[PATCH 0/6] MIPS: Use thin archives & dead code elimination

[PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Paul Burton]

This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
leads to a typical generic kernel build becoming ~5% smaller:

  add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
  ...
  Total: Before=9001030, After=8543686, chg -5.08%

Applies atop v4.9-rc1.


Paul Burton (6):
  kbuild: Keep device tree tables though dead code elimination
  kbuild: Keep .init.setup section through dead code elimination
  kbuild: Keep PCI fixups through dead code elimination
  kbuild: Keep earlycon table through dead code elimination
  MIPS: Ensure bss section ends on a long-aligned address
  MIPS: Use thin archives & dead code elimination

 arch/mips/Kconfig                 |  2 ++
 arch/mips/kernel/vmlinux.lds.S    |  7 ++++---
 include/asm-generic/vmlinux.lds.h | 24 ++++++++++++------------
 3 files changed, 18 insertions(+), 15 deletions(-)

-- 
2.10.0

[PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Paul Burton]

This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
leads to a typical generic kernel build becoming ~5% smaller:

  add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
  ...
  Total: Before=9001030, After=8543686, chg -5.08%

Applies atop v4.9-rc1.


Paul Burton (6):
  kbuild: Keep device tree tables though dead code elimination
  kbuild: Keep .init.setup section through dead code elimination
  kbuild: Keep PCI fixups through dead code elimination
  kbuild: Keep earlycon table through dead code elimination
  MIPS: Ensure bss section ends on a long-aligned address
  MIPS: Use thin archives & dead code elimination

 arch/mips/Kconfig                 |  2 ++
 arch/mips/kernel/vmlinux.lds.S    |  7 ++++---
 include/asm-generic/vmlinux.lds.h | 24 ++++++++++++------------
 3 files changed, 18 insertions(+), 15 deletions(-)

-- 
2.10.0

[PATCH 1/6] kbuild: Keep device tree tables though dead code elimination

[Paul Burton]

When CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is enabled we must ensure that
we still keep the device tree tables, otherwise we end up unable to
probe drivers on systems that do so using device tree & we throw away
the bulk of those drivers because they cease to be referenced.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 3074796..b648d97 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -169,8 +169,8 @@
 #define _OF_TABLE_1(name)						\
 	. = ALIGN(8);							\
 	VMLINUX_SYMBOL(__##name##_of_table) = .;			\
-	*(__##name##_of_table)						\
-	*(__##name##_of_table_end)
+	KEEP(*(__##name##_of_table))					\
+	KEEP(*(__##name##_of_table_end))
 
 #define CLKSRC_OF_TABLES()	OF_TABLE(CONFIG_CLKSRC_OF, clksrc)
 #define IRQCHIP_OF_MATCH_TABLE() OF_TABLE(CONFIG_IRQCHIP, irqchip)
-- 
2.10.0

[PATCH 1/6] kbuild: Keep device tree tables though dead code elimination

[Paul Burton]

When CONFIG_LD_DEAD_CODE_DATA_ELIMINATION is enabled we must ensure that
we still keep the device tree tables, otherwise we end up unable to
probe drivers on systems that do so using device tree & we throw away
the bulk of those drivers because they cease to be referenced.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 3074796..b648d97 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -169,8 +169,8 @@
 #define _OF_TABLE_1(name)						\
 	. = ALIGN(8);							\
 	VMLINUX_SYMBOL(__##name##_of_table) = .;			\
-	*(__##name##_of_table)						\
-	*(__##name##_of_table_end)
+	KEEP(*(__##name##_of_table))					\
+	KEEP(*(__##name##_of_table_end))
 
 #define CLKSRC_OF_TABLES()	OF_TABLE(CONFIG_CLKSRC_OF, clksrc)
 #define IRQCHIP_OF_MATCH_TABLE() OF_TABLE(CONFIG_IRQCHIP, irqchip)
-- 
2.10.0

[PATCH 2/6] kbuild: Keep .init.setup section through dead code elimination

[Paul Burton]

With CONFIG_LD_DEAD_CODE_DATA_ELIMINATION we must keep the .init.setup
section or we'll throw away support for kernel parameters.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index b648d97..06673d6 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -688,7 +688,7 @@
 #define INIT_SETUP(initsetup_align)					\
 		. = ALIGN(initsetup_align);				\
 		VMLINUX_SYMBOL(__setup_start) = .;			\
-		*(.init.setup)						\
+		KEEP(*(.init.setup))					\
 		VMLINUX_SYMBOL(__setup_end) = .;
 
