diff mbox series

gcc7/gcc6: Fix unaligned STRD issue on ARM

Message ID 20171107233438.33636-1-raj.khem@gmail.com
State Superseded
Headers show
Series gcc7/gcc6: Fix unaligned STRD issue on ARM | expand

Commit Message

Khem Raj Nov. 7, 2017, 11:34 p.m. UTC
Backport
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=82445

Fixes [YOCTO 12297]

Signed-off-by: Khem Raj <raj.khem@gmail.com>

---
 meta/recipes-devtools/gcc/gcc-6.4.inc              |   1 +
 ...5-suppress-32-bit-aligned-ldrd-strd-peeph.patch | 194 +++++++++++++++++++++
 meta/recipes-devtools/gcc/gcc-7.2.inc              |   1 +
 ...5-suppress-32-bit-aligned-ldrd-strd-peeph.patch | 194 +++++++++++++++++++++
 4 files changed, 390 insertions(+)
 create mode 100644 meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch
 create mode 100644 meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch

-- 
2.15.0

-- 
_______________________________________________
Openembedded-core mailing list
Openembedded-core@lists.openembedded.org
http://lists.openembedded.org/mailman/listinfo/openembedded-core
diff mbox series

Patch

diff --git a/meta/recipes-devtools/gcc/gcc-6.4.inc b/meta/recipes-devtools/gcc/gcc-6.4.inc
index a42b7d83ca..37e996afb7 100644
--- a/meta/recipes-devtools/gcc/gcc-6.4.inc
+++ b/meta/recipes-devtools/gcc/gcc-6.4.inc
@@ -80,6 +80,7 @@  SRC_URI = "\
 "
 BACKPORTS = "\
            file://CVE-2016-6131.patch \
+           file://0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch \
 "
 SRC_URI[md5sum] = "11ba51a0cfb8471927f387c8895fe232"
 SRC_URI[sha256sum] = "850bf21eafdfe5cd5f6827148184c08c4a0852a37ccf36ce69855334d2c914d4"
diff --git a/meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch b/meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch
new file mode 100644
index 0000000000..0214ab83d9
--- /dev/null
+++ b/meta/recipes-devtools/gcc/gcc-6.4/0057-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch
@@ -0,0 +1,194 @@ 
+From ad5bf450aef2ffee6d57ed193fabc5f72f8eaa65 Mon Sep 17 00:00:00 2001
+From: rearnsha <rearnsha@138bc75d-0d04-0410-961f-82ee72b054a4>
+Date: Thu, 19 Oct 2017 13:16:42 +0000
+Subject: [PATCH] [ARM] PR 82445 - suppress 32-bit aligned ldrd/strd peepholing
+ with -mno-unaligned-access
+
+Peephole patterns exist in the arm backend to spot load/store
+operations to adjacent memory operations in order to convert them into
+ldrd/strd instructions.  However, when we have strict alignment
+enforced, then we can only do this if the accesses are known to be
+64-bit aligned; this is unlikely to be the case for most loads.  The
+patch adds some alignment checking to the code that validates the
+addresses for use in the peephole patterns.  This should also fix
+incorrect generation of ldrd/strd with unaligned accesses that could
+previously have occurred on ARMv5e where all such operations must be
+64-bit aligned.
+
+I've added some new tests as well.  In doing so I discovered that the
+ldrd/strd peephole tests could never fail since they would match the
+source file name in the scanned assembly as well as any instructions
+of the intended type.  I've fixed those by tightening the scan results
+slightly.
+
+gcc:
+
+* config/arm/arm.c (align_ok_ldrd_strd): New function.
+(mem_ok_for_ldrd_strd): New parameter align.  Extract the alignment of the
+mem into it.
+(gen_operands_ldrd_strd): Validate the alignment of the accesses.
+
+testsuite:
+
+* gcc.target/arm/peep-ldrd-1.c: Tighten test scan pattern.
+* gcc.target/arm/peep-strd-1.c: Likewise.
+* gcc.target/arm/peep-ldrd-2.c: New test.
