1

当我RTFM时,AFAIK 中断处理必须在汇编中实现。

eret 指令用于在预异常地址处恢复执行。

这个原因是否也可以推广到 x86,即无论芯片是什么,您都无法编写没有汇编指令的操作系统?这是实际结果还是理论上证明?汇编语言和其他语言之间是否存在一些导致这种差异的主要差异?我推测汇编语言没有所谓的BNF,这是真的吗?是不是说汇编语言没有所谓的上下文无关文法,汇编语言不是用yacc、bison、flex、lex实现的,而是更像是按照硬件芯片来实现的?

我正在使用的低级代码看起来像这样,我想知道为什么它不能在 C(或 Java 等)中完成,因为这些结果似乎与 2 个图灵完整实现可以解决相同的问题相矛盾,所以如果语言 A 是图灵完备且语言 B 也是图灵完备的,那么 A 能够解决的任何问题也可以用语言 B 解决。

请如果你能教我这些概念,因为我认为我可以自己学习个别指令,但我不知道如何回答关于汇编编程的一般问题,例如被问到为什么我用汇编而不是 C 来做,那么为什么用 C而不是 OOP 等等。

################################################################
#
# Definitions for important devices and addresses in this system.
#

# Uart_0 at 0x860

.equ de2_uart_0_base,0x860

# Timer_1 at 0x920, interrupt index 10 (mask 2^10 = 0x400)

.equ de2_timer_1_base,0x920
.equ de2_timer_1_intmask,0x400

# Timeout value for 0,1 ms tick-count interval (CHANGED in every version)

.equ de2_timer_1_timeout_value,4999

# Required tick count per time-slice, meaning
# the number of timer-interrupts before a thread-switch is performed

.equ oslab_ticks_per_timeslice,100

# Interrupt address at 0x800020

.equ de2_nios2_interrupt_address,0x800020

#
# End of device-address definitions
#
################################################################

################################################################
#
# Definition of variables for keeping system time etcetera.
#

.data
.align 2
.global oslab_internal_globaltime
oslab_internal_globaltime:  .word 0

# Definition of variable for remembering the number of
# timer-interrupts since the last thread-switch

.data
.align 2
.global oslab_internal_tickcount
oslab_internal_tickcount:   .word 0

# Definition of system (interrupt) stack, sp, and gp

.data
.align 2
oslab_internal_gp:  .word 0
oslab_internal_sp:  .word 0
oslab_system_stack: .fill 256,1,0
oslab_system_stacktop:

# Definition of the end-of-timeslice message.

oslab_internal_yield_message:
                    .asciz "\n#### Thread yielded after using %d tick%c."

#
# End of system-time variable definitions.
#
################################################################

################################################################
#
# Interrupt handling code.
#

# Stub for interrupt handler

.text
oslab_internal_stub:
    movia   et,oslab_exception_handler
    jmp     et

# The interrupt handler

oslab_exception_handler:
    # Check source of exception, following the procedure
    # described in the Nios II Processor Reference Handbook.

    rdctl   et,estatus      # Check ESTATUS
    andi    et,et,1         # Test EPIE
    beq     et,r0,oslab_exception_was_not_an_interrupt
    rdctl   et,ipending     # Check IPENDING
    beq     et,r0,oslab_exception_was_not_an_interrupt

    # If control comes here, we have established that the
    # exception was caused by an interrupt.
    # Subtract 4 from ea, so that the interrupted instruction
    # will be re-run when we return.

    subi    ea,ea,4

    # Check the source of the interrupt.
    # Possible source No. 1: Timer_1 (currently the only source).

    rdctl   et,ipending
    andi    et,et,de2_timer_1_intmask
    bne     et,r0,oslab_timer_1_interrupt

    # If control comes here, we have an interrupt from an unknown source.
    # This condition is IGNORED in this version of OSLAB.

    eret

oslab_exception_was_not_an_interrupt:

    # Test if the interrupted instruction was a TRAP

    subi    sp,sp,4         # PUSH r8 (instruction 1)
    stw     r8,0(sp)        # PUSH r8 (instruction 2)

    movia   r8,0x003b683a   # binary code for TRAP
    ldw     et,-4(ea)       # Load interrupted instruction
    cmpeq   et,et,r8        # Compare to binary code for TRAP

    # Result from comparison is now in et.

    ldw     r8,0(sp)        # POP r8 (instruction 1)
    addi    sp,sp,4         # POP r8 (instruction 2)

