| Message ID | 20110205170424.GC30899@tyan-ft48-01.lab.bos.redhat.com |
|---|---|
| State | New |
| Headers | show |
--- gcc/varasm.c.jj 2011-02-03 20:09:38.000000000 +0100 +++ gcc/varasm.c 2011-02-05 00:01:59.911796344 +0100 @@ -6060,7 +6060,12 @@ default_section_type_flags (tree decl, c flags = SECTION_WRITE; } else - flags = SECTION_WRITE; + { + flags = SECTION_WRITE; + if (strcmp (name, ".data.rel.ro") == 0 + || strcmp (name, ".data.rel.ro.local") == 0) + flags |= SECTION_RELRO; + } if (decl && DECL_ONE_ONLY (decl)) flags |= SECTION_LINKONCE;
Hi! My recent PR31490 patch caused a regression on ppc and pa at least, where gcc started complaining about section type conflicts on .data.rel.ro{,.local} sections. The problem is if get_section is called both from default_elf_select_rtx_section where it is called with .data.rel.ro{,.local} section but NULL decl, and with a decl that belongs into the same section. default_section_type_flags would set SECTION_RELRO only if decl was non-NULL. Fixed by oring in SECTION_RELRO for those two explicit sections too. Bootstrapped/regtested on x86_64-linux and i686-linux, tested with a cross on the pa testcase. Ok for trunk? 2011-02-05 Jakub Jelinek <jakub@redhat.com> PR middle-end/47610 * varasm.c (default_section_type_flags): If decl is NULL, and name is .data.rel.ro or .data.rel.ro.local, set SECTION_RELRO bit. Jakub