From owner-svn-src-head@freebsd.org Wed Oct 19 18:15:45 2016 Return-Path: Delivered-To: svn-src-head@mailman.ysv.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2001:1900:2254:206a::19:1]) by mailman.ysv.freebsd.org (Postfix) with ESMTP id BD269C19C1D; Wed, 19 Oct 2016 18:15:45 +0000 (UTC) (envelope-from kib@FreeBSD.org) Received: from repo.freebsd.org (repo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:0]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (Client did not present a certificate) by mx1.freebsd.org (Postfix) with ESMTPS id 6C623924; Wed, 19 Oct 2016 18:15:45 +0000 (UTC) (envelope-from kib@FreeBSD.org) Received: from repo.freebsd.org ([127.0.1.37]) by repo.freebsd.org (8.15.2/8.15.2) with ESMTP id u9JIFiHt077438; Wed, 19 Oct 2016 18:15:44 GMT (envelope-from kib@FreeBSD.org) Received: (from kib@localhost) by repo.freebsd.org (8.15.2/8.15.2/Submit) id u9JIFi1d077437; Wed, 19 Oct 2016 18:15:44 GMT (envelope-from kib@FreeBSD.org) Message-Id: <201610191815.u9JIFi1d077437@repo.freebsd.org> X-Authentication-Warning: repo.freebsd.org: kib set sender to kib@FreeBSD.org using -f From: Konstantin Belousov Date: Wed, 19 Oct 2016 18:15:44 +0000 (UTC) To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r307649 - head/sys/dev/e1000 X-SVN-Group: head MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit X-BeenThere: svn-src-head@freebsd.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: SVN commit messages for the src tree for head/-current List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 19 Oct 2016 18:15:45 -0000 Author: kib Date: Wed Oct 19 18:15:44 2016 New Revision: 307649 URL: https://svnweb.freebsd.org/changeset/base/307649 Log: Partial workaround for Intel PCI adapters reading past the end of the host-programmed DMA regions. This change seemingly fixes the descriptor fetches, but the packet memory accesses are left problematic. Reviewed by: emaste, erj, sbruno Sponsored by: The FreeBSD Foundation MFC after: 2 weeks Differential revision: https://reviews.freebsd.org/D8282 Modified: head/sys/dev/e1000/if_lem.c Modified: head/sys/dev/e1000/if_lem.c ============================================================================== --- head/sys/dev/e1000/if_lem.c Wed Oct 19 17:42:45 2016 (r307648) +++ head/sys/dev/e1000/if_lem.c Wed Oct 19 18:15:44 2016 (r307649) @@ -543,8 +543,16 @@ lem_attach(device_t dev) */ adapter->hw.mac.report_tx_early = 1; - tsize = roundup2(adapter->num_tx_desc * sizeof(struct e1000_tx_desc), - EM_DBA_ALIGN); + /* + * It seems that the descriptor DMA engine on some PCI cards + * fetches memory past the end of the last descriptor in the + * ring. These reads are problematic when VT-d (DMAR) busdma + * is used. Allocate the scratch space to avoid getting + * faults from DMAR, by requesting scratch memory for one more + * descriptor. + */ + tsize = roundup2((adapter->num_tx_desc + 1) * + sizeof(struct e1000_tx_desc), EM_DBA_ALIGN); /* Allocate Transmit Descriptor ring */ if (lem_dma_malloc(adapter, tsize, &adapter->txdma, BUS_DMA_NOWAIT)) { @@ -555,8 +563,11 @@ lem_attach(device_t dev) adapter->tx_desc_base = (struct e1000_tx_desc *)adapter->txdma.dma_vaddr; - rsize = roundup2(adapter->num_rx_desc * sizeof(struct e1000_rx_desc), - EM_DBA_ALIGN); + /* + * See comment above txdma allocation for rationale behind +1. + */ + rsize = roundup2((adapter->num_rx_desc + 1) * + sizeof(struct e1000_rx_desc), EM_DBA_ALIGN); /* Allocate Receive Descriptor ring */ if (lem_dma_malloc(adapter, rsize, &adapter->rxdma, BUS_DMA_NOWAIT)) {