Skip to contentSkip to conversation

Separate user/kernel QP creation logic

PATCH In review Leon Romanovsky 3 participants 20 messages Jul 21, 2021

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover Separate user/kernel QP creation logic

From: Leon Romanovsky <[email protected]>

Hi,

The "QP allocation" series shows clearly how convoluted the create QP flow and especially in XRC_TGT flows, where it called to kernel verb just to pass some parameters as NULL to the user create QP verb.

This series is a small step to make clean XRC_TGT flow by providing more clean user/kernel create QP verb separation.

It is based on the "QP allocation" series.

Thanks

Leon Romanovsky (7):
RDMA/mlx5: Delete not-available udata check RDMA/core: Delete duplicated and unreachable code RDMA/core: Remove protection from wrong in-kernel API usage RDMA/core: Reorganize create QP low-level functions RDMA/core: Configure selinux QP during creation RDMA/core: Properly increment and decrement QP usecnts RDMA/core: Create clean QP creations interface for uverbs

 drivers/infiniband/core/core_priv.h           |  67 ++----
 drivers/infiniband/core/uverbs_cmd.c          |  30 +--
 drivers/infiniband/core/uverbs_std_types_qp.c |  28 +--
 drivers/infiniband/core/verbs.c               | 216 ++++++++++++------
 drivers/infiniband/hw/mlx5/qp.c               |   3 -
 include/rdma/ib_verbs.h                       |   8 +-
 6 files changed, 166 insertions(+), 186 deletions(-)
Signature · 1 line
2.31.1
#2 lore raw

PATCH 2/7 RDMA/core: Delete duplicated and unreachable code

From: Leon Romanovsky <[email protected]>

The ib_create_named_qp() is kernel verb and no kernel users exist that use XRC_INI QP. Hence such QP path is not reachable. In addition, delete duplicated assignments of QP attributes from the initialization structure.

---
 drivers/infiniband/core/core_priv.h |  1 +
 drivers/infiniband/core/verbs.c     | 22 ++++------------------
 2 files changed, 5 insertions(+), 18 deletions(-)
drivers/infiniband/core/core_priv.h +1 -0
index 5dfa1190e3ea..cc54d74930d6 100644
@@ -342,6 +342,7 @@ _ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 	qp->rwq_ind_tbl = attr->rwq_ind_tbl;
 	qp->event_handler = attr->event_handler;
 	qp->port = attr->port_num;
+	qp->qp_context = attr->qp_context;
 
 	spin_lock_init(&qp->mr_lock);
 	INIT_LIST_HEAD(&qp->rdma_mrs);
drivers/infiniband/core/verbs.c +4 -18
index 89c6987cb5eb..635642a3ecbc 100644
@@ -1257,28 +1257,14 @@ struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
 		return xrc_qp;
 	}
 
-	qp->event_handler = qp_init_attr->event_handler;
-	qp->qp_context = qp_init_attr->qp_context;
-	if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
-		qp->recv_cq = NULL;
-		qp->srq = NULL;
-	} else {
-		qp->recv_cq = qp_init_attr->recv_cq;
-		if (qp_init_attr->recv_cq)
-			atomic_inc(&qp_init_attr->recv_cq->usecnt);
-		qp->srq = qp_init_attr->srq;
-		if (qp->srq)
-			atomic_inc(&qp_init_attr->srq->usecnt);
-	}
-
-	qp->send_cq = qp_init_attr->send_cq;
-	qp->xrcd    = NULL;
+	if (qp_init_attr->recv_cq)
+		atomic_inc(&qp_init_attr->recv_cq->usecnt);
+	if (qp->srq)
+		atomic_inc(&qp_init_attr->srq->usecnt);
 
 	atomic_inc(&pd->usecnt);
 	if (qp_init_attr->send_cq)
 		atomic_inc(&qp_init_attr->send_cq->usecnt);
-	if (qp_init_attr->rwq_ind_tbl)
-		atomic_inc(&qp->rwq_ind_tbl->usecnt);
 
 	if (qp_init_attr->cap.max_rdma_ctxs) {
 		ret = rdma_rw_init_mrs(qp, qp_init_attr);
Signature · 1 line
2.31.1
#3 lore raw

PATCH 3/7 RDMA/core: Remove protection from wrong in-kernel API usage

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

The ib_create_named_qp() is kernel verb that is not used for user supplied attributes. In such case, it is ULP responsibility to provide valid QP attributes.

