Skip to content

Latest commit

 

History

History
3346 lines (2696 loc) · 211 KB

aaelf32.rst

File metadata and controls

3346 lines (2696 loc) · 211 KB

ELF for the Arm® Architecture

2020Q4

Date of Issue: 21st December 2020

Arm_logo_blue_150MN.png

This document describes the processor-specific definitions for ELF for the Application Binary Interface (ABI) for the Arm architecture.

Object files, file formats, linking, EABI, ELF

Please check Application Binary Interface for the Arm® Architecture for the latest release of this document.

Please report defects in this specification to the issue tracker page on GitHub.

This work is licensed under the Creative Commons Attribution-ShareAlike 4.0 International License. To view a copy of this license, visit http://creativecommons.org/licenses/by-sa/4.0/ or send a letter to Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

Grant of Patent License. Subject to the terms and conditions of this license (both the Public License and this Patent License), each Licensor hereby grants to You a perpetual, worldwide, non-exclusive, no-charge, royalty-free, irrevocable (except as stated in this section) patent license to make, have made, use, offer to sell, sell, import, and otherwise transfer the Licensed Material, where such license applies only to those patent claims licensable by such Licensor that are necessarily infringed by their contribution(s) alone or by combination of their contribution(s) with the Licensed Material to which such contribution(s) was submitted. If You institute patent litigation against any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Licensed Material or a contribution incorporated within the Licensed Material constitutes direct or contributory patent infringement, then any licenses granted to You under this license for that Licensed Material shall terminate as of the date such litigation is filed.

As identified more fully in the License section, this project is licensed under CC-BY-SA-4.0 along with an additional patent license. The language in the additional patent license is largely identical to that in Apache-2.0 (specifically, Section 3 of Apache-2.0 as reflected at https://www.apache.org/licenses/LICENSE-2.0) with two exceptions.

First, several changes were made related to the defined terms so as to reflect the fact that such defined terms need to align with the terminology in CC-BY-SA-4.0 rather than Apache-2.0 (e.g., changing “Work” to “Licensed Material”).

Second, the defensive termination clause was changed such that the scope of defensive termination applies to “any licenses granted to You” (rather than “any patent licenses granted to You”). This change is intended to help maintain a healthy ecosystem by providing additional protection to the community against patent litigation claims.

Contributions to this project are licensed under an inbound=outbound model such that any such contributions are licensed by the contributor under the same terms as those in the License section.

The text of and illustrations in this document are licensed by Arm under a Creative Commons Attribution–Share Alike 4.0 International license ("CC-BY-SA-4.0”), with an additional clause on patents. The Arm trademarks featured here are registered trademarks or trademarks of Arm Limited (or its subsidiaries) in the US and/or elsewhere. All rights reserved. Please visit https://www.arm.com/company/policies/trademarks for more information about Arm’s trademarks.

Copyright (c) 2005-2020, Arm Limited and its affiliates. All rights reserved.

Contents

The following support level definitions are used by the Arm ABI specifications:

Release
Arm considers this specification to have enough implementations, which have received sufficient testing, to verify that it is correct. The details of these criteria are dependent on the scale and complexity of the change over previous versions: small, simple changes might only require one implementation, but more complex changes require multiple independent implementations, which have been rigorously tested for cross-compatibility. Arm anticipates that future changes to this specification will be limited to typographical corrections, clarifications and compatible extensions.
Beta
Arm considers this specification to be complete, but existing implementations do not meet the requirements for confidence in its release quality. Arm may need to make incompatible changes if issues emerge from its implementation.
Alpha
The content of this specification is a draft, and Arm considers the likelihood of future incompatible changes to be significant.

All content in this document is at the Release quality level.

Issue Date Change
1.0 24th March 2005 First public release.
1.01 5th July 2005 Defined in Section Types, Special Sections SHT_ARM_PREEMPTMAP; corrected the erroneous value of SHT_ARM_ATTRIBUTES.
1.02 6th January 2006 Minor correction to definition of e_entry (ELF Header). Clarified restrictions on local symbol removal in relocatable files (Symbol names). Clarified the definition of R_ARM_RELATIVE when S = 0 (Dynamic relocations). Added material describing architecture compatibility for executable files (Platform architecture compatibility data).
1.03 5th May 2006 Clarified that bit[0] of [e_entry] controls the instruction set selection on entry. Added rules governing SHF_MERGE optimizations (Merging of objects in sections with SHF_MERGE). Added material describing initial addends for REL-type relocations (Addends and PC-bias compensation).
1.04 25th January 2007 In Relocation corrected the definition of R_ARM_ALU_(PC|SB)_Gn_NC, R_ARM_THM_PC8, R_ARM_THM_PC12, and R_ARM_THM_ALU_PREL_11_0. Added a table of 32-bit thumb relocations. In Relocation types and Relocations for thread-local storage, added new relocations to support an experimental Linux TLS addressing model. In Platform architecture compatibility data reduced the field masked by PT_ARM_ARCHEXT_ARCHMSK to 8 bits (no current value exceeds 4 bits).
1.05 25th September 2007 Correct definition of Pa in Relocation types (the bit-mask was incorrect). Corrected spelling of TLS relocations in Relocations for thread-local storage.
A 25th October 2007 Document renumbered (formerly GENC-003538 v1.05).
B 2nd April 2008 Corrected error in Static Thumb-32 instruction relocations where instructions for R_ARM_THM_PC12 and R_ARM_THM_ALU_PREL_11_0 had been transposed.
C 10th October 2008 In Static Arm relocations, specified which relocations are permitted to generate veneers corrupting ip. In Dynamic relocations specified the meaning of dynamic meaning of dynamic relocations relocations R_ARM_TLS_DTPMOD32 and R_ARM_TLS_TPOFF32 when the symbol is NULL. Reserved vendor-specific section numbers and names to the DBGOVL32 ABI extension. Clarified use of the symbol by R_ARM_V4BX.
D 28th October 2009 Added http://infocenter.arm.com/ references to the recently published [ARMARM] and the [ARMv5ARM]; in Static Thumb32 relocations cross-referenced permitted veneer-generation. In Static Thumb-16 relocations, extended R_ARM_THM_PC8 to ADR as well as LDR(literal). Updated and tidied Platform architecture compatibility data and added Platform architecture compatibility data (ABI format) as a proposal for recording executable file attributes.
E 30th November 2012 In Arm-specific e_flags, added ELF header e_flags to indicate floating point PCS conformance and a mask for legacy bits. In Relocation, standardized instruction descriptions to use Arm ARM terminology. In Addends and PC-bias compensation, clarified initial addend formulation for MOVW/MOVT and R_ARM_THM_PC8. In Relocation codes table, reserved relocation 140 for a specific future use. In Arm relocation actions by instruction type, added entries for MOVW and MOVT; in subsection Call and Jump Relocations: grouped R_ARM_THM_CALL with the other Thumb relocations, and in the final paragraph changed the behaviour of jump relocations to unresolved weak references to be implementation-defined rather than undefined. In Static Thumb-16 relocations, added Overflow column. In Static Thumb-32 instruction relocations, corrected Result Mask for R_ARM_THM_PC12; added Thumb relocation actions by instruction type; corrected final paragraph to clarify the cross-reference to call and jump relocations. In Relocation types, Static Thumb32 relocations, Proxy generating relocations, added R_ARM_THM_GOT_BREL12. In Dynamic relocations, clarified the wording for R_ARM_RELATIVE. In Platform architecture compatibility data (ABI format), corrected off-by-one error in size of array.
F 24th November 2015 In Relocation codes table, changed the subdivisions within the reserved/unallocated relocation space (136-255). Renumbered R_ARM_IRELATIVE from 140 to 160 (the number agreed with stakeholders; publication as 140 was incorrect). In Static Arm instruction relocations, removed incorrect overflow check on R_ARM_MOVT_ABS, R_ARM_MOVT_PREL and R_ARM_MOVT_BREL. Clarified in Relocation types that relocation expression values are computed mod 232. In Relocation, added R_ARM_THM_ALU_ABS_Gn[_NC] relocations. In Section Attribute Flags, added SHF_ARM_NOREAD processor specific section attribute flag.
2018Q4 21st December 2018

In Section Attribute Flags, renamed SHF_ARM_NOREAD to SHF_ARM_PURECODE, relaxed definition.

In Private relocations, expanded private relocation space to 32 relocations, and clarified relationship with EI_OSABI.

In ELF Identification, added EI_OSABI value for ELFOSABI_ARM_FDPIC.

2020Q4 21st December 2020

This document refers to, or is referred to by, the documents listed in the following table.

Ref External URL Title
AAPCS32   Procedure Call Standard for the Arm Architecture
AAELF32 This document ELF for the Arm Architecture
BSABI32   ABI for the Arm Architecture (Base Standard)
EHABI32   Exception Handling ABI for the Arm Architecture
Addenda32   Addenda to, and errata in, the ABI for the Arm Architecture
DBGOVL32   Support for Debugging Overlaid Programs
ARMARM https://developer.arm.com/docs/ddi0406/c/arm-architecture-reference-manual-armv7-a-and-armv7-r-edition Arm DDI 0406: Arm Architecture Reference Manual Arm v7-A and Arm v7-R edition
https://developer.arm.com/products/architecture/m-profile/docs/ddi0403/e/armv7-m-architecture-reference-manual Arm DDI 0403C: Armv7-M Architecture Reference Manual
ARMv5ARM https://developer.arm.com/docs/ddi0100/latest/armv5-architecture-reference-manual Arm DDI 0100I: Armv5 Architecture Reference Manual
GDWARF http://dwarfstd.org/Dwarf3Std.php DWARF 3.0, the generic debug table format
LSB http://refspecs.linuxfoundation.org/lsb.shtml Linux Standards Base
SCO-ELF http://www.sco.com/developers/gabi/2003-12-17/contents.html System V Application Binary Interface – DRAFT – 17 December 2003
SYM-VER http://www.akkadia.org/drepper/symbol-versioning GNU Symbol Versioning
FDPIC https://github.com/mickael-guene/fdpic_doc FDPIC ABI

The ABI for the Arm Architecture uses the following terms and abbreviations.

AAPCS
Procedure Call Standard for the Arm Architecture
ABI

Application Binary Interface:

  1. The specifications to which an executable must conform in order to execute in a specific execution environment. For example, the Linux ABI for the Arm Architecture.
  2. A particular aspect of the specifications to which independently produced relocatable files must conform in order to be statically linkable and executable. For example, the C++ ABI for the Arm Architecture, the Run-time ABI for the Arm Architecture, the C Library ABI for the Arm Architecture.
AEABI
(Embedded) ABI for the Arm architecture (this ABI…)
Arm-based
... based on the Arm architecture ...
core registers
The general purpose registers visible in the Arm architecture’s programmer’s model, typically r0-r12, SP, LR, PC, and CPSR.
EABI
An ABI suited to the needs of embedded, and deeply embedded (sometimes called free standing), applications.
Q-o-I
Quality of Implementation – a quality, behavior, functionality, or mechanism not required by this standard, but which might be provided by systems conforming to it. Q-o-I is often used to describe the tool-chain-specific means by which a standard requirement is met.
VFP
The Arm architecture’s Floating Point architecture and instruction set. In this ABI, this abbreviation includes all floating point variants regardless of whether or not vector (V) mode is supported.

This specification has been developed with the active support of the following organizations. In alphabetical order: Arm, CodeSourcery, Intel, Metrowerks, Montavista, Nexus Electronics, PalmSource, Symbian, Texas Instruments, and Wind River.

This specification provides the processor-specific definitions required by ELF [SCO-ELF] for Arm based systems.

The ELF specification is part of the larger System V ABI specification where it forms chapters 4 and 5. However, the specification can be used in isolation as a generic object and executable format.

Platform Standards of this document covers ELF related matters that are platform specific. Most of this material is related to the Base Platform ABI.

Object Files and Program Loading and Dynamic Linking of this document are structured to correspond to chapters 4 and 5 of the ELF specification. Specifically:

There are several drafts of the ELF specification on the SCO web site. This specification is based on the December 2003 draft, which was the most recent stable draft at the time this specification was developed.

The BPABI is an abstract platform standard. Platforms conforming to the BPABI can generally share a common toolchain with minimal post-processing requirements.

The BPABI uses the GNU-extended Solaris symbol versioning mechanism [SYM-VER].

Concrete data structure descriptions can be found in /usr/include/sys/link.h (Solaris), /usr/include/elf.h (Linux), in the Linux base specifications [LSB], and in Drepper’s paper [SYM-VER]. Drepper provides more detail than the summary here.

An object or executable file using symbol versioning shall set the EI_OSABI field in the ELF header to ELFOSABI_ARM_AEABI or some other appropriate operating-system specific value.

4.1.1.1   Symbol versioning sections

Symbol versioning adds three sections to an executable file (under the SVr4 ABI these are included in the RO program segment). Each section can be located via a DT_xxx entry in the file’s dynamic section.

  • The version definitions section. This section defines:

    • The symbol versions associated with symbols exported from this executable file.
    • The version of the file itself.
  • The version section.

    This section extends the dynamic symbol table with an extra Elf32_Half field for each symbol. The Nth entry gives the index in the virtual table of versions (described below) of the version associated with the Nth symbol.

  • The versions needed section.

