Mesa 7i90 in SPI mode

So, mesaflash compiled well, but it can't see 7i90 on my Pi 2.

Another question: what are symptoms of lost connection?
Say, I loaded hm2_rpspi and addf-ed it to a thread, so I can control CR leds. After 3-4 hrs, I can still issue setp and it works. Does it mean that the connection has been stable and never lost?

i noticed on my machines that sometimes mesaflash would only work after a fresh boot, if i had started and stopped linuxcnc, eg modules were loaded. then the card wouldnt be detected. what is the output of mesaflash?

you can also try running this tool too see if the card is responding with a "AA AA AA AA" preamble on /dev/spidev0.0
github.com/torvalds/linux/blob/master/tools/spi/spidev_test.c

also wrt to kernel build times, i used to have a separate box running distcc to cross compile kernels but it fried, i just compiled 4.9.15 directly on some boards, it took about 4 hours on a pi 3 and 16 hours on a pi zero.

One thing to note, mesaflash and spidev_test both assume the standard spi driver is loaded and available. The hm2_rpspi solution goes directly to the low-level hardware and at least originally required disabling the standard spi driver.

Right, I did not enable SPI in config.txt before, thanks for the hint!
Now mesaflash works (at least readhmid is OK).

And yes, loading hm2_spi breaks it. Mesaflash shows something about wrong cookie FFFF FFFF FFFF. Not a big deal, of course.

If interested there is an updated hm2_spi.c which allows for 8-bit words here,
github.com/resojoe/linuxcnc/tree/master/...rivers/mesa-hostmot2

Thanks, I'll try that!
Why wouldn't you send a PR against the main repo?

BTW I noticed that mesaflash says something about 32-bit failing, using 8-bit...

Have personally seen a few SPI timeout messages when running. Likely need to update standard RPi spidev driver code to provide better determinism.

Also need to completely comment these changes prior to submission.

The 7i90 expects 32-bit transfers and was originally developed using hardware which provides 32-bit words when using SPI. Thus mesaflash configures the hardware for 32-bit words. The RPi, Pine and others only allow for 8-bit SPI "word" sizes. These SBCs reject the 32-bit request. Instead of failing and exiting mesaflash now sets 8-bit words and orders the bytes for correct transmission.

OK I connected a stepper drive (with optocoupled input) to 7i90 pins and now watchdog bites shortly after the start. Pretty unusable setup...

The SPI interface is quite susceptible to noise
This can be improved by:

1. Keeping the SPI cable short (a few inches max)
2. Using all the 26 pin cable grounds
(to lower inductance so stray HF currents dont generate much ground bumping)

Its probably possible to improve the noise situation (at the expense of speed) on the
FPGA RX side by digital filtering the RX data and clock
and on the RSPI RX side with a RC filter on the data from the 7I90