From: Leon Romanovsky <leon@kernel.org>
To: Cheng Xu <chengyou@linux.alibaba.com>
Cc: linux-kernel@vger.kernel.org, linux-rdma@vger.kernel.org,
linux-hyperv@vger.kernel.org
Subject: Re: [PATCH rdma-next 18/50] RDMA/erdma: Separate user and kernel CQ creation paths
Date: Tue, 24 Feb 2026 12:57:02 +0200 [thread overview]
Message-ID: <20260224105702.GK10607@unreal> (raw)
In-Reply-To: <d21833fa-a737-3b46-dda3-92837f78f8e4@linux.alibaba.com>
On Tue, Feb 24, 2026 at 01:51:41PM +0800, Cheng Xu wrote:
>
>
> On 2/13/26 6:57 PM, Leon Romanovsky wrote:
> > From: Leon Romanovsky <leonro@nvidia.com>
> >
> > Split CQ creation into distinct kernel and user flows. The erdma driver,
> > inherited from mlx4, uses a problematic pattern that shares and caches
> > umem in erdma_map_user_dbrecords(). This design blocks the driver from
> > supporting generic umem sources (VMA, dmabuf, memfd, and others).
> >
> > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > ---
> > drivers/infiniband/hw/erdma/erdma_main.c | 1 +
> > drivers/infiniband/hw/erdma/erdma_verbs.c | 97 ++++++++++++++++++++-----------
> > drivers/infiniband/hw/erdma/erdma_verbs.h | 2 +
> > 3 files changed, 67 insertions(+), 33 deletions(-)
> >
> > diff --git a/drivers/infiniband/hw/erdma/erdma_main.c b/drivers/infiniband/hw/erdma/erdma_main.c
> > index f35b30235018..1b6426e89d80 100644
> > --- a/drivers/infiniband/hw/erdma/erdma_main.c
> > +++ b/drivers/infiniband/hw/erdma/erdma_main.c
> > @@ -505,6 +505,7 @@ static const struct ib_device_ops erdma_device_ops = {
> > .alloc_pd = erdma_alloc_pd,
> > .alloc_ucontext = erdma_alloc_ucontext,
> > .create_cq = erdma_create_cq,
> > + .create_user_cq = erdma_create_user_cq,
> > .create_qp = erdma_create_qp,
> > .dealloc_pd = erdma_dealloc_pd,
> > .dealloc_ucontext = erdma_dealloc_ucontext,
>
> <...>
>
> > +
> > +int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr,
> > + struct uverbs_attr_bundle *attrs)
>
> create_cq will be used for kernel CQ creation, and the third input parameter
> 'struct uverbs_attr_bundle *attrs' will be useless, so it can be removed? Same to
> all drivers.
Yes, but only after conversion of all drivers. I have that removal patch
in my v2.
>
>
> > +{
>
> <...>
>
> > + ret = create_cq_cmd(NULL, cq);
> > + if (ret)
> > + goto err_free_res;
>
>
> In create_cq_cmd, should add the following change:
I took slightly different approach and inlined create_cq_cmd() into erdma_create_*_cq().
Thanks
>
> diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.c b/drivers/infiniband/hw/erdma/erdma_verbs.c
> index 8c30df61ae3d..eca28524e04b 100644
> --- a/drivers/infiniband/hw/erdma/erdma_verbs.c
> +++ b/drivers/infiniband/hw/erdma/erdma_verbs.c
> @@ -240,7 +240,7 @@ static int create_cq_cmd(struct erdma_ucontext *uctx, struct erdma_cq *cq)
> req.first_page_offset = mem->page_offset;
> req.cq_dbrec_dma = cq->user_cq.dbrec_dma;
>
> - if (uctx->ext_db.enable) {
> + if (uctx && uctx->ext_db.enable) {
> req.cfg1 |= FIELD_PREP(
> ERDMA_CMD_CREATE_CQ_MTT_DB_CFG_MASK, 1);
> req.cfg2 = FIELD_PREP(ERDMA_CMD_CREATE_CQ_DB_CFG_MASK,
>
>
> Thanks,
> Cheng Xu
>
next prev parent reply other threads:[~2026-02-24 10:57 UTC|newest]
Thread overview: 73+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-02-13 10:57 [PATCH rdma-next 00/50] RDMA: Ensure CQ UMEMs are managed by ib_core Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 01/50] RDMA: Move DMA block iterator logic into dedicated files Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 02/50] RDMA/umem: Allow including ib_umem header from any location Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 03/50] RDMA/umem: Remove unnecessary includes and defines from ib_umem header Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 04/50] RDMA/core: Promote UMEM to a core component Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 05/50] RDMA/core: Manage CQ umem in core code Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 06/50] RDMA/efa: Rely on CPU address in create‑QP Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 07/50] RDMA/core: Prepare create CQ path for API unification Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 08/50] RDMA/core: Reject zero CQE count Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 09/50] RDMA/efa: Remove check for " Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 10/50] RDMA/mlx5: Save 4 bytes in CQ structure Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 11/50] RDMA/mlx5: Provide a modern CQ creation interface Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 12/50] RDMA/mlx4: Inline mlx4_ib_get_cq_umem into callers Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 13/50] RDMA/mlx4: Introduce a modern CQ creation interface Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 14/50] RDMA/mlx4: Remove unused create_flags field from CQ structure Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 15/50] RDMA/bnxt_re: Convert to modern CQ interface Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 16/50] RDMA/cxgb4: Separate kernel and user CQ creation paths Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 17/50] RDMA/mthca: Split user and kernel " Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 18/50] RDMA/erdma: Separate " Leon Romanovsky
