Homework Assignment 5
                               CS 350c

                         Unique Number: 52140
                             Spring, 2017

                       Given:  February 14, 2017
                        Due:  February 23, 2017


This homework concerns exploring the x86 memory management mechanisms
and some questions about the specificity of the x86 ISA.  You may need
to refer to the x86 documentation to be able to complete this
assignment.

In addition, this assignment asks that you carefully read Section 4.1
of "Computer Systems, A Programmer's Perspective".  Your next homework
assignment will involve writing a y86 ISA simulator.


1. If you want to create a set of page tables to map 21,357,113 bytes
   of contiguous memory, how many 4K pages would there need to be
   assuming that all of the memory is of the same type?

2. What is the minimum amount of memory that would be required for all
   of the page-table, memory-resident, mapping information necessary
   to make mapping possible for 21,357,113 bytes of contiguous
   read/write data memory when using 4K pages?

3. What is the minimum amount of memory that would be required for all
   of the page-table, memory-resident, mapping information necessary
   to make mapping possible for 21,357,113 bytes of contiguous
   read/write data memory when using 2M pages?

4. Does it matter where the 21,357,113 bytes start in the memory as it
   concerns the required number of page-table entries?

5. Using 4K pages, how large would the TLB need to be so that we could
   address all 21,357,113 bytes without ever needing to reload the TLB?

6. Using 2M pages, how large would the TLB need to be so that we could
   address all 21,357,113 bytes without ever needing to reload the TLB?


Here are some questions that concern the x86 instruction-set
architecture.

7. Are there x86 instructions that (potentially) produce
   non-repeatable answers?  What do I mean?  If you use the RDTSC
   and RDTSCP instructions to measure the duration of a program, why
   would you get different answers when running the same program
   repeatedly?  Would one get a better result if one were to use a
   raw x86 machine; that is, a x86 system without an operating system?

8. If you write a "timing loop" program -- a program that repeats some
   calculation some number of times -- will it provide an accurate
   delay if you know the clock frequency of the processor on which you
   you are running your experiments?  Why or why not?  Explain your
   answer.

9. Given a simulator state like this:

          typedef unsigned char u8;
          typedef long i64;

          i64 rip;             // y86 program counter
          i64 rgf[ 15 ];       // y86 registers
          u8  zf;              // y86 zero flag
          u8  sf;              // y86 sign flag
          u8  of;              // y86 overflow flag
          u8  mem[ MEMSIZE ];  // y86 memory
          u8  status;          // y86 processor status

   where MEMSIZE < 2^64, could we create a faithful simulator with
   respect to the y86 instruction-set architecture specification?