[Bug middle-end/68414] New: gcc doesn't emit .skip for .vtable_map_vars sections

[ro at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68414

            Bug ID: 68414
           Summary: gcc doesn't emit .skip for .vtable_map_vars sections
           Product: gcc
           Version: 6.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: middle-end
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ro at gcc dot gnu.org
                CC: ctice at gcc dot gnu.org
  Target Milestone: ---
              Host: *-*-solaris2.*
            Target: *-*-solaris2.*
             Build: *-*-solaris2.*

Created attachment 36760
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=36760&action=edit
minimal patch, untested

While doing the libvtv port to Solaris, all testcases were failing on
Solaris/SPARC
with Solaris as.  It turned out that as doesn't set sh_addralign in the section
header for .vtable_map_vars.  This is not a problem with gas in this case, but
for a regular .bss section gcc emits .skip 1 which avoids this.

Consider the following:

$ cat align.bss.c
#define VTV_PAGE_SIZE 4096

char _vtable_map_vars_start []
__attribute__ ((aligned(VTV_PAGE_SIZE)))
  = { };
$ gcc -S align.bss.c
$ cat align.vmv.c
#define VTV_PAGE_SIZE 4096

char _vtable_map_vars_start []
__attribute__ ((aligned(VTV_PAGE_SIZE),section(".vtable_map_vars")))
  = { };
$ gcc -S align.vmv.c
$ diff -u align.{bss,vmv}.s
--- align.bss.s 2015-11-18 16:20:37.628513468 +0100
+++ align.vmv.s 2015-11-18 16:21:43.411691226 +0100
@@ -1,9 +1,9 @@
-       .file   "align.bss.c"
+       .file   "align.vmv.c"
        .global _vtable_map_vars_start
-       .section        ".bss"
+       .section       
".vtable_map_vars%_vtable_map_vars_start",#alloc,#write,#progbits
+       .group 
_vtable_map_vars_start,".vtable_map_vars%_vtable_map_vars_start",#comdat
        .align 4096
        .type   _vtable_map_vars_start, #object
        .size   _vtable_map_vars_start, 0
 _vtable_map_vars_start:
-       .skip 1
        .ident  "GCC: (GNU) 5.1.0"

I think this needs to be fixed in varasm.c (assemble_variable) in the special
case for .vtable_map_vars, with something like the attached patch.  I'm just
not sure which conditions are necessary to avoid emitting .skip in
unappropriate
cases.

  Rainer