From patchwork Thu Aug 1 12:05:10 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750236 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 462F8C3DA7F for ; Thu, 1 Aug 2024 12:05:38 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 994B610E92F; Thu, 1 Aug 2024 12:05:36 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="f8tlBfJo"; dkim-atps=neutral Received: from dfw.source.kernel.org (dfw.source.kernel.org [139.178.84.217]) by gabe.freedesktop.org (Postfix) with ESMTPS id 86B2C10E92E for ; Thu, 1 Aug 2024 12:05:35 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by dfw.source.kernel.org (Postfix) with ESMTP id C012962857; Thu, 1 Aug 2024 12:05:34 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 16003C32786; Thu, 1 Aug 2024 12:05:33 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513934; bh=yHsuGHxiLjgWyN5xswDwNMl+wTJR3b2GStP/fjg5kt4=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=f8tlBfJo2NqAZZS65pSrGZVWv9Qzy47YMB34cU0sxgOaihlTEZdZKqm2VEHPgGkDm YBtp5Ezayj34aO17JzWcQiD5bNmVM89D5Vgbz5cZ9uwEhjfAaIQU5b5OuUECtL9ApT wAtu1mg0zSnUHuEM9YvjVumzl2TjaxbdZKYycWliUZZ3YVm2ndz6zkklroAVWWFGKC q90fY+es675tl6BfcZEXSK7eu7fl0o5nzAw6ZFbNsAKMp0fFvFVZEcnD5eo0a6DErx 3HH2Rp+5F2L+uEvv7OTpE6xyg/0UExPdXLn2TUoERC8loOHJ+SUz742M8AmjqjIKjX KNZEXveWw24yg== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH mlx5-next 1/8] net/mlx5: Add IFC related stuff for data direct Date: Thu, 1 Aug 2024 15:05:10 +0300 Message-ID: <82da7f578a567909bb5858a64ba844fe4cc298fa.1722512548.git.leon@kernel.org> X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Add IFC related stuff for data direct. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- include/linux/mlx5/mlx5_ifc.h | 51 +++++++++++++++++++++++++++++++---- 1 file changed, 46 insertions(+), 5 deletions(-) diff --git a/include/linux/mlx5/mlx5_ifc.h b/include/linux/mlx5/mlx5_ifc.h index cab228cf51c6..970c9d8473ef 100644 --- a/include/linux/mlx5/mlx5_ifc.h +++ b/include/linux/mlx5/mlx5_ifc.h @@ -313,6 +313,7 @@ enum { MLX5_CMD_OP_MODIFY_VHCA_STATE = 0xb0e, MLX5_CMD_OP_SYNC_CRYPTO = 0xb12, MLX5_CMD_OP_ALLOW_OTHER_VHCA_ACCESS = 0xb16, + MLX5_CMD_OPCODE_QUERY_VUID = 0xb22, MLX5_CMD_OP_MAX }; @@ -1885,7 +1886,8 @@ struct mlx5_ifc_cmd_hca_cap_bits { u8 reserved_at_5a0[0x10]; u8 enhanced_cqe_compression[0x1]; - u8 reserved_at_5b1[0x2]; + u8 reserved_at_5b1[0x1]; + u8 crossing_vhca_mkey[0x1]; u8 log_max_dek[0x5]; u8 reserved_at_5b8[0x4]; u8 mini_cqe_resp_stride_index[0x1]; @@ -1954,7 +1956,9 @@ struct mlx5_ifc_cmd_hca_cap_bits { u8 dynamic_msix_table_size[0xc]; u8 reserved_at_740[0xc]; u8 min_dynamic_vf_msix_table_size[0x4]; - u8 reserved_at_750[0x4]; + u8 reserved_at_750[0x2]; + u8 data_direct[0x1]; + u8 reserved_at_753[0x1]; u8 max_dynamic_vf_msix_table_size[0xc]; u8 reserved_at_760[0x3]; @@ -1982,7 +1986,9 @@ struct mlx5_ifc_cmd_hca_cap_2_bits { u8 reserved_at_0[0x80]; u8 migratable[0x1]; - u8 reserved_at_81[0x1f]; + u8 reserved_at_81[0x11]; + u8 query_vuid[0x1]; + u8 reserved_at_93[0xd]; u8 max_reformat_insert_size[0x8]; u8 max_reformat_insert_offset[0x8]; @@ -4154,6 +4160,7 @@ enum { MLX5_MKC_ACCESS_MODE_KSM = 0x3, MLX5_MKC_ACCESS_MODE_SW_ICM = 0x4, MLX5_MKC_ACCESS_MODE_MEMIC = 0x5, + MLX5_MKC_ACCESS_MODE_CROSSING = 0x6, }; struct mlx5_ifc_mkc_bits { @@ -4196,7 +4203,10 @@ struct mlx5_ifc_mkc_bits { u8 bsf_octword_size[0x20]; - u8 reserved_at_120[0x80]; + u8 reserved_at_120[0x60]; + + u8 crossing_target_vhca_id[0x10]; + u8 reserved_at_190[0x10]; u8 translations_octword_size[0x20]; @@ -5124,6 +5134,36 @@ struct mlx5_ifc_query_vport_state_out_bits { u8 state[0x4]; }; +struct mlx5_ifc_array1024_auto_bits { + u8 array1024_auto[32][0x20]; +}; + +struct mlx5_ifc_query_vuid_in_bits { + u8 opcode[0x10]; + u8 uid[0x10]; + + u8 reserved_at_20[0x40]; + + u8 query_vfs_vuid[0x1]; + u8 data_direct[0x1]; + u8 reserved_at_62[0xe]; + u8 vhca_id[0x10]; +}; + +struct mlx5_ifc_query_vuid_out_bits { + u8 status[0x8]; + u8 reserved_at_8[0x18]; + + u8 syndrome[0x20]; + + u8 reserved_at_40[0x1a0]; + + u8 reserved_at_1e0[0x10]; + u8 num_of_entries[0x10]; + + struct mlx5_ifc_array1024_auto_bits vuid[]; +}; + enum { MLX5_VPORT_STATE_OP_MOD_VNIC_VPORT = 0x0, MLX5_VPORT_STATE_OP_MOD_ESW_VPORT = 0x1, @@ -8989,7 +9029,8 @@ struct mlx5_ifc_create_mkey_in_bits { u8 pg_access[0x1]; u8 mkey_umem_valid[0x1]; - u8 reserved_at_62[0x1e]; + u8 data_direct[0x1]; + u8 reserved_at_63[0x1d]; struct mlx5_ifc_mkc_bits memory_key_mkey_entry; From patchwork Thu Aug 1 12:05:11 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750240 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 06AEDC3DA4A for ; Thu, 1 Aug 2024 12:05:54 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 5844210E931; Thu, 1 Aug 2024 12:05:53 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="gvfiWWaE"; dkim-atps=neutral Received: from dfw.source.kernel.org (dfw.source.kernel.org [139.178.84.217]) by gabe.freedesktop.org (Postfix) with ESMTPS id B73BB10E933 for ; Thu, 1 Aug 2024 12:05:51 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by dfw.source.kernel.org (Postfix) with ESMTP id 2EA3662767; Thu, 1 Aug 2024 12:05:51 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 54545C4AF0A; Thu, 1 Aug 2024 12:05:50 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513950; bh=tmgwM3dgC5/IejCxV/AIPdrvW4LX0UJNSQHZ02gdLW4=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=gvfiWWaEChlbMPz+AfspkLZ61MuYBS7G+wsFCQGbi6P0yjWyI2HNKdfHGs4lmN1/v lGn98TonRKMkPC+XvCJ8ar0fnRGygm60hoAJRGmTD392OLrnmsbfJWxFhXJe8JM2tK xJtCQt44Vu3jzdXZWnrxuQEfj3pmdIFSXmTM62twE2bp7cuUfwQafCSZk60FfPZtkn nBgyt0ey7zigfvljc/cePxmzfcduMf8vexja7jP8SdAnfk1ENZ1fEXyd6QJif6XsG7 LUzIIOws25GfY9NrGqbdiOe3JuvoJFohiJZjNcloecJNpMm75lxItomAJPaq1t5Agt /9Cdrq2lFMS4w== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 2/8] RDMA/mlx5: Introduce the 'data direct' driver Date: Thu, 1 Aug 2024 15:05:11 +0300 Message-ID: X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Introduce the 'data direct' driver for a ConnectX-8 Data Direct device. The 'data direct' driver functions as the affiliated DMA device for one or more capable mlx5_ib devices. This DMA device, as the name suggests, is used exclusively for DMA operations. It can be considered a DMA engine managed by a PF/VF, lacking network capabilities and having minimal overall capabilities. Consequently, the DMA NIC PF will not be exposed to or directly used by software applications. The driver will not have any direct interface or interaction with the firmware (no command interface, no capabilities, etc.). It will operate solely over PCI to enable its DMA functionality. Registration and un-registration of the driver are handled as part of the mlx5_ib initialization and exit processes, as the mlx5_ib devices will effectively be its clients. The driver will serve as the DMA device for accessing another PCI device to achieve optimal performance (both on the same NUMA node, P2P access, etc.). Upon probing, it will read its VUID over PCI to handle mlx5_ib device registrations with the same VUID. Upon removal, it will notify its clients to allow them to clean up the resources that were mmaped with its DMA device. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/mlx5/Makefile | 1 + drivers/infiniband/hw/mlx5/data_direct.c | 227 +++++++++++++++++++++++ drivers/infiniband/hw/mlx5/data_direct.h | 23 +++ drivers/infiniband/hw/mlx5/main.c | 24 +++ drivers/infiniband/hw/mlx5/mlx5_ib.h | 6 + 5 files changed, 281 insertions(+) create mode 100644 drivers/infiniband/hw/mlx5/data_direct.c create mode 100644 drivers/infiniband/hw/mlx5/data_direct.h diff --git a/drivers/infiniband/hw/mlx5/Makefile b/drivers/infiniband/hw/mlx5/Makefile index 72a526236c2e..b38961f5058e 100644 --- a/drivers/infiniband/hw/mlx5/Makefile +++ b/drivers/infiniband/hw/mlx5/Makefile @@ -6,6 +6,7 @@ mlx5_ib-y := ah.o \ cong.o \ counters.o \ cq.o \ + data_direct.o \ dm.o \ doorbell.o \ gsi.o \ diff --git a/drivers/infiniband/hw/mlx5/data_direct.c b/drivers/infiniband/hw/mlx5/data_direct.c new file mode 100644 index 000000000000..b9ba84afaae2 --- /dev/null +++ b/drivers/infiniband/hw/mlx5/data_direct.c @@ -0,0 +1,227 @@ +// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB +/* + * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved + */ + +#include "mlx5_ib.h" +#include "data_direct.h" + +static LIST_HEAD(mlx5_data_direct_dev_list); +static LIST_HEAD(mlx5_data_direct_reg_list); + +/* + * This mutex should be held when accessing either of the above lists + */ +static DEFINE_MUTEX(mlx5_data_direct_mutex); + +struct mlx5_data_direct_registration { + struct mlx5_ib_dev *ibdev; + char vuid[MLX5_ST_SZ_BYTES(array1024_auto) + 1]; + struct list_head list; +}; + +static const struct pci_device_id mlx5_data_direct_pci_table[] = { + { PCI_VDEVICE(MELLANOX, 0x2100) }, /* ConnectX-8 Data Direct */ + { 0, } +}; + +static int mlx5_data_direct_vpd_get_vuid(struct mlx5_data_direct_dev *dev) +{ + struct pci_dev *pdev = dev->pdev; + unsigned int vpd_size, kw_len; + u8 *vpd_data; + int start; + int ret; + + vpd_data = pci_vpd_alloc(pdev, &vpd_size); + if (IS_ERR(vpd_data)) { + pci_err(pdev, "Unable to read VPD, err=%ld\n", PTR_ERR(vpd_data)); + return PTR_ERR(vpd_data); + } + + start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size, "VU", &kw_len); + if (start < 0) { + ret = start; + pci_err(pdev, "VU keyword not found, err=%d\n", ret); + goto end; + } + + dev->vuid = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL); + ret = dev->vuid ? 