Re: [PATCH v4 1/6] of: iommu: add ptr to OF node arg to of_iommu_configure()

[Laurent Pinchart <laurent.pinchart@ideasonboard.com>]

Hi Will,

On Wednesday 28 January 2015 13:32:19 Will Deacon wrote:
> On Wed, Jan 28, 2015 at 01:15:10PM +0000, Laurent Pinchart wrote:
> > On Wednesday 28 January 2015 12:29:42 Will Deacon wrote:
> >> On Wed, Jan 28, 2015 at 12:23:03PM +0000, Laurent Pinchart wrote:
> >>> On Wednesday 28 January 2015 11:33:00 Will Deacon wrote:
> >>>> On Mon, Jan 26, 2015 at 06:49:01PM +0000, Murali Karicheri wrote:
> >>>>> On 01/25/2015 08:32 AM, Laurent Pinchart wrote:
> >>>>>> On Friday 23 January 2015 17:32:34 Murali Karicheri wrote:
> >>>>>>> Function of_iommu_configure() is called from of_dma_configure() to
> >>>>>>> setup iommu ops using DT property. This API is currently used for
> >>>>>>> platform devices for which DMA configuration (including iommu ops)
> >>>>>>> may come from device's parent. To extend this functionality for
> >>>>>>> PCI devices, this API need to take a parent node ptr as an argument
> >>>>>>> instead of assuming device's parent. This is needed since for PCI,
> >>>>>>> the dma configuration may be defined in the DT node of the root
> >>>>>>> bus bridge's parent device. Currently only dma-range is used for PCI
> >>>>>>> and iommu is not supported. So return error if the device is PCI.
> >>>>>>> 
> >>>>>>> Cc: Joerg Roedel<joro@8bytes.org>
> >>>>>>> Cc: Grant Likely<grant.likely@linaro.org>
> >>>>>>> Cc: Rob Herring<robh+dt@kernel.org>
> >>>>>>> Cc: Bjorn Helgaas<bhelgaas@google.com>
> >>>>>>> Cc: Will Deacon<will.deacon@arm.com>
> >>>>>>> Cc: Russell King<linux@arm.linux.org.uk>
> >>>>>>> Cc: Arnd Bergmann<arnd@arndb.de>
> >>>>>>> Cc: Suravee Suthikulpanit<Suravee.Suthikulpanit@amd.com>
> >>>>>>> 
> >>>>>>> Signed-off-by: Murali Karicheri<m-karicheri2@ti.com>
> >>>>>>> ---
> >>>>>>> 
> >>>>>>>   drivers/iommu/of_iommu.c |   10 ++++++++--
> >>>>>>>   drivers/of/platform.c    |    2 +-
> >>>>>>>   include/linux/of_iommu.h |    6 ++++--
> >>>>>>>   3 files changed, 13 insertions(+), 5 deletions(-)
> >>>>>>> 
> >>>>>>> diff --git a/drivers/iommu/of_iommu.c b/drivers/iommu/of_iommu.c
> >>>>>>> index af1dc6a..439235b 100644
> >>>>>>> --- a/drivers/iommu/of_iommu.c
> >>>>>>> +++ b/drivers/iommu/of_iommu.c
> >>>>>>> @@ -133,19 +133,25 @@ struct iommu_ops *of_iommu_get_ops(struct
> >>>>>>> device_node *np)
> >>>>>>>   	return ops;
> >>>>>>>  }
> >>>>>>> 
> >>>>>>> -struct iommu_ops *of_iommu_configure(struct device *dev)
> >>>>>>> +struct iommu_ops *of_iommu_configure(struct device *dev,
> >>>>>>> +				     struct device_node *iommu_np)
> >>>>>>>  {
> >>>>>>>   	struct of_phandle_args iommu_spec;
> >>>>>>>   	struct device_node *np;
> >>>>>>>   	struct iommu_ops *ops = NULL;
> >>>>>>>   	int idx = 0;
> >>>>>>> 
> >>>>>>> +	if (dev_is_pci(dev)) {
> >>>>>>> +		dev_err(dev, "iommu is currently not supported for PCI\n");
> >>>>>>> +		return NULL;
> >>>>>>> +	}
> >>>>>>> +
> >>>>>>>   	/*
> >>>>>>>   	 * We don't currently walk up the tree looking for a parent
> >>>>>>>   	 IOMMU.
> >>>>>>>   	 * See the `Notes:' section of
> >>>>>>>   	 * Documentation/devicetree/bindings/iommu/iommu.txt
> >>>>>>>   	 */
> >>>>>> 
> >>>>>> Wouldn't it be better to fix this rather than passing the device
> >>>>>> node pointer to the function ? The solution would be more generic.
> >>>>> 
> >>>>> Will Deacon (Copied here) is working on this as we alteady discussed
> >>>>> in this thread. So it will be addressed by him in another series.
> >>>> 
> >>>> Well, "working on this" equates to "has it somewhere near the bottom
> >>>> of a very long list" :)
> >>> 
> >>> What's your opinion on this patch then, do you think adding the
> >>> iommu_np argument makes sense as a temporary hack, or should we instead
> >>> walk up the tree looking for an iommus attribute in parent nodes ? I
> >>> don't expect that to be insanely difficult to code.
> >> 
> >> If walking up the tree is useful, then I think we should do that and
> >> update the Documentation to reflect the new behaviour.
> > 
> > If I understand Murali's patch set right (please correct me if that's not
> > the case) the PCI code walks up the DT nodes hierarchy to the parent node
> > that contains the iommus attribute and passes a pointer to that node to
> > this function. It's thus a PCI-specific solution. As a temporary hack
> > that's OK I suppose, but if implementing it right straight away isn't
> > difficult that would be better.
> 
> It looks to me like the code walks the PCI topology to get the DT node for
> the host controller, and passes *that* to of_dma_configure. That sounds
> like the right thing to do to me, especially since the PCI topology is
> likely not encoded in the device-tree. So actually, it is passing the
> first parent node afaict.

Indeed, that's right. I forgot for a moment that we have non-DT devices ;-)

Acked-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>

Murali, nitpicking a bit, shouldn't the iommu_np parameter be renamed ? It 
points to the node containing the iommus parameter, not to the iommu node, so 
the current name is slightly confusing. A brief kerneldoc above the function 
would also help. This can be the subject of a separate patch.

> The part I'm more interested in is the mapping of RequesterID (PCI dma
> alias) to IOMMU ID when we transition from the PCI topology to the DT
> topology. Currently, it seems to be 1:1 on the platforms I have, but I
> doubt this will always be the case.

-- 
Regards,

Laurent Pinchart