+* gcc.target/arm/peep-strd-2.c: New test.
+
+
+
+git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gcc-6-branch@253892 138bc75d-0d04-0410-961f-82ee72b054a4
+---
+Upstream-Status: Backport
+Signed-off-by: Khem Raj <raj.khem@gmail.com>
+
+ gcc/ChangeLog                                      |  8 +++++++
+ gcc/config/arm/arm.c                               | 27 ++++++++++++++++++----
+ gcc/testsuite/ChangeLog                            |  8 +++++++
+ gcc/testsuite/gcc.target/arm/peep-ldrd-1.c         |  2 +-
+ .../arm/{peep-ldrd-1.c => peep-ldrd-2.c}           |  4 ++--
+ gcc/testsuite/gcc.target/arm/peep-strd-1.c         |  2 +-
+ .../arm/{peep-strd-1.c => peep-strd-2.c}           |  4 ++--
+ 7 files changed, 44 insertions(+), 11 deletions(-)
+ copy gcc/testsuite/gcc.target/arm/{peep-ldrd-1.c => peep-ldrd-2.c} (63%)
+ copy gcc/testsuite/gcc.target/arm/{peep-strd-1.c => peep-strd-2.c} (58%)
+
+diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
+index 9c0813d598d..e3da9f77fb6 100644
+--- a/gcc/config/arm/arm.c
++++ b/gcc/config/arm/arm.c
+@@ -15926,12 +15926,23 @@ operands_ok_ldrd_strd (rtx rt, rtx rt2, rtx rn, HOST_WIDE_INT offset,
+   return true;
+ }
+ 
++/* Return true if a 64-bit access with alignment ALIGN and with a
++   constant offset OFFSET from the base pointer is permitted on this
++   architecture.  */
++static bool
++align_ok_ldrd_strd (HOST_WIDE_INT align, HOST_WIDE_INT offset)
++{
++  return (unaligned_access
++	  ? (align >= BITS_PER_WORD && (offset & 3) == 0)
++	  : (align >= 2 * BITS_PER_WORD && (offset & 7) == 0));
++}
++
+ /* Helper for gen_operands_ldrd_strd.  Returns true iff the memory
+    operand MEM's address contains an immediate offset from the base
+-   register and has no side effects, in which case it sets BASE and
+-   OFFSET accordingly.  */
++   register and has no side effects, in which case it sets BASE,
++   OFFSET and ALIGN accordingly.  */
+ static bool
+-mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset)
++mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset, HOST_WIDE_INT *align)
+ {
+   rtx addr;
+ 
+@@ -15950,6 +15961,7 @@ mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset)
+   gcc_assert (MEM_P (mem));
+ 
+   *offset = const0_rtx;
++  *align = MEM_ALIGN (mem);
+ 
+   addr = XEXP (mem, 0);
+ 
+@@ -15990,7 +16002,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+                         bool const_store, bool commute)
+ {
+   int nops = 2;
+-  HOST_WIDE_INT offsets[2], offset;
++  HOST_WIDE_INT offsets[2], offset, align[2];
+   rtx base = NULL_RTX;
+   rtx cur_base, cur_offset, tmp;
+   int i, gap;
+@@ -16002,7 +16014,8 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+      registers, and the corresponding memory offsets.  */
+   for (i = 0; i < nops; i++)
+     {
+-      if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset))
++      if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset,
++				 &align[i]))
+         return false;
+ 
+       if (i == 0)
+@@ -16114,6 +16127,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+       /* Swap the instructions such that lower memory is accessed first.  */
+       std::swap (operands[0], operands[1]);
+       std::swap (operands[2], operands[3]);
++      std::swap (align[0], align[1]);
+       if (const_store)
+         std::swap (operands[4], operands[5]);
+     }
+@@ -16127,6 +16141,9 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+   if (gap != 4)
+     return false;
+ 
++  if (!align_ok_ldrd_strd (align[0], offset))
