia/rodinia_3.1/data/hotspot/temp_16384 /home/badia/rodinia_3.1/data/hotspot/power_16384 . 
 # CSV output file. It will be overwrite at each injection
 csvFile = results/hotspot_RF.csv
 
 # You should create a script on the benchmark source folder to verify GOLD_OUTPUT x INJ_OUTPUT
-goldenCheckScript = ../rodinia_3.1/cuda/hotspot/sdc_check.sh
+goldenCheckScript = codes/hotspot/sdc_check.sh
 
 # Number of signals that will be sent to the application
 seqSignals = 2 

codes/lavaMD/lavaMD.conf

@@ -18,8 +18,8 @@ faultModel = 0
 # INST_OUT -> Instruction Output (NOT IMPLEMENTED YET)
 # INST_composed -> Instruction Adress (NOT IMPLEMENTED YET)
 
-injectionSite =  RF
-#injectionSite = INST_OUT
+#injectionSite =  RF
+injectionSite = INST_OUT
 # Max time factor to finish the app, this will be multiplied by the application running time
 # For example if your app spend 2s, and the maxWaitTimes is 5, the max running time before it is
 # Considered as a crash is 10s

codes/nw/nw.conf

@@ -17,7 +17,7 @@ faultModel = 0
 # RF -> Register File
 # INST_OUT -> Instruction Output (NOT IMPLEMENTED YET)
 # INST_composed -> Instruction Adress (NOT IMPLEMENTED YET)
-injectionSite = INST_OUT 
+injectionSite = RF 
 
 # Max time factor to finish the app, this will be multiplied by the application running time
 # For example if your app spend 2s, and the maxWaitTimes is 5, the max running time before it is
@@ -26,9 +26,9 @@ maxWaitTimes = 5
 
 # binary file of the application
 # Must be full path
-benchmarkBinary = /home/badia/rodinia_3.1/cuda/nw/needle
+benchmarkBinary = /home/badia/mycarol-fi/codes/nw/needle
 
-benchmarkBinary_noverificar = /home/badia/rodinia_3.1/cuda/nw/needle_MUDO
+benchmarkBinary_noverificar = /home/badia/mycarol-fi/codes/nw/needle_MUDO
 #benchmarkBinary_noverificar = /home/badia/rodinia_3.1/cuda/lavaMD/lavaMD 
 
 # Commands to set the session inside GDB environment
@@ -37,10 +37,10 @@ benchmarkArgs =  32000 1
  #benchmarkArgs_noverificar =  -s 2048
 
 # CSV output file. It will be overwrite at each injection
-csvFile = /home/badia/rodinia_3.1/cuda/nw/fi_nw_single_bit.csv
+csvFile = results/nw_RF.csv
 
 # You should create a script on the benchmark source folder to verify GOLD_OUTPUT x INJ_OUTPUT
-goldenCheckScript = ../rodinia_3.1/cuda/nw/sdc_check.sh
+goldenCheckScript = codes/nw/sdc_check.sh
 
 # Number of signals that will be sent to the application
 seqSignals = 2

codes/nw/sdc_check.sh

@@ -9,7 +9,6 @@
 # are environment variables defined by the fault_injector.py
 
 # diff stdout
-diff -B ${INJ_OUTPUT_PATH} ${GOLD_OUTPUT_PATH} > ${DIFF_LOG}
 
 # Special comparison like the following one can be done in this script
 grep -q "Result = FAIL" ${INJ_OUTPUT_PATH} >> ${DIFF_LOG}
@@ -18,9 +17,9 @@ grep -q "Result = FAIL" ${INJ_OUTPUT_PATH} >> ${DIFF_LOG}
 diff -B ${INJ_ERR_PATH} ${GOLD_ERR_PATH} > ${DIFF_ERR_LOG}
 echo $pwd
 ls -l result*
-diff ./result.txt ~/rodinia_3.1/cuda/nw/gold/result.1.txt >> ${INJ_OUTPUT_PATH}
+diff ./result.txt codes/nw/gold/result.1.txt >> ${INJ_OUTPUT_PATH}
 
-diff ./result.txt ~/rodinia_3.1/cuda/nw/gold/result.1.txt #>> ${INJ_OUTPUT_PATH}
+#diff ./result.txt ~/rodinia_3.1/cuda/nw/gold/result.1.txt #>> ${INJ_OUTPUT_PATH}
 if [ $? -ne 0 ]
 then
        	echo "Result = FAIL" >> ${INJ_OUTPUT_PATH}
@@ -29,4 +28,6 @@ else
 fi	
 rm result.txt
 # Must exit 0
+p
+diff -B ${INJ_OUTPUT_PATH} ${GOLD_OUTPUT_PATH} > ${DIFF_LOG}
 exit 0

common_parameters.py

@@ -9,24 +9,11 @@ MAX_TIMES_TO_PROFILE = 2
 # Log path to store all injections info
 LOGS_PATH = 'logs'
 
