Re: [RFC 00/18] remoteproc: Decorelate virtio from core

[Mathieu Poirier <mathieu.poirier@linaro.org>]

On Thu, Apr 16, 2020 at 06:13:13PM +0200, Arnaud Pouliquen wrote:
> This series proposes to introduce the notion of platform device for the
> rproc virtio management. One obective is to allow virtio to declare is
> own memory resources without the usage of dma_declare_coherent_memory
> that seems deprecated since the introduction of the device tree.

Just to follow up with the rest of the community...

During the openAMP remoteproc sub-group conference call [1] Arnaud and I have agreed
the best way forward for this patchset is to split it up and make a few
adjustment that will make it easier for people to review the work.

Thanks,
Mathieu

[1]. These conference call are open to anyone who wishes to participate.

> 
> Proposal:
> - the rproc virtio is processed in a separate platform and could be handled
>   as a generic platform device.
> - Several vdev devices can be declared in DT:
>     - which allows to declare their own memory regions and answer to [1].
>     - as next steps it would be also possible to:
>        - declare their own mailboxes, rpmsg drivers, ...
>        - use dma-range to handle the pa<->da translation at virtio level
> 
> Several notions are introduced here:
> - Virtio platform registration which allows to decorelate virtio from the
>   remote proc core device. 
> - Synchronization of the child devices relying on component bind/unbind.
>   This mechanism ensures the synchronization of the child devices before
>   the boot of the remote processor and before the release of the resources
>   on the remote processor shutdown.
> - Ability to populate child devices declared in rproc device tree node.
> - Possibility to declare the memory regions reserved to a virtio devices in
>   the devicetree.
> 
> Known limitations:
> - the patchset has been tested on a st32mP1 plaform only
> - it is based on the v5.6 (need to evoluate depending on V5.7 and on going
>   patchsets).
> - The virtio memory allocation does not take into account memory
>   controllers such as IOMMU and MPU.
> 
> Arnaud Pouliquen (18):
>   remoteproc: Store resource table address in rvdev
>   remoteproc: Introduce virtio device add/remove functions in core.
>   remoteproc: Move rvdev management in rproc_virtio
>   remoteproc: Add rproc_get_by_node helper
>   remoteproc: Create platform device for vdev
>   remoteproc: Add component in core for child devices synchronization
>   remoteproc: Add component bind/unbind for virtio platform
>   remoteproc: Externalize carveout functions
>   remoteproc: Move vring management from core to virtio
>   remoteproc: Add capability to populate rproc subnode devices
>   remoteproc: Add child node component in rproc match list
>   remoteproc: Support of pre-registered virtio device
>   remoteproc: Add memory default allocator helper
>   remoteproc: Add pa to da translation API
>   remoteproc: associate memory entry to a device
>   remoteproc: Parse virtio node for memory region
>   remoteproc: stm32: add the pa to da ops.
>   ARM: dts: stm32: Declare a virtio device
> 
>  arch/arm/boot/dts/stm32mp15xx-dkx.dtsi   |  10 +
>  drivers/remoteproc/remoteproc_core.c     | 469 ++++++++++++-----------
>  drivers/remoteproc/remoteproc_internal.h |  23 +-
>  drivers/remoteproc/remoteproc_virtio.c   | 415 ++++++++++++++++++--
>  drivers/remoteproc/stm32_rproc.c         |   1 +
>  include/linux/remoteproc.h               |  27 +-
>  6 files changed, 673 insertions(+), 272 deletions(-)
> 
> -- 
> 2.17.1
>