On Wed, Mar 21, 2018 at 09:41:11PM +0100, Jakub Jelinek wrote:
> I've committed my last patch to unbreak aarch64 and powerpc* bootstrap
> in a rush just after watching it passed bootstrap, not waiting for test
> results.
> Unfortunately, it seems the DECL_INITIAL (decl) = decl; hack wasn't used as
> the comment suggested only by dw2_force_const_mem, but also by asan.c.
> In that case forcing flag_section_anchors temporarily to 0 is too fragile,
> so this patch instead reverts the varasm.c change (just provides a different
> comment) and for good measure adds the DECL_INITIAL hack on the dw2 decl
> after it is emitted (we still use the temporary flag_section_anchors = 0
> on the assemble_variable, because at that point it really should have better
> DECL_INITIAL to choose section right).
>
> Bootstrapped/regtested on {x86_64,i686,powerpc64,powerpc64le}-linux,
> bootstrapped on aarch64-linux (regtest still pending).
And successfully tested also on aarch64-linux.
> Ok for trunk?
>
> 2018-03-21 Jakub Jelinek <ja...@redhat.com>
>
> PR sanitizer/85018
> * dwarf2asm.c (dw2_output_indirect_constant_1): Set
> DECL_INITIAL (decl) to decl at the end.
> * varasm.c (use_blocks_for_decl_p): Revert the 2018-03-20 change,
> adjust the comment.
>
> --- gcc/dwarf2asm.c.jj 2018-03-20 14:00:19.863928902 +0100
> +++ gcc/dwarf2asm.c 2018-03-21 11:52:13.523869188 +0100
> @@ -982,6 +982,7 @@ dw2_output_indirect_constant_1 (const ch
> flag_section_anchors = save_flag_section_anchors;
> flag_sanitize = save_flag_sanitize;
> assemble_integer (sym_ref, POINTER_SIZE_UNITS, POINTER_SIZE, 1);
> + DECL_INITIAL (decl) = decl;
>
> return 0;
> }
> --- gcc/varasm.c.jj 2018-03-20 14:00:20.278928898 +0100
> +++ gcc/varasm.c 2018-03-21 11:51:01.838881084 +0100
> @@ -1253,6 +1253,12 @@ use_blocks_for_decl_p (tree decl)
> if (!VAR_P (decl) && TREE_CODE (decl) != CONST_DECL)
> return false;
>
> + /* DECL_INITIAL (decl) set to decl is a hack used for some decls that
> + are never used from code directly and we never want object block
> handling
> + for those. */
> + if (DECL_INITIAL (decl) == decl)
> + return false;
> +
> /* If this decl is an alias, then we don't want to emit a
> definition. */
> if (VAR_P (decl)
Jakub
