> -----Original Message----- > From: Matthew Wilcox [mailto:matthew@xxxxxx] > Sent: Wednesday, March 01, 2006 1:12 PM > To: Tarte, Robert > Cc: Jeff Garzik; Mike Anderson; James Bottomley; linux- > scsi@xxxxxxxxxxxxxxx; Jeff Garzik > Subject: Re: [PATCH] aic94xx: driver assertion in non-x86 BIOS env > > On Wed, Mar 01, 2006 at 12:57:12PM -0800, Tarte, Robert wrote: > > I have located an HP machine that is the same flavor as James' (HP > > Integrity rx2600: ia64 with zx1 chipset). It exhibits the same problem. > > I believe that the architecture may differ with traditional PC/PCI > > architectures in how it deals with non-aligned PCI access. When I > > converted the OCM initialization code to write on qwords instead of > > dwords, the code was able to pass OCM initialization. It is possible > > that the machine is doing a qword read/modify/write when accessing OCM > > memory, which would explain the hang. I haven't hooked up a PCI > > analyzer. The code now gets to the point where it is trying to read > > NVRAM in order to get the SAS address. It seems that the machine is > > unable to retrieve non-aligned byte accesses correctly. I will convert > > the code to pull the correct byte/word/dword from the aligned qword. > > We'll see if that fixes it. None of these behaviors are exhibited on > > x86, x86_64, powerpc, and ia64 without the zx1 chipset. I understand > > that there may be a few ARM implementations that have similar issues. > > You're saying that you do an 8-byte non-DMA write to an address > that's congruent to 4 mod 8? Based on briefly skimming the internal > documentation for zx1, I'd say you're invoking undefined behaviour. Thanks for the confirmation. Internal documentation for zx1? Where can I locate this? Thanks in advance. > > - > > : send the line "unsubscribe linux-scsi" in > > the body of a message to majordomo@xxxxxxxxxxxxxxx > > More majordomo info at http://vger.kernel.org/majordomo-info.html - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html