FPGA, SoC, And CPLD Boards And Kits
FPGA Evaluation and Development Kits
5892 Discussions

[Arria10] (mSGDMA <=> PCIE_Hip_Txs) txs_waitrequest signal show "big delay" when reading operation

JET60200
New Contributor I
575 Views

hello expert,

On our own Arrria10 board, we use "PCIE gen3x8 with AVMM-DMA (64bit)" setting. 

Per our system specific scenario, we have to add "MSGDMA + TSE " for ethernet i/f.  But we met a "Txs_waitRequest_o deassert has too big delay " problem , detail as following :

hip_Txs_waitrequest_too_big_delay_(168_cycles)_.png

 

[1]  let's talk that marked number (3)  in above picture,  [Hip_Txs] connects to [TSE_subsystem -> dma_rx_mm_write] , which is  the ethernet packet receiving path. 

      Firstly this recv path work fine as expect.  From SigTap picture(below), we see per around ~10 clock cycles, "Txs_waitrequest" signal was deasserted to activate the mSGDMA write data through "Hip_Txs"  into Host memory space. thus ethernet were transfered from fpga tse sub-system into Host memory correctly, and the transferring performance is correct. 

mSGDMA_writedata.png   【2】unfortunately for the Ethernet XMit path, where [TSE_subsystem -> dma_tx_mm_read] connects to [Hip_Txs], the performance is very bad.  From SigTap picture (below), we found per around ~168 clock cycles,  "Txs_waitrequest" signal was deasserted to activate the mSGDMA read data through "Hip_Txs"  from HOST (x86 cpu). 【168 cycles】 is TOO BIG delay between every mSGDMA reading operation, thus the result is " the sending performance is very bad".

mSGDMA_readdata.png

 

[Question] :  why this big delay (~168 cycles) of "Txs_waitrequest" happens in mSGDMA reading direction?  while the reverse writing direction the "Txs_waitrequest" behaviour good ?  

 

 

[BTW],  we uses "AVMM-DMA" setting in our system.  We made an experiment to change from "AVMM-DMA"  to  "AVMM w/o DMA" in PCIe HIP core,  in this case both "mSGDMA writing" and mSGDMA reading works fine, that says the “~168 cycles delay" disappears, this maybe some clues for your expertised analysis.        

 

 

   

0 Kudos
3 Replies
JET60200
New Contributor I
557 Views

@SengKok_L_Intel

hi  SK,  you'r expert of A10 PCIe HIP Core, would you please help on this issue ?  

If need more debugging trace, we can provide.  

 

thanks very much~

0 Kudos
JET60200
New Contributor I
537 Views

Hi Intel,  we didn't receive any reply about the problem we reported, could any expert give an advice how to move forward?  we 're stuck really since our performance  test failed now.  Thanks in advance ~

 

cc @KhaiChein_Y_Intel ,

0 Kudos
JET60200
New Contributor I
516 Views

kindly reply

0 Kudos
Reply