Merge branch 'odp_fixes' into rdma.git for-next

Jason Gunthorpe says:

====================
This is a collection of general cleanups for ODP to clarify some of the
flows around umem creation and use of the interval tree.
====================

The branch is based on v5.3-rc5 due to dependencies

* odp_fixes:
  RDMA/mlx5: Use odp instead of mr->umem in pagefault_mr
  RDMA/mlx5: Use ib_umem_start instead of umem.address
  RDMA/core: Make invalidate_range a device operation
  RDMA/odp: Use kvcalloc for the dma_list and page_list
  RDMA/odp: Check for overflow when computing the umem_odp end
  RDMA/odp: Provide ib_umem_odp_release() to undo the allocs
  RDMA/odp: Split creating a umem_odp from ib_umem_get
  RDMA/odp: Make the three ways to create a umem_odp clear
  RMDA/odp: Consolidate umem_odp initialization
  RDMA/odp: Make it clearer when a umem is an implicit ODP umem
  RDMA/odp: Iterate over the whole rbtree directly
  RDMA/odp: Use the common interval tree library instead of generic
  RDMA/mlx5: Fix MR npages calculation for IB_ACCESS_HUGETLB

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

1 files changed, 69 insertions, 34 deletions

diff --git a/drivers/infiniband/core/device.c b/drivers/infiniband/core/device.c
index 8892862fb759..99c4a55545cf 100644
--- a/drivers/infiniband/core/device.c
+++ b/drivers/infiniband/core/device.c
@@ -93,11 +93,17 @@ static DEFINE_XARRAY_FLAGS(devices, XA_FLAGS_ALLOC);
 static DECLARE_RWSEM(devices_rwsem);
 #define DEVICE_REGISTERED XA_MARK_1
 
-static LIST_HEAD(client_list);
+static u32 highest_client_id;
 #define CLIENT_REGISTERED XA_MARK_1
 static DEFINE_XARRAY_FLAGS(clients, XA_FLAGS_ALLOC);
 static DECLARE_RWSEM(clients_rwsem);
 
+static void ib_client_put(struct ib_client *client)
+{
+	if (refcount_dec_and_test(&client->uses))
+		complete(&client->uses_zero);
+}
+
 /*
  * If client_data is registered then the corresponding client must also still
  * be registered.
@@ -653,6 +659,14 @@ static int add_client_context(struct ib_device *device,
 
 	down_write(&device->client_data_rwsem);
 	/*
+	 * So long as the client is registered hold both the client and device
+	 * unregistration locks.
+	 */
+	if (!refcount_inc_not_zero(&client->uses))
+		goto out_unlock;
+	refcount_inc(&device->refcount);
+
+	/*
 	 * Another caller to add_client_context got here first and has already
 	 * completely initialized context.
 	 */
@@ -675,6 +689,9 @@ static int add_client_context(struct ib_device *device,
 	return 0;
 
 out:
+	ib_device_put(device);
+	ib_client_put(client);
+out_unlock:
 	up_write(&device->client_data_rwsem);
 	return ret;
 }
@@ -694,7 +711,7 @@ static void remove_client_context(struct ib_device *device,
 	client_data = xa_load(&device->client_data, client_id);
 	xa_clear_mark(&device->client_data, client_id, CLIENT_DATA_REGISTERED);
 	client = xa_load(&clients, client_id);
-	downgrade_write(&device->client_data_rwsem);
+	up_write(&device->client_data_rwsem);
 
 	/*
 	 * Notice we cannot be holding any exclusive locks when calling the
@@ -704,17 +721,13 @@ static void remove_client_context(struct ib_device *device,
 	 *
 	 * For this reason clients and drivers should not call the
 	 * unregistration functions will holdling any locks.
-	 *
-	 * It tempting to drop the client_data_rwsem too, but this is required
-	 * to ensure that unregister_client does not return until all clients
-	 * are completely unregistered, which is required to avoid module
-	 * unloading races.
 	 */
 	if (client->remove)
 		client->remove(device, client_data);
 
 	xa_erase(&device->client_data, client_id);
-	up_read(&device->client_data_rwsem);
+	ib_device_put(device);
+	ib_client_put(client);
 }
 
 static int alloc_port_data(struct ib_device *device)
@@ -1223,7 +1236,7 @@ static int setup_device(struct ib_device *device)
 
 static void disable_device(struct ib_device *device)
 {
-	struct ib_client *client;
+	u32 cid;
 
 	WARN_ON(!refcount_read(&device->refcount));
 
