clGetDeviceIDs() failure

I have a tiny script to launch cgminer which works fine.

go:

#!/bin/sh
set -x
cd /x/y/z
/x/y/z/cgminer ...

UNLESS I launch it from rc.local:

nohup /x/y/z/go > /dev/null 2>&1 &

in this case clGetDeviceIDs() fails, which causes clDevicesNum() in ocl.c to fail, which aborts everything.

[2013-05-13 19:56:01] Started cgminer 3.1.0
[2013-05-13 19:56:01] Started cgminer 3.1.0 No protocol specified
[2013-05-13 19:56:02] CL Platform 0 vendor: Advanced Micro Devices, Inc.
[2013-05-13 19:56:02] CL Platform 0 name: AMD Accelerated Parallel Processing
[2013-05-13 19:56:02] CL Platform 0 version: OpenCL 1.2 AMD-APP (1124.2)
[2013-05-13 19:56:02] Error -1: Getting Device IDs (num)
[2013-05-13 19:56:02] clDevicesNum returned error, no GPUs usable
[2013-05-13 19:56:02] CL Platform 0 vendor: Advanced Micro Devices, Inc.
[2013-05-13 19:56:02] CL Platform 0 name: AMD Accelerated Parallel Processing
[2013-05-13 19:56:02] CL Platform 0 version: OpenCL 1.2 AMD-APP (1124.2)
[2013-05-13 19:56:02] Error -1: Getting Device IDs (num)
[2013-05-13 19:56:02] clDevicesNum returned error, no GPUs usable
[2013-05-13 19:56:02] All devices disabled, cannot mine!
[2013-05-13 19:56:02] All devices disabled, cannot mine!

(I edited out the ASCII escape sequences used for screen display)

rc.local is launched as the final thing coming up into run level 3 - S99rc.local in rc3.d starts it up.

If I run rc.local directly, it also works, if I reboot it does not, and fails as above.

What's different between the two cases? I captured the environment for the working case:

env > xxx

then sourced it from go:

#!/bin/sh
set -x
cd /x/y/z
. /x/y/z/xxx
/x/y/z/cgminer ...

but that did not help either.

One side note is the "No protocol specified" error in the does-not-work case, which I don't think comes from
cgminer source.

I thought I'd ask before digging deeper.

thanks