[PATCH rdma-next v2 0/5] RDMA/mana_ib: Implement RNIC CQs

[PATCH rdma-next v2 0/5] RDMA/mana_ib: Implement RNIC CQs

[Konstantin Taranov]

From: Konstantin Taranov <kotaranov@microsoft.com>

This patch series implements creation and destruction of CQs
which can be used with RC QPs.

Patches with RC QPs will be sent in the next patch series.

To create a CQ for RNIC, mana_ib requires creation of EQs
within mana_ib device. An EQ of mana ethernet cannot be used.

To make the implementation of create_cq cleaner, this series
also introduces a helper to remove CQ callbacks.

Mana ethernet and mana_ib CQs are different entities which are
created in different isolation zones (ethernet vs rnic).
As a result, RNIC cannot use ethenet CQs and ethernet cannot
use RNIC CQs.
That is why, we use existing udata request for creation of
ethernet CQs. If the request has an extra flag, then we create
an RNIC CQ. The kernel-level CQs will be RNIC CQs (in future
patches).

To preserve backward and forward compatibility with RDMA-CORE,
we will make the following changes to mana provider in RDMA-CORE:

The rdma-core will request RNIC CQs by default, with the proposed
request format and the special flag.
If the mana has installed an allocator with manadv_set_context_attr,
then the rdma-core understands that this is a DPDK use-case and
requests an ethernet CQ, by not setting the flag.

If the user has a new RDMA-core and an old kernel, then the user can
detect it as the response to create RNIC cq will not have queue id.

If the user has an old RDMA-core, then the flags will be 0 and ethernet
CQ will be created (as expected by the user).

v1->v2:
1) removed patch that replace cqe with buf_size
2) added aditional check of queue id in the remove cb helper
3) removed buf_size from uapi request and added flags instead. It seems
to be a better proposal that will not require to increase the ABI version.

Konstantin Taranov (5):
  RDMA/mana_ib: create EQs for RNIC CQs
  RDMA/mana_ib: create and destroy RNIC cqs
  RDMA/mana_ib: introduce a helper to remove cq callbacks
  RDMA/mana_ib: boundary check before installing cq callbacks
  RDMA/mana_ib: implement uapi for creation of rnic cq

 drivers/infiniband/hw/mana/cq.c      | 74 +++++++++++++++++++----
 drivers/infiniband/hw/mana/main.c    | 88 +++++++++++++++++++++++++++-
 drivers/infiniband/hw/mana/mana_ib.h | 34 +++++++++++
 drivers/infiniband/hw/mana/qp.c      | 26 ++------
 include/uapi/rdma/mana-abi.h         | 12 ++++
 5 files changed, 200 insertions(+), 34 deletions(-)

-- 
2.43.0

[PATCH rdma-next v2 1/5] RDMA/mana_ib: create EQs for RNIC CQs

[Konstantin Taranov]

From: Konstantin Taranov <kotaranov@microsoft.com>

Create EQs within mana_ib device. Such EQs are required
for creation of RNIC CQs.

Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>
Reviewed-by: Long Li <longli@microsoft.com>
---
 drivers/infiniband/hw/mana/main.c    | 34 ++++++++++++++++++++++++++--
 drivers/infiniband/hw/mana/mana_ib.h |  1 +
 2 files changed, 33 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/mana/main.c b/drivers/infiniband/hw/mana/main.c
index f540147..546d059 100644
--- a/drivers/infiniband/hw/mana/main.c
+++ b/drivers/infiniband/hw/mana/main.c
@@ -658,7 +658,7 @@ int mana_ib_create_eqs(struct mana_ib_dev *mdev)
 {
 	struct gdma_context *gc = mdev_to_gc(mdev);
 	struct gdma_queue_spec spec = {};
-	int err;
+	int err, i;
 
 	spec.type = GDMA_EQ;
 	spec.monitor_avl_buf = false;
@@ -672,12 +672,42 @@ int mana_ib_create_eqs(struct mana_ib_dev *mdev)
 	if (err)
 		return err;
 
+	mdev->eqs = kcalloc(mdev->ib_dev.num_comp_vectors, sizeof(struct gdma_queue *),
+			    GFP_KERNEL);
+	if (!mdev->eqs) {
+		err = -ENOMEM;
+		goto destroy_fatal_eq;
+	}
+
+	for (i = 0; i < mdev->ib_dev.num_comp_vectors; i++) {
+		spec.eq.msix_index = (i + 1) % gc->num_msix_usable;
+		err = mana_gd_create_mana_eq(mdev->gdma_dev, &spec, &mdev->eqs[i]);
+		if (err)
+			goto destroy_eqs;
+	}
+
 	return 0;
+
+destroy_eqs:
+	while (i-- > 0)
+		mana_gd_destroy_queue(gc, mdev->eqs[i]);
+	kfree(mdev->eqs);
+destroy_fatal_eq:
+	mana_gd_destroy_queue(gc, mdev->fatal_err_eq);
+	return err;
 }
 
