The nano_realloc() function in newlib/libc/stdlib/mallocr.c does a memcpy with size equal
to the realloc parameter 'size', this can be a problem as we are copying from the original location more data than needed with the risk of triggering some faults in case the access to that memory is protected or not accessible. The implementation should use the old size for copying when the size increases.

It'd be great to see in revision history what is the difference from newlib upstream tree... What I would do is

1. check in the newlib as is first with revision hash in commit message
2. copy over the changes and make another commit.
   This way it's would easier to maintain the fork on let others diff it easily between the initial check-in and the head.

Although, it would be a good idea to try and pull-request your changes into upstream with preprocessor directives around the parts you consider cutting out.

The putchar macro is not calling the __sinit() function.
If putchar is called before any call to stdio function calling __sinit() (like printf()) ; putchar attemps to write into the const struct __sf_fake_stdout.

This is current implementation in code base:

if a is an even integer or any integer with 0s at lower significant followed by 1s. then it will end up in an infinite loop.

Correct code should be

while(a)
{
    if (a & 1) r += b;
    b <<= 1;
    a >>= 1;
}