 #define INIT_CALLS_LEVEL(level)						\
-- 
2.10.0

[PATCH 2/6] kbuild: Keep .init.setup section through dead code elimination

[Paul Burton]

With CONFIG_LD_DEAD_CODE_DATA_ELIMINATION we must keep the .init.setup
section or we'll throw away support for kernel parameters.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index b648d97..06673d6 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -688,7 +688,7 @@
 #define INIT_SETUP(initsetup_align)					\
 		. = ALIGN(initsetup_align);				\
 		VMLINUX_SYMBOL(__setup_start) = .;			\
-		*(.init.setup)						\
+		KEEP(*(.init.setup))					\
 		VMLINUX_SYMBOL(__setup_end) = .;
 
 #define INIT_CALLS_LEVEL(level)						\
-- 
2.10.0

[PATCH 3/6] kbuild: Keep PCI fixups through dead code elimination

[Paul Burton]

With CONFIG_LD_DEAD_CODE_DATA_ELIMINATION enabled we must ensure that we
keep PCI fixups or they will be discarded during the link & we'll lose
support for any PCI fixups.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 06673d6..e710593 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -288,28 +288,28 @@
 	/* PCI quirks */						\
 	.pci_fixup        : AT(ADDR(.pci_fixup) - LOAD_OFFSET) {	\
 		VMLINUX_SYMBOL(__start_pci_fixups_early) = .;		\
-		*(.pci_fixup_early)					\
+		KEEP(*(.pci_fixup_early))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_early) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_header) = .;		\
-		*(.pci_fixup_header)					\
+		KEEP(*(.pci_fixup_header))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_header) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_final) = .;		\
-		*(.pci_fixup_final)					\
+		KEEP(*(.pci_fixup_final))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_final) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_enable) = .;		\
-		*(.pci_fixup_enable)					\
+		KEEP(*(.pci_fixup_enable))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_enable) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_resume) = .;		\
-		*(.pci_fixup_resume)					\
+		KEEP(*(.pci_fixup_resume))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_resume) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_resume_early) = .;	\
-		*(.pci_fixup_resume_early)				\
+		KEEP(*(.pci_fixup_resume_early))			\
 		VMLINUX_SYMBOL(__end_pci_fixups_resume_early) = .;	\
 		VMLINUX_SYMBOL(__start_pci_fixups_suspend) = .;		\
-		*(.pci_fixup_suspend)					\
+		KEEP(*(.pci_fixup_suspend))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_suspend) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_suspend_late) = .;	\
-		*(.pci_fixup_suspend_late)				\
+		KEEP(*(.pci_fixup_suspend_late))			\
 		VMLINUX_SYMBOL(__end_pci_fixups_suspend_late) = .;	\
 	}								\
 									\
-- 
2.10.0

[PATCH 3/6] kbuild: Keep PCI fixups through dead code elimination

[Paul Burton]

With CONFIG_LD_DEAD_CODE_DATA_ELIMINATION enabled we must ensure that we
keep PCI fixups or they will be discarded during the link & we'll lose
support for any PCI fixups.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index 06673d6..e710593 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -288,28 +288,28 @@
 	/* PCI quirks */						\
 	.pci_fixup        : AT(ADDR(.pci_fixup) - LOAD_OFFSET) {	\
 		VMLINUX_SYMBOL(__start_pci_fixups_early) = .;		\
-		*(.pci_fixup_early)					\
+		KEEP(*(.pci_fixup_early))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_early) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_header) = .;		\
-		*(.pci_fixup_header)					\
+		KEEP(*(.pci_fixup_header))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_header) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_final) = .;		\
-		*(.pci_fixup_final)					\
+		KEEP(*(.pci_fixup_final))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_final) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_enable) = .;		\
-		*(.pci_fixup_enable)					\
+		KEEP(*(.pci_fixup_enable))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_enable) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_resume) = .;		\
-		*(.pci_fixup_resume)					\
+		KEEP(*(.pci_fixup_resume))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_resume) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_resume_early) = .;	\
-		*(.pci_fixup_resume_early)				\
+		KEEP(*(.pci_fixup_resume_early))			\
 		VMLINUX_SYMBOL(__end_pci_fixups_resume_early) = .;	\
 		VMLINUX_SYMBOL(__start_pci_fixups_suspend) = .;		\
-		*(.pci_fixup_suspend)					\
+		KEEP(*(.pci_fixup_suspend))				\
 		VMLINUX_SYMBOL(__end_pci_fixups_suspend) = .;		\
 		VMLINUX_SYMBOL(__start_pci_fixups_suspend_late) = .;	\
-		*(.pci_fixup_suspend_late)				\
+		KEEP(*(.pci_fixup_suspend_late))			\
 		VMLINUX_SYMBOL(__end_pci_fixups_suspend_late) = .;	\
 	}								\
 									\
-- 
2.10.0

