From mboxrd@z Thu Jan 1 00:00:00 1970 From: Krzysztof Halasa Subject: Re: Strange network timeouts w/ 2.6.30.5 Date: Thu, 20 Aug 2009 15:45:21 +0200 Message-ID: References: <985009134.71250769263099.JavaMail.root@mail.holmansrus.com> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-2 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: David Miller , linux-kernel@vger.kernel.org, netdev@vger.kernel.org To: Walt Holman Return-path: In-Reply-To: <985009134.71250769263099.JavaMail.root@mail.holmansrus.com> (Walt Holman's message of "Thu\, 20 Aug 2009 06\:54\:23 -0500 \(CDT\)") Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Walt Holman writes: > dmesg is attached. This box does have >2GB Ram (6GB total). The dmes= g > will show e100 init'd 3 times since the first is the stock modprobe, > 2nd was forced with use_io and the 3rd modprobe was after reverting > the patch. You most probably can't test without swiotlb (RAM has to be limited to 2 GB or so), can you? That would (dis)prove my theory. Alternatively (o= r better), a test on IOMMU-equipped system would do. Since swiotlb is x86-only thing (though other 64-bit archs may have something similar), I think the correct work around is to enable the "for_device" handoff on !X86. Something like maybe: Signed-off-by: Krzysztof Ha=B3asa diff --git a/drivers/net/e100.c b/drivers/net/e100.c index 014dfb6..b610088 100644 --- a/drivers/net/e100.c +++ b/drivers/net/e100.c @@ -1762,9 +1762,12 @@ static int e100_rx_indicate(struct nic *nic, str= uct rx *rx, =20 if (ioread8(&nic->csr->scb.status) & rus_no_res) nic->ru_running =3D RU_SUSPENDED; +#ifndef CONFIG_X86 + /* FIXME interferes with swiotlb. */ pci_dma_sync_single_for_device(nic->pdev, rx->dma_addr, sizeof(struct rfd), PCI_DMA_BIDIRECTIONAL); +#endif return -ENODATA; } =20 --=20 Krzysztof Halasa