liteeth questions

[Joel Stanley]

Litex team,

I've spent some time on the litex Ethernet driver, and with Benh's
help I've gotten to the point where it sends and receives some but not
all packets, and only crashes after a while.

Florent, I assume the write to READER_START causes the hardware to
take the data in the slot indicated by READER_SLOT and send it out?
How can software learn when we can tell the hardware that the other
slot has data?

I get the TX interrupt, which I assume is when a slot has finished
transmission. Do I need to keep track of which slot has most recently
been sent, or is there a way to ask the hardware?

Cheers,

Joel

--

Details:

Working:
- Interrupts
- Tx packets
- Rx packets
- tftp, wget and ping work

TODO:
- Tx on both queues
- Free Tx skbs so we don't run out of memory
- udhcpc and SSH
- MDIO support. We map the memory but don't attempt to talk to it

When you run out of memory, it looks like this:

[ 2561.590000] skbuff: skb_over_panic: text:c021c280 len:64 0
[ 2561.590000] BUG: failure at net/core/skbuff.c:104/skb_panic!
[ 2561.590000] Kernel panic - not syncing: BUG!
[ 2561.590000] ---[ end Kernel panic - not syncing: BUG!

[Tim 'mithro' Ansell]

Hi!

Was wondering how this was going? I believe that Florent replied off list?

Tim 'mithro' Ansell

--
You received this message because you are subscribed to the Google Groups "Linux for LiteX FPGA SoC" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-litex+unsubscribe@googlegroups.com.
To post to this group, send an email to linux...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/linux-litex/CACPK8Xdz1kYEJ%2B0KUXO1M-5WVsOKJGkARzU1dOM_%2Bx4etQ%3DpOQ%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

[Jan Schmidt]

Here's a small fixup patch that makes the litex eth driver stable for me.

Performance is not stellar, as one might expect on a 50MHz CPU. With
wget, I see about 380KB/sec.

Cheers,

Jan

[Tim 'mithro' Ansell]

Awesome work Jan! Could someone who understands the kernel review this patch?

In the terms of performance, I have a start of a proposal for a DMA engine for LiteEth here -> https://docs.google.com/document/d/16GfKhTI6OD07oObcxq31JeKApqx4q4YHZIcNfv63gQo/edit?usp=sharing

The primary idea is that you can construct a FIFO of memory read commands and then get interrupts when various points are completed. Really need some feedback from Kernel developers if this type of DMA interface would be useful for accelerating LiteEth.

We really want to keep this DMA as simple as possible but one possible extension is to calculate CRCs as part of the DMA transfer.

Tim 'mithro' Ansell

--

You received this message because you are subscribed to the Google Groups "Linux for LiteX FPGA SoC" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-litex+unsubscribe@googlegroups.com.
To post to this group, send an email to linux...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/linux-litex/e5d78fc3-afe0-4d79-36f9-850602d3fc7e%40noraisin.net.

[Tim Ansell]

So ao

Tim 'mithro' Ansell

--
You received this message because you are subscribed to the Google Groups "Linux for LiteX FPGA SoC" group.
To unsubscribe from this group and stop receiving emails from it, send an email to linux-litex+unsubscribe@googlegroups.com.
To post to this group, send an email to linux...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/linux-litex/CACPK8Xdz1kYEJ%2B0KUXO1M-5WVsOKJGkARzU1dOM_%2Bx4etQ%3DpOQ%40mail.gmail.com.

[Olof Kindgren]

Den tis 30 jan. 2018 19:14Tim 'mithro' Ansell <m...@mith.ro> skrev:

Awesome work Jan! Could someone who understands the kernel review this patch?

In the terms of performance, I have a start of a proposal for a DMA engine for LiteEth here -> https://docs.google.com/document/d/16GfKhTI6OD07oObcxq31JeKApqx4q4YHZIcNfv63gQo/edit?usp=sharing

The primary idea is that you can construct a FIFO of memory read commands and then get interrupts when various points are completed. Really need some feedback from Kernel developers if this type of DMA interface would be useful for accelerating LiteEth.

That sounds very much like a chained DMA (or a scatter/gather DMA). I believe there should be kernel infrastructure for that but I haven't checked

//Olof

We really want to keep this DMA as simple as possible but one possible extension is to calculate CRCs as part of the DMA transfer.

Tim 'mithro' Ansell

On 30 January 2018 at 10:00, Jan Schmidt <tha...@noraisin.net> wrote:

Here's a small fixup patch that makes the litex eth driver stable for me.

Performance is not stellar, as one might expect on a 50MHz CPU. With
wget, I see about 380KB/sec.

Cheers,

Jan


On 28/12/17 11:25, Joel Stanley wrote:
> Litex team,
>
> I've spent some time on the litex Ethernet driver, and with Benh's
> help I've gotten to the point where it sends and receives some but not
> all packets, and only crashes after a while.
>
> Florent, I assume the write to READER_START causes the hardware to
> take the data in the slot indicated by READER_SLOT and send it out?
> How can software learn when we can tell the hardware that the other
> slot has data?
>
> I get the TX interrupt, which I assume is when a slot has finished
> transmission. Do I need to keep track of which slot has most recently
> been sent, or is there a way to ask the hardware?
>
> Cheers,
>
> Joel
>

--

You received this message because you are subscribed to the Google Groups "Linux for LiteX FPGA SoC" group.

To unsubscribe from this group and stop receiving emails from it, send an email to linux-litex...@googlegroups.com.

To post to this group, send an email to linux...@googlegroups.com.

To view this discussion on the web, visit https://groups.google.com/d/msgid/linux-litex/e5d78fc3-afe0-4d79-36f9-850602d3fc7e%40noraisin.net.

For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "Linux for LiteX FPGA SoC" group.

To unsubscribe from this group and stop receiving emails from it, send an email to linux-litex...@googlegroups.com.
To post to this group, send email to linux...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/linux-litex/CAHLUNMyASrq040BiwC22Lcty99t9JZQ5K1bkoWtrH5o-JYrNjg%40mail.gmail.com.