Why does GDB default to using compressed breakpoint instructions for s/w BPs? about riscv-binutils-gdb HOT 9 CLOSED

isn't gdb aware of the architecture details?

from riscv-binutils-gdb.

I think it should be from misa and maybe other csrs but at the moment to debug our specific RV32IM I have to do this otherwise gdb will assume a 64 bit target and ability to use compressed BP instructions:

set arch riscv:rv32
set riscv use_compressed_breakpoints no

from riscv-binutils-gdb.

gdb attempts to read MISA, and if C is set it will use compressed breakpoints: https://github.com/riscv/riscv-binutils-gdb/blob/riscv-binutils-2.29/gdb/riscv-tdep.c#L231

In older versions of the code it would be unable to read MISA if it was at the old location (it moved in some priv spec update). 697d5b8 fixed that issue.

from riscv-binutils-gdb.

But why bother trying to use compressed BP instructions when uncompressed BP instructions are universally compatible?
Seems like unnecessary complication for the implementation and the end user to use compressed BP instructions?

from riscv-binutils-gdb.

The software breakpoint instruction can be no longer than the instruction it is replacing, in case the program contains a jump to the instruction right after it. Rather than figuring out the size of the instruction, it's much simpler just to use the smallest supported breakpoint instruction.

from riscv-binutils-gdb.

The software breakpoint instruction can be no longer than the instruction it is replacing, in case the program contains a jump to the instruction right after it. Rather than figuring out the size of the instruction, it's much simpler just to use the smallest supported breakpoint instruction.

Thanks Tim - that explains matters alright.

gdb attempts to read MISA, and if C is set it will use compressed breakpoints: https://github.com/riscv/riscv-binutils-gdb/blob/riscv-binutils-2.29/gdb/riscv-tdep.c#L231

In older versions of the code it would be unable to read MISA if it was at the old location (it moved in some priv spec update). 697d5b8 fixed that issue.

So a build of gdb with that mod no longer needs:

set riscv use_compressed_breakpoints no

regardless of the draft privilege spec that the target adheres to (1.9, 1.9.1 or 1.10)?

from riscv-binutils-gdb.

So a build of gdb with that mod no longer needs:

set riscv use_compressed_breakpoints no

I just tried it with an RV32IM adhering to the 1.9 priv spec and that seems to be the case - i.e. it does the right thing (uses non compressed BP instruction) even when set riscv use_compressed_breakpoints no) is NOT specified.

Unfortunately set arch riscv:rv32 is still needed - shouldn't it be able to figure out the XLEN from misa or some other CSR to obviate the need for this?

from riscv-binutils-gdb.

Unfortunately set arch riscv:rv32 is still needed - shouldn't it be able to figure out the XLEN from misa or some other CSR to obviate the need for this?

Yes, but unfortunately it's not so easily implemented. This is #54

from riscv-binutils-gdb.

Since the rationale for using compressed bps where supported was explained above I'm closing this issue now.

from riscv-binutils-gdb.