| Message ID | 20161012070020.30170-1-npiggin@gmail.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
Nicholas Piggin <npiggin@gmail.com> writes: > Add --orphan-handling=error to final link flags. This ensures we have to > handle all sections. This would have caught subtle breakage such as > 7de3b27bac47da9de08409df1d69664acbb72197 at build-time. > > Also bring some wayward sections into the fold: > - .text.hot and .text.unlikely are compiler generated sections. > - .sfpr is a linker generated section for register save functions. > - .sdata2, .dynsbss, .plt are used by PPC32 > - We previously did not specify DWARF_DEBUG or STABS_DEBUG > - DWARF_DEBUG did not include DWARF3 .debug_ranges > - A number of sections are unused. > > I don't know if I've exactly got everything right here, particularly > with ppc32, so would appreciate people casting their eye over it. > > Signed-off-by: Nicholas Piggin <npiggin@gmail.com> > --- > arch/powerpc/Makefile | 2 +- > arch/powerpc/kernel/vmlinux.lds.S | 16 ++++++++++++++-- > include/asm-generic/vmlinux.lds.h | 3 +++ > 3 files changed, 18 insertions(+), 3 deletions(-) > > diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile > index 50d020a..a3f2784 100644 > --- a/arch/powerpc/Makefile > +++ b/arch/powerpc/Makefile > @@ -90,7 +90,7 @@ endif > > LDFLAGS_vmlinux-y := -Bstatic > LDFLAGS_vmlinux-$(CONFIG_RELOCATABLE) := -pie > -LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) > +LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) --orphan-handling=error At least some old(er) toolchains don't support that: /opt/cross/kisskb/gcc-4.6.3-nolibc/powerpc-linux/bin/powerpc-linux-ld: unrecognized option '--orphan-handling=error' So we'll need an ld-option wrapper around it. cheers
On 12/10/16 18:00, Nicholas Piggin wrote: > Add --orphan-handling=error to final link flags. This ensures we have to > handle all sections. This would have caught subtle breakage such as > 7de3b27bac47da9de08409df1d69664acbb72197 at build-time. > > Also bring some wayward sections into the fold: > - .text.hot and .text.unlikely are compiler generated sections. > - .sfpr is a linker generated section for register save functions. > - .sdata2, .dynsbss, .plt are used by PPC32 > - We previously did not specify DWARF_DEBUG or STABS_DEBUG > - DWARF_DEBUG did not include DWARF3 .debug_ranges > - A number of sections are unused. > > I don't know if I've exactly got everything right here, particularly > with ppc32, so would appreciate people casting their eye over it. > > Signed-off-by: Nicholas Piggin <npiggin@gmail.com> > --- > arch/powerpc/Makefile | 2 +- > arch/powerpc/kernel/vmlinux.lds.S | 16 ++++++++++++++-- > include/asm-generic/vmlinux.lds.h | 3 +++ > 3 files changed, 18 insertions(+), 3 deletions(-) > > diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile > index 50d020a..a3f2784 100644 > --- a/arch/powerpc/Makefile > +++ b/arch/powerpc/Makefile > @@ -90,7 +90,7 @@ endif > > LDFLAGS_vmlinux-y := -Bstatic > LDFLAGS_vmlinux-$(CONFIG_RELOCATABLE) := -pie > -LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) > +LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) --orphan-handling=error > > ifeq ($(CONFIG_PPC64),y) > ifeq ($(call cc-option-yn,-mcmodel=medium),y) > diff --git a/arch/powerpc/kernel/vmlinux.lds.S b/arch/powerpc/kernel/vmlinux.lds.S > index 8295f51..9f4d85e 100644 > --- a/arch/powerpc/kernel/vmlinux.lds.S > +++ b/arch/powerpc/kernel/vmlinux.lds.S > @@ -97,7 +97,7 @@ SECTIONS > .text : AT(ADDR(.text) - LOAD_OFFSET) { > ALIGN_FUNCTION(); > /* careful! __ftr_alt_* sections need to be close to .text */ > - *(.text .fixup __ftr_alt_* .ref.text) > + *(.text.hot .text .text.fixup .text.unlikely .fixup __ftr_alt_* .ref.text .sfpr) \ > SCHED_TEXT Is the "\" to pack the sched_text with the rest of .text? > CPUIDLE_TEXT > LOCK_TEXT > @@ -256,7 +256,9 @@ SECTIONS > .data : AT(ADDR(.data) - LOAD_OFFSET) { > DATA_DATA > *(.sdata) > + *(.sdata2) > *(.got.plt) *(.got) > + *(.plt) > } > #else > .data : AT(ADDR(.data) - LOAD_OFFSET) { > @@ -317,6 +319,16 @@ SECTIONS > _end = . ; > PROVIDE32 (end = .); > > - /* Sections to be discarded. */ > + STABS_DEBUG > + > + DWARF_DEBUG > + Are the debug sections discarded or is the above comment misplaced? I guess they are discarded because of them being outside of _end at relocation. > DISCARDS > + /DISCARD/ : { > + *(*.EMB.apuinfo) > + *(.glink .iplt .plt .rela* .comment) > + *(.gnu.version*) > + *(.gnu.attributes) > + *(.eh_frame) > + } > } > diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h > index 3e42bcd..264ebb3 100644 > --- a/include/asm-generic/vmlinux.lds.h > +++ b/include/asm-generic/vmlinux.lds.h > @@ -581,6 +581,7 @@ > #define SBSS(sbss_align) \ > . = ALIGN(sbss_align); \ > .sbss : AT(ADDR(.sbss) - LOAD_OFFSET) { \ > + *(.dynsbss) \ > *(.sbss) \ > *(.scommon) \ > } > @@ -627,6 +628,8 @@ > .debug_str 0 : { *(.debug_str) } \ > .debug_loc 0 : { *(.debug_loc) } \ > .debug_macinfo 0 : { *(.debug_macinfo) } \ > + /* DWARF 3 */ \ > + .debug_ranges 0 : { *(.debug_ranges) } \ > /* SGI/MIPS DWARF 2 extensions */ \ > .debug_weaknames 0 : { *(.debug_weaknames) } \ > .debug_funcnames 0 : { *(.debug_funcnames) } \ > Balbir Singh.
Nicholas Piggin <npiggin@gmail.com> writes: > diff --git a/arch/powerpc/kernel/vmlinux.lds.S b/arch/powerpc/kernel/vmlinux.lds.S > index 8295f51..9f4d85e 100644 > --- a/arch/powerpc/kernel/vmlinux.lds.S > +++ b/arch/powerpc/kernel/vmlinux.lds.S > @@ -317,6 +319,16 @@ SECTIONS > _end = . ; > PROVIDE32 (end = .); > > - /* Sections to be discarded. */ > + STABS_DEBUG > + > + DWARF_DEBUG With DEBUG_INFO=y I'm seeing: /opt/toolchains/6/powerpc/bin/ld: error: unplaced orphan section `.debug_addr' from `arch/powerpc/kernel/prom_init.o'. /opt/toolchains/6/powerpc/bin/ld: error: unplaced orphan section `.debug_gnu_pubnames' from `arch/powerpc/kernel/prom_init.o'. /opt/toolchains/6/powerpc/bin/ld: error: unplaced orphan section `.debug_gnu_pubtypes' from `arch/powerpc/kernel/prom_init.o'. And many more for the same sections from other files. cheers
On Fri, 14 Oct 2016 13:35:54 +1100 Balbir Singh <bsingharora@gmail.com> wrote: > On 12/10/16 18:00, Nicholas Piggin wrote: > > Add --orphan-handling=error to final link flags. This ensures we have to > > handle all sections. This would have caught subtle breakage such as > > 7de3b27bac47da9de08409df1d69664acbb72197 at build-time. > > > > Also bring some wayward sections into the fold: > > - .text.hot and .text.unlikely are compiler generated sections. > > - .sfpr is a linker generated section for register save functions. > > - .sdata2, .dynsbss, .plt are used by PPC32 > > - We previously did not specify DWARF_DEBUG or STABS_DEBUG > > - DWARF_DEBUG did not include DWARF3 .debug_ranges > > - A number of sections are unused. > > > > I don't know if I've exactly got everything right here, particularly > > with ppc32, so would appreciate people casting their eye over it. > > > > Signed-off-by: Nicholas Piggin <npiggin@gmail.com> > > --- > > arch/powerpc/Makefile | 2 +- > > arch/powerpc/kernel/vmlinux.lds.S | 16 ++++++++++++++-- > > include/asm-generic/vmlinux.lds.h | 3 +++ > > 3 files changed, 18 insertions(+), 3 deletions(-) > > > > diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile > > index 50d020a..a3f2784 100644 > > --- a/arch/powerpc/Makefile > > +++ b/arch/powerpc/Makefile > > @@ -90,7 +90,7 @@ endif > > > > LDFLAGS_vmlinux-y := -Bstatic > > LDFLAGS_vmlinux-$(CONFIG_RELOCATABLE) := -pie > > -LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) > > +LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) --orphan-handling=error > > > > ifeq ($(CONFIG_PPC64),y) > > ifeq ($(call cc-option-yn,-mcmodel=medium),y) > > diff --git a/arch/powerpc/kernel/vmlinux.lds.S b/arch/powerpc/kernel/vmlinux.lds.S > > index 8295f51..9f4d85e 100644 > > --- a/arch/powerpc/kernel/vmlinux.lds.S > > +++ b/arch/powerpc/kernel/vmlinux.lds.S > > @@ -97,7 +97,7 @@ SECTIONS > > .text : AT(ADDR(.text) - LOAD_OFFSET) { > > ALIGN_FUNCTION(); > > /* careful! __ftr_alt_* sections need to be close to .text */ > > - *(.text .fixup __ftr_alt_* .ref.text) > > + *(.text.hot .text .text.fixup .text.unlikely .fixup __ftr_alt_* .ref.text .sfpr) \ > > SCHED_TEXT > > Is the "\" to pack the sched_text with the rest of .text? Actually it looks like a typo that slipped in. Good catch. > > @@ -256,7 +256,9 @@ SECTIONS > > .data : AT(ADDR(.data) - LOAD_OFFSET) { > > DATA_DATA > > *(.sdata) > > + *(.sdata2) > > *(.got.plt) *(.got) > > + *(.plt) > > } > > #else > > .data : AT(ADDR(.data) - LOAD_OFFSET) { > > @@ -317,6 +319,16 @@ SECTIONS > > _end = . ; > > PROVIDE32 (end = .); > > > > - /* Sections to be discarded. */ > > + STABS_DEBUG > > + > > + DWARF_DEBUG > > + > > Are the debug sections discarded or is the above > comment misplaced? I guess they are discarded because of them > being outside of _end at relocation. I just got rid of the comment. Having it right above "DISCARDS" seemed redundant. I believe the debug sections will be linked into the file, because they were not put in the /DISCARD/ output. But they will be outside the _end symbol. Thanks, Nick
From: Michael Ellerman > Sent: 14 October 2016 01:46 ... > > +LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) --orphan-handling=error > > At least some old(er) toolchains don't support that: > > /opt/cross/kisskb/gcc-4.6.3-nolibc/powerpc-linux/bin/powerpc-linux-ld: unrecognized option '-- > orphan-handling=error' > > So we'll need an ld-option wrapper around it. In the past I've caused orphan sections to error by assigning them to the same address as something that exists. Works with all linkers, even if the error message isn't as useful. David
David Laight <David.Laight@ACULAB.COM> writes: > From: Michael Ellerman >> Sent: 14 October 2016 01:46 > ... >> > +LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) --orphan-handling=error >> >> At least some old(er) toolchains don't support that: >> >> /opt/cross/kisskb/gcc-4.6.3-nolibc/powerpc-linux/bin/powerpc-linux-ld: unrecognized option '-- >> orphan-handling=error' >> >> So we'll need an ld-option wrapper around it. > > In the past I've caused orphan sections to error by assigning them > to the same address as something that exists. > Works with all linkers, even if the error message isn't as useful. How do you assign them an address without knowing what they are? With a wild card rule? Anyway I think I'm happy for this to only work on newer toolchains. cheers
