haskell.compiler: Apply patches to fix PPC64 ELFv2 support

[OPNA2608]

Fixes pkgsCross.ppc64-elfv2.haskellPackages.hello

Things done

• Built on platform:
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• Tested, as applicable:
  • NixOS tests in nixos/tests.
  • Package tests at passthru.tests.
  • Tests in lib/tests or pkgs/test for functions and "core" functionality.
• Ran nixpkgs-review on this PR. See nixpkgs-review usage.
• Tested basic functionality of all binary files, usually in ./result/bin/.
• Nixpkgs Release Notes
  • Package update: when the change is major or breaking.
• NixOS Release Notes
  • Module addition: when adding a new NixOS module.
  • Module update: when the change is significant.
• Fits CONTRIBUTING.md, pkgs/README.md, maintainers/README.md and other READMEs.

---

Add a 👍 reaction to pull requests you find important.

[alexfmpe]

Dunno what branch this needs to go against yet

Should target staging going by the "5001+" rebuilds label

Doesn't apply on GHC 9.8, needs manual backport

As in, patch fails? Then let's add a versionAtLeast. 9.10 is the default anyway (which is why this causes so many rebuilds) so I think the few if any people people wanting this platform/ghc can just update ghc

[OPNA2608]

Should target staging going by the "5001+" rebuilds label

Yepyep, I'll fix that before undrafting. I just don't have the spoons for this rn, so it'll take abit. Hence, draft.

As in, patch fails?

Fails to apply, yeah.

---

Edit:

9.10 is the default anyway (which is why this causes so many rebuilds) so I think the few if any people people wanting this platform/ghc can just update ghc

Actually, 9.4 is kind of being used for testing cross-compilation of GHC, and maybe future bootstrapping of our GHCs via tarballs of cross-compiled static GHC builds. On powerpc64 that GHC would have to be ELFv2.

[OPNA2608]

Made the fetchpatched version of the patch work with Hadrian-using GHCs <=9.8, and manually backported it to GHC 9.4 to allow building of a cross-compiled GHC. Should be fine now, I think.

[wolfgangwalther]

Diff LGTM; was not able to test.

[alexfmpe]

I can test a couple builds sometime this week if we want more confidence?

[wolfgangwalther]

Sure, that would be appreciated.

[alexfmpe]

$ nix-build -A pkgsCross.ppc64-musl.haskellPackages.hello
/nix/store/xfklsd7j6ks01ybz4rk5jmx4zdl06xfr-hello-powerpc64-unknown-linux-musl-1.0.0.2

$(nix-build -A qemu-user)/bin/qemu-ppc64 /nix/store/xfklsd7j6ks01ybz4rk5jmx4zdl06xfr-hello-powerpc64-unknown-linux-musl-1.0.0.2/bin/hello
Hello, World!

$ nix-build -A pkgsCross.ppc64-elfv2.haskellPackages.hello
/nix/store/8ymx3z3g75sgfsfn5pfqvl43g6k1fbj2-hello-powerpc64-unknown-linux-gnuabielfv2-1.0.0.2

$ $(nix-build -A qemu-user)/bin/qemu-ppc64 /nix/store/8ymx3z3g75sgfsfn5pfqvl43g6k1fbj2-hello-powerpc64-unknown-linux-gnuabielfv2-1.0.0.2/bin/hello
qemu: uncaught target signal 11 (Segmentation fault) - core dumped
[1]    2457937 segmentation fault (core dumped)  ./result-qemu-user/bin/qemu-ppc64 result/bin/hello

$ nix-build -A pkgsCross.ppc64-elfv2.haskell.compiler.ghc94 --keep-going
error: builder for '/nix/store/lhivz1xls2qlz8mhi7l1s064hqrfif90-gcc-15.2.0-powerpc64-unknown-linux-gnuabielfv2.drv' failed with exit code 2;

[OPNA2608]

$ $(nix-build -A qemu-user)/bin/qemu-ppc64 /nix/store/8ymx3z3g75sgfsfn5pfqvl43g6k1fbj2-hello-powerpc64-unknown-linux-gnuabielfv2-1.0.0.2/bin/hello
qemu: uncaught target signal 11 (Segmentation fault) - core dumped
[1]    2457937 segmentation fault (core dumped)  ./result-qemu-user/bin/qemu-ppc64 result/bin/hello

Can confirm the crash on hardware. Unfortunately, we seem to have more fundamental issues with glibc-elfv2 when it comes to native building, so my focus is mainly on glibc-elfv1 (default glibc ABI on this platform, our default for pkgs) & musl-elfv2 (only ABI musl supports, pkgsStatic).

