Tool/software:
Dear Expert
I have met a question while developing with am2732. I use the enet to send the point result to the PC and found there is a checksum error within a certain UDP packet and moreover certain bytes in the packet.I have checked my datapath,and disabled almost function in the c66x core and called a memset function in the end of datapath process.However,this problem still occur.Besides，I found the error data is bound to an certain address in the L3 memory space.I have adopt ping-pong buffer to store point result for even and odd frame respectively,and the error data is located in the ping buffer.Since the C66x only process the basic memset,so it seems this error is not caused by data overlay.
I have checked the cache，and there is no cache problem.The C66x execute the CacheP_wbInv function before send IPC to inform R5F0_0 core that the data is ready to be handled.The R5F0_0 core send the point result out by Enet using DMA,and there is no need to take Cache Process.
The enet takes about 30ms to transfer the data out,and I have print the data in the error data's address and found the data in L3 memory was not changed and is confilt with the captured udp checksum error packet.And below is the polling code.
uint8_t* scanRes = (uint8_t*)0x882ffa58;
for(loopCnt2=0;loopCnt2<10;loopCnt2++)
{
CacheP_inv((void*)scanRes, 64,CacheP_TYPE_ALL);
logger("%d times scan\r\n",loopCnt2);
for(loopCnt=0;loopCnt<16;loopCnt++)
{
logger("%d\t",scanRes[loopCnt]);
}
logger("\r\n");
ClockP_usleep(3*1000);
}
The address of error data is not changed When I changed the start address of Ping buffer.And in some cases,when I change the start address only forward or backward for 8 bytes,this error disappeared.
Please give me some advance on this issue.
regards
Zhiqiang Lin
