Advice on connecting this code to a larger design

[cpitclaudel]

Hi there,

Thanks for putting this project together; it's great :) I've successfully programmed a BX with the sample echo device, and verified that it works nicely.

I'd like to get some advice on connecting this to a larger project. When I hook this up to my design, I observe all sorts of transmission errors (duplicated and missing data values, hangs). I'm guessing it's because of timing issues, as I get the following output from nextpnr:

Info: Device utilisation:
Info:            ICESTORM_LC:  5124/ 7680    66%
Info:           ICESTORM_RAM:    18/   32    56%
Info:                  SB_IO:     4/  256     1%
Info:                  SB_GB:     3/    8    37%
Info:           ICESTORM_PLL:     1/    2    50%
Info:            SB_WARMBOOT:     0/    1     0%

[…]

Warning: Max frequency for clock 'CLK$SB_IO_IN_$glb_clk': 23.41 MHz (FAIL at 48.00 MHz)
Warning: Max frequency for clock    'clk_48mhz_$glb_clk': 42.82 MHz (FAIL at 48.00 MHz)

Info: Max delay <async>                       -> posedge clk_48mhz_$glb_clk   : 2.48 ns
Info: Max delay posedge CLK$SB_IO_IN_$glb_clk -> posedge clk_48mhz_$glb_clk   : 37.36 ns
Info: Max delay posedge clk_48mhz_$glb_clk    -> <async>                      : 3.14 ns
Info: Max delay posedge clk_48mhz_$glb_clk    -> posedge CLK$SB_IO_IN_$glb_clk: 8.36 ns

I assume the communication issues stem from that; correct?

I have a toplevel clock that I use to run my main design, and as input to the PLL to derive the 48MHz clock. Is this how I should do it? Or should I divide down the 48MHz clock to get a 12MHz back from it? (I tried that approach and got to 46MHz).

Is there any way to run the USB module at a lower speed?

[davidthings]

Hi! Thanks for taking a look at the code.

First... I assume you are using NextPNR. Older code, like Arachne, won't work. Arachne doesn't optimize for clock time! NextPNR does.

One thing I should have emphasized in the doc is that the module doesn't like getting overrun. I always use it with a FIFO or two to try to avoid a situation where it gets blocked or overrun.

USB_SERIAL -> FIFO -> modules

USB_SERIAL <- FIFO <- modules

Also, if the whole design fails timing, the whole thing is in trouble. The usual symptom is that the USB port just fails to enumerate.

Finally I see that you're at 66% fullness. This is a biggish project for our beloved TinyFPGA BX. Can you try some smaller subsets to see if pure code magnitude is part of the problem.

Let me know how things go!

[cpitclaudel]

Hi ! Thanks for the quick response :)

Yup, I'm using nextpnr. I do have a FIFO between my design and usb_serial. Interestingly, failing timing gives me these data curruption errors, but no enumeration failure just yet ^^

The smaller designs I've tried passed timing, and things worked fine there.

Thanks again for your help!

[cpitclaudel]

I've tried to shrink my design further. On very small designs I pass timing and things work, but as I add complexity things start breaking again.

Do you have experience with Low-Speed USB (running at 1.5Mbits/s instead of 12, with a clock speed of 24MHz)? How hard would it be to adjust the existing code to run with a 24MHz clock? Or am I approaching this wrong?

[ulixxe]

Maybe you can try my implementation at: https://github.com/ulixxe/usb_cdc