[PATCH 4/6] kbuild: Keep earlycon table through dead code elimination

[Paul Burton]

With CONFIG_LD_DEAD_CODE_DATA_ELIMINATION enabled we must ensure that we
keep the earlycon table or we discard all entries leading to no earlycon
devices being found.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index e710593..abe79e3 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -156,7 +156,7 @@
 #ifdef CONFIG_SERIAL_EARLYCON
 #define EARLYCON_TABLE() STRUCT_ALIGN();			\
 			 VMLINUX_SYMBOL(__earlycon_table) = .;	\
-			 *(__earlycon_table)			\
+			 KEEP(*(__earlycon_table))		\
 			 VMLINUX_SYMBOL(__earlycon_table_end) = .;
 #else
 #define EARLYCON_TABLE()
-- 
2.10.0

[PATCH 4/6] kbuild: Keep earlycon table through dead code elimination

[Paul Burton]

With CONFIG_LD_DEAD_CODE_DATA_ELIMINATION enabled we must ensure that we
keep the earlycon table or we discard all entries leading to no earlycon
devices being found.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 include/asm-generic/vmlinux.lds.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h
index e710593..abe79e3 100644
--- a/include/asm-generic/vmlinux.lds.h
+++ b/include/asm-generic/vmlinux.lds.h
@@ -156,7 +156,7 @@
 #ifdef CONFIG_SERIAL_EARLYCON
 #define EARLYCON_TABLE() STRUCT_ALIGN();			\
 			 VMLINUX_SYMBOL(__earlycon_table) = .;	\
-			 *(__earlycon_table)			\
+			 KEEP(*(__earlycon_table))		\
 			 VMLINUX_SYMBOL(__earlycon_table_end) = .;
 #else
 #define EARLYCON_TABLE()
-- 
2.10.0

[PATCH 5/6] MIPS: Ensure bss section ends on a long-aligned address

[Paul Burton]

When clearing the .bss section in kernel_entry we do so using LONG_S
instructions, and branch whilst the current write address doesn't equal
the end of the .bss section minus the size of a long integer. The .bss
section always begins at a long-aligned address and we always increment
the write pointer by the size of a long integer - we therefore rely upon
the .bss section ending at a long-aligned address. If this is not the
case then the long-aligned write address can never be equal to the
non-long-aligned end address & we will continue to increment past the
end of the .bss section, attempting to zero the rest of memory.

Despite this requirement that .bss end at a long-aligned address we pass
0 as the end alignment requirement to the BSS_SECTION macro and thus
don't guarantee any particular alignment, allowing us to hit the error
condition described above.

Fix this by instead passing LONGSIZE as the end alignment argument to
the BSS_SECTION macro.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 arch/mips/kernel/vmlinux.lds.S | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
index d5de675..d1f5401 100644
--- a/arch/mips/kernel/vmlinux.lds.S
+++ b/arch/mips/kernel/vmlinux.lds.S
@@ -1,3 +1,4 @@
+#include <asm/asm.h>
 #include <asm/asm-offsets.h>
 #include <asm/thread_info.h>
 
@@ -182,7 +183,7 @@ SECTIONS
 	 * Force .bss to 64K alignment so that .bss..swapper_pg_dir
 	 * gets that alignment.	 .sbss should be empty, so there will be
 	 * no holes after __init_end. */
-	BSS_SECTION(0, 0x10000, 0)
+	BSS_SECTION(0, 0x10000, LONGSIZE)
 
 	_end = . ;
 
-- 
2.10.0

[PATCH 5/6] MIPS: Ensure bss section ends on a long-aligned address

[Paul Burton]

When clearing the .bss section in kernel_entry we do so using LONG_S
instructions, and branch whilst the current write address doesn't equal
the end of the .bss section minus the size of a long integer. The .bss
section always begins at a long-aligned address and we always increment
the write pointer by the size of a long integer - we therefore rely upon
the .bss section ending at a long-aligned address. If this is not the
case then the long-aligned write address can never be equal to the
non-long-aligned end address & we will continue to increment past the
end of the .bss section, attempting to zero the rest of memory.