 void mana_ib_destroy_eqs(struct mana_ib_dev *mdev)
 {
-	mana_gd_destroy_queue(mdev_to_gc(mdev), mdev->fatal_err_eq);
+	struct gdma_context *gc = mdev_to_gc(mdev);
+	int i;
+
+	mana_gd_destroy_queue(gc, mdev->fatal_err_eq);
+
+	for (i = 0; i < mdev->ib_dev.num_comp_vectors; i++)
+		mana_gd_destroy_queue(gc, mdev->eqs[i]);
+
+	kfree(mdev->eqs);
 }
 
 int mana_ib_gd_create_rnic_adapter(struct mana_ib_dev *mdev)
diff --git a/drivers/infiniband/hw/mana/mana_ib.h b/drivers/infiniband/hw/mana/mana_ib.h
index 4c1240d..bfcf6df 100644
--- a/drivers/infiniband/hw/mana/mana_ib.h
+++ b/drivers/infiniband/hw/mana/mana_ib.h
@@ -56,6 +56,7 @@ struct mana_ib_dev {
 	struct gdma_dev *gdma_dev;
 	mana_handle_t adapter_handle;
 	struct gdma_queue *fatal_err_eq;
+	struct gdma_queue **eqs;
 	struct mana_ib_adapter_caps adapter_caps;
 };
 
-- 
2.43.0

[PATCH rdma-next v2 2/5] RDMA/mana_ib: create and destroy RNIC cqs

[Konstantin Taranov]

From: Konstantin Taranov <kotaranov@microsoft.com>

Implement RNIC requests for creation and destruction of RNIC CQs.

Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>
Reviewed-by: Long Li <longli@microsoft.com>
---
 drivers/infiniband/hw/mana/main.c    | 54 ++++++++++++++++++++++++++++
 drivers/infiniband/hw/mana/mana_ib.h | 32 +++++++++++++++++
 2 files changed, 86 insertions(+)

diff --git a/drivers/infiniband/hw/mana/main.c b/drivers/infiniband/hw/mana/main.c
index 546d059..2a41135 100644
--- a/drivers/infiniband/hw/mana/main.c
+++ b/drivers/infiniband/hw/mana/main.c
@@ -834,3 +834,57 @@ int mana_ib_gd_config_mac(struct mana_ib_dev *mdev, enum mana_ib_addr_op op, u8
 
 	return 0;
 }
+
+int mana_ib_gd_create_cq(struct mana_ib_dev *mdev, struct mana_ib_cq *cq, u32 doorbell)
+{
+	struct gdma_context *gc = mdev_to_gc(mdev);
+	struct mana_rnic_create_cq_resp resp = {};
+	struct mana_rnic_create_cq_req req = {};
+	int err;
+
+	mana_gd_init_req_hdr(&req.hdr, MANA_IB_CREATE_CQ, sizeof(req), sizeof(resp));
+	req.hdr.dev_id = gc->mana_ib.dev_id;
+	req.adapter = mdev->adapter_handle;
+	req.gdma_region = cq->queue.gdma_region;
+	req.eq_id = mdev->eqs[cq->comp_vector]->id;
+	req.doorbell_page = doorbell;
+
+	err = mana_gd_send_request(gc, sizeof(req), &req, sizeof(resp), &resp);
+
+	if (err) {
+		ibdev_err(&mdev->ib_dev, "Failed to create cq err %d", err);
+		return err;
+	}
+
+	cq->queue.id  = resp.cq_id;
+	cq->cq_handle = resp.cq_handle;
+	/* The GDMA region is now owned by the CQ handle */
+	cq->queue.gdma_region = GDMA_INVALID_DMA_REGION;
+
+	return 0;
+}
+
+int mana_ib_gd_destroy_cq(struct mana_ib_dev *mdev, struct mana_ib_cq *cq)
+{
+	struct gdma_context *gc = mdev_to_gc(mdev);
+	struct mana_rnic_destroy_cq_resp resp = {};
+	struct mana_rnic_destroy_cq_req req = {};
+	int err;
+
+	if (cq->cq_handle == INVALID_MANA_HANDLE)
+		return 0;
+
+	mana_gd_init_req_hdr(&req.hdr, MANA_IB_DESTROY_CQ, sizeof(req), sizeof(resp));
+	req.hdr.dev_id = gc->mana_ib.dev_id;
+	req.adapter = mdev->adapter_handle;
+	req.cq_handle = cq->cq_handle;
+
+	err = mana_gd_send_request(gc, sizeof(req), &req, sizeof(resp), &resp);
+
+	if (err) {
+		ibdev_err(&mdev->ib_dev, "Failed to destroy cq err %d", err);
+		return err;
+	}
+
+	return 0;
+}
diff --git a/drivers/infiniband/hw/mana/mana_ib.h b/drivers/infiniband/hw/mana/mana_ib.h
index bfcf6df..9162f29 100644
--- a/drivers/infiniband/hw/mana/mana_ib.h
+++ b/drivers/infiniband/hw/mana/mana_ib.h
@@ -92,6 +92,7 @@ struct mana_ib_cq {
 	struct mana_ib_queue queue;
 	int cqe;
 	u32 comp_vector;
+	mana_handle_t  cq_handle;
 };
 
