On Tue, 26 May 2026 22:30:00 +0200,
"Theo de Raadt" <deraadt@openbsd.org> wrote:
> 
> > It was poor wording, I mean that on 2-3 reads I actually read the report and
> > it works. After series of additional test to challenge 100ms sleep I
> > reallized that I never saw that I beed 2rd read. Second time is enough.
> 
> Do you just read, or do you write a command followed by a new read?
> 
> If the latter, is it possible that the first write gets gibbled on the
> way to the device?
> 
> Are we use the i2c bus driver first byte works correctly?
> 

I had instrumented qciic_fdt_write() and dumped a word which was written by
HWRITE4(sc, GENI_TX_FIFO, word) by printf() it.

Both call sends the same word to the device, but the first one reads series
of 0xff, and the next write and read reads the report.

Just second read, without write command, reads the same series of 0xff.

To get report it writes 0x02320005 and 0x00000006, and for good reports it
reads 0x04000215 and 0xffffffff as a bad one.

I don't see anything in a code path that can explain that behaviour.

-- 
wbr, Kirill