Despite this requirement that .bss end at a long-aligned address we pass
0 as the end alignment requirement to the BSS_SECTION macro and thus
don't guarantee any particular alignment, allowing us to hit the error
condition described above.

Fix this by instead passing LONGSIZE as the end alignment argument to
the BSS_SECTION macro.

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Nicholas Piggin <npiggin@gmail.com>
---

 arch/mips/kernel/vmlinux.lds.S | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
index d5de675..d1f5401 100644
--- a/arch/mips/kernel/vmlinux.lds.S
+++ b/arch/mips/kernel/vmlinux.lds.S
@@ -1,3 +1,4 @@
+#include <asm/asm.h>
 #include <asm/asm-offsets.h>
 #include <asm/thread_info.h>
 
@@ -182,7 +183,7 @@ SECTIONS
 	 * Force .bss to 64K alignment so that .bss..swapper_pg_dir
 	 * gets that alignment.	 .sbss should be empty, so there will be
 	 * no holes after __init_end. */
-	BSS_SECTION(0, 0x10000, 0)
+	BSS_SECTION(0, 0x10000, LONGSIZE)
 
 	_end = . ;
 
-- 
2.10.0

[PATCH 6/6] MIPS: Use thin archives & dead code elimination

[Paul Burton]

Enable CONFIG_THIN_ARCHIVES & CONFIG_LD_DEAD_CODE_DATA_ELIMINATION for
MIPS, ensuring that we keep the data bus exception table & the machine
list which would be discarded without marking them with KEEP.

This shrinks a typical generic kernel build with drivers enabled for
Boston, Ci40 & SEAD-3 by around 5%, or ~450kb. As reported by
bloat-o-meter:

  add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
  ...
  Total: Before=9001030, After=8543686, chg -5.08%

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Nicholas Piggin <npiggin@gmail.com>

---

 arch/mips/Kconfig              | 2 ++
 arch/mips/kernel/vmlinux.lds.S | 4 ++--
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index b3c5bde..8557667 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -66,6 +66,8 @@ config MIPS
 	select HAVE_EXIT_THREAD
 	select HAVE_REGS_AND_STACK_ACCESS_API
 	select HAVE_ARCH_HARDENED_USERCOPY
+	select THIN_ARCHIVES
+	select LD_DEAD_CODE_DATA_ELIMINATION
 
 menu "Machine selection"
 
diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
index d1f5401..a43ba2a 100644
--- a/arch/mips/kernel/vmlinux.lds.S
+++ b/arch/mips/kernel/vmlinux.lds.S
@@ -72,7 +72,7 @@ SECTIONS
 	/* Exception table for data bus errors */
 	__dbe_table : {
 		__start___dbe_table = .;
-		*(__dbe_table)
+		KEEP(*(__dbe_table))
 		__stop___dbe_table = .;
 	}
 
@@ -122,7 +122,7 @@ SECTIONS
 	. = ALIGN(4);
 	.mips.machines.init : AT(ADDR(.mips.machines.init) - LOAD_OFFSET) {
 		__mips_machines_start = .;
-		*(.mips.machines.init)
+		KEEP(*(.mips.machines.init))
 		__mips_machines_end = .;
 	}
 
-- 
2.10.0

[PATCH 6/6] MIPS: Use thin archives & dead code elimination

[Paul Burton]

Enable CONFIG_THIN_ARCHIVES & CONFIG_LD_DEAD_CODE_DATA_ELIMINATION for
MIPS, ensuring that we keep the data bus exception table & the machine
list which would be discarded without marking them with KEEP.