    This section describes the versions referred to by symbols not defined in this executable file. Each entry names a DSO and points to a list of versions needed from it. In effect this represents FROM DSO IMPORT Ver1, Ver2, …. This section provides a record of the symbol bindings used by the static linker when the executable file was created.

In standard ELF style, both the version definitions section and the versions needed section identify (via the sh_link field in their section headers) a string table section (often .dynstr) containing the textual values they refer to.

The (virtual) table of versions

When an executable file uses symbol versioning there is also a virtual table of versions. This is not represented in the file (there is no corresponding file component). It contains a row for each distinct version defined by, and needed by, this file.

Each version defined, and each version needed, by this file carries its row index in this virtual table, so the table can be constructed on demand. Indexes 2, 3, 4, and so on, are local to this file. Indexes 0 and 1 have predefined global meanings, as do indexes with the top bit (0x8000) set.

4.1.1.2   Locating symbol versioning sections

The version definition section can be located via keys in the dynamic section, as follows.

DT_VERDEF (0x6FFFFFFC) address
DT_VERDEFNUM (0x6FFFFFFD) count

This key pair identifies the head and length, of a list of version definitions exported from this executable file. The list is not contiguous – each member points to its successor.

The versions needed section can be located via keys in the dynamic section, as follows.

DT_VERNEED (0x6FFFFFFE) address
DT_VERNEEDNUM (0x6FFFFFFF) count

This key pair identifies the head and length of a list of needed versions. Each list member identifies a DSO imported from, and points to a sub-list of versions used by symbols imported from that DSO at the time this executable file was created by the static linker. Neither list need be contiguous – each member points to its successor.

The version section can be located via a key in the dynamic section, as follows.

DT_VERSYM (0x6FFFFFF0) address

The version section adds a field to each dynamic symbol that identifies the version of that symbol’s definition, or the version of that symbol needed to satisfy that reference. The number of entries must be same as the number of entries in the dynamic symbol table identified by DT_SYMTAB and DT_HASH (and by the Arm-specific tag DT_ARM_SYMTABSZ).

4.1.1.3   Version definition section

The version definition section has the name .XXX_verdef and the section type SHT_XXX_verdef (the names vary but the section type – 0x6FFFFFFD – is the same for Solaris and Linux). Its sh_link field identifies the string table section (often .dynstr) it refers to.

The version definition section defines a set of versions exported from this file and the successor relationships among them.

Each version has a textual name, and two versions are the same if their names compare equal. Textual names are represented by offsets into the associated string table section. Names that must be processed during dynamic linking are also hashed using the standard ELF hash function [SCO-ELF].

Each version definition is linked to the next version definition via it vd_next field which contains the byte offset from the start of this version definition to the start of the next one. Zero marks the end of the list.

Each symbol exported from this shared object refers, via an index in the version section, to one of these version definitions. If bit 15 of the index is set, the symbol is hidden from static binding because it has an old version.

During static linking against this shared object, an undefined symbol can only match an identically named STB_GLOBAL definition which refers to one of these version definitions via an index with bit 15 clear.

Each top-level version definition links via its vd_aux field to a list of version names. Each link contains the byte offset between the start of the structure containing it and the start of the structure linked to. Zero marks the end of the list. The first member of the list names the latest version, hashed in the version definition’s vd_hash field. Subsequent members name predecessor versions, but these are irrelevant to both static and dynamic linking.

4.1.1.4   Symbol version section

The symbol version section has the name .XXX_versym and the section type SHT_XXX_versym (the names vary but the section type – 0x6FFFFFFF – is the same for Solaris and Linux).

The symbol version section is a table of ELF32_Half values. The Nth entry in the section corresponds to the Nth symbol in the dynamic symbol table.

  • 0 if the symbol is local to this executable file.
  • 1 if the symbol is undefined and unbound (to be bound dynamically), or if the symbol is defined and names the version of the executable file (usually a shared object) itself.
  • The index (> 1) of the corresponding version definition, or version needed, in the virtual table of versions (described in Symbol versioning sections).

This is the same value as is stored in the vd_ndx field of a version definition structure and the vna_other field of a version needed auxiliary structure.

Bit 15 of the index is set to denote that this is an old version of the symbol. Such symbols are not used during static binding, but may be linked to during dynamic linking.

4.1.1.5   Versions needed section

The versions needed section has the name .XXX_verneed and the section type SHT_XXX_verneed (the names vary but the section type – 0x6FFFFFFE – is the same for Solaris and Linux). Its sh_link field identifies the string table section (often .dynstr) it refers to.

The versions needed section contains a list of needed DSOs, and the symbol versions needed from them.

Within each version needed structure, the vn_file field is the offset in the associated string section of the SONAME of the needed DSO, and the vn_next field contains the byte offset from the start of this version needed structure to the start of its successor.

Each version needed structure links to a sub-list of needed versions via a byte offset to the start of the first member in its vn_aux field. In effect this represents FROM DSO IMPORT Ver1, Ver2, ...

Each version needed auxiliary structure contains its index in the virtual table of versions in its vna_other field. The vna_name field contains the offset in the associated string table of the name of the required version.

Under SVr4, symbol pre-emption occurs at dynamic link time, controlled by the dynamic linker, so there is nothing to encode in a DSO.

In the DLL-creating tool flow, pre-emption happens off line and must be recorded in a BPABI executable file in a form that can be conveniently processed by a post linker. If there is to be any pre-emption when a process is created, what to do must be recorded in the platform executable produced by the post linker.

4.1.2.1   Pre-emption Map Format

Static preemption data is recorded in a special section in the object file. The map is recorded in the dynamic section with the tag DT_ARM_PREEMPTMAP, which contains the virtual address of the map.

In the section view, the pre-emption map special section is called .ARM.preemptmap. It has type SHT_ARM_PREEMPTMAP. In common with other sections that refer to a string table, its sh_link field contains the section index of an associated string table.

The map contains a sequence of entries of the form:

Elf32_Word count                      // Count of pre-empted definitions following
Elf32_Word symbol-name                // Offset in the associated string table
Elf32_Word pre-empting-DLL            // Offset in the associated string table
Elf32_Word pre-empted-DLL             // Offset in the associated string table
...                                   //

The map is terminated by a count of zero.

If count is non-zero, the next two words identify the name of the symbol being pre-empted and the name (SONAME) of the executable file providing the pre-empting definition. This structure is followed by count words each of which identifies the SONAME of an executable file whose definition of symbol-name is pre-empted.

Symbol-name is the offset in the associated string table section of a NUL-terminated byte string (NTBS) that names a symbol defined in a dynamic symbol table. This value must not be 0.

Each of pre-empting-DLL and pre-empted-DLL is an offset in the associated string table section of an NTBS naming a DLL. The name used is the shared object name (SONAME) cited by DT_NEEDED dynamic tags. The root executable file does not have a SONAME, so its name is encoded as 0.

4.1.3.1   Symbols for which a PLT entry must be generated

A PLT entry implements a long-branch to a destination outside of this executable file. In general, the static linker knows only the name of the destination. It does not know its address or instruction-set state. Such a location is called an imported location or imported symbol.

Some targets (specifically SVr4-based DSOs) also require functions exported from an executable file to have PLT entries. In effect, exported functions are treated as if they were imported, so that their definitions can be overridden (pre-empted) at dynamic link time.

A linker must generate a PLT entry for each candidate symbol cited by a BL-class relocation directive.

  • For an SVr4-based DSO, each STB_GLOBAL symbol with STV_DEFAULT visibility is a candidate.
  • For all other platforms conforming to this ABI, only non-WEAK, not hidden (by STV_HIDDEN), undefined, STB_GLOBAL symbols are candidates.

Note

When targeting DLL-based and bare platforms, relocations that cite WEAK undefined symbols must be performed by the static linker using the appropriate NULL value of the relocation. No WEAK undefined symbols are copied to the dynamic symbol table. WEAK definitions may be copied to the dynamic table, but it is Q-o-I whether a dynamic linker will take any account of the WEAK attribute. In contrast, SVr4-based platforms process WEAK at dynamic link time.

4.1.3.2   Overview of PLT entry code generation

A PLT entry must be able to branch any distance to either instruction-set state. The span and state are fixed when the executable is linked dynamically. A PLT entry must therefore end with code similar to the following.

Arm V5 and later Arm V4T
LDR pc, Somewhere

LDR ip, Somewhere

BX ip

Somewhere: DCD Destination

Note

There is no merit in making the final step PC-relative. A location must be written at dynamic link time and at that time the target address must be known [even if dynamic linking is performed off line]. Similarly, it is generally pointless trying to construct a PLT entry entirely in 16-bit Thumb instructions. Even with the overhead of an inline Thumb-to-Arm state change, an Arm-state entry is usually smaller and always faster.

The table below summarizes the code generation variants a static linker must support. PLT refers to the read-only component of the veneer and PLTGOT to the corresponding writable function pointer.

PLT code generation options
Platform family Neither ROM replaceable nor free of dynamic relocations ROM replaceable, or PLT is free of dynamic relocations
DLL-like, single address space (Palm OS-like)

PLT code loads a function pointer from the PLT, for example:

      LDR pc, LX,
LX    DCD R_ARM_GLOB_DAT(X)
PLT code loads the PLTGOT entry SB-relative (DLL-like, single address space, PLT linkage)
DLL-like, multiple virtual address spaces (Symbian OS-like) PLT code loads a function pointer from the PLT (code and dynamic relocation as shown above). PLT code loads the PLTGOT entry via an address constant in the PLT (DLL-like, multiple virtual address space, PLT linkage)
SVr4-like (Linux-like) Not applicable, but as above if it were. PLT code loads the PLTGOT entry PC-relative (SVr4 DSO-like PLT linkage).

Following subsections present specimen Arm code sequences appropriate to the right hand column. In each case simplification to the direct (no PLTGOT) case is shown in the left hand column.

Note also that:

  • In each case we assume Arm architecture V5 or later, and omit the 4-byte Thumb-to-Arm prelude that is needed to support Thumb-state callers.
  • Under Arm architecture V4T, in the two DLL cases shown in the first column above, the final LDR pc, …, can be replaced by LDR ip, …; BX ip.
  • In the case of SVr4 linkage there is an additional constraint to support incremental dynamic linking, namely that ip must address the corresponding PLTGOT entry. This constraint is most easily met under architecture V4T by requiring DSOs to be entered in Arm state (but more complex solutions are possible).
  • Other platforms are free to impose the same constraint if they support incremental dynamic linking.
4.1.3.3   PLT relocation

A post linker may need to distinguish PLTGOT-generating relocations from GOT-generating ones.

If the static linker were generating a relocatable ELF file it would naturally generate the PLT into its own section (.plt, say), subject to relocations from a corresponding relocation section (.rel.plt say). No other GOT-generating relocations can occur in .rel.plt, so that section would contain all the PLTGOT-generating relocations. By the usual collation rules of static linking, in a subsequent executable file-producing link step those relocations would end up in a contiguous sub-range of the dynamic relocation section.

The ELF standard requires that the GOT-generating relocations of the PLT are emitted into a contiguous sub-range of the dynamic relocation section. That sub-range is denoted by the standard tags DT_JMPREL and DT_PLTRELSZ. The type of relocations (REL or RELA) is stored in the DT_PLTREL tag.

Various symbols and names may require a vendor-specific name to avoid the potential for name-space conflicts. The list of currently registered vendors and their preferred short-hand name is given in the following table. Tools developers not listed are requested to co-ordinate with Arm to avoid the potential for conflicts.

Registered Vendors
Name Vendor
ADI Analog Devices
acle Reserved for use by Arm C Language Extensions.
aeabi Reserved to the ABI for the Arm Architecture (EABI pseudo-vendor)
AnonXyz
anonXyz
Reserved to private experiments by the Xyz vendor. Guaranteed not to clash with any registered vendor name.
ARM Arm Ltd (Note: the company, not the processor).
cxa C++ ABI pseudo-vendor
FSL Freescale Semiconductor Inc.
GHS Green Hills Systems
gnu GNU compilers and tools (Free Software Foundation)
iar IAR Systems
icc ImageCraft Creations Inc (ImageCraft C Compiler)
intel Intel Corporation
ixs Intel Xscale
llvm The LLVM/Clang projects
PSI PalmSource Inc.
RAL Rowley Associates Ltd
SEGGER SEGGER Microcontroller GmbH
somn SOMNIUM Technologies Limited.
TASKING Altium Ltd.
TI TI Inc.
tls Reserved for use in thread-local storage routines.
WRS Wind River Systems.

To register a vendor prefix with Arm, please E-mail your request to arm.eabi at arm.com.

The ELF header provides a number of fields that assist in interpretation of the file. Most of these are specified in the base standard. The following fields have Arm-specific meanings.

e_type
There are currently no Arm-specific object file types. All values between ET_LOPROC and ET_HIPROC are reserved to future revisions of this specification.
e_machine
An object file conforming to this specification must have the value EM_ARM (40, 0x28).
e_entry

The value stored in this field is treated like any other code pointer. Specifically, if bit[0] is 0b1 then the entry point contains Thumb code; while bit[1:0] = 0b00 implies that the entry point contains Arm code. The combination bit[1:0] = 0b10 is reserved.

The base ELF specification requires this field to be zero if an application does not have an entry point. Nonetheless, some applications may require an entry point of zero (for example, via the reset vector).