2026-02-24 5:51 ` Cheng Xu
2026-02-24 10:57 ` Leon Romanovsky [this message]
2026-02-13 10:57 ` [PATCH rdma-next 19/50] RDMA/ionic: Split " Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 20/50] RDMA/qedr: Convert to modern CQ interface Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 21/50] RDMA/vmw_pvrdma: Provide a modern CQ creation interface Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 22/50] RDMA/ocrdma: Split user and kernel CQ creation paths Leon Romanovsky
2026-02-13 10:57 ` [PATCH rdma-next 23/50] RDMA/irdma: " Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 24/50] RDMA/usnic: Provide a modern CQ creation interface Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 25/50] RDMA/mana: " Leon Romanovsky
2026-02-24 22:30 ` [EXTERNAL] " Long Li
2026-02-25 8:24 ` Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 26/50] RDMA/erdma: Separate user and kernel CQ creation paths Leon Romanovsky
2026-02-24 2:20 ` Cheng Xu
2026-02-24 10:46 ` Leon Romanovsky
2026-02-26 6:17 ` Junxian Huang
2026-02-26 6:54 ` Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 27/50] RDMA/rdmavt: Split " Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 28/50] RDMA/siw: " Leon Romanovsky
2026-02-13 16:56 ` Bernard Metzler
2026-02-13 21:17 ` Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 29/50] RDMA/rxe: " Leon Romanovsky
2026-02-13 23:22 ` yanjun.zhu
2026-02-15 7:06 ` Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 30/50] RDMA/core: Remove legacy CQ creation fallback path Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 31/50] RDMA/core: Remove unused ib_resize_cq() implementation Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 32/50] RDMA: Clarify that CQ resize is a user‑space verb Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 33/50] RDMA/bnxt_re: Drop support for resizing kernel CQs Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 34/50] RDMA/irdma: Remove resize support for " Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 35/50] RDMA/mlx4: Remove support for kernel CQ resize Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 36/50] RDMA/mlx5: Remove support for resizing kernel CQs Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 37/50] RDMA/mthca: Remove resize support for " Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 38/50] RDMA/rdmavt: " Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 39/50] RDMA/rxe: Remove unused kernel‑side CQ resize support Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 40/50] RDMA: Properly propagate the number of CQEs as unsigned int Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 41/50] RDMA/core: Generalize CQ resize locking Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 42/50] RDMA/bnxt_re: Complete CQ resize in a single step Leon Romanovsky
2026-02-16 3:59 ` Selvin Xavier
2026-02-16 8:07 ` Leon Romanovsky
2026-02-17 5:02 ` Selvin Xavier
2026-02-17 7:56 ` Leon Romanovsky
2026-02-17 10:52 ` Selvin Xavier
2026-02-19 8:02 ` Selvin Xavier
2026-02-24 8:15 ` Selvin Xavier
2026-02-24 10:59 ` Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 43/50] RDMA/bnxt_re: Rely on common resize‑CQ locking Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 44/50] RDMA/bnxt_re: Reduce CQ memory footprint Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 45/50] RDMA/mlx4: Use generic resize-CQ lock Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 46/50] RDMA/mlx4: Use on‑stack variables instead of storing them in the CQ object Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 47/50] RDMA/mlx5: Use generic resize-CQ lock Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 48/50] RDMA/mlx5: Select resize‑CQ callback based on device capabilities Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 49/50] RDMA/mlx5: Reduce CQ memory footprint Leon Romanovsky
2026-02-13 10:58 ` [PATCH rdma-next 50/50] RDMA/mthca: Use generic resize-CQ lock Leon Romanovsky
2026-02-25 13:51 ` (subset) [PATCH rdma-next 00/50] RDMA: Ensure CQ UMEMs are managed by ib_core Leon Romanovsky
2026-02-25 13:53 ` Leon Romanovsky
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260224105702.GK10607@unreal \
--to=leon@kernel.org \
--cc=chengyou@linux.alibaba.com \
--cc=linux-hyperv@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rdma@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox