I found the solution for this problem. The problem is in PLL.
If you use 12MHz quartz, then you should set PLLCFG = 0x00000007;
Fcco = (2*M*Fin)/N, M=8, N=1 -> Fcco = 196MHz
Fcpu = 48MHz = 196MHz/4
Tell me if works, on your boards.
Borut
--- In l..., Edwin Olson wrote:
>
> Hi folks,
>
> I have about 20 Keil MCB2300 boards with an LPC2378 on them. Of
these,
> 15 seem to be running our application software with
no problems at
all.
> The remainder consistently (some right away, others
after a short
amount
> of uptime) exhibit significant ethernet packet loss--
around 6-10%
when
> doing a flood ping.) The "good" boards drop packets
at well under
0.01%:
> about what I'd expect from a private switched
ethernet and
comparable to
> my desktop.
>
> This seems to indicate one of two possible problems:
>
> 1. There is a hardware problem affecting some fraction of the
> MCB2300's. It could be a problem with the PHY, MCU, PCB, or even
assembly.
>
> 2. I am doing something that is pushing the parts near or beyond
some
> limit they have, and with 20 boards, I'm seeing
the failure cases
that
> would result from doing this.
>
> Note that these boards are running at 48MHz in order to avoid the
> previously-discussed 72MHz+MAM problems affecting the LPC2378
silicon.
> Setting MAMCR=0 also has no effect: the "bad" boards
still drop
packets.
>
> Any ideas? Aside from the packet loss, the boards seem to be
running
> fine (no crashes, no obvious memory corruption). The
problem is
> reminiscent of the behavior when the power supply browns out (the
first
> thing to go is the ability to transmit packets, it
seems. Not sure
> whether it's the PHY or the MCU that's the limiting factor, but
I'd
> guess the PHY since it's a 3.3V device and would be hit by a brown
out
> first.). However, these boards exhibit poor behavior
even when all
the
> power rails are good.
>
> There's no obvious correlation between the MCU date code and
good/bad
> boards.
>
> -Ed
>