[PATCH 00/22] ELF: correct handling of simplified section directives

[PATCH 00/22] ELF: correct handling of simplified section directives

[Jan Beulich]

Various targets override .bss (many), .text, or .data (a few) handling,
and sadly more often than not these overrides break things: In some
cases .previous doesn't work as intended, while in other cases the
sub-section specifier that ELF allows also for .bss for isn't recognized.
Fix this (first primarily for .bss, then for .text/.data), putting in
place tests to cover these pretty universal ELF aspects.

Speaking of "universal ELF aspects": Three targets (csky, mcore, and
spu) handle .bss much like .lcomm. There surely are reasons for this,
but it feels at best odd.

01: Arm: drop .bss override
02: Arm64: drop .bss override
03: RISC-V: drop .bss override
04: IA64: drop .bss override
05: bfin: drop .bss override
06: m32c: drop .bss override
07: m68k: drop .bss override
08: microblaze: drop/restrict override of .text, .data, and .bss
09: rl78: drop .bss override
10: rx: drop .bss override
11: s390: drop .bss override
12: score: drop .bss override
13: visium: drop .bss and .skip overrides
14: z80: drop .bss override
15: ELF: test certain .bss usages
16: v850: drop .bss override
17: gas: correct .bss documentation for non-ELF
18: d30v: fix .text/.data interaction with .previous
19: hppa/ELF: fix .text/.data interaction with .previous
20: nios2: fix .text/.data interaction with .previous
21: pru: fix .text/.data interaction with .previous
22: ELF: test certain .text/.data usages

Jan

[PATCH 01/22] Arm: drop .bss override

[Jan Beulich]

The comment looks bogus (perhaps simply stale), and there are also no
other precautions against subsections being used on ELF with .bss. It
also doesn't look to be a good idea to override the custom handlers that
ELF and COFF have (afaict doing so further broke .previous on ELF).

--- a/gas/config/tc-arm.c
+++ b/gas/config/tc-arm.c
@@ -3347,20 +3347,7 @@ s_syntax (int unused ATTRIBUTE_UNUSED)
   demand_empty_rest_of_line ();
 }
 
-/* Directives: sectioning and alignment.  */
-
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  /* We don't support putting frags in the BSS segment, we fake it by
-     marking in_bss, then looking at s_skip for clues.	*/
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-
-#ifdef md_elf_section_change_hook
-  md_elf_section_change_hook ();
-#endif
-}
+/* Directives: alignment.  */
 
 static void
 s_even (int ignore ATTRIBUTE_UNUSED)
@@ -5187,7 +5174,6 @@ const pseudo_typeS md_pseudo_table[] =
   { "dn",	   s_dn,          0 },
   { "qn",          s_qn,          0 },
   { "unreq",	   s_unreq,	  0 },
-  { "bss",	   s_bss,	  0 },
   { "align",	   s_align_ptwo,  2 },
   { "arm",	   s_arm,	  0 },
   { "thumb",	   s_thumb,	  0 },
--- a/gas/doc/c-arm.texi
+++ b/gas/doc/c-arm.texi
@@ -870,11 +870,6 @@ incrementally to the architecture being
 This performs the same action as @var{.code 32}.
 
 @c BBBBBBBBBBBBBBBBBBBBBBBBBB
-
-@cindex @code{.bss} directive, ARM
-@item .bss
-This directive switches to the @code{.bss} section.
-
 @c CCCCCCCCCCCCCCCCCCCCCCCCCC
 
 @cindex @code{.cantunwind} directive, ARM

[PATCH 02/22] Arm64: drop .bss override

[Jan Beulich]

The comment looks bogus (perhaps simply stale, perhaps wrongly copied
from Arm in the first place), and there are also no other precautions
against subsections being used on ELF with .bss. It also doesn't look
to be a good idea to override the custom handlers that ELF and COFF
have (afaict doing so further broke .previous on ELF).

As to the mapping state update - such also doesn't appear to be done
for other section switching, so its original purpose was at best
questionable as well.

--- a/gas/config/tc-aarch64.c
+++ b/gas/config/tc-aarch64.c
@@ -1866,17 +1866,7 @@ mapping_state_2 (enum mstate state, int
 #define mapping_state_2(x, y)	/* nothing */
 #endif
 
-/* Directives: sectioning and alignment.  */
-
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  /* We don't support putting frags in the BSS segment, we fake it by
-     marking in_bss, then looking at s_skip for clues.  */
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-  mapping_state (MAP_DATA);
-}
+/* Directives: alignment.  */
 
 static void
 s_even (int ignore ATTRIBUTE_UNUSED)
@@ -2396,7 +2386,6 @@ const pseudo_typeS md_pseudo_table[] = {
   /* Never called because '.req' does not start a line.  */
   {"req", s_req, 0},
   {"unreq", s_unreq, 0},
-  {"bss", s_bss, 0},
   {"even", s_even, 0},
   {"ltorg", s_ltorg, 0},
   {"pool", s_ltorg, 0},
--- a/gas/doc/c-aarch64.texi
+++ b/gas/doc/c-aarch64.texi
@@ -389,11 +389,6 @@ extensions by the @option{-mcpu} command
 incrementally to the architecture being compiled for.
 
 @c BBBBBBBBBBBBBBBBBBBBBBBBBB
-
-@cindex @code{.bss} directive, AArch64
-@item .bss
-This directive switches to the @code{.bss} section.
-
 @c CCCCCCCCCCCCCCCCCCCCCCCCCC
 
 @cindex @code{.cpu} directive, AArch64

[PATCH 03/22] RISC-V: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handler that
ELF has; afaict doing so broke .previous, and a sub-section specifier
wasn't accepted either.

