MarkT:
Choose a regulator that can handle at least 8 or 9V on the input. Make sure its got enough input decoupling
too.
Mark makes some excellent points here. A higher voltage rated LDO is important, also you may not be able to use an LDO at all and should try a standard regulator. Read below for why that might be.
There are several things to keep in mind when coupling SMPS and Linear regulators. The first thing to remember is that while we like to think of linear regulators as resistive devices (they just dissipate the extra voltage as heat). The reality is that linear regulators are actually better thought of as variable gain amplifiers. What I mean by this is that they are amplifiers which adjust their gain to hold the output at a particular value. This distinction is important because just like amplifiers they are prone to oscillation when certain conditions are met. LDO regulators are MUCH MUCH MUCH more prone to oscillation than standard types. This is why the decoupling capacitors before and after the regulator are so necessary. These are not to smooth out the inputs and outputs per se, the regulator should be handling the output already, but are to dampen any signals that would cause the regulator to oscillate. Also, more importantly if the regulator does enter an oscillation mode, the caps serve to dampen the oscillation such that it stops quickly instead of continuing on indefinitely, or even worse increasing, damaging the LDO. The math on all of this is really damn complex and to be honest stuff I do not fully understand. Stuff about nodes and frequency zeros, phase angles etc.
Moving on to interfacing SMPS with linear regulators. It's very common to see an SMPS drive a regulator into oscillation. There are a wide variety of high frequency switching transients involved which linear regulators have to dissipate. On top of this, its not uncommon to see an LDO wreak havoc on an SMPS feedback circuit. SMPS devices are much happier seeing pure resistive loads, with a regulator its not that way due to its amplifier nature. You can get issues where the two feed back circuits (the one in the SMPS and the one in the regulator) cause each other to misbehave. You can think of it as constructive and destructive interference. Sometimes they work together where both are pulling low, causing the output to swing too low, which causes both to pull to high, causing the output to swing too high. This basically results in the system and/or the linear regulator oscillating, eventually killing it. Other times they work against each other causing huge swings on the input to the linear regulator killing it. Of course, other times where the frequencies between the two are not a problem you can them working together just fine.
As for what to do, first thing. As Mark said first off give your self some voltage head room by using a higher voltage rated regulator. If you don't have to use an LDO then don't; check the drop out versus current chart in the datasheet to see what your real dropout would be versus using the max value. As for the decoupling capacitors, the best thing to do is start off using the datasheets guidelines for decoupling and going from there. It may be as simple as adding a small and large value decoupling cap before the regulator. On the other hand, you may find that one of the various types of Low Pass filters is necessary before the regulator. By these types I mean CLC, CRC, LC, RC, etc. There is no one way to do this but start with the simple options first and go from there.
Hopefully this gives you a better idea of what is likely going on here.
PS. It might be more than just the 5V SMPS and LDO that are oscillating. You mention another circuit that is for USB power. If there are any LDO's or SMPS topogies in that section as well, they could causes the 5V SMPS to start having a less than idea output which then causes oscillation between the SMPS and LDO. This type of stuff would be QUITE obvious if you have an oscilloscope.
Also, a regulator with NO load is a REALLY bad idea. Any built up charge on the output has NO WHERE to go. The only way is via the feedback loop, making oscillation REALLY easy to happen. During testing, throw on around 10mA-50mA of load on the output. A 500 ohm 1/8W to 100ohm 1/2W resistor should do the trick.
Can anyone explain what oscillating in relation to a LDO is? Is it like an endless loop in code?