TI PRU_ADC_onChip example

265 views
Skip to first unread message

Cheng Chen

unread,
Apr 30, 2021, 4:09:02 PM4/30/21
to BeagleBoard
Hi all, 

I am practicing PRU skills through some TI examples. I am playing with PRU_ADC_onChip example and ran into a few questions. I wonder if you can help me with. 

The nice thing about this example is it has a README.txt with all the procedures. The idea is to use rpmsg to communicate between arm and pru module and read out ADC value. 
I am using a Beaglebone Black wireless.
Here is the basic procedures.

FILE STRUCTURE
PRU_ADC
  |
  |--pru_adc_firmware.c firmware loaded into PRU0
  |
  |--pru_adc_userspace
       |--pru_adc_userspace.c userspace code that interacts with PRU0

BUILD FIRMWARE & USERSPACE CODE
$ cd <PATH_TO_PRU_SW_SUPPORT_PACKAGE>/examples/am335x/PRU_ADC_onChip/
$ make clean
$ make
$ cd pru_adc_userspace/
$ make clean
$ make

My BBB wireless can compile pru code successfully because I installed PRU_CGT compiler. But it is unable to compile ARM code. I think that is because ARM_CCT cross-compiler toochain environment is missing, in another word, I need to install processor-sdk-am335x

My first questions is can I install processor-sdk-am335x  into Debian system I currently have (Linux beaglebone 4.19.94-ti-r62 ) ? I am a little confused about the relationship between this SDK and Debian system. Why is the tutorial asking me to compile  pru_adc_userspace.c  in the Beaglebone. I thought it is supposed to be executed in a cross-compilation environment.

I ended up installing processor-sdk-am335x on my linux desktop and compiled successfully. Then I copied the generated file back to BBB wireless. But when I tried to run the program, it shows the following error. 

Reading voltage at ADC Channel: 5
/dev/rpmsg_pru30 could not be opened.
Trying to initialize PRU using sysfs interface.
ERROR: Could not open /dev/rpmsg_pru30

Attached is the excerpt where the problem happened. Could anyone help me with some hints of what is going on here? Much appreciated. 


struct shared_struct message;

/* use character device /dev/rpmsg_pru30 */
char outputFilename[] = "/dev/rpmsg_pru30";

/* test that /dev/rpmsg_pru30 exists */
FILE *ofp;
uint16_t returnedVoltage;
ofp = fopen(outputFilename, "r");

if (ofp == NULL) {

printf("/dev/rpmsg_pru30 could not be opened. \n");
printf("Trying to initialize PRU using sysfs interface.\n");

FILE *sysfs_node;
char firmware[] = "/sys/class/remoteproc/remoteproc1/firmware";
char firmwareName[] = "PRU_ADC_onChip.out";
sysfs_node = fopen(firmware, "r+");
if (sysfs_node == NULL) {
printf("cannot open firmware sysfs_node");
return EXIT_FAILURE;
}
fwrite(&firmwareName, sizeof(uint8_t), sizeof(firmwareName),
sysfs_node);
fclose(sysfs_node);

char pruState[] = "/sys/class/remoteproc/remoteproc1/state";
char start[] = "start";
sysfs_node = fopen(pruState, "r+");
if (sysfs_node == NULL) {
printf("cannot open state sysfs_node");
return EXIT_FAILURE;
}
fwrite(&start, sizeof(uint8_t), sizeof(start), sysfs_node);
fclose(sysfs_node);

/* give RPMSG time to initialize */
sleep(3);

ofp = fopen(outputFilename, "r");

if (ofp == NULL) {
printf("ERROR: Could not open /dev/rpmsg_pru30\n");
exit(EXIT_FAILURE);
}
}





Mark Lazarewicz

unread,
Apr 30, 2021, 5:09:01 PM4/30/21
to Cheng Chen, BeagleBoard
Hello 

I know the code. It's all explained in the SDK documention. I also like these examples.
Your asking questions about an SDK that's supported by Texas Instruments. You do understand this .org group you posted in may contain TI employees but is NOT TI support it's Beagle board Debian.

 I think if you read the documents you will understand the answers

 I'm sure you could compile this with some work the sdk instructions talk about This. 

Hypothetical question ❓
If the instructions told you a virtual box build was not supported and not recommended would you use one anyway and then ask the person who told you not too do this why it doesn't work.

I'm struggling to be nice in this reply.

 I remember asking questions as a young engineer that clearly showed I made zero effort to research anything.

Then one day I got some really critical replies about my skills.

Do some reading then if stuck ask questions

Or better yet follow what the sdk instructions suggest.

If you found this code  on internet and don't have a TI account or are not eligible for ITAR restrictions to download the sdk  you have a big problem.

I see you have a Gmail account




--
For more options, visit http://beagleboard.org/discuss
---
You received this message because you are subscribed to the Google Groups "BeagleBoard" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beagleboard...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beagleboard/1d43aa7b-0e94-4431-9e31-13545213940bn%40googlegroups.com.

Dennis Lee Bieber

unread,
Apr 30, 2021, 8:45:57 PM4/30/21
to Beagleboard
On Fri, 30 Apr 2021 13:09:02 -0700 (PDT), in
gmane.comp.hardware.beagleboard.user Cheng Chen
<chenc2442-Re5JQE...@public.gmane.org> wrote:


>My BBB wireless can compile pru code successfully because I installed
>PRU_CGT compiler. But it is unable to compile ARM code. I think that is
>because ARM_CCT cross-compiler toochain environment is missing, in another
>word, I need to install processor-sdk-am335x
>
The key is probably in the phrase "cross-compiler"... If you are
compiling ON the Beagle, you are using the "native" compiler, not a
cross-compiler ("cross" implies a toolchain that runs on a different
OS/architecture but which builds binary files meant for the target of the
toolchain -- but you are ON ARM, building FOR ARM).