This shrinks a typical generic kernel build with drivers enabled for
Boston, Ci40 & SEAD-3 by around 5%, or ~450kb. As reported by
bloat-o-meter:

  add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
  ...
  Total: Before=9001030, After=8543686, chg -5.08%

Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Nicholas Piggin <npiggin@gmail.com>

---

 arch/mips/Kconfig              | 2 ++
 arch/mips/kernel/vmlinux.lds.S | 4 ++--
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index b3c5bde..8557667 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -66,6 +66,8 @@ config MIPS
 	select HAVE_EXIT_THREAD
 	select HAVE_REGS_AND_STACK_ACCESS_API
 	select HAVE_ARCH_HARDENED_USERCOPY
+	select THIN_ARCHIVES
+	select LD_DEAD_CODE_DATA_ELIMINATION
 
 menu "Machine selection"
 
diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
index d1f5401..a43ba2a 100644
--- a/arch/mips/kernel/vmlinux.lds.S
+++ b/arch/mips/kernel/vmlinux.lds.S
@@ -72,7 +72,7 @@ SECTIONS
 	/* Exception table for data bus errors */
 	__dbe_table : {
 		__start___dbe_table = .;
-		*(__dbe_table)
+		KEEP(*(__dbe_table))
 		__stop___dbe_table = .;
 	}
 
@@ -122,7 +122,7 @@ SECTIONS
 	. = ALIGN(4);
 	.mips.machines.init : AT(ADDR(.mips.machines.init) - LOAD_OFFSET) {
 		__mips_machines_start = .;
-		*(.mips.machines.init)
+		KEEP(*(.mips.machines.init))
 		__mips_machines_end = .;
 	}
 
-- 
2.10.0

Re: [PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Nicholas Piggin]

Hey Paul,

On Thu, 20 Oct 2016 21:26:59 +0100
Paul Burton <paul.burton@imgtec.com> wrote:

> This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
> leads to a typical generic kernel build becoming ~5% smaller:
> 
>   add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
>   ...
>   Total: Before=9001030, After=8543686, chg -5.08%
> 
> Applies atop v4.9-rc1.


Very nice, and thanks for the kbuild fixes, I think they all look sane.

Let's try to get those kbuild fixes in through the kbuild tree first
(which has some other fixes required for 4.9). I can take them and send
them to kbuild maintainer if you like.

On powerpc we'll likely provide an option to select these manually for
4.9 because there has been the odd toolchain issue come up, so that's
something to consider.

For your linker script, you may consider putting the function sections
into the same input as other text. TEXT_TEXT does not include .text.*,
so mips's .text.* below it will catch those.

You may just open-code your TEXT_TEXT, and have:

*(.text.hot .text .text.fixup .text.unlikely .text.[0-9a-zA-Z_]*)

or similar.

Thanks,
Nick

Re: [PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Nicholas Piggin]

Hey Paul,

On Thu, 20 Oct 2016 21:26:59 +0100
Paul Burton <paul.burton@imgtec.com> wrote:

> This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
> leads to a typical generic kernel build becoming ~5% smaller:
> 
>   add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
>   ...
>   Total: Before=9001030, After=8543686, chg -5.08%
> 
> Applies atop v4.9-rc1.


Very nice, and thanks for the kbuild fixes, I think they all look sane.

Let's try to get those kbuild fixes in through the kbuild tree first
(which has some other fixes required for 4.9). I can take them and send
them to kbuild maintainer if you like.

On powerpc we'll likely provide an option to select these manually for
4.9 because there has been the odd toolchain issue come up, so that's
something to consider.

For your linker script, you may consider putting the function sections
into the same input as other text. TEXT_TEXT does not include .text.*,
so mips's .text.* below it will catch those.

You may just open-code your TEXT_TEXT, and have:

*(.text.hot .text .text.fixup .text.unlikely .text.[0-9a-zA-Z_]*)

or similar.

Thanks,
Nick

Re: [PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Paul Burton]

[-- Attachment #1: Type: text/plain, Size: 1625 bytes --]

Hi Nick,

On Friday, 21 October 2016 11:51:47 BST Nicholas Piggin wrote:
> Paul Burton <paul.burton@imgtec.com> wrote:
> > This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> > and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
> > 
> > leads to a typical generic kernel build becoming ~5% smaller:
> >   add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
> >   ...
> >   Total: Before=9001030, After=8543686, chg -5.08%
> > 
> > Applies atop v4.9-rc1.
> 
> Very nice, and thanks for the kbuild fixes, I think they all look sane.
> 
> Let's try to get those kbuild fixes in through the kbuild tree first
> (which has some other fixes required for 4.9). I can take them and send
> them to kbuild maintainer if you like.