++    return false;
++
+   /* Make sure we generate legal instructions.  */
+   if (operands_ok_ldrd_strd (operands[0], operands[1], base, offset,
+                              false, load))
+diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
+index eb2b86ee7b6..d49eff6b87e 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
+@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q)
+   *p = a;
+   return a;
+ }
+-/* { dg-final { scan-assembler "ldrd" } } */
++/* { dg-final { scan-assembler "ldrd\\t" } } */
+diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c
+similarity index 63%
+copy from gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
+copy to gcc/testsuite/gcc.target/arm/peep-ldrd-2.c
+index eb2b86ee7b6..6822c2b1454 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c
+@@ -1,6 +1,6 @@
+ /* { dg-do compile } */
+ /* { dg-require-effective-target arm_prefer_ldrd_strd } */
+-/* { dg-options "-O2" }  */
++/* { dg-options "-O2 -mno-unaligned-access" }  */
+ int foo(int a, int b, int* p, int *q)
+ {
+   a = p[2] + p[3];
+@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q)
+   *p = a;
+   return a;
+ }
+-/* { dg-final { scan-assembler "ldrd" } } */
++/* { dg-final { scan-assembler-not "ldrd\\t" } } */
+diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-1.c
+index bd330769599..fe1beac7229 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-strd-1.c
+@@ -6,4 +6,4 @@ void foo(int a, int b, int* p)
+   p[2] = a;
+   p[3] = b;
+ }
+-/* { dg-final { scan-assembler "strd" } } */
++/* { dg-final { scan-assembler "strd\\t" } } */
+diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-2.c
+similarity index 58%
+copy from gcc/testsuite/gcc.target/arm/peep-strd-1.c
+copy to gcc/testsuite/gcc.target/arm/peep-strd-2.c
+index bd330769599..bfc5ebe9eec 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-strd-2.c
+@@ -1,9 +1,9 @@
+ /* { dg-do compile } */
+ /* { dg-require-effective-target arm_prefer_ldrd_strd } */
+-/* { dg-options "-O2" }  */
++/* { dg-options "-O2 -mno-unaligned-access" }  */
+ void foo(int a, int b, int* p)
+ {
+   p[2] = a;
+   p[3] = b;
+ }
+-/* { dg-final { scan-assembler "strd" } } */
++/* { dg-final { scan-assembler-not "strd\\t" } } */
+-- 
+2.15.0
+
diff --git a/meta/recipes-devtools/gcc/gcc-7.2.inc b/meta/recipes-devtools/gcc/gcc-7.2.inc
index 04edad77cb..02b8e03482 100644
--- a/meta/recipes-devtools/gcc/gcc-7.2.inc
+++ b/meta/recipes-devtools/gcc/gcc-7.2.inc
@@ -79,6 +79,7 @@  SRC_URI = "\
            ${BACKPORTS} \
 "
 BACKPORTS = "\
+           file://0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch \
 "
 SRC_URI[md5sum] = "ff370482573133a7fcdd96cd2f552292"
 SRC_URI[sha256sum] = "1cf7adf8ff4b5aa49041c8734bbcf1ad18cc4c94d0029aae0f4e48841088479a"
diff --git a/meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch b/meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch
new file mode 100644
index 0000000000..021250700d
--- /dev/null
+++ b/meta/recipes-devtools/gcc/gcc-7.2/0051-ARM-PR-82445-suppress-32-bit-aligned-ldrd-strd-peeph.patch
@@ -0,0 +1,194 @@ 
+From 8c18b422211878ba02503462cb22a2cc25a0a325 Mon Sep 17 00:00:00 2001
+From: rearnsha <rearnsha@138bc75d-0d04-0410-961f-82ee72b054a4>
+Date: Thu, 19 Oct 2017 13:14:55 +0000
+Subject: [PATCH] [ARM] PR 82445 - suppress 32-bit aligned ldrd/strd peepholing
+ with -mno-unaligned-access
+
+Peephole patterns exist in the arm backend to spot load/store
+operations to adjacent memory operations in order to convert them into
+ldrd/strd instructions.  However, when we have strict alignment
