SPARC Segments

• External variables
• The text segment
• The data segment
• The bss segment
• Initialized variable declarations:
  • numeric
  • ASCII
  • pointers

C Variables

• "extern": globally visible
• "static": only visible within file/function
• "automatic": on stack
• "register": in registers (do not have address)
• "volatile": cannot be cached
• constants: read-only (e.g. strings)

SPARC segments

• variables initialized to 0 (default):
  -- bss segment
• variables initialized to non-zero values:
  -- data segment
• read-only constants and executable code, initialized to non-zero values:
  -- text segment
• stack, variable-sized area, uninitialized:
  -- stack segment

SPARC segments

• segments are aligned to page boundaries (multiples of 0x2000 or 8192)
• anything declared in an assembly program goes into the text segment unless otherwise declared.
• declarations .text, .data, .bss, tell the assembler that what follows goes into the specified segment

SPARC memory organization

text segment

• loaded from binary file
• read-only
• starts at location 0x2000
• mostly location-independent:
  • stack locations are relative to %sp or %fp
  • all branches are relative
  • main must be known (which is why we use .global)

data segment

• loaded from binary file
• read-write
• .word for one or more 32-bit quantities
• .half for one or more 16-bit quantities
• .byte for one or more 8-bit quantities:
  • .byte 150, 145, 154, 154, 157
  • .byte "h", "e", "l", "l", "o"
  • .ascii "hello"
• .ascii for strings of characters (unterminated)
• .asciz for null-terminated strings of characters (C strings)

Example: Hello World

main () { printf("hello world\n"); }

include(macrodefs.m)
        .global printf
format: .asciz  "hello world\n"
        .align  4
        beginmain
        set     format, %o0
        call    printf
        nop
        endmain

.global

• used for symbols visible outside this file
• code labels (entry points), variable labels
• only .global symbols are visible to the debugger
• all non-global symbols are only visible within the file they are declared in
• each global symbol should only be declared once
• .common name, size lets us declare name any number of times; each occurrence refers to the same memory

.align

• assembler context for each line includes:
  • current segment s
  • current segment counter c
• .align n increases c by a number i \in (0... n - 1) so c is a multiple of n

Example: Array of Pointers

char * wkday [] = {"mon",
                   "tue", "wed", "thu",
                   "fri", "sat", "sun"}

        .align  4
        .global wkday
wkday:  .word   monm, tuem, wedm
        .word   thum, frim, satm,
        .word   sunm
monm:  .asciz  "mon"
        ...
sunm:  .asciz  "sun"
        set     wkday + (3 << 2), %o0
        ld      [%o0], %o0

bss segment

• size loaded from binary file
• automatically initialized to zero
• read-write
• .skip size to reserve space

bss segment examples

	.bss
	.align	4
arraym:.skip	8 * 100
cm:	.skip	1
	.align	4
im:	.skip	4
	.align	2
	.common	shared, 50