 struct mana_ib_qp {
@@ -119,6 +120,8 @@ enum mana_ib_command_code {
 	MANA_IB_DESTROY_ADAPTER = 0x30003,
 	MANA_IB_CONFIG_IP_ADDR	= 0x30004,
 	MANA_IB_CONFIG_MAC_ADDR	= 0x30005,
+	MANA_IB_CREATE_CQ       = 0x30008,
+	MANA_IB_DESTROY_CQ      = 0x30009,
 };
 
 struct mana_ib_query_adapter_caps_req {
@@ -202,6 +205,31 @@ struct mana_rnic_config_mac_addr_resp {
 	struct gdma_resp_hdr hdr;
 }; /* HW Data */
 
+struct mana_rnic_create_cq_req {
+	struct gdma_req_hdr hdr;
+	mana_handle_t adapter;
+	u64 gdma_region;
+	u32 eq_id;
+	u32 doorbell_page;
+}; /* HW Data */
+
+struct mana_rnic_create_cq_resp {
+	struct gdma_resp_hdr hdr;
+	mana_handle_t cq_handle;
+	u32 cq_id;
+	u32 reserved;
+}; /* HW Data */
+
+struct mana_rnic_destroy_cq_req {
+	struct gdma_req_hdr hdr;
+	mana_handle_t adapter;
+	mana_handle_t cq_handle;
+}; /* HW Data */
+
+struct mana_rnic_destroy_cq_resp {
+	struct gdma_resp_hdr hdr;
+}; /* HW Data */
+
 static inline struct gdma_context *mdev_to_gc(struct mana_ib_dev *mdev)
 {
 	return mdev->gdma_dev->gdma_context;
@@ -321,4 +349,8 @@ int mana_ib_gd_add_gid(const struct ib_gid_attr *attr, void **context);
 int mana_ib_gd_del_gid(const struct ib_gid_attr *attr, void **context);
 
 int mana_ib_gd_config_mac(struct mana_ib_dev *mdev, enum mana_ib_addr_op op, u8 *mac);
+
+int mana_ib_gd_create_cq(struct mana_ib_dev *mdev, struct mana_ib_cq *cq, u32 doorbell);
+
+int mana_ib_gd_destroy_cq(struct mana_ib_dev *mdev, struct mana_ib_cq *cq);
 #endif
-- 
2.43.0

[PATCH rdma-next v2 3/5] RDMA/mana_ib: introduce a helper to remove cq callbacks

[Konstantin Taranov]

From: Konstantin Taranov <kotaranov@microsoft.com>

Intoduce the mana_ib_remove_cq_cb helper to remove cq callbacks.
The helper removes code duplicates.

Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>
Reviewed-by: Long Li <longli@microsoft.com>
---
 drivers/infiniband/hw/mana/cq.c      | 19 ++++++++++++-------
 drivers/infiniband/hw/mana/mana_ib.h |  1 +
 drivers/infiniband/hw/mana/qp.c      | 26 ++++----------------------
 3 files changed, 17 insertions(+), 29 deletions(-)

diff --git a/drivers/infiniband/hw/mana/cq.c b/drivers/infiniband/hw/mana/cq.c
index dc931b9..298e8f1 100644
--- a/drivers/infiniband/hw/mana/cq.c
+++ b/drivers/infiniband/hw/mana/cq.c
@@ -48,16 +48,10 @@ int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
 	struct mana_ib_cq *cq = container_of(ibcq, struct mana_ib_cq, ibcq);
 	struct ib_device *ibdev = ibcq->device;
 	struct mana_ib_dev *mdev;
-	struct gdma_context *gc;
 
 	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
-	gc = mdev_to_gc(mdev);
-
-	if (cq->queue.id != INVALID_QUEUE_ID) {
-		kfree(gc->cq_table[cq->queue.id]);
-		gc->cq_table[cq->queue.id] = NULL;
-	}
 
+	mana_ib_remove_cq_cb(mdev, cq);
 	mana_ib_destroy_queue(mdev, &cq->queue);
 
 	return 0;
@@ -89,3 +83,14 @@ int mana_ib_install_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq)
 	gc->cq_table[cq->queue.id] = gdma_cq;
 	return 0;
 }
+
+void mana_ib_remove_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq)
+{
+	struct gdma_context *gc = mdev_to_gc(mdev);
+
+	if (cq->queue.id >= gc->max_num_cqs || cq->queue.id == INVALID_QUEUE_ID)
+		return;
+
+	kfree(gc->cq_table[cq->queue.id]);
+	gc->cq_table[cq->queue.id] = NULL;
+}
diff --git a/drivers/infiniband/hw/mana/mana_ib.h b/drivers/infiniband/hw/mana/mana_ib.h
index 9162f29..68c3b4f 100644
--- a/drivers/infiniband/hw/mana/mana_ib.h
+++ b/drivers/infiniband/hw/mana/mana_ib.h
@@ -255,6 +255,7 @@ static inline void copy_in_reverse(u8 *dst, const u8 *src, u32 size)
 }
 