>My first questions is can I install processor-sdk-am335x into Debian
>system I currently have (Linux beaglebone 4.19.94-ti-r62 ) ? I am a little
>confused about the relationship between this SDK and Debian system. Why is
>the tutorial asking me to compile pru_adc_userspace.c in the Beaglebone.
>I thought it is supposed to be executed in a cross-compilation environment.
>

Cross compilation is used mostly for a couple of reasons: either the
target system does not support a native compiler (you aren't going to run a
compiler on something like a Tiva C board, or an Arduino Due [both of which
run ARM M-series processors] -- though the native compiler on a
Beagle/Debian might be able to build files that can be downloaded to such
boards); the target board is not yet built -- in which case one is likely
going to be "running" the code using something like QEMU or other software
emulation of the hardware system; the target board has a compiler, but the
speed of the cross-compiler environment is much faster (consider a
hyper-threaded 64-bit processor -- which is seen as 8-cores by the OS, each
running at 2.5-3.5GHz, with 12GB of RAM, vs a single 32-bit core running at
1GB with less than 1GB of RAM).

>I ended up installing processor-sdk-am335x on my linux desktop and compiled
>successfully. Then I copied the generated file back to BBB wireless. But
>when I tried to run the program, it shows the following error.
>
>Reading voltage at ADC Channel: 5
>/dev/rpmsg_pru30 could not be opened.
>Trying to initialize PRU using sysfs interface.
>ERROR: Could not open /dev/rpmsg_pru30
>

This is a different matter, and likely means the device tree
configuration is not set correctly -- over the last few years there have
been two INCOMPATIBLE means of accessing the PRU. RemoteProc/RPMsg is the
newer system, pru_uio is older -- current images for the Beagle load
RemoteProc -- check /boot/uEnv.txt to see what version your system is
loading...

###PRUSS OPTIONS
###pru_rproc (4.14.x-ti kernel)
#uboot_overlay_pru=/lib/firmware/AM335X-PRU-RPROC-4-14-TI-00A0.dtbo
###pru_rproc (4.19.x-ti kernel)
uboot_overlay_pru=/lib/firmware/AM335X-PRU-RPROC-4-19-TI-00A0.dtbo
###pru_uio (4.14.x-ti, 4.19.x-ti & mainline/bone kernel)
#uboot_overlay_pru=/lib/firmware/AM335X-PRU-UIO-00A0.dtbo
###

... however, I don't seem to have any /dev/rpmsg* entries either.

I'm sure this have been brought up in this forum a few times in the
last year or so, but I don't recall the solution (if any). I did find that
the default firmware was not "running"... But starting it didn't add the
/dev/rpmsg*

debian@beaglebone:~$ cat /sys/class/remoteproc/remoteproc1/*
cat: /sys/class/remoteproc/remoteproc1/device: Is a directory
am335x-pru0-fw
4a334000.pru
cat: /sys/class/remoteproc/remoteproc1/power: Is a directory
offline <*****************************************
cat: /sys/class/remoteproc/remoteproc1/subsystem: Is a directory
DEVTYPE=remoteproc
debian@beaglebone:~$ echo 'start' > /sys/class/remoteproc/remoteproc1/state
debian@beaglebone:~$ cat /sys/class/remoteproc/remoteproc1/*
cat: /sys/class/remoteproc/remoteproc1/device: Is a directory
am335x-pru0-fw
4a334000.pru
cat: /sys/class/remoteproc/remoteproc1/power: Is a directory
running <*********************************************
cat: /sys/class/remoteproc/remoteproc1/subsystem: Is a directory
DEVTYPE=remoteproc



--
Dennis L Bieber

Cheng Chen

unread,
Apr 30, 2021, 8:52:03 PM4/30/21
to BeagleBoard
Hey Mark, 

Thanks for spending time for replying. I really appreciate it. 
I totally agree with you that one should spend time investigating first. I apologize if they are dumb questions, but I have stuck there for two weeks. I am more a circuit guy and just started picking up Beaglebone as a hobby and potential career expansion. 

My first intention is to post in TI E2E support forums, but it requires a company email to do so. If this particular .org is not a good place for rookie questions, would you please advise some place for suitable for discussion?

Regarding to the documents, are you referring to processor SDK Linux Software Developer's guide? If that is the one, I did follow its instructions, but maybe I missed something. I will go over it again. If that's not the one, which document are you referring to? I also went through PRUcookbook and Exploring BeagleBone by Derek Molly. Not a lot are mentioned regarding this topic. 

The code is built-in in the Beaglebone Black, this is one of the reasons I am confused why it cannot be compiled. One can also download freely from TI github (https://git.ti.com/cgit/pru-software-support-package/pru-software-support-package/). 

Again thanks for the advice and suggestion. I am very glad that there is such a nice place that I can call for help and I hope after some practice I am also able to contribute here. 

