Data lost in MessageQ

jkeegg

hi experts,

I have a problem about messageq useage on linux-c6x.

I created a MessageQ queue and a HeapMemMP on core 0, and also regist it on other cores.

When I use other cores alloc message then send it to core 0, the message can reach to core 0 successful, but the payload of message always get lost, about 10% can keep complete.

The strange thing is when I use MessageQ_getMsgSize(), MessageQ_getMsgPri() and MessageQ_getMsgId() to get the info I had set before, I could get the right size, pri and ID.

over 13 years ago

0 Rex Chang over 13 years ago

TI__Guru 50170 points

Hi, Jerry,

I am trying to understand your issue. Does it mean that you can get the size, ID, and pri correctly for every message, but only 10% of the payload?

Rex

0 Rex Chang over 13 years ago in reply to Rex Chang

TI__Guru 50170 points

Hi, Jerry,

Have you tried with the example in the Software Ecosystem Features of the linux-c6x wiki page? Does the example work for you?

Rex

0 jkeegg over 13 years ago in reply to Rex Chang

Prodigy 240 points

for example, I send message "Hello" from core 1 to core 0 ten times.

I can always get the right sender at core 0 side, but about 2~3 messages lost its payload "Hello"

0 jkeegg over 13 years ago in reply to Rex Chang

Prodigy 240 points

I did, but I found that example only have 2 cores communication at same time.

In my application, I need 8 cores can communicate with each other when I want.

0 Rex Chang over 13 years ago in reply to jkeegg

TI__Guru 50170 points

Hi, Jerry,

When the payload is lost, what exactly the symptom is? Are you seeing garbage data? What data are in the memory at address msg with msgQ header offset? In the syslink example, 1000 messages are sent and data integrity are checked too.

Syslink is on top of the IPC and is 1 to 1. For 8 cores communicating to each other, you will need to set up one syslink per pair.

Rex

0 jkeegg over 13 years ago in reply to Rex Chang

Prodigy 240 points

Data in memory at address msg with msgQ header offset are 0.

I did set up one syslink per pair.

0 Rex Chang over 13 years ago in reply to jkeegg

TI__Guru 50170 points

I assume your setup is core 0 running Linux and core 1 BIOS, correct? How do you check the data in the payload, debugger, printf, or other ways?

0 jkeegg over 13 years ago in reply to Rex Chang

Prodigy 240 points

Yes, I use printf to check the payload.

0 Rex Chang over 13 years ago in reply to jkeegg

TI__Guru 50170 points

Is this little endian and what platform is it, c6678? I'll see what I get when I print the payload out in the example code on the same platform.

Rex

0 jkeegg over 13 years ago in reply to Rex Chang

Prodigy 240 points

Yes, I had test that example before and run well, but I make something difference.

When the msg arrived in core 0, it was freed by core 0.

I do more trace in my code, and find the msg's addr always be 0x0c01c180 when I free it in core 0 then alloc new msg on other cores, in this situation, only the first msg alloced can keep integrity.

When I do not free the msg on core 0, the msg's addr increase, and the payload do not lose any more.

0 Rex Chang over 13 years ago in reply to jkeegg

TI__Guru 50170 points

Hi, Jerry,

Are you sending messages asynchronously? In our example, the sender frees the message upon receiving the message back from the receiver and after verifying the payload integrity. If the receiver frees the message, it will need to allocate a new message to send back to the sender, so it will free the buffer. If it is not the case, the sender will use the same buffer to send the next message.

Rex

Processors

Processors forum

Data lost in MessageQ