+enforced, then we can only do this if the accesses are known to be
+64-bit aligned; this is unlikely to be the case for most loads.  The
+patch adds some alignment checking to the code that validates the
+addresses for use in the peephole patterns.  This should also fix
+incorrect generation of ldrd/strd with unaligned accesses that could
+previously have occurred on ARMv5e where all such operations must be
+64-bit aligned.
+
+I've added some new tests as well.  In doing so I discovered that the
+ldrd/strd peephole tests could never fail since they would match the
+source file name in the scanned assembly as well as any instructions
+of the intended type.  I've fixed those by tightening the scan results
+slightly.
+
+gcc:
+
+* config/arm/arm.c (align_ok_ldrd_strd): New function.
+(mem_ok_for_ldrd_strd): New parameter align.  Extract the alignment of the
+mem into it.
+(gen_operands_ldrd_strd): Validate the alignment of the accesses.
+
+testsuite:
+
+* gcc.target/arm/peep-ldrd-1.c: Tighten test scan pattern.
+* gcc.target/arm/peep-strd-1.c: Likewise.
+* gcc.target/arm/peep-ldrd-2.c: New test.
+* gcc.target/arm/peep-strd-2.c: New test.
+
+
+
+git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/branches/gcc-7-branch@253891 138bc75d-0d04-0410-961f-82ee72b054a4
+---
+Upstream-Status: Backport
+Signed-off-by: Khem Raj <raj.khem@gmail.com>
+
+ gcc/ChangeLog                                      |  8 +++++++
+ gcc/config/arm/arm.c                               | 27 ++++++++++++++++++----
+ gcc/testsuite/ChangeLog                            |  8 +++++++
+ gcc/testsuite/gcc.target/arm/peep-ldrd-1.c         |  2 +-
+ .../arm/{peep-ldrd-1.c => peep-ldrd-2.c}           |  4 ++--
+ gcc/testsuite/gcc.target/arm/peep-strd-1.c         |  2 +-
+ .../arm/{peep-strd-1.c => peep-strd-2.c}           |  4 ++--
+ 7 files changed, 44 insertions(+), 11 deletions(-)
+ copy gcc/testsuite/gcc.target/arm/{peep-ldrd-1.c => peep-ldrd-2.c} (63%)
+ copy gcc/testsuite/gcc.target/arm/{peep-strd-1.c => peep-strd-2.c} (58%)
+
+diff --git a/gcc/config/arm/arm.c b/gcc/config/arm/arm.c
+index 1ded0d2a17d..989957f048e 100644
+--- a/gcc/config/arm/arm.c
++++ b/gcc/config/arm/arm.c
+@@ -15199,12 +15199,23 @@ operands_ok_ldrd_strd (rtx rt, rtx rt2, rtx rn, HOST_WIDE_INT offset,
+   return true;
+ }
+ 
++/* Return true if a 64-bit access with alignment ALIGN and with a
++   constant offset OFFSET from the base pointer is permitted on this
++   architecture.  */
++static bool
++align_ok_ldrd_strd (HOST_WIDE_INT align, HOST_WIDE_INT offset)
++{
++  return (unaligned_access
++	  ? (align >= BITS_PER_WORD && (offset & 3) == 0)
++	  : (align >= 2 * BITS_PER_WORD && (offset & 7) == 0));
++}
++
+ /* Helper for gen_operands_ldrd_strd.  Returns true iff the memory
+    operand MEM's address contains an immediate offset from the base
+-   register and has no side effects, in which case it sets BASE and
+-   OFFSET accordingly.  */
++   register and has no side effects, in which case it sets BASE,
++   OFFSET and ALIGN accordingly.  */
+ static bool
+-mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset)
++mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset, HOST_WIDE_INT *align)
+ {
+   rtx addr;
+ 
+@@ -15223,6 +15234,7 @@ mem_ok_for_ldrd_strd (rtx mem, rtx *base, rtx *offset)
+   gcc_assert (MEM_P (mem));
+ 
+   *offset = const0_rtx;
++  *align = MEM_ALIGN (mem);
+ 
+   addr = XEXP (mem, 0);
+ 
+@@ -15263,7 +15275,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+                         bool const_store, bool commute)
