sys_write doesn't print a message to the terminal

Sunless · 12-07-2019, 03:49 AM

Hi guys!

I'm trying to explore GNU Assembler using this example: https://0xax.github.io/asm_1/

My code of main.s is

Quote:

.data
message: .ascii "ASSEMBLY OUTPUT"

.text
.globl _start

_start:
movq $1, %rdi
movq $1, %rax
movq $15, %rdx
movq $message, %rsi
syscall
movq $60, %rax
movq $0, %rdi
syscall

Command line preparation:

Quote:

as -g -o main.o main.s

ld -o main main.o

Now, I run ./main But it doesn't print anything and terminal is waiting my next command.

What's going wrong?

In GNU Debugger numbers were successfully stored in registers.

I see result of first syscall "-14" in RAX.

NevemTeve · 12-07-2019, 05:16 AM

It does work for me (linux/amd64). The only thing I miss is a line-feed:

Code:

$ as -g -o hello_linux.o hello_linux.s
$ ld -g -o hello_linux hello_linux.o
$ ./hello_linux
ASSEMBLY OUTPUT$

GazL · 12-07-2019, 05:39 AM

Your code works here.

Code:

$ cat hello.s
.data
message: .ascii "ASSEMBLY OUTPUT"

.text
.globl _start

_start:
movq $1, %rdi
movq $1, %rax
movq $15, %rdx
movq $message, %rsi
syscall
movq $60, %rax
movq $0, %rdi
syscall 
$ as -o hello hello.s
$ ld -o hello hello.o
$ ./hello 
ASSEMBLY OUTPUT$

errno 14 is EFAULT "bad address", which suggests it can't read the message text.

I wrote a similar example here https://www.linuxquestions.org/quest...5/#post6020662
Maybe see if that gives the same error.

Sunless · 12-07-2019, 05:53 AM

Thank you. I didn't saw result message in terminal because there is no '\n' in the message text. Program works. With .string I have 18 in RAX result (do you mean it?). Where did you found the error code description?

GazL · 12-07-2019, 05:59 AM

Quote:

I see result of first syscall "-14" in RAX.

error codes are negative of errno. Positive in %rax means it worked, returning the number of characters written.

The errno values are (eventually - includes that include includes, that include includes...) in:
/usr/include/asm-generic/errno-base.h
/usr/include/asm-generic/errno.h

man 2 write, gives explanation of the various returns that sys_write may give.

Sunless · 12-07-2019, 07:15 AM

Could I optimize my program with more code using these compilation instructions on my corei7?

Quote:

as -mtune=corei7 -o main.o main.s
ld -Ofast -o main main.o

Does the linker work with Ofast like GCC? There is no -Ofast in ld -help.

dugan · 12-07-2019, 09:56 AM

Quote:

Originally Posted by Sunless

Thank you. I didn't saw result message in terminal because there is no '\n' in the message text.

I spotted this immediately. Standard output is buffered, and typically only flushed when you output a line feed. The writes (EDITING TO CLARIFY: as in the actual display of characters on the screen) happen during flushes.

Here are some explanatory hits I got from google:

https://eklitzke.org/stdout-buffering
https://www.turnkeylinux.org/blog/unix-buffering
http://www.pixelbeat.org/programming/stdio_buffering/
https://c-for-dummies.com/blog/?p=3701

Sunless · 12-08-2019, 06:48 AM

Thank you. I have found these pages more clear to me:
https://www.computerhope.com/jargon/...descriptor.htm
https://www.computerhope.com/jargon/s/stdout.htm
https://en.wikipedia.org/wiki/Control_character

Is there something lightweight than fsync syscall?

Code:

 .data
    message: .ascii "ASSEMBLY OUTPUT MESSAGE"
    .equ one, 1
    len = . - message

     .text
    .globl _start

    _start:
    movq $one, %rdi
    movq $one, %rax
    movq $len, %rdx
    movq $message, %rsi
    syscall #sys_write
    movq $74, %rax #rdi == 1
    syscall #fsync
    movq $60, %rax
    xor %rdi, %rdi
    syscall #exit

github syscall reference

Where could I find official reference to control characters influence on stdout and stderr? Did that limited by bash in a terminal?

GazL · 12-08-2019, 07:33 AM

libc's stdout FILE stream is certainly buffered, so you might need to fflush(stdout) in some C programs, but for a asm program making a direct call of the write(2) syscall writing to the tty on FD 1, I believe the write is unbuffered and therefore you don't need to flush/fsync. (someone please correct me if I'm wrong here).

You usually only use fsync() when you need to be sure something has gotten to disk before continuing, such as writing data to disk files, and even then you normally only do it when your program needs to periodically checkpoint or when you're done with the file and are about to close the FD.

Geist · 12-16-2019, 01:02 PM

On a somewhat sidenotey sidenote:

If you are learning assembler just for fun, then maybe do it for old computers, like the c64, spectrum and co.
Why? Because, at least in my opinion, at this current point in time it is incredibly unlikely for you to do things faster than the established compilers for C, C++ etc when it comes to the PC.

This is also true for the C64 et al, yes, people have been hacking these things for 40+ years now, but you get to play with awesome things like sid chips, vic chips, actual limitations to gauge your skill in making things lightweight against, and you also have more or less a faster product cycle.

I wish I could have posted this without bumping. I could have private messaged you, yes, but I also think that assembler is really not very much needed these days, as in, highly diminishing returns, and I think that anyone should be at least somewhat aware of this.
Even arduinos and co are mostly programmed with C, etc, these days.

My post is not intended to rain on your parade or anything, but, on the offchance that you're just chasing that assembler feeling, why not go with something that has, at least, again, in my opinion, a more fun yield.

Plus, it can still be lucrative! The "8bit-guy" made a C64 game recently and I think it sold rather well. I doubt a program, or game written in assembler has that kind of marketability on the PC these days.