A platform standard may specify that an executable file always has an entry point, in which case e_entry specifies that entry point, even if zero.

e_flags
The processor-specific flags are shown in the following table. Unallocated bits, and bits allocated in previous versions of this specification, are reserved to future revisions of this specification.
Arm-specific e_flags
Value Meaning
EF_ARM_ABIMASK (0xFF000000) (current version is 0x05000000) This masks an 8-bit version number, the version of the ABI to which this ELF file conforms. This ABI is version 5. A value of 0 denotes unknown conformance.
EF_ARM_BE8 (0x00800000) The ELF file contains BE-8 code, suitable for execution on an Arm Architecture v6 processor. This flag must only be set on an executable file.
EF_ARM_GCCMASK (0x00400FFF) Legacy code (ABI version 4 and earlier) generated by gcc-arm-xxx might use these bits.
EF_ARM_ABI_FLOAT_HARD (0x00000400) (ABI version 5 and later)

Set in executable file headers (e_type = ET_EXEC or ET_DYN) to note that the executable file was built to conform to the hardware floating-point procedure-call standard.

Compatible with legacy (pre version 5) gcc use as EF_ARM_VFP_FLOAT.

EF_ARM_ABI_FLOAT_SOFT (0x00000200) (ABI version 5 and later)

Set in executable file headers (e_type = ET_EXEC or ET_DYN) to note explicitly that the executable file was built to conform to the software floating-point procedure-call standard (the base standard). If both EF_ARM_ABI_FLOAT_XXXX bits are clear, conformance to the base procedure-call standard is implied.

Compatible with legacy (pre version 5) gcc use as EF_ARM_SOFT_FLOAT.

The 16-byte ELF identification (e_ident) provides information on how to interpret the file itself. The following values shall be used on Arm systems

EI_CLASS
An Arm ELF file shall contain ELFCLASS32 objects.
EI_DATA
This field may be either ELFDATA2LSB or ELFDATA2MSB. The choice will be governed by the default data order in the execution environment. On an Architecture v6 processor operating in BE8 mode all instructions are in little-endian format. An executable image suitable for operation in this mode will have EF_ARM_BE8 set in the e_flags field.
EI_OSABI
This field shall be zero unless the file uses objects that have flags which have OS-specific meanings (for example, it makes use of a section index in the range SHN_LOOS through SHN_HIOS). Processor-specific values for this field are defined in the following table.
Arm-specific EI_OSABI values
Value Meaning
ELFOSABI_ARM_AEABI (64) The object contains symbol versioning extensions as described in Symbol Versioning.
ELFOSABI_ARM_FDPIC (65) The object uses relocations in the private range, with semantics defined by [FDPIC].

There are no processor-specific special section indexes defined. All processor-specific values are reserved to future revisions of this specification.

The defined processor-specific section types are listed in the following table. All other processor-specific values are reserved to future revisions of this specification.

Processor specific section types
Name Value Comment
SHT_ARM_EXIDX 0x70000001 Exception Index table
SHT_ARM_PREEMPTMAP 0x70000002 BPABI DLL dynamic linking pre-emption map
SHT_ARM_ATTRIBUTES 0x70000003 Object file compatibility attributes
SHT_ARM_DEBUGOVERLAY 0x70000004 See DBGOVL32 for details.
SHT_ARM_OVERLAYSECTION 0x70000005

Pointers in sections of types SHT_INIT_ARRAY, SHT_PREINIT_ARRAY and SHT_FINI_ARRAY shall be expressed either as absolute values or relative to the address of the pointer; the choice is platform defined. In object files the relocation type R_ARM_TARGET1 may be used to indicate this target-specific relocation processing.

SHT_ARM_EXIDX marks a section containing index information for exception unwinding. See EHABI32 for details.

SHT_ARM_PREEMPTMAP marks a section containing a BPABI DLL dynamic linking pre-emption map. See Pre-emption Map Format.

SHT_ARM_ATTRIBUTES marks a section containing object compatibility attributes. See Build Attributes.

The defined processor-specific section attribute flags are listed in the following table. All other processor-specific values are reserved to future revisions of this specification.

Processor specific section attribute flags
Name Value Comment
SHF_ARM_PURECODE 0x20000000 The contents of this section contains only program instructions and no program data

If any section contained by a segment does not have the SHF_ARM_PURECODE section flag set, the PF_R segment flag must be set in the program header for the segment. If all sections contained by a segment have the SHF_ARM_PURECODE section flag, a linker may optionally clear the PF_R segment flag in the program header of the segment, to signal to the runtime that the program does not rely on being able to read that segment.

5.3.3.1   Merging of objects in sections with SHF_MERGE

In a section with the SHF_MERGE flag set, duplicate used objects may be merged and unused objects may be removed. An object is used if:

  • A relocation directive addresses the object via the section symbol with a suitable addend to point to the object.
  • A relocation directive addresses a symbol within the section. The used object is the one addressed by the symbol irrespective of the addend used.

The following table lists the special sections defined by this ABI.

Arm special sections
Name Type Attributes
.ARM.exidx* SHT_ARM_EXIDX SHF_ALLOC + SHF_LINK_ORDER
.ARM.extab* SHT_PROGBITS SHF_ALLOC
.ARM.preemptmap SHT_ARM_PREEMPTMAP SHF_ALLOC
.ARM.attributes SHT_ARM_ATTRIBUTES none
.ARM.debug_overlay SHT_ARM_DEBUGOVERLAY none
.ARM.overlay_table SHT_ARM_OVERLAYSECTION See DBGOVL32 for details

Names beginning .ARM.exidx name sections containing index entries for section unwinding. Names beginning .ARM.extab name sections containing exception unwinding information. See [EHABI] for details.

.ARM.preemptmap names a section that contains a BPABI DLL dynamic linking pre-emption map. See Pre-emption Map Format.

.ARM.attributes names a section that contains build attributes. See Build Attributes.

.ARM.debug_overlay and .ARM.overlay_table name sections used by the Debugging Overlaid Programs ABI extension described in DBGOVL32.

Additional special sections may be required by some platforms standards.

There is no minimum alignment required for a section. However, sections containing thumb code must be at least 16-bit aligned and sections containing Arm code must be at least 32-bit aligned.

Platform standards may set a limit on the maximum alignment that they can guarantee (normally the page size).

Build attributes are encoded in a section of type SHT_ARM_ATTRIBUTES, and name .ARM.attributes.

The content of the section is a stream of bytes. Numbers other than subsection sizes are encoded numbers using unsigned LEB128 encoding (ULEB128), DWARF-3 style [GDWARF].

Attributes are divided into sub-sections. Each subsection is prefixed by the name of the vendor. There is one subsection that is defined by the “aeabi” pseudo-vendor and contains general information about compatibility of the object file. Attributes defined in vendor-specific sections are private to the vendor. In a conforming object file the information recorded in a vendor-specific section may be safely ignored if it is not understood.

Most build attributes naturally apply to a whole translation unit; however, others might apply more naturally to a section or to a function (symbol of type STT_FUNC). To permit precise description of attributes the syntax permits three granularities of translation at which an attribute can be expressed.

A section inherits the attributes of the file of which it is a component. A symbol definition inherits the attributes of the section in which it is defined. Attributes that cannot apply to the smaller entity are not inherited.

Note

Attributes that naturally apply to a translation unit may, nonetheless, end up applying to a section if sections from distinct relocatable files are combined into a single relocatable file by "partial linking". Similar exceptions may occur at the function level through use of #pragma and other Q-o-I tool chain behavior.

Explicit per-section and per-symbol data should be generated only when it cannot be implied by this inheritance. Being explicit is more verbose, and the explicit options are intended to capture exceptions.

5.3.6.1   Syntactic structure

The overall syntactic structure of an attributes section is:

<format-version>
[ <section-length> "vendor-name"
      [ <file-tag> <size> <attribute>*
      | <section-tag> <size> <section-number>* 0 <attribute>*
      | <symbol-tag> <size> <symbol-number>* 0 <attribute>*
      ]+
]*

Format-version describes the format of the following data. It is a single byte (not ULEB128). This is version 'A' (0x41). This field exists to permit future incompatible changes in format.

Section-length is a 4-byte unsigned integer in the byte order of the ELF file. It contains the length of the vendor-specific data, including the length field itself, the vendor name string and its terminating NUL byte, and the following attribute data. That is, it is the offset from the start of this vendor subsection to the start of the next vendor subsection.

Vendor-name is a NUL-terminated byte string in the style of a C string. Vendor names begining “Anon” or “anon” are reserved to unregistered private use.

Note

In general, a .ARM.attributes section in a relocatable file will contain a vendor subsection from the "aeabi" pseudo vendor and, optionally, one from the generating tool chain (e.g. "Arm", "gnu", "WRS", etc) as listed in Registered Vendor Names.

It is required that:

  • Attributes that record facts about the compatibility of this relocatable file with other relocatable files are recorded in the public "aeabi" subsection.
  • Attributes meaningful only to the producer are recorded in the private vendor subsection. These must not affect compatibility between relocatable files unless that is recorded in the "aeabi" subsection using generic compatibility tags.
  • Generic compatibility tags must record a "safe" approximation. A tool chain may record more precise information that only that tool chain comprehends.

Note

The intent is that a "foreign" tool chain should not mistakenly link incompatible binary files. The consequence is that a foreign tool chain might sometimes refuse to link files that could be safely linked, because their incompatibility has been crudely approximated.

There are no constraints on the order or number of vendor subsections. A consumer can collect the public ("aeabi") attributes in a single pass over the section, then all of its private data in a second pass.

A vendor-attributes subsection may contain any number of sub-subsections. Each records attributes relating to:

  • The whole relocatable file. These sub-subsections contain just a list of attributes.
  • A set of sections within the relocatable file. These sub-subsections contain a list of section numbers followed by a list of attributes.
  • A set of (defined) symbols in the relocatable file. These sub-subsections contain a list of symbol numbers followed by a list of attributes.

A sub-subsection starts with a tag that identifies the type of the sub-subsection (file, section, or symbol), followed by a 4-byte unsigned integer size in the byte-order of the ELF file. The size is the total size of the sub-subsection including the tag, the size itself, and the sub-subsection content.

Both section indexes and defined symbol indexes are non-zero, so a NUL byte ends a string and a list of indexes.

There are no constraints on the order or number of sub-subsections in a vendor subsection. A consumer that needs the data in inheritance order can obtain the file attributes, the section-related attributes, and the symbol-related attributes, by making three passes over the subsection.

A public attribute is encoded as a tag (non zero, ULEB128-encoded followed by a value. A public value is either an enumeration constant (ULEB128-encoded) or a NUL-terminated string.

Some examples of tags and their argument sorts include:

Tag_CPU_raw_name <string>  -- 0x04, "ML692000"
Tag_CPU_name     <string>  -- 0x05, "Arm946E-S"
Tag_PCS_R9_use   <uleb128> -- 0x0E, 0x01 (R9 used as SB)
Tag_PCS_config   <uleb128> -- 0x0D, 0x03 (Linux DSO [/fpic] configuration)
5.3.6.2   Top level structure tags

The following tags are defined globally:

Tag_File, (=1), uleb128:byte-size
Tag_Section, (=2), uleb128:byte-size
Tag_Symbol, (=3), uleb128:byte-size

There are no processor-specific extensions to the string table.

There are no processor-specific symbol types or symbol bindings. All processor-specific values are reserved to future revisions of this specification.

There are two forms of weak symbol:

  • A weak reference — This is denoted by st_shndx=SHN_UNDEF, ELF32_ST_BIND()=STB_WEAK.
  • A weak definition — This is denoted by st_shndx!=SHN_UNDEF, ELF32_ST_BIND()=STB_WEAK.
5.5.1.1   Weak References

Libraries are not searched to resolve weak references. It is not an error for a weak reference to remain unsatisfied.

During linking, the value of an undefined weak reference is:

  • Zero if the relocation type is absolute
  • The address of the place if the relocation type is pc-relative
  • The nominal base address if the relocation type is base-relative.

See Relocation for further details.

5.5.1.2   Weak Definitions

A weak definition does not change the rules by which object files are selected from libraries. However, if a link set contains both a weak definition and a non-weak definition, the non-weak definition will always be used.

All code symbols exported from an object file (symbols with binding STB_GLOBAL) shall have type STT_FUNC.

All extern data objects shall have type STT_OBJECT. No STB_GLOBAL data symbol shall have type STT_FUNC.

The type of an undefined symbol shall be STT_NOTYPE or the type of its expected definition.

The type of any other symbol defined in an executable section can be STT_NOTYPE. The linker is only required to provide interworking support for symbols of type STT_FUNC (interworking for untyped symbols must be encoded directly in the object file).

In addition to the normal rules for symbol values the following rules shall also apply to symbols of type STT_FUNC:

  • If the symbol addresses an Arm instruction, its value is the address of the instruction (in a relocatable object, the offset of the instruction from the start of the section containing it).
  • If the symbol addresses a Thumb instruction, its value is the address of the instruction with bit zero set (in a relocatable object, the section offset with bit zero set).
  • For the purposes of relocation the value used shall be the address of the instruction (st_value & ~1).

Note

This allows a linker to distinguish Arm and Thumb code symbols without having to refer to the map. An Arm symbol will always have an even value, while a Thumb symbol will always have an odd value. However, a linker should strip the discriminating bit from the value before using it for relocation.

A symbol that names a C or assembly language entity should have the name of that entity. For example, a C function called calculate generates a symbol called calculate (not _calculate).