@@ -1231,10 +1244,19 @@ static void disable_device(struct ib_device *device)
 	xa_clear_mark(&devices, device->index, DEVICE_REGISTERED);
 	up_write(&devices_rwsem);
 
+	/*
+	 * Remove clients in LIFO order, see assign_client_id. This could be
+	 * more efficient if xarray learns to reverse iterate. Since no new
+	 * clients can be added to this ib_device past this point we only need
+	 * the maximum possible client_id value here.
+	 */
 	down_read(&clients_rwsem);
-	list_for_each_entry_reverse(client, &client_list, list)
-		remove_client_context(device, client->client_id);
+	cid = highest_client_id;
 	up_read(&clients_rwsem);
+	while (cid) {
+		cid--;
+		remove_client_context(device, cid);
+	}
 
 	/* Pairs with refcount_set in enable_device */
 	ib_device_put(device);
@@ -1661,30 +1683,31 @@ static int assign_client_id(struct ib_client *client)
 	/*
 	 * The add/remove callbacks must be called in FIFO/LIFO order. To
 	 * achieve this we assign client_ids so they are sorted in
-	 * registration order, and retain a linked list we can reverse iterate
-	 * to get the LIFO order. The extra linked list can go away if xarray
-	 * learns to reverse iterate.
+	 * registration order.
 	 */
-	if (list_empty(&client_list)) {
-		client->client_id = 0;
-	} else {
-		struct ib_client *last;
-
-		last = list_last_entry(&client_list, struct ib_client, list);
-		client->client_id = last->client_id + 1;
-	}
+	client->client_id = highest_client_id;
 	ret = xa_insert(&clients, client->client_id, client, GFP_KERNEL);
 	if (ret)
 		goto out;
 
+	highest_client_id++;
 	xa_set_mark(&clients, client->client_id, CLIENT_REGISTERED);
-	list_add_tail(&client->list, &client_list);
 
 out:
 	up_write(&clients_rwsem);
 	return ret;
 }
 
+static void remove_client_id(struct ib_client *client)
+{
+	down_write(&clients_rwsem);
+	xa_erase(&clients, client->client_id);
+	for (; highest_client_id; highest_client_id--)
+		if (xa_load(&clients, highest_client_id - 1))
+			break;
+	up_write(&clients_rwsem);
+}
+
 /**
  * ib_register_client - Register an IB client
  * @client:Client to register
@@ -1704,6 +1727,8 @@ int ib_register_client(struct ib_client *client)
 	unsigned long index;
 	int ret;
 
+	refcount_set(&client->uses, 1);
+	init_completion(&client->uses_zero);
 	ret = assign_client_id(client);
 	if (ret)
 		return ret;
@@ -1739,21 +1764,30 @@ void ib_unregister_client(struct ib_client *client)
 	unsigned long index;
 
 	down_write(&clients_rwsem);
+	ib_client_put(client);
 	xa_clear_mark(&clients, client->client_id, CLIENT_REGISTERED);
 	up_write(&clients_rwsem);
-	/*
-	 * Every device still known must be serialized to make sure we are
-	 * done with the client callbacks before we return.
-	 */
-	down_read(&devices_rwsem);
-	xa_for_each (&devices, index, device)
+
+	/* We do not want to have locks while calling client->remove() */
+	rcu_read_lock();
+	xa_for_each (&devices, index, device) {
+		if (!ib_device_try_get(device))
+			continue;
+		rcu_read_unlock();
+
 		remove_client_context(device, client->client_id);
-	up_read(&devices_rwsem);
 
-	down_write(&clients_rwsem);
-	list_del(&client->list);
-	xa_erase(&clients, client->client_id);
-	up_write(&clients_rwsem);
+		ib_device_put(device);
+		rcu_read_lock();
+	}
+	rcu_read_unlock();
+
+	/*
+	 * remove_client_context() is not a fence, it can return even though a
+	 * removal is ongoing. Wait until all removals are completed.
+	 */
+	wait_for_completion(&client->uses_zero);
+	remove_client_id(client);
 }
 EXPORT_SYMBOL(ib_unregister_client);
 
@@ -2582,6 +2616,7 @@ void ib_set_device_ops(struct ib_device *dev, const struct ib_device_ops *ops)
 	SET_DEVICE_OP(dev_ops, get_vf_config);
 	SET_DEVICE_OP(dev_ops, get_vf_stats);
 	SET_DEVICE_OP(dev_ops, init_port);
+	SET_DEVICE_OP(dev_ops, invalidate_range);
 	SET_DEVICE_OP(dev_ops, iw_accept);
 	SET_DEVICE_OP(dev_ops, iw_add_ref);
 	SET_DEVICE_OP(dev_ops, iw_connect);