From patchwork Wed Dec 12 09:56:22 2018 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Selvin Xavier X-Patchwork-Id: 10725907 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 2C8F891E for ; Wed, 12 Dec 2018 09:56:38 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 1AE0B29A93 for ; Wed, 12 Dec 2018 09:56:38 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id 0EB562AEDE; Wed, 12 Dec 2018 09:56:38 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-8.0 required=2.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,MAILING_LIST_MULTI,RCVD_IN_DNSWL_HI autolearn=ham version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 5E51029A93 for ; Wed, 12 Dec 2018 09:56:37 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1726984AbeLLJ4h (ORCPT ); Wed, 12 Dec 2018 04:56:37 -0500 Received: from mail-pl1-f196.google.com ([209.85.214.196]:36787 "EHLO mail-pl1-f196.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1726519AbeLLJ4g (ORCPT ); Wed, 12 Dec 2018 04:56:36 -0500 Received: by mail-pl1-f196.google.com with SMTP id g9so8370268plo.3 for ; Wed, 12 Dec 2018 01:56:36 -0800 (PST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=broadcom.com; s=google; h=from:to:cc:subject:date:message-id:in-reply-to:references; bh=AA7aQCU++QxbB7MuxlTUfXxtZ+lzYeixiXeJIJjZcsA=; b=F9yCPgBWmjr80X1qL8RWRoZZrer3+1UvcSSVRtcQui8GAcX46XtQJN80uDXgdFsM1S XRrwH1bphXjRMHS+lmbJ//hnCnMuourPTuuVGeP011Q/LOzAcjPzpewIuh5H606g25Wm P27WMaSl4rHFHBa5CP3EAgBlaRAtvvVheh5WA= X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references; bh=AA7aQCU++QxbB7MuxlTUfXxtZ+lzYeixiXeJIJjZcsA=; b=GZGA+qK6N6vmhld8gwkn8VUUb11Z/TRYQI2tAhHtprX5rIiY5u7kywLokixij8KGwa 4Jm4QMZ0FCquymMpFQGxk3f4bm417qSY6Zn5NrJ18WSU3u6i7NUq3zQEgrOPhjllvVD7 M3A0n78BdEq3N6V5/lWNwhfIImQt+3ic562MJKri3TUKgbgPRS7XPbWUOjNMIT37ePxG iQf2vQWIZkqTiBkRPqD9gVBt1ozSzOoagyyDnVjM6muxsjqoqgiMPmJmzK1ShEO6FjPm OS9BfPMqtFQls8U5ExGxypGanCS1K7h7cIDzsBmMIuTP/fFKnDBgD1+eGl0U04brw2i2 0ZOg== X-Gm-Message-State: AA+aEWYx66RPlhQ73495q3h4fu/VgyjJr6xfaG4IKMR7QFIWJRd6OeZ3 G7Tkzh6jTaDWhRpfH6aWpKSaQg== X-Google-Smtp-Source: AFSGD/XWexx9oXvZApX4LOUsTZ4Ayu411xHt6BnUt70ANAbx/Zg4AhtqjjBryBMylmDWpE5j0ZmJvA== X-Received: by 2002:a17:902:2868:: with SMTP id e95mr19368886plb.317.1544608596092; Wed, 12 Dec 2018 01:56:36 -0800 (PST) Received: from dhcp-10-192-206-197.dhcp.broadcom.net ([192.19.234.250]) by smtp.gmail.com with ESMTPSA id g15sm81656840pfj.131.2018.12.12.01.56.34 (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Wed, 12 Dec 2018 01:56:35 -0800 (PST) From: Selvin Xavier To: dledford@redhat.com, jgg@mellanox.com Cc: linux-rdma@vger.kernel.org, Somnath Kotur , Selvin Xavier Subject: [for-next PATCH V2 1/3] RDMA/bnxt_re: Increase memory registration size of a single MR Date: Wed, 12 Dec 2018 01:56:22 -0800 Message-Id: <1544608584-30998-2-git-send-email-selvin.xavier@broadcom.com> X-Mailer: git-send-email 2.5.5 In-Reply-To: <1544608584-30998-1-git-send-email-selvin.xavier@broadcom.com> References: <1544608584-30998-1-git-send-email-selvin.xavier@broadcom.com> Sender: linux-rdma-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-rdma@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP From: Somnath Kotur Currently bnxt_re driver is not utilizing the full limits of our h/w to register single MR. Increase the maximum MR size to 256G. The patch includes the following changes. - Use 'vmalloc' instead of 'kmalloc' while trying to allocate a temporary place holder for all the pointers to user memory pages. - Determine how many logical PDEs are required to store the pointers to user