 int mana_ib_install_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq);
+void mana_ib_remove_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq);
 
 int mana_ib_create_zero_offset_dma_region(struct mana_ib_dev *dev, struct ib_umem *umem,
 					  mana_handle_t *gdma_region);
diff --git a/drivers/infiniband/hw/mana/qp.c b/drivers/infiniband/hw/mana/qp.c
index 280e85a..ba13c5a 100644
--- a/drivers/infiniband/hw/mana/qp.c
+++ b/drivers/infiniband/hw/mana/qp.c
@@ -95,11 +95,9 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd,
 	struct mana_ib_qp *qp = container_of(ibqp, struct mana_ib_qp, ibqp);
 	struct mana_ib_dev *mdev =
 		container_of(pd->device, struct mana_ib_dev, ib_dev);
-	struct gdma_context *gc = mdev_to_gc(mdev);
 	struct ib_rwq_ind_table *ind_tbl = attr->rwq_ind_tbl;
 	struct mana_ib_create_qp_rss_resp resp = {};
 	struct mana_ib_create_qp_rss ucmd = {};
-	struct gdma_queue **gdma_cq_allocated;
 	mana_handle_t *mana_ind_table;
 	struct mana_port_context *mpc;
 	unsigned int ind_tbl_size;
@@ -173,13 +171,6 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd,
 		goto fail;
 	}
 
-	gdma_cq_allocated = kcalloc(ind_tbl_size, sizeof(*gdma_cq_allocated),
-				    GFP_KERNEL);
-	if (!gdma_cq_allocated) {
-		ret = -ENOMEM;
-		goto fail;
-	}
-
 	qp->port = port;
 
 	for (i = 0; i < ind_tbl_size; i++) {
@@ -229,8 +220,6 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd,
 		ret = mana_ib_install_cq_cb(mdev, cq);
 		if (ret)
 			goto fail;
-
-		gdma_cq_allocated[i] = gc->cq_table[cq->queue.id];
 	}
 	resp.num_entries = i;
 
@@ -250,7 +239,6 @@ static int mana_ib_create_qp_rss(struct ib_qp *ibqp, struct ib_pd *pd,
 		goto fail;
 	}
 
-	kfree(gdma_cq_allocated);
 	kfree(mana_ind_table);
 
 	return 0;
@@ -262,13 +250,10 @@ fail:
 		wq = container_of(ibwq, struct mana_ib_wq, ibwq);
 		cq = container_of(ibcq, struct mana_ib_cq, ibcq);
 
-		gc->cq_table[cq->queue.id] = NULL;
-		kfree(gdma_cq_allocated[i]);
-
+		mana_ib_remove_cq_cb(mdev, cq);
 		mana_destroy_wq_obj(mpc, GDMA_RQ, wq->rx_object);
 	}
 
-	kfree(gdma_cq_allocated);
 	kfree(mana_ind_table);
 
 	return ret;
@@ -287,10 +272,8 @@ static int mana_ib_create_qp_raw(struct ib_qp *ibqp, struct ib_pd *ibpd,
 	struct mana_ib_ucontext *mana_ucontext =
 		rdma_udata_to_drv_context(udata, struct mana_ib_ucontext,
 					  ibucontext);
-	struct gdma_context *gc = mdev_to_gc(mdev);
 	struct mana_ib_create_qp_resp resp = {};
 	struct mana_ib_create_qp ucmd = {};
-	struct gdma_queue *gdma_cq = NULL;
 	struct mana_obj_spec wq_spec = {};
 	struct mana_obj_spec cq_spec = {};
 	struct mana_port_context *mpc;
@@ -395,14 +378,13 @@ static int mana_ib_create_qp_raw(struct ib_qp *ibqp, struct ib_pd *ibpd,
 		ibdev_dbg(&mdev->ib_dev,
 			  "Failed copy udata for create qp-raw, %d\n",
 			  err);
-		goto err_release_gdma_cq;
+		goto err_remove_cq_cb;
 	}
 
 	return 0;
 
-err_release_gdma_cq:
-	kfree(gdma_cq);
-	gc->cq_table[send_cq->queue.id] = NULL;
+err_remove_cq_cb:
+	mana_ib_remove_cq_cb(mdev, send_cq);
 
 err_destroy_wq_obj:
 	mana_destroy_wq_obj(mpc, GDMA_SQ, qp->qp_handle);
-- 
2.43.0

[PATCH rdma-next v2 4/5] RDMA/mana_ib: boundary check before installing cq callbacks

[Konstantin Taranov]

From: Konstantin Taranov <kotaranov@microsoft.com>

Add a boundary check inside mana_ib_install_cq_cb to prevent index overflow.

