From 501de306269eab0c386580827def702d3d9a46e9 Mon Sep 17 00:00:00 2001
From: whitequark <whitequark@whitequark.org>
Date: Fri, 18 Mar 2016 22:29:42 +0000
Subject: [PATCH] Report watchdog expiration and RTIO clock failure as
 exceptions.

Fixes #316.
---
 artiq/coredevice/comm_generic.py |  7 ++++++
 artiq/coredevice/exceptions.py   |  6 +++++
 artiq/runtime/net_server.c       |  8 ++++---
 artiq/runtime/net_server.h       |  2 +-
 artiq/runtime/session.c          | 39 ++++++++++++++++++++------------
 artiq/runtime/session.h          |  2 +-
 6 files changed, 44 insertions(+), 20 deletions(-)

diff --git a/artiq/coredevice/comm_generic.py b/artiq/coredevice/comm_generic.py
index f4e1d2eee..31f3ccccb 100644
--- a/artiq/coredevice/comm_generic.py
+++ b/artiq/coredevice/comm_generic.py
@@ -50,6 +50,9 @@ class _D2HMsgType(Enum):
     FLASH_OK_REPLY = 12
     FLASH_ERROR_REPLY = 13
 
+    WATCHDOG_EXPIRED = 14
+    CLOCK_FAILURE = 15
+
 
 class UnsupportedDevice(Exception):
     pass
@@ -522,6 +525,10 @@ class CommGeneric:
                 self._serve_rpc(object_map)
             elif self._read_type == _D2HMsgType.KERNEL_EXCEPTION:
                 self._serve_exception(object_map, symbolizer)
+            elif self._read_type == _D2HMsgType.WATCHDOG_EXPIRED:
+                raise exceptions.WatchdogExpired
+            elif self._read_type == _D2HMsgType.CLOCK_FAILURE:
+                raise exceptions.ClockFailure
             else:
                 self._read_expect(_D2HMsgType.KERNEL_FINISHED)
                 return
diff --git a/artiq/coredevice/exceptions.py b/artiq/coredevice/exceptions.py
index 6317edfe6..5c43bbd6d 100644
--- a/artiq/coredevice/exceptions.py
+++ b/artiq/coredevice/exceptions.py
@@ -129,3 +129,9 @@ class DDSBatchError(Exception):
 class I2CError(Exception):
     """Raised with a I2C transaction fails."""
     artiq_builtin = True
+
+class WatchdogExpired(Exception):
+    """Raised when a watchdog expires."""
+
+class ClockFailure(Exception):
+    """Raised when RTIO PLL is unable to lock."""
diff --git a/artiq/runtime/net_server.c b/artiq/runtime/net_server.c
index dc27f7e4b..88e69f0a1 100644
--- a/artiq/runtime/net_server.c
+++ b/artiq/runtime/net_server.c
@@ -158,9 +158,9 @@ static void tcp_pcb_service(void *arg, struct tcp_pcb *pcb)
     /* Writer interface */
     if(cs == instance->open_session_cs) {
         void *data;
-        int len, sndbuf;
+        int len, sndbuf, close_flag;
 
-        cs->instance->poll(&data, &len);
+        cs->instance->poll(&data, &len, &close_flag);
         if(len > 0) {
             sndbuf = tcp_sndbuf(pcb);
             if(len > sndbuf)
@@ -168,8 +168,10 @@ static void tcp_pcb_service(void *arg, struct tcp_pcb *pcb)
             tcp_write(pcb, data, len, 0);
             instance->ack_consumed(len);
         }
-        if(len < 0)
+        if(close_flag) {
+            tcp_output(pcb);
             net_server_close(cs, pcb);
+        }
     }
 }
 
diff --git a/artiq/runtime/net_server.h b/artiq/runtime/net_server.h
index d3f6a8726..9bb57d6c6 100644
--- a/artiq/runtime/net_server.h
+++ b/artiq/runtime/net_server.h
@@ -10,7 +10,7 @@ struct net_server_instance {
     void (*start)(void);
     void (*end)(void);
     int (*input)(void *data, int length);
-    void (*poll)(void **data, int *length);
+    void (*poll)(void **data, int *length, int *close_flag);
     void (*ack_consumed)(int length);
     void (*ack_sent)(int length);
 
diff --git a/artiq/runtime/session.c b/artiq/runtime/session.c
index fb6e4dcaf..f2f8a8881 100644
--- a/artiq/runtime/session.c
+++ b/artiq/runtime/session.c
@@ -392,7 +392,10 @@ enum {
 
     REMOTEMSG_TYPE_FLASH_READ_REPLY,
     REMOTEMSG_TYPE_FLASH_OK_REPLY,
-    REMOTEMSG_TYPE_FLASH_ERROR_REPLY
+    REMOTEMSG_TYPE_FLASH_ERROR_REPLY,
+
+    REMOTEMSG_TYPE_WATCHDOG_EXPIRED,
+    REMOTEMSG_TYPE_CLOCK_FAILURE,
 };
 
 static int receive_rpc_value(const char **tag, void **slot);
@@ -1083,30 +1086,36 @@ int session_input(void *data, int length)
 /* *length is set to -1 in case of irrecoverable error
  * (the session must be dropped and session_end called)
  */
-void session_poll(void **data, int *length)
+void session_poll(void **data, int *length, int *close_flag)
 {
+    *close_flag = 0;
+
     if(user_kernel_state == USER_KERNEL_RUNNING) {
         if(watchdog_expired()) {
             core_log("Watchdog expired\n");
-            *length = -1;
-            return;
+
+            *close_flag = 1;
+            out_packet_empty(REMOTEMSG_TYPE_WATCHDOG_EXPIRED);
         }
         if(!rtiocrg_check()) {
             core_log("RTIO clock failure\n");
-            *length = -1;
-            return;
+
+            *close_flag = 1;
+            out_packet_empty(REMOTEMSG_TYPE_CLOCK_FAILURE);
         }
     }
 
-    /* If the output buffer is available,
-     * check if the kernel CPU has something to transmit.
-     */
-    if(out_packet_available()) {
-        struct msg_base *umsg = mailbox_receive();
-        if(umsg) {
-            if(!process_kmsg(umsg)) {
-                *length = -1;
-                return;
+    if(!*close_flag) {
+        /* If the output buffer is available,
+         * check if the kernel CPU has something to transmit.
+         */
+        if(out_packet_available()) {
+            struct msg_base *umsg = mailbox_receive();
+            if(umsg) {
+                if(!process_kmsg(umsg)) {
+                    *length = -1;
+                    return;
+                }
             }
         }
     }
diff --git a/artiq/runtime/session.h b/artiq/runtime/session.h
index 4184b1758..06ee421ed 100644
--- a/artiq/runtime/session.h
+++ b/artiq/runtime/session.h
@@ -6,7 +6,7 @@ void session_start(void);
 void session_end(void);
 
 int session_input(void *data, int length);
-void session_poll(void **data, int *length);
+void session_poll(void **data, int *length, int *close_flag);
 void session_ack_consumed(int length);
 void session_ack_sent(int length);