create workload to measure tlb flush performance

[kemicoco]

These workload are designed to measure the performance of tlb flush.
For tlb flush1 test case, each thread/process maps a separated private file mapping to memory, and
writes to a page before initiating a tlb flush by madvise(MADV_DONTNEED), subsequent access of
this page will result in repopulating the memory contents of the up-to-date of the underlying file.

For tlb flush2 test case, each thread/process has an anonymous private mappings, and writes to a 
page before initiationg a tlb flush by madvise(MADV_DONTNEED), subsequent access
of this page will result in zero-fill-on-demand pages for anonymous private
mapping.

Additionally, they are also tlb flush sensitive workload to verify the
behavior of tlb hardware, the assertion fails immediately once tlb flush
failure. We have double checked it by hacking Linux kernel to skip tlb
flush manually.

 This work is done with Dave Hansen. These are the best Linux workloads that we have that could
detect issues with the TLB either in software, or potentially in hardware.
This could provide validation for the ICX "Split TLB" behavior. 

[amitay]

In tlb_flush1.c, tmpfile should be "/tmp/..." instead of "/tmpfs/...".

[amitay]

Pull request #10 includes the fixes for the first patch and additional test for shared mapping.

[antonblanchard]

This was resolved in #10, closing.