[PATCH v5] devicetree: Add generic IOMMU device tree bindings

-----Original Message-----
From: Hiroshi Doyu [mailto:hdoyu at nvidia.com]
Sent: Tuesday, August 19, 2014 3:34 PM
To: Sethi Varun-B16395; Will Deacon
Cc: Hiroshi Doyu; Thierry Reding; Stephen Warren; Arnd Bergmann; Mark
Rutland; devicetree at vger.kernel.org; Olof Johansson; iommu at lists.linux-
foundation.org; Rob Herring; linux-tegra at vger.kernel.org; linux-arm-
kernel at lists.infradead.org; Yoder Stuart-B08248
Subject: Re: [PATCH v5] devicetree: Add generic IOMMU device tree bindings


Varun Sethi [off-list ref] writes:

quoted

Hi Hiroshi,

quoted

-----Original Message-----
From: Hiroshi Doyu [mailto:hdoyu at nvidia.com]
Sent: Thursday, August 14, 2014 9:35 PM
To: Sethi Varun-B16395
Cc: Hiroshi Doyu; Thierry Reding; Stephen Warren; Arnd Bergmann; Will
Deacon; Mark Rutland; devicetree at vger.kernel.org; Olof Johansson;
iommu at lists.linux-foundation.org; Rob Herring;
linux-tegra at vger.kernel.org; linux-arm-kernel at lists.infradead.org
Subject: Re: [PATCH v5] devicetree: Add generic IOMMU device tree
bindings

Hi Varun,

Varun Sethi [off-list ref] writes:

quoted

quoted

-----Original Message-----
From: iommu-bounces at lists.linux-foundation.org [mailto:iommu-
bounces at lists.linux-foundation.org] On Behalf Of Hiroshi Doyu
Sent: Thursday, August 14, 2014 12:18 PM
To: Thierry Reding; Stephen Warren; Arnd Bergmann; Will Deacon
Cc: Mark Rutland; devicetree at vger.kernel.org; Olof Johansson;
iommu at lists.linux-foundation.org; Rob Herring;
linux-tegra at vger.kernel.org; linux-arm-kernel at lists.infradead.org
Subject: Re: [PATCH v5] devicetree: Add generic IOMMU device tree
bindings


Thierry Reding [off-list ref] writes:

quoted

+Multiple-master IOMMU:
+----------------------
+
+       iommu {
+               /* the specifier represents the ID of the master */
+               #iommu-cells = <1>;
+       };
+
+       master at 1 {
+               /* device has master ID 42 in the IOMMU */
+               iommus = <&{/iommu} 42>;
+       };
+
+       master at 2 {
+               /* device has master IDs 23 and 24 in the IOMMU */
+               iommus = <&{/iommu} 23>, <&{/iommu} 24>;
+       };

I think that this "master ID" will be parsed in IOMMU driver. For
example, ARM,SMMU expects "streamID" as "master ID", right?

If a SoC has a feature to configure to assign streamID to devices
at runtime, "streamID" is not equal to "master ID".

  iommus = <&{/smmu} "soc specific master ID">;

"soc master ID" needs to be translated into "streamID" by SoC SW.
It seems that ARM,SMMU kernel driver doesn't expect this kind of
ID translation. If ARM,SMMU kernel driver is used as is, "soc master ID"
would be incompatible? ARM,SMMU needs such translation before
parsing. Is this my understanding right?

If so I think that this master ID configuration/translation may be
quite reasonable requirment for SoC using ARM,SMMU.

Can we consider this ID translation within ARM,SMMU compatibility?

IOW, is it possible to implement some SoC specific hook for ID
translation/configuration in ARM,SMMU kernel driver?

Can the id translation be done using a SMR mask?

No, "SoC master ID" is completely independenf of SMR.

quoted

Also, for dynamic stream ID allocation we would need to represent
the specific master register (to store the stream ID) in the device tree.

I assmue that the above means that iMX has such configuration
register to map steramID and a device dynamically.

We have per master registers for setting the stream ID on the
Layerscape platforms. My point was that we would need the iommu master
node to include a reference to the master id register.

master at 1 {
               /* device has master ID 42 in the IOMMU */
             iommus = <&{/iommu} 42>;
             master-id-reg = <phandle offset> };

In the above, for "iommus=" bindings, you wouldn't need to break
ARM,SMMU compatibility at all if you set "streamID" exactly as below.

  master at 1 {
                 /* device has master ID 42 in the IOMMU */
               iommus = <&{/iommu} 'any given streamID'>;
               master-id-reg = <phandle offset>
  };

And your SoC needs to register bus_notifier and ADD_DEVICE should configure
to map 'any given streamID' to a device via the above register. This wouldn't
need any modification from ARM,SMMU driver and keep the iommus bindings
as it is.

IOW, SoC only needs to register ADD_DEVICE in bus_notifier to map StreamID
to a device. This needs to be executed earlier than IOMMU bus's ADD_DEVICE,
though.

Is my understanding right?

I don't think that SOC specific code needs a bus notifier for setting the stream ID. It can be done as a part of SOC specific initialization. The device tree can be updated to reflect the correct stream ID (SMMU driver can get the updated stream ID from device tree). 
I was thinking more on the lines of updating the device stream id while attaching a device to the domain.  

-Varun