In-kernel API shouldn't check it, exactly like other functions that don't check device capabilities.

---
 drivers/infiniband/core/verbs.c | 10 ----------
 1 file changed, 10 deletions(-)
drivers/infiniband/core/verbs.c +0 -10
index 635642a3ecbc..2090f3c9f689 100644
@@ -1219,16 +1219,6 @@ struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
 	struct ib_qp *qp;
 	int ret;
 
-	if (qp_init_attr->rwq_ind_tbl &&
-	    (qp_init_attr->recv_cq ||
-	    qp_init_attr->srq || qp_init_attr->cap.max_recv_wr ||
-	    qp_init_attr->cap.max_recv_sge))
-		return ERR_PTR(-EINVAL);
-
-	if ((qp_init_attr->create_flags & IB_QP_CREATE_INTEGRITY_EN) &&
-	    !(device->attrs.device_cap_flags & IB_DEVICE_INTEGRITY_HANDOVER))
-		return ERR_PTR(-EINVAL);
-
 	/*
 	 * If the callers is using the RDMA API calculate the resources
 	 * needed for the RDMA READ/WRITE operations.
Signature · 1 line
2.31.1
#12 lore raw

PATCH 6/7 RDMA/core: Properly increment and decrement QP usecnts

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

The QP usecnts were incremented through QP attributes structure while decreased through QP itself. Rely on the ib_creat_qp_user() code that initialized all QP parameters prior returning to the user and increment exactly like destroy does.

---
 drivers/infiniband/core/core_priv.h           |  2 +
 drivers/infiniband/core/uverbs_cmd.c          | 13 +---
 drivers/infiniband/core/uverbs_std_types_qp.c | 13 +---
 drivers/infiniband/core/verbs.c               | 60 ++++++++++---------
 4 files changed, 39 insertions(+), 49 deletions(-)
drivers/infiniband/core/core_priv.h +2 -0
index d28ced053222..d8f464b43dbc 100644
@@ -320,6 +320,8 @@ struct ib_qp *_ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *attr,
 			    struct ib_udata *udata, struct ib_uqp_object *uobj,
 			    const char *caller);
+void ib_qp_usecnt_inc(struct ib_qp *qp);
+void ib_qp_usecnt_dec(struct ib_qp *qp);
 
 struct rdma_dev_addr;
 int rdma_resolve_ip_route(struct sockaddr *src_addr,
drivers/infiniband/core/uverbs_cmd.c +2 -11
index b5153200b8a8..62cafd768d89 100644
@@ -1445,18 +1445,9 @@ static int create_qp(struct uverbs_attr_bundle *attrs,
 		ret = PTR_ERR(qp);
 		goto err_put;
 	}
+	ib_qp_usecnt_inc(qp);
 
-	if (cmd->qp_type != IB_QPT_XRC_TGT) {
-		atomic_inc(&pd->usecnt);
-		if (attr.send_cq)
-			atomic_inc(&attr.send_cq->usecnt);
-		if (attr.recv_cq)
-			atomic_inc(&attr.recv_cq->usecnt);
-		if (attr.srq)
-			atomic_inc(&attr.srq->usecnt);
-		if (ind_tbl)
-			atomic_inc(&ind_tbl->usecnt);
-	} else {
+	if (cmd->qp_type == IB_QPT_XRC_TGT) {
 		/* It is done in _ib_create_qp for other QP types */
 		qp->uobject = obj;
 	}
drivers/infiniband/core/uverbs_std_types_qp.c +2 -11
index 92812f6a21b0..a0e734735ba5 100644
@@ -258,18 +258,9 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QP_CREATE)(
 		ret = PTR_ERR(qp);
 		goto err_put;
 	}
+	ib_qp_usecnt_inc(qp);
 
