How the linker resolves multiple matches when processing scatter files

The linker assignment of the input section is based on a module_select_pattern and input_section_selector pair that is the most specific. However, if a unique match cannot be found, the linker faults the scatter-loading description.

The following variables describe how the linker matches multiple input sections:

• m1 and m2 represent module selector patterns.
• s1 and s2 represent input section selectors.

For example, if input section A matches m1,s1 for execution region R1, and A matches m2,s2 for execution region R2, the linker:

• Assigns A to R1 if m1,s1 is more specific than m2,s2.
• Assigns A to R2 if m2,s2 is more specific than m1,s1.
• Diagnoses the scatter-loading description as faulty if m1,s1 is not more specific than m2,s2 and m2,s2 is not more specific than m1,s1.

armlink uses the following strategy to determine the most specific module_select_pattern, input_section_selector pair:

Resolving the priority of two module_selector, section_selector pairs m1, s1 and m2, s2

The strategy starts with two module_select_pattern, input_section_selector pairs. m1,s1 is more specific than m2,s2 only if any of the following are true:

1. s1 is either a literal input section name, that is it contains no pattern characters, or a section type and s2 matches input section attributes.
2. m1 is more specific than m2.
3. s1 is more specific than s2.

The conditions are tested in order so condition 1 takes precedence over condition 2 and 3, and condition 2 takes precedence over condition 3.

Resolving the priority of two module selectors m1 and m2 in isolation

For the module selector patterns, m1 is more specific than m2 if the text string m1 matches pattern m2 and the text string m2 does not match pattern m1.

Resolving the priority of two section selectors s1 and s2 in isolation

For the input section selectors:

• If one of s1 or s2 matches the input section name or type and the other matches the input section attributes, s1 and s2 are unordered and the description is diagnosed as faulty. For example, using *(+RO) gives ambiguous matches with *(.rodata*).

• If both s1 and s2 match the input section name or type, the following relationships determine whether s1 is more specific than s2:

  • Section type is more specific than section name.
  • If both s1 and s2 match input section type, s1 and s2 are unordered and the description is diagnosed as faulty.
  • If s1 and s2 are both patterns matching section names, the same definition as for module selector patterns is used.

• If both s1 and s2 match input section attributes, the following relationships determine whether s1 is more specific than s2s:

  • ENTRY is more specific than RO-CODE, RO-DATA, RW-CODE, or RW-DATA.
  • RO-CODE is more specific than RO.
  • RO-DATA is more specific than RO.
  • RW-CODE is more specific than RW.
  • RW-DATA is more specific than RW.
  • There are no other members of the (s1 more specific than s2) relationship between section attributes.

This matching strategy has the following consequences:

• Descriptions do not depend on the order they are written in the file.
• Generally, the more specific the description of an object, the more specific the description of the input sections it contains.

• The input_section_selectors are not examined unless:

  • Object selection is inconclusive.
  • One selector specifies a literal input section name or a section type and the other selects by attribute. In this case, the explicit input section name or type is more specific than any attribute. This is true even if the object selector associated with the input section name is less specific than that of the attribute.

The .ANY module selector is available to assign any sections that cannot be resolved from the scatter-loading description.

Example

The following example shows multiple execution regions and pattern matching:

LR_1 0x040000            
{                            
    ER_ROM 0x040000              ; The startup exec region address is the same
    {                            ; as the load address.  
        application.o (+ENTRY)   ; The section containing the entry point from
    }                            ; the object is placed here.
    ER_RAM1 0x048000      
    {
        application.o (+RO-CODE) ; Other RO code from the object goes here
    }
    ER_RAM2 0x050000     
    {
        application.o (+RO-DATA) ; The RO data goes here
    }
    ER_RAM3 0x060000 
    {
        application.o (+RW)      ; RW code and data go here
    }
    ER_RAM4 +0                   ; Follows on from end of ER_R3
    {
        *.o (+RO, +RW, +ZI)      ; Everything except for application.o goes here
    }
}