RDMA/cxgb4: release hw resources on device removal

index feeb8ee6f4a2..44161ca4d2a8 100644
@@ -875,6 +875,11 @@ static int c4iw_rdev_open(struct c4iw_rdev *rdev)
 
 	rdev->status_page->db_off = 0;
 
+	init_completion(&rdev->rqt_compl);
+	init_completion(&rdev->pbl_compl);
+	kref_init(&rdev->rqt_kref);
+	kref_init(&rdev->pbl_kref);
+
 	return 0;
 err_free_status_page_and_wr_log:
 	if (c4iw_wr_log && rdev->wr_log)

@@ -893,13 +898,15 @@ static int c4iw_rdev_open(struct c4iw_rdev *rdev)
 
 static void c4iw_rdev_close(struct c4iw_rdev *rdev)
 {
-	destroy_workqueue(rdev->free_workq);
 	kfree(rdev->wr_log);
 	c4iw_release_dev_ucontext(rdev, &rdev->uctx);
 	free_page((unsigned long)rdev->status_page);
 	c4iw_pblpool_destroy(rdev);
 	c4iw_rqtpool_destroy(rdev);
+	wait_for_completion(&rdev->pbl_compl);
+	wait_for_completion(&rdev->rqt_compl);
 	c4iw_ocqp_pool_destroy(rdev);
+	destroy_workqueue(rdev->free_workq);
 	c4iw_destroy_resource(&rdev->resource);
 }
 

index cc929002c05e..a60def23e9ef 100644
@@ -185,6 +185,10 @@ struct c4iw_rdev {
 	struct wr_log_entry *wr_log;
 	int wr_log_size;
 	struct workqueue_struct *free_workq;
+	struct completion rqt_compl;
+	struct completion pbl_compl;
+	struct kref rqt_kref;
+	struct kref pbl_kref;
 };
 
 static inline int c4iw_fatal_error(struct c4iw_rdev *rdev)

index 3cf25997ed2b..0ef25ae05e6f 100644
@@ -260,12 +260,22 @@ u32 c4iw_pblpool_alloc(struct c4iw_rdev *rdev, int size)
 		rdev->stats.pbl.cur += roundup(size, 1 << MIN_PBL_SHIFT);
 		if (rdev->stats.pbl.cur > rdev->stats.pbl.max)
 			rdev->stats.pbl.max = rdev->stats.pbl.cur;
+		kref_get(&rdev->pbl_kref);
 	} else
 		rdev->stats.pbl.fail++;
 	mutex_unlock(&rdev->stats.lock);
 	return (u32)addr;
 }
 
+static void destroy_pblpool(struct kref *kref)
+{
+	struct c4iw_rdev *rdev;
+
+	rdev = container_of(kref, struct c4iw_rdev, pbl_kref);
+	gen_pool_destroy(rdev->pbl_pool);
+	complete(&rdev->pbl_compl);
+}
+
 void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
 {
 	pr_debug("addr 0x%x size %d\n", addr, size);

@@ -273,6 +283,7 @@ void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
 	rdev->stats.pbl.cur -= roundup(size, 1 << MIN_PBL_SHIFT);
 	mutex_unlock(&rdev->stats.lock);
 	gen_pool_free(rdev->pbl_pool, (unsigned long)addr, size);
+	kref_put(&rdev->pbl_kref, destroy_pblpool);
 }
 
 int c4iw_pblpool_create(struct c4iw_rdev *rdev)

@@ -310,7 +321,7 @@ int c4iw_pblpool_create(struct c4iw_rdev *rdev)
 
 void c4iw_pblpool_destroy(struct c4iw_rdev *rdev)
 {
-	gen_pool_destroy(rdev->pbl_pool);
+	kref_put(&rdev->pbl_kref, destroy_pblpool);
 }
 
 /*

@@ -331,12 +342,22 @@ u32 c4iw_rqtpool_alloc(struct c4iw_rdev *rdev, int size)
 		rdev->stats.rqt.cur += roundup(size << 6, 1 << MIN_RQT_SHIFT);
 		if (rdev->stats.rqt.cur > rdev->stats.rqt.max)
 			rdev->stats.rqt.max = rdev->stats.rqt.cur;
+		kref_get(&rdev->rqt_kref);
 	} else
 		rdev->stats.rqt.fail++;
 	mutex_unlock(&rdev->stats.lock);
 	return (u32)addr;
 }
 
+static void destroy_rqtpool(struct kref *kref)
+{
+	struct c4iw_rdev *rdev;
+
+	rdev = container_of(kref, struct c4iw_rdev, rqt_kref);
+	gen_pool_destroy(rdev->rqt_pool);
+	complete(&rdev->rqt_compl);
+}
+
 void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
 {
 	pr_debug("addr 0x%x size %d\n", addr, size << 6);

@@ -344,6 +365,7 @@ void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
 	rdev->stats.rqt.cur -= roundup(size << 6, 1 << MIN_RQT_SHIFT);
 	mutex_unlock(&rdev->stats.lock);
 	gen_pool_free(rdev->rqt_pool, (unsigned long)addr, size << 6);
+	kref_put(&rdev->rqt_kref, destroy_rqtpool);
 }
 
 int c4iw_rqtpool_create(struct c4iw_rdev *rdev)

@@ -380,7 +402,7 @@ int c4iw_rqtpool_create(struct c4iw_rdev *rdev)
 
 void c4iw_rqtpool_destroy(struct c4iw_rdev *rdev)
 {
-	gen_pool_destroy(rdev->rqt_pool);
+	kref_put(&rdev->rqt_kref, destroy_rqtpool);
 }
 
 /*