0 : -ENOMEM; + +end: + kfree(vpd_data); + return ret; +} + +static void mlx5_data_direct_shutdown(struct pci_dev *pdev) +{ + pci_disable_device(pdev); +} + +static int mlx5_data_direct_set_dma_caps(struct pci_dev *pdev) +{ + int err; + + err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); + if (err) { + dev_warn(&pdev->dev, + "Warning: couldn't set 64-bit PCI DMA mask, err=%d\n", err); + err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)); + if (err) { + dev_err(&pdev->dev, "Can't set PCI DMA mask, err=%d\n", err); + return err; + } + } + + dma_set_max_seg_size(&pdev->dev, SZ_2G); + return 0; +} + +int mlx5_data_direct_ib_reg(struct mlx5_ib_dev *ibdev, char *vuid) +{ + struct mlx5_data_direct_registration *reg; + struct mlx5_data_direct_dev *dev; + + reg = kzalloc(sizeof(*reg), GFP_KERNEL); + if (!reg) + return -ENOMEM; + + reg->ibdev = ibdev; + strcpy(reg->vuid, vuid); + + mutex_lock(&mlx5_data_direct_mutex); + list_for_each_entry(dev, &mlx5_data_direct_dev_list, list) { + if (strcmp(dev->vuid, vuid) == 0) { + mlx5_ib_data_direct_bind(ibdev, dev); + break; + } + } + + /* Add the registration to its global list, to be used upon bind/unbind + * of its affiliated data direct device + */ + list_add_tail(®->list, &mlx5_data_direct_reg_list); + mutex_unlock(&mlx5_data_direct_mutex); + return 0; +} + +void mlx5_data_direct_ib_unreg(struct mlx5_ib_dev *ibdev) +{ + struct mlx5_data_direct_registration *reg; + + mutex_lock(&mlx5_data_direct_mutex); + list_for_each_entry(reg, &mlx5_data_direct_reg_list, list) { + if (reg->ibdev == ibdev) { + list_del(®->list); + kfree(reg); + goto end; + } + } + + WARN_ON(true); +end: + mutex_unlock(&mlx5_data_direct_mutex); +} + +static void mlx5_data_direct_dev_reg(struct mlx5_data_direct_dev *dev) +{ + struct mlx5_data_direct_registration *reg; + + mutex_lock(&mlx5_data_direct_mutex); + list_for_each_entry(reg, &mlx5_data_direct_reg_list, list) { + if (strcmp(dev->vuid, reg->vuid) == 0) + mlx5_ib_data_direct_bind(reg->ibdev, dev); + } + + /* Add the data direct device to the global list, further IB devices may + * use it later as well + */ + list_add_tail(&dev->list, &mlx5_data_direct_dev_list); + mutex_unlock(&mlx5_data_direct_mutex); +} + +static void mlx5_data_direct_dev_unreg(struct mlx5_data_direct_dev *dev) +{ + struct mlx5_data_direct_registration *reg; + + mutex_lock(&mlx5_data_direct_mutex); + /* Prevent any further affiliations */ + list_del(&dev->list); + list_for_each_entry(reg, &mlx5_data_direct_reg_list, list) { + if (strcmp(dev->vuid, reg->vuid) == 0) + mlx5_ib_data_direct_unbind(reg->ibdev); + } + mutex_unlock(&mlx5_data_direct_mutex); +} + +static int mlx5_data_direct_probe(struct pci_dev *pdev, const struct pci_device_id *id) +{ + struct mlx5_data_direct_dev *dev; + int err; + + dev = kzalloc(sizeof(*dev), GFP_KERNEL); + if (!dev) + return -ENOMEM; + + dev->device = &pdev->dev; + dev->pdev = pdev; + + pci_set_drvdata(dev->pdev, dev); + err = pci_enable_device(pdev); + if (err) { + dev_err(dev->device, "Cannot enable PCI device, err=%d\n", err); + goto err; + } + + pci_set_master(pdev); + err = mlx5_data_direct_set_dma_caps(pdev); + if (err) + goto err_disable; + + if (pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP32) && + pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP64) && + pci_enable_atomic_ops_to_root(pdev, PCI_EXP_DEVCAP2_ATOMIC_COMP128)) + dev_dbg(dev->device, "Enabling pci atomics failed\n"); + + err = mlx5_data_direct_vpd_get_vuid(dev); + if (err) + goto err_disable; + + mlx5_data_direct_dev_reg(dev); + return 0; + +err_disable: + pci_disable_device(pdev); +err: + kfree(dev); + return err; +} + +static void mlx5_data_direct_remove(struct pci_dev *pdev) +{ + struct mlx5_data_direct_dev *dev = pci_get_drvdata(pdev); + + mlx5_data_direct_dev_unreg(dev); + pci_disable_device(pdev); + kfree(dev->vuid); + kfree(dev); +} + +static struct pci_driver mlx5_data_direct_driver = { + .name = KBUILD_MODNAME, + .id_table = mlx5_data_direct_pci_table, + .probe = mlx5_data_direct_probe, + .remove = mlx5_data_direct_remove, + .shutdown = mlx5_data_direct_shutdown, +}; + +int mlx5_data_direct_driver_register(void) +{ + return pci_register_driver(&mlx5_data_direct_driver); +} + +void mlx5_data_direct_driver_unregister(void) +{ + pci_unregister_driver(&mlx5_data_direct_driver); +} diff --git a/drivers/infiniband/hw/mlx5/data_direct.h b/drivers/infiniband/hw/mlx5/data_direct.h new file mode 100644 index 000000000000..2fd2bdbe8f69 --- /dev/null +++ b/drivers/infiniband/hw/mlx5/data_direct.h @@ -0,0 +1,23 @@ +/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ +/* + * Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved + */ + +#ifndef _MLX5_IB_DATA_DIRECT_H +#define _MLX5_IB_DATA_DIRECT_H + +struct mlx5_ib_dev; + +struct mlx5_data_direct_dev { + struct device *device; + struct pci_dev *pdev; + char *vuid; + struct list_head list; +}; + +int mlx5_data_direct_ib_reg(struct mlx5_ib_dev *ibdev, char *vuid); +void mlx5_data_direct_ib_unreg(struct mlx5_ib_dev *ibdev); +int mlx5_data_direct_driver_register(void); +void mlx5_data_direct_driver_unregister(void); + +#endif diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 6048b9ad13bb..de254cf03173 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -48,6 +48,7 @@ #include #include #include "macsec.h" +#include "data_direct.h" #define UVERBS_MODULE_NAME mlx5_ib #include @@ -3866,6 +3867,7 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) dev->ib_dev.num_comp_vectors = mlx5_comp_vectors_max(mdev); mutex_init(&dev->cap_mask_mutex); + mutex_init(&dev->data_direct_lock); INIT_LIST_HEAD(&dev->qp_list); spin_lock_init(&dev->reset_flow_resource_lock); xa_init(&dev->odp_mkeys); @@ -4293,6 +4295,21 @@ static void mlx5_ib_stage_dev_notifier_cleanup(struct mlx5_ib_dev *dev) mlx5_notifier_unregister(dev->mdev, &dev->mdev_events); } +void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev, + struct mlx5_data_direct_dev *dev) +{ + mutex_lock(&ibdev->data_direct_lock); + ibdev->data_direct_dev = dev; + mutex_unlock(&ibdev->data_direct_lock); +} + +void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev) +{ + mutex_lock(&ibdev->data_direct_lock); + ibdev->data_direct_dev = NULL; + mutex_unlock(&ibdev->data_direct_lock); +} + void __mlx5_ib_remove(struct mlx5_ib_dev *dev, const struct mlx5_ib_profile *profile, int stage) @@ -4715,17 +4732,23 @@ static int __init mlx5_ib_init(void) ret = mlx5r_rep_init(); if (ret) goto rep_err; + ret = mlx5_data_direct_driver_register(); + if (ret) + goto dd_err; ret = auxiliary_driver_register(&mlx5r_mp_driver); if (ret) goto mp_err; ret = auxiliary_driver_register(&mlx5r_driver); if (ret) goto drv_err; + return 0; drv_err: auxiliary_driver_unregister(&mlx5r_mp_driver); mp_err: + mlx5_data_direct_driver_unregister(); +dd_err: mlx5r_rep_cleanup(); rep_err: mlx5_ib_qp_event_cleanup(); @@ -4737,6 +4760,7 @@ static int __init mlx5_ib_init(void) static void __exit mlx5_ib_cleanup(void) { + mlx5_data_direct_driver_unregister(); auxiliary_driver_unregister(&mlx5r_driver); auxiliary_driver_unregister(&mlx5r_mp_driver); mlx5r_rep_cleanup(); diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index d5eb1b726675..b0d7d8b9e672 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -1131,6 +1131,9 @@ struct mlx5_macsec { struct mlx5_ib_dev { struct ib_device ib_dev; struct mlx5_core_dev *mdev; + struct mlx5_data_direct_dev *data_direct_dev; + /* protect accessing data_direct_dev */ + struct mutex data_direct_lock; struct notifier_block mdev_events; int num_ports; /* serialize update of capability mask @@ -1425,6 +1428,9 @@ int mlx5_ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm, struct ib_dm_mr_attr *attr, struct uverbs_attr_bundle *attrs); +void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev, + struct mlx5_data_direct_dev *dev); +void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev); #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev); From patchwork Thu Aug 1 12:05:12 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750237 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 9266EC3DA64 for ; Thu, 1 Aug 2024 12:05:41 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 0A5BA10E93D; Thu, 1 Aug 2024 12:05:41 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="i2fIT0Ue"; dkim-atps=neutral Received: from dfw.source.kernel.org (dfw.source.kernel.org [139.178.84.217]) by gabe.freedesktop.org (Postfix) with ESMTPS id 6A6F010E93B for ; Thu, 1 Aug 2024 12:05:39 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by dfw.source.kernel.org (Postfix) with ESMTP id BBE95626EE; Thu, 1 Aug 2024 12:05:38 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 0DB47C4AF12; Thu, 1 Aug 2024 12:05:37 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513938; bh=4+zvJheZBxArcPGen/lVnVGy2SZe0eVmEnWE+k/DFek=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=i2fIT0Uerago63KAdzKvUmR44i3jX30NFQbUny4OTnawb9JzIBvQSX1O7bUGQYZhk ntug4roJPlh8gCt1c51H/T05ON9CNk7eThfdR/MihasyvwAGRAprYnbO2NvcHiSj7S jzP9pVpto4rkBnkA5Bv5HYSpETclvr8acR0xqAEMoPjzb0m/FpRMyYEF7UVa2K8yys y2vfoAs4k2FDRP4a2YtSi6l2BSIYBM0G9Ogam+4ao00IFS1f25yDRQ1Ey2aFvBfGvc AO4sy+S2mW/5F8IRiIt9g1CYEcKRqDj5g1r76CeuK7jt/eFFfLAU6Jl3dQl8733aLP xA3X9hcu+tsvg== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 3/8] RDMA/mlx5: Add the initialization flow to utilize the 'data direct' device Date: Thu, 1 Aug 2024 15:05:12 +0300 Message-ID: X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Add the NET device initialization flow to utilize the 'data direct' device. When a NET mlx5_ib device is capable of 'data direct', the following sequence of actions will occur: - Find its affiliated 'data direct' VUID via a firmware command. - Create its own private PD and 'data direct' mkey. - Register to be notified when its 'data direct' driver is probed or removed. The DMA device of the affiliated 'data direct' device, including the private PD and the 'data direct' mkey, will be used later during MR registrations that request the data direct functionality. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/mlx5/cmd.c | 21 +++++++ drivers/infiniband/hw/mlx5/cmd.h | 2 + drivers/infiniband/hw/mlx5/main.c | 90 ++++++++++++++++++++++++++++ drivers/infiniband/hw/mlx5/mlx5_ib.h | 6 ++ 4 files changed, 119 insertions(+) diff --git a/drivers/infiniband/hw/mlx5/cmd.c b/drivers/infiniband/hw/mlx5/cmd.c index 895b62cc528d..7c08e3008927 100644 --- a/drivers/infiniband/hw/mlx5/cmd.c +++ b/drivers/infiniband/hw/mlx5/cmd.c @@ -245,3 +245,24 @@ int mlx5_cmd_uar_dealloc(struct mlx5_core_dev *dev, u32 uarn, u16 uid) MLX5_SET(dealloc_uar_in, in, uid, uid); return mlx5_cmd_exec_in(dev, dealloc_uar, in); } + +int mlx5_cmd_query_vuid(struct mlx5_core_dev *dev, bool data_direct, + char *out_vuid) +{ + u8 out[MLX5_ST_SZ_BYTES(query_vuid_out) + + MLX5_ST_SZ_BYTES(array1024_auto)] = {}; + u8 in[MLX5_ST_SZ_BYTES(query_vuid_in)] = {}; + char *vuid; + int err; + + MLX5_SET(query_vuid_in, in, opcode, MLX5_CMD_OPCODE_QUERY_VUID); + MLX5_SET(query_vuid_in, in, vhca_id, MLX5_CAP_GEN(dev, vhca_id)); + MLX5_SET(query_vuid_in, in, data_direct, data_direct); + err = mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out)); + if (err) + return err; + + vuid = MLX5_ADDR_OF(query_vuid_out, out, vuid); + memcpy(out_vuid, vuid, MLX5_ST_SZ_BYTES(array1024_auto)); + return 0; +} diff --git a/drivers/infiniband/hw/mlx5/cmd.h b/drivers/infiniband/hw/mlx5/cmd.h index e5cd31270443..e6c88b6ebd0d 100644 --- a/drivers/infiniband/hw/mlx5/cmd.h +++ b/drivers/infiniband/hw/mlx5/cmd.h @@ -58,4 +58,6 @@ int mlx5_cmd_mad_ifc(struct mlx5_ib_dev *dev, const void *inb, void *outb, u16 opmod, u8 port); int mlx5_cmd_uar_alloc(struct mlx5_core_dev *dev, u32 *uarn, u16 uid); int mlx5_cmd_uar_dealloc(struct mlx5_core_dev *dev, u32 uarn, u16 uid); +int mlx5_cmd_query_vuid(struct mlx5_core_dev *dev, bool data_direct, + char *out_vuid); #endif /* MLX5_IB_CMD_H */ diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index de254cf03173..fc0562f07249 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -3025,6 +3025,59 @@ static void mlx5_ib_dev_res_cleanup(struct mlx5_ib_dev *dev) mutex_destroy(&devr->srq_lock); } +static int +mlx5_ib_create_data_direct_resources(struct mlx5_ib_dev *dev) +{ + int inlen = MLX5_ST_SZ_BYTES(create_mkey_in); + struct mlx5_core_dev *mdev = dev->mdev; + void *mkc; + u32 mkey; + u32 pdn; + u32 *in; + int err; + + err = mlx5_core_alloc_pd(mdev, &pdn); + if (err) + return err; + + in = kvzalloc(inlen, GFP_KERNEL); + if (!in) { + err = -ENOMEM; + goto err; + } + + MLX5_SET(create_mkey_in, in, data_direct, 1); + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_PA); + MLX5_SET(mkc, mkc, lw, 1); + MLX5_SET(mkc, mkc, lr, 1); + MLX5_SET(mkc, mkc, rw, 1); + MLX5_SET(mkc, mkc, rr, 1); + MLX5_SET(mkc, mkc, a, 1); + MLX5_SET(mkc, mkc, pd, pdn); + MLX5_SET(mkc, mkc, length64, 1); + MLX5_SET(mkc, mkc, qpn, 0xffffff); + err = mlx5_core_create_mkey(mdev, &mkey, in, inlen); + kvfree(in); + if (err) + goto err; + + dev->ddr.mkey = mkey; + dev->ddr.pdn = pdn; + return 0; + +err: + mlx5_core_dealloc_pd(mdev, pdn); + return err; +} + +static void +mlx5_ib_free_data_direct_resources(struct mlx5_ib_dev *dev) +{ + mlx5_core_destroy_mkey(dev->mdev, dev->ddr.mkey); + mlx5_core_dealloc_pd(dev->mdev, dev->ddr.pdn); +} + static u32 get_core_cap_flags(struct ib_device *ibdev, struct mlx5_hca_vport_context *rep) { @@ -3421,6 +3474,38 @@ static bool mlx5_ib_bind_slave_port(struct mlx5_ib_dev *ibdev, return false; } +static int mlx5_ib_data_direct_init(struct mlx5_ib_dev *dev) +{ + char vuid[MLX5_ST_SZ_BYTES(array1024_auto) + 1] = {}; + int ret; + + if (!MLX5_CAP_GEN(dev->mdev, data_direct)) + return 0; + + ret = mlx5_cmd_query_vuid(dev->mdev, true, vuid); + if (ret) + return ret; + + ret = mlx5_ib_create_data_direct_resources(dev); + if (ret) + return ret; + + ret = mlx5_data_direct_ib_reg(dev, vuid); + if (ret) + mlx5_ib_free_data_direct_resources(dev); + + return ret; +} + +static void mlx5_ib_data_direct_cleanup(struct mlx5_ib_dev *dev) +{ + if (!MLX5_CAP_GEN(dev->mdev, data_direct)) + return; + + mlx5_data_direct_ib_unreg(dev); + mlx5_ib_free_data_direct_resources(dev); +} + static int mlx5_ib_init_multiport_master(struct mlx5_ib_dev *dev) { u32 port_num = mlx5_core_native_port_num(dev->mdev) - 1; @@ -3814,6 +3899,7 @@ static const struct uapi_definition mlx5_ib_defs[] = { static void mlx5_ib_stage_init_cleanup(struct mlx5_ib_dev *dev) { + mlx5_ib_data_direct_cleanup(dev); mlx5_ib_cleanup_multiport_master(dev); WARN_ON(!xa_empty(&dev->odp_mkeys)); mutex_destroy(&dev->cap_mask_mutex); @@ -3876,6 +3962,10 @@ static int mlx5_ib_stage_init_init(struct mlx5_ib_dev *dev) spin_lock_init(&dev->dm.lock); dev->dm.dev = mdev; + err = mlx5_ib_data_direct_init(dev); + if (err) + goto err_mp; + return 0; err_mp: mlx5_ib_cleanup_multiport_master(dev); diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index b0d7d8b9e672..b2ebea173547 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -835,6 +835,11 @@ struct mlx5_ib_port_resources { struct work_struct pkey_change_work; }; +struct mlx5_data_direct_resources { + u32 pdn; + u32 mkey; +}; + struct mlx5_ib_resources { struct ib_cq *c0; struct mutex cq_lock; @@ -1188,6 +1193,7 @@ struct mlx5_ib_dev { u16 pkey_table_len; u8 lag_ports; struct mlx5_special_mkeys mkeys; + struct mlx5_data_direct_resources ddr; #ifdef CONFIG_MLX5_MACSEC struct mlx5_macsec macsec; From patchwork Thu Aug 1 12:05:13 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750238 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id DA091C3DA64 for ; Thu, 1 Aug 2024 12:05:47 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 4696810E942; Thu, 1 Aug 2024 12:05:47 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="QUovD0uI"; dkim-atps=neutral Received: from sin.source.kernel.org (sin.source.kernel.org [145.40.73.55]) by gabe.freedesktop.org (Postfix) with ESMTPS id 6C09C10E92E for ; Thu, 1 Aug 2024 12:05:45 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by sin.source.kernel.org (Postfix) with ESMTP id 68697CE18E6; Thu, 1 Aug 2024 12:05:43 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 37A25C4AF09; Thu, 1 Aug 2024 12:05:41 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513942; bh=wcxbwl6CKHuMSz23l9nyG/Jwc9VoDjJQ+eyO66Z4CGQ=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=QUovD0uI+exPoZj7LG0ZbfynjF9a9xLRNtYqhlWc+52nyTGMP3++fgWGzw7DYDNXU K7DvKNcALnGoaxyQES1Lqh+VDdXy6uspMKr9EhK6ZJMGCD2kPXQSnSf2Om/lC2vFDC Re7eo7FDpVG1zwtChmP759+aiaNno/k9bvE0ot2fwqTVkB0bTjwzqxrTuNZ970qzFp UZRr54RLQH33lHhQCaod99Ki1F+u4KZ7X85KwaN0GxZ2DGRrifk1VhdTg7frLDT3Px 68JWSYEbNztemCJ0qqB0uqobpg1HYDbAgqUzad9DkI4UFW1bWXRPMam8ePei44HcSN X91b1LXqbcf7A== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 4/8] RDMA/umem: Add support for creating pinned DMABUF umem with a given dma device Date: Thu, 1 Aug 2024 15:05:13 +0300 Message-ID: <038aad36a43797e5591b20ba81051fc5758124f9.1722512548.git.leon@kernel.org> X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Add support for creating pinned DMABUF umem with a specified DMA device instead of the DMA device of the given IB device. This API will be utilized in the upcoming patches of the series when multiple path DMAs are implemented. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/core/umem_dmabuf.c | 45 ++++++++++++++++++++------- include/rdma/ib_umem.h | 15 +++++++++ 2 files changed, 49 insertions(+), 11 deletions(-) diff --git a/drivers/infiniband/core/umem_dmabuf.c b/drivers/infiniband/core/umem_dmabuf.c index 39357dc2d229..726a09786547 100644 --- a/drivers/infiniband/core/umem_dmabuf.c +++ b/drivers/infiniband/core/umem_dmabuf.c @@ -110,10 +110,12 @@ void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf) } EXPORT_SYMBOL(ib_umem_dmabuf_unmap_pages); -struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device, - unsigned long offset, size_t size, - int fd, int access, - const struct dma_buf_attach_ops *ops) +static struct ib_umem_dmabuf * +ib_umem_dmabuf_get_with_dma_device(struct ib_device *device, + struct device *dma_device, + unsigned long offset, size_t size, + int fd, int access, + const struct dma_buf_attach_ops *ops) { struct dma_buf *dmabuf; struct ib_umem_dmabuf *umem_dmabuf; @@ -152,7 +154,7 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device, umem_dmabuf->attach = dma_buf_dynamic_attach( dmabuf, - device->dma_device, + dma_device, ops, umem_dmabuf); if (IS_ERR(umem_dmabuf->attach)) { @@ -168,6 +170,15 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device, dma_buf_put(dmabuf); return ret; } + +struct ib_umem_dmabuf *ib_umem_dmabuf_get(struct ib_device *device, + unsigned long offset, size_t size, + int fd, int access, + const struct dma_buf_attach_ops *ops) +{ + return ib_umem_dmabuf_get_with_dma_device(device, device->dma_device, + offset, size, fd, access, ops); +} EXPORT_SYMBOL(ib_umem_dmabuf_get); static void @@ -184,16 +195,18 @@ static struct dma_buf_attach_ops ib_umem_dmabuf_attach_pinned_ops = { .move_notify = ib_umem_dmabuf_unsupported_move_notify, }; -struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device, - unsigned