Re: [PATCH v18 4/5] dt-bindings: remoteproc: Add documentation for ZynqMP R5 rproc bindings

[Linus Walleij]

Hi Ben,

thanks for your patch! I noticed this today  and pay some interest
because in the past I used with implementing the support for
TCM memory on ARM32.

On Mon, Oct 5, 2020 at 6:06 PM Ben Levinsky <ben.levinsky@xxxxxxxxxx> wrote:

> Add binding for ZynqMP R5 OpenAMP.
>
> Represent the RPU domain resources in one device node. Each RPU
> processor is a subnode of the top RPU domain node.
>
> Signed-off-by: Jason Wu <j.wu@xxxxxxxxxx>
> Signed-off-by: Wendy Liang <jliang@xxxxxxxxxx>
> Signed-off-by: Michal Simek <michal.simek@xxxxxxxxxx>
> Signed-off-by: Ben Levinsky <ben.levinsky@xxxxxxxxxx>
(...)

> +title: Xilinx R5 remote processor controller bindings
> +
> +description:
> +  This document defines the binding for the remoteproc component that loads and
> +  boots firmwares on the Xilinx Zynqmp and Versal family chipset.

... firmwares for the on-board Cortex R5 of the Zynqmp .. (etc)

> +
> +  Note that the Linux has global addressing view of the R5-related memory (TCM)
> +  so the absolute address ranges are provided in TCM reg's.

Please do not refer to Linux in bindings, they are also for other
operating systems.

Isn't that spelled out "Tightly Coupled Memory" (please expand the acronym).

I had a hard time to parse this description, do you mean:

"The Tightly Coupled Memory (an on-chip SRAM) used by the Cortex R5
is double-ported and visible in both the physical memory space of the
Cortex A5 and the memory space of the main ZynqMP processor
cluster. This is visible in the address space of the ZynqMP processor
at the address indicated here."

That would make sense, but please confirm/update.

> +  memory-region:
> +    description:
> +      collection of memory carveouts used for elf-loading and inter-processor
> +      communication. each carveout in this case should be in DDR, not
> +      chip-specific memory. In Xilinx case, this is TCM, OCM, BRAM, etc.
> +    $ref: /schemas/types.yaml#/definitions/phandle-array

This is nice, you're reusing the infrastructure we already
have for these carveouts, good design!

> +  meta-memory-regions:
> +    description:
> +      collection of memories that are not present in the top level memory
> +      nodes' mapping. For example, R5s' TCM banks. These banks are needed
> +      for R5 firmware meta data such as the R5 firmware's heap and stack.
> +      To be more precise, this is on-chip reserved SRAM regions, e.g. TCM,
> +      BRAM, OCM, etc.
> +    $ref: /schemas/types.yaml#/definitions/phandle-array

Is this in the memory space of the main CPU cluster?

It sure looks like that.

> +     /*
> +      * Below nodes are required if using TCM to load R5 firmware
> +      * if not, then either do not provide nodes are label as disabled in
> +      * status property
> +      */
> +     tcm0a: tcm_0a@ffe00000 {
> +         reg = <0xffe00000 0x10000>;
> +         pnode-id = <0xf>;
> +         no-map;
> +         status = "okay";
> +         phandle = <0x40>;
> +     };
> +     tcm0b: tcm_1a@ffe20000 {
> +         reg = <0xffe20000 0x10000>;
> +         pnode-id = <0x10>;
> +         no-map;
> +         status = "okay";
> +         phandle = <0x41>;
> +     };

All right so this looks suspicious to me. Please explain
what we are seeing in those reg entries? Is this the address
seen by the main CPU cluster?

Does it mean that the main CPU see the memory of the
R5 as "some kind of TCM" and that TCM is physically
mapped at 0xffe00000 (ITCM) and 0xffe20000 (DTCM)?

If the first is ITCM and the second DTCM that is pretty
important to point out, since this reflects the harvard
architecture properties of these two memory areas.

The phandle = thing I do not understand at all, but
maybe there is generic documentation for it that
I've missed?

Last time I checked (which was on the ARM32) the physical
address of the ITCM and DTCM could be changed at
runtime with CP15 instructions. I might be wrong
about this, but if that (or something similar) is still the case
you can't just say hardcode these addresses here, the
CPU can move that physical address somewhere else.
See the code in
arch/arm/kernel/tcm.c

It appears the ARM64 Linux kernel does not have any
TCM handling today, but that could change.

So is this just regular ARM TCM memory (as seen by
the main ARM64 cluster)?

If this is the case, you should probably add back the
compatible string, add a separate device tree binding
for TCM memories along the lines of
compatible = "arm,itcm";
compatible = "arm,dtcm";
The reg address should then ideally be interpreted by
the ARM64 kernel and assigned to the I/DTCM.

I'm paging Catalin on this because I do not know if
ARM64 really has [I|D]TCM or if this is some invention
of Xilinx's.

Yours,
Linus Walleij