---

ppc64-elfv1 & ppc64-musl hellos work.

puna@Yubel ~> /nix/store/n56nxj73hw6s45cix6mjfs207iqqcjyq-hello-powerpc64-unknown-linux-gnuabielfv1-1.0.0.2/bin/hello 
Hello, World!
puna@Yubel ~> /nix/store/wfi73nz25xvcrhia9vzdcxrh4jgn6676-hello-powerpc64-unknown-linux-musl-1.0.0.2/bin/hello 
Hello, World!

ghc94s build, but don't run.

puna@Yubel ~> /nix/store/dhv857lm0ahyfbg7np9jprg5j0b9nkz1-ghc-powerpc64-unknown-linux-gnuabielfv1-9.4.8/bin/ghci
fish: Job 1, '/nix/store/dhv857lm0ahyfbg7np9j…' terminated by signal SIGSEGV (Address boundary error)
puna@Yubel ~ [SIGSEGV]> /nix/store/p5s6ai4z34m72qr30174vwbdcq6qhc26-ghc-powerpc64-unknown-linux-gnuabielfv2-9.4.8/bin/ghci
fish: Job 1, '/nix/store/p5s6ai4z34m72qr30174…' terminated by signal SIGSEGV (Address boundary error)
puna@Yubel ~ [SIGSEGV]> /nix/store/lfqpkyr81rld6sb25z8h8lzhisssh88m-ghc-musl-powerpc64-unknown-linux-musl-9.4.8/bin/ghci
fish: Job 1, '/nix/store/lfqpkyr81rld6sb25z8h…' terminated by signal SIGSEGV (Address boundary error)

Identical to current behaviour of ppc64-elfv1 on master. I'm not gonna try it, but flipping enableUnregisterised to true might fix them. Noted in #439258 (comment), I think the suitability of cross-compiled ghc94 is still getting looked into though?

[wolfgangwalther]

ghc94s build, but don't run.

I'm unsure about the patch for GHC 9.4, if we can't verify that it works. Should we postpone that until these run in the first place?

[OPNA2608]

Seems like ppc64le has similar issues with cross-compiled registerised GHC segfaulting. Maybe a general issue on (64-bit) POWER? #439258 (comment)

I'm fine with pulling the GHC 9.4 changes out for now until they're actually needed (bootstrapping off of cross-compiled GHC).

Edit: Extracted GHC 9.4 changes into #475036.

[vcunat]

Apparently you don't want to apply it to 9.12.3?

applying patch /nix/store/frl3xlnjgsiisd57qzhsl64ch3l7p8yk-ghc-rts-Fix-compile-on-powerpc64-elf-v1.patch
patching file rts/StgCRunAsm.S
Reversed (or previously applied) patch detected!  Assume -R? [n] 

It seems that it's the only version with this problem, at least among those built on Hydra:
https://hydra.nixos.org/eval/1822157?filter=haskell.compiler&compare=1822139#tabs-now-fail

[OPNA2608]

Looks like this was backported upstream, yeah.

https://gitlab.haskell.org/ghc/ghc/-/merge_requests/13894#note_642957
https://gitlab.haskell.org/ghc/ghc/-/merge_requests/14835

      ++ lib.optionals ((lib.versions.majorMinor haskell.compiler.ghc9123.version == "9.12" && lib.versionOlder haskell.compiler.ghc9123.version "9.12.3") || (lib.versions.majorMinor haskell.compiler.ghc9123.version != "9.12" && lib.versionOlder haskell.compiler.ghc9123.version "9.14.1")) [

I guess?