[Macchiato] [EXT] Network driver crash

[Jeremy Linton]

Hi,

On 03/06/2018 05:17 PM, Marcin Wojtas wrote:
> HI Jeremy,
> 
> 2018-03-06 22:08 GMT+01:00 Jeremy Linton <Jeremy.Linton at arm.com>:
>> Hi,
>>
>>
>>
>> So I was running Leif’s firmware (as well as a build of my own from early
>> January) and they were having the problems with the 1G port as well.
>>
>>
>>
>> I tossed the latest 4.16 DTB in the firmware build and that fixed the
>> network problem, except I also made the mistake of pulling forward the
>> edk2/openplatform/atf/marvell-binary tree’s too, and ended up disabling the
>> flash based UEFI variable store because It seems the mem mapping to the SPI
>> flash is somehow broken. Basically it spews “Firmware Volume for Variable
>> Store is corrupted”.
>>
> 
> Thanks for confirming the DTBs.
> 
> I updated marvell-armada-wip branch with 3 new commits:
> 150e1b4 - Marvell/Armada: Ensure proper access to memory mapped SPI
> e3c5b71 - Revert "Platforms/Armada80x0McBin: Use GIC interrupts directly"
> 61bf1b6 - Revert "Platforms/Armada80x0McBin: Restore old
> system-controller binding"

Sure, will do. Although, it might take me a couple days as I'm trying to 
test a couple other things here too.


> 
> I'd appreciate giving a try to boot from SD and checking the network
> in your setup.
> 
>>
>>
>> PCIe remains broken, as do the 10G ports, but I don’t need the USB->Ethernet
>> adapter anymore.
>>
>>
> 
> What do you mean by:
> 1. "PCIe remains broken" - kernel or firmware? What are the symptoms?

This is one of the older boards, where I soldered a resistor. Since that 
I can see the x1 GT710 in the PCIe config space, but accessing the MMIO 
mapping seems to either return 0's or sometimes the driver will crash 
accessing the board (this board/kernel config works fine in another 
machine using an edk2 based firmware).

> 
> 2. "as do the 10G ports" - ditto?

The 10gig ports seem to have something wrong with the phy, they appear 
in linux, but the link up events never happen, and I am unable to pass 
traffic.