linux-mips
[Top] [All Lists]

Re: Please pull drivers/scsi/dec_esp.c from Linus' git

To: Ralf Baechle <ralf@linux-mips.org>
Subject: Re: Please pull drivers/scsi/dec_esp.c from Linus' git
From: "Maciej W. Rozycki" <macro@linux-mips.org>
Date: Thu, 16 Feb 2006 15:48:52 +0000 (GMT)
Cc: Martin Michlmayr <tbm@cyrius.com>, linux-mips@linux-mips.org
In-reply-to: <20060216145931.GA1633@linux-mips.org>
Original-recipient: rfc822;linux-mips@linux-mips.org
References: <20060213225331.GA5315@deprecation.cyrius.com> <20060215150839.GA27719@linux-mips.org> <Pine.LNX.4.64N.0602161016260.7169@blysk.ds.pg.gda.pl> <20060216145931.GA1633@linux-mips.org>
Sender: linux-mips-bounce@linux-mips.org
On Thu, 16 Feb 2006, Ralf Baechle wrote:

> That still leaves below gem to sort out.

 Yeah -- there is that mmiowb() macro that is supposed to fit here, but 
some MIPS-based hardware is ordered weakly (and strangely) enough for this 
single macro to be a bit insufficient.  I think we should have at least 
mmiowb() and mmiob() (corresponding to wmb() and mb(), respectively) as 
there is a system we support that does writes in order, but snoops the 
writeback buffer (the R3220).  Another one is worse yet as does all of 
that plus byte gathering (the R2020).  At least the latter cannot have the 
NCR/Emulex SCSI chip and uses DEC's own design instead (a DC7061 gate 
array highly suspected to also support DSSI if appropriately configured).

 I'm not sure if we really need mmiorb() -- probably not.

  Maciej

<Prev in Thread] Current Thread [Next in Thread>