Symbol names are case sensitive and are matched exactly by linkers.

Any symbol with binding STB_LOCAL may be removed from an object and replaced with an offset from another symbol in the same section under the following conditions:

  • The original symbol and replacement symbol are not of type STT_FUNC, or both symbols are of type STT_FUNC and describe code of the same execution type (either both Arm or both Thumb).
  • All relocations referring to the symbol can accommodate the adjustment in the addend field (it is permitted to convert a REL type relocation to a RELA type relocation).
  • The symbol is not described by the debug information.
  • The symbol is not a mapping symbol.
  • The resulting object, or image, is not required to preserve accurate symbol information to permit decompilation or other post-linking optimization techniques.
  • If the symbol labels an object in a section with the SHF_MERGE flag set, the relocation using symbol may be changed to use the section symbol only if the initial addend of the relocation is zero.

No tool is required to perform the above transformations; an object consumer must be prepared to do this itself if it might find the additional symbols confusing.

Note

Multiple conventions exist for the names of compiler temporary symbols (for example, ARMCC uses Lxxx.yyy, while GNU uses .Lxxx).

5.5.4.1   Reserved symbol names

The following symbols are reserved to this and future revisions of this specification:

  • Local symbols (STB_LOCAL) beginning with $
  • Global symbols (STB_GLOBAL, STB_WEAK) beginning with __aeabi_ (double _ at start).
  • Global symbols (STB_GLOBAL, STB_WEAK) ending with any of $$base, $$length or $$limit
  • Symbols matching the pattern ${Ven|other}${AA|AT|TA|TT}${I|L|S}[$PI]$$symbol
  • Local symbols (STB_LOCAL) beginning with Lib$Request$$ or BuildAttributes$$
  • Symbols beginning with $Sub$$ or $Super$$

Note that global symbols beginning with __vendor_ (double _ at start), where vendor is listed in Registered Vendor Names, Registered Vendor Names, are reserved to the named vendor for the purpose of providing vendor-specific tool-chain support functions.

Conventions for reserved symbols for which support is not required by this ABI are described in Appendix: Conventions for symbols containing $.

A section of an ELF file can contain a mixture of Arm code, Thumb code and data.

There are inline transitions between code and data at literal pool boundaries. There can also be inline transitions between Arm code and Thumb code, for example in Arm-Thumb inter-working veneers.

Linkers, and potentially other tools, need to map images correctly (for example, to support byte swapping to produce a BE-8 image from a BE-32 object file). To support this, a number of symbols, termed mapping symbols appear in the symbol table to denote the start of a sequence of bytes of the appropriate type. All mapping symbols have type STT_NOTYPE and binding STB_LOCAL. The st_size field is unused and must be zero.

The mapping symbols are defined in the following table. It is an error for a relocation to reference a mapping symbol. Two forms of mapping symbol are supported:

  • a short form, that uses a dollar character and a single letter denoting the class. This form can be used when an object producer creates mapping symbols automatically, and minimizes symbol table space * a longer form, where the short form is extended with a period and then any sequence of characters that are legal for a symbol. This form can be used when assembler files have to be annotated manually and the assembler does not support multiple definitions of symbols.
Mapping symbols
Name Meaning
$a
$a.<any...>
Start of a sequence of Arm instructions
$d
$d.<any...>
Start of a sequence of data items (for example, a literal pool)
$t
$t.<any...>
Start of a sequence of Thumb instructions
5.5.5.1   Section-relative mapping symbols

Mapping symbols defined in a section define a sequence of half-open address intervals that cover the address range of the section. Each interval starts at the address defined by the mapping symbol, and continues up to, but not including, the address defined by the next (in address order) mapping symbol or the end of the section. A section must have a mapping symbol defined at the beginning of the section; however, if the section contains only data then the mapping symbol may be omitted.

5.5.5.2   Absolute mapping symbols

Mapping symbols are no-longer required for the absolute section. The equivalent information is now conveyed by the type of the absolute symbol.

Relocation information is used by linkers in order to bind symbols and addresses that could not be determined when the initial object was generated. In these descriptions, references in the style LDR(1) refer to the Armv5 Architecture Reference Manual [Armv5 ARM] while those in the style LDR(immediate, Thumb) give the corresponding reference to the Arm Architecture Reference Manual Arm v7-A and Arm v7-R edition [ARMARM].

The relocation codes for Arm are divided into four categories:

  • Mandatory relocations that must be supported by all static linkers
  • Platform-specific relocations that are required for specific virtual platforms
  • Private relocations that are guaranteed never to be allocated in future revisions of this specification, but which must never be used in portable object files.
  • Unallocated relocations that are reserved for use in future revisions of this specification.
5.6.1.1   Addends and PC-bias compensation

A binary file may use REL or RELA relocations or a mixture of the two (but multiple relocations for the same address must use only one type). If the relocation is pc-relative then compensation for the PC bias (the PC value is 8 bytes ahead of the executing instruction in Arm state and 4 bytes in Thumb state) must be encoded in the relocation by the object producer.

Unless specified otherwise, the initial addend for REL type relocations is formed according to the following rules.

  • If the place is subject to a data-type relocation, the initial value in the place is sign-extended to 32 bits.
  • If the place contains an instruction, the immediate field for the instruction is extracted from it and used as the initial addend. If the instruction is a SUB, or an LDR/STR type instruction with the ‘up’ bit clear, then the initial addend is formed by negating the unsigned immediate value encoded in the instruction.

Some examples are shown in the following table.

Examples of REL format initial addends
Instruction Relocation Encoding Initial Addend
SUB R0, R1, #1020 R_ARM_ALU_PC_G0 0xe2410fff -1020
LDR R0, [R2, #16] R_ARM_LDR_PC_G2 0xe59f0010 16
BL . R_ARM_THM_CALL 0xf7ff, 0xfffe -4
DCB 0xf0 R_ARM_ABS8 0xf0 -16

If the initial addend cannot be encoded in the space available then a RELA format relocation must be used.

There are three special cases for forming the initial addend of REL-type relocations where the immediate field cannot normally hold small signed integers:

  • For relocations processing MOVW and MOVT instructions (in both Arm and Thumb state), the initial addend is formed by interpreting the 16-bit literal field of the instruction as a 16-bit signed value in the range -32768 <= A < 32768. The interpretation is the same whether the relocated place contains a MOVW instruction or a MOVT instruction.
  • For R_ARM_THM_JUMP6 the initial addend is formed by the formula (((imm + 4) & 0x7f) – 4), where imm is the concatenation of bit[9]:bit[7:3]:’0’ from the Thumb CBZ or CBNZ instruction being relocated.
  • For R_ARM_THM_PC8 the initial addend is formed by the formula (((imm + 4) & 0x3ff) – 4), where imm is the 32-bit value encoded in the 8-bit place, as defined in the LDR(3)/LDR(literal) Thumb instructions section of the [ARMARM].
5.6.1.2   Relocation types

Relocation codes table, below, lists the relocation codes for Arm. The table shows:

  • The code which is stored in the ELF32_R_TYPE component of the r_info field.
  • The mnemonic name for the relocation.
  • The type of the relocation. This field substantially divides the relocations into Static and Dynamic relocations. Static relocations are processed by a static linker; they are normally either fully resolved or used to produce dynamic relocations for processing by a post-linking step or a dynamic loader. A well formed image will have no static relocations after static linking is complete, so a post-linker or dynamic loader will normally only have to deal with dynamic relocations. This field is also used to describe deprecated, obsolete, private and unallocated relocation codes. Deprecated codes should not be generated by fully conforming toolchains; however it is recognized that there may be substantial existing code that makes use of these forms, so it is expected that a linker may well be required to handle them at this time. Obsolete codes should not be used, and it is believed that there is little or no common use of these values. All unallocated codes are reserved for future allocation.
  • The class of the relocation describes the type of place being relocated: these are Data, Arm, Thumb16 and Thumb32 (32-bit long-format instructions). A special class of Miscellaneous is used when the operation is not a simple mathematical expression.
  • The operation field describes how the symbol and addend are processed by the relocation code. It does not describe how the addend is formed (for a REL type relocation), what overflow checking is done, or how the value is written back into the place: this information is given in subsequent sections. In all cases, relocation expression values are computed mod 232.

The following nomenclature is used for the operation:

