From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tejun Heo Subject: Re: Gigabyte GA-MA69VM-S2 can't do 64bit DMA Date: Fri, 18 Sep 2009 17:09:19 +0900 Message-ID: <4AB3402F.5010609@gmail.com> References: <4AAE8252.8000206@web.de> <4AAFE66D.8050807@gmail.com> <4AB0ADB2.8090009@web.de> <4AB0C2AD.6020503@gmail.com> <4AB2B463.4010408@web.de> <4AB2D7A1.8060909@gmail.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 7bit Return-path: Received: from mail-yx0-f199.google.com ([209.85.210.199]:40590 "EHLO mail-yx0-f199.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751672AbZIRIIl (ORCPT ); Fri, 18 Sep 2009 04:08:41 -0400 Received: by yxe37 with SMTP id 37so668019yxe.33 for ; Fri, 18 Sep 2009 01:08:44 -0700 (PDT) In-Reply-To: Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: "Huang, Shane" Cc: Holger Hasselmann , linux-ide@vger.kernel.org Huang, Shane wrote: > Hi Tejun, > >> -----Original Message----- >> From: Tejun Heo [mailto:htejun@gmail.com] >>> ahci 0000:00:12.0: version 3.0_ >>> ahci 0000:00:12.0: PCI INT A -> GSI 22 (level, low) -> IRQ 22 >>> ahci 0000:00:12.0: Fujitsu-Siemens AMILO Pa 2510: this >> board can't do 64bit >>> DMA, forcing 32bit >>> ahci 0000:00:12.0: controller can't do 64bit DMA, forcing 32bit >>> ahci 0000:00:12.0: AHCI 0001.0100 32 slots 4 ports 3 Gbps >> 0xf impl SATA mode >>> ahci 0000:00:12.0: flags: ncq sntf ilck pm led clo pmp pio slum part >>> It's correct now. >> Cool, patch posted with the id string shortened. > > Just some small questions about this topic: > > What is the actual impact of 64-bit DMA not being available? If IO happens to memory above 4G it will either be bounced (ie. DMA will be performed to a bounce buffer < 4G which is copied from / to the original buffer or more likely iommu will kick in. > How important is it to have this enabled? > > In other words, will the performance be improved significantly > after it is enabled? Do you have any experimental data for it? :-) There probably won't be any noticeable performance regression if iommu is working properly and no, I don't have any concrete numbers. Thanks.. -- tejun