From mboxrd@z Thu Jan 1 00:00:00 1970 From: Randy Dunlap Subject: Re: Why is AHA152X_CS !64BIT? Date: Tue, 07 Sep 2010 14:41:56 -0700 Message-ID: <4C86B1A4.8030308@oracle.com> References: <4C72D310.4040004@suse.cz> <4C7B924D.6030703@panasas.com> <201009071712.54867.konrad@darnok.org> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 7bit Return-path: Received: from rcsinet10.oracle.com ([148.87.113.121]:19069 "EHLO rcsinet10.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750957Ab0IGVnl (ORCPT ); Tue, 7 Sep 2010 17:43:41 -0400 In-Reply-To: <201009071712.54867.konrad@darnok.org> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Konrad Rzeszutek Wilk Cc: Boaz Harrosh , Jiri Slaby , fischer@linux-buechse.de, linux-scsi@vger.kernel.org On 09/07/10 14:12, Konrad Rzeszutek Wilk wrote: > On Monday 30 August 2010 07:13:17 Boaz Harrosh wrote: >> On 08/23/2010 10:59 PM, Jiri Slaby wrote: >>> Hi, >>> >>> I see that the aha152x driver for pcmcia is marked as unsupported on >>> 64bit. But I also see a patch [1] which removes the restriction based on >>> user's testing in bugzilla [2]. >>> >>> Is there a reason why it would have to be marked as !64BIT? I'm asking >>> because there is an opensuse user with this card who updated to 64-bit >>> distro and lost this driver thereafter. >>> >>> [1] http://kerneltrap.org/mailarchive/linux-scsi/2010/3/6/6832393 >>> [2] https://bugzilla.kernel.org/show_bug.cgi?id=14333 >>> >>> thanks, >> >> If memory serves correctly, it might be that you need more then 4 Gbyte >> of memory installed to exercise the bug, something about IO bouncing >> addresses > 4G. > > If the machine is using SWIOTLB, then the bounce buffer would be activated. By > default if your machine has more than 4GB compiled under x86_64 the SWIOTLB > is turned on - but if you have an Intel/AMD IOMMU it gets turned off. Which > is OK as the Intel/AMD IOMMUs would handle the 4GB restricted devices. So as > long as the driver has pci_dma_mask_set. > > Looking at the git gui blame tool history, the reason that was added was > for 'allow drivers to be built non-modular'. 023ae619 (Robert P. J. Day 2007-03-26 16:06:45 -0400 14) depends on !64BIT That commit just removed the "depends on m" part: - depends on m && !64BIT + depends on !64BIT > So, does this driver build if you make it non-modular? It shouldn't since it still depends on !64BIT. I expect someone thought or had evidence that the driver was not 64-bit clean. Is the bitkeeper kernel repo still visible somewhere? Looks like we would need to look at it for patch history that far back. -- ~Randy *** Remember to use Documentation/SubmitChecklist when testing your code ***