  • S (when used on its own) is the address of the symbol.
  • A is the addend for the relocation.
  • P is the address of the place being relocated (derived from r_offset).
  • Pa is the adjusted address of the place being relocated, defined as (P & 0xFFFFFFFC).
  • T is 1 if the target symbol S has type STT_FUNC and the symbol addresses a Thumb instruction; it is 0 otherwise.
  • B(S) is the addressing origin of the output segment defining the symbol S. The origin is not required to be the base address of the segment. This value must always be word-aligned.
  • GOT_ORG is the addressing origin of the Global Offset Table (the indirection table for imported data addresses). This value must always be word-aligned. See Proxy generating relocations.
  • GOT(S) is the address of the GOT entry for the symbol S.
Relocation codes
Code Name Type Class Operation
0 R_ARM_NONE Static Miscellaneous  
1 R_ARM_PC24 Deprecated Arm ((S + A) | T) – P
2 R_ARM_ABS32 Static Data (S + A) | T
3 R_ARM_REL32 Static Data ((S + A) | T) | – P
4 R_ARM_LDR_PC_G0 Static Arm S + A – P
5 R_ARM_ABS16 Static Data S + A
6 R_ARM_ABS12 Static Arm S + A
7 R_ARM_THM_ABS5 Static Thumb16 S + A
8 R_ARM_ABS8 Static Data S + A
9 R_ARM_SBREL32 Static Data ((S + A) | T) – B(S)
10 R_ARM_THM_CALL Static Thumb32 ((S + A) | T) – P
11 R_ARM_THM_PC8 Static Thumb16 S + A – Pa
12 R_ARM_BREL_ADJ Dynamic Data ChangeIn[B(S)] + A
13 R_ARM_TLS_DESC Dynamic Data  
14 R_ARM_THM_SWI8 Obsolete Encodings reserved for future Dynamic relocations
15 R_ARM_XPC25 Obsolete
16 R_ARM_THM_XPC22 Obsolete
17 R_ARM_TLS_DTPMOD32 Dynamic Data Module[S]
18 R_ARM_TLS_DTPOFF32 Dynamic Data S + A – TLS
19 R_ARM_TLS_TPOFF32 Dynamic Data S + A – tp
20 R_ARM_COPY Dynamic Miscellaneous  
21 R_ARM_GLOB_DAT Dynamic Data (S + A) | T
22 R_ARM_JUMP_SLOT Dynamic Data (S + A) | T
23 R_ARM_RELATIVE Dynamic Data B(S) + A [Note: see Dynamic relocations]
24 R_ARM_GOTOFF32 Static Data ((S + A) | T) – GOT_ORG
25 R_ARM_BASE_PREL Static Data B(S) + A – P
26 R_ARM_GOT_BREL Static Data GOT(S) + A – GOT_ORG
27 R_ARM_PLT32 Deprecated Arm ((S + A) | T) – P
28 R_ARM_CALL Static Arm ((S + A) | T) – P
29 R_ARM_JUMP24 Static Arm ((S + A) | T) – P
30 R_ARM_THM_JUMP24 Static Thumb32 ((S + A) | T) – P
31 R_ARM_BASE_ABS Static Data B(S) + A
32 R_ARM_ALU_PCREL_7_0 Obsolete Note - Legacy (Arm ELF B02) names have been retained for these obsolete relocations.
33 R_ARM_ALU_PCREL_15_8 Obsolete
34 R_ARM_ALU_PCREL_23_15 Obsolete
35 R_ARM_LDR_SBREL_11_0_NC Deprecated Arm S + A – B(S)
36 R_ARM_ALU_SBREL_19_12_NC Deprecated Arm S + A – B(S)
37 R_ARM_ALU_SBREL_27_20_CK Deprecated Arm S + A – B(S)
38 R_ARM_TARGET1 Static Miscellaneous (S + A) | T or ((S + | A) | T) – P
39 R_ARM_SBREL31 Deprecated Data ((S + A) | T) – B(S)
40 R_ARM_V4BX Static Miscellaneous  
41 R_ARM_TARGET2 Static Miscellaneous  
42 R_ARM_PREL31 Static Data ((S + A) | T) – P
43 R_ARM_MOVW_ABS_NC Static Arm (S + A) | T
44 R_ARM_MOVT_ABS Static Arm S + A
45 R_ARM_MOVW_PREL_NC Static Arm ((S + A) | T) – P
46 R_ARM_MOVT_PREL Static Arm S + A – P
47 R_ARM_THM_MOVW_ABS_NC Static Thumb32 (S + A) | T
48 R_ARM_THM_MOVT_ABS Static Thumb32 S + A
49 R_ARM_THM_MOVW_PREL_NC Static Thumb32 ((S + A) | T) – P
50 R_ARM_THM_MOVT_PREL Static Thumb32 S + A – P
51 R_ARM_THM_JUMP19 Static Thumb32 ((S + A) | T) – P
52 R_ARM_THM_JUMP6 Static Thumb16 S + A – P
53 R_ARM_THM_ALU_PREL_11_0 Static Thumb32 ((S + A) | T) – Pa
54 R_ARM_THM_PC12 Static Thumb32 S + A – Pa
55 R_ARM_ABS32_NOI Static Data S + A
56 R_ARM_REL32_NOI Static Data S + A – P
57 R_ARM_ALU_PC_G0_NC Static Arm ((S + A) | T) – P
58 R_ARM_ALU_PC_G0 Static Arm ((S + A) | T) – P
59 R_ARM_ALU_PC_G1_NC Static Arm ((S + A) | T) – P
60 R_ARM_ALU_PC_G1 Static Arm ((S + A) | T) – P
61 R_ARM_ALU_PC_G2 Static Arm ((S + A) | T) – P
62 R_ARM_LDR_PC_G1 Static Arm S + A – P
63 R_ARM_LDR_PC_G2 Static Arm S + A – P
64 R_ARM_LDRS_PC_G0 Static Arm S + A – P
65 R_ARM_LDRS_PC_G1 Static Arm S + A – P
66 R_ARM_LDRS_PC_G2 Static Arm S + A – P
67 R_ARM_LDC_PC_G0 Static Arm S + A – P
68 R_ARM_LDC_PC_G1 Static Arm S + A – P
69 R_ARM_LDC_PC_G2 Static Arm S + A – P
70 R_ARM_ALU_SB_G0_NC Static Arm ((S + A) | T) – B(S)
71 R_ARM_ALU_SB_G0 Static Arm ((S + A) | T) – B(S)
72 R_ARM_ALU_SB_G1_NC Static Arm ((S + A) | T) – B(S)
73 R_ARM_ALU_SB_G1 Static Arm ((S + A) | T) – B(S)
74 R_ARM_ALU_SB_G2 Static Arm ((S + A) | T) – B(S)
75 R_ARM_LDR_SB_G0 Static Arm S + A – B(S)
76 R_ARM_LDR_SB_G1 Static Arm S + A – B(S)
77 R_ARM_LDR_SB_G2 Static Arm S + A – B(S)
78 R_ARM_LDRS_SB_G0 Static Arm S + A – B(S)
79 R_ARM_LDRS_SB_G1 Static Arm S + A – B(S)
80 R_ARM_LDRS_SB_G2 Static Arm S + A – B(S)
81 R_ARM_LDC_SB_G0 Static Arm S + A – B(S)
82 R_ARM_LDC_SB_G1 Static Arm S + A – B(S)
83 R_ARM_LDC_SB_G2 Static Arm S + A – B(S)
84 R_ARM_MOVW_BREL_NC Static Arm ((S + A) | T) – B(S)
85 R_ARM_MOVT_BREL Static Arm S + A – B(S)
86 R_ARM_MOVW_BREL Static Arm ((S + A) | T) – B(S)
87 R_ARM_THM_MOVW_BREL_NC Static Thumb32 ((S + A) | T) – B(S)
88 R_ARM_THM_MOVT_BREL Static Thumb32 S + A – B(S)
89 R_ARM_THM_MOVW_BREL Static Thumb32 ((S + A) | T) – B(S)
90 R_ARM_TLS_GOTDESC Static Data  
91 R_ARM_TLS_CALL Static Arm  
92 R_ARM_TLS_DESCSEQ Static Arm TLS relaxation
93 R_ARM_THM_TLS_CALL Static Thumb32  
94 R_ARM_PLT32_ABS Static Data PLT(S) + A
95 R_ARM_GOT_ABS Static Data GOT(S) + A
96 R_ARM_GOT_PREL Static Data GOT(S) + A – P
97 R_ARM_GOT_BREL12 Static Arm GOT(S) + A – GOT_ORG
98 R_ARM_GOTOFF12 Static Arm S + A – GOT_ORG
99 R_ARM_GOTRELAX Static Miscellaneous  
100 R_ARM_GNU_VTENTRY Deprecated Data ???
101 R_ARM_GNU_VTINHERIT Deprecated Data ???
102 R_ARM_THM_JUMP11 Static Thumb16 S + A – P
103 R_ARM_THM_JUMP8 Static Thumb16 S + A – P
104 R_ARM_TLS_GD32 Static Data GOT(S) + A – P
105 R_ARM_TLS_LDM32 Static Data GOT(S) + A – P
106 R_ARM_TLS_LDO32 Static Data S + A – TLS
107 R_ARM_TLS_IE32 Static Data GOT(S) + A – P
108 R_ARM_TLS_LE32 Static Data S + A – tp
109 R_ARM_TLS_LDO12 Static Arm S + A – TLS
110 R_ARM_TLS_LE12 Static Arm S + A – tp
111 R_ARM_TLS_IE12GP Static Arm GOT(S) + A – GOT_ORG
112-127 R_ARM_PRIVATE_<n> Private (n = 0, 1, ... 15)
128 R_ARM_ME_TOO Obsolete
129 R_ARM_THM_TLS_DESCSEQ16 Static Thumb16  
130 R_ARM_THM_TLS_DESCSEQ32 Static Thumb32  
131 R_ARM_THM_GOT_BREL12 Static Thumb32 GOT(S) + A – GOT_ORG
132 R_ARM_THM_ALU_ABS_G0_NC Static Thumb16 (S + A) | T
133 R_ARM_THM_ALU_ABS_G1_NC Static Thumb16 S + A
134 R_ARM_THM_ALU_ABS_G2_NC Static Thumb16 S + A
135 R_ARM_THM_ALU_ABS_G3 Static Thumb16 S + A
136 R_ARM_THM_BF16 Static Arm ((S + A) | T) – P
137 R_ARM_THM_BF12 Static Arm ((S + A) | T) – P
138 R_ARM_THM_BF18 Static Arm ((S + A) | T) – P
139-159   Static Reserved for future allocation
160 R_ARM_IRELATIVE Dynamic Reserved for future functionality
161-176 R_ARM_PRIVATE_<n> Private (n = 16, 17, ... 31)
177-255   Dynamic Reserved for future allocation
5.6.1.3   Static Data relocations

Except as indicated in the following table, all static data relocations have size 4, alignment 1 and write the full 32-bit result to the place; there is thus no need for overflow checking.

The overflow ranges for R_ARM_ABS16 and R_ARM_ABS8 permit either signed or unsigned results. It is therefore not possible to detect an unsigned value that has underflowed by a small amount, or a signed value that has overflowed by a small amount.

Static Data relocations with non-standard size or processing
Code Name Size REL Addend Overflow
5 R_ARM_ABS16 2 sign_extend(P[16:0]) -32768 ≤ X ≤ 65535
8 R_ARM_ABS8 1 sign_extend(P[8:0]) -128 ≤ X ≤ 255
42 R_ARM_PREL31 4 sign_extend(P[30:0]) 31-bit 2’s complement
5.6.1.4   Static Arm relocations

The relocations that can modify fields of an Arm instruction are listed in the following table. All relocations in this class relocate a 32-bit aligned Arm instruction by modifying part of the instruction. In most cases the modification is to change the offset, but in some cases the opcode itself may be changed (for example, an ADD may be converted to a SUB and vice-versa). In the table:

  • X is the 32-bit result of normal relocation processing
  • Gn is a mask operation that is instruction dependent. See Group Relocations below for rules on how the mask is formed for each case.
Static Arm instruction relocations
Code Name Overflow Instruction Result Mask
4 R_ARM_LDR_PC_G0 Yes LDR, STR, LDRB, STRB ABS(X) & G0 (LDR)
6 R_ARM_ABS12 Yes LDR, STR ABS(X) & 0xFFF
28 R_ARM_CALL Yes BL/BLX X & 0x03FFFFFE
29 R_ARM_JUMP24 Yes B/BL<cond> X & 0x03FFFFFE
43 R_ARM_MOVW_ABS_NC No MOVW X & 0xFFFF
44 R_ARM_MOVT_ABS No MOVT X & 0xFFFF0000
45 R_ARM_MOVW_PREL_NC No MOVW X & 0xFFFF
46 R_ARM_MOVT_PREL No MOVT X & 0xFFFF0000
57 R_ARM_ALU_PC_G0_NC No ADD, SUB ABS(X) & G0
58 R_ARM_ALU_PC_G0 Yes ADD, SUB ABS(X) & G0
59 R_ARM_ALU_PC_G1_NC No ADD, SUB ABS(X) & G1
60 R_ARM_ALU_PC_G1 Yes ADD, SUB ABS(X) & G1
61 R_ARM_ALU_PC_G2 Yes ADD, SUB ABS(X) & G2
62 R_ARM_LDR_PC_G1 Yes LDR, STR, LDRB, STRB ABS(X) & G1 (LDR)
63 R_ARM_LDR_PC_G2 Yes LDR, STR, LDRB, STRB ABS(X) & G2 (LDR)
64 R_ARM_LDRS_PC_G0 Yes LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ABS(X) & G0 (LDRS)
65 R_ARM_LDRS_PC_G1 Yes LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ABS(X) & G1 (LDRS)
66 R_ARM_LDRS_PC_G2 Yes LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ABS(X) & G2 (LDRS)
67 R_ARM_LDC_PC_G0 Yes LDC, STC ABS(X) & G0 (LDC)
68 R_ARM_LDC_PC_G1 Yes LDC, STC ABS(X) & G1 (LDC)
69 R_ARM_LDC_PC_G2 Yes LDC, STC ABS(X) & G2 (LDC)
70 R_ARM_ALU_SB_G0_NC No ADD, SUB ABS(X) & G0
71 R_ARM_ALU_SB_G0 Yes ADD, SUB ABS(X) & G0
72 R_ARM_ALU_SB_G1_NC No ADD, SUB ABS(X) & G1
73 R_ARM_ALU_SB_G1 Yes ADD, SUB ABS(X) & G1
74 R_ARM_ALU_SB_G2 Yes ADD, SUB ABS(X) & G2
75 R_ARM_LDR_SB_G0 Yes LDR, STR, LDRB, STRB ABS(X) & G0 (LDR)
76 R_ARM_LDR_SB_G1 Yes LDR, STR, LDRB, STRB ABS(X) & G1 (LDR)
77 R_ARM_LDR_SB_G2 Yes LDR, STR, LDRB, STRB ABS(X) & G2 (LDR)
78 R_ARM_LDRS_SB_G0 Yes LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ABS(X) & G0 (LDRS)
79 R_ARM_LDRS_SB_G1 Yes LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ABS(X) & G1 (LDRS)
80 R_ARM_LDRS_SB_G2 Yes LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ABS(X) & G2 (LDRS)
81 R_ARM_LDC_SB_G0 Yes LDC, STC ABS(X) & G0 (LDC)
82 R_ARM_LDC_SB_G1 Yes LDC, STC ABS(X) & G1 (LDC)
83 R_ARM_LDC_SB_G2 Yes LDC, STC ABS(X) & G2 (LDC)
84 R_ARM_MOVW_BREL_NC No MOVW X & 0xFFFF
85 R_ARM_MOVT_BREL No MOVT X & 0xFFFF0000
86 R_ARM_MOVW_BREL Yes MOVW X & 0xFFFF
97 R_ARM_GOT_BREL12 Yes LDR ABS(X) & 0xFFF
98 R_ARM_GOTOFF12 Yes LDR, STR ABS(X) & 0xFFF
109 R_ARM_TLS_LDO12 Yes LDR, STR ABS(X) & 0xFFF
110 R_ARM_TLS_LE12 Yes LDR, STR ABS(X) & 0xFFF
111 R_ARM_TLS_IE12GP Yes LDR ABS(X) & 0xFFF

The formation of the initial addend in a REL type relocation for the various instruction classes is described in the following table. Insn modification describes how the 32-bit result X is written back to the instruction; Result_Mask is the value of X after the masking operation described in Static Arm instruction relocations has been applied.

Arm relocation actions by instruction type
Instruction REL Addend Insn modification
BL, BLX sign_extend (insn[23:0] << 2) See Call and jump relocations
B, BL<cond> sign_extend (insn[23:0] << 2) See Call and jump relocations
LDR, STR, LDRB, STRB insn[11:0] * -1(insn[23] == 0)
insn[23] = (X >= 0)
insn[11:0] = Result_Mask(X)
LDRD, STRD, LDRH, STRH, LDRSH, LDRSB ((insn[11:8] << 4) | insn[3:0]) * -1(insn[23] == 0)
insn[23] = (X >= 0)
insn[11:0] = Result_Mask(X)
LDC, STC (insn[7:0] << 2) * -1(insn[23] == 0)
insn[23] = (X >= 0)
insn[7:0] = Result_Mask(X) >> 2
ADD, SUB Imm(insn) * -1(opcode(insn) == SUB)
opcode(insn) = X >= 0 ? ADD : SUB
Imm(insn) = Result_Mask(X)
MOVW See Addends and PC-bias compensation
insn[19:16] = Result_Mask(X) >> 12
insn[11:0] = Result_Mask(X) & 0xFFF
MOVT See Addends and PC-bias compensation. The effect permits executing MOVW and later MOVT to create a 32-bit link-time constant in a register.
insn[19:16] = (Result_Mask(X) >> 16) >> 12
insn[11:0] = (Result_Mask(X) >> 16) & 0xFFF

Call and Jump relocations

There is one relocation (R_ARM_CALL) for unconditional function call instructions (BLX and BL with the condition field set to 0xe), and one for jump instructions (R_ARM_JUMP24). The principal difference between the two relocation values is the handling of Arm/Thumb inter-working: on Arm architecture 5 and above, an instruction relocated by R_ARM_CALL that calls a function that is entered in Thumb state may be relocated by changing the instruction to BLX; an instruction relocated by R_ARM_JUMP24 must use a veneer to effect the transition to Thumb state. Conditional function call instructions (BL<cond>) must be relocated using R_ARM_JUMP24.

