PowerPC64 --plt-align

This changes the PowerPC64 --plt-align option to perform the usual
alignment of code as suggested by its name, as well as the previous
behaviour of padding so as to reduce boundary crossing.  The old
behaviour is had by using a negative parameter.

The default is also changed to align plt stub code by default to 32
byte boundaries, the point being to get better bctr branch prediction
on power8 and power9 hardware.

bfd/
	* elf64-ppp.c (plt_stub_pad): Handle positive and negative
	plt_stub_align.
ld/
	* ld.texinfo (--plt-align): Describe new behaviour of option.
	* emultempl/ppc64elf.em (params): Default plt_stub_align to 5.
	* testsuite/ld-powerpc/powerpc.exp: Pass --no-plt-align for
	selected tests.
	* testsuite/ld-powerpc/relbrlt.d: Pass --no-plt-align.
	* testsuite/ld-powerpc/elfv2so.d: Adjust expected output.

8 files changed, 55 insertions, 22 deletions

diff --git a/bfd/ChangeLog b/bfd/ChangeLog
index 06b9278ec9c..852190274ba 100644
--- a/bfd/ChangeLog
+++ b/bfd/ChangeLog
@@ -1,3 +1,8 @@
+2017-09-09  Alan Modra  <amodra@gmail.com>
+
+	* elf64-ppp.c (plt_stub_pad): Handle positive and negative
+	plt_stub_align.
+
 2017-09-09  H.J. Lu  <hongjiu.lu@intel.com>
 
 	* elf32-i386.c (elf_i386_relocate_section): Update usage of
diff --git a/bfd/elf64-ppc.c b/bfd/elf64-ppc.c
index cf7c178c0ff..4821801c689 100644
--- a/bfd/elf64-ppc.c
+++ b/bfd/elf64-ppc.c
@@ -10610,17 +10610,30 @@ plt_stub_size (struct ppc_link_hash_table *htab,
   return size;
 }
 
-/* If this stub would cross fewer 2**plt_stub_align boundaries if we align,
-   then return the padding needed to do so.  */
+/* Depending on the sign of plt_stub_align:
+   If positive, return the padding to align to a 2**plt_stub_align
+   boundary.
+   If negative, if this stub would cross fewer 2**plt_stub_align
+   boundaries if we align, then return the padding needed to do so.  */
+
 static inline unsigned int
 plt_stub_pad (struct ppc_link_hash_table *htab,
 	      struct ppc_stub_hash_entry *stub_entry,
 	      bfd_vma plt_off)
 {
-  int stub_align = 1 << htab->params->plt_stub_align;
+  int stub_align;
   unsigned stub_size = plt_stub_size (htab, stub_entry, plt_off);
   bfd_vma stub_off = stub_entry->group->stub_sec->size;
 
+  if (htab->params->plt_stub_align >= 0)
+    {
+      stub_align = 1 << htab->params->plt_stub_align;
+      if ((stub_off & (stub_align - 1)) != 0)
+	return stub_align - (stub_off & (stub_align - 1));
+      return 0;
+    }
+
+  stub_align = 1 << -htab->params->plt_stub_align;
   if (((stub_off + stub_size - 1) & -stub_align) - (stub_off & -stub_align)
       > ((stub_size - 1) & -stub_align))
     return stub_align - (stub_off & (stub_align - 1));
diff --git a/ld/ChangeLog b/ld/ChangeLog
index 7c178680d3b..1d052f79eeb 100644
--- a/ld/ChangeLog
+++ b/ld/ChangeLog
@@ -1,3 +1,12 @@
+2017-09-09  Alan Modra  <amodra@gmail.com>
+
+	* ld.texinfo (--plt-align): Describe new behaviour of option.
+	* emultempl/ppc64elf.em (params): Default plt_stub_align to 5.
+	* testsuite/ld-powerpc/powerpc.exp: Pass --no-plt-align for
+	selected tests.
+	* testsuite/ld-powerpc/relbrlt.d: Pass --no-plt-align.
+	* testsuite/ld-powerpc/elfv2so.d: Adjust expected output.
+
 2017-09-09  H.J. Lu  <hongjiu.lu@intel.com>
 