Regards,
Cheng

Mark Lazarewicz

unread,
Apr 30, 2021, 9:06:24 PM4/30/21
to beagl...@googlegroups.com
Cheng

I'm actually using the SDK now so be careful about generic advise
the instructions are for  Ubuntu not Debian
'Since I had questions myself right now  I realized this PRU code you have just a small part of the SDK
My point was you never read the SDK quick start
Please do so here is the doc tarball
'

If you search the group on X15 and DSP Jeff Aused the SDK and ran the binary on Debian(For the DSP with remore proc)

So yes you can use Debian to load the PRU firmware you need a corresponding ARM binary for that code. Jeff used the SDK to build that

I dont know where you read to compile SDK examples on BBB its not something I read

So use Ubuntu box as instructions tell you too

OR

wait to see if Jeff Andlich can tell you if its feasable to attempt to install the SDK on the board itself and compile the code  that doesn't sound right to me

OR

create an account on E2E ask the people who designed the SDK especially if your confused

I can answer CCS JTAG barebone ARM questiion about SDK not interested in Linux

Please read the tarball in  the link

Regard
'


Cheng Chen

unread,
Apr 30, 2021, 9:10:02 PM4/30/21
to BeagleBoard
Hey Dennis, 

Thanks for the reply. It makes a lot sense for cross-compiler. Thanks for the explanation. 
I am pretty sure I am running on 4.19.x-ti kernel. 
uboot_overlay_pru=/lib/firmware/AM335X-PRU-RPROC-4-19-TI-00A0.dtbo

I don't have any problem running other PRU examples. I also noticed that the original firmware is not running at first. But once it is loaded, it seems ok. I tried creating /dev/rpmsg by using "echo hello > /dev/rpmsg_pru30" before running example code. But it would leads to program freeze. I am also exploring direct memory access method to save acquired values in shared memory, but I haven't got any results yet. 

I think I will investigate into the device tree issue as you suggested. Thanks!

Regards,
Cheng

Cheng Chen

unread,
May 1, 2021, 1:21:45 AM5/1/21
to BeagleBoard
Hi Mark, 

Thanks a lot for the response.
I think the idea in this processor SDK Linux tutorial is to create Linux SD card from the Processor SDK Linux package, which somewhat echoes with your suggestion. 
I guess what confuses me is that in README.txt of the example, it just says "build firmware & userspace code" in the directory, which I assume both are done in BBB. Firmware is straightforward to build, because there is PRU code generation tools - compiler. I already used cross-compiler to build Userspace (ARM side) code. In the meantime, I just wonder if that can also be done in current BBB system. I will definitely study the document you suggest. Thanks!

In addition, the bigger problem is actually rpmsg_pru30 issue. That literally prevents me from running any rpmsg functions. As Dennis mentioned, this might be a device tree problem? I need to spend some time on this. 

Regards,
Cheng

Mark Lazarewicz

unread,
May 1, 2021, 2:06:06 AM5/1/21
to beagl...@googlegroups.com
Hi Cheng

The tarball has step by step instructions for that example you mentioned in initial post.
you need that when starting out.

Why?   because few in this group use SDK. Unfortunately you have no choice to ask questions here.


When code doesn't work on ARM you will get advise to use cookbook and Debian or even worse use libpruio😉😣 in this group then you will really be confused.


 You asked howto to run sdk examples  on Debian linux

Your problems are 

1) the instructions say use Ubuntu to install the SDK  and build PRU code and Linux Application code in the SDK ie it's built on a Ubuntu LTS PC a specific version of Ubuntu as well.

2) the ARM  example code you build to talk to PRU is designed for SDK Linux not Debian


#1 above may be possible to overcome . BUT instead 


you could just take the SDK ARM example binary 
and the PRU firmware binary if it exist in SDK binary directory and put them on the ARM. What's in SDK directories how the example works step by step are described in SDK pdf.( I sent a link  to tarball docs start there read how to install apSDK and host requirements )

#2
Jeff built both sperm and PRU binary in his SDK and put them on Debian and said it  it worked on Debian

I don't know where Jeff  installed his SDK but I doubt he put it on the BBB I don't recommend that.

No body seems to understand my instructions in this group ive been told so if its not clear ask. Its all described in SDK 

Lastly I repeat 



Your going to be told don't use SDK use cookbook or libpruio🤔 and Debian on ARM if you try this SDK example and it doesn't work it's already started your having problems asking questions correct?




If you got an account on E2E

They would say why aren't you using SDK Linux we don't support Debian 

Understand?

You have nowhere to go you can't get in E2E

Understand?

I like the SDK myself  but I did see Cookbook has similar RPM Message  example and Marks documented that really well 

What happened there? Isn't there an RPMSG ADC example?

Anyway you would not have any problems if you ran SDK linux with the code example you have  and in theory it should work on Debian the differences are the SDK  linux vs the Yocto SDK Linux maybe muxing I don't know.



My am335x starter kit board i have at home  came with SDK linux running on the  ARM  and my Beaglebone white I'm using with CCS has JTAG built in and has Ubuntu on the SD card I just pull it out.

That's why I understand what's happening. 

 I started out 5 years ago I tried Ubuntu on bone  white followed this group's instructions it was easy then I bought EVM and learned Yocto SDK Linux building at that time if I had questions answers were in the SDK off tutorial and E2E forum would support me. Their board their SDK.

Don't feel bad every month someone asks the same questions you did.

 they say "I found this PRU example in the  SDK can I run this on Debian?"

I say yes it will work what you are attempting.


but I can't help you when  Linux doesn't run your code and gets errors and when you ask for help in this group I expect you will be told use the Cookbook examples.

Do yourself a favor make a choice again I give you 3 choices 

1)  install sdk on Ubuntu box build code for ARM and PRU use SDK Linux on SD card to test this example code.

Or

2) use Debian and cookbook examples.

