[PATCH 00/03] USB Host 2.0 and IOMMU trial on r8a7790 Lager

[PATCH 00/03] USB Host 2.0 and IOMMU trial on r8a7790 Lager

[Magnus Damm]

USB Host 2.0 and IOMMU trial on r8a7790 Lager

[PATCH 01/03] PCI: rcar: Use notifier to hook up IOMMU groups
[PATCH 02/03] iommu/ipmmu-vmsa: Create mapping via group notifier
[PATCH 03/03] ARM: shmobile: lager-reference: IOMMUSY1 and USB tests

This patch series modifies the R-Car Gen2 PCI driver and the IPMMU-VMSA
driver to coexist and assign one IOMMU group per PCI host controller.

The PCI host controllers and the IOMMU device are glued together via
prototype code that adds the IOMMU as a platform device and ties in
the 3 PCI host controllers to the IPMMUSY1 device. Patch 3/3 is not
intended for upstream merge, however patch 1/3 and 2/3 may be useful
with or without further fixes. Comments are very welcome.

The code in this series has been lightly tested on r8a7790 Lager using
USB1 on the CN5 connector. A USB smart card reader has been hot plugged
and notifications and IOMMU map/unmap operations have been verified.

IOMMU groups have been used to separate the PCI host controllers into
groups of devices that need to use the same UTLB number. Perhaps it
would make sense to also use per-device ASIDs somehow, but this needs
further discussions.

After bootup the OHCI and EHCI devices associated with the IOMMU groups
can be located like this:

# cd /sys/kernel/iommu_groups/
# ls */*  
2/devices:
0000:02:02.0  0000:02:01.0  ee0d0000.pci

1/devices:
ee0b0000.pci  0000:01:02.0  0000:01:01.0

0/devices:
ee090000.pci  0000:00:02.0  0000:00:01.0

Signed-off-by: Magnus Damm <damm+renesas@opensource.se>
---

 Built on top of renesas-devel-20140911-v3.17-rc4
 
 arch/arm/mach-shmobile/board-lager-reference.c |   30 +++++
 drivers/iommu/ipmmu-vmsa.c                     |  121 ++++++++++++++----------
 drivers/pci/host/pci-rcar-gen2.c               |   52 ++++++++++
 3 files changed, 157 insertions(+), 46 deletions(-)

Re: [PATCH 00/03] USB Host 2.0 and IOMMU trial on r8a7790 Lager

[Laurent Pinchart]

Hi Magnus,

Thank you for the patches.

On Tuesday 16 September 2014 20:20:01 Magnus Damm wrote:
> USB Host 2.0 and IOMMU trial on r8a7790 Lager
> 
> [PATCH 01/03] PCI: rcar: Use notifier to hook up IOMMU groups
> [PATCH 02/03] iommu/ipmmu-vmsa: Create mapping via group notifier
> [PATCH 03/03] ARM: shmobile: lager-reference: IOMMUSY1 and USB tests
> 
> This patch series modifies the R-Car Gen2 PCI driver and the IPMMU-VMSA
> driver to coexist and assign one IOMMU group per PCI host controller.
> 
> The PCI host controllers and the IOMMU device are glued together via
> prototype code that adds the IOMMU as a platform device and ties in
> the 3 PCI host controllers to the IPMMUSY1 device. Patch 3/3 is not
> intended for upstream merge, however patch 1/3 and 2/3 may be useful
> with or without further fixes. Comments are very welcome.

How the IOMMU subsystem hooks up devices with IOMMUs is being reworked by Will 
Deacon. I'm following that work, and plan to update the IPMMU driver 
accordingly. This will likely conflict with patch 2/3, and I would rather 
clean up the current code before extending it. Would that be fine schedule-
wise with you ?

> The code in this series has been lightly tested on r8a7790 Lager using
> USB1 on the CN5 connector. A USB smart card reader has been hot plugged
> and notifications and IOMMU map/unmap operations have been verified.
> 
> IOMMU groups have been used to separate the PCI host controllers into
> groups of devices that need to use the same UTLB number. Perhaps it
> would make sense to also use per-device ASIDs somehow, but this needs
> further discussions.
> 
> After bootup the OHCI and EHCI devices associated with the IOMMU groups
> can be located like this:
> 
> # cd /sys/kernel/iommu_groups/
> # ls */*
> 2/devices:
> 0000:02:02.0  0000:02:01.0  ee0d0000.pci
> 
> 1/devices:
> ee0b0000.pci  0000:01:02.0  0000:01:01.0
> 
> 0/devices:
> ee090000.pci  0000:00:02.0  0000:00:01.0
> 
> Signed-off-by: Magnus Damm <damm+renesas@opensource.se>
> ---
> 
>  Built on top of renesas-devel-20140911-v3.17-rc4
> 
>  arch/arm/mach-shmobile/board-lager-reference.c |   30 +++++
>  drivers/iommu/ipmmu-vmsa.c                     |  121
> ++++++++++++++---------- drivers/pci/host/pci-rcar-gen2.c               |  
> 52 ++++++++++
>  3 files changed, 157 insertions(+), 46 deletions(-)

-- 
Regards,

Laurent Pinchart

Re: [PATCH 00/03] USB Host 2.0 and IOMMU trial on r8a7790 Lager

[Magnus Damm]

Hi Laurent,

Thanks for your comments!

On Tue, Sep 16, 2014 at 9:03 PM, Laurent Pinchart
<laurent.pinchart@ideasonboard.com> wrote:
> Hi Magnus,
>
> Thank you for the patches.
>
> On Tuesday 16 September 2014 20:20:01 Magnus Damm wrote:
>> USB Host 2.0 and IOMMU trial on r8a7790 Lager
>>
>> [PATCH 01/03] PCI: rcar: Use notifier to hook up IOMMU groups
>> [PATCH 02/03] iommu/ipmmu-vmsa: Create mapping via group notifier
>> [PATCH 03/03] ARM: shmobile: lager-reference: IOMMUSY1 and USB tests
>>
>> This patch series modifies the R-Car Gen2 PCI driver and the IPMMU-VMSA
>> driver to coexist and assign one IOMMU group per PCI host controller.
>>
>> The PCI host controllers and the IOMMU device are glued together via
>> prototype code that adds the IOMMU as a platform device and ties in
>> the 3 PCI host controllers to the IPMMUSY1 device. Patch 3/3 is not
>> intended for upstream merge, however patch 1/3 and 2/3 may be useful
>> with or without further fixes. Comments are very welcome.
>
> How the IOMMU subsystem hooks up devices with IOMMUs is being reworked by Will
> Deacon. I'm following that work, and plan to update the IPMMU driver
> accordingly. This will likely conflict with patch 2/3, and I would rather
> clean up the current code before extending it. Would that be fine schedule-
> wise with you ?

Reworking based on code from WIll Deacon sounds good. I believe it is
fine schedule-wise. It would be good to know how to tie in the PCI
bits with the new code. I assume the PCI host driver needs some
modifications even with the new code.

We've touched this topic before face-to-face, but it would be good to
discuss more about IPMMU features. For instance, how to handle things
like multiple devices associated with one UTLB and perhaps also UTLB
to ASID mapping. And then there is the per-channel DMA stuff with the
associated headache!

Thanks,

/ magnus