--- a/gas/config/tc-riscv.c
+++ b/gas/config/tc-riscv.c
@@ -4471,15 +4471,6 @@ s_dtprel (int bytes)
   demand_empty_rest_of_line ();
 }
 
-/* Handle the .bss pseudo-op.  */
-
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-}
-
 static void
 riscv_make_nops (char *buf, bfd_vma bytes)
 {
@@ -5202,7 +5193,6 @@ static const pseudo_typeS riscv_pseudo_t
   {"dword", cons, 8},
   {"dtprelword", s_dtprel, 4},
   {"dtpreldword", s_dtprel, 8},
-  {"bss", s_bss, 0},
   {"uleb128", s_riscv_leb128, 0},
   {"sleb128", s_riscv_leb128, 1},
   {"insn", s_riscv_insn, 0},
--- a/gas/doc/c-riscv.texi
+++ b/gas/doc/c-riscv.texi
@@ -138,10 +138,6 @@ Emits a DTP-relative word (or double-wor
 meant to be used by the compiler in shared libraries for DWARF debug info for
 thread local variables.
 
-@cindex BSS directive
-@item .bss
-Sets the current section to the BSS section.
-
 @cindex LEB128 directives
 @item .uleb128 @var{value}
 @itemx .sleb128 @var{value}

[PATCH 04/22] IA64: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handlers that
ELF and COFF have. While in this case interaction with ELF's .previous
wasn't screwed, the sub-section specifier wasn't permitted.

--- a/gas/config/tc-ia64.c
+++ b/gas/config/tc-ia64.c
@@ -68,8 +68,7 @@
 enum special_section
   {
     /* IA-64 ABI section pseudo-ops.  */
-    SPECIAL_SECTION_BSS = 0,
-    SPECIAL_SECTION_SBSS,
+    SPECIAL_SECTION_SBSS = 0,
     SPECIAL_SECTION_SDATA,
     SPECIAL_SECTION_RODATA,
     SPECIAL_SECTION_COMMENT,
@@ -645,7 +644,7 @@ static const bfd_vma nop[IA64_NUM_UNITS]
    habit of setting temporary sentinels.  */
 static char special_section_name[][20] =
   {
-    {".bss"}, {".sbss"}, {".sdata"}, {".rodata"}, {".comment"},
+    {".sbss"}, {".sdata"}, {".rodata"}, {".comment"},
     {".IA_64.unwind"}, {".IA_64.unwind_info"},
     {".init_array"}, {".fini_array"}
   };
@@ -3172,7 +3171,7 @@ dot_loc (int x)
   dwarf2_directive_loc (x);
 }
 
-/* .sbss, .bss etc. are macros that expand into ".section SECNAME".  */
+/* .sbss, .srodata etc. are macros that expand into ".section SECNAME".  */
 static void
 dot_special_section (int which)
 {
@@ -5201,7 +5200,6 @@ const pseudo_typeS md_pseudo_table[] =
     { "radix", dot_radix, 0 },
     { "lcomm", s_lcomm_bytes, 1 },
     { "loc", dot_loc, 0 },
-    { "bss", dot_special_section, SPECIAL_SECTION_BSS },
     { "sbss", dot_special_section, SPECIAL_SECTION_SBSS },
     { "sdata", dot_special_section, SPECIAL_SECTION_SDATA },
     { "rodata", dot_special_section, SPECIAL_SECTION_RODATA },

[PATCH 05/22] bfin: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handler that
ELF has; afaict doing so broke .previous.

--- a/gas/config/tc-bfin.c
+++ b/gas/config/tc-bfin.c
@@ -106,16 +106,6 @@ bfin_pic_ptr (int nbytes)
   demand_empty_rest_of_line ();
 }
 
-static void
-bfin_s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  int temp;
-
-  temp = get_absolute_expression ();
-  subseg_set (bss_section, (subsegT) temp);
-  demand_empty_rest_of_line ();
-}
-
 const pseudo_typeS md_pseudo_table[] = {
   {"align", s_align_bytes, 0},
   {"byte2", cons, 2},
@@ -128,7 +118,6 @@ const pseudo_typeS md_pseudo_table[] = {
   {"p", s_ignore, 0},
   {"pdata", s_ignore, 0},
   {"var", s_ignore, 0},
-  {"bss", bfin_s_bss, 0},
   {0, 0, 0}
 };
 

[PATCH 06/22] m32c: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handler that
ELF has; afaict doing so broke .previous.

--- a/gas/config/tc-m32c.c
+++ b/gas/config/tc-m32c.c
@@ -100,8 +100,6 @@ set_isa (enum isa_attr isa_num)
   cgen_bitset_set (& m32c_isa, isa_num);
 }
 
-static void s_bss (int);
-
 int
 md_parse_option (int c, const char * arg ATTRIBUTE_UNUSED)
 {
@@ -141,20 +139,9 @@ md_show_usage (FILE * stream)
   fprintf (stream, _(" M32C specific command line options:\n"));
 }
 
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  int temp;
-
-  temp = get_absolute_expression ();
-  subseg_set (bss_section, (subsegT) temp);
-  demand_empty_rest_of_line ();
-}
-
 /* The target specific pseudo-ops which we support.  */
 const pseudo_typeS md_pseudo_table[] =
 {
-  { "bss",	s_bss, 		0},
   { "3byte",	cons,		3 },
   { "word",	cons,		4 },
   { NULL, 	NULL, 		0 }

[PATCH 07/22] m68k: drop .bss override

[Jan Beulich]

The comment looks bogus (perhaps simply stale), and there are also no
other precautions against subsections being used on ELF with .bss. It
also doesn't look to be a good idea to override the custom handler that
ELF has (afaict doing so further broke .previous).

