Re: [Fwd: Re: [linker] relaxation question]

[Nick Clifton <nickc at redhat dot com>]

Hi Swami,

> Any hint to get the relocs for .uleb128 also ?

These are much harder to handle.  (Which is probably why the mn10300 
port does not do anything special for them at the moment).

The problem is that LEB128 values are variable in length.  In theory the 
correct thing to do is:

  * If linker relaxation is not enabled then emit the LEB128 value
    as normal.

  * If the expression does not involve any symbols then emit the
    LEB128 value as normal.

  * If the expression is the difference of two symbols which are
    in the same section and this sections will not be relaxed then
    emit the LEB128 value as normal.

  * Otherwise emit a LEB128 value which is the maximum possible
    length for the target architecture.  (Ie: 16-bit, 32-bit or
    64-bit).  Then also emit one or two relocations to indicate
    that an LEB128 value needs to be handled at this location.
    (You will probably have to create new relocations for this).
    Emit one relocation if there is only one symbol in the original
    expression.  Emit two relocations if the original expression
    is the difference of two symbols.

  * In the linker you must make sure that in addition to relaxing
    the normal sections you also relax any section that contains
    any of the LEB128 relocations.  Ie you relax the LEB128 values.

  * Finally in the linker, when you are processing relocations
    you must add code to handle these LEB128 values and compute
    the correct value to be placed into the binary.  (You may
    actually choose to do this computation and insertion at the
    same time as you relax the LEB128 relocations, but
    conceptually there are two separate operations).

Cheers
  Nick