Skip to contentSkip to conversation

RDMA/mana_ib: take CQ type from the device type

PATCH Superseded Konstantin Taranov 4 participants 6 messages Jan 9 – Jan 13

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 RDMA/mana_ib: take CQ type from the device type

From: Konstantin Taranov <[email protected]>

Get CQ type from the used gdma device. The MANA_IB_CREATE_RNIC_CQ flag is ignored. It was used in older kernel versions where the mana_ib was shared between ethernet and rnic.

Fixes d4293f96ce0b ("RDMA/mana_ib: unify mana_ib functions to support any gdma device")
---
 drivers/infiniband/hw/mana/cq.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/infiniband/hw/mana/cq.c +1 -3
index 1becc8779..2dce1b677 100644
@@ -24,6 +24,7 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 
 	cq->comp_vector = attr->comp_vector % ibdev->num_comp_vectors;
 	cq->cq_handle = INVALID_MANA_HANDLE;
+	is_rnic_cq = mana_ib_is_rnic(mdev);
 
 	if (udata) {
 		if (udata->inlen < offsetof(struct mana_ib_create_cq, flags))
@@ -35,8 +36,6 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 			return err;
 		}
 
-		is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);
-
 		if ((!is_rnic_cq && attr->cqe > mdev->adapter_caps.max_qp_wr) ||
 		    attr->cqe > U32_MAX / COMP_ENTRY_SIZE) {
 			ibdev_dbg(ibdev, "CQE %d exceeding limit\n", attr->cqe);
@@ -55,7 +54,6 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 							  ibucontext);
 		doorbell = mana_ucontext->doorbell;
 	} else {
-		is_rnic_cq = true;
 		buf_size = MANA_PAGE_ALIGN(roundup_pow_of_two(attr->cqe * COMP_ENTRY_SIZE));
 		cq->cqe = buf_size / COMP_ENTRY_SIZE;
 		err = mana_ib_create_kernel_queue(mdev, buf_size, GDMA_CQ, &cq->queue);
Signature · 1 line
2.43.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 5 lines

Subject: [PATCH rdma-next 1/1] RDMA/mana_ib: take CQ type from the device type

From: Konstantin Taranov <[email protected]>

Quoted 7 lines

Get CQ type from the used gdma device. The MANA_IB_CREATE_RNIC_CQ flag is ignored. It was used in older kernel versions where the mana_ib was shared between ethernet and rnic.

Fixes d4293f96ce0b ("RDMA/mana_ib: unify mana_ib functions to support

any gdma device")

Quoted 41 lines
---
 drivers/infiniband/hw/mana/cq.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
(quoted diff) +0 -0

b/drivers/infiniband/hw/mana/cq.c index 1becc8779..2dce1b677 100644

drivers/infiniband/hw/mana/cq.c +1 -0
@@ -24,6 +24,7 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
 ib_cq_init_attr *attr,
 
 	cq->comp_vector = attr->comp_vector % ibdev->num_comp_vectors;
 	cq->cq_handle = INVALID_MANA_HANDLE;
+	is_rnic_cq = mana_ib_is_rnic(mdev);
 
 	if (udata) {
drivers/infiniband/hw/mana/cq.c +0 -2
 		if (udata->inlen < offsetof(struct mana_ib_create_cq, flags))
@@ -35,8 +36,6 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct
 ib_cq_init_attr *attr,
 			return err;
 		}
 
-		is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);
-
 		if ((!is_rnic_cq && attr->cqe > mdev-
Quoted 1 line

adapter_caps.max_qp_wr) ||

drivers/infiniband/hw/mana/cq.c +0 -0
 		    attr->cqe > U32_MAX / COMP_ENTRY_SIZE) {
 			ibdev_dbg(ibdev, "CQE %d exceeding limit\n", attr-
Quoted 1 line

cqe); @@ -55,7 +54,6 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const

struct ib_cq_init_attr *attr,

drivers/infiniband/hw/mana/cq.c +0 -1
 							  ibucontext);
 		doorbell = mana_ucontext->doorbell;
 	} else {
-		is_rnic_cq = true;
 		buf_size = MANA_PAGE_ALIGN(roundup_pow_of_two(attr-
Quoted 1 line

cqe * COMP_ENTRY_SIZE));

drivers/infiniband/hw/mana/cq.c +0 -0
 		cq->cqe = buf_size / COMP_ENTRY_SIZE;
 		err = mana_ib_create_kernel_queue(mdev, buf_size,

GDMA_CQ, &cq->queue);
--
2.43.0

#2 lore raw
In reply to Konstantin Taranov
Quoted 25 lines · Konstantin Taranov

From: Konstantin Taranov <[email protected]>

Get CQ type from the used gdma device. The MANA_IB_CREATE_RNIC_CQ flag is ignored. It was used in older kernel versions where the mana_ib was shared between ethernet and rnic.

