MAC-NET not sending eth frames

[hmijail]

I am trying to use the MAC-NET core in the Vybrid uC in a TWR-VF65GS10 board (connected to a TWR-SER2 board). And I believe I have done everything that should be done to be able to send a random ethernet frame:

• clocks ungated and multiplexers set to use the external RMII clock,
• IO pins configured and multiplexers set,
• FIFO descriptors and buffers created and aligned with better-than-recommended alignment,
• disabled ENET, set registers to my liking (store&forward, CRC added by MAC-NET, RMII mode)
• initialized the PHY and set its registers to my liking,
• got the PHY to autonegotiate and detect the link conditions forced by the other side and report good link,
• fed back the negotiated speed to ENET,
• re-enabled ENET,
• prepared an ethernet frame into one of the buffers, set its flags,
• and set ENET0->TDAR = ENET_TDAR_TDAR_MASK.

If I understand correctly, now DMA should take over and either send the frame or signal any error, and then set TDAR to 0.

But nothing is happening. TDAR stays ==1 indefinitely. No changes in EIR either. And Wireshark also doesn't see anything.

I have tried using enhanced buffers and requested interrupt flags to see if any error flags appeared on their status fields. Nothing.

I have tried checking what the BSP does, and I don't see anything I could be missing.

I have tried different alignment sizes, either the minimum required ones or much bigger than recommended.

I have tried ungating all the CCGR clocks just in case I was missing something..., setting the MII_SPEED in MSCR slower than necessary..., tried different sizes of ethernet frames, ... nothing happens.

U-boot in the accelerometer demo boot disk does manage to ping the network, so the hardware is functional.

The RAM I am using is set to be non-cached.

Does anyone have any idea of what could be wrong?

[hmijail · Accepted solution]

Just in case someone finds him/herself in this situation, my problem was that the MMU had been activated and I was passing virtual addresses to the MAC-NET. The moment I passed the physical addresses it started working.

A colleague says the Vybrid USB controller did something similar because of alignment problems, so in general, looks like the module-private DMA engines have this tendency to just sit there if there is any problem with their work parameters. For the MAC-NET case, I expected at the very least that the EIR[EBERR] bit would flag a problem with the DMA transaction (as its description says), but... no.

[karina_valencia]

timesyssupport can you attend this case?