    # Use the comparison result in et as branch condition.
    # The value in et will also be used later, to tell if the
    # exception was a trap or an interrupt.

    bne     et,r0,oslab_trap_handler

    # If control comes here, we have an exception which was not a TRAP.
    # This should not normally happen.
    # However, someone writing programs for the OSLAB micro-operating system
    # could perhaps use unimplemented instructions. To catch unimplemented
    # instructions, we insert a BREAK instruction here. This will stop execution
    # unless the program is run through the debugger.

    break 0
    eret

oslab_timer_1_interrupt:

    # Acknowledge the timer_1 interrupt.

    movia   et,de2_timer_1_base
    stw     r0,0(et)

    # Save contents of R8, to get a free register for
    # temporary values.

    subi    sp,sp,4
    stw     r8,0(sp)        # PUSH r8

    # Increase system clock.

    movia   r8,oslab_internal_globaltime
    ldw     et,0(r8)
    addi    et,et,1
    stw     et,0(r8)

    # Increase tick counter.

    movia   r8,oslab_internal_tickcount
    ldw     et,0(r8)
    addi    et,et,1
    stw     et,0(r8)

    # Restore original contents of R8.

    ldw     r8,0(sp)        # POP r8
    addi    sp,sp,4

    # Check value of tick counter,
    # against the required number of ticks per time-slice.
    # Note: oslab_ticks_per_timeslice is an assembler constant,
    # and not a variable. Hence, no load/store-instructions here.

    subi    et,et,oslab_ticks_per_timeslice

    # If the result from the subtraction is zero (or perhaps positive),
    # then it is time to switch threads.

    bge     et,r0,oslab_time_to_switch

    # If we fall-through here, then we have had one of those many
    # timer interrupts on which we should not switch threads.
    # Return to caller.

    eret

oslab_time_to_switch:

    # This code will now fall-through into the TRAP handler
    # which performs a context switch.
    #
    # We will print out a message for each timer interrupt.
    # To be able to tell that we had a timer interrupt, and not
    # a TRAP, we set et to zero.

    movi    et,0

oslab_trap_handler:

    # Save registers r1 through r23, plus fp, gp, ra and ea

    .set noat               # R1 is used here.
    subi    sp,sp,108       # Make room for all registers. 
    stw     r1, 4(sp)       # R1 is saved in slot 1, not slot 0.
    stw     r2, 8(sp)
    stw     r3,12(sp)
    stw     r4,16(sp)
    stw     r5,20(sp)
    stw     r6,24(sp)
    stw     r7,28(sp)
    stw     r8,32(sp)
    stw     r9,36(sp)
    stw    r10,40(sp)
    stw    r11,44(sp)
    stw    r12,48(sp)
    stw    r13,52(sp)
    stw    r14,56(sp)
    stw    r15,60(sp)
    stw    r16,64(sp)
    stw    r17,68(sp)
    stw    r18,72(sp)
    stw    r19,76(sp)
    stw    r20,80(sp)
    stw    r21,84(sp)
    stw    r22,88(sp)
    stw    r23,92(sp)
    stw    r26,96(sp)
    stw    r28,100(sp)
    stw    r31,104(sp)
    stw     ea,0(sp)        # Special case, saved in slot 0.

    mov     r4,sp           # Copy stack pointer to param1 register
    movia   sp,oslab_system_stacktop     # Use system stack instead

    # Test et to see if this was a timeout event or a TRAP.

    beq     et,r0,oslab_not_a_trap

    # If this was a trap event, we fall through here.
    # Our simplified printf is used to print a message,
    # saying that the previous thread yielded parts of its time-slice.

################################################################
#
#   The following code prints a nice message. Nothing more.
#   This code saves and restores all registers it uses.
#   You can safely ignore the following code, up to
#   (but NOT including) the label oslab_not_a_trap.
#
    subi    sp,sp,4         # Contents of r4 must be preserved.
    stw     r4,0(sp)        # PUSH r4.

    movia   r4,oslab_internal_yield_message
    movia   r5,oslab_internal_tickcount
    ldw     r5,0(r5)
    movi    r6,0            # Gold-plating: check if 1 tick or several ticks.
    subi    et,r5,1         # Do not print the s if only 1 tick.
    beq     et,r0,oslab_no_plural_ticks
    movi    r6,'s'          # If 0 ticks, or 2 or more ticks, print the s.
oslab_no_plural_ticks:
    call    printf

    ldw     r4,0(sp)        # POP r4
    addi    sp,sp,4
#
#   This comment marks the end of the code for printing a nice message.
#   Now comes other code, which is potentially much more interesting.
#
################################################################

    # Move on to thread-switch code.

oslab_not_a_trap:

    # Clear tick counter, since we are going to switch threads.

    movia   et,oslab_internal_tickcount
    stw     r0,0(et)

    # Now it is time to execute the thread-switch code.
    # We use the more general callr, rather than call.

    movia   et,oslab_internal_threadswitch
    callr   et              # Call thread switch routine written in C

    mov     sp,r2           # Copy return value to stack pointer
                            # Yes, the system stack pointer is lost,
                            # but who cares? We will not need it any more.

