Skip to content

Document exception handling, add R_RISCV_32_PCREL description #110

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 3 commits into
base: master
Choose a base branch
from
Open

Document exception handling, add R_RISCV_32_PCREL description #110

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
fd4df20
Add a description of DWARF exception handling
edward-jones Jul 16, 2019
e701d04
Document R_RISCV_PCREL_32 relocation
edward-jones Jul 16, 2019
cd79667
Describe registers used when passing values through to a landing pad
edward-jones Jul 18, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
81 changes: 79 additions & 2 deletions riscv-elf.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,10 @@
* [Note Sections](#note-sections)
* [Dynamic Table](#dynamic-table)
* [Hash Table](#hash-table)
4. [DWARF](#dwarf)
4. [Exception Handling](#exception-handling)
* [Encodings](#exception-handling-encodings)
* [Registers](#exception-handling-registers)
5. [DWARF](#dwarf)
* [Dwarf Register Numbers](#dwarf-register-numbers)

## Copyright and license information
Expand Down Expand Up @@ -436,7 +439,8 @@ Enum | ELF Reloc Type | Description | Details
54 | R_RISCV_SET8 | Local label subtraction |
55 | R_RISCV_SET16 | Local label subtraction |
56 | R_RISCV_SET32 | Local label subtraction |
57-191 | *Reserved* | Reserved for future standard use |
57 | R_RISCV_32_PCREL | PC-relative 32-bit offset | Primarily used for FDE start location in .eh_frame
58-191 | *Reserved* | Reserved for future standard use |
192-255 | *Reserved* | Reserved for nonstandard ABI extensions |

Nonstandard extensions are free to use relocation numbers 192-255 for any
Expand Down Expand Up @@ -791,6 +795,79 @@ typedef struct

## <a name=hash-table></a>Hash Table

# <a name=exception-handling></a>Exception Handling

Exception handling is achieved using C++ exceptions based on DWARF-derived
CIE (Common Information Entry) and FDE (Frame Descriptor Entry) data
structures saved in `.eh_frame`, plus additional data
structures in the `.gcc_except_table` section of the ELF. Some relevant
fields in these data structures are defined below:

## <a name=exception-handling-encodings></a>Encodings

### Frame Descriptor Entry (FDE) encoding

Used to encode the range of code addresses which the current Frame
Descriptor Entry applies to. Generally encoded as:

```
DW_EH_PE_pcrel | DW_EH_PE_sdata4
Copy link
Collaborator

@jim-wilson jim-wilson Jul 17, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't have support for 64-bit DWARF in the gcc RISC-V port, so yes, this is always DW_EH_PE_sdata4 currently. I don't anticipate that we will ever need 64-bit DWARF. If someone does insist on adding support for that, then we will need a R_RISCV_64_PCREL relocation which does not exist yet. Most gcc ports don't bother to support 64-bit DWARF. It is only useful if you have binaries over 4GB in size, and if you do have binaries that big then you probably have other more important problems than unwind info.

```

This encoding is used both to specify the start address for the code
that the FDE is applicable to, and the length of the code which
the FDE applies to. In the first instance, a 32-bit signed offset
is emitted from the current position in the FDE to the code address
itself, and this generates a `R_RISCV_32_PCREL` relocation. In the
second instance, a `R_RISCV_ADD32` and `R_RISCV_SUB32` pair is
Copy link
Collaborator

@jim-wilson jim-wilson Jul 17, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Technically, the R_RISCV_ADD32 and R_RISCV_SUB32 relocations are only required when linker relaxation is enabled, since otherwise we can't know the size of the function until after linker relaxation. Unfortunately, the toolchain currently emits the relocs always. There is an open bug report somewhere asking us to fix this, so in the future these relocs may not be present when linker relaxation is disabled.

emitted which encodes the difference between the end address and
start address in the code.

### Language Specific Data Area (LSDA) encoding

Encodes refences to the LSDA for the current FDE. Like the FDE encoding
this is generally encoded as:

```
DW_EH_PE_pcrel | DW_EH_PE_sdata4
```

This causes a `R_RISCV_ADD32` and `R_RISCV_SUB32` relocation pair
to be emitted, encoding the offset from the position the LSDA is
encoded in the FDE (in the `.eh_frame` section) to the LSDA data
structure itself (which is found in the `.gcc_except_table` section).

### Personality Function encoding

Use to encode references to the personality function for the current
CIE. Unlike the LSDA and FDE, the personality function is encoded
with an additional level of indirection.

```
DW_EH_PE_indirect | DW_EH_PE_pcrel | DW_EH_PE_sdata4
```

This is also encoded as a `R_RISCV_ADD32` and `R_RISCV_SUB32` pair,
however the target is not the direct address of the
`__gxx_personality_v0` function, but instead references a temporary
symbol `DW.ref.__gxx_personality_v0` which contains the full address
of `__gxx_personality_v0` and is found in a special section
`.sdata.DW.ref.__gxx_personality_v0`.

The reason for this is that it allows the `.eh_frame` section to
contain no dynamic relocations - since `__gxx_personality_v0` may
be in a shared object - and mean the `.eh_frame` section does not
need to be writeable at runtime. Instead, only the new temporary
`.sdata.DW.ref.__gxx_personality_v0` section is writeable.

## <a name=exception-handling-registers></a>Registers

When an exception is caught, data needs to be passed from the
personality function to the landing pad. Four scratch registers are
used for this purpose, and these parameters are passed to the landing
pad according to the integer calling convention; they are passed in
registers `a0`, `a1`, `a2` and `a3`.

# <a name=dwarf></a>DWARF

Dwarf Register Numbers <a name=dwarf-register-numbers>
Expand Down