memory pages and allocate contigous pages of sufficient size to store all of them. Signed-off-by: Somnath Kotur Signed-off-by: Selvin Xavier --- drivers/infiniband/hw/bnxt_re/bnxt_re.h | 3 ++ drivers/infiniband/hw/bnxt_re/ib_verbs.c | 24 ++++++-------- drivers/infiniband/hw/bnxt_re/qplib_res.c | 54 ++++++++++++++++++++++++++----- drivers/infiniband/hw/bnxt_re/qplib_res.h | 4 ++- drivers/infiniband/hw/bnxt_re/qplib_sp.c | 8 ++--- 5 files changed, 65 insertions(+), 28 deletions(-) diff --git a/drivers/infiniband/hw/bnxt_re/bnxt_re.h b/drivers/infiniband/hw/bnxt_re/bnxt_re.h index 31baa893..d2fa2a6b 100644 --- a/drivers/infiniband/hw/bnxt_re/bnxt_re.h +++ b/drivers/infiniband/hw/bnxt_re/bnxt_re.h @@ -39,6 +39,9 @@ #ifndef __BNXT_RE_H__ #define __BNXT_RE_H__ + +#include + #define ROCE_DRV_MODULE_NAME "bnxt_re" #define BNXT_RE_DESC "Broadcom NetXtreme-C/E RoCE Driver" diff --git a/drivers/infiniband/hw/bnxt_re/ib_verbs.c b/drivers/infiniband/hw/bnxt_re/ib_verbs.c index 54fdd4c..9ff6810 100644 --- a/drivers/infiniband/hw/bnxt_re/ib_verbs.c +++ b/drivers/infiniband/hw/bnxt_re/ib_verbs.c @@ -3604,32 +3604,26 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, } mr->qplib_mr.total_size = length; - pbl_tbl = kcalloc(umem_pgs, sizeof(u64 *), GFP_KERNEL); - if (!pbl_tbl) { - rc = -ENOMEM; - goto free_umem; - } - page_shift = umem->page_shift; if (!bnxt_re_page_size_ok(page_shift)) { dev_err(rdev_to_dev(rdev), "umem page size unsupported!"); rc = -EFAULT; - goto fail; + goto free_umem; } - if (!umem->hugetlb && length > BNXT_RE_MAX_MR_SIZE_LOW) { - dev_err(rdev_to_dev(rdev), "Requested MR Sz:%llu Max sup:%llu", - length, (u64)BNXT_RE_MAX_MR_SIZE_LOW); - rc = -EINVAL; - goto fail; - } if (umem->hugetlb && length > BNXT_RE_PAGE_SIZE_2M) { page_shift = BNXT_RE_PAGE_SHIFT_2M; dev_warn(rdev_to_dev(rdev), "umem hugetlb set page_size %x", 1 << page_shift); } + pbl_tbl = vmalloc(umem_pgs * sizeof(u64 *)); + if (!pbl_tbl) { + rc = -EINVAL; + goto free_umem; + } + /* Map umem buf ptrs to the PBL */ umem_pgs = fill_umem_pbl_tbl(umem, pbl_tbl, page_shift); rc = bnxt_qplib_reg_mr(&rdev->qplib_res, &mr->qplib_mr, pbl_tbl, @@ -3639,7 +3633,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, goto fail; } - kfree(pbl_tbl); + vfree(pbl_tbl); mr->ib_mr.lkey = mr->qplib_mr.lkey; mr->ib_mr.rkey = mr->qplib_mr.lkey; @@ -3647,7 +3641,7 @@ struct ib_mr *bnxt_re_reg_user_mr(struct ib_pd *ib_pd, u64 start, u64 length, return &mr->ib_mr; fail: - kfree(pbl_tbl); + vfree(pbl_tbl); free_umem: ib_umem_release(umem); free_mrw: diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.c b/drivers/infiniband/hw/bnxt_re/qplib_res.c index 59eeac5..13fa2d4 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_res.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_res.c @@ -164,7 +164,7 @@ int bnxt_qplib_alloc_init_hwq(struct pci_dev *pdev, struct bnxt_qplib_hwq *hwq, u32 *elements, u32 element_size, u32 aux, u32 pg_size, enum bnxt_qplib_hwq_type hwq_type) { - u32 pages, slots, size, aux_pages = 0, aux_size = 0; + u32 pages, slots, size, aux_pages = 0, aux_size = 0, num_pdes, alloc_sz; dma_addr_t *src_phys_ptr, **dst_virt_ptr; int i, rc; @@ -192,18 +192,41 @@ int bnxt_qplib_alloc_init_hwq(struct pci_dev *pdev, struct bnxt_qplib_hwq *hwq, } /* Alloc the 1st memory block; can be a PDL/PTL/PBL */ - if (sghead && (pages == MAX_PBL_LVL_0_PGS)) + if (sghead && pages == MAX_PBL_LVL_0_PGS) { rc = __alloc_pbl(pdev, &hwq->pbl[PBL_LVL_0], sghead, pages, pg_size); - else - rc = __alloc_pbl(pdev, &hwq->pbl[PBL_LVL_0], NULL, 1, pg_size); + } else { + /* + * Find out how many PDEs it takes to store all the PBLs(ptrs to + * actual user pages). + * We still need to allocate 1 contigous page to store all these + * effectively logical PDEs though for HW access + */ + if (hwq_type == HWQ_TYPE_MR) { + num_pdes = pages >> MAX_PDL_LVL_SHIFT; + if (num_pdes) { + alloc_sz = num_pdes * pg_size; + dev_dbg(&pdev->dev, "num_pdes = %d alloc_sz 0x%x\n", + num_pdes, alloc_sz); + rc = __alloc_pbl(pdev, &hwq->pbl[PBL_LVL_0], + NULL, 1, alloc_sz); + + } else { + rc = __alloc_pbl(pdev, &hwq->pbl[PBL_LVL_0], + NULL, 1, pg_size); + } + } else { + rc = __alloc_pbl(pdev, &hwq->pbl[PBL_LVL_0], NULL, + 1, pg_size); + } + } if (rc) goto fail; hwq->level = PBL_LVL_0; if (pages > MAX_PBL_LVL_0_PGS) { - if (pages > MAX_PBL_LVL_1_PGS) { + if (pages > MAX_PBL_LVL_1_PGS && hwq_type != HWQ_TYPE_MR) { /* 2 levels of indirection */ rc = __alloc_pbl(pdev, &hwq->pbl[PBL_LVL_1], NULL, MAX_PBL_LVL_1_PGS_FOR_LVL_2, pg_size); @@ -255,9 +278,24 @@ int bnxt_qplib_alloc_init_hwq(struct pci_dev *pdev, struct bnxt_qplib_hwq *hwq, dst_virt_ptr = (dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr; src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr; - for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; i++) { - dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] = - src_phys_ptr[i] | flag; + + if (hwq_type == HWQ_TYPE_MR) { + /* + * For MR it is expected that we supply only + * 1 contigous page i.e only 1 entry in the PDL + * that will contain all the PBLs for the user + * supplied memory region + */ + for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; + i++) { + dst_virt_ptr[0][i] = src_phys_ptr[i] | + flag; + } + } else { + for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; + i++) + dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] = + src_phys_ptr[i] | flag; } if (hwq_type == HWQ_TYPE_QUEUE) { /* Find the last pg of the size */ diff --git a/drivers/infiniband/hw/bnxt_re/qplib_res.h b/drivers/infiniband/hw/bnxt_re/qplib_res.h index 2e5c052..a783bfa 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_res.h +++ b/drivers/infiniband/hw/bnxt_re/qplib_res.h @@ -55,12 +55,14 @@ extern const struct bnxt_qplib_gid bnxt_qplib_gid_zero; enum bnxt_qplib_hwq_type { HWQ_TYPE_CTX, HWQ_TYPE_QUEUE, - HWQ_TYPE_L2_CMPL + HWQ_TYPE_L2_CMPL, + HWQ_TYPE_MR }; #define MAX_PBL_LVL_0_PGS 1 #define MAX_PBL_LVL_1_PGS 512 #define MAX_PBL_LVL_1_PGS_SHIFT 9 +#define MAX_PDL_LVL_SHIFT 9 #define MAX_PBL_LVL_1_PGS_FOR_LVL_2 256 #define MAX_PBL_LVL_2_PGS (256 * 512) diff --git a/drivers/infiniband/hw/bnxt_re/qplib_sp.c b/drivers/infiniband/hw/bnxt_re/qplib_sp.c index 5216b5f..566b70a 100644 --- a/drivers/infiniband/hw/bnxt_re/qplib_sp.c +++ b/drivers/infiniband/hw/bnxt_re/qplib_sp.c @@ -668,11 +668,11 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr, pages = pg_ptrs >> MAX_PBL_LVL_1_PGS_SHIFT; if (!pages) pages++; - - if (pages > MAX_PBL_LVL_1_PGS) { + /* Limit max MR size to 256 GB eventhough HW supports more */ + if (pages > MAX_PBL_LVL_2_PGS) { dev_err(&res->pdev->dev, "SP: Reg MR pages requested (0x%x) exceeded max (0x%x)\n", - pages, MAX_PBL_LVL_1_PGS); + pages, MAX_PBL_LVL_2_PGS); return -ENOMEM; } /* Free the hwq if it already exist, must be a rereg */ @@ -684,7 +684,7 @@ int bnxt_qplib_reg_mr(struct bnxt_qplib_res *res, struct bnxt_qplib_mrw *mr, rc = bnxt_qplib_alloc_init_hwq(res->pdev, &mr->hwq, NULL, 0, &mr->hwq.max_elements, PAGE_SIZE, 0, PAGE_SIZE, - HWQ_TYPE_CTX); + HWQ_TYPE_MR); if (rc) { dev_err(&res->pdev->dev, "SP: Reg MR memory allocation failed\n");