-	if (attr.qp_type != IB_QPT_XRC_TGT) {
-		atomic_inc(&pd->usecnt);
-		if (attr.send_cq)
-			atomic_inc(&attr.send_cq->usecnt);
-		if (attr.recv_cq)
-			atomic_inc(&attr.recv_cq->usecnt);
-		if (attr.srq)
-			atomic_inc(&attr.srq->usecnt);
-		if (attr.rwq_ind_tbl)
-			atomic_inc(&attr.rwq_ind_tbl->usecnt);
-	} else {
+	if (attr.qp_type == IB_QPT_XRC_TGT) {
 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
 					  uobject);
 		atomic_inc(&obj->uxrcd->refcnt);
drivers/infiniband/core/verbs.c +33 -27

90 lines not on this page · raw message

Signature · 1 line
2.31.1
#6 lore raw

PATCH 4/7 RDMA/core: Reorganize create QP low-level functions

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

The low-level create QP function grew to be larger than any sensible incline function should be. The inline attribute is not really needed for that function and can be implemented as exported symbol.

---
 drivers/infiniband/core/core_priv.h | 59 ++-------------------
 drivers/infiniband/core/verbs.c     | 82 ++++++++++++++++++++++++++---
 include/rdma/ib_verbs.h             |  8 +--
 3 files changed, 82 insertions(+), 67 deletions(-)
drivers/infiniband/core/core_priv.h +4 -55
index cc54d74930d6..d28ced053222 100644
@@ -316,61 +316,10 @@ struct ib_device *ib_device_get_by_index(const struct net *net, u32 index);
 void nldev_init(void);
 void nldev_exit(void);
 
-static inline struct ib_qp *
-_ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
-	      struct ib_qp_init_attr *attr, struct ib_udata *udata,
-	      struct ib_uqp_object *uobj, const char *caller)
-{
-	struct ib_qp *qp;
-	int ret;
-
-	if (!dev->ops.create_qp)
-		return ERR_PTR(-EOPNOTSUPP);
-
-	qp = rdma_zalloc_drv_obj_numa(dev, ib_qp);
-	if (!qp)
-		return ERR_PTR(-ENOMEM);
-
-	qp->device = dev;
-	qp->pd = pd;
-	qp->uobject = uobj;
-	qp->real_qp = qp;
-
-	qp->qp_type = attr->qp_type;
-	qp->rwq_ind_tbl = attr->rwq_ind_tbl;
-	qp->srq = attr->srq;
-	qp->rwq_ind_tbl = attr->rwq_ind_tbl;
-	qp->event_handler = attr->event_handler;
-	qp->port = attr->port_num;
-	qp->qp_context = attr->qp_context;
-
-	spin_lock_init(&qp->mr_lock);
-	INIT_LIST_HEAD(&qp->rdma_mrs);
-	INIT_LIST_HEAD(&qp->sig_mrs);
-
-	rdma_restrack_new(&qp->res, RDMA_RESTRACK_QP);
-	WARN_ONCE(!udata && !caller, "Missing kernel QP owner");
-	rdma_restrack_set_name(&qp->res, udata ? NULL : caller);
-	ret = dev->ops.create_qp(qp, attr, udata);
-	if (ret)
-		goto err_create;
-
-	/*
-	 * TODO: The mlx4 internally overwrites send_cq and recv_cq.
-	 * Unfortunately, it is not an easy task to fix that driver.
-	 */
-	qp->send_cq = attr->send_cq;
-	qp->recv_cq = attr->recv_cq;
-
-	rdma_restrack_add(&qp->res);
-	return qp;
-
-err_create:
-	rdma_restrack_put(&qp->res);
-	kfree(qp);
-	return ERR_PTR(ret);
-
-}
+struct ib_qp *_ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
+			    struct ib_qp_init_attr *attr,
+			    struct ib_udata *udata, struct ib_uqp_object *uobj,
+			    const char *caller);
 
 struct rdma_dev_addr;
 int rdma_resolve_ip_route(struct sockaddr *src_addr,
drivers/infiniband/core/verbs.c +74 -8

100 lines not on this page · raw message

include/rdma/ib_verbs.h +4 -4
index 8cd7d1fc719f..6514110168c1 100644
@@ -3688,13 +3688,13 @@ static inline int ib_post_srq_recv(struct ib_srq *srq,
 					      bad_recv_wr ? : &dummy);
 }
 
