ludo@gnu.org (Ludovic Courtès) writes:

> Hello Marius!
>
> Marius Bakke <mbakke@fastmail.com> skribis:
>
>> * gnu/packages/patches/binutils-aarch64-symbol-relocation.patch: New file.
>> * gnu/local.mk (dist_patch_DATA): Register it.
>> * gnu/packages/base.scm (binutils-sans-bug-22764): New public variable.
>> * gnu/packages/linux.scm (make-linux-libre)[native-inputs]: On aarch64, define
>> new ld-wrapper with the above binutils and use it.
>
> Woohoo, excellent!
>
>> +++ b/gnu/packages/patches/binutils-aarch64-symbol-relocation.patch
>> @@ -0,0 +1,108 @@
>> +Fix a regression in Binutils 2.30 where some symbols are incorrectly assumed
>> +to be addresses:
>> +
>> +https://sourceware.org/bugzilla/show_bug.cgi?id=22764
>> +
>> +Patch taken from upstream (with ChangeLog entries omitted):
>> +
>> +https://sourceware.org/git/gitweb.cgi?p=binutils-gdb.git;a=commitdiff;h=279b2f94168ee91e02ccd070d27c983fc001fe12
>> +
>> +diff --git a/bfd/elfnn-aarch64.c b/bfd/elfnn-aarch64.c
>> +index af448f9..2737773 100644
>> +--- a/bfd/elfnn-aarch64.c
>> ++++ b/bfd/elfnn-aarch64.c
>> +@@ -7189,10 +7189,19 @@ elfNN_aarch64_check_relocs (bfd *abfd, struct bfd_link_info *info,
>> + #if ARCH_SIZE == 64
>> + 	case BFD_RELOC_AARCH64_32:
>> + #endif
>> +-	  if (bfd_link_pic (info)
>> +-	      && (sec->flags & SEC_ALLOC) != 0
>> +-	      && (sec->flags & SEC_READONLY) != 0)
>> ++	  if (bfd_link_pic (info) && (sec->flags & SEC_ALLOC) != 0)
>> + 	    {
>> ++	      if (h != NULL
>> ++		  /* This is an absolute symbol.  It represents a value instead
>> ++		     of an address.  */
>> ++		  && ((h->root.type == bfd_link_hash_defined
>> ++		       && bfd_is_abs_section (h->root.u.def.section))
>> ++		      /* This is an undefined symbol.  */
>> ++		      || h->root.type == bfd_link_hash_undefined))
>> ++		break;
>> ++
>> ++	      /* For local symbols, defined global symbols in a non-ABS section,
>> ++		 it is assumed that the value is an address.  */
>> + 	      int howto_index = bfd_r_type - BFD_RELOC_AARCH64_RELOC_START;
>> + 	      _bfd_error_handler
>> + 		/* xgettext:c-format */
>
> Perhaps you can omit the ld/testsuite bits that follow (I don’t think we
> run this test suite, do we?).

Good idea.  I dropped the tests since we don't use them and pushed as
cdbcfc113bfc02e58bdf6ec42c545c479681602f.