Hi, I've been struggling to make a RISC-V cross compiler. This `manifest.scm`file failed to create because there was no available dynamic linker forthe targets which, in the end, don't need any dynamic linker: manifest.scm: ``` scm (use-modules (gnu packages cross-base) (gnu packages embedded) (gnu packages man)) (packages->manifest `(,@(let* ((triplet "riscv32-unknown-elf") (binutils (cross-binutils triplet))) (list binutils (cross-gcc triplet #:xbinutils binutils #:libc #f))) ,man-db))``` I realized there were some targets missing (and there are more!) in thebootstrap.scm file, so I think adding them is harmless and lets usmake cross compilers that are not available otherwise. I'd like to add riscv32-linux too, but that involves setting the correctpath and I don't know it. Hope this patch is helpful.
From 58785556b7448cfe32916d3842bce398b360c2de Mon Sep 17 00:00:00 2001From: Ekaitz Zarraga <email@example.com>Date: Sat, 23 Jan 2021 22:50:44 +0100Subject: [PATCH] gnu: glibc-dynamic-linker: Add riscv*-elf targets * gnu/packages/bootstrap.scm: (glibc-dynamic-linker): Set riscv64-*-elf targets to "no-ld"--- gnu/packages/bootstrap.scm | 2 ++ 1 file changed, 2 insertions(+)
Toggle diff (13 lines)diff --git a/gnu/packages/bootstrap.scm b/gnu/packages/bootstrap.scmindex c39e60ec8b..8566d11284 100644--- a/gnu/packages/bootstrap.scm+++ b/gnu/packages/bootstrap.scm@@ -302,6 +302,8 @@ or false to signal an error." ((string=? system "i686-mingw") "no-ld.so") ((string=? system "x86_64-mingw") "no-ld.so") ((string=? system "vc4-elf") "no-ld.so")+ ((string=? system "riscv32-elf") "no-ld.so")+ ((string=? system "riscv64-elf") "no-ld.so") (else (error "dynamic linker name not known for this system" system))))
For more information on the dynamic linker names I didn't know: lp32, ilp32f and ilp32d are the ABI. i -> integerl -> longp -> pointerf -> floatd -> double The chars in the beginning indicate what kind of data is sent to 32 bit registers. The suffix says which ones are sent to floating point registers. So in ilp32 floating point numbers are sent to integer registers, while in ilp32d double precision numbers are sent to floating point registers. Not every ABI is compatible with every single RISC-V configuration. RV32I for instance, doesn't support floating point numbers (they are supported in the F extension) so it doesn't provide floating point registers meaning that it's not able to use any ABI that has a `d` or an `f` as suffix. But RISC-V configurations with floating point modules can actually run the simplest ABI if they want so we probably need to add many entries here :) Does this make sense?