--- a/gas/config/tc-m68k.c
+++ b/gas/config/tc-m68k.c
@@ -455,7 +455,6 @@ static int reverse_16_bits (int in);
 static int reverse_8_bits (int in);
 static void install_gen_operand (int mode, int val);
 static void install_operand (int mode, int val);
-static void s_bss (int);
 static void s_data1 (int);
 static void s_data2 (int);
 static void s_even (int);
@@ -862,7 +861,6 @@ const pseudo_typeS md_pseudo_table[] =
 {
   {"data1", s_data1, 0},
   {"data2", s_data2, 0},
-  {"bss", s_bss, 0},
   {"even", s_even, 0},
   {"skip", s_space, 0},
   {"proc", s_proc, 0},
@@ -5493,16 +5491,6 @@ s_data2 (int ignore ATTRIBUTE_UNUSED)
   demand_empty_rest_of_line ();
 }
 
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  /* We don't support putting frags in the BSS segment, we fake it
-     by marking in_bss, then looking at s_skip for clues.  */
-
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-}
-
 static void
 s_even (int ignore ATTRIBUTE_UNUSED)
 {

[PATCH 08/22] microblaze: drop/restrict override of .text, .data, and .bss

[Jan Beulich]

While only ELF is supported right now, (stub) code generally is in place
for the non-ELF case as well. Don't override .bss for ELF - that's
unlikely to be a good idea anyway and prevented the sub-section
specifier from being usable. Don't override .text and .data at all - for
.data and ELF for the same reason, while for .text and ELF obj-elf.c's is
all we need, and for (hypothetical) non-ELF read.c's identical handling
would have been invoked anyway.
---
Wiring .bss to s_data() is somewhat odd, too.

--- a/gas/config/tc-microblaze.c
+++ b/gas/config/tc-microblaze.c
@@ -135,27 +135,6 @@ microblaze_generate_symbol (char *sym)
 
 /* Handle the section changing pseudo-ops. */
 
-static void
-microblaze_s_text (int ignore ATTRIBUTE_UNUSED)
-{
-#ifdef OBJ_ELF
-  obj_elf_text (ignore);
-#else
-  s_text (ignore);
-#endif
-}
-
-static void
-microblaze_s_data (int ignore ATTRIBUTE_UNUSED)
-{
-#ifdef OBJ_ELF
-  obj_elf_change_section (".data", SHT_PROGBITS, SHF_ALLOC+SHF_WRITE,
-			  0, 0, false);
-#else
-  s_data (ignore);
-#endif
-}
-
 /* Things in the .sdata segment are always considered to be in the small data section.  */
 
 static void
@@ -298,20 +277,13 @@ microblaze_s_rdata (int localvar)
 }
 
 static void
-microblaze_s_bss (int localvar)
+microblaze_s_sbss (int ignore ATTRIBUTE_UNUSED)
 {
 #ifdef OBJ_ELF
-  if (localvar == 0) /* bss.  */
-    obj_elf_change_section (".bss", SHT_NOBITS, SHF_ALLOC+SHF_WRITE,
-			    0, 0, false);
-  else if (localvar == 1)
-    {
-      /* sbss.  */
-      obj_elf_change_section (".sbss", SHT_NOBITS, SHF_ALLOC+SHF_WRITE,
-			      0, 0, false);
-      if (sbss_segment == 0)
-	sbss_segment = subseg_new (".sbss", 0);
-    }
+  obj_elf_change_section (".sbss", SHT_NOBITS, SHF_ALLOC+SHF_WRITE,
+			  0, 0, false);
+  if (sbss_segment == 0)
+    sbss_segment = subseg_new (".sbss", 0);
 #else
   s_data (ignore);
 #endif
@@ -385,7 +357,6 @@ microblaze_s_weakext (int ignore ATTRIBU
 const pseudo_typeS md_pseudo_table[] =
 {
   {"lcomm", microblaze_s_lcomm, 1},
-  {"data", microblaze_s_data, 0},
   {"data8", cons, 1},      /* Same as byte.  */
   {"data16", cons, 2},     /* Same as hword.  */
   {"data32", cons, 4},     /* Same as word.  */
@@ -396,9 +367,10 @@ const pseudo_typeS md_pseudo_table[] =
   {"rodata", microblaze_s_rdata, 0},
   {"sdata2", microblaze_s_rdata, 1},
   {"sdata", microblaze_s_sdata, 0},
-  {"bss", microblaze_s_bss, 0},
-  {"sbss", microblaze_s_bss, 1},
-  {"text", microblaze_s_text, 0},
+#ifndef OBJ_ELF
+  {"bss", s_data, 0},
+#endif
+  {"sbss", microblaze_s_sbss, 0},
   {"word", cons, 4},
   {"frame", s_ignore, 0},
   {"mask", s_ignore, 0}, /* Emitted by gcc.  */

[PATCH 09/22] rl78: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handler that
ELF has; afaict doing so broke .previous.

--- a/gas/config/tc-rl78.c
+++ b/gas/config/tc-rl78.c
@@ -375,16 +375,6 @@ md_show_usage (FILE * stream)
 }
 
 static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  int temp;
-
-  temp = get_absolute_expression ();
-  subseg_set (bss_section, (subsegT) temp);
-  demand_empty_rest_of_line ();
-}
-
-static void
 rl78_float_cons (int ignore ATTRIBUTE_UNUSED)
 {
   if (elf_flags & E_FLAG_RL78_64BIT_DOUBLES)
@@ -397,7 +387,6 @@ const pseudo_typeS md_pseudo_table[] =
 {
   /* Our "standard" pseudos.  */
   { "double", rl78_float_cons,	'd' },
-  { "bss",    s_bss, 		0 },
   { "3byte",  cons,		3 },
   { "int",    cons,		4 },
   { "word",   cons,		4 },

[PATCH 10/22] rx: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handler that
ELF has; afaict doing so broke .previous.