Or best solution in my opinion

3) do #1 first 
get it working  correctly
buy 2nd SD card put Debian on it
then try the binaries you build from #1 on Debian 



Everybody asks to mix both solutions your asking for trouble and wasted time trust me. 

Always Use the recommended linux host to build SDK or Debian and the SDK. 

To ignore instructions that describe host then ask for help is insane 

If you want grief use Fedora virtual machine on win 3 pc  to build Debian and ask in this group why it doesn't work. You will get silence or be sent in 100 directions at once 😥🤣😄😀


Mark 




--
For more options, visit http://beagleboard.org/discuss
---
You received this message because you are subscribed to the Google Groups "BeagleBoard" group.
To unsubscribe from this group and stop receiving emails from it, send an email to beagleboard...@googlegroups.com.
To view this discussion on the web visit

Mark Lazarewicz

unread,
May 1, 2021, 11:20:41 AM5/1/21
to beagl...@googlegroups.com
<In addition, the bigger problem is actually rpmsg_pru30 issue. That literally <prevents me from running any rpmsg functions. As Dennis mentioned, this <might be a device tree problem? I need to spend some time on this. 

Which Linux are you using on ARM when you see this error ?

 I also sent another earlier reply trying  explain this was likely to happen if you are not running the SDK linux on ARM.

Can you confirm what you are running on ARM please?

Mark Lazarewicz

unread,
May 1, 2021, 12:46:03 PM5/1/21
to beagl...@googlegroups.com
<<<My first intention is to post in TI E2E support forums, but it requires a company email to do so. 

Hello Cheng 

I want to Help you  but it appears my message is lost in translation

What it sounds like to me  is TI pays these engineers $$ to answer questions and does not want to waste time and $$$ with users that dont follow their well written instructions which say use SDK Yocto Linux on the ARM for this example.

For me I start with a working example with instructions and documentation then modifyit

If I undertsand correctly you have never had an example working

If you like breaking examples and working on projects that ONLY works from a unix script and hides all the details then this is the correct group to to ask questions (-:

I suggest you try the example you found  following the intructions exactly.  if you cant or wont do that switch to DEbian/Cookbook

But if you do the latter I suggest don't change things follow the instructions

What is interesting is a Linux C application program should work correctly if it was coded generic especially when both Linux variants are for the same chip. Trying to figure out what is different between Linux variants to me is not productive its not my focus for you maybe it is.

Perhaps the Linux in the SDK was configured differently which implies it handle PRU slightly different Im not surprised by this.

Perhaps you can find what's different that may be a valid approach that works for you and maybe someone can help. I think Dennis gave you a good clue.

I will watch the thread hopefully I can be of help should you choose to follow the path E2E and the SDK layed out for you

Cheers

Walter Cromer

unread,
May 3, 2021, 11:24:23 AM5/3/21
to BeagleBoard
I just went through this pain and the people in this group were awesome help.

I needed to read three analog inputs and it was a bear but once I got it, it has been going well.

You don't mention the OS of your development platform or I may have missed it.  I'm on a Windows 10 box and using a BBB Wireless.  TI's entire learning system for this expects a Linux development platform so it's not as useful as it could be if you are on Windows.  I didn't want to go to the trouble of even bringing up a virtual Linux on my Windows box for this.  I did install Code Composer Studio (CCS) from TI, but gave up on it.  There's no easy way to transfer your compiled firmware to the BBB from within it according to TI.   So I just do everything on the Beagle and it meets my needs.

I use the cloud9 IDE that ships with the BeagleBones for coding both the ARM and PRU code.  This environment compiles the ARM-side code and executes it just like any normal host (aka Linux, aka ARM) program just fine.   However, to compile the PRU code and load it I go over to a PUTTY session and use the make files from Mark Yoders' PRU Cookbook (https://markayoder.github.io/PRUCookbook/) .  My process is clunky but my programs aren't huge or extremely complex (yet) and this works for me.  I don't have and don't want to invest in a debug probe so debugging the PRU code can be a pain and slow.  The only option really is to use RPMsg almost like printf to send messages back at key points in the PRU code to let me know where it is executing and what's happening.  (Purists and more advanced developers are barfing and laughing at me right now I am sure.)  

I strongly encourage you to get the Technical Reference Manual for the TI ARM335x and specifically spend time with the Touchscreen Controller chapter.  If you need precise timing, you'll want to spend time in the Industrial Ethernet Peripheral chapter too.   These are powerful tools once you understand them.

Also, get the PRU Optimizing C/C++ Compiler manual from TI and go through it.  One thing that really tripped me up was their implementation of the __far keyword.   GCC doesn't recognize that and I didn't understand what was going on at first.  

As Mark commented, some people encourage using remoteproc, some encourage using libpruio and some still use the uio. TI supports remoteproc and I expect them to be around so I went with remoteproc.  It may have its drawbacks but it is working just fine for me.  I can't comment on the other two as I have not tried them.   Also, I've found the TI E2E forum's moderators to be patient with me as a neophyte.   But the Google group's members do have wider experience.
   
FYI - watch out for how TI puts some register settings that cross nibble or byte boundaries.  It can really bite you!  Take a look at the STEPCONFIGn registers implementation of averaging to see what I mean.

I hope this helps!

Mark Lazarewicz

unread,
May 3, 2021, 1:53:50 PM5/3/21
to Walter Cromer, BeagleBoard
#  I did install Code Composer Studio (CCS) from #TI, but gave up on it.  There's no easy way to #transfer your compiled firmware to the BBB from #within it according to TI.   