-struct ib_qp *ib_create_named_qp(struct ib_pd *pd,
-				 struct ib_qp_init_attr *qp_init_attr,
-				 const char *caller);
+struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
+				  struct ib_qp_init_attr *qp_init_attr,
+				  const char *caller);
 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd,
 					 struct ib_qp_init_attr *init_attr)
 {
-	return ib_create_named_qp(pd, init_attr, KBUILD_MODNAME);
+	return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME);
 }
 
 /**
Signature · 1 line
2.31.1
#1 lore raw

PATCH 1/7 RDMA/mlx5: Delete not-available udata check

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

XRC_TGT QPs are created through kernel verbs and don't have udata at all.

Fixes 6eefa839c4dd ("RDMA/mlx5: Protect from kernel crash if XRC_TGT doesn't have udata")
Fixes e383085c2425 ("RDMA/mlx5: Set ECE options during QP create")
---
 drivers/infiniband/hw/mlx5/qp.c | 3 ---
 1 file changed, 3 deletions(-)
drivers/infiniband/hw/mlx5/qp.c +0 -3
index 18b018f1db83..81e3170a1ae6 100644
@@ -1908,7 +1908,6 @@ static int get_atomic_mode(struct mlx5_ib_dev *dev,
 static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 			     struct mlx5_create_qp_params *params)
 {
-	struct mlx5_ib_create_qp *ucmd = params->ucmd;
 	struct ib_qp_init_attr *attr = params->attr;
 	u32 uidx = params->uidx;
 	struct mlx5_ib_resources *devr = &dev->devr;
@@ -1928,8 +1927,6 @@ static int create_xrc_tgt_qp(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp,
 	if (!in)
 		return -ENOMEM;
 
-	if (MLX5_CAP_GEN(mdev, ece_support) && ucmd)
-		MLX5_SET(create_qp_in, in, ece, ucmd->ece_options);
 	qpc = MLX5_ADDR_OF(create_qp_in, in, qpc);
 
 	MLX5_SET(qpc, qpc, st, MLX5_QP_ST_XRC);
Signature · 1 line
2.31.1
#11 lore raw

PATCH 5/7 RDMA/core: Configure selinux QP during creation

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

All QP creation flows called ib_create_qp_security(), but differently. This caused to the need to provide exclusion conditions for the XRC_TGT, because such QP already had selinux configuration call.

In order to fix it, move ib_create_qp_security() to the general QP creation routine.

---
 drivers/infiniband/core/uverbs_cmd.c          |  7 -------
 drivers/infiniband/core/uverbs_std_types_qp.c |  6 ------
 drivers/infiniband/core/verbs.c               | 11 +++++++----
 3 files changed, 7 insertions(+), 17 deletions(-)
drivers/infiniband/core/uverbs_cmd.c +0 -7
index 8c8ca7bce3ca..b5153200b8a8 100644
@@ -1447,10 +1447,6 @@ static int create_qp(struct uverbs_attr_bundle *attrs,
 	}
 
 	if (cmd->qp_type != IB_QPT_XRC_TGT) {
-		ret = ib_create_qp_security(qp, device);
-		if (ret)
-			goto err_cb;
-
 		atomic_inc(&pd->usecnt);
 		if (attr.send_cq)
 			atomic_inc(&attr.send_cq->usecnt);
@@ -1502,9 +1498,6 @@ static int create_qp(struct uverbs_attr_bundle *attrs,
 	resp.response_length = uverbs_response_length(attrs, sizeof(resp));
 	return uverbs_response(attrs, &resp, sizeof(resp));
 
-err_cb:
-	ib_destroy_qp_user(qp, uverbs_get_cleared_udata(attrs));
-
 err_put:
 	if (!IS_ERR(xrcd_uobj))
 		uobj_put_read(xrcd_uobj);
drivers/infiniband/core/uverbs_std_types_qp.c +0 -6
index c00cfb5ed387..92812f6a21b0 100644
@@ -280,12 +280,6 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QP_CREATE)(
 	obj->uevent.uobject.object = qp;
 	uverbs_finalize_uobj_create(attrs, UVERBS_ATTR_CREATE_QP_HANDLE);
 