--- a/gas/config/tc-rx.c
+++ b/gas/config/tc-rx.c
@@ -222,16 +222,6 @@ md_show_usage (FILE * stream)
 }
 
 static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  int temp;
-
-  temp = get_absolute_expression ();
-  subseg_set (bss_section, (subsegT) temp);
-  demand_empty_rest_of_line ();
-}
-
-static void
 rx_float_cons (int ignore ATTRIBUTE_UNUSED)
 {
   if (elf_flags & E_FLAG_RX_64BIT_DOUBLES)
@@ -650,7 +640,6 @@ const pseudo_typeS md_pseudo_table[] =
 
   /* Our "standard" pseudos. */
   { "double",   rx_float_cons,  0 },
-  { "bss",	s_bss, 		0 },
   { "3byte",	cons,		3 },
   { "int",	cons,		4 },
   { "word",	cons,		4 },

[PATCH 11/22] s390: drop .bss override

[Jan Beulich]

The comment looks bogus (perhaps simply stale), and there are also no
other precautions against subsections being used on ELF with .bss. It
also doesn't look to be a good idea to override the custom handler that
ELF has (afaict doing so further broke .previous).

--- a/gas/config/tc-s390.c
+++ b/gas/config/tc-s390.c
@@ -88,7 +88,6 @@ int s390_cie_data_alignment;
 /* Define the prototypes for the pseudo-ops */
 static void s390_byte (int);
 static void s390_elf_cons (int);
-static void s390_bss (int);
 static void s390_insn (int);
 static void s390_literals (int);
 static void s390_machine (int);
@@ -98,7 +97,6 @@ const pseudo_typeS md_pseudo_table[] =
 {
   { "align",        s_align_bytes,      0 },
   /* Pseudo-ops which must be defined.  */
-  { "bss",          s390_bss,           0 },
   { "insn",         s390_insn,          0 },
   /* Pseudo-ops which must be overridden.  */
   { "byte",	    s390_byte,	        0 },
@@ -1734,16 +1732,6 @@ md_create_long_jump (ptr, from_addr, to_
 }
 #endif
 
-void
-s390_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  /* We don't support putting frags in the BSS segment, we fake it
-     by marking in_bss, then looking at s_skip for clues.  */
-
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-}
-
 /* Pseudo-op handling.  */
 
 void

[PATCH 12/22] score: drop .bss override

[Jan Beulich]

The comment looks bogus (perhaps simply stale, or wrongly copied from
another target). It also doesn't look to be a good idea to override the
custom handler that ELF has (afaict doing so broke .previous as well as
sub-section specification).

While there also fold the identical handlers for .text (there likely is
more room for such folding).

--- a/gas/config/tc-score.c
+++ b/gas/config/tc-score.c
@@ -24,8 +24,6 @@
 
 #include "tc-score7.c"
 
-static void s3_s_score_bss (int ignore ATTRIBUTE_UNUSED);
-static void s3_s_score_text (int ignore);
 static void s3_score_s_section (int ignore);
 static void s3_s_change_sec (int sec);
 static void s3_s_score_mask (int reg_type ATTRIBUTE_UNUSED);
@@ -39,7 +37,6 @@ static void s3_s_score_gpword (int ignor
 static void s3_s_score_cpadd (int ignore ATTRIBUTE_UNUSED);
 static void s3_s_score_lcomm (int bytes_p);
 
-static void s_score_bss (int ignore ATTRIBUTE_UNUSED);
 static void s_score_text (int ignore);
 static void s_section (int ignore);
 static void s_change_sec (int sec);
@@ -196,7 +193,6 @@ symbolS *GOT_symbol;
 
 const pseudo_typeS md_pseudo_table[] =
 {
-  {"bss", s_score_bss, 0},
   {"text", s_score_text, 0},
   {"word", cons, 4},
   {"long", cons, 4},
@@ -5532,22 +5528,6 @@ s3_do_dsp3 (char *str)
     s3_inst.relax_inst = 0x8000;
 }
 
-
-/* If we change section we must dump the literal pool first.  */
-static void
-s3_s_score_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  subseg_set (bss_section, (subsegT) get_absolute_expression ());
-  demand_empty_rest_of_line ();
-}
-
-static void
-s3_s_score_text (int ignore)
-{
-  obj_elf_text (ignore);
-  record_alignment (now_seg, 2);
-}
-
 static void
 s3_score_s_section (int ignore)
 {
@@ -6324,21 +6304,10 @@ s3_build_dependency_insn_hsh (void)
 }
 
 static void
-s_score_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  if (score3)
-    return s3_s_score_bss (ignore);
-  else
-    return s7_s_score_bss (ignore);
-}
-
-static void
 s_score_text (int ignore)
 {
-  if (score3)
-    return s3_s_score_text (ignore);
-  else
-    return s7_s_score_text (ignore);
+  obj_elf_text (ignore);
+  record_alignment (now_seg, 2);
 }
 
 static void
--- a/gas/config/tc-score7.c
+++ b/gas/config/tc-score7.c
@@ -5367,21 +5367,6 @@ s7_build_reg_hsh (struct s7_reg_map *map
 
 \f
 
-/* If we change section we must dump the literal pool first.  */
-static void
-s7_s_score_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  subseg_set (bss_section, (subsegT) get_absolute_expression ());
-  demand_empty_rest_of_line ();
-}
-
-static void
-s7_s_score_text (int ignore)
-{
-  obj_elf_text (ignore);
-  record_alignment (now_seg, 2);
-}
-
 static void
 s7_s_section (int ignore)
 {

[PATCH 13/22] visium: drop .bss and .skip overrides

[Jan Beulich]

The comment in s_bss() looks bogus (perhaps simply stale, or wrongly
copied from another target). It also doesn't look to be a good idea to
override the custom handler that ELF has (afaict doing so broke
.previous as well as sub-section specification).

The override for .skip is simply pointless, for read.c having exactly
the same.

While there also drop two adjacent redundant (with read.h) declarations
(which would be outright dangerous if read.h wasn't included anyway).

--- a/gas/config/tc-visium.c
+++ b/gas/config/tc-visium.c
@@ -163,40 +163,19 @@ tc_gen_reloc (asection *section ATTRIBUT
   return reloc;
 }
 
-extern char *input_line_pointer;
-
-
-static void s_bss (int);
 static void visium_rdata (int);
 
 static void visium_update_parity_bit (char *);
 static char *parse_exp (char *, expressionS *);
 
-/* These are the back-ends for the various machine dependent pseudo-ops.  */
-void demand_empty_rest_of_line (void);
-
-
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  /* We don't support putting frags in the BSS segment, we fake it
-     by marking in_bss, then looking at s_skip for clues.  */
-
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-}
-
-
 /* This table describes all the machine specific pseudo-ops the assembler
-   has to support. The fields are:
+   has to support, and that aren't handled elsewhere. The fields are:
 
    1: Pseudo-op name without dot.
    2: Function to call to execute this pseudo-op.
    3: Integer arg to pass to the function.  */
 const pseudo_typeS md_pseudo_table[] =
 {
-  {"bss", s_bss, 0},
-  {"skip", s_space, 0},
   {"align", s_align_bytes, 0},
   {"noopt", s_ignore, 0},
   {"optim", s_ignore, 0},

[PATCH 14/22] z80: drop .bss override

[Jan Beulich]

It doesn't look to be a good idea to override the custom handlers that
ELF and COFF have; afaict doing so broke .previous on ELF, and a sub-
section specifier wasn't accepted either.

--- a/gas/config/tc-z80.c
+++ b/gas/config/tc-z80.c
@@ -3483,15 +3483,6 @@ area (int arg)
     psect (arg);
 }
 
-/* Handle the .bss pseudo-op.  */
-
-static void
-s_bss (int ignore ATTRIBUTE_UNUSED)
-{
-  subseg_set (bss_section, 0);
-  demand_empty_rest_of_line ();
-}
-
 /* Port specific pseudo ops.  */
 const pseudo_typeS md_pseudo_table[] =
 {
@@ -3507,7 +3498,6 @@ const pseudo_typeS md_pseudo_table[] =
   { ".hd64", set_inss, INS_Z180},
   { ".z80", set_inss, INS_Z80},
   { ".z80n", set_inss, INS_Z80N},
-  { "bss", s_bss, 0},
   { "db" , emit_data, 1},
   { "d24", z80_cons, 3},
   { "d32", z80_cons, 4},

[PATCH 15/22] ELF: test certain .bss usages

[Jan Beulich]

Various targets have / had overrides for .bss. Make sure that in such
cases
- .previous still works correctly (requiring such targets to invoke
  obj_elf_section_change_hook() from their overriding handlers),
- sub-section specifiers are accepted as far as feasible (mandated by
  the doc).

--- /dev/null
+++ b/gas/testsuite/gas/elf/bss-prev.d
@@ -0,0 +1,13 @@
+#name: .bss / .previous interaction
+#as: --no-pad-sections
+#readelf: -S --wide
+
+There are [0-9]+ section headers, starting at offset 0x[0-9a-f]+:
+
+Section Headers:
+ +\[Nr\] Name +Type +Addr(ess|) +Off +Size .*
+#...
+ *\[ [1-9]\] *\.text +PROGBITS +0*0 +0[0-9a-f]* 0+ .*
+ *\[ [1-9]\] *\.data +PROGBITS +0*0 +0[0-9a-f]* 0*1 .*
+ *\[ [1-9]\] *\.bss +NOBITS +0*0 +0[0-9a-f]* 0*1 .*
+#pass
--- /dev/null
+++ b/gas/testsuite/gas/elf/bss-prev.s
@@ -0,0 +1,6 @@
+	.text
+	.data
+	.bss
+	.byte 0
+	.previous
+	.byte 1
--- /dev/null
+++ b/gas/testsuite/gas/elf/bss-subsect.d
@@ -0,0 +1,11 @@
+#name: .bss <subsection>
+#as: --no-pad-sections
+#readelf: -S --wide
+
+There are [0-9]+ section headers, starting at offset 0x[0-9a-f]+:
+
+Section Headers:
+ +\[Nr\] Name +Type +Addr(ess|) +Off +Size .*
+#...
+ *\[ [1-9]\] *\.bss +NOBITS +0*0 +0[0-9a-f]* 0*a .*
+#pass
--- /dev/null
+++ b/gas/testsuite/gas/elf/bss-subsect.s
@@ -0,0 +1,10 @@
+	.bss
+	.byte 0
+
+	.bss 2
+	.balign 2
+	.skip 2
+
+	.bss 1
+	.balign 4
+	.skip 4
--- a/gas/testsuite/gas/elf/elf.exp
+++ b/gas/testsuite/gas/elf/elf.exp
@@ -336,6 +336,13 @@ if { [is_elf_format] } then {
 
     run_dump_test "pr25917"
     run_dump_test "bss"
+    # Some targets treat .bss similar to .lcomm.
+    if { ![istarget "csky-*-*"]
+	 && ![istarget "mcore-*-*"]
+	 && ![istarget "spu-*-*"] } then {
+	run_dump_test "bss-prev" $dump_opts
+	run_dump_test "bss-subsect" $dump_opts
+    }
     run_dump_test "bad-bss"
     run_dump_test "bad-section-flag"
     run_dump_test "bad-size"

[PATCH 16/22] gas: correct .bss documentation for non-ELF

[Jan Beulich]

Only ELF permits the specification of a subsection, and even there not
consistently: csky, mcore, and spu handle .bss similar to .lcomm.
---
Except for H8 this doesn't look to have any effect, because all.texi
blindly enables various object formats. Yet still the documentation ends
up wrong for non-ELF targets.
---
v1: Mention exceptions.

--- a/gas/doc/as.texi
+++ b/gas/doc/as.texi
@@ -4456,7 +4456,12 @@ Some machine configurations provide addi
 * Asciz::                       @code{.asciz "@var{string}"}@dots{}
 * Attach_to_group::             @code{.attach_to_group @var{name}}
 * Balign::                      @code{.balign [@var{abs-expr}[, @var{abs-expr}]]}
+@ifset ELF
 * Bss::                         @code{.bss @var{subsection}}
+@end ifset
+@ifclear ELF
+* Bss::                         @code{.bss}
+@end ifclear
 * Bundle directives::           @code{.bundle_align_mode @var{abs-expr}}, etc
 * Byte::                        @code{.byte @var{expressions}}
 * CFI directives::		@code{.cfi_startproc [simple]}, @code{.cfi_endproc}, etc.
@@ -4801,14 +4806,20 @@ filled in with the value 0x368d (the exa
 the endianness of the processor).  If it skips 1 or 3 bytes, the fill value is
 undefined.
 
+@ifset ELF
 @node Bss
 @section @code{.bss @var{subsection}}
+@end ifset
+@ifclear ELF
+@node Bss
+@section @code{.bss}
+@end ifclear
 @cindex @code{bss} directive
 
 @code{.bss} tells @command{@value{AS}} to assemble the following statements
 onto the end of the bss section.
 @ifset ELF
-For ELF based targets an optional @var{subsection} expression (which must
+For most ELF based targets an optional @var{subsection} expression (which must
 evaluate to a positive integer) can be provided.  In this case the statements
 are appended to the end of the indicated bss subsection.
 @end ifset

[PATCH 17/22] v850: drop .bss override

[Jan Beulich]

While there doesn't look to be anything wrong with this override,
there's also no apparent reason why this override would be needed. Drop
it, reducing overall size a tiny bit.

--- a/gas/config/tc-v850.c
+++ b/gas/config/tc-v850.c
@@ -198,8 +198,6 @@ struct v850_seg_entry v850_seg_table[] =
   { NULL, ".call_table_text",
     SEC_ALLOC | SEC_LOAD | SEC_RELOC | SEC_READONLY | SEC_CODE
     | SEC_HAS_CONTENTS},
-  { NULL, ".bss",
-    SEC_ALLOC }
 };
 
 #define SDATA_SECTION		0
@@ -215,7 +213,6 @@ struct v850_seg_entry v850_seg_table[] =
 #define ZCOMMON_SECTION		10
 #define CALL_TABLE_DATA_SECTION	11
 #define CALL_TABLE_TEXT_SECTION	12
-#define BSS_SECTION		13
 
 static void
 do_v850_seg (int i, subsegT sub)
@@ -578,7 +575,6 @@ const pseudo_typeS md_pseudo_table[] =
   { "zbss",		v850_seg,		ZBSS_SECTION		},
   { "rosdata",		v850_seg,		ROSDATA_SECTION 	},
   { "rozdata",		v850_seg,		ROZDATA_SECTION 	},
-  { "bss",		v850_seg,		BSS_SECTION		},
   { "offset",		v850_offset,		0			},
   { "word",		cons,			4			},
   { "zcomm",		v850_comm,		ZCOMMON_SECTION 	},
@@ -1969,7 +1965,6 @@ md_begin (void)
       op++;
     }
 
-  v850_seg_table[BSS_SECTION].s = bss_section;
   bfd_set_arch_mach (stdoutput, v850_target_arch, machine);
   bfd_set_private_flags (stdoutput, v850_e_flags);
 }

[PATCH 18/22] d30v: fix .text/.data interaction with .previous

[Jan Beulich]

Just like obj_elf_section() is called for .section, obj_elf_{text,data}()
need calling for .text/.data.

--- a/gas/config/tc-d30v.c
+++ b/gas/config/tc-d30v.c
@@ -2055,7 +2055,7 @@ static void
 s_d30v_text (int i)
 
 {
-  s_text (i);
+  obj_elf_text (i);
   d30v_last_label = NULL;
   d30v_current_align = 0;
   d30v_current_align_seg = now_seg;
@@ -2067,7 +2067,7 @@ s_d30v_text (int i)
 static void
 s_d30v_data (int i)
 {
-  s_data (i);
+  obj_elf_data (i);
   d30v_last_label = NULL;
   d30v_current_align = 0;
   d30v_current_align_seg = now_seg;

[PATCH 19/22] hppa/ELF: fix .text/.data interaction with .previous

[Jan Beulich]

For some ELF targets .text/.data are overridden. In that case
obj_elf_{text,data}() need calling, just like .code vectors to that
function for the remaining ELF targets.

While there also hand on the function arguments, even if right now
they're meaningless. This matches what other targets' code does.

--- a/gas/config/tc-hppa.c
+++ b/gas/config/tc-hppa.c
@@ -6199,7 +6199,7 @@ pa_callinfo (int unused ATTRIBUTE_UNUSED
    label when finished.  */
 
 static void
-pa_text (int unused ATTRIBUTE_UNUSED)
+pa_text (int arg)
 {
 #ifdef OBJ_SOM
   current_space = is_defined_space ("$TEXT$");
@@ -6207,21 +6207,32 @@ pa_text (int unused ATTRIBUTE_UNUSED)
     = pa_subsegment_to_subspace (current_space->sd_seg, 0);
 #endif
 
-  s_text (0);
+#ifdef OBJ_ELF
+  obj_elf_text (arg);
+#else
+  s_text (arg);
+#endif
+
   pa_undefine_label ();
 }
 
 /* Switch to the data space.  As usual delete our label.  */
 
 static void
-pa_data (int unused ATTRIBUTE_UNUSED)
+pa_data (int arg)
 {
 #ifdef OBJ_SOM
   current_space = is_defined_space ("$PRIVATE$");
   current_subspace
     = pa_subsegment_to_subspace (current_space->sd_seg, 0);
 #endif
-  s_data (0);
+
+#ifdef OBJ_ELF
+  obj_elf_data (arg);
+#else
+  s_data (arg);
+#endif
+
   pa_undefine_label ();
 }
 

[PATCH 20/22] nios2: fix .text/.data interaction with .previous

[Jan Beulich]

Just like obj_elf_section() is called for .section, obj_elf_{text,data}()
need calling for .text/.data.

--- a/gas/config/tc-nios2.c
+++ b/gas/config/tc-nios2.c
@@ -507,7 +507,7 @@ s_nios2_align (int ignore ATTRIBUTE_UNUS
 static void
 s_nios2_text (int i)
 {
-  s_text (i);
+  obj_elf_text (i);
   nios2_last_label = NULL;
   nios2_current_align = 0;
   nios2_current_align_seg = now_seg;
@@ -518,7 +518,7 @@ s_nios2_text (int i)
 static void
 s_nios2_data (int i)
 {
-  s_data (i);
+  obj_elf_data (i);
   nios2_last_label = NULL;
   nios2_current_align = 0;
   nios2_current_align_seg = now_seg;

[PATCH 21/22] pru: fix .text/.data interaction with .previous

[Jan Beulich]

Just like obj_elf_section() is called for .section, obj_elf_{text,data}()
need calling for .text/.data.

--- a/gas/config/tc-pru.c
+++ b/gas/config/tc-pru.c
@@ -368,7 +368,7 @@ s_pru_align (int ignore ATTRIBUTE_UNUSED
 static void
 s_pru_text (int i)
 {
-  s_text (i);
+  obj_elf_text (i);
   pru_last_label = NULL;
   pru_current_align = 0;
   pru_current_align_seg = now_seg;
@@ -379,7 +379,7 @@ s_pru_text (int i)
 static void
 s_pru_data (int i)
 {
-  s_data (i);
+  obj_elf_data (i);
   pru_last_label = NULL;
   pru_current_align = 0;
   pru_current_align_seg = now_seg;

[PATCH 22/22] ELF: test certain .text/.data usages

[Jan Beulich]

Various targets have / had overrides for .text and/or .data. Make sure
that in such cases sub-section specifiers are accepted, as mandated by
the doc.

--- /dev/null
+++ b/gas/testsuite/gas/elf/data-prev.d
@@ -0,0 +1,14 @@
+#name: .data / .previous interaction
+#as: --no-pad-sections
+#readelf: -S --wide
+
+There are [0-9]+ section headers, starting at offset 0x[0-9a-f]+:
+
+Section Headers:
+ +\[Nr\] Name +Type +Addr(ess|) +Off +Size .*
+#...
+ *\[ [1-9]\] *\.text +PROGBITS +0*0 +0[0-9a-f]* 0* .*
+ *\[ [1-9]\] *\.data +PROGBITS +0*0 +0[0-9a-f]* 0*1 .*
+#...
+ *\[ [1-9]\] *\.aux +PROGBITS +0*0 +0[0-9a-f]* 0*1 .*
+#pass
--- /dev/null
+++ b/gas/testsuite/gas/elf/data-prev.s
@@ -0,0 +1,6 @@
+	.text
+	.section .aux, "a", %progbits
+	.data
+	.byte 0
+	.previous
+	.byte 1
--- /dev/null
+++ b/gas/testsuite/gas/elf/data-subsect.d
@@ -0,0 +1,11 @@
+#name: .data <subsection>
+#as: --no-pad-sections
+#readelf: -S --wide
+
+There are [0-9]+ section headers, starting at offset 0x[0-9a-f]+:
+
+Section Headers:
+ +\[Nr\] Name +Type +Addr(ess|) +Off +Size .*
+#...
+ *\[ [1-9]\] *\.data +PROGBITS +0*0 +0[0-9a-f]* 0*a .*
+#pass
--- /dev/null
+++ b/gas/testsuite/gas/elf/data-subsect.s
@@ -0,0 +1,14 @@
+	.data
+	.byte 1
+
+	.data 2
+	.balign 2
+	.byte 3
+	.byte 3
+
+	.data 1
+	.balign 4
+	.byte 2
+	.byte 2
+	.byte 2
+	.byte 2
--- a/gas/testsuite/gas/elf/elf.exp
+++ b/gas/testsuite/gas/elf/elf.exp
@@ -194,6 +194,10 @@ if { [is_elf_format] } then {
 	}
     }
     run_dump_test "pseudo"
+    run_dump_test "text-prev" $dump_opts
+    run_dump_test "text-subsect" $dump_opts
+    run_dump_test "data-prev" $dump_opts
+    run_dump_test "data-subsect" $dump_opts
     run_dump_test "section0"
     run_dump_test "section1"
     # The h8300 port issues a warning message for
--- /dev/null
+++ b/gas/testsuite/gas/elf/text-prev.d
@@ -0,0 +1,14 @@
+#name: .text / .previous interaction
+#as: --no-pad-sections
+#readelf: -S --wide
+
+There are [0-9]+ section headers, starting at offset 0x[0-9a-f]+:
+
+Section Headers:
+ +\[Nr\] Name +Type +Addr(ess|) +Off +Size .*
+#...
+ *\[ [1-9]\] *\.text +PROGBITS +0*0 +0[0-9a-f]* 0*([1248]|10) .*
+ *\[ [1-9]\] *\.data +PROGBITS +0*0 +0[0-9a-f]* 0* .*
+#...
+ *\[ [1-9]\] *\.aux +PROGBITS +0*0 +0[0-9a-f]* 0*1 .*
+#pass
--- /dev/null
+++ b/gas/testsuite/gas/elf/text-prev.s
@@ -0,0 +1,6 @@
+	.data
+	.section .aux, "a", %progbits
+	.text
+	.nop
+	.previous
+	.byte 0
--- /dev/null
+++ b/gas/testsuite/gas/elf/text-subsect.d
@@ -0,0 +1,11 @@
+#name: .text <subsection>
+#as: --no-pad-sections
+#readelf: -S --wide
+
+There are [0-9]+ section headers, starting at offset 0x[0-9a-f]+:
+
+Section Headers:
+ +\[Nr\] Name +Type +Addr(ess|) +Off +Size .*
+#...
+ *\[ [1-9]\] *\.text +PROGBITS +0*0 +0[0-9a-f]* 0*(6[1248]|70) .*
+#pass
--- /dev/null
+++ b/gas/testsuite/gas/elf/text-subsect.s
@@ -0,0 +1,10 @@
+	.text
+	.nop
+
+	.text 2
+	.balign 32
+	.nop
+
+	.text 1
+	.balign 64
+	.nop

Re: [PATCH 19/22] hppa/ELF: fix .text/.data interaction with .previous

[John David Anglin]

Tested on hppa-linux.  No regressions were observed.

Dave

On 2023-12-15 7:19 a.m., Jan Beulich wrote:
> For some ELF targets .text/.data are overridden. In that case
> obj_elf_{text,data}() need calling, just like .code vectors to that
> function for the remaining ELF targets.
>
> While there also hand on the function arguments, even if right now
> they're meaningless. This matches what other targets' code does.
>
> --- a/gas/config/tc-hppa.c
> +++ b/gas/config/tc-hppa.c
> @@ -6199,7 +6199,7 @@ pa_callinfo (int unused ATTRIBUTE_UNUSED
>      label when finished.  */
>   
>   static void
> -pa_text (int unused ATTRIBUTE_UNUSED)
> +pa_text (int arg)
>   {
>   #ifdef OBJ_SOM
>     current_space = is_defined_space ("$TEXT$");
> @@ -6207,21 +6207,32 @@ pa_text (int unused ATTRIBUTE_UNUSED)
>       = pa_subsegment_to_subspace (current_space->sd_seg, 0);
>   #endif
>   
> -  s_text (0);
> +#ifdef OBJ_ELF
> +  obj_elf_text (arg);
> +#else
> +  s_text (arg);
> +#endif
> +
>     pa_undefine_label ();
>   }
>   
>   /* Switch to the data space.  As usual delete our label.  */
>   
>   static void
> -pa_data (int unused ATTRIBUTE_UNUSED)
> +pa_data (int arg)
>   {
>   #ifdef OBJ_SOM
>     current_space = is_defined_space ("$PRIVATE$");
>     current_subspace
>       = pa_subsegment_to_subspace (current_space->sd_seg, 0);
>   #endif
> -  s_data (0);
> +
> +#ifdef OBJ_ELF
> +  obj_elf_data (arg);
> +#else
> +  s_data (arg);
> +#endif
> +
>     pa_undefine_label ();
>   }
>   
>


-- 
John David Anglin  dave.anglin@bell.net

Re: [PATCH 20/22] nios2: fix .text/.data interaction with .previous

[Sandra Loosemore]

On 12/15/23 05:19, Jan Beulich wrote:
> Just like obj_elf_section() is called for .section, obj_elf_{text,data}()
> need calling for .text/.data.

Looks fine to me.

-Sandra

Re: [PATCH 03/22] RISC-V: drop .bss override

[Nelson Chu]

[-- Attachment #1: Type: text/plain, Size: 1519 bytes --]

I passed the regressions of riscv-gnu-toolchain, so if no one objects, then
it seems safe to commit.

Thanks
Nelson


On Fri, Dec 15, 2023 at 8:07 PM Jan Beulich <jbeulich@suse.com> wrote:

> It doesn't look to be a good idea to override the custom handler that
> ELF has; afaict doing so broke .previous, and a sub-section specifier
> wasn't accepted either.
>
> --- a/gas/config/tc-riscv.c
> +++ b/gas/config/tc-riscv.c
> @@ -4471,15 +4471,6 @@ s_dtprel (int bytes)
>    demand_empty_rest_of_line ();
>  }
>
> -/* Handle the .bss pseudo-op.  */
> -
> -static void
> -s_bss (int ignore ATTRIBUTE_UNUSED)
> -{
> -  subseg_set (bss_section, 0);
> -  demand_empty_rest_of_line ();
> -}
> -
>  static void
>  riscv_make_nops (char *buf, bfd_vma bytes)
>  {
> @@ -5202,7 +5193,6 @@ static const pseudo_typeS riscv_pseudo_t
>    {"dword", cons, 8},
>    {"dtprelword", s_dtprel, 4},
>    {"dtpreldword", s_dtprel, 8},
> -  {"bss", s_bss, 0},
>    {"uleb128", s_riscv_leb128, 0},
>    {"sleb128", s_riscv_leb128, 1},
>    {"insn", s_riscv_insn, 0},
> --- a/gas/doc/c-riscv.texi
> +++ b/gas/doc/c-riscv.texi
> @@ -138,10 +138,6 @@ Emits a DTP-relative word (or double-wor
>  meant to be used by the compiler in shared libraries for DWARF debug info
> for
>  thread local variables.
>
> -@cindex BSS directive
> -@item .bss
> -Sets the current section to the BSS section.
> -
>  @cindex LEB128 directives
>  @item .uleb128 @var{value}
>  @itemx .sleb128 @var{value}
>
>

Re: [PATCH 00/22] ELF: correct handling of simplified section directives

[Alan Modra]

On Fri, Dec 15, 2023 at 12:59:39PM +0100, Jan Beulich wrote:
> Various targets override .bss (many), .text, or .data (a few) handling,
> and sadly more often than not these overrides break things: In some
> cases .previous doesn't work as intended, while in other cases the
> sub-section specifier that ELF allows also for .bss for isn't recognized.
> Fix this (first primarily for .bss, then for .text/.data), putting in
> place tests to cover these pretty universal ELF aspects.

Looks good to me.

-- 
Alan Modra
Australia Development Lab, IBM