 	PR ld/22115
diff --git a/ld/emultempl/ppc64elf.em b/ld/emultempl/ppc64elf.em
index 58cb798cb12..cf75957fa9c 100644
--- a/ld/emultempl/ppc64elf.em
+++ b/ld/emultempl/ppc64elf.em
@@ -37,7 +37,7 @@ static struct ppc64_elf_params params = { NULL,
 					  &ppc_add_stub_section,
 					  &ppc_layout_sections_again,
 					  1, -1, 0,
-					  ${DEFAULT_PLT_STATIC_CHAIN-0}, -1, 0,
+					  ${DEFAULT_PLT_STATIC_CHAIN-0}, -1, 5,
 					  -1, 0, -1, -1, 0};
 
 /* Fake input file for stubs.  */
diff --git a/ld/ld.texinfo b/ld/ld.texinfo
index ba19cd7cb69..90a745accde 100644
--- a/ld/ld.texinfo
+++ b/ld/ld.texinfo
@@ -7603,10 +7603,12 @@ off this feature.
 @item --plt-align
 @itemx --no-plt-align
 Use these options to control whether individual PLT call stubs are
-padded so that they don't cross a 32-byte boundary, or to the
-specified power of two boundary when using @code{--plt-align=}.  Note
-that this isn't alignment in the usual sense.  By default PLT call
-stubs are packed tightly.
+aligned to a 32-byte boundary, or to the specified power of two
+boundary when using @code{--plt-align=}.  A negative value may be
+specified to pad PLT call stubs so that they do not cross the
+specified power of two boundary (or the minimum number of boundaries
+if a PLT stub is so large that it must cross a boundary).  By default
+PLT call stubs are aligned to 32-byte boundaries.
 
 @cindex PowerPC64 PLT call stub static chain
 @kindex --plt-static-chain
diff --git a/ld/testsuite/ld-powerpc/elfv2so.d b/ld/testsuite/ld-powerpc/elfv2so.d
index 2c1fa3234bd..a577b2a7ea8 100644
--- a/ld/testsuite/ld-powerpc/elfv2so.d
+++ b/ld/testsuite/ld-powerpc/elfv2so.d
@@ -12,24 +12,28 @@ Disassembly of section \.text:
 .*:	(e9 82 80 38|38 80 82 e9) 	ld      r12,-32712\(r2\)
 .*:	(7d 89 03 a6|a6 03 89 7d) 	mtctr   r12
 .*:	(4e 80 04 20|20 04 80 4e) 	bctr
+	\.\.\.
 
 .* <.*\.plt_call\.f3>:
 .*:	(f8 41 00 18|18 00 41 f8) 	std     r2,24\(r1\)
 .*:	(e9 82 80 28|28 80 82 e9) 	ld      r12,-32728\(r2\)
 .*:	(7d 89 03 a6|a6 03 89 7d) 	mtctr   r12
 .*:	(4e 80 04 20|20 04 80 4e) 	bctr
+	\.\.\.
 
 .* <.*\.plt_call\.f2>:
 .*:	(f8 41 00 18|18 00 41 f8) 	std     r2,24\(r1\)
 .*:	(e9 82 80 30|30 80 82 e9) 	ld      r12,-32720\(r2\)
 .*:	(7d 89 03 a6|a6 03 89 7d) 	mtctr   r12
 .*:	(4e 80 04 20|20 04 80 4e) 	bctr
+	\.\.\.
 
 .* <.*\.plt_call\.f1>:
 .*:	(f8 41 00 18|18 00 41 f8) 	std     r2,24\(r1\)
 .*:	(e9 82 80 40|40 80 82 e9) 	ld      r12,-32704\(r2\)
 .*:	(7d 89 03 a6|a6 03 89 7d) 	mtctr   r12
 .*:	(4e 80 04 20|20 04 80 4e) 	bctr
+	\.\.\.
 