-# Temporary file to store kernel information
-KERNEL_INFO_DIR = LOGS_PATH + '/tmp/carol-fi-kernel-info.txt'
-
-# For golden generation
-GOLD_ERR_PATH = LOGS_PATH + '/tmp/carol_fi_golden_bench_err.txt'
-GOLD_OUTPUT_PATH = LOGS_PATH + '/tmp/carol_fi_golden_bench_output.txt'
-
-# Files that will be compared to golden ones
-INJ_OUTPUT_PATH = LOGS_PATH + '/tmp/carol_fi_inj_bench_output_{}.txt'
-INJ_ERR_PATH = LOGS_PATH + '/tmp/carol_fi_inj_bench_err_{}.txt'
 
 # Internal python scripts
 FLIP_SCRIPT = 'flip_value.py'
 PROFILER_SCRIPT = 'profiler.py'
 
-# Temporary difference logs
-DIFF_LOG = LOGS_PATH + '/tmp/diff_{}.log'
-DIFF_ERR_LOG = LOGS_PATH + '/tmp/diff_err_{}.log'
 
 # Debug env vars
 # Debug FI process
@@ -34,14 +21,12 @@ DEBUG = True
 # Debug profiler process
 DEBUG_PROFILER = True
 
-# Log file for SignalApp thread
-SIGNAL_APP_LOG = LOGS_PATH + '/tmp/signal_app_thread_{}.txt'
 
 # Num of sleep time divisor
 NUM_DIVISION_TIMES = 100.0
 
 # Common body of log filename
-LOG_DEFAULT_NAME = LOGS_PATH + '/tmp/carolfi-flipvalue-{}.log'
+
 
 
 # MAX INT 32 bits
@@ -110,3 +95,30 @@ LEAST_8_BITS = 5
 
 # Focus error string
 FOCUS_ERROR_STRING = "Focus not set on any active CUDA kernel."
+
+def rewrite_path():
+  # Temporary file to store kernel information
+  global KERNEL_INFO_DIR,GOLD_ERR_PATH,GOLD_OUTPUT_PATH,INJ_OUTPUT_PATH,INJ_ERR_PATH,DIFF_LOG,DIFF_ERR_LOG, SIGNAL_APP_LOG, LOG_DEFAULT_NAME   
+  KERNEL_INFO_DIR = LOGS_PATH + '/tmp/carol-fi-kernel-info.txt'
+  # For golden generation
+  GOLD_ERR_PATH = LOGS_PATH + '/tmp/carol_fi_golden_bench_err.txt'
+  GOLD_OUTPUT_PATH = LOGS_PATH + '/tmp/carol_fi_golden_bench_output.txt'
+
+  # Files that will be compared to golden ones
+  INJ_OUTPUT_PATH = LOGS_PATH + '/tmp/carol_fi_inj_bench_output_{}.txt'
+  INJ_ERR_PATH = LOGS_PATH + '/tmp/carol_fi_inj_bench_err_{}.txt'
+
+
+
+  # Temporary difference logs
+  DIFF_LOG = LOGS_PATH + '/tmp/diff_{}.log'
+  DIFF_ERR_LOG = LOGS_PATH + '/tmp/diff_err_{}.log'
+  SIGNAL_APP_LOG = LOGS_PATH + '/tmp/signal_app_thread_{}.txt'
+
+
+
+   # Common body of log filename
+   
+  LOG_DEFAULT_NAME = LOGS_PATH + '/tmp/carolfi-flipvalue-{}.log'
+
+  

fault_injector.py

@@ -69,6 +69,7 @@ def signal_handler(sig, frame):
         os.system(cmd + " > /dev/null 2>&1")
  
     os.system("rm -f {}/bin/*".format(current_path))
+    print("Current_path "+current_path)
     for th in gpus_threads:
       th.join()
     #sys.exit(0)
@@ -309,19 +310,26 @@ def check_injection_outcome(host_thread, logging, injection_site):
         # Check specific outcomes
         # No need to process for RF
         instruction = 'register'
+        assm_line = logging.search("ASSM_LINE")
+        
+        dpc={}
+        dpc['absoluto']="0x"+re.match(r".*0x([0-9a-fA-F]+) <.*",assm_line).group(1)
+        dpc['relativo']=assm_line.split('<')[1].split('>')[0]
+        print("---PC: "+dpc['absoluto']+ "PC rel"+dpc['relativo'])
+        pc=dpc['absoluto']+"<"+dpc['relativo']+">"
         if cp.INJECTION_SITES[injection_site] in [cp.INST_OUT, cp.INST_ADD]:
             # if fault was injected ASSM_LINE MUST be in the logfile          
-            assm_line = logging.search("ASSM_LINE")
             instruction = re.match(r".*:\t(\S+) .*", assm_line).group(1)
-            #instruction = assm_line
+           
 
     except TypeError as te:
         instruction = new_value = old_value = None
+        pc = instruction 
         fi_successful = False
         if cp.DEBUG:
             cf.printf("THREAD {} FAULT WAS NOT INJECTED. ERROR {}".format(host_thread, te))
 