Hi Walter

This doesn't look correct or sound like TI.
JTAG loads code extremely fast especially on the ARM. If you're referring to PRU code I've also done that as well.

Your overall  approach is fine imo just slow and a work in progress and yes they are helpful in E2E .

CCS/ JTAG is for barebones/ RTOS ARM it will speed up PRU development.

 unfortunately again in  my opinion a complex control loop would run on a DSP this PRU is too resource limited.

It's purpose is offloading 

Glad to see your making progress 

I don't know if you saw a comment I completed your project all on the ARM side barebones very quickly unfortunately I don't recommend this for a beginner. To attempt this  you really need a low level understanding of ARM so your current approach is probably your best choice







Walter Cromer

unread,
May 3, 2021, 3:34:56 PM5/3/21
to BeagleBoard
It was disappointing, to say the least that they said there wasn't an easy way to transfer the firmware but I believe that's what they meant.   

I had my application running on the ARM side just fine but we just couldn't get the deterministic timing we require for our application.  That's why I went to the PRUs.    We're getting awesome results on the timing now and RPMsg is fast enough on the transfer to get us the data for analysis that we need.  Right now my problem is that I need to put 10 to 15 samples in an array and do some basic slope and standard deviation calculations on the values in the array but when I add that, I'm getting that my program won't fit into memory.  I'm working now to learn how to initialize all my variables in the shared memory space.  That's locking up the BBB every time right now.   I'm following Molloy's book but it's not getting me there so far. 

Mark Lazarewicz

unread,
May 3, 2021, 3:44:10 PM5/3/21
to Walter Cromer, BeagleBoard
Walter

Yes I remember everything about your application including the Debian  ARM Linux application delays nobody seemed to have answers on fixing.

Your running windows 10 not using the SDK using cloud 9 and Debian  as I understand. What is E2E saying about your compiler errors your asking about in this group???

Mark

Walter Cromer

unread,
May 3, 2021, 4:02:25 PM5/3/21
to BeagleBoard
I'm using Debian and Cloud9 running on the Beagle.  The Windows 10 box is really just providing a browser 'terminal' to the Beagle.

What compiler errors?  Do you mean __far?   I don't think I asked about that on E2E.  I just realized my mistake and did some reading to understand it.  It's not an issue anymore.

Mark Lazarewicz

unread,
May 3, 2021, 8:29:03 PM5/3/21
to beagl...@googlegroups.com
Walter

This below code you posted in another thread was what I was referring to. Does the E2E forum support cloud 9 dev on BBB??? 


I'm also curious how you actually build/modify your Linux kernel with no Linux box.

Walter Cromer
changed the code to this and get the same error.

fd = open ("/dev/mem", O_RDONLY | O_SYNC);   // not sure O_SYNC is necessary since this is a read-only open situation

Servo tester!!!
ERROR: could not open /dev/mem.

make: *** [/var/lib/cloud9/common/Makefile:173: start] Error 1
rm /tmp/cloud9-examples/pwm-test.o




On Mon, May 3, 2021 at 3:02 PM, Walter Cromer

Walter Cromer

unread,
May 4, 2021, 9:35:31 AM5/4/21
to BeagleBoard
I haven't posted the question about this code to the E2E forum as I think it must have something to do with permissions.

I don't know if they support Cloud9 on E2E.  I haven't asked any questions specific to Cloud9.

And, I don't do any kernel development.  It's above me right now.  I'm grateful for all those who do work in this space!
But I write all my code directly on the BBB using cloud9.  I guess until recently I didn't look for any other way to do it. 

Walter

Cheng Chen

unread,
May 4, 2021, 11:33:51 AM5/4/21
to BeagleBoard
Hi Walter,

Thank you so much for the reply. I think my setup is exactly the same as what you have (win10 and BBB wireless). That really motivated me to see somebody else successfully run the system with the same setup.
I actually made some preliminary progress after two nights brooding and I am able to read out ADC data through rpmsg. Thanks a lot for your tips.

I think the problem is still in the makefile and environment. As you mentioned, you are using makefile from PRUcookbook while I started off with TI's built-in makefile.  I believe there is a couple of slight difference between Debian system and TI SDK environment which turns out to be critical in compiling. I carefully compared the difference between two makefiles and created one which is close to the makefile in the PRUcookbook. That works like a charm.

Next step I also want to explore precise timing and see how fast the adc can run. May I ask what is the speed you are reading out from ADC?

Regards,
Cheng

Cheng Chen

unread,
May 4, 2021, 11:45:57 AM5/4/21
to BeagleBoard
Hi Mark,

Thanks a lot for the detailed explanation regarding the SDK. I actually made the examples work in the end. Like you said, there is difference between SDK Linux and Debian Linux. My approach is similar to what Walter refers. My understanding now is examples from TI needs to be modified properly (mainly makefile) so that it can be applied to the Debian system without any problem. I haven't finished going through the reference you provided, but they have been tremendous help.
Thanks again!

Regards,
Cheng

Walter Cromer

unread,
May 4, 2021, 4:02:38 PM5/4/21
to BeagleBoard
I am glad to have helped a little bit.   Stick with it.  When it clicks and you start really moving forward I think you'll be pleased.

Can you share the changes to the Makefile that you made?  I'm curious if it might help me.

Right now I am reading the ADC every 2.7ms.  I'm reading three channels.  I include the step id too and check that.  I am using switch-case to check the step and put the analog input into the right variable in my code.   It might be slighter faster with an if-elseif but I like the neatness of the switch-case and until it causes me timing problems, I'm sticking with it.

