CBZ CBNZ instruction usage in inline assembly

Hi,

I am trying to compile the following inline assembly.
I am compiling for cortex-m4 (stm32f407 device).
I tried different toolchain version, all give the same result.

__ASM volatile
(
  "ldr r2, =0x42418234\n\t"
  "label1:\n\t"
  "ldr.w r3, [r2, #0]\n\t" //load TXE bit value using bit band register
  "cbnz r3, label1\n\t"

     : // no output
     : [ptr] "r" (buf_ptr),
       [ptr_end] "r" (buf_end)
     : "cc", "r2","r3"
);

and I am getting an error:
/tmp/ccOYDN40.s: Assembler messages:
/tmp/ccOYDN40.s:397: Error: branch out of range
make: *** [src/com_proc.o] Error 1
src/subdir.mk:81: recipe for target 'src/com_proc.o' failed

I am aware that CBNZ can only branch within 4_130 bytes, which I think is the case in my code.
replacing "cbnz r3, label1\n\t" with
"cmp r3, #0\n\t"
"bne label1\n\t"

works fine.

I am not sure what I am doing wrong with cbnz instruction?