Disapearing pl2303 usb serial adapter on rpi2

[bob prohaska]

On Fri, Dec 22, 2017 at 12:01:54PM -0800, Rodney W. Grimes wrote:
> 
> I would need a circuit layout diagram to be certain, but I
> suspect that if this device does latch up it is going to
> stay latched until all sources of voltage are removed.
> 
If that's the case I can understand the pl2303's bad reputation.
It does not obviously explain the pl2303's erratic operation on
an RPI2 running FreeBSD compared to an RPI3 running Raspbian or
a Mac running OSX.

> 
> Try lifting everything but the data end gnd, that would be
> the USB connector, and RX and TX.
> 
If that didn't cause the adapter to reset it would be a miracle 8-)
I'll try it next time things lock up.

> > It might help if I indicate the layout of the system, since I gather it's
> > perahps a little unusual. There are four hosts, call them com, net,
> > ns1 and ns2. They're all networked to a hub, with the usb/serial cables
> > arranged so that each RPI2 provides terminal service to the next host
> > in the ring:
> > 
> > com-usb/serial-ns1-usb/serial/-ns2-usb/serial-net-usb/serial-com
> 
> Ok, that is good additional data.
> 
> > There was a faint tendency for hosts net and ns1 to have more usb/serial
> > adapter lockups, so those hosts got FTDI adapters and all was well until
> > the host called com (the test box) was upgraded to r326951. Serial link
> > uptimes went from weeks or months to hours, for that host only.
> > 
> > The serial cables obviously ground all four RPI2s together, I think the
> > ethernet likewise distributes ground. Wall warts are connected to a single
> > outlet strip, are isolated and do not distribute ground. All the cables are 
> > confined to a small shelf about one by two feet in size.
> 
> I think you are correct in that the serial cables are tying all 4 systems
> grounds togeather, I am unclear on what the USB ground isolation rules
> are, I would actually suspect they are isolated.  I put an ohm meter on
> my ftdi USB serial adapter and its gnd pin is infact tied to the shell
> of the USB connector, so that is a ground path.
> 
Doing the same test on my pl2303 adapters finds the shell poorly grounded. 
The metal USB shell is only loosely crimped to the ground plane of the
circuit board, and not soldered. Pin 4 is supposed to be ground also, so
maybe it doesn't matter. Still, I'm surprised enough to try soldering one
of the shells to ground and will try it presently. 

> Ethernet does NOT have any ground wires, they are differential pairs
> and isolated via the transceiver.
>

Ok, that's good, I think.... 

> My pedantic side does not like this stringy "using a signal cable"
> as a system ground.  I would rather see all 4 RPI's tied with a
> good ground to each other, and disconnect the serial cable gnd
> wire.  
>
Basically I agree with you. I didn't like the ring topology either, but
it was the only way to get serial access to all the Pi's without some
sort of terminal server.
  
> > It wouldn't be surprising if this turned out to be a wiring issue, but
> > proof has so far proved elusive. Just for completeness, there's a photo
> > of the setup at 
> > http://www.zefox.net/~fbsd/com_net_ns1_ns2
> > Left to right, the hosts are com, net, ns1 and ns2. I admit it isn't 
> > pretty, but is far from the biggest cable nightmare seen elsewhere.
> 
> Looks okay, but this is still a poor system ground in my mind.  This
> is not an earth ground, but a signal ground, and those are usually
> best done in a star topology, not a string.

It occurs to me that lifting _one_ ground will break the ground loop. The
serial pins at the lifted ground will have to return current through the
other Pi's, which isn't good. At this point if I could make the problem
dependably _worse_ I'd be at least slightly the wiser.

Thanks for reading!

bob prohaska