From nobody Thu Apr 2 22:08:38 2026 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 5C08F3559F8; Fri, 13 Feb 2026 11:00:04 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=10.30.226.201 ARC-Seal: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1770980404; cv=none; b=jxFDZ66bTNz69Kd+28stA8ZPpjn2L4vdZ/JMHaWWeQ+SlSiyZgYFrvKX65XSvdxyjmPk9YlXvkNGwNGvMTLV3KfFStgLml7qIRyH7oTzm7n5yto3EcnDNUpR6RUrU6dNg/lG9NEn0XlVQltmVxhwIAcl8HwzDWQFxNx9wgZk4v4= ARC-Message-Signature: i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1770980404; c=relaxed/simple; bh=hn7YzjvK8jxXt9+ur6KEPT/X9OAdVySLRDWe9dlWIoU=; h=From:To:Cc:Subject:Date:Message-ID:In-Reply-To:References: MIME-Version:Content-Type; b=uSZiw0/6Lj8yKrfwz2tAxyIYf/E1DAlm31fgs7v/L0hrrJo7QrLMHZgICAHldqVaO+XtSuTngChAUs+IZr3wMUvPb8cBqnkwr1qIpmM7/0zbMCM0TqrmebmLMFYUIVrzErbyJNUDflaEWga6ysNWAgJtezV2ToD3+6EjSd2u2HU= ARC-Authentication-Results: i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=FUFtQUL9; arc=none smtp.client-ip=10.30.226.201 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="FUFtQUL9" Received: by smtp.kernel.org (Postfix) with ESMTPSA id 4D54CC19423; Fri, 13 Feb 2026 11:00:03 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1770980404; bh=hn7YzjvK8jxXt9+ur6KEPT/X9OAdVySLRDWe9dlWIoU=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=FUFtQUL9JdM3h8II8cr+oS/z4A3tqdTDCJQg9NkOOBhopVK1gyJOFlNPMRUAx99lq mBkEuUBhinWZcAUITdRxLb9jvbW2D/RdyUNskf/H5u3Fcxe+H9ErtrhkqoWT/Nzcqj MKHuQUkz+kOxu5SD3xiTvVBNtTpkknBNqJyD2pBB6iMxY9CgsA9sesktV8XZp/x5W8 37U3HqWnHjd7/YtOWvHbcIkbxmqZjJcopIrBu8TV2C91uxLk3h9I+O6Gqe0cIEHkrr fyzug4yNYCJrwTK75XV464doH3PK2gdITx/ObI04grdqx5/QvOiO8rkxPQk6Xyd3I9 CsdggKu1iR+Pg== From: Leon Romanovsky To: Jason Gunthorpe , Leon Romanovsky , Selvin Xavier , Kalesh AP , Potnuri Bharat Teja , Michael Margolin , Gal Pressman , Yossi Leybovich , Cheng Xu , Kai Shen , Chengchang Tang , Junxian Huang , Abhijit Gangurde , Allen Hubbe , Krzysztof Czurylo , Tatyana Nikolova , Long Li , Konstantin Taranov , Yishai Hadas , Michal Kalderon , Bryan Tan , Vishnu Dasa , Broadcom internal kernel review list , Christian Benvenuti , Nelson Escobar , Dennis Dalessandro , Bernard Metzler , Zhu Yanjun Cc: linux-kernel@vger.kernel.org, linux-rdma@vger.kernel.org, linux-hyperv@vger.kernel.org Subject: [PATCH rdma-next 18/50] RDMA/erdma: Separate user and kernel CQ creation paths Date: Fri, 13 Feb 2026 12:57:54 +0200 Message-ID: <20260213-refactor-umem-v1-18-f3be85847922@nvidia.com> X-Mailer: git-send-email 2.53.0 In-Reply-To: <20260213-refactor-umem-v1-0-f3be85847922@nvidia.com> References: <20260213-refactor-umem-v1-0-f3be85847922@nvidia.com> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset="utf-8" X-Mailer: b4 0.15-dev-47773 Content-Transfer-Encoding: quoted-printable From: Leon Romanovsky 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 --- 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 =3D { .alloc_pd =3D erdma_alloc_pd, .alloc_ucontext =3D erdma_alloc_ucontext, .create_cq =3D erdma_create_cq, + .create_user_cq =3D erdma_create_user_cq, .create_qp =3D erdma_create_qp, .dealloc_pd =3D erdma_dealloc_pd, .dealloc_ucontext =3D erdma_dealloc_ucontext, diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.c b/drivers/infiniband= /hw/erdma/erdma_verbs.c index 058edc42de58..6f809907fec5 100644 --- a/drivers/infiniband/hw/erdma/erdma_verbs.c +++ b/drivers/infiniband/hw/erdma/erdma_verbs.c @@ -1952,8 +1952,8 @@ static int erdma_init_kernel_cq(struct erdma_cq *cq) return -ENOMEM; } =20 -int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, - struct uverbs_attr_bundle *attrs) +int erdma_create_user_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr = *attr, + struct uverbs_attr_bundle *attrs) { struct ib_udata *udata =3D &attrs->driver_udata; struct erdma_cq *cq =3D to_ecq(ibcq); @@ -1962,6 +1962,11 @@ int