-	if (attr.qp_type != IB_QPT_XRC_TGT) {
-		ret = ib_create_qp_security(qp, device);
-		if (ret)
-			return ret;
-	}
-
 	set_caps(&attr, &cap, false);
 	ret = uverbs_copy_to_struct_or_zero(attrs,
 					UVERBS_ATTR_CREATE_QP_RESP_CAP, &cap,
drivers/infiniband/core/verbs.c +7 -4
index 645962490fa4..612c73861e0d 100644
@@ -1216,6 +1216,7 @@ struct ib_qp *_ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 			    struct ib_udata *udata, struct ib_uqp_object *uobj,
 			    const char *caller)
 {
+	struct ib_udata dummy = {};
 	struct ib_qp *qp;
 	int ret;
 
@@ -1257,9 +1258,15 @@ struct ib_qp *_ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
 	qp->send_cq = attr->send_cq;
 	qp->recv_cq = attr->recv_cq;
 
+	ret = ib_create_qp_security(qp, dev);
+	if (ret)
+		goto err_security;
+
 	rdma_restrack_add(&qp->res);
 	return qp;
 
+err_security:
+	qp->device->ops.destroy_qp(qp, udata ? &dummy : NULL);
 err_create:
 	rdma_restrack_put(&qp->res);
 	kfree(qp);
@@ -1298,10 +1305,6 @@ struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd,
 	if (IS_ERR(qp))
 		return qp;
 
-	ret = ib_create_qp_security(qp, device);
-	if (ret)
-		goto err;
-
 	if (qp_init_attr->qp_type == IB_QPT_XRC_TGT) {
 		struct ib_qp *xrc_qp =
 			create_xrc_qp_user(qp, qp_init_attr);
Signature · 1 line
2.31.1
#13 lore raw

PATCH 7/7 RDMA/core: Create clean QP creations interface for uverbs

In reply to Leon Romanovsky

From: Leon Romanovsky <[email protected]>

Unify create QP creation interface to make clean approach to create XRC_TGT and regular QPs.

---
 drivers/infiniband/core/core_priv.h           | 21 +++++-
 drivers/infiniband/core/uverbs_cmd.c          | 12 +--
 drivers/infiniband/core/uverbs_std_types_qp.c |  9 +--
 drivers/infiniband/core/verbs.c               | 73 +++++++++++--------
 4 files changed, 63 insertions(+), 52 deletions(-)
drivers/infiniband/core/core_priv.h +17 -4
index d8f464b43dbc..dd4c2e560b59 100644
@@ -316,10 +316,23 @@ struct ib_device *ib_device_get_by_index(const struct net *net, u32 index);
 void nldev_init(void);
 void nldev_exit(void);
 
-struct ib_qp *_ib_create_qp(struct ib_device *dev, struct ib_pd *pd,
-			    struct ib_qp_init_attr *attr,
-			    struct ib_udata *udata, struct ib_uqp_object *uobj,
-			    const char *caller);
+struct ib_qp *ib_create_qp_user(struct ib_device *dev, struct ib_pd *pd,
+				struct ib_qp_init_attr *attr,
+				struct ib_udata *udata,
+				struct ib_uqp_object *uobj, const char *caller);
+static inline struct ib_qp *ib_create_qp_uverbs(struct ib_device *dev,
+						struct ib_pd *pd,
+						struct ib_qp_init_attr *attr,
+						struct ib_udata *udata,
+						struct ib_uqp_object *uobj)
+{
+	if (attr->qp_type == IB_QPT_XRC_TGT)
+		return ib_create_qp_user(dev, pd, attr, NULL, uobj,
+					 KBUILD_MODNAME);
+
+	return ib_create_qp_user(dev, pd, attr, udata, uobj, NULL);
+}
+
 void ib_qp_usecnt_inc(struct ib_qp *qp);
 void ib_qp_usecnt_dec(struct ib_qp *qp);
 
drivers/infiniband/core/uverbs_cmd.c +1 -11
index 62cafd768d89..39206ecdeaad 100644
@@ -1435,23 +1435,13 @@ static int create_qp(struct uverbs_attr_bundle *attrs,
 		attr.source_qpn = cmd->source_qpn;
 	}
 
-	if (cmd->qp_type == IB_QPT_XRC_TGT)
-		qp = ib_create_qp(pd, &attr);
-	else
-		qp = _ib_create_qp(device, pd, &attr, &attrs->driver_udata, obj,
-				   NULL);
-
+	qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
 	if (IS_ERR(qp)) {
 		ret = PTR_ERR(qp);
 		goto err_put;
 	}
 	ib_qp_usecnt_inc(qp);
 
-	if (cmd->qp_type == IB_QPT_XRC_TGT) {
-		/* It is done in _ib_create_qp for other QP types */
-		qp->uobject = obj;
-	}
-
 	obj->uevent.uobject.object = qp;
 	obj->uevent.event_file = READ_ONCE(attrs->ufile->default_async_file);
 	if (obj->uevent.event_file)
drivers/infiniband/core/uverbs_std_types_qp.c +1 -8
index a0e734735ba5..15ea00188363 100644
@@ -248,12 +248,7 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QP_CREATE)(
 	set_caps(&attr, &cap, true);
 	mutex_init(&obj->mcast_lock);
 
-	if (attr.qp_type == IB_QPT_XRC_TGT)
-		qp = ib_create_qp(pd, &attr);
-	else
-		qp = _ib_create_qp(device, pd, &attr, &attrs->driver_udata, obj,
-				   NULL);
-
+	qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
 	if (IS_ERR(qp)) {
 		ret = PTR_ERR(qp);
 		goto err_put;
@@ -264,8 +259,6 @@ static int UVERBS_HANDLER(UVERBS_METHOD_QP_CREATE)(
 		obj->uxrcd = container_of(xrcd_uobj, struct ib_uxrcd_object,
 					  uobject);
 		atomic_inc(&obj->uxrcd->refcnt);
-		/* It is done in _ib_create_qp for other QP types */
-		qp->uobject = obj;
 	}
 
 	obj->uevent.uobject.object = qp;
drivers/infiniband/core/verbs.c +44 -29

100 lines not on this page · raw message

Signature · 1 line
2.31.1
#4 lore raw

Re: [PATCH rdma-next 3/7] RDMA/core: Remove protection from wrong in-kernel API usage

In reply to Leon Romanovsky
Quoted 8 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

The ib_create_named_qp() is kernel verb that is not used for user supplied attributes. In such case, it is ULP responsibility to provide valid QP attributes.

In-kernel API shouldn't check it, exactly like other functions that don't check device capabilities.

Hmm. These still looks like useful debugging checks.

#7 lore raw

Re: [PATCH rdma-next 4/7] RDMA/core: Reorganize create QP low-level functions

In reply to Leon Romanovsky
Quoted 2 lines

+/**
+ * ib_create_qp_kernel - Creates a kernel QP associated with the specified

Any reason this function is renamed? This seems rather unrelated to the rest of th patch.

Quoted 9 lines

+ * protection domain.
* @pd: The protection domain associated with the QP.
* @qp_init_attr: A list of initial attributes required to create the
* QP. If QP creation succeeds, then the attributes are updated to
* the actual capabilities of the created QP.
* @caller: caller's build-time module name
- *
- * NOTE: for user qp use ib_create_qp_user with valid udata! */

Also a kerneldoc comment for a function that is an implementation detail is actively harmful. Please document ib_create_qp instead.