That sounds great :)

> On powerpc we'll likely provide an option to select these manually for
> 4.9 because there has been the odd toolchain issue come up, so that's
> something to consider.

I imagine the MIPS bits will probably be v4.10 material, but hopefully Ralf 
can get them into -next as soon as possible after the kbuild bits are in, 
which should give us some time to discover any toolchain issues.

> For your linker script, you may consider putting the function sections
> into the same input as other text. TEXT_TEXT does not include .text.*,
> so mips's .text.* below it will catch those.
> 
> You may just open-code your TEXT_TEXT, and have:
> 
> *(.text.hot .text .text.fixup .text.unlikely .text.[0-9a-zA-Z_]*)
> 
> or similar.

Ah, so are you saying that would give the linker more scope for discarding 
things?

Thanks,
    Paul

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 801 bytes --]

Re: [PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Paul Burton]

[-- Attachment #1: Type: text/plain, Size: 1625 bytes --]

Hi Nick,

On Friday, 21 October 2016 11:51:47 BST Nicholas Piggin wrote:
> Paul Burton <paul.burton@imgtec.com> wrote:
> > This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> > and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
> > 
> > leads to a typical generic kernel build becoming ~5% smaller:
> >   add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
> >   ...
> >   Total: Before=9001030, After=8543686, chg -5.08%
> > 
> > Applies atop v4.9-rc1.
> 
> Very nice, and thanks for the kbuild fixes, I think they all look sane.
> 
> Let's try to get those kbuild fixes in through the kbuild tree first
> (which has some other fixes required for 4.9). I can take them and send
> them to kbuild maintainer if you like.

That sounds great :)

> On powerpc we'll likely provide an option to select these manually for
> 4.9 because there has been the odd toolchain issue come up, so that's
> something to consider.

I imagine the MIPS bits will probably be v4.10 material, but hopefully Ralf 
can get them into -next as soon as possible after the kbuild bits are in, 
which should give us some time to discover any toolchain issues.

> For your linker script, you may consider putting the function sections
> into the same input as other text. TEXT_TEXT does not include .text.*,
> so mips's .text.* below it will catch those.
> 
> You may just open-code your TEXT_TEXT, and have:
> 
> *(.text.hot .text .text.fixup .text.unlikely .text.[0-9a-zA-Z_]*)
> 
> or similar.

Ah, so are you saying that would give the linker more scope for discarding 
things?

Thanks,
    Paul

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 801 bytes --]

Re: [PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Nicholas Piggin]

On Fri, 21 Oct 2016 08:45:47 +0100
Paul Burton <paul.burton@imgtec.com> wrote:

> Hi Nick,
> 
> On Friday, 21 October 2016 11:51:47 BST Nicholas Piggin wrote:
> > Paul Burton <paul.burton@imgtec.com> wrote:  
> > > This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> > > and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
> > > 
> > > leads to a typical generic kernel build becoming ~5% smaller:
> > >   add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
> > >   ...
> > >   Total: Before=9001030, After=8543686, chg -5.08%
> > > 
> > > Applies atop v4.9-rc1.  
> > 
> > Very nice, and thanks for the kbuild fixes, I think they all look sane.
> > 
> > Let's try to get those kbuild fixes in through the kbuild tree first
> > (which has some other fixes required for 4.9). I can take them and send
> > them to kbuild maintainer if you like.  
> 
> That sounds great :)
> 
> > On powerpc we'll likely provide an option to select these manually for
> > 4.9 because there has been the odd toolchain issue come up, so that's
> > something to consider.  
> 
> I imagine the MIPS bits will probably be v4.10 material, but hopefully Ralf 
> can get them into -next as soon as possible after the kbuild bits are in, 
> which should give us some time to discover any toolchain issues.

Okay, whatever works for you.


> > For your linker script, you may consider putting the function sections
> > into the same input as other text. TEXT_TEXT does not include .text.*,
> > so mips's .text.* below it will catch those.
> > 
> > You may just open-code your TEXT_TEXT, and have:
> > 
> > *(.text.hot .text .text.fixup .text.unlikely .text.[0-9a-zA-Z_]*)
> > 
> > or similar.  
> 
> Ah, so are you saying that would give the linker more scope for discarding 
> things?

I don't think discarding, but it will allow those sections to be placed
together with more scope for reordering. This can reduce the amount of
branch trampolines required.

Aside from that, it just gives less change in behaviour. Without
-ffunction-sections, those functions would be placed in .text section,
so you ideally want to keep them there when enabling it.

