Selftest is not supported on revC

[electroniceel]

Selftest can give false-positive results on revC, see for example https://twitter.com/szeloof/status/1161346534098833408

The main reason is probably that cmos inputs are left floating in high-z and their value is used for the test results. This seems to be a leftover from revA and revB which had hold circuits.

So selftest for revC has to be reworked to utilize the pullups/pulldowns.

For the pins-int test the internal pullups of the ice40 should be used. This is not yet supported in nmigen.

[whitequark]

One problem is that there could actually be shorts on the side of pull drivers too.

[electroniceel]

There could be shorts on the side of the pull drivers, there could also be a bridge over the 10k resistor, changing pullup/pulldown to push-pull.

Some of this can be detected on the other port utilizing the pins-loop and hoping that there is not the same error on the same pins.

But in the end you can detect all shorts with current measurement. This is why I wanted to have it on the test jig pcb.

[esden]

Just for added information. I just ran into this issue too.

The pins-int selftest consistently results in:

E: g.applet.internal.selftest: self-test: FAIL
E: g.applet.internal.selftest: pins-int: fail high: A0 A1 A2 A3 A4 A5 A6 A7 B0 B1 B2 B3 B4 B5 B6 B7

All other selftests pass. This is consistent on 3 revC1 boards I could get tested.

[FFY00]

Also ran into this in a new board, I had tested before with an older version and it passed.

I: g.device.hardware: device already has bitstream ID b690670e14e6d9e9ae51bd8ddae5ad4d
I: g.cli: running handler for applet 'selftest'
I: g.applet.internal.selftest: running self-test mode pins-int
I: g.applet.internal.selftest: running self-test mode loopback
E: g.applet.internal.selftest: self-test: FAIL
E: g.applet.internal.selftest: pins-int: fail high: A0 A1 A2 A3 A4 A5 A6 A7 B0 B1 B2 B3 B4 B5 B6 B7

[whitequark]

Yes, the selftest is completely broken on revC, please stop reiterating this over and over again. I should probably just remove it.

[attie]

@whitequark can this be closed after 13a3509?

[whitequark]

No, we still have to implement some form of selftest on revC. I've renamed the issue.

[TomKeddie]

@whitequark would you like me to look at this? I think the first steps would be to get the existing tests to be sane on revC?

[rwhitby]

@TomKeddie I'd be happy to assist you with that if desired (not that I think you need assistance to do it, but two sets of heads and hands may useful).

[TomKeddie]

@rwhitby I'll look at this toward the end of the week. I have some fixes already on a branch, I made some changes/hacks to find the shorts in the boards I built.

[RoganDawes]

I just started trying to use my CrowdSupply Glasgow C3, and am running into odd behaviour on the selftests. Even the pins-loop test fails, surprisingly, with different pins reported each run.

Not sure how to proceed, but it does feel like more/working selftests would help to diagnose the problem?