#14 lore raw

Re: [PATCH rdma-next 7/7] RDMA/core: Create clean QP creations interface for uverbs

In reply to Leon Romanovsky
Quoted 5 lines
drivers/infiniband/core/core_priv.h +5 -0
+struct ib_qp *ib_create_qp_user(struct ib_device *dev, struct ib_pd *pd,
+				struct ib_qp_init_attr *attr,
+				struct ib_udata *udata,
+				struct ib_uqp_object *uobj, const char *caller);
+static inline struct ib_qp *ib_create_qp_uverbs(struct ib_device *dev,

drivers/infiniband/core/core_priv.h struct ib_device *ib_device_get_by_index(const struct net *net, u32 index); · line 333

Quoted 10 lines
drivers/infiniband/core/core_priv.h +10 -0
+						struct ib_pd *pd,
+						struct ib_qp_init_attr *attr,
+						struct ib_udata *udata,
+						struct ib_uqp_object *uobj)
+{
+	if (attr->qp_type == IB_QPT_XRC_TGT)
+		return ib_create_qp_user(dev, pd, attr, NULL, uobj,
+					 KBUILD_MODNAME);
+
+	return ib_create_qp_user(dev, pd, attr, udata, uobj, NULL);

Why not always pass along the udata and caller and just not use them in the low-level code?

#5 lore raw

Re: [PATCH rdma-next 3/7] RDMA/core: Remove protection from wrong in-kernel API usage

In reply to Christoph Hellwig
Quoted 11 lines · Christoph Hellwig
Quoted 8 lines · Leon Romanovsky

From: Leon Romanovsky <[email protected]>

The ib_create_named_qp() is kernel verb that is not used for user supplied attributes. In such case, it is ULP responsibility to provide valid QP attributes.

In-kernel API shouldn't check it, exactly like other functions that don't check device capabilities.

Hmm. These still looks like useful debugging checks.

All in-tree ULPs already set properly based on device capabilities.

Thanks

#8 lore raw

Re: [PATCH rdma-next 4/7] RDMA/core: Reorganize create QP low-level functions

In reply to Christoph Hellwig
Quoted 5 lines · Christoph Hellwig
Quoted 2 lines

+/**
+ * ib_create_qp_kernel - Creates a kernel QP associated with the specified

Any reason this function is renamed? This seems rather unrelated to the rest of th patch.

I wanted to make two functions: ib_create_qp_kernel and ib_create_qp_user.

Quoted 11 lines
Quoted 10 lines

+ * protection domain.
* @pd: The protection domain associated with the QP.
* @qp_init_attr: A list of initial attributes required to create the
* QP. If QP creation succeeds, then the attributes are updated to
* the actual capabilities of the created QP.
* @caller: caller's build-time module name
- *
- * NOTE: for user qp use ib_create_qp_user with valid udata! */

Quoted 2 lines

Also a kerneldoc comment for a function that is an implementation detail is actively harmful. Please document ib_create_qp instead.

Sure, will do.

#15 lore raw

Re: [PATCH rdma-next 7/7] RDMA/core: Create clean QP creations interface for uverbs

In reply to Christoph Hellwig
Quoted 10 lines · Christoph Hellwig
Quoted 10 lines
drivers/infiniband/core/core_priv.h +10 -0
+struct ib_qp *ib_create_qp_user(struct ib_device *dev, struct ib_pd *pd,
+				struct ib_qp_init_attr *attr,
+				struct ib_udata *udata,
+				struct ib_uqp_object *uobj, const char *caller);
+static inline struct ib_qp *ib_create_qp_uverbs(struct ib_device *dev,
+						struct ib_pd *pd,
+						struct ib_qp_init_attr *attr,
+						struct ib_udata *udata,
+						struct ib_uqp_object *uobj)
+{
Quoted 8 lines

drivers/infiniband/core/core_priv.h struct ib_device *ib_device_get_by_index(const struct net *net, u32 index); · line 333

Quoted 5 lines
drivers/infiniband/core/core_priv.h +5 -0
+	if (attr->qp_type == IB_QPT_XRC_TGT)
+		return ib_create_qp_user(dev, pd, attr, NULL, uobj,
+					 KBUILD_MODNAME);
+
+	return ib_create_qp_user(dev, pd, attr, udata, uobj, NULL);

Why not always pass along the udata and caller and just not use them in the low-level code?

You will need to add some sort of "if qp tpye" for ib_create_qp_uverbs() callers, because they always provide udata != NULL.

After this series, the callers look like this:

 1438         qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
                                                          ^^^^^^^^^ not NULL

So instead of bothering callers, I implemented it here with one "if".