A linker may use a veneer (a sequence of instructions) to implement the relocated branch if the relocation is one of R_ARM_PC24, R_ARM_CALL, R_ARM_JUMP24, (or, in Thumb state, R_ARM_THM_CALL, R_ARM_THM_JUMP24, or R_ARM_THM_JUMP19) and:

  • The target symbol has type STT_FUNC
  • Or, the target symbol and relocated place are in separate sections input to the linker

In all other cases a linker shall diagnose an error if relocation cannot be effected without a veneer. A linker generated veneer may corrupt register r12 (IP) and the condition flags, but must preserve all other registers. On M-profile processors a veneer may also assume the presence of a stack with at least 8 bytes (2 words) of memory. Linker veneers may be needed for a number of reasons, including, but not limited to:

  • Target is outside the addressable span of the branch instruction (± 32Mb)
  • Target address and execution state will not be known until run time, or the address might be pre-empted

In some systems indirect calls may also use veneers in order to support dynamic linkage while preserving pointer equivalence. On platforms that do not support dynamic pre-emption of symbols an unresolved weak reference to a symbol relocated by R_ARM_CALL (or, in Thumb state, R_ARM_THM_CALL) shall be treated as a jump to the next instruction (the call becomes a no-op). The behaviour of R_ARM_JUMP24 and static Thumb jump relocations in these conditions is implementation-defined.

Group relocations

Relocation codes 4 and 57-83 are intended to relocate sequences of instructions that generate a single address. They are encoded to extract the maximum flexibility from the Arm ADD- and SUB-immediate instructions without need to determine during linking the full sequence being used. The relocations operate by performing the basic relocation calculation and then partitioning the result into a set of groups of bits that can be statically determined. All processing for the formation of the groups is done on the absolute value of X; the sign of X is used to determine whether ADD or SUB instructions are used, or, if the sequence concludes with a load/store operation, the setting of the U bit (bit 23) in the instruction.

A group, Gn, is formed by examining the residual value, Yn, after the bits for group Gn–1 have been masked off. Processing for group G0 starts with the absolute value of X. For ALU-type relocations a group is formed by determining the most significant bit (MSB) in the residual and selecting the smallest constant Kn such that

MSB(Yn) & (255 << 2Kn) != 0,

except that if Yn is 0, then Kn is 0. The value Gn is then

Yn & (255 << 2Kn),

and the residual, Yn+1, for the next group is

Yn & ~Gn.

Note that if Yn is 0, then Gn will also be 0.

For group relocations that access memory the residual value is examined in its entirety (i.e. after the appropriate sequence of ALU groups have been removed): if the relocation has not overflowed, then the residual for such an instruction will always be a valid offset for the indicated type of memory access.

Overflow checking is always performed on the highest-numbered group in a sequence. For ALU-type relocations the result has overflowed if Yn+1 is not zero. For memory access relocations the result has overflowed if the residual is not a valid offset for the type of memory access.

Note

The unchecked (_NC) group relocations all include processing of the Thumb bit of a symbol. However, the memory forms of group relocations (eg R_ARM_LDR_G0) ignore this bit. Therefore the use of the memory forms with symbols of type STT_FUNC is unpredictable.

5.6.1.5   Static Thumb16 relocations

Relocations for 16-bit thumb instructions are shown in the following table. In general the addressing range of these relocations is too small for them to reference external symbols and they are documented here for completeness. A linker is not required to generate trampoline sequences (or veneers) to extend the branching range of the jump relocations.

Relocation R_ARM_THM_JUMP6 is only applicable to the Thumb-2 instruction set.

Static Thumb-16 relocations
Code Name Overflow Instruction Result Mask
7 R_ARM_THM_ABS5 Yes
LDR(1)/LDR(immediate, Thumb), STR(1)/STR(immediate, Thumb)
X & 0x7C
11 R_ARM_THM_PC8 Yes
LDR(3)/LDR(literal), ADD(5)/ADR
X & 0x3FC
52 R_ARM_THM_JUMP6 Yes CBZ, CBNZ X & 0x7E
102 R_ARM_THM_JUMP11 Yes B(2)/B X & 0xFFE
103 R_ARM_THM_JUMP8 Yes B(1)/B<cond> X & 0x1FE
132 R_ARM_THM_ALU_ABS_G0_NC No
ADD(2)/ADD (immediate, Thumb, 8-bit immediate), MOV(1)/MOV (immediate)
X & 0x000000FF
133 R_ARM_THM_ALU_ABS_G1_NC No
ADD(2)/ADD (immediate, Thumb, 8-bit immediate), MOV(1)/MOV (immediate)
X & 0x0000FF00
134 R_ARM_THM_ALU_ABS_G2_NC No
ADD(2)/ADD (immediate, Thumb, 8-bit immediate), MOV(1)/MOV (immediate)
X & 0x00FF0000
135 R_ARM_THM_ALU_ABS_G3 No
ADD(2)/ADD (immediate, Thumb, 8-bit immediate), MOV(1)/MOV (immediate)
X & 0xFF000000
5.6.1.6   Static Thumb32 relocations

Relocations for 32-bit Thumb instructions are shown in the following table. With the exception of R_ARM_THM_CALL, these relocations are only applicable to 32-bit Thumb instructions.

Static Thumb-32 instruction relocations
Code Name Overflow Instruction Result Mask
10 R_ARM_THM_CALL Yes BL X & 0x01FFFFFE
30 R_ARM_THM_JUMP24 Yes B.W X & 0x01FFFFFE
47 R_ARM_THM_MOVW_ABS_NC No MOVW X & 0x0000FFFF
48 R_ARM_THM_MOVT_ABS No MOVT X & 0xFFFF0000
49 R_ARM_THM_MOVW_PREL_NC No MOVW X & 0x0000FFFF
50 R_ARM_THM_MOVT_PREL No MOVT X & 0xFFFF0000
51 R_ARM_THM_JUMP19 Yes B<cond>.W X & 0x001FFFFE
53 R_ARM_THM_ALU_PREL_11_0 Yes ADR.W X & 0x00000FFF
54 R_ARM_THM_PC12 Yes LDR<,B,SB,H,SH> (literal) ABS(X) & 0x00000FFF
87 R_ARM_THM_MOVW_BREL_NC No MOVW X & 0x0000FFFF
88 R_ARM_THM_MOVT_BREL No MOVT X & 0xFFFF0000
89 R_ARM_THM_MOVW_BREL Yes MOVW X & 0x0000FFFF
131 R_ARM_THM_GOT_BREL12 Yes
LDR(immediate, Thumb) 12-bit immediate
X & 0x00000FFF

The formation of the initial addend in a REL type relocation for the various instruction classes is described in the following table. Insn modification describes how the result X is written back to the instruction; Result_Mask is the value of X after the masking operation described in Static Thumb-16 relocations or Static Thumb-32 instruction relocations has been applied.

Thumb relocation actions by instruction type
Instruction REL Addend Insn modification
Thumb-16 instructions
LDR(1)/LDR(immediate, Thumb), STR(1)/STR(immediate, Thumb) insn[10:6] << 2 insn[10:6] = Result_Mask(X) >> 2
LDR(3)/LDR(literal), ADD(5)/ADR See Addends and PC-bias compensation insn[7:0] = Result_Mask(X) >> 2
CBZ, CBNZ See Addends and PC-bias compensation
insn [9] = Result_Mask(X) >> 6
insn[7:0] = (Result_Mask(X) >> 1) & 0x1F
B(2)/B sign_extend(insn[10:0] << 1) insn[10:0] = Result_Mask(X) >> 1
B(1)/B<cond> sign_extend(insn[7:0] << 1) insn[7:0] = Result_Mask(X) >> 1
ADD(2)/ADD(immediate, Thumb, 8-bit immediate), MOV(1)/MOV(immediate) insn[7:0]
insn[7:0] = Result_Mask(X) >> (8*n)
when relocated by R_ARM_THM_ALU_ABS_Gn[_NC]
Thumb-32 instructions
BL See Thumb call and jump relocations See Thumb call and jump relocations
B.W See Thumb call and jump relocations See Thumb call and jump relocations
B<cond>.W See Thumb call and jump relocations See Thumb call and jump relocations
MOVW See Addends and PC-bias compensation
insn[19:16] = Result_Mask(X) >> 12
insn[26] = (Result_Mask(X) >> 11) & 0x1
insn[14:12] = (Result_Mask(X) >> 8) & 0x7
insn[7:0] = Result_Mask(X) & 0xFF
(encodes the least significant 16 bits)
MOVT See Addends and PC-bias compensation. The effect permits executing MOVW and later MOVT to create a 32-bit link-time constant in a register.
insn[19:16] = Result_Mask(X) >> 28
insn[26] = (Result_Mask(X) >> 27) & 0x1
insn[14:12] = (Result_Mask(X) >> 24) & 0x7
insn[7:0] = (Result_Mask(X) >> 16) & 0xFF
(encodes the most significant 16 bits)
ADR.W (insn[26] << 11) | (insn[14:12] << 8) | insn[7:0]
insn[26] = Result_Mask(X) >> 11
insn[14:12] = (Result_Mask(X) >> 8) & 0x7
insn[7:0] = Result_Mask(X) & 0xFF
LDR<,B,SB,H,SH> (literal) insn[11:0] * -1(insn[23] ==0)
insn[23] = (X >= 0)
insn[11:0] = Result_Mask(X)
LDR (immediate, Thumb) 12-bit immediate insn[11:0] insn[11:0] = Result_Mask(X)

Thumb call and jump relocations

R_ARM_THM_CALL is used to relocate Thumb BL (and Armv5 Thumb BLX) instructions. It is the Thumb equivalent of R_ARM_CALL and the same rules on conversion apply. Bits 0-10 of the first half-word encode the most significant bits of the branch offset, bits 0-10 of the second half-word encode the least significant bits and the offset is in units of half-words. Thus 22 bits encode a branch offset of ± 222 bytes. When linking Armv6 (and later, see [ARM ARM]) Thumb code the range of the branch is increased by 2 bits, increasing the offset range to ± 224 bytes. The same relocation is used for both cases since a linker need only know that the code will run on a Thumb-2 (Armv6 and later) capable processor to exploit the additional range.

The addend for B.W and B<cond>.W is the signed immediate quantity encoded in the instruction, extracted in a similar way to BL; for details see [ARMARM].

The conditions under which call and jump relocations are permitted to generate an ip-corrupting intra-call veneer, and their behaviour in conjunction with unresolved weak references, are specified in Static Arm relocations under the heading Call and Jump relocations.

5.6.1.7   Static miscellaneous relocations

R_ARM_NONE records that the section containing the place to be relocated depends on the section defining the symbol mentioned in the relocation directive in a way otherwise invisible to the static linker. The effect is to prevent removal of sections that might otherwise appear to be unused.

R_ARM_V4BX records the location of an Armv4t BX instruction. This enables a static linker to generate Armv4 compatible images from Armv4t objects containing only Arm code by converting the instruction to MOV PC, r, where r is the register used in the BX instruction. See AAPCS32 for details. The symbol is unused and may even be the NULL symbol (index 0).

R_ARM_TARGET1 is processed in a platform-specific manner. It may only be used in sections with the types SHT_INIT_ARRAY, SHT_PREINIT_ARRAY, and SHT_FINI_ARRAY. The relocation must be processed either in the same way as R_ARM_REL32 or as R_ARM_ABS32: a virtual platform must specify which method is used. If the relocation is processed as R_ARM_REL32 then the section may be marked read-only and coalesced with other read-only data, otherwise it may only be marked read-only if it does not require dynamic linking.

R_ARM_TARGET2 is processed in a platform-specific manner. It is used to encode a data dependency that will only be dereferenced by code in the run-time support library.

5.6.1.8   Proxy generating relocations

A number of relocations generate proxy locations that are then subject to dynamic relocation. The proxies are normally gathered together in a single table, called the Global Offset Table or GOT. The following table lists the relocations that generate proxy entries.

Proxy generating relocations
Code Relocation Comment
26 R_ARM_GOT_BREL Offset of the GOT entry relative to the GOT origin
95 R_ARM_GOT_ABS Absolute address of the GOT entry
96 R_ARM_GOT_PREL Offset of the GOT entry from the place
97 R_ARM_GOT_BREL12 Offset of the GOT entry from the GOT origin. Stored in the offset field of an Arm LDR instruction
131 R_ARM_THM_GOT_BREL12 Offset of the GOT entry from the GOT origin. Stored in the offset field of a Thumb LDR instruction

All of the GOT entries generated by these relocations are subject to dynamic relocation by R_ARM_GLOB_DAT of the symbol indicated in the generating relocation. There is no provision for generating an addend for the dynamic entry. GOT entries must always be 32-bit aligned words. Multiple GOT-generating relocations referencing the same symbol may share a single entry in the GOT.

R_ARM_GOT_BREL, R_ARM_GOT_BREL12 and R_ARM_THM_GOT_BREL12 generate an offset from the addressing origin of the GOT. To calculate the absolute address of an entry it is necessary to add in the GOT’s addressing origin. How the origin is established depends on the execution environment and several relocations are provided in support of it.