I am also fairly certain the ADC can be read faster.  I have the ADC in one-shot mode and delay before I kick it off again.   I've also run this with no averaging, 4 averages, and 16 averages and it makes very little difference in timing for me.

Walter

Mark Lazarewicz

unread,
May 4, 2021, 5:20:46 PM5/4/21
to beagl...@googlegroups.com
Cheng

Yes difference between the two Linux is why the E2E wants to know which Linux you running.

Walter

Here is a shared RAM CCS JTAG PRU discussion but as Cheng stated the user is using SDK Linux.

Perhaps the code examples will help you solve your freeze up. It's possible ARM Linux is using that memory.

Looks like the TI PRU expert is very helpful as long as your using the SDK.

Jeff Andich

unread,
May 4, 2021, 8:24:15 PM5/4/21
to beagl...@googlegroups.com
Very informative thread guys!

Interesting article in E2E on accessing shared memory and RPMsg..

This statement by Jason Reader 

“…Also, if you are testing an RPMsg project you will need to use the pruss_remoteproc Linux module to load the code and not CCS over JTAG. …”

could be a clue as to the issue I was having attempting to load the C66 with JTAG after remoteproc initially loaded the RPMsg example project.  The stock example program was referred to in a post a couple of months ago  ( ). If I allowed remoteproc to load the C66 example program, and downloaded only the symbol file for DSP1/C66 via CCS/JTAG, I was able to set and reach breakpoints.  However, if I disabled remoteproc’s loading of the exe and attempted to load the entire C66 executable via JTAG, I got an error message ( from memory, remoteproc ID not found).  I will reproduce and post up the detail  in a separate post…




You received this message because you are subscribed to a topic in the Google Groups "BeagleBoard" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/beagleboard/W5jMCO3fu84/unsubscribe.
To unsubscribe from this group and all its topics, send an email to beagleboard...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beagleboard/1105921283.166452.1620163222532%40mail.yahoo.com.

Mark Lazarewicz

unread,
May 4, 2021, 11:43:31 PM5/4/21
to beagl...@googlegroups.com
Hi Jeff

Hope all is well good to here from you 

Yes saw that I think that's why someone told Walter there's no way to get code into PRU with CCS. 
It's obvious that the poster was using CCS fine to debug PRU code it's only the examples with RPMSg where JTAG is blocked. This to me means  the ARM has a Lock on some shared resources. I've also seen comments take the Linux SD card out🤔 when using CCS JTAG. 

The PRU resources are so limited I'm thinking any PRU  application would need to remove RPMSG after the application is working. it's really a glorified printf debugger .

In the event you see PRU freezes or crashes JTAG is invaluable and you could remove the Rpmsg from the code while debugging.


Back to your  DSP comments Jeff. 

On OMAP4 the key to debugging the DSP with CCS was a gel file that basically holds ARM off from interfering. If you look at the link I provided and re-read the user provides his gel script in theory if your PRU or DSP code needed no muxing done  by the  ARM you debug your DSP or PRU code with CCS and JTAG and you dummy up input Data from ARM but you hold the ARM in reset while you debug the PRU or DSP. We did that on a OMAP L 138 the DSP ran the Matlab control theory loop. DSP can access DDR and a big shared SRAM.

Once our ARM DSP IPC all worked you really can't debug one side it's like halting a wifi transmission in the middle the communication will fail right.

My point is you hold off the ARM with a gel script hold it in reset then load a gel script that initializes the DDR Ram and loads the code and starts the DSP. The problem these gel files are inhouse TI  tribal knowledge or seem to be lost between CCS version.

A good example is the Am335x SDK has those same example you were running on 57x with a PRU CCS JTAG tutorial. I went through that on a Beaglebone white and CCS 6.0 the PRU_CAPE. Gel file was flakey I had no Cape. It was good enough to re learn CCS 6 on am335x on windows 7 but I quickly punted I'm running CCS 10 on windows 10 on the Beaglebone white which has JTAG built in. But get this they abandoned Starter ware they folded it into the SDK Linux PDK the .org page says starterware is supported it's not. It also says that you can use the Linux  SDK on any version of Linux so I think Cheng started installing it on the Beaglebone bone yikes.

This am335x PRU is very limited. We used one PRU for a UART for DSP and the other PRU was an LCD segment controller.

We also a TI 28xx quadrature encoder the DSP read over SPI. The clock on these DSP is way faster than a PRU. The DSP control loop ran ever 100 nS. 

This PRU at 200MHZ that's 50 instructions in assembler it couldn't work as hard as a DSP.

Walter has a fairly simple loop he had working on ARM side reading ADC but he was getting unacceptable delay so he's learning PRU.

I don't know my Beaglebone white really isn't supported anymore I've got a barebones ARM ADC running on it I'm unwilling to give up JTAG and CCS I do have several black's and the JTAG sockets but don't feel soldering ambitious.

 my Am335x SK EVM already runs SDK Linux has JTAG onboard and has more memory like the black. Looks like I was building uboot on a Linux Ubuntu laptop  I definitely remember building SDK  Linux kernel and running defconfig and finally learned how to add remove module's and features from the kernel that was in 2015. 

I guess it's better late than never my only concern is that pin muxing of that SDK kernel has to be outdated but I'll take the risk if I can build my Kernel's source's. You can't deliver a product if you can't rebuild source code right Jeff??? Still there.🤔

On  barebones ARM pinmuxin depending on what peripherals you need on your project that's why starterware is no longer supported it becomes ifdef nightmare but the c code is obvious.

 it's the cache, MMU code that's complex then fold in open source TCP/IP wow no way you going to get a barebones ARM project going in less a year's using console uart printfs. 