    # restore registers
    ldw     r1, 4(sp)
    ldw     r2, 8(sp)
    ldw     r3,12(sp)
    ldw     r4,16(sp)
    ldw     r5,20(sp)
    ldw     r6,24(sp)
    ldw     r7,28(sp)
    ldw     r8,32(sp)
    ldw     r9,36(sp)
    ldw    r10,40(sp)
    ldw    r11,44(sp)
    ldw    r12,48(sp)
    ldw    r13,52(sp)
    ldw    r14,56(sp)
    ldw    r15,60(sp)
    ldw    r16,64(sp)
    ldw    r17,68(sp)
    ldw    r18,72(sp)
    ldw    r19,76(sp)
    ldw    r20,80(sp)
    ldw    r21,84(sp)
    ldw    r22,88(sp)
    ldw    r23,92(sp)
    ldw    r26,96(sp)
    ldw    r28,100(sp)
    ldw    r31,104(sp)
    ldw     ea,0(sp)        # Special case
    addi    sp,sp,108

    eret                    # Return from exception

#
# End of exception handling code.
#
################################################################

################################################################
#
# Startup code.
#
# When the system is started, Altera-supplied code initializes the
# Nios II CPU and cache memories, and then calls alt_main.
#

.global alt_main
alt_main:
    wrctl   status,r0       # Disable interrupts.
    wrctl   ienable,r0      # Clear all bits in IENABLE.

    # Now copy the stub.

    movia   r8,oslab_internal_stub
    movia   r9,de2_nios2_interrupt_address
    ldw     r10,0(r8)
    stw     r10,0(r9)
    ldw     r10,4(r8)
    stw     r10,4(r9)
    ldw     r10,8(r8)
    stw     r10,8(r9)

    # Initialize timer_1.

    movia   r8,de2_timer_1_base
    movia   r9,de2_timer_1_timeout_value
    srli    r10,r9,16
    stw     r10,12(r8)      # Write periodh
    andi    r10,r9,0xffff
    stw     r10,8(r8)       # Write periodl
    movi    r10,7           # Continuous, interrupt on timeout, and start
    stw     r10,4(r8)

    # Initialize CPU for interrupts from timer_1.

    movi    r10,de2_timer_1_intmask
    wrctl   ienable,r10
    movi    r10,1
    wrctl   status,r10

    # Call to main. Do not jump, main is a subroutine,
    # and may execute a ret instruction.

    subi    sp,sp,4
    stw     ra,0(sp)        # PUSH r31
    movia   r8,main
    callr   r8
    ldw     ra,0(sp)        # POP r31
    addi    sp,sp,4

    # If main returns, we will return directly to the routine
    # that called us (that called alt_main).

    ret

#
# End of startup code.
#
################################################################

################################################################
#
# Helper functions for initialization and thread handling.
#

.text
.align 2
.global oslab_internal_get_gp
oslab_internal_get_gp:
    mov     r2,gp
    ret

.global oslab_begin_critical_region
oslab_begin_critical_region:
    wrctl   status,r0
    ret

.global oslab_end_critical_region
oslab_end_critical_region:
    movi    r8,1
    wrctl   status,r8
    ret

.global oslab_get_internal_globaltime
oslab_get_internal_globaltime:
    movia   r2,oslab_internal_globaltime
    ldw     r2,0(r2)
    ret

.global oslab_get_internal_tickcount
oslab_get_internal_tickcount:
    movia   r2,oslab_internal_tickcount
    ldw     r2,0(r2)
    ret

.global oslab_yield
oslab_yield:
    trap
    ret

#
# End of helper functions.
#
################################################################
#
# ********************************************************
# *** You don't have to study the code below this line ***
# ********************************************************
#
################################################################
#
# A simplified printf() replacement.
# Implements the following conversions: %c, %d, %s and %x.
# No format-width specifications are allowed,
# for example "%08x" is not implemented.
# Up to four arguments are accepted, i.e. the format string
# and three more. Any extra arguments are silently ignored.
#
# The printf() replacement relies on routines
# out_char_uart_0, out_hex_uart_0,
# out_number_uart_0 and out_string_uart_0
# in file oslab_lowlevel_c.c
#
# We need the macros PUSH and POP - definitions follow.