Fixes: 2a31c5a7e0d8 ("RDMA/mana_ib: Introduce mana_ib_install_cq_cb helper function")
Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>
Reviewed-by: Long Li <longli@microsoft.com>
---
 drivers/infiniband/hw/mana/cq.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/infiniband/hw/mana/cq.c b/drivers/infiniband/hw/mana/cq.c
index 298e8f1..688ffe6 100644
--- a/drivers/infiniband/hw/mana/cq.c
+++ b/drivers/infiniband/hw/mana/cq.c
@@ -70,6 +70,8 @@ int mana_ib_install_cq_cb(struct mana_ib_dev *mdev, struct mana_ib_cq *cq)
 	struct gdma_context *gc = mdev_to_gc(mdev);
 	struct gdma_queue *gdma_cq;
 
+	if (cq->queue.id >= gc->max_num_cqs)
+		return -EINVAL;
 	/* Create CQ table entry */
 	WARN_ON(gc->cq_table[cq->queue.id]);
 	gdma_cq = kzalloc(sizeof(*gdma_cq), GFP_KERNEL);
-- 
2.43.0

[PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation of rnic cq

[Konstantin Taranov]

From: Konstantin Taranov <kotaranov@microsoft.com>

Enable users to create RNIC CQs using a corresponding flag.
With the previous request size, an ethernet CQ is created.
As a response, return ID of the created CQ.

Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>
---
 drivers/infiniband/hw/mana/cq.c | 55 ++++++++++++++++++++++++++++++---
 include/uapi/rdma/mana-abi.h    | 12 +++++++
 2 files changed, 63 insertions(+), 4 deletions(-)

diff --git a/drivers/infiniband/hw/mana/cq.c b/drivers/infiniband/hw/mana/cq.c
index 688ffe6..c6a3fd5 100644
--- a/drivers/infiniband/hw/mana/cq.c
+++ b/drivers/infiniband/hw/mana/cq.c
@@ -9,17 +9,22 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 		      struct ib_udata *udata)
 {
 	struct mana_ib_cq *cq = container_of(ibcq, struct mana_ib_cq, ibcq);
+	struct mana_ib_create_cq_resp resp = {};
+	struct mana_ib_ucontext *mana_ucontext;
 	struct ib_device *ibdev = ibcq->device;
 	struct mana_ib_create_cq ucmd = {};
 	struct mana_ib_dev *mdev;
+	bool is_rnic_cq;
+	u32 doorbell;
 	int err;
 
 	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
 
-	if (udata->inlen < sizeof(ucmd))
-		return -EINVAL;
-
 	cq->comp_vector = attr->comp_vector % ibdev->num_comp_vectors;
+	cq->cq_handle = INVALID_MANA_HANDLE;
+
+	if (udata->inlen < offsetof(struct mana_ib_create_cq, flags))
+		return -EINVAL;
 
 	err = ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata->inlen));
 	if (err) {
@@ -28,7 +33,9 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 		return err;
 	}
 
-	if (attr->cqe > mdev->adapter_caps.max_qp_wr) {
+	is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);
+
+	if (!is_rnic_cq && attr->cqe > mdev->adapter_caps.max_qp_wr) {
 		ibdev_dbg(ibdev, "CQE %d exceeding limit\n", attr->cqe);
 		return -EINVAL;
 	}
@@ -40,7 +47,41 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 		return err;
 	}
 
+	mana_ucontext = rdma_udata_to_drv_context(udata, struct mana_ib_ucontext,
+						  ibucontext);
+	doorbell = mana_ucontext->doorbell;
+
+	if (is_rnic_cq) {
+		err = mana_ib_gd_create_cq(mdev, cq, doorbell);
+		if (err) {
+			ibdev_dbg(ibdev, "Failed to create RNIC cq, %d\n", err);
+			goto err_destroy_queue;
+		}
+
+		err = mana_ib_install_cq_cb(mdev, cq);
+		if (err) {
+			ibdev_dbg(ibdev, "Failed to install cq callback, %d\n", err);
+			goto err_destroy_rnic_cq;
+		}
+	}
+
+	resp.cqid = cq->queue.id;
+	err = ib_copy_to_udata(udata, &resp, min(sizeof(resp), udata->outlen));
+	if (err) {
+		ibdev_dbg(&mdev->ib_dev, "Failed to copy to udata, %d\n", err);
+		goto err_remove_cq_cb;
+	}
+
 	return 0;
+
+err_remove_cq_cb:
+	mana_ib_remove_cq_cb(mdev, cq);
+err_destroy_rnic_cq:
+	mana_ib_gd_destroy_cq(mdev, cq);
+err_destroy_queue:
+	mana_ib_destroy_queue(mdev, &cq->queue);
+
+	return err;
 }
 
 int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
@@ -52,6 +93,12 @@ int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
 	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
 
 	mana_ib_remove_cq_cb(mdev, cq);
+
+	/* Ignore return code as there is not much we can do about it.
+	 * The error message is printed inside.
+	 */
+	mana_ib_gd_destroy_cq(mdev, cq);
+
 	mana_ib_destroy_queue(mdev, &cq->queue);
 
 	return 0;
diff --git a/include/uapi/rdma/mana-abi.h b/include/uapi/rdma/mana-abi.h
index 5fcb31b..2c41cc3 100644
--- a/include/uapi/rdma/mana-abi.h
+++ b/include/uapi/rdma/mana-abi.h
@@ -16,8 +16,20 @@
 
