4302 stack overflow from squeue_drain()

Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Dan McDonald <danmcd@nexenta.com>
Reviewed by: Sebastien Roy <seb@delphix.com>
Approved by: Eric Schrock <eric.schrock@delphix.com>

5 files changed, 31 insertions, 0 deletions

diff --git a/usr/src/uts/common/inet/squeue.c b/usr/src/uts/common/inet/squeue.c
index 4108931c36..2e08dc359b 100644
--- a/usr/src/uts/common/inet/squeue.c
+++ b/usr/src/uts/common/inet/squeue.c
@@ -23,6 +23,10 @@
  */
 
 /*
+ * Copyright 2012 Joyent, Inc.  All rights reserved.
+ */
+
+/*
  * Squeues: General purpose serialization mechanism
  * ------------------------------------------------
  *
@@ -120,6 +124,8 @@
 #include <sys/sdt.h>
 #include <sys/ddi.h>
 #include <sys/sunddi.h>
+#include <sys/stack.h>
+#include <sys/archsystm.h>
 
 #include <inet/ipclassifier.h>
 #include <inet/udp_impl.h>
@@ -142,6 +148,9 @@ int squeue_workerwait_ms = 0;
 static int squeue_drain_ns = 0;
 static int squeue_workerwait_tick = 0;
 
+uintptr_t squeue_drain_stack_needed = 10240;
+uint_t squeue_drain_stack_toodeep;
+
 #define	MAX_BYTES_TO_PICKUP	150000
 
 #define	ENQUEUE_CHAIN(sqp, mp, tail, cnt) {			\
@@ -712,6 +721,20 @@ squeue_drain(squeue_t *sqp, uint_t proc_type, hrtime_t expire)
 	boolean_t	sq_poll_capable;
 	ip_recv_attr_t	*ira, iras;
 
+	/*
+	 * Before doing any work, check our stack depth; if we're not a
+	 * worker thread for this squeue and we're beginning to get tight on
+	 * on stack, kick the worker, bump a counter and return.
+	 */
+	if (proc_type != SQS_WORKER && STACK_BIAS + (uintptr_t)getfp() -
+	    (uintptr_t)curthread->t_stkbase < squeue_drain_stack_needed) {
+		ASSERT(mutex_owned(&sqp->sq_lock));
+		sqp->sq_awaken = ddi_get_lbolt();
+		cv_signal(&sqp->sq_worker_cv);
+		squeue_drain_stack_toodeep++;
+		return;
+	}
+
 	sq_poll_capable = (sqp->sq_state & SQS_POLL_CAPAB) != 0;
 again:
 	ASSERT(mutex_owned(&sqp->sq_lock));
diff --git a/usr/src/uts/intel/ip/ip.global-objs.debug64 b/usr/src/uts/intel/ip/ip.global-objs.debug64
index ed9165dc7f..b05d399caf 100644
--- a/usr/src/uts/intel/ip/ip.global-objs.debug64
+++ b/usr/src/uts/intel/ip/ip.global-objs.debug64
@@ -254,6 +254,8 @@ sqset_lock
 squeue_cache
 squeue_drain_ms
 squeue_drain_ns
+squeue_drain_stack_needed
+squeue_drain_stack_toodeep
 squeue_workerwait_ms
 squeue_workerwait_tick
 tcp_acceptor_rinit
diff --git a/usr/src/uts/intel/ip/ip.global-objs.obj64 b/usr/src/uts/intel/ip/ip.global-objs.obj64
index 11a0ba651f..7dd30502a0 100644
--- a/usr/src/uts/intel/ip/ip.global-objs.obj64
+++ b/usr/src/uts/intel/ip/ip.global-objs.obj64
@@ -251,6 +251,8 @@ sqset_lock
 squeue_cache
 squeue_drain_ms
 squeue_drain_ns
+squeue_drain_stack_needed
+squeue_drain_stack_toodeep
 squeue_workerwait_ms
 squeue_workerwait_tick
 tcp_acceptor_rinit
diff --git a/usr/src/uts/sparc/ip/ip.global-objs.debug64 b/usr/src/uts/sparc/ip/ip.global-objs.debug64
index ed9165dc7f..b05d399caf 100644
--- a/usr/src/uts/sparc/ip/ip.global-objs.debug64
+++ b/usr/src/uts/sparc/ip/ip.global-objs.debug64
@@ -254,6 +254,8 @@ sqset_lock
 squeue_cache
 squeue_drain_ms
 squeue_drain_ns
+squeue_drain_stack_needed
+squeue_drain_stack_toodeep
 squeue_workerwait_ms
 squeue_workerwait_tick
 tcp_acceptor_rinit
diff --git a/usr/src/uts/sparc/ip/ip.global-objs.obj64 b/usr/src/uts/sparc/ip/ip.global-objs.obj64
index 11a0ba651f..7dd30502a0 100644
--- a/usr/src/uts/sparc/ip/ip.global-objs.obj64
+++ b/usr/src/uts/sparc/ip/ip.global-objs.obj64
@@ -251,6 +251,8 @@ sqset_lock
 squeue_cache
 squeue_drain_ms
 squeue_drain_ns
+squeue_drain_stack_needed
+squeue_drain_stack_toodeep
 squeue_workerwait_ms
 squeue_workerwait_tick
 tcp_acceptor_rinit