> Comments?
Numbers:
unconditional copy of 2 tls descs: 5 cycles
this patch with 1 tls desc: 26 cycles
this patch with 8 tls descs: 52 cycles
lldt: 51 cycles
lgdt: 50 cycles
context switch: 2000 cycles (measured with pipe read/write and vmstat so
it's not very accurate)

So this patch causes a 1% context switch performance drop for
multithreaded applications.

Note: the benchmark doesn't include the initial test for non-zero
nr_tls_bytes and doesn't include setting the LDT descriptor