# PUSH reg - push a single register on the stack

.macro PUSH reg
    subi sp,sp,4    # reserve space on stack
    stw  \reg,0(sp) # store register
.endm

# POP  reg - pop a single register from the stack

.macro POP  reg
    ldw  \reg,0(sp) # fetch top of stack contents
    addi sp,sp,4    # return previously reserved space
.endm

.text
.global printf
printf:
    PUSH    ra      # PUSH return address register r31.
    PUSH    r16     # R16 will point into format string.
    PUSH    r17     # R17 will contain the argument number.
    PUSH    r18     # R18 will contain a copy of r5.
    PUSH    r19     # R19 will contain a copy of r6.
    PUSH    r20     # R20 will contain a copy of r7.
    mov     r16,r4  # Get format string argument
    movi    r17,0   # Clear argument number.
    mov     r18,r5  # Copy r5 to safe place.
    mov     r19,r6  # Copy r6 to safe place.
    mov     r20,r7  # Copy r7 to safe place.
asm_printf_loop:
    ldb     r4,0(r16)   # Get a byte of format string.
    addi    r16,r16,1   # Point to next byte
    # End of format string is marked by a zero-byte.
    beq     r4,r0,asm_printf_end
    cmpeqi  r9,r4,92    # Check for backslash escape.
    bne     r9,r0,asm_printf_backslash
    cmpeqi  r9,r4,'%'   # Check for percent-sign escape.
    bne     r9,r0,asm_printf_percentsign
asm_printf_doprint:
    # No escapes present, just print the character.
    movia   r8,out_char_uart_0
    callr   r8
    br      asm_printf_loop
asm_printf_backslash:
    # Preload address to out_char_uart_0 into r8.
    movia   r8,out_char_uart_0
    ldb     r4,0(r16)   # Get byte after backslash
    addi    r16,r16,1   # Increase byte count.
    # Having a backslash at the end of the format string
    # is illegal, but must not crash our printf code.
    beq     r4,r0,asm_printf_end
    cmpeqi  r9,r4,'n'   # Newline
    beq     r9,r0,asm_printf_backslash_not_newline
    movi    r4,10       # Newline
    callr   r8
    br      asm_printf_loop
asm_printf_backslash_not_newline:
    cmpeqi  r9,r4,'r'   # Return
    beq     r9,r0,asm_printf_backslash_not_return
    movi    r4,13       # Return
    callr   r8
    br      asm_printf_loop
asm_printf_backslash_not_return:
    # Unknown character after backslash - ignore.
    br      asm_printf_loop
asm_printf_percentsign:
    addi    r17,r17,1   # Increase argument count.
    cmpgei  r8,r17,4    # Check against maximum argument count.
    # If maximum argument count exceeded, print format string.
    bne     r8,r0,asm_printf_doprint
    cmpeqi  r9,r17,1    # Is argument number equal to 1?
    beq     r9,r0,asm_printf_not_r5 # beq jumps if cmpeqi false
    mov     r4,r18      # If yes, get argument from saved copy of r5.
    br      asm_printf_do_conversion
asm_printf_not_r5:
    cmpeqi  r9,r17,2    # Is argument number equal to 2?
    beq     r9,r0,asm_printf_not_r6 # beq jumps if cmpeqi false
    mov     r4,r19      # If yes, get argument from saved copy of r6.
    br      asm_printf_do_conversion
asm_printf_not_r6:
    cmpeqi  r9,r17,3    # Is argument number equal to 3?
    beq     r9,r0,asm_printf_not_r7 # beq jumps if cmpeqi false
    mov     r4,r20       # If yes, get argument from saved copy of r7.
    br      asm_printf_do_conversion
asm_printf_not_r7:
    # This should not be possible.
    # If this strange error happens, print format string.
    br      asm_printf_doprint
asm_printf_do_conversion:
    ldb     r8,0(r16)   # Get byte after percent-sign.
    addi    r16,r16,1   # Increase byte count.
    cmpeqi  r9,r8,'x'   # Check for %x (hexadecimal).
    beq     r9,r0,asm_printf_not_x
    movia   r8,out_hex_uart_0
    callr   r8
    br      asm_printf_loop
asm_printf_not_x:
    cmpeqi  r9,r8,'d'   # Check for %d (decimal).
    beq     r9,r0,asm_printf_not_d
    movia   r8,out_number_uart_0
    callr   r8
    br      asm_printf_loop
asm_printf_not_d:
    cmpeqi  r9,r8,'c'   # Check for %c (character).
    beq     r9,r0,asm_printf_not_c
    # Print character argument.
    br      asm_printf_doprint
asm_printf_not_c:
    cmpeqi  r9,r8,'s'   # Check for %s (string).
    beq     r9,r0,asm_printf_not_s
    movia   r8,out_string_uart_0
    callr   r8
    br      asm_printf_loop
asm_printf_not_s:
asm_printf_unknown:
    # We do not know what to do with other formats.
    # Print the format string text.
    movi    r4,'%'
    movia   r8,out_char_uart_0
    callr   r8
    ldb     r4,-1(r16)
    br      asm_printf_doprint
asm_printf_end:
    POP     r20
    POP     r19
    POP     r18
    POP     r17
    POP     r16
    POP     ra
    ret

