Re: dd miniroot.fs into CF card from non-amiga NetBSD box

[Rin Okuyama <rokuyama%rk.phys.keio.ac.jp@localhost>]

On 2017/02/24 19:56, Michael van Elst wrote:
> It is supposed to calibrate the delay loop by measuring loop speed
> against the hardware clocks. I don't know why this fails for your
> board.
>
> delay(x) loops (1024 * x / delaydivisor) times.
>
> The initial delaydivisor for 68060 is 1024 * 1 / 80 which is truncated
> to 12. So you loop about 85 * x times and this needs to take more than
> about 1.4us (eclock) and less than about 10ms (hz interval).
>
> If your machine is too fast, you can make the initial divisor smaller
> so that the test fits into this range.
>
> The dmesg output shows the value of delaydivisor, a Cyberstorm Mk2
> (68060 50MHz) shows a value of 21. So the calibration started with
> 12 and increased it to 21 which makes it unlikely that your board
> is too fast for the test.
>
> What calibration value is printed for you?

Thank you so much for kind instruction. I enabled DEBUG codes in
dev/clock.c and observed calibration values:

68060 @ 50MHz:
  Calibration delay loop...
  diff 510 us, new divisor 21/1024 us
  diff 1008 us, new divisor 21/1024 us
  diff 1008 us, new divisor 21/1024 us

68030 @ 50MHz:
  Calibration delay loop...
  diff 995 us, new divisor 167/1024 us
  diff 1017 us, new divisor 166/1024 us
  diff 1024 us, new divisor 166/1024 us

It seems that calibration can be done in accordance with your scenario.
I don't understand why it fails. In addition, when newfs was done with
68030, timeout errors sometimes happen with 68060, but they became no
longer critical; the system boots up into multi-user mode, and I can
extract xfont.tgz successfully. I'm not sure yet whether the problem
has gone or not. I will test it further...

By the way,

On 2017/02/23 23:57, Michael van Elst wrote:
> rokuyama%rk.phys.keio.ac.jp@localhost (Rin Okuyama) writes:
>
> > By the way, I'm envy of Debian's RDB support. We need to implement
> > DKWEDGE_METHOD_RDB option to NetBSD in the future!
>
> That's trivial to do.

I've written DKWEDGE_METHOD_RDB option. You can find the main patch
as well as that for kernel configration files (enabled for x86 by
default, commented out for other platforms) here:

  http://www.netbsd.org/~rin/dkwedge_rdb_20170226/

It works just fine for me on amd64. Could you please review it? Would
it be better to send-pr?

Thanks,
rin