I220A-Lab 10 Solved

Exercise 1: Rotate Instructions

Most modern architectures provide some kind of rotate instruction which rotate a register circularly. C does not provide any rotate semantics and it seems necessary to use assembly language if rotations are needed in performance-critical code (like encryption applications).

However, existing C compilers are extremely capable and are setup to recognize certain idioms for rotation. Change over to the rotate directory and look at the rotate.c le. It provides implementation of rotates in terms of shifts: a rotate is computed as a shift in the rotate direction or’d with bits rotated around; the bits rotated around are computed by shifting bit-length - shift-length bits in the opposite direction.

Build an assembly language le by typing make and look at the generated rotate.s. You will see that the generated code does make use of rotate instructions. Since the rotl() and rotr() functions have been declared inline, the compiler has replaced the calls with single rotl and rotr instructions respectively.

The takeaway from this exercise is that when a situation seems to require the use of assembly language, it is worth checking whether the situation can be handled satisfactorily while staying entirely within C.

1.3.3               Exercise 2: cpuid for Vendor Information

Since around the time Pentium processors were introducted, the x86 architecture has provided a cpuid instruction which provides information about the CPU. The operation of the cpuid instruction is controlled by the incoming value in the eax register. Information is returned in the eax, ebx, ecx and edx registers with the details of the returned information depending on what was requested by the incoming eax register.

When the incoming eax register is speci ed as 0, the returned eax value is the largest value which can be used for an incoming eax value. ebx, edx and ecx (in that order) spell out a vendor string corresponding to the CPU manufacturer. Non-zero incoming values for eax, returns many low-level details of the CPU.

Change over to the cpuid1 directory and look at the les contained there. The cpuid.s le provides a get_cpuid() function which executes the cpuid instruction with the incoming eax set to 0 and returns the returned information via 4-incoming pointer arguments in rdi, rsi, rdx and rcx. This function is exercised by a program in main.c which decodes and prints out the vendor string.

Build the program by typing make and run it by typing ./cpuid.

1.3.4                Exercise 3: Arbitrary cpuid Information

The previous exercise only used the cpuid instruction with the incoming eax register set to 0. Change over to the cpuid2 directory: it contains a main.c which is setup to treat its rst command-line argument as the value of eax used to select the cpuid operation. However, the provided cpuid.s le does not provide this functionality and has the same code as that from the previous exercise. Modify this le to take an additional initial parameter which is the desired cpuid op.

After making your changes, build and test. Check out its operation for values of eax other than 0, verifying your result with the information given in wikipedia.

1.3.5           Exercise 4: Parity

A bit-word is de ned to have even parity if the number of bits in the word is even; otherwise it has odd parity. Parity is one of the most simple error-detecting mechanisms; for example, memory words often have an extra parity bit to allow detecting single-bit errors (for example, a 32-bit memory word will have 33-bits with the extra parity bit used to detect errors in the 32 data bits).

Computing the parity of a word in C would require a sequence of bit operations. However, in x86 assembly language, the parity of a word is available by testing the parity ag P after simply test’ing the word.

Change over to the parity directory and look at the provided les. Complete the code in the parity.s le to return eax as 1 i the incoming argument has even parity. You should use the testl instruction to set the parity ag. You can then test the ag using the jpe ("jump if even parity") instruction to ensure that the correct return value is set up in register eax.

Build using make and test by checking the parity of the command-line arguments to the program.

$ ./parity usage: ./parity INT1... $ ./parity 1 2 3 4 5 6 7 15 21

parity(1) = 0 parity(2) = 0 parity(3) = 1 parity(4) = 0 parity(5) = 1 parity(6) = 1 parity(7) = 0 parity(15) = 1 parity(21) = 0

$ ./parity -1 0 -2 parity(-1) = 1 parity(0) = 1 parity(-2) = 0 $

1.3.6                Exercise 5: Accessing the Instruction Pointer

gcc allows inline assembly language as a compiler extension. The inline assembly must be written within a string which is the rst argument of the __asm__(). This argument can contain assembly instructions which must be separated by a ; or \n\t sequence.

This string can also be used as a template. In that case, the remaining arguments to __asm__() (separated by :) provide constraints on the template arguments. The details of the format are given in the references.

Change over to the rip directory and look at the rip.c le. This le provides a function get_rip() which returns the instruction pointer for the the return in the function. It uses inline assembly to access the current instruction pointer. The exact syntax:

__asm__("leaq (%%rip), %0": =r(rip));

speci es the template string as "leaq (%%rip), %0" (the holes in the template are indicated using %d for small integer d; note the clumsiness of having to quote the % used in the register speci er by doubling as in %%rip). The second argument (after the :) speci es the output values a ected by the template: in this case we want it to correspond to the template variable %0 and make it correspond to the register allocated to the rip C variable.

Note that it would have been more direct to have had a template which resulted in something like movq %rip, %rax. However, that is not a legal instruction as none of the x86 instructions allow reading rip directly. Instead, the leaq (%rip), %rax gets the value of rip indirectly by using it as a base register in an address computation.

Build the program by typing make and run it using ./rip. Observe that the printed rip pointer is consistent with the address printed for the get_rip() function (look at the rip.dump le produced by the make and verify the values).

1.3.7               Exercise 6: Accessing the Stack Pointer

Change over to the rsp directory and look at the les contained there. Then go into the rsp.c le and add inline assembly to the get_rsp() function to return the value of the rsp register just before the return. Note that since x86 allows direct reading of the rsp register, getting hold of its value is much easier than reading the value of the rip register in the previous exercise.