Thanks,
Nick

Re: [PATCH 0/6] MIPS: Use thin archives & dead code elimination

[Nicholas Piggin]

On Fri, 21 Oct 2016 08:45:47 +0100
Paul Burton <paul.burton@imgtec.com> wrote:

> Hi Nick,
> 
> On Friday, 21 October 2016 11:51:47 BST Nicholas Piggin wrote:
> > Paul Burton <paul.burton@imgtec.com> wrote:  
> > > This series fixes a few issues with CONFIG_LD_DEAD_CODE_DATA_ELIMINATION
> > > and then enables it, along with CONFIG_THIN_ARCHIVES, for MIPS. This
> > > 
> > > leads to a typical generic kernel build becoming ~5% smaller:
> > >   add/remove: 0/3028 grow/shrink: 1/14 up/down: 18/-457362 (-457344)
> > >   ...
> > >   Total: Before=9001030, After=8543686, chg -5.08%
> > > 
> > > Applies atop v4.9-rc1.  
> > 
> > Very nice, and thanks for the kbuild fixes, I think they all look sane.
> > 
> > Let's try to get those kbuild fixes in through the kbuild tree first
> > (which has some other fixes required for 4.9). I can take them and send
> > them to kbuild maintainer if you like.  
> 
> That sounds great :)
> 
> > On powerpc we'll likely provide an option to select these manually for
> > 4.9 because there has been the odd toolchain issue come up, so that's
> > something to consider.  
> 
> I imagine the MIPS bits will probably be v4.10 material, but hopefully Ralf 
> can get them into -next as soon as possible after the kbuild bits are in, 
> which should give us some time to discover any toolchain issues.

Okay, whatever works for you.


> > For your linker script, you may consider putting the function sections
> > into the same input as other text. TEXT_TEXT does not include .text.*,
> > so mips's .text.* below it will catch those.
> > 
> > You may just open-code your TEXT_TEXT, and have:
> > 
> > *(.text.hot .text .text.fixup .text.unlikely .text.[0-9a-zA-Z_]*)
> > 
> > or similar.  
> 
> Ah, so are you saying that would give the linker more scope for discarding 
> things?

I don't think discarding, but it will allow those sections to be placed
together with more scope for reordering. This can reduce the amount of
branch trampolines required.

Aside from that, it just gives less change in behaviour. Without
-ffunction-sections, those functions would be placed in .text section,
so you ideally want to keep them there when enabling it.

Thanks,
Nick

Re: [PATCH 1/6] kbuild: Keep device tree tables though dead code elimination

[Sergei Shtylyov]

Hello.

    s/though/through/ in the subject.

MBR, Sergei

Re: [PATCH 5/6] MIPS: Ensure bss section ends on a long-aligned address

[Marcin Nowakowski]

Hi Paul,

On 20.10.2016 22:27, Paul Burton wrote:
> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
> index d5de675..d1f5401 100644
> --- a/arch/mips/kernel/vmlinux.lds.S
> +++ b/arch/mips/kernel/vmlinux.lds.S
> @@ -1,3 +1,4 @@
> +#include <asm/asm.h>
>  #include <asm/asm-offsets.h>
>  #include <asm/thread_info.h>

This patch (and specifically including asm/asm.h) causes a build error 
when CONFIG_RELOCATABLE=y as there's a LONG() linker command that gets 
converted to '.[d]word' as a result of including the assembly macros ...

Marcin

Re: [PATCH 5/6] MIPS: Ensure bss section ends on a long-aligned address

[Marcin Nowakowski]

Hi Paul,

On 20.10.2016 22:27, Paul Burton wrote:
> diff --git a/arch/mips/kernel/vmlinux.lds.S b/arch/mips/kernel/vmlinux.lds.S
> index d5de675..d1f5401 100644
> --- a/arch/mips/kernel/vmlinux.lds.S
> +++ b/arch/mips/kernel/vmlinux.lds.S
> @@ -1,3 +1,4 @@
> +#include <asm/asm.h>
>  #include <asm/asm-offsets.h>
>  #include <asm/thread_info.h>

This patch (and specifically including asm/asm.h) causes a build error 
when CONFIG_RELOCATABLE=y as there's a LONG() linker command that gets 
converted to '.[d]word' as a result of including the assembly macros ...

Marcin