#
# End of simplified printf() replacement code.
#
################################################################
.end
4

2 回答 2

2

为什么要组装?

组装通常与芯片的ISA具有 1:1 的关系。汇编指令通常具有以下形式:

<opcode> <lhs>,<rhs>

这在尽可能低的级别上工作,因此程序员可以处理各个处理器功能。

汇编仍然需要一个汇编程序才能由处理器运行。汇编器将符号汇编转换为处理器可以理解的二进制表示。

这类似于编译器的工作方式。在最高级别,编译器通常将源代码转换为抽象语法树,并从那里生成代码,这些代码可能是汇编、另一种语言的代码或机器代码。

为什么不是高级语言?

理论上可以为任何语言编写编译器以生成任意机器代码。这意味着理论上您还可以处理汇编可以访问的任何指令(例如中断),只要您的编译器知道如何解释您的更高级别的代码。

正如在这个答案中一样,大多数操作系统都会为您处理中断,并用信号将它们抽象出来。但是没有什么能阻止您自己处理它们,例如,Arduino 使这在 C++ 中成为可能。然而,问题是操作系统不允许任何程序访问中断,特别是因为某些中断需要特权CPU 模式

此外,没有什么可以阻止您使用 OO 语言(或函数式等)来实现像内核这样的低级语言,但是语言越复杂,生成高效的机器代码就越困难,当您构建其他软件将在其上运行的软件,它需要尽可能快。您将无法使用诸如垃圾收集(这在许多 OO 语言中很常见)之类的某些功能,因为在您之后没有什么可以清理的。

为什么选择 C?

OO 本身并没有使它变慢的固有特性,而汇编也没有使它变快的固有特性。当您确切知道代码在做什么时,为处理器编写代码会更容易。

当你用 C 语言编写时,你只比汇编高一级。它足够高级,可以为您提供函数、结构和变量等概念,但又足够低级,您可以对生成的代码做出合理的假设。事实上,反汇编程序是优化 C 代码的好方法。尝试使用 Java 之类的东西来做到这一点!

于 2013-03-22T06:24:01.940 回答
1

编程语言倾向于避免或最多平均化它们运行的​​处理器的能力。大多数系统上都有一些操作是必需的(从中断返回就是一个很好的例子),但不能直接用高级语言实现。有一些解决方案,一个是编译器特定的指令,在函数声明中加入一些单词中断的味道会告诉一些编译器这个函数是特殊的并且需要被包装以便它可以是一个中断处理程序,即通常不同于正常功能的框架。

因为永远不会有通用的汇编语言/指令集,你无法创建通用的高级语言特性来匹配硬件,这没有意义,仅仅用几行汇编来解决这些问题是如此微不足道,即使有一个通用指令集,我怀疑高级语言会在这些功能上浪费任何精力。

即使使用 C,也需要一定程度的引导,可能设置堆栈指针,将 .data 放在正确的位置,将 .bss 清零等。有趣的是,Cortex-M 处理器内核已经将一些东西硬编码到硬件中,这样你就可以创建几乎没有汇编的应用程序,除了 C 代码之外,只需要向量表,并且通常该向量表只是 asm 指令,但该表中可能没有 asm 指令。如果/当编译器的调用约定发生变化时,突然之间硬件不匹配并且您遇到问题,因此任何人都可以猜测它会存在多长时间。因为这些是微控制器而不是更高级别的系统,所以该解决方案无法转换。

于 2013-03-22T13:44:22.520 回答