 #define MANA_IB_UVERBS_ABI_VERSION 1
 
+enum mana_ib_create_cq_flags {
+	MANA_IB_CREATE_RNIC_CQ	= 1 << 0,
+};
+
 struct mana_ib_create_cq {
 	__aligned_u64 buf_addr;
+	__u16	flags;
+	__u16	reserved0;
+	__u32	reserved1;
+};
+
+struct mana_ib_create_cq_resp {
+	__u32 cqid;
+	__u32 reserved;
 };
 
 struct mana_ib_create_qp {
-- 
2.43.0

RE: [PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation of rnic cq

[Long Li]

> Subject: [PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation
> of rnic cq
> 
> From: Konstantin Taranov <kotaranov@microsoft.com>
> 
> Enable users to create RNIC CQs using a corresponding flag.
> With the previous request size, an ethernet CQ is created.
> As a response, return ID of the created CQ.
> 
> Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>
> ---
>  drivers/infiniband/hw/mana/cq.c | 55 ++++++++++++++++++++++++++++++---
>  include/uapi/rdma/mana-abi.h    | 12 +++++++
>  2 files changed, 63 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/infiniband/hw/mana/cq.c b/drivers/infiniband/hw/mana/cq.c
> index 688ffe6..c6a3fd5 100644
> --- a/drivers/infiniband/hw/mana/cq.c
> +++ b/drivers/infiniband/hw/mana/cq.c
> @@ -9,17 +9,22 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
> ib_cq_init_attr *attr,
>  		      struct ib_udata *udata)
>  {
>  	struct mana_ib_cq *cq = container_of(ibcq, struct mana_ib_cq, ibcq);
> +	struct mana_ib_create_cq_resp resp = {};
> +	struct mana_ib_ucontext *mana_ucontext;
>  	struct ib_device *ibdev = ibcq->device;
>  	struct mana_ib_create_cq ucmd = {};
>  	struct mana_ib_dev *mdev;
> +	bool is_rnic_cq;
> +	u32 doorbell;
>  	int err;
> 
>  	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
> 
> -	if (udata->inlen < sizeof(ucmd))
> -		return -EINVAL;
> -
>  	cq->comp_vector = attr->comp_vector % ibdev->num_comp_vectors;
> +	cq->cq_handle = INVALID_MANA_HANDLE;
> +
> +	if (udata->inlen < offsetof(struct mana_ib_create_cq, flags))
> +		return -EINVAL;
> 
>  	err = ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata-
> >inlen));
>  	if (err) {
> @@ -28,7 +33,9 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
> ib_cq_init_attr *attr,
>  		return err;
>  	}
> 
> -	if (attr->cqe > mdev->adapter_caps.max_qp_wr) {
> +	is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);
> +
> +	if (!is_rnic_cq && attr->cqe > mdev->adapter_caps.max_qp_wr) {
>  		ibdev_dbg(ibdev, "CQE %d exceeding limit\n", attr->cqe);
>  		return -EINVAL;
>  	}
> @@ -40,7 +47,41 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
> ib_cq_init_attr *attr,
>  		return err;
>  	}
> 
> +	mana_ucontext = rdma_udata_to_drv_context(udata, struct
> mana_ib_ucontext,
> +						  ibucontext);
> +	doorbell = mana_ucontext->doorbell;
> +
> +	if (is_rnic_cq) {
> +		err = mana_ib_gd_create_cq(mdev, cq, doorbell);
> +		if (err) {
> +			ibdev_dbg(ibdev, "Failed to create RNIC cq, %d\n", err);
> +			goto err_destroy_queue;
> +		}
> +
> +		err = mana_ib_install_cq_cb(mdev, cq);
> +		if (err) {
> +			ibdev_dbg(ibdev, "Failed to install cq callback, %d\n",
> err);
> +			goto err_destroy_rnic_cq;
> +		}
> +	}
> +
> +	resp.cqid = cq->queue.id;
> +	err = ib_copy_to_udata(udata, &resp, min(sizeof(resp), udata->outlen));
> +	if (err) {
> +		ibdev_dbg(&mdev->ib_dev, "Failed to copy to udata, %d\n", err);
> +		goto err_remove_cq_cb;
> +	}
> +
>  	return 0;
> +
> +err_remove_cq_cb:
> +	mana_ib_remove_cq_cb(mdev, cq);
> +err_destroy_rnic_cq:
> +	mana_ib_gd_destroy_cq(mdev, cq);
> +err_destroy_queue:
> +	mana_ib_destroy_queue(mdev, &cq->queue);
> +
> +	return err;
>  }
> 
>  int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata) @@ -52,6
> +93,12 @@ int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
>  	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
> 
>  	mana_ib_remove_cq_cb(mdev, cq);
> +
> +	/* Ignore return code as there is not much we can do about it.
> +	 * The error message is printed inside.
> +	 */
> +	mana_ib_gd_destroy_cq(mdev, cq);
> +
>  	mana_ib_destroy_queue(mdev, &cq->queue);
> 
>  	return 0;
> diff --git a/include/uapi/rdma/mana-abi.h b/include/uapi/rdma/mana-abi.h
> index 5fcb31b..2c41cc3 100644
> --- a/include/uapi/rdma/mana-abi.h
> +++ b/include/uapi/rdma/mana-abi.h
> @@ -16,8 +16,20 @@
> 
>  #define MANA_IB_UVERBS_ABI_VERSION 1
> 
> +enum mana_ib_create_cq_flags {
> +	MANA_IB_CREATE_RNIC_CQ	= 1 << 0,
> +};
> +
>  struct mana_ib_create_cq {
>  	__aligned_u64 buf_addr;
> +	__u16	flags;
> +	__u16	reserved0;
> +	__u32	reserved1;
> +};
> +
> +struct mana_ib_create_cq_resp {
> +	__u32 cqid;
> +	__u32 reserved;
>  };
> 
>  struct mana_ib_create_qp {
> --
> 2.43.0

For this review, it will be helpful if you can also post a link to the rdma-core changes.

Long

RE: [PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation of rnic cq

[Konstantin Taranov]

> 
> For this review, it will be helpful if you can also post a link to the rdma-core
> changes.
> 
> Long

Here is the PR to rdma-core with the changes: https://github.com/linux-rdma/rdma-core/pull/1455
The code was tested in 4 variations (old + new kernels against old + new rdma-cores) to confirm compatibility.

Konstantin

RE: [PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation of rnic cq

[Long Li]

> Here is the PR to rdma-core with the changes:
> https://github.co/
> m%2Flinux-rdma%2Frdma-
> core%2Fpull%2F1455&data=05%7C02%7Clongli%40microsoft.com%7C15661647
> 58b54b0d382508dc69e72a37%7C72f988bf86f141af91ab2d7cd011db47%7C1%7
> C0%7C638501688834172236%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLj
> AwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C
> &sdata=lZPsMJvU%2BeoVBajN5HZBNZaADKnwOfROv5OqXL52%2FF4%3D&reser
> ved=0
> The code was tested in 4 variations (old + new kernels against old + new rdma-
> cores) to confirm compatibility.
>
> Konstantin

There are minor issues with rdma-core change. This kernel patch looks good to me.

I have added "Reviewed-by" on this patch.

Long

RE: [PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation of rnic cq

[Long Li]

> Subject: [PATCH rdma-next v2 5/5] RDMA/mana_ib: implement uapi for creation
> of rnic cq
> 
> From: Konstantin Taranov <kotaranov@microsoft.com>
> 
> Enable users to create RNIC CQs using a corresponding flag.
> With the previous request size, an ethernet CQ is created.
> As a response, return ID of the created CQ.
> 
> Signed-off-by: Konstantin Taranov <kotaranov@microsoft.com>

