[fedora-arm] smsc95xx performance bug: eth vs usb

[DJ Delorie]

> A comment left on my blog [1] pointed me to a Google Groups discussion 
> [2] which pointed to a problem with gcc-4.6.0 [3].  Fedora 13 had 
> gcc-4.4.5, though, and the kernel I'm using was compiled with gcc-4.3.3, 
> so this may not be related.

I saw that, but my kernels have the "good" generated code, yet still
have the slow performance.  It seems to me that the "bad" generated
code results in a USB that just doesn't work at all.