PROCESSOR-SDK-AM57X: Beyond SDK v08.02.00.04, where did the PRU doc and examples go?

Hello Nathan,

The featured was de-scoped from the SDK 9.2 release. Therefore it was removed from the documentation. We are working on bringing the feature back in future releases.

I believe the features are still supported in the second release of SDK 8.2x. It might just be an oversight in the documentation. 

Do you have any issues recreating the functionality you need?

-Josue

Thanks Josue,

Is there a specific tag or branch that is most compatible with version 9 of the SDK?

Nathan

Hold on, are you stating that PRU functionality will not work at all with version 9 of the SDK?

If that is true I cannot upgrade to version 9 at all.  When do you expect the feature to function again?

Correct. See https://software-dl.ti.com/processor-sdk-linux/esd/AM57X/09_02_00_133/exports/docs/devices/AM57X/linux/Release_Specific_Release_Notes.html#release-09-02-00

We are working on a plan for this functionality to be fully back in SDK 11 sometime mid year 2025. Plan is also in motion for 9.3 release that will support PRU eth on its own, for this December.

-Josue

Apparently not.

I have tried both versions of SDK 8, and neither compiles.  It looks like pru_types.h is broken.  They added some structs, removed another, but kept pointers to the removed struct around.  So compilation failure.

Hello Nathan,

Could you please specify exactly what you tried and what version of the Programmable Real-time Unit (PRU) Software Support Package release you are using?

Please see the following: https://git.ti.com/cgit/pru-software-support-package/pru-software-support-package/tree/Release_Notes.txt

I believe you should be using Release 6.0.

Best,

Josue

I have tried SDK versions:

• ti-processor-sdk-linux-rt-am57xx-evm-08_02_00_04
• ti-processor-sdk-linux-rt-am57xx-evm-08_02_01_00

All of them are on pru-icss-6.0.1.

pru_types.h does not compile because it has structs that have a pointer to an undefined type `struct ch_map`.  That struct was removed and replaced by `struct pruss_int_map` because the remote_proc driver changed.  However 2 structs below it in the same file (`struct fw_rsc_custom_ints` & `struct fw_rsc_custom_ints_k3`) both contain pointers to the now undefined `struct ch_map`.

Looking at the git site, pru_types.h would have last compiled in release 5.9.0.  git.ti.com/.../pru_types.h

It would appear we cannot move our tool chain beyond SDK v6 because RP messaging is broken.

Nathan,

Does this mean that you would not be migrating your project to SDK 8.x?

Nathan Wright said:

It would appear we cannot move our tool chain beyond SDK v6 because RP messaging is broken.

By RP you mean Remote Proc?

-Josue

Hello Nathan, I'm the guy who maintains the PRU Software Support Package (PSSP). I'm in a different team than Josue so I have not tested AM57x recently, but I can comment on other stuff going on with the PSSP.

To clarify:

support for PRU Ethernet was dropped in AM57x Linux SDK 9.2.

Josue, can you confirm whether support for Linux initializing the PRU subsystem with the remoteproc driver, and using RPMsg (aka RP Message) to communicate with the PRU subsystem, was also dropped?

When I run tests on AM335x, it looks like the general remoteproc loading of PRU works fine, as long as it is not using RPMsg with Linux. The issue is with setting up the virtio interface. This was a known issue on AM335x that I thought the team had fixed sometime after the AM335x SDK 9.1 release?

root@am335x-evm:/lib/firmware# uname -a
Linux am335x-evm 6.1.46-g1d4b5da681 #1 PREEMPT Thu Oct 19 10:19:08 UTC 2023 armv7l armv7l armv7l GNU/Linux

// load with RPMsg fails
root@am335x-evm:/lib/firmware# ln -sf ~/PRU_RPMsg_Echo_Interrupt1.out am335x-pru1-fw
root@am335x-evm:/lib/firmware# ln -sf ~/PRU_RPMsg_Echo_Interrupt0.out am335x-pru0-fw

root@am335x-evm:/lib/firmware# echo start > /sys/class/remoteproc/remoteproc0/state
[ 1176.431474] remoteproc remoteproc0: powering up 4a334000.pru
[ 1176.438269] remoteproc remoteproc0: Booting fw image am335x-pru0-fw, size 70984
[ 1176.468574] pru-rproc 4a334000.pru: error -ENXIO: IRQ vring not found
[ 1176.487350] remoteproc remoteproc0: unable to get vring interrupt, status = -6
[ 1176.500537] remoteproc remoteproc0: can't start rproc 4a334000.pru: -6
[ 1176.516707] remoteproc remoteproc0: Boot failed: -6

