From mboxrd@z Thu Jan  1 00:00:00 1970
From: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Date: Fri, 02 Oct 2015 08:13:32 +0000
Subject: Re: [PATCH] Gen2 IPMMU and DU prototype
Message-Id: <4312084.V9UEz2QYFm@avalon>
List-Id: <linux-sh.vger.kernel.org>
References: <20150930142242.25755.81946.sendpatchset@little-apple>
In-Reply-To: <20150930142242.25755.81946.sendpatchset@little-apple>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: linux-sh@vger.kernel.org

Hi Magnus,

On Friday 02 October 2015 12:12:22 Magnus Damm wrote:
> On Thu, Oct 1, 2015 at 11:30 PM, Laurent Pinchart wrote:
> > On Wednesday 30 September 2015 23:22:42 Magnus Damm wrote:
> >> From: Magnus Damm <damm+renesas@opensource.se>
> >> 
> >> Here is a simple hack to enable the IPMMU on R-Car Gen2.
> >> 
> >> The VGA port on r8a7790 Lager may be driven by the DU
> >> and the IPMMU via the modetest utility to output some
> >> test image.
> >> 
> >> Actually the r8a7790 DU device is connected to the
> >> IPMMU using two separate uTLBs, but the DU driver
> >> is currently supporting all DU instances using a
> >> single device. To make testing easily the r8a7790 DU
> >> driver code is with this patch hacked to just cover
> >> a single DU instance with a single IPMMU uTLB.
> > 
> > You can reference multiple uTLBs in the DT ipmmus property. The ipmmu-vmsa
> > driver currently requires all uTLBs to belong to the same IPMMU instance,
> > which shouldn't be an issue here.
> 
> Sure, from a DT property point of view it seems possible to assign
> multiple uTLBs to the DU node. I would like us to start integrating
> the IPMMU on Gen2 and passing multiple uTLBs in case of r8a7790 must
> be the right way to describe the hardware.
> 
> When it comes to the DU driver and memory management, from my point of
> view Gen2 also seems to pass a single device pointer passed to
> dma_alloc_writecombine(). I may be wrong, but to support multiple
> uTLBs it looks like we need to pass different device pointers to
> dma_alloc_writecombine() - unless some index is also passed how can
> the IPMMU code otherwise know which uTLB to use?
> 
> So it looks like the device pointer issue both affects Gen2 and Gen3.
> Of course I wish that wasn't the case, so please prove me wrong. =)

The IPMMU driver currently enables all uTLBs associated with a bus master 
device. As the driver uses a single page table for all uTLBs we shouldn't have 
any issue on Gen2. Problems will of course appear if/when we start using 
multiple page tables for a single IPMMU instance.

-- 
Regards,

Laurent Pinchart