From 4b3f76e5ad6bdde7af15559401c47d3a64262db9 Mon Sep 17 00:00:00 2001
From: Luca Colagrande <luca.colagrande3@gmail.com>
Date: Thu, 22 Feb 2024 14:41:45 +0100
Subject: [PATCH] ipc: Handle failing RTL simulations (#97)

---
 target/common/test/SnitchSim.py | 34 +++++++++++++++++++++++++++++++--
 1 file changed, 32 insertions(+), 2 deletions(-)

diff --git a/target/common/test/SnitchSim.py b/target/common/test/SnitchSim.py
index c5d12536f..d98c1ec14 100755
--- a/target/common/test/SnitchSim.py
+++ b/target/common/test/SnitchSim.py
@@ -4,6 +4,7 @@
 # SPDX-License-Identifier: SHL-0.51
 #
 # Paul Scheffler <paulsc@iis.ee.ethz.ch>
+# Luca Colagrande <colluca@iis.ee.ethz.ch>
 #
 # This class implements a minimal wrapping IPC server for `tb_lib`.
 # `__main__` shows a demonstrator for it, running a simulation and accessing its memory.
@@ -14,6 +15,12 @@
 import subprocess
 import struct
 import functools
+import threading
+import time
+import signal
+
+# Simulation monitor polling period (in seconds)
+SIM_MONITOR_POLL_PERIOD = 2
 
 
 class SnitchSim:
@@ -38,6 +45,10 @@ def start(self):
         # Open FIFOs
         self.tx = open(tx_fd, 'wb', buffering=0)  # Unbuffered
         self.rx = open(rx_fd, 'rb')
+        # Create thread to monitor simulation
+        self.stop_sim_monitor = threading.Event()
+        self.sim_monitor = threading.Thread(target=self.__monitor_sim)
+        self.sim_monitor.start()
 
     def __sim_active(func):
         @functools.wraps(func)
@@ -45,9 +56,23 @@ def inner(self, *args, **kwargs):
             if self.sim is None:
                 raise RuntimeError(f'Snitch is not running (simulation `{self.sim_bin}`'
                                    f'binary `{self.snitch_bin}`)')
-            return func(self, *args, **kwargs)
+            # Catch SIGINT raised by simulation monitor
+            try:
+                return func(self, *args, **kwargs)
+            except KeyboardInterrupt:
+                print('Simulation monitor detected a simulation failure.')
+                self.stop_sim_monitor.set()
+                self.sim_monitor.join()
+                sys.exit(1)
         return inner
 
+    def __monitor_sim(self):
+        while not self.stop_sim_monitor.is_set():
+            if self.sim.poll() is not None:
+                # Raise SIGINT to interrupt main thread, could be blocked on a read
+                os.kill(os.getpid(), signal.SIGINT)
+            time.sleep(SIM_MONITOR_POLL_PERIOD)
+
     @__sim_active
     def read(self, addr: int, length: int) -> bytes:
         op = struct.pack('=QQQ', 0, addr, length)
@@ -73,15 +98,20 @@ def poll(self, addr: int, mask32: int, exp32: int):
         bytestring = self.rx.read(4)
         return int.from_bytes(bytestring, byteorder='little')
 
-    # Simulator can exit only once TX FIFO closes
     @__sim_active
     def finish(self, wait_for_sim: bool = True):
+        # Close FIFOs (simulator can exit only once TX FIFO closes)
         self.rx.close()
         self.tx.close()
+        # Close simulation monitor
+        self.stop_sim_monitor.set()
+        self.sim_monitor.join()
+        # Wait for simulation or terminate
         if (wait_for_sim):
             self.sim.wait()
         else:
             self.sim.terminate()
+        # Cleanup
         self.tmpdir.cleanup()
         self.sim = None