// load with INTC configuration (but no RPMsg) succeeds
root@am335x-evm:/lib/firmware# ln -sf /lib/firmware/pru/PRU_Direct_Connect0.out am335x-pru0-fw
root@am335x-evm:/lib/firmware# echo start > /sys/class/remoteproc/remoteproc0/state
[ 1296.221395] remoteproc remoteproc0: powering up 4a334000.pru
[ 1296.346828] remoteproc remoteproc0: Booting fw image am335x-pru0-fw, size 43452
[ 1296.356730] remoteproc remoteproc0: remote processor 4a334000.pru is now up
root@am335x-evm:/lib/firmware# ln -sf /lib/firmware/pru/PRU_Direct_Connect1.out am335x-pru1-fw
root@am335x-evm:/lib/firmware# echo start > /sys/class/remoteproc/remoteproc1/state
[ 1314.621599] remoteproc remoteproc1: powering up 4a338000.pru
[ 1314.747563] remoteproc remoteproc1: Booting fw image am335x-pru1-fw, size 25760
[ 1314.763044] remoteproc remoteproc1: remote processor 4a338000.pru is now up

We can see the commit causing this behavior here:
https://git.ti.com/cgit/ti-linux-kernel/ti-linux-kernel/commit/drivers/remoteproc?h=ti-linux-6.1.y-cicd&id=3b4bcb5e26a5fb9312cceaeef9065164868d0e3f
but I do not see any commits on the Linux kernel 6.1, 6.6, or Linux upstream branch that look like a fix. Let us check with the developers.

Regards,

Nick

Nathan, Nick,

RPMsg in SDK 8.2 should be working. There are a few issues with the release in terms of remote proc working including making sure a v-ring patch is applied and one is using the correct IPC version since the RTOS and Linux Release were offset. 

See: https://e2e.ti.com/support/processors-group/processors/f/processors-forum/1304306/faq-processor-sdk-am57x-what-ipc-version-for-do-i-use-when-using-linux-sdk-08-02-01-00

-Josue

Thanks to both of you for your continued support.

What we have is essentially an extension of `PRU_RPMsg_Echo_Interrupt1_0` example. 

We have a small c program that runs on the PRU that generates a square wave synced to time provided via a message.  Up in user space we grab the time from `/dev/ptp0` and send it to the PRU via messaging to `/dev/rpmsg_pru30`. 

That's all it is. 

For future readers, we document how interrupts between Linux and PRUs should be defined here:
https://software-dl.ti.com/processor-sdk-linux/esd/AM57X/08_02_00_04/exports/docs/common/PRU-ICSS/INTC_Configuration.html

I also added an example of what your Linux-side code should look like in the PRU firmware's intc_map file:
https://git.ti.com/cgit/pru-software-support-package/pru-software-support-package/tree/examples/am572x/PRU_RPMsg_Echo_Interrupt1_0/intc_map_0.h

Regards,

Nick

For Nathan & future readers, let me make a quick disclaimer:

PRU remoteproc and PRU RPMsg was not tested on AM57x for kernel 6.1. So if things do not work for you, Josue will be limited in the assistance he can offer. 

With that said... we did not make a bunch of kernel changes to PRU remoteproc & RPMsg between kernel 5.10 and kernel 6.1. So I would EXPECT that you can still load non-RPMsg firmware without making any modifications (e.g., PRU_Halt, PRU_Direct_Connect).

And I would EXPECT that you should be able to get RPMsg firmware to load as well, after applying the patch for your specific processor:
https://git.ti.com/cgit/ti-linux-kernel/ti-linux-kernel/log/arch/arm/boot/dts?h=ti-linux-5.10.y-cicd&qt=grep&q=Add+PRU+system+events+for+virtio

Regards,

Nick