Thanks

#16 lore raw

Re: [PATCH rdma-next 7/7] RDMA/core: Create clean QP creations interface for uverbs

Quoted 9 lines · Leon Romanovsky

You will need to add some sort of "if qp tpye" for ib_create_qp_uverbs() callers, because they always provide udata != NULL.

After this series, the callers look like this:

 1438         qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
                                                          ^^^^^^^^^ not NULL

So instead of bothering callers, I implemented it here with one "if".

Sorry if my mail was confusing. I don't want it in the callers, I want it as deep down in the stack as possible instead of having the strange wrapper.

#17 lore raw

Re: [PATCH rdma-next 7/7] RDMA/core: Create clean QP creations interface for uverbs

Quoted 6 lines · Christoph Hellwig
Quoted 5 lines · Leon Romanovsky

You will need to add some sort of "if qp tpye" for ib_create_qp_uverbs() callers, because they always provide udata != NULL.

After this series, the callers look like this:

Quoted 8 lines
Quoted 4 lines
 1438         qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
                                                          ^^^^^^^^^ not NULL

So instead of bothering callers, I implemented it here with one "if".

Sorry if my mail was confusing. I don't want it in the callers, I want it as deep down in the stack as possible instead of having the strange wrapper.

In fact ib_create_qp_user already sets udata to NULL for IB_QPT_XRC_TGT, and _ib_create_qp/create_qp ignores the caller if the udata is NULL. So I think you can just remove the wrapper and we're already fine.

#19 lore raw

Re: [PATCH rdma-next 7/7] RDMA/core: Create clean QP creations interface for uverbs

In reply to Christoph Hellwig
Quoted 6 lines · Christoph Hellwig
Quoted 5 lines · Leon Romanovsky

You will need to add some sort of "if qp tpye" for ib_create_qp_uverbs() callers, because they always provide udata != NULL.

After this series, the callers look like this:

Quoted 8 lines
Quoted 4 lines
 1438         qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
                                                          ^^^^^^^^^ not NULL

So instead of bothering callers, I implemented it here with one "if".

Sorry if my mail was confusing. I don't want it in the callers, I want it as deep down in the stack as possible instead of having the strange wrapper.

I see, will change and resend.

Thanks

#18 lore raw

Re: [PATCH rdma-next 7/7] RDMA/core: Create clean QP creations interface for uverbs

In reply to Christoph Hellwig
Quoted 12 lines · Christoph Hellwig
Quoted 11 lines · Christoph Hellwig
Quoted 9 lines · Leon Romanovsky

You will need to add some sort of "if qp tpye" for ib_create_qp_uverbs() callers, because they always provide udata != NULL.

After this series, the callers look like this:

 1438         qp = ib_create_qp_uverbs(device, pd, &attr, &attrs->driver_udata, obj);
                                                          ^^^^^^^^^ not NULL

So instead of bothering callers, I implemented it here with one "if".

Quoted 7 lines
Quoted 3 lines

Sorry if my mail was confusing. I don't want it in the callers, I want it as deep down in the stack as possible instead of having the strange wrapper.

In fact ib_create_qp_user already sets udata to NULL for IB_QPT_XRC_TGT, and _ib_create_qp/create_qp ignores the caller if the udata is NULL. So I think you can just remove the wrapper and we're already fine.

Yeah, thanks

#9 lore raw

Re: [PATCH rdma-next 4/7] RDMA/core: Reorganize create QP low-level functions

In reply to Leon Romanovsky
Quoted 8 lines · Leon Romanovsky

External email: Use caution opening links or attachments

From: Leon Romanovsky <[email protected]>

The low-level create QP function grew to be larger than any sensible incline function should be. The inline attribute is not really needed for that function and can be implemented as exported symbol.

incline -> inline?

#10 lore raw

Re: [PATCH rdma-next 4/7] RDMA/core: Reorganize create QP low-level functions

Quoted 11 lines · Mark Zhang
Quoted 8 lines · Leon Romanovsky

External email: Use caution opening links or attachments

From: Leon Romanovsky <[email protected]>

The low-level create QP function grew to be larger than any sensible incline function should be. The inline attribute is not really needed for that function and can be implemented as exported symbol.

incline -> inline?

Thanks

Quoted 2 lines