 .* <f1>:
 .*:	(3c 4c 00 02|02 00 4c 3c) 	addis   r2,r12,2
@@ -37,14 +41,14 @@ Disassembly of section \.text:
 .*:	(7c 08 02 a6|a6 02 08 7c) 	mflr    r0
 .*:	(f8 21 ff e1|e1 ff 21 f8) 	stdu    r1,-32\(r1\)
 .*:	(f8 01 00 30|30 00 01 f8) 	std     r0,48\(r1\)
-.*:	(4b ff ff dd|dd ff ff 4b) 	bl      .*\.plt_call\.f1>
+.*:	(4b ff ff cd|cd ff ff 4b) 	bl      .*\.plt_call\.f1>
 .*:	(e8 62 80 08|08 80 62 e8) 	ld      r3,-32760\(r2\)
-.*:	(4b ff ff c5|c5 ff ff 4b) 	bl      .*\.plt_call\.f2>
+.*:	(4b ff ff a5|a5 ff ff 4b) 	bl      .*\.plt_call\.f2>
 .*:	(e8 41 00 18|18 00 41 e8) 	ld      r2,24\(r1\)
 .*:	(e8 62 80 10|10 80 62 e8) 	ld      r3,-32752\(r2\)
-.*:	(4b ff ff a9|a9 ff ff 4b) 	bl      .*\.plt_call\.f3>
+.*:	(4b ff ff 79|79 ff ff 4b) 	bl      .*\.plt_call\.f3>
 .*:	(e8 41 00 18|18 00 41 e8) 	ld      r2,24\(r1\)
-.*:	(4b ff ff 91|91 ff ff 4b) 	bl      .*\.plt_call\.f4>
+.*:	(4b ff ff 51|51 ff ff 4b) 	bl      .*\.plt_call\.f4>
 .*:	(e8 41 00 18|18 00 41 e8) 	ld      r2,24\(r1\)
 .*:	(e8 01 00 30|30 00 01 e8) 	ld      r0,48\(r1\)
 .*:	(38 21 00 20|20 00 21 38) 	addi    r1,r1,32
diff --git a/ld/testsuite/ld-powerpc/powerpc.exp b/ld/testsuite/ld-powerpc/powerpc.exp
index 0735ab71c2c..ef3b718a2d9 100644
--- a/ld/testsuite/ld-powerpc/powerpc.exp
+++ b/ld/testsuite/ld-powerpc/powerpc.exp
@@ -163,26 +163,26 @@ set ppcelftests {
 }
 
 set ppc64elftests {
-    {"TLS static exec (markers)" "-melf64ppc" ""
+    {"TLS static exec (markers)" "-melf64ppc --no-plt-align" ""
      "-a64 --defsym TLSMARK=1"  {tls.s tlslib.s}
      {{objdump -dr tls.d} {objdump -sj.got tls.g} {objdump -sj.tdata tls.t}}
      "tlsm"}
-    {"TLS static exec" "-melf64ppc" "" "-a64"  {tls.s tlslib.s}
+    {"TLS static exec" "-melf64ppc --no-plt-align" "" "-a64"  {tls.s tlslib.s}
      {{objdump -dr tls.d} {objdump -sj.got tls.g} {objdump -sj.tdata tls.t}}
      "tls"}
     {"TLS helper shared library" "-shared -melf64ppc tmpdir/tlslib.o" "" "" {}
      {} "libtlslib.so"}
     {"TLS helper old shared lib" "-shared -melf64ppc" "" "-a64" {oldtlslib.s}
      {} "liboldlib.so"}
-    {"TLS dynamic exec" "-melf64ppc --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tls.o tmpdir/libtlslib.so" "" "" {}
+    {"TLS dynamic exec" "-melf64ppc --no-plt-align --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tls.o tmpdir/libtlslib.so" "" "" {}
      {{readelf -WSsrl tlsexe.r} {objdump -dr tlsexe.d}
       {objdump -sj.got tlsexe.g} {objdump -sj.tdata tlsexe.t}}
      "tlsexe"}
-    {"TLS dynamic old" "-melf64ppc --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tls.o tmpdir/liboldlib.so" "" "" {}
+    {"TLS dynamic old" "-melf64ppc --no-plt-align --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tls.o tmpdir/liboldlib.so" "" "" {}
      {{readelf -WSsrl tlsexe.r} {objdump -dr tlsexe.d}
       {objdump -sj.got tlsexe.g} {objdump -sj.tdata tlsexe.t}}
      "tlsexeold"}
-    {"TLS shared" "-shared -melf64ppc --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tls.o" "" "" {}
+    {"TLS shared" "-shared -melf64ppc --no-plt-align --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tls.o" "" "" {}
      {{readelf -WSsrl tlsso.r} {objdump -dr tlsso.d}
       {objdump -sj.got tlsso.g} {objdump -sj.tdata tlsso.t}}
      "tls.so"}
@@ -190,17 +190,17 @@ set ppc64elftests {
      {{objdump -dr tlstoc.d} {objdump -sj.got tlstoc.g}
       {objdump -sj.tdata tlstoc.t}}
      "tlstoc"}
-    {"TLSTOC dynamic exec" "-melf64ppc --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tlstoc.o tmpdir/libtlslib.so" ""
+    {"TLSTOC dynamic exec" "-melf64ppc --no-plt-align --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tlstoc.o tmpdir/libtlslib.so" ""
      "" {}
      {{readelf -WSsrl tlsexetoc.r} {objdump -dr tlsexetoc.d}
       {objdump -sj.got tlsexetoc.g} {objdump -sj.tdata tlsexetoc.t}}
      "tlsexetoc"}
-    {"TLSTOC dynamic old" "-melf64ppc --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tlstoc.o tmpdir/liboldlib.so" ""
+    {"TLSTOC dynamic old" "-melf64ppc --no-plt-align --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tlstoc.o tmpdir/liboldlib.so" ""
      "" {}
      {{readelf -WSsrl tlsexetoc.r} {objdump -dr tlsexetoc.d}
       {objdump -sj.got tlsexetoc.g} {objdump -sj.tdata tlsexetoc.t}}
      "tlsexetocold"}
-    {"TLSTOC shared" "-shared -melf64ppc --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tlstoc.o" "" "" {}
+    {"TLSTOC shared" "-shared -melf64ppc --no-plt-align --no-ld-generated-unwind-info --hash-style=sysv tmpdir/tlstoc.o" "" "" {}
      {{readelf -WSsrl tlstocso.r} {objdump -dr tlstocso.d}
       {objdump -sj.got tlstocso.g} {objdump -sj.tdata tlstocso.t}}
      "tlstoc.so"}
@@ -221,7 +221,7 @@ set ppc64elftests {
      "tlsopt4"}
     {"TLS DLL" "-shared -melf64ppc --version-script tlsdll.ver" "" "-a64" {tlsdll.s}
      {} "tlsdll.so"}
-    {"TLS opt 5" "-melf64ppc -shared --gc-sections --no-plt-localentry tmpdir/tlsdll.so" "" "-a64"  {tlsopt5.s}
+    {"TLS opt 5" "-melf64ppc --no-plt-align -shared --gc-sections --no-plt-localentry tmpdir/tlsdll.so" "" "-a64"  {tlsopt5.s}
      {{objdump -dr tlsopt5.d} {readelf -wf tlsopt5.wf}}
      "tlsopt5"}
     {"sym@tocbase" "-shared -melf64ppc" "" "-a64" {symtocbase-1.s symtocbase-2.s}
diff --git a/ld/testsuite/ld-powerpc/relbrlt.d b/ld/testsuite/ld-powerpc/relbrlt.d
index a5f2437d083..a00b1ffd0c5 100644
--- a/ld/testsuite/ld-powerpc/relbrlt.d
+++ b/ld/testsuite/ld-powerpc/relbrlt.d
@@ -1,6 +1,6 @@
 #source: relbrlt.s
 #as: -a64
-#ld: -melf64ppc --no-ld-generated-unwind-info --emit-relocs
+#ld: -melf64ppc --no-plt-align --no-ld-generated-unwind-info --emit-relocs
 #objdump: -Dr
 
 .*