That FTDI UART/ JTAG onboard  the devboard the hardware guys were old school they definitely knew what someone doing low level development needed. That was the Beaglebone white and the $200 am335x SK 

The black ditched the JTAG onboard and doubled the RAM.

Unfortunately everyone using anything after the bone  black hasn't hair Left 🤣 because they spend their Life waiting on printf from DSP on x15 and PRU. 

Win 10 CCS10 Black hawk x15 your blazing Jeff unfortunately I don't see anybody beyond you in this group seeing the value of CCS. 

Most of these guys on this group  are ARM Linux application guys they don't need CCS for that. And anybody else using the  X15 DSP will probably use RPMSg. 

I want to stay positive but I can't make a $$$ investment in any .org board again that doesn't have a working JTAG connector on it so if I want to do DSP I gotta go backwards to my OMAP L138 board.

For you I can check out all the .gel that came with  CSS 10 as well as what gels come in the 57x SDK you should not have any problem loading up DSP CCS code. Pull the Linux SD card out on x15 I mean you're learning signal processing on 67x right. CCS and JTAG are the  default TI DSP tools for at least 10 years



Mark

Cheng Chen

unread,
May 6, 2021, 6:28:45 PM5/6/21
to beagl...@googlegroups.com
Hi Walter,

Sorry for the late reply.
The most important part I modified for TI's makefile is to make sure that the firmware is loaded into remoteproc1. I basically replaced the loading procedure in the shell script with Molly's version which I attached below. I also added the entire include file and modified some of the constants and variable names in the c code because compiler reports errors of unrecognized header file and variables. But except those, it runs pretty well.

start:
ifneq ($(PRU_DIR),)
@echo write_init_pins.sh
@$(COMMON)/write_init_pins.sh /lib/firmware/$(CHIP)-pru$(PRUN)-fw
@echo "-    Starting PRU $(PRUN)"
@echo start > $(PRU_DIR)/state
else ifeq ($(PROC),tidl)
ti-mct-heap-check -c
sudo mjpg_streamer -i "input_opencv.so -r 640x480 -d /dev/$(shell fgrep -v vpe /sys/class/video4linux/video*/name | perl -ne '/\/(video\d+)\/name/ && print $$1') \
--filter ./$(TARGET)$(EXE)" -o "output_http.so -p 8090 -w /usr/share/mjpg-streamer/www"
else
./$(TARGET)$(EXE)
endif


Regards,
Cheng


Walter Cromer <wal...@edenconceptsllc.com> 于2021年5月4日周二 下午4:02写道：
You received this message because you are subscribed to a topic in the Google Groups "BeagleBoard" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/beagleboard/W5jMCO3fu84/unsubscribe.
To unsubscribe from this group and all its topics, send an email to beagleboard...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/beagleboard/ca296f29-c476-45fc-906d-e24595453270n%40googlegroups.com.

Mark Lazarewicz

unread,
May 8, 2021, 7:52:41 PM5/8/21
to beagl...@googlegroups.com
Hello Cheng

I learned a few things this weekend I thought I would share

The PRU  Labs examples 1 to 3 can be loaded with CCS and JTAG from Windows 10 

You can even debug both PRU0 and PRU1 at the same time examine memory and use HW  uart debug to speed up development

The RPMSG example LAB  must be loaded by the linux or TI RTOS drivers running on ARM side

I have win 7 and Windows 10 CCS/JTAG installations working one  a Beaglebone White and the other AM335X SK 
BBB is also supported 

My Last step is building the SDK  linux kernel from scratch  with rproc kernel modules loaded  from a VirtualBox Ubuntu VM on Win 7

The Linux  SDK has binaries for the ARM side so a  Native Linux BOX is NOT REQUIRED(but recommended )

CCS DOES NOT require linux to load and debug PRU 

For those that want to learn ARM TI RTOS Win 10 is required to build

The SDK documents I saved as the wiki is disappearing are awesome I found some very detailed PRU documentation that talks about everything from running TI PRU firmware On PRU ICCS for complex Industrial protocols as well a custom PRU code

The facts about clock cycles also were discussed. All of it in can be found by following the documentation tar ball I sent I you.

I think too many people panic dont want to run SDK Linux or use CCS /JTAG dont read through all the documents as they dont want to go that route.

Thats fine but the basic fundamentals of the whole SOC are then missed leading to confusion 

The SDK has done an excellent job of documenting this unfortunately unless your actually using the SDK all of this is hidden and I guess they are taking down some wikis so kind of sad this data will be lost.

The approach in this group is wonderful especially for Linux App types that don't care about details
 they just want a working kernel and actually making one script to build linux makes sense as supporting keystroke errors and inability to read docs in a chronological orderly way would be a nightmare.

So in summary In my humble Opinion if your goal is writing Linux apps on a open source board the SDK probally is NOT for you.

If your goal is barebones, TI  RTOS , board bring up, custom hardware , DSP programming Cortex M4 programming , advanced PRU apps or learning or adding Linux Device drivers the SDK is a great asset.