erdma_create_cq(struct ib_cq *ibcq, const struct= ib_cq_init_attr *attr, int ret; struct erdma_ucontext *ctx =3D rdma_udata_to_drv_context( udata, struct erdma_ucontext, ibucontext); + struct erdma_ureq_create_cq ureq; + struct erdma_uresp_create_cq uresp; + + if (ibcq->umem) + return -EOPNOTSUPP; =20 if (depth > dev->attrs.max_cqe) return -EINVAL; @@ -1977,31 +1982,22 @@ int erdma_create_cq(struct ib_cq *ibcq, const struc= t ib_cq_init_attr *attr, if (ret < 0) return ret; =20 - if (!rdma_is_kernel_res(&ibcq->res)) { - struct erdma_ureq_create_cq ureq; - struct erdma_uresp_create_cq uresp; - - ret =3D ib_copy_from_udata(&ureq, udata, - min(udata->inlen, sizeof(ureq))); - if (ret) - goto err_out_xa; + ret =3D ib_copy_from_udata(&ureq, udata, + min(udata->inlen, sizeof(ureq))); + if (ret) + goto err_out_xa; =20 - ret =3D erdma_init_user_cq(ctx, cq, &ureq); - if (ret) - goto err_out_xa; + ret =3D erdma_init_user_cq(ctx, cq, &ureq); + if (ret) + goto err_out_xa; =20 - uresp.cq_id =3D cq->cqn; - uresp.num_cqe =3D depth; + uresp.cq_id =3D cq->cqn; + uresp.num_cqe =3D depth; =20 - ret =3D ib_copy_to_udata(udata, &uresp, - min(sizeof(uresp), udata->outlen)); - if (ret) - goto err_free_res; - } else { - ret =3D erdma_init_kernel_cq(cq); - if (ret) - goto err_out_xa; - } + ret =3D ib_copy_to_udata(udata, &uresp, + min(sizeof(uresp), udata->outlen)); + if (ret) + goto err_free_res; =20 ret =3D create_cq_cmd(ctx, cq); if (ret) @@ -2010,19 +2006,54 @@ int erdma_create_cq(struct ib_cq *ibcq, const struc= t ib_cq_init_attr *attr, return 0; =20 err_free_res: - if (!rdma_is_kernel_res(&ibcq->res)) { - erdma_unmap_user_dbrecords(ctx, &cq->user_cq.user_dbr_page); - put_mtt_entries(dev, &cq->user_cq.qbuf_mem); - } else { - dma_free_coherent(&dev->pdev->dev, depth << CQE_SHIFT, - cq->kern_cq.qbuf, cq->kern_cq.qbuf_dma_addr); - dma_pool_free(dev->db_pool, cq->kern_cq.dbrec, - cq->kern_cq.dbrec_dma); - } + erdma_unmap_user_dbrecords(ctx, &cq->user_cq.user_dbr_page); + put_mtt_entries(dev, &cq->user_cq.qbuf_mem); =20 err_out_xa: xa_erase(&dev->cq_xa, cq->cqn); + return ret; +} + +int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, + struct uverbs_attr_bundle *attrs) +{ + struct erdma_cq *cq =3D to_ecq(ibcq); + struct erdma_dev *dev =3D to_edev(ibcq->device); + unsigned int depth =3D attr->cqe; + int ret; + + if (depth > dev->attrs.max_cqe) + return -EINVAL; =20 + depth =3D roundup_pow_of_two(depth); + cq->ibcq.cqe =3D depth; + cq->depth =3D depth; + cq->assoc_eqn =3D attr->comp_vector + 1; + + ret =3D xa_alloc_cyclic(&dev->cq_xa, &cq->cqn, cq, + XA_LIMIT(1, dev->attrs.max_cq - 1), + &dev->next_alloc_cqn, GFP_KERNEL); + if (ret < 0) + return ret; + + ret =3D erdma_init_kernel_cq(cq); + if (ret) + goto err_out_xa; + + ret =3D create_cq_cmd(NULL, cq); + if (ret) + goto err_free_res; + + return 0; + +err_free_res: + dma_free_coherent(&dev->pdev->dev, depth << CQE_SHIFT, + cq->kern_cq.qbuf, cq->kern_cq.qbuf_dma_addr); + dma_pool_free(dev->db_pool, cq->kern_cq.dbrec, + cq->kern_cq.dbrec_dma); + +err_out_xa: + xa_erase(&dev->cq_xa, cq->cqn); return ret; } =20 diff --git a/drivers/infiniband/hw/erdma/erdma_verbs.h b/drivers/infiniband= /hw/erdma/erdma_verbs.h index 7d8d3fe501d5..21a4fb404806 100644 --- a/drivers/infiniband/hw/erdma/erdma_verbs.h +++ b/drivers/infiniband/hw/erdma/erdma_verbs.h @@ -435,6 +435,8 @@ int erdma_get_port_immutable(struct ib_device *dev, u32= port, struct ib_port_immutable *ib_port_immutable); int erdma_create_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr *attr, struct uverbs_attr_bundle *attrs); +int erdma_create_user_cq(struct ib_cq *ibcq, const struct ib_cq_init_attr = *attr, + struct uverbs_attr_bundle *attrs); int erdma_query_port(struct ib_device *dev, u32 port, struct ib_port_attr *attr); int erdma_query_gid(struct ib_device *dev, u32 port, int idx, --=20 2.52.0