Reviewed-by: Long Li <longli@microsoft.com>

> ---
>  drivers/infiniband/hw/mana/cq.c | 55 ++++++++++++++++++++++++++++++---
>  include/uapi/rdma/mana-abi.h    | 12 +++++++
>  2 files changed, 63 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/infiniband/hw/mana/cq.c b/drivers/infiniband/hw/mana/cq.c
> index 688ffe6..c6a3fd5 100644
> --- a/drivers/infiniband/hw/mana/cq.c
> +++ b/drivers/infiniband/hw/mana/cq.c
> @@ -9,17 +9,22 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
> ib_cq_init_attr *attr,
>  		      struct ib_udata *udata)
>  {
>  	struct mana_ib_cq *cq = container_of(ibcq, struct mana_ib_cq, ibcq);
> +	struct mana_ib_create_cq_resp resp = {};
> +	struct mana_ib_ucontext *mana_ucontext;
>  	struct ib_device *ibdev = ibcq->device;
>  	struct mana_ib_create_cq ucmd = {};
>  	struct mana_ib_dev *mdev;
> +	bool is_rnic_cq;
> +	u32 doorbell;
>  	int err;
> 
>  	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
> 
> -	if (udata->inlen < sizeof(ucmd))
> -		return -EINVAL;
> -
>  	cq->comp_vector = attr->comp_vector % ibdev->num_comp_vectors;
> +	cq->cq_handle = INVALID_MANA_HANDLE;
> +
> +	if (udata->inlen < offsetof(struct mana_ib_create_cq, flags))
> +		return -EINVAL;
> 
>  	err = ib_copy_from_udata(&ucmd, udata, min(sizeof(ucmd), udata-
> >inlen));
>  	if (err) {
> @@ -28,7 +33,9 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
> ib_cq_init_attr *attr,
>  		return err;
>  	}
> 
> -	if (attr->cqe > mdev->adapter_caps.max_qp_wr) {
> +	is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);
> +
> +	if (!is_rnic_cq && attr->cqe > mdev->adapter_caps.max_qp_wr) {
>  		ibdev_dbg(ibdev, "CQE %d exceeding limit\n", attr->cqe);
>  		return -EINVAL;
>  	}
> @@ -40,7 +47,41 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
> ib_cq_init_attr *attr,
>  		return err;
>  	}
> 
> +	mana_ucontext = rdma_udata_to_drv_context(udata, struct
> mana_ib_ucontext,
> +						  ibucontext);
> +	doorbell = mana_ucontext->doorbell;
> +
> +	if (is_rnic_cq) {
> +		err = mana_ib_gd_create_cq(mdev, cq, doorbell);
> +		if (err) {
> +			ibdev_dbg(ibdev, "Failed to create RNIC cq, %d\n", err);
> +			goto err_destroy_queue;
> +		}
> +
> +		err = mana_ib_install_cq_cb(mdev, cq);
> +		if (err) {
> +			ibdev_dbg(ibdev, "Failed to install cq callback, %d\n",
> err);
> +			goto err_destroy_rnic_cq;
> +		}
> +	}
> +
> +	resp.cqid = cq->queue.id;
> +	err = ib_copy_to_udata(udata, &resp, min(sizeof(resp), udata->outlen));
> +	if (err) {
> +		ibdev_dbg(&mdev->ib_dev, "Failed to copy to udata, %d\n", err);
> +		goto err_remove_cq_cb;
> +	}
> +
>  	return 0;
> +
> +err_remove_cq_cb:
> +	mana_ib_remove_cq_cb(mdev, cq);
> +err_destroy_rnic_cq:
> +	mana_ib_gd_destroy_cq(mdev, cq);
> +err_destroy_queue:
> +	mana_ib_destroy_queue(mdev, &cq->queue);
> +
> +	return err;
>  }
> 
>  int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata) @@ -52,6
> +93,12 @@ int mana_ib_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata)
>  	mdev = container_of(ibdev, struct mana_ib_dev, ib_dev);
> 
>  	mana_ib_remove_cq_cb(mdev, cq);
> +
> +	/* Ignore return code as there is not much we can do about it.
> +	 * The error message is printed inside.
> +	 */
> +	mana_ib_gd_destroy_cq(mdev, cq);
> +
>  	mana_ib_destroy_queue(mdev, &cq->queue);
> 
>  	return 0;
> diff --git a/include/uapi/rdma/mana-abi.h b/include/uapi/rdma/mana-abi.h
> index 5fcb31b..2c41cc3 100644
> --- a/include/uapi/rdma/mana-abi.h
> +++ b/include/uapi/rdma/mana-abi.h
> @@ -16,8 +16,20 @@
> 
>  #define MANA_IB_UVERBS_ABI_VERSION 1
> 
> +enum mana_ib_create_cq_flags {
> +	MANA_IB_CREATE_RNIC_CQ	= 1 << 0,
> +};
> +
>  struct mana_ib_create_cq {
>  	__aligned_u64 buf_addr;
> +	__u16	flags;
> +	__u16	reserved0;
> +	__u32	reserved1;
> +};
> +
> +struct mana_ib_create_cq_resp {
> +	__u32 cqid;
> +	__u32 reserved;
>  };
> 
>  struct mana_ib_create_qp {
> --
> 2.43.0

Re: [PATCH rdma-next v2 0/5] RDMA/mana_ib: Implement RNIC CQs

[Leon Romanovsky]

On Fri, 26 Apr 2024 06:12:35 -0700, Konstantin Taranov wrote:
> From: Konstantin Taranov <kotaranov@microsoft.com>
> 
> This patch series implements creation and destruction of CQs
> which can be used with RC QPs.
> 
> Patches with RC QPs will be sent in the next patch series.
> 
> [...]

Applied, thanks!

[1/5] RDMA/mana_ib: create EQs for RNIC CQs
      https://git.kernel.org/rdma/rdma/c/e73c882f0a0149
[2/5] RDMA/mana_ib: create and destroy RNIC cqs
      https://git.kernel.org/rdma/rdma/c/58434159168529
[3/5] RDMA/mana_ib: introduce a helper to remove cq callbacks
      https://git.kernel.org/rdma/rdma/c/3e41105263d5d7
[4/5] RDMA/mana_ib: boundary check before installing cq callbacks
      https://git.kernel.org/rdma/rdma/c/f79edef79b6a21
[5/5] RDMA/mana_ib: implement uapi for creation of rnic cq
      https://git.kernel.org/rdma/rdma/c/44b607ad4cdf23

Best regards,
-- 
Leon Romanovsky <leon@kernel.org>