long offset, - size_t size, int fd, - int access) +struct ib_umem_dmabuf * +ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device, + struct device *dma_device, + unsigned long offset, size_t size, + int fd, int access) { struct ib_umem_dmabuf *umem_dmabuf; int err; - umem_dmabuf = ib_umem_dmabuf_get(device, offset, size, fd, access, - &ib_umem_dmabuf_attach_pinned_ops); + umem_dmabuf = ib_umem_dmabuf_get_with_dma_device(device, dma_device, offset, + size, fd, access, + &ib_umem_dmabuf_attach_pinned_ops); if (IS_ERR(umem_dmabuf)) return umem_dmabuf; @@ -217,6 +230,16 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device, ib_umem_release(&umem_dmabuf->umem); return ERR_PTR(err); } +EXPORT_SYMBOL(ib_umem_dmabuf_get_pinned_with_dma_device); + +struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device, + unsigned long offset, + size_t size, int fd, + int access) +{ + return ib_umem_dmabuf_get_pinned_with_dma_device(device, device->dma_device, + offset, size, fd, access); +} EXPORT_SYMBOL(ib_umem_dmabuf_get_pinned); void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h index 565a85044541..de05268ed632 100644 --- a/include/rdma/ib_umem.h +++ b/include/rdma/ib_umem.h @@ -150,6 +150,11 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device, unsigned long offset, size_t size, int fd, int access); +struct ib_umem_dmabuf * +ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device, + struct device *dma_device, + unsigned long offset, size_t size, + int fd, int access); int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf); void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf); void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf); @@ -196,6 +201,16 @@ ib_umem_dmabuf_get_pinned(struct ib_device *device, unsigned long offset, { return ERR_PTR(-EOPNOTSUPP); } + +static inline struct ib_umem_dmabuf * +ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device, + struct device *dma_device, + unsigned long offset, size_t size, + int fd, int access) +{ + return ERR_PTR(-EOPNOTSUPP); +} + static inline int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) { return -EOPNOTSUPP; From patchwork Thu Aug 1 12:05:14 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750239 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id C18E2C3DA7F for ; Thu, 1 Aug 2024 12:05:50 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 26AF710E93A; Thu, 1 Aug 2024 12:05:50 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="cwCLvLdx"; dkim-atps=neutral Received: from sin.source.kernel.org (sin.source.kernel.org [145.40.73.55]) by gabe.freedesktop.org (Postfix) with ESMTPS id 6608610E937 for ; Thu, 1 Aug 2024 12:05:49 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by sin.source.kernel.org (Postfix) with ESMTP id 872C6CE193B; Thu, 1 Aug 2024 12:05:47 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 326DEC32786; Thu, 1 Aug 2024 12:05:45 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513946; bh=Lrwpx2uyOMMd30miKkZzGs29UF0PiPnGLNhz0BxCTdY=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=cwCLvLdxcKpwb7cQ70CLYC6NFRTEF+QAC/gzwzp4xw5PgesySI7OoWnQUVVb+R1rK ZwcCI0ocdUczf4Fmj8OpGqIbdDJ5saXfxCLVfeNWkYZO9jrSrZ/z2YdppZrmkGWaM9 DX8eT8AXuBJjqIfTEbGqYVl9QT17vPHC1SQO0Qf7ajeF7javo7gLVsqFeD3w+nbUKW x+kFkd6LxG0b7WgGrlsTT0lMsgTQmFZlHVgIisqy3VXYMS7yMBclo2ZnJkfR9JgfzN q4EEKDj9qEEDtL6IYqmuDOdnpBcdIBW8Ax0XJzY9cCNxqbqrwuEsJkVbbGySuz5KRy /B1kEFuJPW2fQ== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 5/8] RDMA/umem: Introduce an option to revoke DMABUF umem Date: Thu, 1 Aug 2024 15:05:14 +0300 Message-ID: X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Introduce an option to revoke DMABUF umem. This option will retain the umem allocation while revoking its DMA mapping. Furthermore, any subsequent attempts to map the pages should fail once the umem has been revoked. This functionality will be utilized in the upcoming patches in the series, where we aim to delay umem deallocation until the mkey deregistration. However, we must unmap its pages immediately. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/core/umem_dmabuf.c | 21 +++++++++++++++++++-- include/rdma/ib_umem.h | 3 +++ 2 files changed, 22 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/core/umem_dmabuf.c b/drivers/infiniband/core/umem_dmabuf.c index 726a09786547..9fcd37761264 100644 --- a/drivers/infiniband/core/umem_dmabuf.c +++ b/drivers/infiniband/core/umem_dmabuf.c @@ -23,6 +23,9 @@ int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) dma_resv_assert_held(umem_dmabuf->attach->dmabuf->resv); + if (umem_dmabuf->revoked) + return -EINVAL; + if (umem_dmabuf->sgt) goto wait_fence; @@ -242,15 +245,29 @@ struct ib_umem_dmabuf *ib_umem_dmabuf_get_pinned(struct ib_device *device, } EXPORT_SYMBOL(ib_umem_dmabuf_get_pinned); -void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) +void ib_umem_dmabuf_revoke(struct ib_umem_dmabuf *umem_dmabuf) { struct dma_buf *dmabuf = umem_dmabuf->attach->dmabuf; dma_resv_lock(dmabuf->resv, NULL); + if (umem_dmabuf->revoked) + goto end; ib_umem_dmabuf_unmap_pages(umem_dmabuf); - if (umem_dmabuf->pinned) + if (umem_dmabuf->pinned) { dma_buf_unpin(umem_dmabuf->attach); + umem_dmabuf->pinned = 0; + } + umem_dmabuf->revoked = 1; +end: dma_resv_unlock(dmabuf->resv); +} +EXPORT_SYMBOL(ib_umem_dmabuf_revoke); + +void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) +{ + struct dma_buf *dmabuf = umem_dmabuf->attach->dmabuf; + + ib_umem_dmabuf_revoke(umem_dmabuf); dma_buf_detach(dmabuf, umem_dmabuf->attach); dma_buf_put(dmabuf); diff --git a/include/rdma/ib_umem.h b/include/rdma/ib_umem.h index de05268ed632..7dc7b1cc71b5 100644 --- a/include/rdma/ib_umem.h +++ b/include/rdma/ib_umem.h @@ -38,6 +38,7 @@ struct ib_umem_dmabuf { unsigned long last_sg_trim; void *private; u8 pinned : 1; + u8 revoked : 1; }; static inline struct ib_umem_dmabuf *to_ib_umem_dmabuf(struct ib_umem *umem) @@ -158,6 +159,7 @@ ib_umem_dmabuf_get_pinned_with_dma_device(struct ib_device *device, int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf); void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf); void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf); +void ib_umem_dmabuf_revoke(struct ib_umem_dmabuf *umem_dmabuf); #else /* CONFIG_INFINIBAND_USER_MEM */ @@ -217,6 +219,7 @@ static inline int ib_umem_dmabuf_map_pages(struct ib_umem_dmabuf *umem_dmabuf) } static inline void ib_umem_dmabuf_unmap_pages(struct ib_umem_dmabuf *umem_dmabuf) { } static inline void ib_umem_dmabuf_release(struct ib_umem_dmabuf *umem_dmabuf) { } +static inline void ib_umem_dmabuf_revoke(struct ib_umem_dmabuf *umem_dmabuf) {} #endif /* CONFIG_INFINIBAND_USER_MEM */ #endif /* IB_UMEM_H */ From patchwork Thu Aug 1 12:05:15 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750243 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 48140C3DA64 for ; Thu, 1 Aug 2024 12:06:08 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id BBEF610E933; Thu, 1 Aug 2024 12:06:07 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="jV92yMrR"; dkim-atps=neutral Received: from sin.source.kernel.org (sin.source.kernel.org [145.40.73.55]) by gabe.freedesktop.org (Postfix) with ESMTPS id 5CC4A10E933 for ; Thu, 1 Aug 2024 12:06:06 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by sin.source.kernel.org (Postfix) with ESMTP id 95B11CE1939; Thu, 1 Aug 2024 12:06:04 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 36A51C32786; Thu, 1 Aug 2024 12:06:02 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513963; bh=HNVP9vMEpgCGw1ZvFBEVtXJyQ/wcj87OJIJ4nBHuuus=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=jV92yMrRJw6sPhUT18QCfVFYgI1RwBqV25HBO64e2xwh/paz0J66h47xW0JYgU9e3 0P0Dwq8v0yP6qJBSjpW54O5TSH8lQRWNXJkC6vlJrJmmIFmg+3XYGUsjSZR9riLwhE v5flOgNn7Oe3Xej/rJRzQhBF2rPBxVDCGDGbG65VignzqISgMwFaJr17LvQDzoXyUH BdtqqVAbBe2CF2URGnAp0I+xstnDIstNEKiieHnrUpRnJMX6N5zoCWyr46dhsmpuaJ r1LBK4qvwA8X3Y+/B0ZMNUjtTahPtRi747SISrdDBdMh5W9uMVkgZ894EXRr8cXCU4 c5qoeuLTrXpXQ== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 6/8] RDMA: Pass uverbs_attr_bundle as part of '.reg_user_mr_dmabuf' API Date: Thu, 1 Aug 2024 15:05:15 +0300 Message-ID: <9a25b2fc02443f7c36c2d93499ae25252b6afd40.1722512548.git.leon@kernel.org> X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Pass uverbs_attr_bundle as part of '.reg_user_mr_dmabuf' API instead of udata. This enables passing some new ioctl attributes to the drivers, as will be introduced in the next patches for mlx5 driver. Change the involved drivers accordingly. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/core/uverbs_std_types_mr.c | 2 +- drivers/infiniband/hw/bnxt_re/ib_verbs.c | 3 ++- drivers/infiniband/hw/bnxt_re/ib_verbs.h | 2 +- drivers/infiniband/hw/efa/efa.h | 2 +- drivers/infiniband/hw/efa/efa_verbs.c | 4 ++-- drivers/infiniband/hw/irdma/verbs.c | 2 +- drivers/infiniband/hw/mlx5/mlx5_ib.h | 2 +- drivers/infiniband/hw/mlx5/mr.c | 2 +- include/rdma/ib_verbs.h | 2 +- 9 files changed, 11 insertions(+), 10 deletions(-) diff --git a/drivers/infiniband/core/uverbs_std_types_mr.c b/drivers/infiniband/core/uverbs_std_types_mr.c index 03e1db5d1e8c..7ebc7bd3caae 100644 --- a/drivers/infiniband/core/uverbs_std_types_mr.c +++ b/drivers/infiniband/core/uverbs_std_types_mr.c @@ -239,7 +239,7 @@ static int UVERBS_HANDLER(UVERBS_METHOD_REG_DMABUF_MR)( mr = pd->device->ops.reg_user_mr_dmabuf(pd, offset, length, iova, fd, access_flags, - &attrs->driver_udata); + attrs); if (IS_ERR(mr)) return PTR_ERR(mr); diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 7c757351a016..43a68e7de02a 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -4122,7 +4122,8 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, struct ib_mr *bnxt_re_reg_user_mr_dmabuf(struct ib_pd *ib_pd, u64 start, u64 length, u64 virt_addr, int fd, - int mr_access_flags, struct ib_udata *udata) + int mr_access_flags, + struct uverbs_attr_bundle *attrs) { struct bnxt_re_pd *pd = container_of(ib_pd, struct bnxt_re_pd, ib_pd); struct bnxt_re_dev *rdev = pd->rdev; diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.h b/drivers/infiniband/hw/bnxt_re/ib_verbs.h index e98cb1717338..3ddeda312376 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.h +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.h @@ -242,7 +242,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, struct ib_mr *bnxt_re_reg_user_mr_dmabuf(struct ib_pd *ib_pd, u64 start, u64 length, u64 virt_addr, int fd, int mr_access_flags, - struct ib_udata *udata); + struct uverbs_attr_bundle *attrs); int bnxt_re_alloc_ucontext(struct ib_ucontext *ctx, struct ib_udata *udata); void bnxt_re_dealloc_ucontext(struct ib_ucontext *context); int bnxt_re_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); diff --git a/drivers/infiniband/hw/efa/efa.h b/drivers/infiniband/hw/efa/efa.h index e580e087e9da..d7fc9d5eeefd 100644 --- a/drivers/infiniband/hw/efa/efa.h +++ b/drivers/infiniband/hw/efa/efa.h @@ -168,7 +168,7 @@ struct ib_mr *efa_reg_mr(struct ib_pd *ibpd, u64 start, u64 length, struct ib_mr *efa_reg_user_mr_dmabuf(struct ib_pd *ibpd, u64 start, u64 length, u64 virt_addr, int fd, int access_flags, - struct ib_udata *udata); + struct uverbs_attr_bundle *attrs); int efa_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata); int efa_get_port_immutable(struct ib_device *ibdev, u32 port_num, struct ib_port_immutable *immutable); diff --git a/drivers/infiniband/hw/efa/efa_verbs.c b/drivers/infiniband/hw/efa/efa_verbs.c index b1e0a1b7c59d..cc13415ff7e7 100644 --- a/drivers/infiniband/hw/efa/efa_verbs.c +++ b/drivers/infiniband/hw/efa/efa_verbs.c @@ -1684,14 +1684,14 @@ static int efa_register_mr(struct ib_pd *ibpd, struct efa_mr *mr, u64 start, struct ib_mr *efa_reg_user_mr_dmabuf(struct ib_pd *ibpd, u64 start, u64 length, u64 virt_addr, int fd, int access_flags, - struct ib_udata *udata) + struct uverbs_attr_bundle *attrs) { struct efa_dev *dev = to_edev(ibpd->device); struct ib_umem_dmabuf *umem_dmabuf; struct efa_mr *mr; int err; - mr = efa_alloc_mr(ibpd, access_flags, udata); + mr = efa_alloc_mr(ibpd, access_flags, &attrs->driver_udata); if (IS_ERR(mr)) { err = PTR_ERR(mr); goto err_out; diff --git a/drivers/infiniband/hw/irdma/verbs.c b/drivers/infiniband/hw/irdma/verbs.c index fc0ce35da14e..6a107decb704 100644 --- a/drivers/infiniband/hw/irdma/verbs.c +++ b/drivers/infiniband/hw/irdma/verbs.c @@ -3085,7 +3085,7 @@ static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len, static struct ib_mr *irdma_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start, u64 len, u64 virt, int fd, int access, - struct ib_udata *udata) + struct uverbs_attr_bundle *attrs) { struct irdma_device *iwdev = to_iwdev(pd->device); struct ib_umem_dmabuf *umem_dmabuf; diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index b2ebea173547..e915a62da49c 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -1354,7 +1354,7 @@ struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 start, u64 length, u64 virt_addr, int fd, int access_flags, - struct ib_udata *udata); + struct uverbs_attr_bundle *attrs); int mlx5_ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, u32 flags, diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 98bd8eaa393e..1dfd9124bdd1 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -1513,7 +1513,7 @@ static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = { struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, u64 length, u64 virt_addr, int fd, int access_flags, - struct ib_udata *udata) + struct uverbs_attr_bundle *attrs) { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_ib_mr *mr = NULL; diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 6c5712ae559d..a1dcf812d787 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -2476,7 +2476,7 @@ struct ib_device_ops { struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset, u64 length, u64 virt_addr, int fd, int mr_access_flags, - struct ib_udata *udata); + struct uverbs_attr_bundle *attrs); struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length, u64 virt_addr, int mr_access_flags, struct ib_pd *pd, From patchwork Thu Aug 1 12:05:16 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750241 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 51915C3DA4A for ; Thu, 1 Aug 2024 12:05:58 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id CF7F010E92E; Thu, 1 Aug 2024 12:05:57 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="mkFqLKbM"; dkim-atps=neutral Received: from dfw.source.kernel.org (dfw.source.kernel.org [139.178.84.217]) by gabe.freedesktop.org (Postfix) with ESMTPS id 0D49B10E92E for ; Thu, 1 Aug 2024 12:05:56 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by dfw.source.kernel.org (Postfix) with ESMTP id 5BC3262767; Thu, 1 Aug 2024 12:05:55 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 486FBC4AF09; Thu, 1 Aug 2024 12:05:54 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513955; bh=hQGpY8tJlsjoqtAXCFjMYrygBhOgLOEkOqTm/xHQp5c=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=mkFqLKbMW+r5rbXY+wYlFUVrrnN2uG9n5mHJc1Zc7H01GWT6SVteskvcem/ET81Qo slUsrPOOvMnGtEJkPCa/mztJJGjslpcmRp0jo/9Uhvk/Ks2VCKj+/V0/128YpeJySI IBx9n7Sd0WBKLxwXgTFovyhfs9MeY6yXsYiNIAphBZHKNr9h/+o4qsYGqdAdBL3Khm auklwnicpLITndnoXiR8Y5YZ2+AgmQsNqE0WSyVZcleQaBJtNY4Hptv1XtBCppfU6c OyZ6HvBKh1MWHUox+ikwDySpyXJ2VAHSUbmIOkc9xNNwcSyXXkUakKebJzEN8wOch/ SKx1tiatM+q+w== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 7/8] RDMA/mlx5: Add support for DMABUF MR registrations with Data-direct Date: Thu, 1 Aug 2024 15:05:16 +0300 Message-ID: <1f99d8020ed540d9702b9e2252a145a439609ba6.1722512548.git.leon@kernel.org> X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Add support for DMABUF MR registrations with Data-direct device. Upon userspace calling to register a DMABUF MR with the data direct bit set, the below algorithm will be followed. 1) Obtain a pinned DMABUF umem from the IB core using the user input parameters (FD, offset, length) and the DMA PF device. The DMA PF device is needed to allow the IOMMU to enable the DMA PF to access the user buffer over PCI. 2) Create a KSM MKEY by setting its entries according to the user buffer VA to IOVA mapping, with the MKEY being the data direct device-crossed MKEY. This KSM MKEY is umrable and will be used as part of the MR cache. The PD for creating it is the internal device 'data direct' kernel one. 3) Create a crossing MKEY that points to the KSM MKEY using the crossing access mode. 4) Manage the KSM MKEY by adding it to a list of 'data direct' MKEYs managed on the mlx5_ib device. 5) Return the crossing MKEY to the user, created with its supplied PD. Upon DMA PF unbind flow, the driver will revoke the KSM entries. The final deregistration will occur under the hood once the application deregisters its MKEY. Notes: - This version supports only the PINNED UMEM mode, so there is no dependency on ODP. - The IOVA supplied by the application must be system page aligned due to HW translations of KSM. - The crossing MKEY will not be umrable or part of the MR cache, as we cannot change its crossed (i.e. KSM) MKEY over UMR. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/mlx5/main.c | 11 + drivers/infiniband/hw/mlx5/mlx5_ib.h | 8 + drivers/infiniband/hw/mlx5/mr.c | 304 +++++++++++++++++++--- drivers/infiniband/hw/mlx5/odp.c | 5 +- drivers/infiniband/hw/mlx5/umr.c | 93 ++++--- drivers/infiniband/hw/mlx5/umr.h | 1 + include/uapi/rdma/mlx5_user_ioctl_cmds.h | 4 + include/uapi/rdma/mlx5_user_ioctl_verbs.h | 4 + 8 files changed, 358 insertions(+), 72 deletions(-) diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index fc0562f07249..b85ad3c0bfa1 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -3490,6 +3490,7 @@ static int mlx5_ib_data_direct_init(struct mlx5_ib_dev *dev) if (ret) return ret; + INIT_LIST_HEAD(&dev->data_direct_mr_list); ret = mlx5_data_direct_ib_reg(dev, vuid); if (ret) mlx5_ib_free_data_direct_resources(dev); @@ -3882,6 +3883,14 @@ ADD_UVERBS_ATTRIBUTES_SIMPLE( dump_fill_mkey), UA_MANDATORY)); +ADD_UVERBS_ATTRIBUTES_SIMPLE( + mlx5_ib_reg_dmabuf_mr, + UVERBS_OBJECT_MR, + UVERBS_METHOD_REG_DMABUF_MR, + UVERBS_ATTR_FLAGS_IN(MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + enum mlx5_ib_uapi_reg_dmabuf_flags, + UA_OPTIONAL)); + static const struct uapi_definition mlx5_ib_defs[] = { UAPI_DEF_CHAIN(mlx5_ib_devx_defs), UAPI_DEF_CHAIN(mlx5_ib_flow_defs), @@ -3891,6 +3900,7 @@ static const struct uapi_definition mlx5_ib_defs[] = { UAPI_DEF_CHAIN(mlx5_ib_create_cq_defs), UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_DEVICE, &mlx5_ib_query_context), + UAPI_DEF_CHAIN_OBJ_TREE(UVERBS_OBJECT_MR, &mlx5_ib_reg_dmabuf_mr), UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_VAR, UAPI_DEF_IS_OBJ_SUPPORTED(var_is_supported)), UAPI_DEF_CHAIN_OBJ_TREE_NAMED(MLX5_IB_OBJECT_UAR), @@ -4396,6 +4406,7 @@ void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev, void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev) { mutex_lock(&ibdev->data_direct_lock); + mlx5_ib_revoke_data_direct_mrs(ibdev); ibdev->data_direct_dev = NULL; mutex_unlock(&ibdev->data_direct_lock); } diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h index e915a62da49c..be83a4d91a34 100644 --- a/drivers/infiniband/hw/mlx5/mlx5_ib.h +++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h @@ -682,6 +682,8 @@ struct mlx5_ib_mr { struct mlx5_ib_mkey mmkey; struct ib_umem *umem; + /* The mr is data direct related */ + u8 data_direct :1; union { /* Used only by kernel MRs (umem == NULL) */ @@ -719,6 +721,10 @@ struct mlx5_ib_mr { } odp_destroy; struct ib_odp_counters odp_stats; bool is_odp_implicit; + /* The affilated data direct crossed mr */ + struct mlx5_ib_mr *dd_crossed_mr; + struct list_head dd_node; + u8 revoked :1; }; }; }; @@ -1169,6 +1175,7 @@ struct mlx5_ib_dev { /* protect resources needed as part of reset flow */ spinlock_t reset_flow_resource_lock; struct list_head qp_list; + struct list_head data_direct_mr_list; /* Array with num_ports elements */ struct mlx5_ib_port *port; struct mlx5_sq_bfreg bfreg; @@ -1437,6 +1444,7 @@ struct ib_mr *mlx5_ib_reg_dm_mr(struct ib_pd *pd, struct ib_dm *dm, void mlx5_ib_data_direct_bind(struct mlx5_ib_dev *ibdev, struct mlx5_data_direct_dev *dev); void mlx5_ib_data_direct_unbind(struct mlx5_ib_dev *ibdev); +void mlx5_ib_revoke_data_direct_mrs(struct mlx5_ib_dev *dev); #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING int mlx5_ib_odp_init_one(struct mlx5_ib_dev *ibdev); diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c index 1dfd9124bdd1..6829e3688b60 100644 --- a/drivers/infiniband/hw/mlx5/mr.c +++ b/drivers/infiniband/hw/mlx5/mr.c @@ -43,6 +43,7 @@ #include "dm.h" #include "mlx5_ib.h" #include "umr.h" +#include "data_direct.h" enum { MAX_PENDING_REG_MR = 8, @@ -54,7 +55,9 @@ static void create_mkey_callback(int status, struct mlx5_async_work *context); static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags, - unsigned int page_size, bool populate); + unsigned int page_size, bool populate, + int access_mode); +static int __mlx5_ib_dereg_mr(struct ib_mr *ibmr); static void set_mkc_access_pd_addr_fields(void *mkc, int acc, u64 start_addr, struct ib_pd *pd) @@ -1126,12 +1129,10 @@ static unsigned int mlx5_umem_dmabuf_default_pgsz(struct ib_umem *umem, static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, struct ib_umem *umem, u64 iova, - int access_flags) + int access_flags, int access_mode) { - struct mlx5r_cache_rb_key rb_key = { - .access_mode = MLX5_MKC_ACCESS_MODE_MTT, - }; struct mlx5_ib_dev *dev = to_mdev(pd->device); + struct mlx5r_cache_rb_key rb_key = {}; struct mlx5_cache_ent *ent; struct mlx5_ib_mr *mr; unsigned int page_size; @@ -1144,6 +1145,7 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, if (WARN_ON(!page_size)) return ERR_PTR(-EINVAL); + rb_key.access_mode = access_mode; rb_key.ndescs = ib_umem_num_dma_blocks(umem, page_size); rb_key.ats = mlx5_umem_needs_ats(dev, umem, access_flags); rb_key.access_flags = get_unchangeable_access_flags(dev, access_flags); @@ -1154,7 +1156,7 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, */ if (!ent) { mutex_lock(&dev->slow_path_mutex); - mr = reg_create(pd, umem, iova, access_flags, page_size, false); + mr = reg_create(pd, umem, iova, access_flags, page_size, false, access_mode); mutex_unlock(&dev->slow_path_mutex); if (IS_ERR(mr)) return mr; @@ -1175,13 +1177,71 @@ static struct mlx5_ib_mr *alloc_cacheable_mr(struct ib_pd *pd, return mr; } +static struct ib_mr * +reg_create_crossing_vhca_mr(struct ib_pd *pd, u64 iova, u64 length, int access_flags, + u32 crossed_lkey) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + int access_mode = MLX5_MKC_ACCESS_MODE_CROSSING; + struct mlx5_ib_mr *mr; + void *mkc; + int inlen; + u32 *in; + int err; + + if (!MLX5_CAP_GEN(dev->mdev, crossing_vhca_mkey)) + return ERR_PTR(-EOPNOTSUPP); + + mr = kzalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) + return ERR_PTR(-ENOMEM); + + inlen = MLX5_ST_SZ_BYTES(create_mkey_in); + in = kvzalloc(inlen, GFP_KERNEL); + if (!in) { + err = -ENOMEM; + goto err_1; + } + + mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); + MLX5_SET(mkc, mkc, crossing_target_vhca_id, + MLX5_CAP_GEN(dev->mdev, vhca_id)); + MLX5_SET(mkc, mkc, translations_octword_size, crossed_lkey); + MLX5_SET(mkc, mkc, access_mode_1_0, access_mode & 0x3); + MLX5_SET(mkc, mkc, access_mode_4_2, (access_mode >> 2) & 0x7); + + /* for this crossing mkey IOVA should be 0 and len should be IOVA + len */ + set_mkc_access_pd_addr_fields(mkc, access_flags, 0, pd); + MLX5_SET64(mkc, mkc, len, iova + length); + + MLX5_SET(mkc, mkc, free, 0); + MLX5_SET(mkc, mkc, umr_en, 0); + err = mlx5_ib_create_mkey(dev, &mr->mmkey, in, inlen); + if (err) + goto err_2; + + mr->mmkey.type = MLX5_MKEY_MR; + set_mr_fields(dev, mr, length, access_flags, iova); + mr->ibmr.pd = pd; + kvfree(in); + mlx5_ib_dbg(dev, "crossing mkey = 0x%x\n", mr->mmkey.key); + + return &mr->ibmr; +err_2: + kvfree(in); +err_1: + kfree(mr); + return ERR_PTR(err); +} + /* * If ibmr is NULL it will be allocated by reg_create. * Else, the given ibmr will be used. */ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, u64 iova, int access_flags, - unsigned int page_size, bool populate) + unsigned int page_size, bool populate, + int access_mode) { struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_ib_mr *mr; @@ -1190,7 +1250,9 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, int inlen; u32 *in; int err; - bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg)); + bool pg_cap = !!(MLX5_CAP_GEN(dev->mdev, pg)) && + (access_mode == MLX5_MKC_ACCESS_MODE_MTT); + bool ksm_mode = (access_mode == MLX5_MKC_ACCESS_MODE_KSM); if (!page_size) return ERR_PTR(-EINVAL); @@ -1213,7 +1275,7 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, } pas = (__be64 *)MLX5_ADDR_OF(create_mkey_in, in, klm_pas_mtt); if (populate) { - if (WARN_ON(access_flags & IB_ACCESS_ON_DEMAND)) { + if (WARN_ON(access_flags & IB_ACCESS_ON_DEMAND || ksm_mode)) { err = -EINVAL; goto err_2; } @@ -1229,14 +1291,22 @@ static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, struct ib_umem *umem, mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); set_mkc_access_pd_addr_fields(mkc, access_flags, iova, populate ? pd : dev->umrc.pd); + /* In case a data direct flow, overwrite the pdn field by its internal kernel PD */ + if (umem->is_dmabuf && ksm_mode) + MLX5_SET(mkc, mkc, pd, dev->ddr.pdn); + MLX5_SET(mkc, mkc, free, !populate); - MLX5_SET(mkc, mkc, access_mode_1_0, MLX5_MKC_ACCESS_MODE_MTT); + MLX5_SET(mkc, mkc, access_mode_1_0, access_mode); MLX5_SET(mkc, mkc, umr_en, 1); MLX5_SET64(mkc, mkc, len, umem->length); MLX5_SET(mkc, mkc, bsf_octword_size, 0); - MLX5_SET(mkc, mkc, translations_octword_size, - get_octo_len(iova, umem->length, mr->page_shift)); + if (ksm_mode) + MLX5_SET(mkc, mkc, translations_octword_size, + get_octo_len(iova, umem->length, mr->page_shift) * 2); + else + MLX5_SET(mkc, mkc, translations_octword_size, + get_octo_len(iova, umem->length, mr->page_shift)); MLX5_SET(mkc, mkc, log_page_size, mr->page_shift); if (mlx5_umem_needs_ats(dev, umem, access_flags)) MLX5_SET(mkc, mkc, ma_translation_mode, 1); @@ -1373,13 +1443,15 @@ static struct ib_mr *create_real_mr(struct ib_pd *pd, struct ib_umem *umem, xlt_with_umr = mlx5r_umr_can_load_pas(dev, umem->length); if (xlt_with_umr) { - mr = alloc_cacheable_mr(pd, umem, iova, access_flags); + mr = alloc_cacheable_mr(pd, umem, iova, access_flags, + MLX5_MKC_ACCESS_MODE_MTT); } else { unsigned int page_size = mlx5_umem_find_best_pgsz( umem, mkc, log_page_size, 0, iova); mutex_lock(&dev->slow_path_mutex); - mr = reg_create(pd, umem, iova, access_flags, page_size, true); + mr = reg_create(pd, umem, iova, access_flags, page_size, + true, MLX5_MKC_ACCESS_MODE_MTT); mutex_unlock(&dev->slow_path_mutex); } if (IS_ERR(mr)) { @@ -1442,7 +1514,8 @@ static struct ib_mr *create_user_odp_mr(struct ib_pd *pd, u64 start, u64 length, if (IS_ERR(odp)) return ERR_CAST(odp); - mr = alloc_cacheable_mr(pd, &odp->umem, iova, access_flags); + mr = alloc_cacheable_mr(pd, &odp->umem, iova, access_flags, + MLX5_MKC_ACCESS_MODE_MTT); if (IS_ERR(mr)) { ib_umem_release(&odp->umem); return ERR_CAST(mr); @@ -1510,35 +1583,31 @@ static struct dma_buf_attach_ops mlx5_ib_dmabuf_attach_ops = { .move_notify = mlx5_ib_dmabuf_invalidate_cb, }; -struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, - u64 length, u64 virt_addr, - int fd, int access_flags, - struct uverbs_attr_bundle *attrs) +static struct ib_mr * +reg_user_mr_dmabuf(struct ib_pd *pd, struct device *dma_device, + u64 offset, u64 length, u64 virt_addr, + int fd, int access_flags, int access_mode) { + bool pinned_mode = (access_mode == MLX5_MKC_ACCESS_MODE_KSM); struct mlx5_ib_dev *dev = to_mdev(pd->device); struct mlx5_ib_mr *mr = NULL; struct ib_umem_dmabuf *umem_dmabuf; int err; - if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM) || - !IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) - return ERR_PTR(-EOPNOTSUPP); - - mlx5_ib_dbg(dev, - "offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x\n", - offset, virt_addr, length, fd, access_flags); - err = mlx5r_umr_resource_init(dev); if (err) return ERR_PTR(err); - /* dmabuf requires xlt update via umr to work. */ - if (!mlx5r_umr_can_load_pas(dev, length)) - return ERR_PTR(-EINVAL); + if (!pinned_mode) + umem_dmabuf = ib_umem_dmabuf_get(&dev->ib_dev, + offset, length, fd, + access_flags, + &mlx5_ib_dmabuf_attach_ops); + else + umem_dmabuf = ib_umem_dmabuf_get_pinned_with_dma_device(&dev->ib_dev, + dma_device, offset, length, + fd, access_flags); - umem_dmabuf = ib_umem_dmabuf_get(&dev->ib_dev, offset, length, fd, - access_flags, - &mlx5_ib_dmabuf_attach_ops); if (IS_ERR(umem_dmabuf)) { mlx5_ib_dbg(dev, "umem_dmabuf get failed (%ld)\n", PTR_ERR(umem_dmabuf)); @@ -1546,7 +1615,7 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, } mr = alloc_cacheable_mr(pd, &umem_dmabuf->umem, virt_addr, - access_flags); + access_flags, access_mode); if (IS_ERR(mr)) { ib_umem_release(&umem_dmabuf->umem); return ERR_CAST(mr); @@ -1556,9 +1625,13 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, atomic_add(ib_umem_num_pages(mr->umem), &dev->mdev->priv.reg_pages); umem_dmabuf->private = mr; - err = mlx5r_store_odp_mkey(dev, &mr->mmkey); - if (err) - goto err_dereg_mr; + if (!pinned_mode) { + err = mlx5r_store_odp_mkey(dev, &mr->mmkey); + if (err) + goto err_dereg_mr; + } else { + mr->data_direct = true; + } err = mlx5_ib_init_dmabuf_mr(mr); if (err) @@ -1566,10 +1639,101 @@ struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, return &mr->ibmr; err_dereg_mr: - mlx5_ib_dereg_mr(&mr->ibmr, NULL); + __mlx5_ib_dereg_mr(&mr->ibmr); return ERR_PTR(err); } +static struct ib_mr * +reg_user_mr_dmabuf_by_data_direct(struct ib_pd *pd, u64 offset, + u64 length, u64 virt_addr, + int fd, int access_flags) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + struct mlx5_data_direct_dev *data_direct_dev; + struct ib_mr *crossing_mr; + struct ib_mr *crossed_mr; + int ret = 0; + + /* As of HW behaviour the IOVA must be page aligned in KSM mode */ + if (!PAGE_ALIGNED(virt_addr) || (access_flags & IB_ACCESS_ON_DEMAND)) + return ERR_PTR(-EOPNOTSUPP); + + mutex_lock(&dev->data_direct_lock); + data_direct_dev = dev->data_direct_dev; + if (!data_direct_dev) { + ret = -EINVAL; + goto end; + } + + /* The device's 'data direct mkey' was created without RO flags to + * simplify things and allow for a single mkey per device. + * Since RO is not a must, mask it out accordingly. + */ + access_flags &= ~IB_ACCESS_RELAXED_ORDERING; + crossed_mr = reg_user_mr_dmabuf(pd, &data_direct_dev->pdev->dev, + offset, length, virt_addr, fd, + access_flags, MLX5_MKC_ACCESS_MODE_KSM); + if (IS_ERR(crossed_mr)) { + ret = PTR_ERR(crossed_mr); + goto end; + } + + mutex_lock(&dev->slow_path_mutex); + crossing_mr = reg_create_crossing_vhca_mr(pd, virt_addr, length, access_flags, + crossed_mr->lkey); + mutex_unlock(&dev->slow_path_mutex); + if (IS_ERR(crossing_mr)) { + __mlx5_ib_dereg_mr(crossed_mr); + ret = PTR_ERR(crossing_mr); + goto end; + } + + list_add_tail(&to_mmr(crossed_mr)->dd_node, &dev->data_direct_mr_list); + to_mmr(crossing_mr)->dd_crossed_mr = to_mmr(crossed_mr); + to_mmr(crossing_mr)->data_direct = true; +end: + mutex_unlock(&dev->data_direct_lock); + return ret ? ERR_PTR(ret) : crossing_mr; +} + +struct ib_mr *mlx5_ib_reg_user_mr_dmabuf(struct ib_pd *pd, u64 offset, + u64 length, u64 virt_addr, + int fd, int access_flags, + struct uverbs_attr_bundle *attrs) +{ + struct mlx5_ib_dev *dev = to_mdev(pd->device); + int mlx5_access_flags = 0; + int err; + + if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM) || + !IS_ENABLED(CONFIG_INFINIBAND_ON_DEMAND_PAGING)) + return ERR_PTR(-EOPNOTSUPP); + + if (uverbs_attr_is_valid(attrs, MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS)) { + err = uverbs_get_flags32(&mlx5_access_flags, attrs, + MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS, + MLX5_IB_UAPI_REG_DMABUF_ACCESS_DATA_DIRECT); + if (err) + return ERR_PTR(err); + } + + mlx5_ib_dbg(dev, + "offset 0x%llx, virt_addr 0x%llx, length 0x%llx, fd %d, access_flags 0x%x, mlx5_access_flags 0x%x\n", + offset, virt_addr, length, fd, access_flags, mlx5_access_flags); + + /* dmabuf requires xlt update via umr to work. */ + if (!mlx5r_umr_can_load_pas(dev, length)) + return ERR_PTR(-EINVAL); + + if (mlx5_access_flags & MLX5_IB_UAPI_REG_DMABUF_ACCESS_DATA_DIRECT) + return reg_user_mr_dmabuf_by_data_direct(pd, offset, length, virt_addr, + fd, access_flags); + + return reg_user_mr_dmabuf(pd, pd->device->dma_device, + offset, length, virt_addr, + fd, access_flags, MLX5_MKC_ACCESS_MODE_MTT); +} + /* * True if the change in access flags can be done via UMR, only some access * flags can be updated. @@ -1665,7 +1829,7 @@ struct ib_mr *mlx5_ib_rereg_user_mr(struct ib_mr *ib_mr, int flags, u64 start, struct mlx5_ib_mr *mr = to_mmr(ib_mr); int err; - if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM)) + if (!IS_ENABLED(CONFIG_INFINIBAND_USER_MEM) || mr->data_direct) return ERR_PTR(-EOPNOTSUPP); mlx5_ib_dbg( @@ -1793,7 +1957,7 @@ mlx5_alloc_priv_descs(struct ib_device *device, static void mlx5_free_priv_descs(struct mlx5_ib_mr *mr) { - if (!mr->umem && mr->descs) { + if (!mr->umem && !mr->data_direct && mr->descs) { struct ib_device *device = mr->ibmr.device; int size = mr->max_descs * mr->desc_size; struct mlx5_ib_dev *dev = to_mdev(device); @@ -1847,6 +2011,34 @@ static int cache_ent_find_and_store(struct mlx5_ib_dev *dev, return ret; } +static int mlx5_ib_revoke_data_direct_mr(struct mlx5_ib_mr *mr) +{ + struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device); + struct ib_umem_dmabuf *umem_dmabuf = to_ib_umem_dmabuf(mr->umem); + int err; + + lockdep_assert_held(&dev->data_direct_lock); + mr->revoked = true; + err = mlx5r_umr_revoke_mr(mr); + if (WARN_ON(err)) + return err; + + ib_umem_dmabuf_revoke(umem_dmabuf); + return 0; +} + +void mlx5_ib_revoke_data_direct_mrs(struct mlx5_ib_dev *dev) +{ + struct mlx5_ib_mr *mr, *next; + + lockdep_assert_held(&dev->data_direct_lock); + + list_for_each_entry_safe(mr, next, &dev->data_direct_mr_list, dd_node) { + list_del(&mr->dd_node); + mlx5_ib_revoke_data_direct_mr(mr); + } +} + static int mlx5_revoke_mr(struct mlx5_ib_mr *mr) { struct mlx5_ib_dev *dev = to_mdev(mr->ibmr.device); @@ -1864,7 +2056,7 @@ static int mlx5_revoke_mr(struct mlx5_ib_mr *mr) return destroy_mkey(dev, mr); } -int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) +static int __mlx5_ib_dereg_mr(struct ib_mr *ibmr) { struct mlx5_ib_mr *mr = to_mmr(ibmr); struct mlx5_ib_dev *dev = to_mdev(ibmr->device); @@ -1931,6 +2123,36 @@ int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) return 0; } +static int dereg_crossing_data_direct_mr(struct mlx5_ib_dev *dev, + struct mlx5_ib_mr *mr) +{ + struct mlx5_ib_mr *dd_crossed_mr = mr->dd_crossed_mr; + int ret; + + ret = __mlx5_ib_dereg_mr(&mr->ibmr); + if (ret) + return ret; + + mutex_lock(&dev->data_direct_lock); + if (!dd_crossed_mr->revoked) + list_del(&dd_crossed_mr->dd_node); + + ret = __mlx5_ib_dereg_mr(&dd_crossed_mr->ibmr); + mutex_unlock(&dev->data_direct_lock); + return ret; +} + +int mlx5_ib_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) +{ + struct mlx5_ib_mr *mr = to_mmr(ibmr); + struct mlx5_ib_dev *dev = to_mdev(ibmr->device); + + if (mr->data_direct) + return dereg_crossing_data_direct_mr(dev, mr); + + return __mlx5_ib_dereg_mr(ibmr); +} + static void mlx5_set_umr_free_mkey(struct ib_pd *pd, u32 *in, int ndescs, int access_mode, int page_shift) { diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c index a524181f34df..44a3428ea342 100644 --- a/drivers/infiniband/hw/mlx5/odp.c +++ b/drivers/infiniband/hw/mlx5/odp.c @@ -710,7 +710,10 @@ static int pagefault_dmabuf_mr(struct mlx5_ib_mr *mr, size_t bcnt, ib_umem_dmabuf_unmap_pages(umem_dmabuf); err = -EINVAL; } else { - err = mlx5r_umr_update_mr_pas(mr, xlt_flags); + if (mr->data_direct) + err = mlx5r_umr_update_data_direct_ksm_pas(mr, xlt_flags); + else + err = mlx5r_umr_update_mr_pas(mr, xlt_flags); } dma_resv_unlock(umem_dmabuf->attach->dmabuf->resv); diff --git a/drivers/infiniband/hw/mlx5/umr.c b/drivers/infiniband/hw/mlx5/umr.c index ffc31b01f690..eb74c163fd83 100644 --- a/drivers/infiniband/hw/mlx5/umr.c +++ b/drivers/infiniband/hw/mlx5/umr.c @@ -632,44 +632,47 @@ static void mlx5r_umr_final_update_xlt(struct mlx5_ib_dev *dev, wqe->data_seg.byte_count = cpu_to_be32(sg->length); } -/* - * Send the DMA list to the HW for a normal MR using UMR. - * Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP - * flag may be used. - */ -int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +static int +_mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags, bool dd) { + size_t ent_size = dd ? sizeof(struct mlx5_ksm) : sizeof(struct mlx5_mtt); struct mlx5_ib_dev *dev = mr_to_mdev(mr); struct device *ddev = &dev->mdev->pdev->dev; struct mlx5r_umr_wqe wqe = {}; struct ib_block_iter biter; + struct mlx5_ksm *cur_ksm; struct mlx5_mtt *cur_mtt; size_t orig_sg_length; - struct mlx5_mtt *mtt; size_t final_size; + void *curr_entry; struct ib_sge sg; + void *entry; u64 offset = 0; int err = 0; - if (WARN_ON(mr->umem->is_odp)) - return -EINVAL; - - mtt = mlx5r_umr_create_xlt( - dev, &sg, ib_umem_num_dma_blocks(mr->umem, 1 << mr->page_shift), - sizeof(*mtt), flags); - if (!mtt) + entry = mlx5r_umr_create_xlt(dev, &sg, + ib_umem_num_dma_blocks(mr->umem, 1 << mr->page_shift), + ent_size, flags); + if (!entry) return -ENOMEM; orig_sg_length = sg.length; - mlx5r_umr_set_update_xlt_ctrl_seg(&wqe.ctrl_seg, flags, &sg); mlx5r_umr_set_update_xlt_mkey_seg(dev, &wqe.mkey_seg, mr, mr->page_shift); + if (dd) { + /* Use the data direct internal kernel PD */ + MLX5_SET(mkc, &wqe.mkey_seg, pd, dev->ddr.pdn); + cur_ksm = entry; + } else { + cur_mtt = entry; + } + mlx5r_umr_set_update_xlt_data_seg(&wqe.data_seg, &sg); - cur_mtt = mtt; + curr_entry = entry; rdma_umem_for_each_dma_block(mr->umem, &biter, BIT(mr->page_shift)) { - if (cur_mtt == (void *)mtt + sg.length) { + if (curr_entry == entry + sg.length) { dma_sync_single_for_device(ddev, sg.addr, sg.length, DMA_TO_DEVICE); @@ -681,23 +684,31 @@ int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) DMA_TO_DEVICE); offset += sg.length; mlx5r_umr_update_offset(&wqe.ctrl_seg, offset); - - cur_mtt = mtt; + if (dd) + cur_ksm = entry; + else + cur_mtt = entry; } - cur_mtt->ptag = - cpu_to_be64(rdma_block_iter_dma_address(&biter) | - MLX5_IB_MTT_PRESENT); - - if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP)) - cur_mtt->ptag = 0; - - cur_mtt++; + if (dd) { + cur_ksm->va = cpu_to_be64(rdma_block_iter_dma_address(&biter)); + cur_ksm->key = cpu_to_be32(dev->ddr.mkey); + cur_ksm++; + curr_entry = cur_ksm; + } else { + cur_mtt->ptag = + cpu_to_be64(rdma_block_iter_dma_address(&biter) | + MLX5_IB_MTT_PRESENT); + if (mr->umem->is_dmabuf && (flags & MLX5_IB_UPD_XLT_ZAP)) + cur_mtt->ptag = 0; + cur_mtt++; + curr_entry = cur_mtt; + } } - final_size = (void *)cur_mtt - (void *)mtt; + final_size = curr_entry - entry; sg.length = ALIGN(final_size, MLX5_UMR_FLEX_ALIGNMENT); - memset(cur_mtt, 0, sg.length - final_size); + memset(curr_entry, 0, sg.length - final_size); mlx5r_umr_final_update_xlt(dev, &wqe, mr, &sg, flags); dma_sync_single_for_device(ddev, sg.addr, sg.length, DMA_TO_DEVICE); @@ -705,10 +716,32 @@ int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) err: sg.length = orig_sg_length; - mlx5r_umr_unmap_free_xlt(dev, mtt, &sg); + mlx5r_umr_unmap_free_xlt(dev, entry, &sg); return err; } +int mlx5r_umr_update_data_direct_ksm_pas(struct mlx5_ib_mr *mr, unsigned int flags) +{ + /* No invalidation flow is expected */ + if (WARN_ON(!mr->umem->is_dmabuf) || (flags & MLX5_IB_UPD_XLT_ZAP)) + return -EINVAL; + + return _mlx5r_umr_update_mr_pas(mr, flags, true); +} + +/* + * Send the DMA list to the HW for a normal MR using UMR. + * Dmabuf MR is handled in a similar way, except that the MLX5_IB_UPD_XLT_ZAP + * flag may be used. + */ +int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags) +{ + if (WARN_ON(mr->umem->is_odp)) + return -EINVAL; + + return _mlx5r_umr_update_mr_pas(mr, flags, false); +} + static bool umr_can_use_indirect_mkey(struct mlx5_ib_dev *dev) { return !MLX5_CAP_GEN(dev->mdev, umr_indirect_mkey_disabled); diff --git a/drivers/infiniband/hw/mlx5/umr.h b/drivers/infiniband/hw/mlx5/umr.h index 5f734dc72bef..4a02c9b5aad8 100644 --- a/drivers/infiniband/hw/mlx5/umr.h +++ b/drivers/infiniband/hw/mlx5/umr.h @@ -95,6 +95,7 @@ int mlx5r_umr_revoke_mr(struct mlx5_ib_mr *mr); int mlx5r_umr_rereg_pd_access(struct mlx5_ib_mr *mr, struct ib_pd *pd, int access_flags); int mlx5r_umr_update_mr_pas(struct mlx5_ib_mr *mr, unsigned int flags); +int mlx5r_umr_update_data_direct_ksm_pas(struct mlx5_ib_mr *mr, unsigned int flags); int mlx5r_umr_update_xlt(struct mlx5_ib_mr *mr, u64 idx, int npages, int page_shift, int flags); diff --git a/include/uapi/rdma/mlx5_user_ioctl_cmds.h b/include/uapi/rdma/mlx5_user_ioctl_cmds.h index 5b74d6534899..106276a4cce7 100644 --- a/include/uapi/rdma/mlx5_user_ioctl_cmds.h +++ b/include/uapi/rdma/mlx5_user_ioctl_cmds.h @@ -274,6 +274,10 @@ enum mlx5_ib_create_cq_attrs { MLX5_IB_ATTR_CREATE_CQ_UAR_INDEX = UVERBS_ID_DRIVER_NS_WITH_UHW, }; +enum mlx5_ib_reg_dmabuf_mr_attrs { + MLX5_IB_ATTR_REG_DMABUF_MR_ACCESS_FLAGS = (1U << UVERBS_ID_NS_SHIFT), +}; + #define MLX5_IB_DW_MATCH_PARAM 0xA0 struct mlx5_ib_match_params { diff --git a/include/uapi/rdma/mlx5_user_ioctl_verbs.h b/include/uapi/rdma/mlx5_user_ioctl_verbs.h index 3189c7f08d17..7c233df475e7 100644 --- a/include/uapi/rdma/mlx5_user_ioctl_verbs.h +++ b/include/uapi/rdma/mlx5_user_ioctl_verbs.h @@ -54,6 +54,10 @@ enum mlx5_ib_uapi_flow_action_packet_reformat_type { MLX5_IB_UAPI_FLOW_ACTION_PACKET_REFORMAT_TYPE_L2_TO_L3_TUNNEL = 0x3, }; +enum mlx5_ib_uapi_reg_dmabuf_flags { + MLX5_IB_UAPI_REG_DMABUF_ACCESS_DATA_DIRECT = 1 << 0, +}; + struct mlx5_ib_uapi_devx_async_cmd_hdr { __aligned_u64 wr_id; __u8 out_data[]; From patchwork Thu Aug 1 12:05:17 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leon Romanovsky X-Patchwork-Id: 13750242 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 8F001C3DA64 for ; Thu, 1 Aug 2024 12:06:03 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 104F510E5D1; Thu, 1 Aug 2024 12:06:03 +0000 (UTC) Authentication-Results: gabe.freedesktop.org; dkim=pass (2048-bit key; unprotected) header.d=kernel.org header.i=@kernel.org header.b="F8jiYAD+"; dkim-atps=neutral Received: from sin.source.kernel.org (sin.source.kernel.org [145.40.73.55]) by gabe.freedesktop.org (Postfix) with ESMTPS id AF10A10E933 for ; Thu, 1 Aug 2024 12:06:01 +0000 (UTC) Received: from smtp.kernel.org (transwarp.subspace.kernel.org [100.75.92.58]) by sin.source.kernel.org (Postfix) with ESMTP id E942DCE1939; Thu, 1 Aug 2024 12:05:59 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id B57CAC4AF0A; Thu, 1 Aug 2024 12:05:58 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1722513959; bh=Vz6K0CseR2CvsnyMziB3zHBkOhk2OUVB/Qm9uhyU7F4=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=F8jiYAD+/8dE1YMWeRzW41azT2Hb3buByI2zt4jSSEsUYzeYaKnzbzOdoMwLYX09i bV9wu6XZQIvCUyvVvpms63V5cBYtisC11FpT/G+LU2psIw724ofxKru6pj44P4A4Xi HWLwg3w3bCW6T7ScasgQxzL2f3A3bGihpysPMt4TioDA2gT9g7lSW4wEpopUEDp93m lsZIzOOw04cTwVr8WJ+qkF2SWcnHL0XnfQwXDjOptvnIT3uG140de3bFQT5ll3YjPP vwnB7qBqneAWK6h6KjWlvKHgURnZsh76LCF4KRvAhzRFbPKAHsi0+NCswmZE5ou/jU Xfpc08G1A0tag== From: Leon Romanovsky To: Jason Gunthorpe Cc: Yishai Hadas , =?utf-8?q?Christian_K=C3=B6nig?= , dri-devel@lists.freedesktop.org, linaro-mm-sig@lists.linaro.org, linux-media@vger.kernel.org, linux-rdma@vger.kernel.org, Michael Margolin , Mustafa Ismail , netdev@vger.kernel.org, Saeed Mahameed , Selvin Xavier , Sumit Semwal , Tariq Toukan , Tatyana Nikolova Subject: [PATCH rdma-next 8/8] RDMA/mlx5: Introduce GET_DATA_DIRECT_SYSFS_PATH ioctl Date: Thu, 1 Aug 2024 15:05:17 +0300 Message-ID: <403745463e0ef52adbef681ff09aa6a29a756352.1722512548.git.leon@kernel.org> X-Mailer: git-send-email 2.45.2 In-Reply-To: References: MIME-Version: 1.0 X-BeenThere: dri-devel@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Direct Rendering Infrastructure - Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dri-devel-bounces@lists.freedesktop.org Sender: "dri-devel" From: Yishai Hadas Introduce the 'GET_DATA_DIRECT_SYSFS_PATH' ioctl to return the sysfs path of the affiliated 'data direct' device for a given device. Signed-off-by: Yishai Hadas Signed-off-by: Leon Romanovsky --- drivers/infiniband/hw/mlx5/std_types.c | 55 +++++++++++++++++++++++- include/uapi/rdma/mlx5_user_ioctl_cmds.h | 5 +++ 2 files changed, 59 insertions(+), 1 deletion(-) diff --git a/drivers/infiniband/hw/mlx5/std_types.c b/drivers/infiniband/hw/mlx5/std_types.c index bbfcce3bdc84..ffeb1e1a1538 100644 --- a/drivers/infiniband/hw/mlx5/std_types.c +++ b/drivers/infiniband/hw/mlx5/std_types.c @@ -10,6 +10,7 @@ #include #include #include "mlx5_ib.h" +#include "data_direct.h" #define UVERBS_MODULE_NAME mlx5_ib #include @@ -183,6 +184,50 @@ static int UVERBS_HANDLER(MLX5_IB_METHOD_QUERY_PORT)( sizeof(info)); } +static int UVERBS_HANDLER(MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH)( + struct uverbs_attr_bundle *attrs) +{ + struct mlx5_data_direct_dev *data_direct_dev; + struct mlx5_ib_ucontext *c; + struct mlx5_ib_dev *dev; + int out_len = uverbs_attr_get_len(attrs, + MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH); + u32 dev_path_len; + char *dev_path; + int ret; + + c = to_mucontext(ib_uverbs_get_ucontext(attrs)); + if (IS_ERR(c)) + return PTR_ERR(c); + dev = to_mdev(c->ibucontext.device); + mutex_lock(&dev->data_direct_lock); + data_direct_dev = dev->data_direct_dev; + if (!data_direct_dev) { + ret = -ENODEV; + goto end; + } + + dev_path = kobject_get_path(&data_direct_dev->device->kobj, GFP_KERNEL); + if (!dev_path) { + ret = -ENOMEM; + goto end; + } + + dev_path_len = strlen(dev_path) + 1; + if (dev_path_len > out_len) { + ret = -ENOSPC; + goto end; + } + + ret = uverbs_copy_to(attrs, MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH, dev_path, + dev_path_len); + kfree(dev_path); + +end: + mutex_unlock(&dev->data_direct_lock); + return ret; +} + DECLARE_UVERBS_NAMED_METHOD( MLX5_IB_METHOD_QUERY_PORT, UVERBS_ATTR_PTR_IN(MLX5_IB_ATTR_QUERY_PORT_PORT_NUM, @@ -193,9 +238,17 @@ DECLARE_UVERBS_NAMED_METHOD( reg_c0), UA_MANDATORY)); +DECLARE_UVERBS_NAMED_METHOD( + MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH, + UVERBS_ATTR_PTR_OUT( + MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH, + UVERBS_ATTR_MIN_SIZE(0), + UA_MANDATORY)); + ADD_UVERBS_METHODS(mlx5_ib_device, UVERBS_OBJECT_DEVICE, - &UVERBS_METHOD(MLX5_IB_METHOD_QUERY_PORT)); + &UVERBS_METHOD(MLX5_IB_METHOD_QUERY_PORT), + &UVERBS_METHOD(MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH)); DECLARE_UVERBS_NAMED_METHOD( MLX5_IB_METHOD_PD_QUERY, diff --git a/include/uapi/rdma/mlx5_user_ioctl_cmds.h b/include/uapi/rdma/mlx5_user_ioctl_cmds.h index 106276a4cce7..fd2e4a3a56b3 100644 --- a/include/uapi/rdma/mlx5_user_ioctl_cmds.h +++ b/include/uapi/rdma/mlx5_user_ioctl_cmds.h @@ -348,6 +348,7 @@ enum mlx5_ib_pd_methods { enum mlx5_ib_device_methods { MLX5_IB_METHOD_QUERY_PORT = (1U << UVERBS_ID_NS_SHIFT), + MLX5_IB_METHOD_GET_DATA_DIRECT_SYSFS_PATH, }; enum mlx5_ib_query_port_attrs { @@ -355,4 +356,8 @@ enum mlx5_ib_query_port_attrs { MLX5_IB_ATTR_QUERY_PORT, }; +enum mlx5_ib_get_data_direct_sysfs_path_attrs { + MLX5_IB_ATTR_GET_DATA_DIRECT_SYSFS_PATH = (1U << UVERBS_ID_NS_SHIFT), +}; + #endif