-    return block, fi_successful, new_value, old_value, register, thread, instruction
+    return block, fi_successful, new_value, old_value, register, thread, instruction, pc
 """
 Function to run one execution of the fault injector
 return old register value, new register value
@@ -444,7 +452,7 @@ def gdb_inject_fault(**kwargs):
 
     # Check if the carolfi logfile contains the information
     # to confirm the fault injection outcome
-    block, fi_successful, new_value, old_value, register, thread, instruction = check_injection_outcome(
+    block, fi_successful, new_value, old_value, register, thread, instruction,pc = check_injection_outcome(
         host_thread=host_thread,
         logging=logging,
         injection_site=injection_site
@@ -460,7 +468,7 @@ def gdb_inject_fault(**kwargs):
         cf.printf("THREAD {} SAVE OUTPUT AND RETURN".format(host_thread))
 
     return_list = [register, old_value, new_value, fi_successful,
-                   is_hang, is_crash, is_sdc, is_masked, signal_init_wait_time, block, thread, instruction, user_defined_string]
+                   is_hang, is_crash, is_sdc, is_masked, signal_init_wait_time, block, thread, instruction, pc, user_defined_string]
     return return_list
 
 
@@ -580,7 +588,7 @@ def fault_injection_by_signal(**kwargs):
             fi_tic = int(time.time())
             [register, old_val, new_val, fault_injected,
              hang, crash, masked,sdc, signal_init_time, block,
-             thread, instruction, user_defined_val] = gdb_inject_fault(**kwargs)
+             thread, instruction, pc, user_defined_val] = gdb_inject_fault(**kwargs)
 
 	 	
             # Time toc
@@ -596,7 +604,7 @@ def fault_injection_by_signal(**kwargs):
             row = [unique_id, register, num_rounds, fault_model, thread,
                        block, old_val, new_val, injection_site,
                        fault_injected, hang, crash[0], sdc, masked ,tmp,injection_time,
-                       signal_init_time, bits_to_flip, instruction, user_defined_val]
+                       signal_init_time, bits_to_flip, instruction, pc, user_defined_val]
                 
             if fault_injected:
                # output_str = "THREAD:{}, FAULT NUM:{}".format(host_thread, num_rounds)
@@ -662,6 +670,9 @@ def main():
     args = parser.parse_args()
     if args.iterations < 1:
         parser.error('Iterations must be greater than zero')
+    if (args.device>0):
+      cp.LOGS_PATH=cp.LOGS_PATH+"-"+str(args.device)    
+    cp.rewrite_path()
 
     # Start with a different seed every time to vary the random numbers generated
     # the seed will be the current number of second since 01/01/70
@@ -689,7 +700,7 @@ def main():
     # Csv log
     fieldnames = ['unique_id', 'register', 'iteration', 'fault_model', 'thread', 'block', 'old_value',
                   'new_value', 'inj_site', 'fault_successful', 'hang', 'crash', 'masked', 'sdc', 'Exception','time',
-                  'inj_time_location', 'bits_flipped', 'instruction', 'user_defined']
+                  'inj_time_location', 'bits_flipped', 'instruction', 'pc', 'user_defined']
     
     if os.path.exists("tmpxxx_num_rounds.conf"):
       mode='a'

test.sh

@@ -5,7 +5,7 @@ set -e
 #uncomment to a more verbose script
 #set -x
 
-FAULTS=1000
+FAULTS=10
 #CONFFILE=codes/matrixMul/matrixmul_16K.conf
 
 #CONFFILE=codes/mmElem/matrixmul_16K.conf
@@ -28,4 +28,4 @@ done
 echo "Fault injection finished"
 date >> tiempos
 echo "Fin..." >> tiempos
-exit 0
+exit 

test_hotspot.sh

@@ -4,8 +4,8 @@ set -e
 
 #uncomment to a more verbose script
 #set -x
-DIR_RODINIA=$HOME/rodinia_3.1/cuda
-FAULTS=1000
+DIR_RODINIA=codes
+FAULTS=100
 #CONFFILE=codes/matrixMul/matrixmul_16K.conf
 
 #CONFFILE=codes/mmElem/matrixmul_16K.conf

test_nw.sh

@@ -4,8 +4,9 @@ set -e
 
 #uncomment to a more verbose script
 #set -x
-DIR_RODINIA=$HOME/rodinia_3.1/cuda
-FAULTS=1000
+#DIR_RODINIA=$HOME/rodinia_3.1/cuda
+DIR_RODINIA=codes
+FAULTS=100
 #CONFFILE=codes/matrixMul/matrixmul_16K.conf
 
 #CONFFILE=codes/mmElem/matrixmul_16K.conf