From mboxrd@z Thu Jan 1 00:00:00 1970 From: Konrad Rzeszutek Wilk Subject: Re: sata_sil24: swiotlb buffer is full ? Date: Mon, 31 Oct 2016 15:09:18 -0400 Message-ID: <20161031190918.GE15811@char.us.oracle.com> References: <8b6da1a2-ca9c-e15a-10a8-f3cda092698c@shaw.ca> <20161031161825.GA26364@mtj.duckdns.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from aserp1040.oracle.com ([141.146.126.69]:35758 "EHLO aserp1040.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932404AbcJaTJ0 (ORCPT ); Mon, 31 Oct 2016 15:09:26 -0400 Content-Disposition: inline In-Reply-To: <20161031161825.GA26364@mtj.duckdns.org> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Tejun Heo Cc: Andrew Ryder , linux-ide@vger.kernel.org, linux-kernel@vger.kernel.org On Mon, Oct 31, 2016 at 10:18:25AM -0600, Tejun Heo wrote: > Hello, > > On Sat, Oct 29, 2016 at 11:40:29PM -0400, Andrew Ryder wrote: > > I have some disks attached to a "Silicon Image, Inc. SiI 3124 PCI-X Serial > > ATA Controller" and it repeatedly locks up the system with the message > > whenever there is heavy disk i/o. The system the controller is attached to > > is a via EPIA-M910 board. > > > > sata_sil24: 0000:06:03.0: swiotlb buffer is full: 65536 bytes) > > DMA: Out of SW-IOMMU space for 65536 bytes at device .." > > sata_sil24 0000:06:03.0: swiotlb buffer is full (sz: 65536 bytes .." > > > > For the past week I have been running with two additional boot parameters > > (iommu=allowdac swiotlb=131072) which seem to have solved the issue, but I > > was curious if this is a driver bug or not? Usually it means that the device (sta_sil24) can only handle certain DMA addresses and hence needs the assistance of the bounce buffers (swiotlb). Increasing the number of them is the right way to make it work. I would call this hardware limitation - if you provide the lspci -n -s 06:03.0 one can look in the driver and see where it sets the DMA mask. > > (cc'ing swiotbl maintainer, hi!) > > That looks like iotlb area running out. I don't think there's much to > be done from driver side and we've traditionally been pretty bad at > handling iotlb errors. Konrad, do you have any ideas? > > Thanks. > > -- > tejun