Skip to contentSkip to conversation

ipmi: Add limits to event and receive message requests

PATCH New Corey Minyard 1 participant 1 message Apr 28

Subsystems (2)
lore
Share & tools
lore

Corey Minyard

author maintainer
#0 lore raw

PATCH v3 1/1 ipmi: Add limits to event and receive message requests

The driver would just fetch events and receive messages until the BMC said it was done. To avoid issues with BMCs that never say they are done, add a limit of 10 fetches at a time.

In addition, an si interface has an attn state it can return from the hardware which is supposed to cause a flag fetch to see if the driver needs to fetch events or message or a few other things. If the attn bit gets stuck, it's a similar problem. So allow messages in between flag fetches so the driver itself doesn't get stuck.

This is a more general fix than the previous fix for the specific bad BMC, but should fix the more general issue of a BMC that won't stop saying it has data.

This has been there from the beginning of the driver. It's not a bug per-se, but it is accounting for bugs in BMCs.

---
I have added this problem as a capability in the openipmi library simulator so I can reproduce the issue and make sure everything works properly.

 drivers/char/ipmi/ipmi_si_intf.c | 54 +++++++++++++++++++++++++-------
 drivers/char/ipmi/ipmi_ssif.c    | 23 ++++++++++++--
 2 files changed, 64 insertions(+), 13 deletions(-)
drivers/char/ipmi/ipmi_si_intf.c +43 -11
index 08c208cc64c5..7c3c463e08da 100644
@@ -168,6 +168,10 @@ struct smi_info {
 			     OEM2_DATA_AVAIL)
 	unsigned char       msg_flags;
 
+	/* When requesting events and messages, don't do it forever. */
+	unsigned int        num_requests_in_a_row;
+	bool		    last_was_flag_fetch;
+
 	/* Does the BMC have an event buffer? */
 	bool		    has_event_buffer;
 
@@ -410,7 +414,10 @@ static void start_getting_msg_queue(struct smi_info *smi_info)
 
 	start_new_msg(smi_info, smi_info->curr_msg->data,
 		      smi_info->curr_msg->data_size);
-	smi_info->si_state = SI_GETTING_MESSAGES;
+	if (smi_info->si_state != SI_GETTING_MESSAGES) {
+		smi_info->num_requests_in_a_row = 0;
+		smi_info->si_state = SI_GETTING_MESSAGES;
+	}
 }
 
 static void start_getting_events(struct smi_info *smi_info)
@@ -421,7 +428,10 @@ static void start_getting_events(struct smi_info *smi_info)
 
 	start_new_msg(smi_info, smi_info->curr_msg->data,
 		      smi_info->curr_msg->data_size);
-	smi_info->si_state = SI_GETTING_EVENTS;
+	if (smi_info->si_state != SI_GETTING_EVENTS) {
+		smi_info->num_requests_in_a_row = 0;
+		smi_info->si_state = SI_GETTING_EVENTS;
+	}
 }
 
 /*
@@ -595,6 +605,7 @@ static void handle_transaction_done(struct smi_info *smi_info)
 			smi_info->si_state = SI_NORMAL;
 		} else {
 			smi_info->msg_flags = msg[3];
+			smi_info->last_was_flag_fetch = true;
 			handle_flags(smi_info);
 		}
 		break;
@@ -646,6 +657,11 @@ static void handle_transaction_done(struct smi_info *smi_info)
 		} else {
 			smi_inc_stat(smi_info, events);
 
+			smi_info->num_requests_in_a_row++;
+			if (smi_info->num_requests_in_a_row > 10)
+				/* Stop if we do this too many times. */
+				smi_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
+
 			/*
 			 * Do this before we deliver the message
 			 * because delivering the message releases the
@@ -684,6 +700,11 @@ static void handle_transaction_done(struct smi_info *smi_info)
 		} else {
 			smi_inc_stat(smi_info, incoming_messages);
 
+			smi_info->num_requests_in_a_row++;
+			if (smi_info->num_requests_in_a_row > 10)
+				/* Stop if we do this too many times. */
+				smi_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
+
 			/*
 			 * Do this before we deliver the message
 			 * because delivering the message releases the
@@ -825,6 +846,26 @@ static enum si_sm_result smi_event_handler(struct smi_info *smi_info,
 		goto out;
 	}
 