+ {
+   int nops = 2;
+-  HOST_WIDE_INT offsets[2], offset;
++  HOST_WIDE_INT offsets[2], offset, align[2];
+   rtx base = NULL_RTX;
+   rtx cur_base, cur_offset, tmp;
+   int i, gap;
+@@ -15275,7 +15287,8 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+      registers, and the corresponding memory offsets.  */
+   for (i = 0; i < nops; i++)
+     {
+-      if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset))
++      if (!mem_ok_for_ldrd_strd (operands[nops+i], &cur_base, &cur_offset,
++				 &align[i]))
+         return false;
+ 
+       if (i == 0)
+@@ -15389,6 +15402,7 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+       /* Swap the instructions such that lower memory is accessed first.  */
+       std::swap (operands[0], operands[1]);
+       std::swap (operands[2], operands[3]);
++      std::swap (align[0], align[1]);
+       if (const_store)
+         std::swap (operands[4], operands[5]);
+     }
+@@ -15402,6 +15416,9 @@ gen_operands_ldrd_strd (rtx *operands, bool load,
+   if (gap != 4)
+     return false;
+ 
++  if (!align_ok_ldrd_strd (align[0], offset))
++    return false;
++
+   /* Make sure we generate legal instructions.  */
+   if (operands_ok_ldrd_strd (operands[0], operands[1], base, offset,
+                              false, load))
+diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
+index eb2b86ee7b6..d49eff6b87e 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
+@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q)
+   *p = a;
+   return a;
+ }
+-/* { dg-final { scan-assembler "ldrd" } } */
++/* { dg-final { scan-assembler "ldrd\\t" } } */
+diff --git a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c
+similarity index 63%
+copy from gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
+copy to gcc/testsuite/gcc.target/arm/peep-ldrd-2.c
+index eb2b86ee7b6..6822c2b1454 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-ldrd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-ldrd-2.c
+@@ -1,6 +1,6 @@
+ /* { dg-do compile } */
+ /* { dg-require-effective-target arm_prefer_ldrd_strd } */
+-/* { dg-options "-O2" }  */
++/* { dg-options "-O2 -mno-unaligned-access" }  */
+ int foo(int a, int b, int* p, int *q)
+ {
+   a = p[2] + p[3];
+@@ -8,4 +8,4 @@ int foo(int a, int b, int* p, int *q)
+   *p = a;
+   return a;
+ }
+-/* { dg-final { scan-assembler "ldrd" } } */
++/* { dg-final { scan-assembler-not "ldrd\\t" } } */
+diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-1.c
+index bd330769599..fe1beac7229 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-strd-1.c
+@@ -6,4 +6,4 @@ void foo(int a, int b, int* p)
+   p[2] = a;
+   p[3] = b;
+ }
+-/* { dg-final { scan-assembler "strd" } } */
++/* { dg-final { scan-assembler "strd\\t" } } */
+diff --git a/gcc/testsuite/gcc.target/arm/peep-strd-1.c b/gcc/testsuite/gcc.target/arm/peep-strd-2.c
+similarity index 58%
+copy from gcc/testsuite/gcc.target/arm/peep-strd-1.c
+copy to gcc/testsuite/gcc.target/arm/peep-strd-2.c
+index bd330769599..bfc5ebe9eec 100644
+--- a/gcc/testsuite/gcc.target/arm/peep-strd-1.c
++++ b/gcc/testsuite/gcc.target/arm/peep-strd-2.c
+@@ -1,9 +1,9 @@
+ /* { dg-do compile } */
+ /* { dg-require-effective-target arm_prefer_ldrd_strd } */
+-/* { dg-options "-O2" }  */
++/* { dg-options "-O2 -mno-unaligned-access" }  */
+ void foo(int a, int b, int* p)
+ {
+   p[2] = a;
+   p[3] = b;
+ }
+-/* { dg-final { scan-assembler "strd" } } */
++/* { dg-final { scan-assembler-not "strd\\t" } } */
+-- 
+2.15.0
+