printf() problem with ported NewLib

[IanSeyler]

I'm wondering if anyone else has seen this behavior.

Example:

Code: Select all

printf("NewLib Test Application\n=======================\n");
printf("%s %d\n", "Output:", 1234);

Result:

Code: Select all

NewLib Test Application
Output: 1234

The NewLib write() function was only called to print 24 characters on the first call to printf().

[Kazinsal]

I would wager it's because of the newline and not an arbitrary 24-character limit. Try removing the newline?

[IanSeyler]

Code:

Code: Select all

printf("NewLib Test Application=======================\n");
printf("%s %d\n", "Output:", 1234);

Results:

Code: Select all

NewLib Test Application=======================
Output: 1234

Looks like it stops printing after the first newline. That first call to printf() should result with one call to write(), correct?

[Kazinsal]

My crystal ball tells me you haven't implemented isatty() and/or fstat(), so newlib is confused as to what buffering to use.

[Antti]

Does your write() correctly handle the newline?

[bluemoon]

Looks like it stops printing after the first newline. That first call to printf() should result with one call to write(), correct?

In my implementation, newlib make 2 calls to write with if there is line feed in middle of string:

Code: Select all

printf ("  PID[%d]: Hello again! counter=%d, foo=%d\n   line2\n", pid, i, g_foo.foo++);


Result:
SYSCALL : fstat(1, 7FBFF7D0)
SYSCALL : sbrk(00000430)
SYSCALL : sbrk(00000BD0)
SYSCALL : write(1, 01000010, 20)
FOO::FOO(0x10d008);
SYSCALL : write(1, 01000010, 21)
FOO::FOO(0x1000420);
SYSCALL : getpid() -> FFFFFFFF:801210D0: 1
SYSCALL : write(1, 01000010, 42)
  PID[1]: Hello again! counter=0, foo=123
SYSCALL : write(1, 01000010, 9)
   line2
SYSCALL : usleep(000F4240)

My fstat:

Code: Select all

long kservice_fstat(int fd, struct stat* st) {
    kprintf ( "SYSCALL : fstat(%d, %p)\n", fd, st);
    if ( st == NULL ) return -1;
    st->st_mode = S_IFCHR; // | S_IWUSR;
                           //    st->st_blksize = 1;
    return 0;
}                   

[IanSeyler]

Blacklight, isatty() looks fine.. it was just returning 1 be default. I changed it to return 1 for STDOUT and STDERR. fstat was just set to minimal implementation of returning 0. I'll take a further look at fstat().

Antti, write() handles the newlines correctly. write(1, "test1\ntest2\n", 12) works as expected.

[IanSeyler]

I think my issue is with sbrk(). A really simple program crashes whereas a more complicated one doesn't.

I get a page-fault with this:

Code: Select all

#include <stdio.h>

int main()
{
    printf("Hello1\nHello2\nHello3\n");
    return 0;
}

I'll do some digging and report back.

[IanSeyler]

My sbrk() is borked (I think)! I'm also thinking it could be due to me not zeroing BSS.

Should I be implementing a crt0.c file to run before calling main?

Should NewLib work correctly with flat binaries?

My linker script:

Code: Select all

OUTPUT_FORMAT("binary")
OUTPUT_ARCH("i386:x86-64")
ENTRY(main)
SECTIONS
{
        . = 0x0000000000200000;
        .text : { *(.text) }
        .data : { *(.data .rodata) QUAD(ADDR(.bss)+SIZEOF(.bss)) }
        .bss : { *(.bss) }
        end = .; _end = .; __end = .;
}

How I compile:

Code: Select all

gcc -I ../../newlib-2.0.0/newlib/libc/include/ -c helloc.c -o helloc.o
ld -T app.ld -o helloc.app helloc.o libc.a

More debugging to do!

[Owen]

I'm also thinking it could be due to me not zeroing BSS.

That is part of the contract for BSS...

[bluemoon]

I'm also thinking it could be due to me not zeroing BSS.

That is part of the contract for BSS...

LOL, I saw at least 5 people on this forum, including myself, made this mistake
I know, when you drag newlib into your OS you just can't wait to make some test application, and missed this (and other) critical parts.

[sortie]

If you don't already, you ought to be using a real cross-compiler to build your kernel and user-space. Please do that.

[dozniak]

I'm also thinking it could be due to me not zeroing BSS.

That is part of the contract for BSS...

LOL, I saw at least 5 people on this forum, including myself, made this mistake
I know, when you drag newlib into your OS you just can't wait to make some test application, and missed this (and other) critical parts.

If you're using multiboot bootloader, it clears out the bss for you. In any other case - it's up to you.

[bluemoon]

If you're using multiboot bootloader, it clears out the bss for you. In any other case - it's up to you.

NewLib Test Application=======================

I presume we are dealing with loading application, not about loading the kernel from boot loader.

[dozniak]

If you're using multiboot bootloader, it clears out the bss for you. In any other case - it's up to you.

NewLib Test Application=======================

I presume we are dealing with loading application, not about loading the kernel from boot loader.

Oh, indeed. In this case the crt should deal with that.