From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Moore, Eric Dean" Subject: Re: [PATCH 2.6 ] MPT Fusion driver 3.01.02 update Date: Wed, 17 Mar 2004 18:38:51 -0700 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <001201c40c89$c4fcef20$221015ac@ericmoore> References: <0E3FA95632D6D047BA649F95DAB60E57040DB25E@exa-atlanta.se.lsil.com> <20040318020948.2f884e89.ak@suse.de> Mime-Version: 1.0 Content-Type: text/plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Return-path: Received: from mail0.lsil.com ([147.145.40.20]:21733 "EHLO mail0.lsil.com") by vger.kernel.org with ESMTP id S262281AbUCRBkJ (ORCPT ); Wed, 17 Mar 2004 20:40:09 -0500 List-Id: linux-scsi@vger.kernel.org To: Andi Kleen Cc: linux-scsi@vger.kernel.org, James.Bottomley@steeleye.com, Alexander.Stohr@gmx.de, rddunlap@osdl.org Andi - The fix in 3.01.02 was your request for commenting out the "queue is full message" which you posted to the mailing list earlier in the last week. This has nothing to do with with todays issue. Regarding todays issue. There is a work around for your issue which I have already suggested earlier today. That is to modify the mpt fusion Makefile, by uncommenting out the FIFO FIX define. This work around will set a dma mask for less 4GB, thus we don't hit the error condition in PrimeFifo's when the two memory allocation are on different 4gb boundaries, and the failure displays "-3" error code as yours did in the original dmesg. I have provided this temp fix to HP a few weeks ago. I plan to start work soon to solve this issue by combining these three FIFOs into one memory allocation( the third FIFO allocation is in mptscsih.c, that is the chain buffers). Eric Moore ----- Original Message ----- From: "Andi Kleen" To: "Moore, Eric Dean" Cc: ; ; ; Sent: Wednesday, March 17, 2004 6:09 PM Subject: Re: [PATCH 2.6 ] MPT Fusion driver 3.01.02 update > On Wed, 17 Mar 2004 14:43:21 -0500 > "Moore, Eric Dean" wrote: > > > This is an update for the MPT Fusion drivers 2.6 kernel. > > Version 3.01.02. > > > This doesn't seem to fix the bug I'm running into all the time: > > you set a 64bit consistent dma mask, but assume in PrimeIocFifos > that two calls to pci_alloc_consistent() end in the same 4GB > segment, which cannot work. > > As a workaround i just reverted it to a 4GB consistent mask for now. > > -Andi > > >