  • R_ARM_BASE_PREL with the NULL symbol (symbol 0) will give the offset of the GOT origin from the address of the place.
  • R_ARM_BASE_ABS with the NULL symbol will give the absolute address of the GOT origin.
  • Other execution environments may require that the GOT origin be congruent with some other base. In these environments the appropriate means of establishing that base will apply.

In addition to the data generating relocations listed above the call and branch relocations (R_ARM_CALL, R_ARM_THM_CALL, R_ARM_JUMP24, R_ARM_THM_JUMP24, R_ARM_THM_JUMP19) may also require a proxy to be generated if the symbol will be defined in an external executable or may be pre-empted at execution time. The details of proxy sequences and locations are described in PLT Sequences and Usage Models.

R_ARM_GOTRELAX is reserved to permit future-linker based optimizations of GOT addressing sequences.

5.6.1.9   Relocations for thread-local storage

The static relocations needed to support thread-local storage in a SVr4-type environment are listed in the following table.

Static TLS relocations
Code Relocation Place Comment
104 R_ARM_TLS_GD32 Data General Dynamic Model
105 R_ARM_TLS_LDM32 Data Local Dynamic Model
106 R_ARM_TLS_LDO32 Data Local Dynamic Model
107 R_ARM_TLS_IE32 Data Initial Exec Model
108 R_ARM_TLS_LE32 Data Local Exec Model
109 R_ARM_TLS_LDO12 Arm LDR Local Dynamic Model
110 R_ARM_TLS_LE12 Arm LDR Local Exec Model
111 R_ARM_TLS_IE12GP Arm LDR Initial Exec Model

R_ARM_TLS_GD32 causes two adjacent entries to be added to the dynamically relocated section (the Global Offset Table, or GOT). The first of these is dynamically relocated by R_ARM_TLS_DTPMOD32, the second by R_ARM_TLS_DTPOFF32. The place resolves to the offset of the first of the GOT entries from the place.

R_ARM_TLS_LDM32 is the same as R_ARM_TLS_GD32 except that the second slot in the GOT is initialized to zero and has no dynamic relocation.

R_ARM_TLS_LDO32 resolves to the offset of the referenced data object (which must be local to the module) from the origin of the TLS block for the current module.

R_ARM_TLS_LDO12 is the same as R_ARM_TLS_LDO32 except that the result of the relocation is encoded as the 12-bit offset of an Arm LDR instruction.

R_ARM_TLS_LE32 resolves to the offset of the referenced data object (which must be in the initial data block) from the thread pointer ($tp).

R_ARM_TLS_LE12 is the same as R_ARM_TLS_LE32 except that the result of the relocation is encoded as the 12-bit offset of an Arm LDR instruction.

R_ARM_TLS_IE32 allocates an entry in the GOT that is dynamically relocated by R_ARM_TLS_TPOFF32. The place resolves to the offset of the GOT entry from the place.

R_ARM_TLS_IE12GP allocates an entry in the GOT that is dynamically relocated by R_ARM_TLS_TPOFF32. The place resolved to the offset of the GOT entry from the origin of the GOT and is encoded in the 12-bit offset of an Arm LDR instruction.

New experimental TLS relocations

TLSDESC contains a proposal for enhanced performance of TLS code. At this stage the proposal is still experimental, but the relocations R_ARM_TLS_DESC, R_ARM_TLS_GOTDESC, R_ARM_TLS_CALL, R_ARM_TLS_DESCSEQ, R_ARM_THM_TLS_CALL, R_ARM_THM_TLS_DESCSEQ16 and R_ARM_THM_TLS_DESCSEQ32 have been reserved to support this.

Note

The relocation R_ARM_TLS_DESC re-uses relocation code from the now-obsolete R_ARM_SWI24, but since the former was a static relocation and the new relocation is dynamic there are no practical conflicts in usage.

5.6.1.10   Dynamic relocations

The dynamic relocations for those execution environments that support only a limited number of run-time relocation types are listed in the following table.

Dynamic relocations
Code Relocation Comment
17 R_ARM_TLS_DTPMOD32

(S ≠ 0) Resolves to the module number of the module defining the specified TLS symbol, S.

(S = 0) Resolves to the module number of the current module (ie. the module containing this relocation).

18 R_ARM_TLS_DTPOFF32 Resolves to the index of the specified TLS symbol within its TLS block
19 R_ARM_TLS_TPOFF32

(S ≠ 0) Resolves to the offset of the specified TLS symbol, S, from the Thread Pointer, TP.

(S = 0) Resolves to the offset of the current module’s TLS block from the Thread Pointer, TP (the addend contains the offset of the local symbol within the TLS block).

20 R_ARM_COPY See below
21 R_ARM_GLOB_DAT Resolves to the address of the specified symbol
22 R_ARM_JUMP_SLOT Resolves to the address of the specified symbol
23 R_ARM_RELATIVE

(S ≠ 0) B(S) resolves to the difference between the address at which the segment defining the symbol S was loaded and the address at which it was linked.

(S = 0) B(S) resolves to the difference between the address at which the segment being relocated was loaded and the address at which it was linked.

With the exception of R_ARM_COPY all dynamic relocations require that the place being relocated is a word-aligned 32-bit object.

R_ARM_JUMP_SLOT is used to mark code targets that will be executed. On platforms that support dynamic binding the relocations may be performed lazily on demand. The unresolved address stored in the place will initially point to the entry sequence stub for the dynamic linker and must be adjusted during initial loading by the offset of the load address of the segment from its link address. Addresses stored in the place of these relocations may not be used for pointer comparison until the relocation has been resolved. In a REL form of this relocation the addend, A, is always 0.

R_ARM_COPY may only appear in executable objects where e_type is set to ET_EXEC. The effect is to cause the dynamic linker to locate the target symbol in a shared library object and then to copy the number of bytes specified by the st_size field to the place. The address of the place is then used to pre-empt all other references to the specified symbol. It is an error if the storage space allocated in the executable is insufficient to hold the full copy of the symbol. If the object being copied contains dynamic relocations then the effect must be as if those relocations were performed before the copy was made.

Note

R_ARM_COPY is normally only used in SVr4 type environments where the executable is not position independent and references by the code and read-only data sections cannot be relocated dynamically to refer to an object that is defined in a shared library.

The need for copy relocations can be avoided if a compiler generates all code references to such objects indirectly through a dynamically relocatable location, and if all static data references are placed in relocatable regions of the image. In practice, however, this is difficult to achieve without source-code annotation; a better approach is to avoid defining static global data in shared libraries.

5.6.1.11   Deprecated relocations

Deprecated relocations are in the process of being retired from the specification and may be removed or marked obsolete in future revisions. An object file containing these codes is still conforming, but producers should be changed to use the new alternatives.

The relocations R_ARM_GNU_VTENTRY and R_ARM_GNU_VTINHERIT have been used by some toolchains to facilitate unused virtual function elimination during linking. This method is not recommended and these relocations may be made obsolete in a future revision of this specification. These relocations may be safely ignored.

Deprecated relocations
Relocation Replacement
R_ARM_PC24 Use R_ARM_CALL or R_ARM_JUMP24
R_ARM_PLT32 Use R_ARM_CALL or R_ARM_JUMP24
R_ARM_LDR_SBREL_11_0_NC Use R_ARM_LDR_SB_Gxxx
R_ARM_ALU_SBREL_19_12_NC Use R_ARM_ALU_SB_Gxxx
R_ARM_ALU_SBREL_27_20_CK Use R_ARM_ALU_SB_Gxxx
R_ARM_SBREL31 Use new exception table format. Previous drafts of this document sometimes referred to this relocation as R_ARM_ROSEGREL32.
R_ARM_GNU_VTENTRY None
R_ARM_GNU_VTINHERIT None
5.6.1.12   Obsolete relocations

Obsolete relocations are no-longer used in this revision of the specification (but had defined meanings in a previous revision). Unlike deprecated relocations, there is no, or little known, use of these relocation codes. Conforming object producers must not generate these relocation codes and conforming linkers are not required to process them. Future revisions of this specification may re-assign these codes for a new relocation type.

5.6.1.13   Private relocations

Relocation types 112-127 and 161-176 are reserved as platform-specific relocations. The interpretation of these relocations is dependent on the value of the EI_OSABI field of the ELF header. If the value of EI_OSABI is zero or ELFOSABI_ARM_AEABI, these relocations are reserved.

5.6.1.14   Armv8.1-M Mainline Branch Future relocations

R_ARM_THM_BF16, R_ARM_THM_BF12 and R_ARM_THM_BF18 are used to relocate branch targets for the Branch Future instructions in Armv8.1-M Mainline. The formation of the initial addend in a REL type relocation for the various instruction classes is described in Armv8.1-M Mainline Branch Future relocation actions by instruction type. The immediates mentioned in this table are the ones used to represent the offset from the targetted branch instruction to the branch target using a two's complement immediate. That is this immediate should be the same as the one being used by the targetted branch instruction. Insn modification describes how the result X is written back to the instruction; Result_Mask is the value of X after the masking operation described in Armv8.1-M Mainline Branch Future relocations table has been applied.

In all cases the following semantics apply:

  • A Branch Future to an out of range label must either be replaced with a NOP or re-use an existing trampoline that jumps to the relocation target, it must not create a new trampoline. (It is up to the implementation to decide which of these options is preferable).
  • No errors are required.
Armv8.1-M Mainline Branch Future relocations
Code Name Overflow Instruction Result Mask
136 R_ARM_THM_BF16 Replace with NOP BF 16-bit immediate X & 0x0001FFFE
137 R_ARM_THM_BF12 Replace with NOP BFCSEL 12-bit immediate X & 0x00001FFE
138 R_ARM_THM_BF18 Replace with NOP BFL 18-bit immediate X & 0x0007FFFE
Armv8.1-M Mainline Branch Future relocation actions by instruction type
Instruction REL Addend Insn modification
BF (label, label) insn[20:16] << 12 | insn[10:1] << 2 | insn[11] << 1 | 1
insn[10:1] = (Result_Mask(X) >> 2) & 0x000003FF
insn[11] = (Result_Mask(X) >> 1) & 0x00000001
insn[20:16] = Result_Mask(X) >> 12
BFCSEL (label, label, label, cond) insn[16] << 12 | insn[10:1] << 2 | insn[11] << 1 | 1
insn[10:1] = (Result_Mask(X) >> 2) & 0x000003FF
insn[11] = (Result_Mask(X) >> 1) & 0x00000001
insn[16] = Result_Mask(X) >> 12
BFL (label, label) insn[22:16] << 12 | insn[10:1] << 2 | insn[11] << 1 | 1
insn[10:1] = (Result_Mask(X) >> 2) & 0x000003FF
insn[11] = (Result_Mask(X) >> 1) & 0x00000001
insn[22:16] = Result_Mask(X) >> 12
5.6.1.15   Unallocated relocations

All unallocated relocation types are reserved for use by future revisions of this specification.

All RELA type relocations are idempotent. They may be reapplied to the place and the result will be the same. This allows a static linker to preserve full relocation information for an image by converting all REL type relocations into RELA type relocations.

Note

A REL type relocation can never be idempotent because the act of applying the relocation destroys the original addend.

This section provides details of Arm-specific definitions and changes relating to executable images.

The Program Header provides a number of fields that assist in interpretation of the file. Most of these are specified in the base standard. The following fields have Arm-specific meanings.

p_type

The following table lists the processor-specific segment types.

Processor-specific segment types
Name p_type Meaning
PT_ARM_ARCHEXT 0x70000000 Platform architecture compatibility information
PT_ARM_EXIDX 0x70000001 Exception unwind tables
PT_ARM_UNWIND 0x70000001 alias for PT_ARM_EXIDX

A segment of type PT_ARM_ARCHEXT contains information describing the platform capabilities required by the executable file. The segment is optional, but if present it must appear before segment of type PT_LOAD. The platform independent parts of this segment are described in Platform architecture compatibility data.

PT_ARM_EXIDX (alias PT_ARM_UNWIND) describes the location of a program’s unwind tables.

p_flags

There are no processor-specific flags. All bits in the PT_MASKPROC part of this field are reserved to future revisions of this specification.

This data describes the platform capabilities required by an executable file. It can be constructed by a linker using the attributes [Addenda32] found in its input relocatable files, or otherwise.

If this segment is present it shall contain at least one 32-bit word with meaning defined by Common architecture compatibility data masks, Architecture compatibility data formats, Architecture profile compatibility data, and Architecture version compatibility data below.

Common architecture compatibility data masks

Name Value Meaning
PT_ARM_ARCHEXT_FMTMSK 0xff000000 Masks bits describing the format of data in subsequent words. The masked value is described in Architecture compatibility data formats, below.
PT_ARM_ARCHEXT_PROFMSK 0x00ff0000 Masks bits describing the architecture profile required by the executable. The masked value is described in Architecture profile compatibility data, below.
PT_ARM_ARCHEXT_ARCHMSK 0x000000ff Masks bits describing the base architecture required by the executable. The masked value is described in Architecture version compatibility data, below.

Architecture compatibility data formats

The following table lists the architecture compatibility data formats defined by this ABI. All other format identifiers are reserved to future revisions of this specification.

Name Value Meaning
PT_ARM_ARCHEXT_FMT_OS 0x00000000 There are no additional words of data. However, if EF_OSABI is non-zero, the relevant platform ABI may define additional data that follows the initial word.
PT_ARM_ARCHEXT_FMT_ABI 0x01000000 Platform architecture compatibility data (ABI format), below, describes the format of the following data words.