Mark



Mark Lazarewicz

unread,
May 9, 2021, 9:51:42 PM5/9/21
to 'Mark Lazarewicz' via BeagleBoard
Update AM335x SDK evaluation

Sucess!! rebuild kernel sources in VBOX  next will test my LInux on ARM on a board with RPROC drivers  and lastly revist CCS and JTAG in Ubuntu VBOX as an option I prefer CCS on Windows but do recall CCS Linux is needed for Linux debugging

Final eval steps use CCS to measure PRU instruction timings following the Labs


 OBJCOPY arch/arm/boot/zImage
  Kernel: arch/arm/boot/zImage is ready
mark@mark-VirtualBox:~/ti-processor-sdk-linux-am335x-evm-06.03.00.106/board-support/linux-4.19.94+gitAUTOINC+be5389fd85-gbe5389fd85$





Walter Cromer

unread,
May 10, 2021, 9:59:16 AM5/10/21
to Beagle Board
Mark makes some excellent points.  I started developing C on the ARM side under Linux before needing to start using PRUs.  I had not spent much time in the TRM while working on the host side because I didn't need to.  It really bit me until I did.  Once I really dug into the TRM and started to understand the SOC at a hardware level, things began to progress much better and faster.  And, going through the PRU Optimizing C/C++ Compiler manual and PRU-ICSS/PRU_ICSSG Getting Started Guide on Linux helped too.   What DID NOT HELP was that TI has not finished moving the wiki pages over to a new resource but at least they made them available for download. 

And, finally, this group of people have been awesome to help!


Virus-free. www.avast.com

To view this discussion on the web visit https://groups.google.com/d/msgid/beagleboard/1205556611.1038015.1620517942188%40mail.yahoo.com.


--
Thanks, 

Walter Cromer
Chief Idea Officer & Founder
Eden Concepts LLC

Live. Grow. Harmonize.


The information contained in this e-mail message may be privileged and confidential. If the reader of this transmission is not the intended recipient, you are hereby notified that any dissemination, distribution or copying of this communication is strictly prohibited. If you have received this communication in error or are not sure whether it is privileged, please immediately notify us by return e-mail and delete or destroy any copies, electronic, paper or otherwise, which you may have of this communication and any attachments.
 

Virus-free. www.avast.com

Cheng Chen

unread,
May 11, 2021, 12:28:41 AM5/11/21
to BeagleBoard
Hi Mark, 

Thanks a lot for the updates. I went through the SDK documents and I actually got a lot inspiration. Thanks for that. 
I bought an TMDSEMU110-U for debugging. That is recommended from TI tutorials of PRU debugging: PRU is connected with TMDSEMU110-U and then to the laptop. Is that how you debugged with JTAG ? I feel this approach is a little cumbersome since I need to set up a lot of environments manually. But it worked, so I just put up with it for now. 
I think both SDK and Beaglebone has a lot of interesting stuffs worth being explored. Appreciate your help!

Regards,
Cheng

Mark Lazarewicz

unread,
May 11, 2021, 3:39:36 PM5/11/21
to beagl...@googlegroups.com
HI Cheng

I have two Beaglebone White and the am335x SK both have JTAG on the board so no soldering. 
The key is to follow the labs 1 to 3 only dont use any RPMsg examples 
The other key is the PRU gel script is crucial there is a typo error in the instructions about correct .gel file name and how to execute it 
I used CCS 6.0 win 7 and CCS 10 on windows 10

Id be glad to help if I can 

The power of the CCS/JTAG is reading memory locations and finding crashes quickly its worth the learning curve

Any issues please ask but tell me your CCS version 

What I dont like about the PRU  tutorials is all use Linux interaction as in interrupts and they assume SDK linux not Debian

Its some work but you could use Windows to create a SDK linux SD card for JTAG if problems

 I have several jtags  one is USBV2 and I have yours I also have two BBB but even being EE I dont solder

Let me know be glad to help trust me after you master CCS/JTAG you will be very happy and have an awesome tool in your belt

Mark


Mark Lazarewicz

unread,
May 11, 2021, 7:01:17 PM5/11/21
to 'Mark Lazarewicz' via BeagleBoard
Hello Walter and Cheng

I spent 2 hours total doing following

1)Created Ubuntu 16.04 LTS  Dev Box using VBOX on Win 7
2) Created SDK linux SD card using image writer on Windows booted Linux on Beaglebone White 
3) Rebuilt kernel sources created images of customized kernel and replaced these on SD card 

I have a full development environ Now !!!!!

Next step add a kernel driver I wrote and understand how use the OCP  shared RAM to get large amounts of data from PRU to linux

***************************************************************

 _____                    _____           _         _
|  _  |___ ___ ___ ___   |  _  |___ ___  |_|___ ___| |_
|     |  _| .'| . | . |  |   __|  _| . | | | -_|  _|  _|
|__|__|_| |__,|_  |___|  |__|  |_| |___|_| |___|___|_|
              |___|                    |___|

Arago Project http://arago-project.org am335x-evm ttyS0

Arago 2019.11 am335x-evm ttyS0

am335x-evm login: [  143.131162] NET: Registered protocol family 15
[  143.545960] Initializing XFRM netlink socket

 _____                    _____           _         _
|  _  |___ ___ ___ ___   |  _  |___ ___  |_|___ ___| |_
|     |  _| .'| . | . |  |   __|  _| . | | | -_|  _|  _|
|__|__|_| |__,|_  |___|  |__|  |_| |___|_| |___|___|_|
              |___|                    |___|

Arago Project http://arago-project.org am335x-evm ttyS0

Arago 2019.11 am335x-evm ttyS0

am335x-evm login:



Jeff Andich

unread,
May 12, 2021, 9:23:41 AM5/12/21
to beagl...@googlegroups.com
This thread doesn’t appear to appear on the revamped beagleboard forums.. correct me if I'm wrong..

Should it be added retroactively or should we create a thread on the new forums with a link to this?

Thanks!


Reply all
Reply to author
Forward
0 new messages