From: Michael Ellerman [mailto:mpe@ellerman.id.au] > > In the past I've caused orphan sections to error by assigning them > > to the same address as something that exists. > > Works with all linkers, even if the error message isn't as useful. > > How do you assign them an address without knowing what they are? With a > wild card rule? Right at the end of the ldscript I have: /* Anything from any unexpected section ends up here and * generates an error because this overlaps another section */ unwanted _gp : { *(*) } } David
diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile index 50d020a..a3f2784 100644 --- a/arch/powerpc/Makefile +++ b/arch/powerpc/Makefile @@ -90,7 +90,7 @@ endif LDFLAGS_vmlinux-y := -Bstatic LDFLAGS_vmlinux-$(CONFIG_RELOCATABLE) := -pie -LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) +LDFLAGS_vmlinux := $(LDFLAGS_vmlinux-y) --orphan-handling=error ifeq ($(CONFIG_PPC64),y) ifeq ($(call cc-option-yn,-mcmodel=medium),y) diff --git a/arch/powerpc/kernel/vmlinux.lds.S b/arch/powerpc/kernel/vmlinux.lds.S index 8295f51..9f4d85e 100644 --- a/arch/powerpc/kernel/vmlinux.lds.S +++ b/arch/powerpc/kernel/vmlinux.lds.S @@ -97,7 +97,7 @@ SECTIONS .text : AT(ADDR(.text) - LOAD_OFFSET) { ALIGN_FUNCTION(); /* careful! __ftr_alt_* sections need to be close to .text */ - *(.text .fixup __ftr_alt_* .ref.text) + *(.text.hot .text .text.fixup .text.unlikely .fixup __ftr_alt_* .ref.text .sfpr) \ SCHED_TEXT CPUIDLE_TEXT LOCK_TEXT @@ -256,7 +256,9 @@ SECTIONS .data : AT(ADDR(.data) - LOAD_OFFSET) { DATA_DATA *(.sdata) + *(.sdata2) *(.got.plt) *(.got) + *(.plt) } #else .data : AT(ADDR(.data) - LOAD_OFFSET) { @@ -317,6 +319,16 @@ SECTIONS _end = . ; PROVIDE32 (end = .); - /* Sections to be discarded. */ + STABS_DEBUG + + DWARF_DEBUG + DISCARDS + /DISCARD/ : { + *(*.EMB.apuinfo) + *(.glink .iplt .plt .rela* .comment) + *(.gnu.version*) + *(.gnu.attributes) + *(.eh_frame) + } } diff --git a/include/asm-generic/vmlinux.lds.h b/include/asm-generic/vmlinux.lds.h index 3e42bcd..264ebb3 100644 --- a/include/asm-generic/vmlinux.lds.h +++ b/include/asm-generic/vmlinux.lds.h @@ -581,6 +581,7 @@ #define SBSS(sbss_align) \ . = ALIGN(sbss_align); \ .sbss : AT(ADDR(.sbss) - LOAD_OFFSET) { \ + *(.dynsbss) \ *(.sbss) \ *(.scommon) \ } @@ -627,6 +628,8 @@ .debug_str 0 : { *(.debug_str) } \ .debug_loc 0 : { *(.debug_loc) } \ .debug_macinfo 0 : { *(.debug_macinfo) } \ + /* DWARF 3 */ \ + .debug_ranges 0 : { *(.debug_ranges) } \ /* SGI/MIPS DWARF 2 extensions */ \ .debug_weaknames 0 : { *(.debug_weaknames) } \ .debug_funcnames 0 : { *(.debug_funcnames) } \
Add --orphan-handling=error to final link flags. This ensures we have to handle all sections. This would have caught subtle breakage such as 7de3b27bac47da9de08409df1d69664acbb72197 at build-time. Also bring some wayward sections into the fold: - .text.hot and .text.unlikely are compiler generated sections. - .sfpr is a linker generated section for register save functions. - .sdata2, .dynsbss, .plt are used by PPC32 - We previously did not specify DWARF_DEBUG or STABS_DEBUG - DWARF_DEBUG did not include DWARF3 .debug_ranges - A number of sections are unused. I don't know if I've exactly got everything right here, particularly with ppc32, so would appreciate people casting their eye over it. Signed-off-by: Nicholas Piggin <npiggin@gmail.com> --- arch/powerpc/Makefile | 2 +- arch/powerpc/kernel/vmlinux.lds.S | 16 ++++++++++++++-- include/asm-generic/vmlinux.lds.h | 3 +++ 3 files changed, 18 insertions(+), 3 deletions(-)