so if this reports unlimited (or at least, larger than anything else) entropy then there are likely situations where it's the only source available (people don't typically have lava lamps wired up). and then an attack seems possible.
[edit: while startup is the case of the bug i remember, you might "consume" entropy faster than it is generated at other times too (i imagine a server running https has a fairly high demand for entropy, for example).]
i was trying to explain why an untrusted hardware rng cannot be improved in some cases (because of limited entropy from elsewhere). and i don't see how what you are saying addresses that.
(1) Chaining rnd generators is fine as long as they are not correlated somehow. If there is any correlaton, the output may be weaker. (Consider chainig two exactly same generator as a corner case.)
(2) In this case we don't chain generators; you would loose the speed of the integrated one otherwise.
The important commit here is:
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.g...
Excerpted:
Change get_random_bytes() to not use the HW RNG, even if it is avaiable.
The reason for this is that the hw random number generator is fast (if it is present), but it requires that we trust the hardware manufacturer to have not put in a back door. (For example, an increasing counter encrypted by an AES key known to the NSA.)
It's unlikely that Intel (for example) was paid off by the US Government to do this, but it's impossible for them to prove otherwise --- especially since Bull Mountain is documented to use AES as a whitener. Hence, the output of an evil, trojan-horse version of RDRAND is statistically indistinguishable from an RDRAND implemented to the specifications claimed by Intel. Short of using a tunnelling electronic microscope to reverse engineer an Ivy Bridge chip and disassembling and analyzing the CPU microcode, there's no way for us to tell for sure.