Troubleshooting the V80 / V80LL card
This section provides suggested solutions to common problems that are faced when working with the V80 or V80LL. Though only the V80 is referenced in the solutions below, they all apply to both the V80 and V80LL cards.
General troubleshooting
Before trying any other solutions, it's always good practice to power down the host machine and then power it back up again. This can solve configuration issues related to PCIe enumeration and can solve many other problems.
Problems programming the V80 over PCIe
The processes below assume that the V80 has already been successfully programmed with a Vollo image at least once,
either via JTAG or PCIe. If you have not yet programmed the card with a Vollo image, please follow the instructions
in Programming the V80 via JTAG.
If you are having trouble while programming the V80 card over PCIe, then the first step is to ensure that all of the card's functions are bound to the Vollo driver:
sudo $VOLLO_SDK/load-kernel-driver.sh vfio
If AMD's ami kernel driver from an earlier release is loaded, it holds the card's management function and
vollo-tool cannot program the flash until it is unloaded with sudo rmmod ami. Then run
load-kernel-driver.sh again.
The driver association with the V80 should now be visible with an lspci command.
lspci -vvvnn -s $BDF
If you don't know what your card's BDF is, it may be found by running the command below.
lspci | grep "Myrtle"
Running the full lspci command should display the V80 card loaded with a Vollo image as below. The card's BDF here is 01:00.
01:00.0 Processing accelerators [1200]: Myrtle.ai Device [1ed9:000a]
Subsystem: Xilinx Corporation Device [10ee:50b4]
Control: I/O- Mem+ BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
IOMMU group: 12
Region 0: Memory at c0000000 (32-bit, non-prefetchable) [size=256M]
Region 2: Memory at d0000000 (32-bit, non-prefetchable) [size=4M]
Region 4: Memory at fce0000000 (64-bit, prefetchable) [size=4M]
Capabilities: <access denied>
01:00.1 Processing accelerators [1200]: Myrtle.ai Device [1ed9:100a]
Subsystem: Xilinx Corporation Device [10ee:50b4]
Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
Latency: 0
IOMMU group: 13
Region 0: Memory at fcd0000000 (64-bit, prefetchable) [size=256M]
Capabilities: <access denied>
Kernel driver in use: pci-pf-stub
The last line shows the pci-pf-stub driver correctly associated with the 1st physical function of the board, 01:00.1.
This provides access to the OSPI programming functions available on the V80 via PCIe and the normal PCIe programming detailed in Programming the V80 over PCIe should be possible. If not, please continue with this guide.
Recovering the V80 from a "bad state"
The V80 card can sometimes enter into a "bad state". This can result in the card not functioning correctly despite following other troubleshooting steps, including powering down and powering up the host machine.
Such a failure might manifest as follows:
If, while running the PCIe programming commands found in step 2 of Programming the FPGA over PCIe,
vollo-tool reports that the card looks "wedged", then the V80 card is in a bad state. vollo-tool checks for
this before it writes anything to the flash, so the image on the card is unchanged.
The suggested solution for stateful failures like this is to:
- Power the host machine down
- Completely remove power from the host machine by either unplugging or removing the card from the host machine completely.
- Leave the host machine and/or V80 card unpowered for around 30 minutes.
- Reinsert the V80 into the host machine (if removed), reconnect power to the host machine.
- Power up the host machine again.
- Re-attempt PCIe programming following the steps provided in Programming the FPGA over PCIe.
This unorthodox solution has been found to work in many cases and is indeed the recommended solution to these problems as provided by AMD in their support forum.