Architecture profile compatibility data

The following table lists the values specifying the architectural profile needed by an executable file.

Name Value Meaning
PT_ARM_ARCHEXT_PROF_NONE 0x00000000 The architecture has no profile variants, or the image has no profile-specific constraints
PT_ARM_ARCHEXT_PROF_ARM

0x00410000

(‘A’<<16)

The executable file requires the Application profile
PT_ARM_ARCHEXT_PROF_RT

0x00520000

(‘R’<<16)

The executable file requires the Real-Time profile
PT_ARM_ARCHEXT_PROF_MC

0x004D0000

(‘M’<<16)

The executable file requires the Microcontroller profile
PT_ARM_ARCHEXT_PROF_CLASSIC

0x00530000

(‘S’<<16)

The executable file requires the ‘classic’ (‘A’ or ‘R’ profile) exception model.

Architecture version compatibility data

The following table defines the values that specify the minimum architecture version needed by this executable file. These values are identical to those of the Tag_CPU_arch attribute used in the attributes section [Addenda32] of a relocatable file.

Name Value Meaning the executable file needs (at least) ...
PT_ARM_ARCHEXT_ARCH_UNKN 0x00 The needed architecture is unknown or specified in some other way
PT_ARM_ARCHEXT_ARCHv4 0x01 Architecture v4
PT_ARM_ARCHEXT_ARCHv4T 0x02 Architecture v4T
PT_ARM_ARCHEXT_ARCHv5T 0x03 Architecture v5T
PT_ARM_ARCHEXT_ARCHv5TE 0x04 Architecture v5TE
PT_ARM_ARCHEXT_ARCHv5TEJ 0x05 Architecture v5TEJ
PT_ARM_ARCHEXT_ARCHv6 0x06 Architecture v6
PT_ARM_ARCHEXT_ARCHv6KZ 0x07 Architecture v6KZ
PT_ARM_ARCHEXT_ARCHv6T2 0x08 Architecture v6T2
PT_ARM_ARCHEXT_ARCHv6K 0x09 Architecture v6K
PT_ARM_ARCHEXT_ARCHv7 0x0A Architecture v7 (in this case the architecture profile may also be required to fully specify the needed execution environment)
PT_ARM_ARCHEXT_ARCHv6M 0x0B Architecture v6M (e.g. Cortex-M0)
PT_ARM_ARCHEXT_ARCHv6SM 0x0C Architecture v6S-M (e.g. Cortex-M0)
PT_ARM_ARCHEXT_ARCHv7EM 0x0D Architecture v7E-M
6.1.1.1   Platform architecture compatibility data (ABI format)

The status of this section is informative. It records a proposal that might be adopted.

The data following the word defined by Platform architecture compatibility data consists of an array of 2-byte signed integers (starting at offset 4 in the architecture compatibility data segment) followed by a number of null-terminated byte strings (NTBS). The p_filesz field of the segment header gives the total size in bytes of the architecture compatibility data.

The integer array maps the ABI public attribute tags [Addenda32] as follows.

  • Array[0] contains the number of elements in array.
  • If tag ≥ array[0], the value of tag for the executable file is 0. Only tags with non-0 values need to be mapped.
  • If 4 ≤ tag < array[0], the value of tag for the executable file is array[tag]. A negative value v denotes that tag has the NTBS value found at offset –v from the start of the segment.
  • Array[1] contains the major version number and array[2] the minor version number of the ABI release to which the data conforms (at least 2, 8). Array[3] is reserved and should be 0.

There are no processor-specific definitions relating to program loading.

The following table lists the processor-specific dynamic array tags.

AArch32-specific dynamic array tags
Name Value d_un Executable Shared Object
DT_ARM_RESERVED1 0x70000000      
DT_ARM_SYMTABSZ 0x70000001 d_val Platform specific Platform specific
DT_ARM_PREEMPTMAP 0x70000002 d_ptr Platform specific Platform specific
DT_ARM_RESERVED2 0x70000003      

DT_ARM_SYMTABSZ gives the number of entries in the dynamic symbol table, including the initial dummy symbol.

DT_ARM_PREEMPTMAP holds the address of the pre-emption map for platforms that use the DLL static binding model. See Symbol Pre-emption in DLLs for details. On platforms that permit use of a pre-emption map, the DT_SONAME tag must be present in all shared objects.

Note

Some executable images may exist that use DT_ARM_RESERVED1 and DT_ARM_RESERVED2 instead of DT_ARM_SYMTABSZ and DT_ARM_PREEMPTMAP respectively. These tags use the d_un field in a manner incompatible with the Generic ELF requirements.

For some execution environments a further processing step may be needed after linking before an executable can be run on the target environment. The precise processing may depend on both the target platform. Depending on the nature of the post-processing it may be done in any of following places

  • As a final step during linking
  • As a preliminary step during execution of the image
  • As a separate post-linking step

In some cases the result may still be an ELF executable image, in others it may produce an image that is in some other format more appropriate to the operating system.

Images that are expected to execute in big-endian mode on processors that implement Architecture version 6 or higher will normally need to be post-processed to convert the instructions that are in big-endian byte order to little-endian as expected by the processor. The mapping symbol information can be used to do this transformation accurately. In all segments that contain executable code:

  • For areas mapped as data ($d or $d.<any…>) no changes are made
  • For areas mapped as Thumb ($t or $t.<any…>) each half-word aligned pair of bytes are swapped
  • For areas mapped as Arm ($a or $a.<any…>) each word-aligned object is swapped so that the first and fourth bytes are exchanged and the second and third exchanged.

An ELF image that has been transformed in this manner is marked by setting EF_ARM_BE8 in the e_flags field.

Note

If BE-8 images are subject to further relocation of instructions (either by a dynamic linker or by further post-linking operations) account must be taken of the fact that the instructions are now in little-endian format.

The status of this appendix is informative.

The simplest code sequence for the PLT entry corresponding to imported symbol X is:

LDR   ip, [pc, #0]         ; Load the 32-bit offset of my PLTGOT entry from SB
LDR   pc, [ip, sb]!        ; Branch indirect through the PLTGOT entry
                           ;   leaving ip addressing the PLTGOT slot
DCD   R_ARM_GOT_BREL(X)    ; GOT_BASE = SB

The final DCD is subject to relocation by a PLTGOT-generating relocation directive. This directive may be processed by a target-specific linker or by a target-specific post-linker. After processing:

  • The place contains the 32-bit offset from the static base (sb) of the PLTGOT entry for X.
  • The PLTGOT entry for X is subject to an R_ARM_JUMP_SLOT(X) dynamic relocation.

A more complicated sequence that avoids one of the memory accesses is:

ADD   ip, sb, #:SB_OFFSET_27_20:__PLTGOT(X)    ; R_ARM_ALU_SB_G0_NC(__PLTGOT(X))
ADD   ip, ip, #:SB_OFFSET_19_12:__PLTGOT(X)    ; R_ARM_ALU_SB_G1_NC(__PLTGOT(X))
LDR   pc, [ip, #:SB_OFFSET_11_0:__PLTGOT(X)]!  ; R_ARM_LDR_SB_G2(__PLTGOT(X))

If the linker can place all PLTGOT entries within 1MB of SB, the sequence becomes:

ADD   ip, sb, #:SB_OFFSET_19_12:__PLTGOT(X)    ; R_ARM_ALU_SB_G0_NC(__PLTGOT(X))
LDR   pc, [ip, #:SB_OFFSET_11_0:__PLTGOT(X)]!  ; R_ARM_LDR_SB_G1(__PLTGOT(X))

The write-back on the final LDR ensures that ip contains the address of the PLTGOT entry. This is critical to incremental dynamic linking.

The code sequence for the PLT entry corresponding to imported symbol X is:

LDR   ip, [pc, #0]      ; Load the 32-bit address of my PLTGOT entry
LDR   pc, [ip]          ; Branch indirect through the PLTGOT entry
DCD   R_ARM_GOT_ABS(X)  ; GOT_BASE = 0

Note that ip addresses the PLTGOT entry, which is critical to incremental dynamic linking.

The final DCD is subject to relocation by a PLTGOT-generating relocation directive. This directive may be processed by a target-specific linker or by a target-specific post-linker. After processing:

  • The place contains the 32-bit address of the PLTGOT entry for X.
  • The PLTGOT entry for X is subject to an R_ARM_JUMP_SLOT(X) dynamic relocation.

Because a DLL has two segments that can be loaded independently, there is no more efficient address generating sequence – analogous to the SB-relative sequence shown in above – that does not require complex instruction field-relocating directives to be processed at dynamic link time.

This ABI requires dynamic relocations to relocate 32-bit fields, so there is no sequence analogous to that of the preceding subsection.

The simplest code sequence for the PLT entry corresponding to imported symbol X is:

     LDR   ip, L2         ; Load the 32-bit pc-relative offset of my PLTGOT entry
L1:  ADD   ip, ip, pc     ; formulate its address...
     LDR   pc, [ip]       ; Branch through the PLTGOT entry addressed by ip
L2:  DCD   R_ARM_GOT_PREL(X) + (L2 – L1 – 8)

The dynamic linker relies on ip addressing the PLTGOT entry for X.

The final DCD is subject to static relocation by a PLTGOT-generating relocation directive. This directive may be processed by a target-specific linker or by a target-specific post-linker. After processing:

  • The place contains the 32-bit offset from L1+8 to the PLTGOT entry for X.
  • The PLTGOT entry for X is subject to an R_ARM_JUMP_SLOT(X) dynamic relocation.

A more complicated, pc-relative, sequence that avoids one of the memory accesses is shown below. Because an SVr4 executable file is compact (usually < 228 bytes) and rigid (it has only one base address, whereas a DLL has two), all the relocations can be fully resolved during static linking.

ADD   ip, pc, #-8:PC_OFFSET_27_20: __PLTGOT(X)    ; R_ARM_ALU_PC_G0_NC(__PLTGOT(X))
ADD   ip, ip, #-4:PC_OFFSET_19_12: __PLTGOT(X)    ; R_ARM_ALU_PC_G1_NC(__PLTGOT(X))
LDR   pc, [ip, #0:PC_OFFSET_11_0: __PLTGOT(X)]!   ; R_ARM_LDR_PC_G2(__PLTGOT(X))

The write-back on the final LDR ensures that ip contains the address of the PLTGOT entry. This is critical to incremental dynamic linking.

In effect, the sequence constructs a 28-bit offset for the LDR. The first relocation does the right thing because pc addresses the LDR, so, in general, it picks out bits [27-20] of that offset. The third relocation picks out bits [11-0] of the same offset. The second relocation needs to construct bits [19-12] of the offset from dot+4 to X., that is, from dot to X-4. Ignoring the -4 sometimes produces the wrong answer!

Encoding such a small addend requires that the initial value not be shifted by the shift applied to the result value. This is expected for a RELA-type relocation that can encode -4 directly. However, a REL-type must encode the initial value of the addend using SUB ip, ip, #4.

In small enough DSOs (< 220 bytes from the PLT to the PLTGOT) the first instruction can be omitted, and the sequence collapses to the following.

SUB   ip, pc, #4:PC_OFFSET_19_12: __PLTGOT(X)    ; R_ARM_ALU_PC_G0_NC(__PLTGOT(X))
LDR   pc, [ip, #0:PC_OFFSET_11_0: __PLTGOT(X)]!  ; R_ARM_LDR_PC_G2(__PLTGOT(X))

An SVr4 executable does not need be position independent, its writable segment can be relocated dynamically, and it is compact and rigid. Therefore, its PLT entries can use the simple, absolute code sketched in DLL-like, multiple virtual address space, PLT linkage or the more complex, pc-relative, versions sketched in SVr4 DSO-like PLT linkage, as the tool chain chooses.

In both cases, ip must address the corresponding PLTGOT slot at the point where the PLT calls through it.

The status of this appendix is informative.

A toolchain is not required to support any of the conventions described in this appendix; however, it is recommended that if symbols matching the patterns described are used, then the following conventions are adhered to.

A number of symbols may be used to delimit the addresses and sizes of aspects of a linked image. These symbols are of the following general forms:

Load$$region_name$$Base
Image$$region_name$${Base|Length|Limit}
Image$$region_name$$ZI$${Base|Length|Limit}
Image$${RO|RW|ZI}$${Base|Limit}
SectionName$${Base|Limit}

A toolchain may define these symbols unconditionally, or only if they are referred to by the application: so a post-linker must not depend on the existence of any of these symbols.

A symbol $Sub$$name is the sub-class version of name. A symbol $Super$$name is the super-class version of name. In the presence of a definition of both name and $Sub$$name:

  • A reference to name resolves to the definition of $Sub$$name.
  • A reference to $Super$$name resolves to the definition of name.

It is an error to refer to $Sub$$name, or to define $Super$$name, or to use $Sub$$… or $Super$$… recursively.

A veneer symbol has the same binding as the symbol it veneers. They are used to label sequences of instructions that are automatically generated during linking. The general format of the symbols is:

${Ven|other}${AA|AT|TA|TT}${I|L|S}[$PI]$$symbol_name

where AA, AT, TA, or TT denotes the type of the veneer — Arm to Arm, Arm to Thumb, etc; I, L, or S denotes inline (the target follows immediately), long reach (32-bit), or short reach (typically 26-bit); and $PI denotes that the veneer is position independent.