[U-Boot] GT-64120 PCI init

[Richard Danter]

2008/8/12 Richard Danter <radanter at googlemail.com>:
> 2008/8/8 Richard Danter <radanter at googlemail.com>:
>> I now have a driver that initialises the 64120 and U-Boot sees the
>> bridge and Ethernet card. So far so good! But the Ethernet card can
>> not ping anything on my network.
>>
>> This is what I see...
>>
>> --8<--
>> U-Boot 1.3.3-svn19 (Aug  8 2008 - 15:32:31)
>>
>> Board: PPMC MIPS
>> DRAM:  64 MB
>> Flash:  4 MB
>> In:    serial
>> Out:   serial
>> Err:   serial
>> Net:   i82559#0
>> => pci
>> Scanning PCI devices on bus 0
>> BusDevFun  VendorId   DeviceId   Device Class       Sub-Class
>> _____________________________________________________________
>> 00.00.00   0x11ab     0x4620     Bridge device           0x00
>> 00.07.00   0x8086     0x1030     Network controller      0x00
>> 00.08.00   0x1011     0x0026     Bridge device           0x04
>> => printenv
>> baudrate=9600
>> ethact=i82559#0
>> ipaddr=192.168.1.84
>> netmask=255.255.255.0
>> stdin=serial
>> stdout=serial
>> stderr=serial
>>
>> Environment size: 113/32764 bytes
>> => ping 192.168.1.1
>> i82559#0: Tx error buffer not ready
>> ping failed; host 192.168.1.1 is not alive
>> --8<--
>>
>> My guess is that there is probably still something I did not do in the
>> PCI init code, but I have looked at as many examples as I can find and
>> I can't see anything amiss. Any thoughts on what would case the buffer
>> not to work even though the Ethernet card itself is seen and appears
>> to be detected correctly?
>
> OK, I think I figured this out but still not sure what the correct
> solution is...
>
> The tx_ring and rx_ring buffers are being placed in SDRAM accessed via
> kseg0 (cached) at 0x8nnnnnnn. After trying to ping and getting the
> timeout I see the following at this location...
>
> 83ff8d30: 00000240 308dff03 16080001 00002203    ... at 0.........".
> 83ff8d40: 01002e00 60088868 0040f284 31050000    ....`..h. at ..1...
>
> But if I look at the same physical address via kseg1 (uncached) at
> 0xAnnnnnnn then I see...
>
> a3ff8d30: 00a00240 308dff03 16080001 00002203    ... at 0.........".
> a3ff8d40: 01002e00 60088868 0040f284 31050000    ....`..h. at ..1...
>
> I think the 00a0 at the start is the correct value (Tx done). Can
> anyone confirm this?

Confirmed myself.

Solution so far is either to run U-Boot from KSeg1 or to pepper the
eepro100 driver with flush_cache() calls. Both seem to work though
running in kseg1 is slower.

Has anyone else used an eepro100 on a MIPS target? Did you hit the same problem?

I can see that some drivers (e.g. rtl8139.c) already have the
flush_cache() calls.

Rich


>
> If so then the problem seems to be that I need to read/write the Tx/Rx
> buffers via kseg1 and not kseg0. But in general I want as much data
> and code in kseg0 as possible to get the best performance.
>
> What is the correct way to do this?
>
> Thanks
> Rich
>