Fixes d4293f96ce0b ("RDMA/mana_ib: unify mana_ib functions to support any gdma device")
---
 drivers/infiniband/hw/mana/cq.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)
drivers/infiniband/hw/mana/cq.c +1 -0
index 1becc8779..2dce1b677 100644
@@ -24,6 +24,7 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 
 	cq->comp_vector = attr->comp_vector % ibdev->num_comp_vectors;
 	cq->cq_handle = INVALID_MANA_HANDLE;
+	is_rnic_cq = mana_ib_is_rnic(mdev);
 
 	if (udata) {
 		if (udata->inlen < offsetof(struct mana_ib_create_cq, flags))

drivers/infiniband/hw/mana/cq.c int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, · line 38

Quoted 4 lines
drivers/infiniband/hw/mana/cq.c +0 -1
@@ -35,8 +36,6 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 			return err;
 		}
 
-		is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);

You need to add code which prohibits future use of this BIT(0) in ucmd.flags for backward compatibility and maybe delete MANA_IB_CREATE_RNIC_CQ from UAPI too.

Thanks

Quoted 15 lines
drivers/infiniband/hw/mana/cq.c +0 -2
-
 		if ((!is_rnic_cq && attr->cqe > mdev->adapter_caps.max_qp_wr) ||
 		    attr->cqe > U32_MAX / COMP_ENTRY_SIZE) {
 			ibdev_dbg(ibdev, "CQE %d exceeding limit\n", attr->cqe);
@@ -55,7 +54,6 @@ int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
 							  ibucontext);
 		doorbell = mana_ucontext->doorbell;
 	} else {
-		is_rnic_cq = true;
 		buf_size = MANA_PAGE_ALIGN(roundup_pow_of_two(attr->cqe * COMP_ENTRY_SIZE));
 		cq->cqe = buf_size / COMP_ENTRY_SIZE;
 		err = mana_ib_create_kernel_queue(mdev, buf_size, GDMA_CQ, &cq->queue);
Signature · 1 line
2.43.0
#3 lore raw
Quoted 9 lines

drivers/infiniband/hw/mana/cq.c int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, · line 38

Quoted 2 lines
drivers/infiniband/hw/mana/cq.c +0 -1
-		is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);

You need to add code which prohibits future use of this BIT(0) in ucmd.flags for backward compatibility and maybe delete MANA_IB_CREATE_RNIC_CQ from UAPI too.

Thanks

Hi Leon. I thought that my proposed change is backward and forward compatible. If I add code that prohibits this flag, then the older rdma-core will fail to create CQ, as it sets this flag. Add rdma-core should set the flag to support older kernels.

So, the current solution is as follows:
rdma-core always sends the flag. The kernels without this patch still use this flag. Newer kernels just ignore the flag and create the CQ according to the client. It is not fully possible to retire this flag now, as we want to be backwards compatible and support older kernels and older rdma-core.
Or did you mean something else? Or do I miss something?

Thanks

#4 lore raw
Quoted 14 lines · Konstantin Taranov
Quoted 9 lines

drivers/infiniband/hw/mana/cq.c int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, · line 38

Quoted 2 lines
drivers/infiniband/hw/mana/cq.c +0 -1
-		is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);

You need to add code which prohibits future use of this BIT(0) in ucmd.flags for backward compatibility and maybe delete MANA_IB_CREATE_RNIC_CQ from UAPI too.

Thanks

Hi Leon. I thought that my proposed change is backward and forward compatible. If I add code that prohibits this flag, then the older rdma-core will fail to create CQ, as it sets this flag. Add rdma-core should set the flag to support older kernels.

Quoted 6 lines

So, the current solution is as follows:
rdma-core always sends the flag. The kernels without this patch still use this flag. Newer kernels just ignore the flag and create the CQ according to the client. It is not fully possible to retire this flag now, as we want to be backwards compatible and support older kernels and older rdma-core.
Or did you mean something else? Or do I miss something?

There needs to be a way to document in the code that this bit is reserved and must not be used.

Thanks

Quoted 2 lines

Thanks

#5 lore raw
Quoted 19 lines · Konstantin Taranov
Quoted 11 lines
Quoted 9 lines

drivers/infiniband/hw/mana/cq.c int mana_ib_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, · line 38

Quoted 2 lines
drivers/infiniband/hw/mana/cq.c +0 -1
-		is_rnic_cq = !!(ucmd.flags & MANA_IB_CREATE_RNIC_CQ);

You need to add code which prohibits future use of this BIT(0) in ucmd.flags for backward compatibility and maybe delete MANA_IB_CREATE_RNIC_CQ from UAPI too.

Thanks

Hi Leon. I thought that my proposed change is backward and forward

compatible.

Quoted 2 lines

If I add code that prohibits this flag, then the older rdma-core will fail to create CQ, as it sets this flag. Add rdma-core should set the flag to

support older kernels.

Quoted 3 lines

So, the current solution is as follows:
rdma-core always sends the flag. The kernels without this patch still use

Quoted 8 lines

this flag.

Quoted 4 lines

Newer kernels just ignore the flag and create the CQ according to the client. It is not fully possible to retire this flag now, as we want to be backwards compatible and support older kernels and older rdma-core. Or did you mean something else? Or do I miss something?

There needs to be a way to document in the code that this bit is reserved and must not be used.

Got it! I will add comments to the code that the bit is reserved and should not be reused.

Thanks

Quoted 5 lines

Thanks

Quoted 2 lines

Thanks