[ofa-general] Re: libmlx4 wc flash

[Roland Dreier]

 > so what it comes down to, is that if we assume that WC
 > will *only be enabled through PAT* then it's safe to use sfence
 > in this case. Right?

I don't think we can really make any assumptions about what
instructions a 32-bit x86 processor has available.  Who knows what
wacky stuff VIA or someone like that will come up with?

The best thing seems to be just to stick to "lock; addl $0,0(%%esp)"
for 32-bit x86.  We now have the infrastructure to support multiple
builds of libraries and have ld.so select automatically at runtime but
I'm not sure it's really worth it.

 - R.