+	/*
+	 * If we are currently idle, or if the last thing that was
+	 * done was a flag fetch and there is a message pending, try
+	 * to start the next message.
+	 *
+	 * We do the waiting message check to avoid a stuck flag
+	 * completely wedging the driver.  Let a message through
+	 * in between flag operations if that happens.
+	 */
+	if (si_sm_result == SI_SM_IDLE ||
+	    (si_sm_result == SI_SM_ATTN && smi_info->waiting_msg &&
+	     smi_info->last_was_flag_fetch)) {
+		smi_info->last_was_flag_fetch = false;
+		smi_inc_stat(smi_info, idles);
+
+		si_sm_result = start_next_msg(smi_info);
+		if (si_sm_result != SI_SM_IDLE)
+			goto restart;
+	}
+
 	/*
 	 * We prefer handling attn over new messages.  But don't do
 	 * this if there is not yet an upper layer to handle anything.
@@ -852,15 +893,6 @@ static enum si_sm_result smi_event_handler(struct smi_info *smi_info,
 		}
 	}
 
-	/* If we are currently idle, try to start the next message. */
-	if (si_sm_result == SI_SM_IDLE) {
-		smi_inc_stat(smi_info, idles);
-
-		si_sm_result = start_next_msg(smi_info);
-		if (si_sm_result != SI_SM_IDLE)
-			goto restart;
-	}
-
 	if ((si_sm_result == SI_SM_IDLE)
 	    && (atomic_read(&smi_info->req_events))) {
 		/*
drivers/char/ipmi/ipmi_ssif.c +21 -2
index b49500a1bd36..f3798f4e6a63 100644
@@ -225,6 +225,9 @@ struct ssif_info {
 	bool		    has_event_buffer;
 	bool		    supports_alert;
 
+	/* When requesting events and messages, don't do it forever. */
+	unsigned int        num_requests_in_a_row;
+
 	/*
 	 * Used to tell what we should do with alerts.  If we are
 	 * waiting on a response, read the data immediately.
@@ -413,7 +416,10 @@ static void start_event_fetch(struct ssif_info *ssif_info, unsigned long *flags)
 	}
 
 	ssif_info->curr_msg = msg;
-	ssif_info->ssif_state = SSIF_GETTING_EVENTS;
+	if (ssif_info->ssif_state != SSIF_GETTING_EVENTS) {
+		ssif_info->num_requests_in_a_row = 0;
+		ssif_info->ssif_state = SSIF_GETTING_EVENTS;
+	}
 	ipmi_ssif_unlock_cond(ssif_info, flags);
 
 	msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
@@ -436,7 +442,10 @@ static void start_recv_msg_fetch(struct ssif_info *ssif_info,
 	}
 
 	ssif_info->curr_msg = msg;
-	ssif_info->ssif_state = SSIF_GETTING_MESSAGES;
+	if (ssif_info->ssif_state != SSIF_GETTING_MESSAGES) {
+		ssif_info->num_requests_in_a_row = 0;
+		ssif_info->ssif_state = SSIF_GETTING_MESSAGES;
+	}
 	ipmi_ssif_unlock_cond(ssif_info, flags);
 
 	msg->data[0] = (IPMI_NETFN_APP_REQUEST << 2);
@@ -843,6 +852,11 @@ static void msg_done_handler(struct ssif_info *ssif_info, int result,
 			ssif_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
 			handle_flags(ssif_info, flags);
 		} else {
+			ssif_info->num_requests_in_a_row++;
+			if (ssif_info->num_requests_in_a_row > 10)
+				/* Stop if we do this too many times. */
+				ssif_info->msg_flags &= ~EVENT_MSG_BUFFER_FULL;
+
 			handle_flags(ssif_info, flags);
 			ssif_inc_stat(ssif_info, events);
 			deliver_recv_msg(ssif_info, msg);
@@ -876,6 +890,11 @@ static void msg_done_handler(struct ssif_info *ssif_info, int result,
 			ssif_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
 			handle_flags(ssif_info, flags);
 		} else {
+			ssif_info->num_requests_in_a_row++;
+			if (ssif_info->num_requests_in_a_row > 10)
+				/* Stop if we do this too many times. */
+				ssif_info->msg_flags &= ~RECEIVE_MSG_AVAIL;
+
 			ssif_inc_stat(ssif_info, incoming_messages);
 			handle_flags(ssif_info, flags);
 			deliver_recv_msg(ssif_info, msg);
Signature · 1 line
2.43.0