Fixes for clang/lld
cover Fixes for clang/lld
These patches fix some compilation / linking issues with clang & lld.
Bill Wendling (2):
powerpc/wrapper: Add "-z notext" flag to disable diagnostic powerpc/boot: Use clang when CC is clang
arch/powerpc/boot/Makefile | 4 ++++ arch/powerpc/boot/wrapper | 6 ++++-- 2 files changed, 8 insertions(+), 2 deletions(-)
Signature · 1 line
2.29.0.rc1.297.gfa9743e501-goog
PATCH 1/2 powerpc/wrapper: Add "-z notext" flag to disable diagnostic
The "-z notext" flag disables reporting an error if DT_TEXTREL is set.
ld.lld: error: can't create dynamic relocation R_PPC64_ADDR64 against
symbol: _start in readonly segment; recompile object files with -fPIC or pass '-Wl,-z,notext' to allow text relocations in the output
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
defined in
referenced by crt0.o:(.text+0x8) in archivearch/powerpc/boot/wrapper.a
The BFD linker disables this by default (though it's configurable in current versions). LLD enables this by default. So we add the flag to keep LLD from emitting the error.
--- arch/powerpc/boot/wrapper | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-)
arch/powerpc/boot/wrapper +4 -2
@@ -46,6 +46,7 @@ compression=.gz uboot_comp=gzip pie= format= +notext= # cross-compilation prefix CROSS=
@@ -353,6 +354,7 @@ epapr) platformo="$object/pseries-head.o $object/epapr.o $object/epapr-wrapper.o" link_address='0x20000000' pie=-pie + notext='-z notext' ;; mvme5100) platformo="$object/fixed-head.o $object/mvme5100.o"
@@ -493,8 +495,8 @@ if [ "$platform" != "miboot" ]; then text_start="-Ttext $link_address" fi #link everything - ${CROSS}ld -m $format -T $lds $text_start $pie $nodl -o "$ofile" $map \ - $platformo $tmp $object/wrapper.a + ${CROSS}ld -m $format -T $lds $text_start $pie $notext $nodl -o "$ofile" \ + $map $platformo $tmp $object/wrapper.a rm $tmp fi
Signature · 1 line
2.29.0.rc1.297.gfa9743e501-goog
PATCH 2/2 powerpc/boot: Use clang when CC is clang
In reply to Bill WendlingThe gcc compiler may not be available if CC is clang.
--- arch/powerpc/boot/Makefile | 4 ++++ 1 file changed, 4 insertions(+)
arch/powerpc/boot/Makefile +4 -0
@@ -21,7 +21,11 @@ all: $(obj)/zImage ifdef CROSS32_COMPILE +ifdef CONFIG_CC_IS_CLANG + BOOTCC := $(CROSS32_COMPILE)clang +else BOOTCC := $(CROSS32_COMPILE)gcc +endif BOOTAR := $(CROSS32_COMPILE)ar else BOOTCC := $(CC)
Signature · 1 line
2.29.0.rc1.297.gfa9743e501-goog
PATCH 1/3 powerpc/wrapper: add "-z notext" flag to disable diagnostic
The "-z notext" flag disables reporting an error if DT_TEXTREL is set.
ld.lld: error: can't create dynamic relocation R_PPC64_ADDR64 against
symbol: _start in readonly segment; recompile object files with -fPIC or pass '-Wl,-z,notext' to allow text relocations in the output
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
defined in
referenced by crt0.o:(.text+0x8) in archivearch/powerpc/boot/wrapper.a
The BFD linker disables this by default (though it's configurable in current versions). LLD enables this by default. So we add the flag to keep LLD from emitting the error.
--- arch/powerpc/boot/wrapper | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/powerpc/boot/wrapper +3 -1
@@ -46,6 +46,7 @@ compression=.gz uboot_comp=gzip pie= format= +notext= rodynamic= # cross-compilation prefix
@@ -354,6 +355,7 @@ epapr) platformo="$object/pseries-head.o $object/epapr.o $object/epapr-wrapper.o" link_address='0x20000000' pie=-pie + notext='-z notext' rodynamic=$(if ${CROSS}ld -V 2>&1 | grep -q LLD ; then echo "-z rodynamic"; fi) ;; mvme5100)
@@ -495,7 +497,7 @@ if [ "$platform" != "miboot" ]; then text_start="-Ttext $link_address" fi #link everything - ${CROSS}ld -m $format -T $lds $text_start $pie $nodl $rodynamic -o "$ofile" $map \ + ${CROSS}ld -m $format -T $lds $text_start $pie $nodl $rodynamic $notext -o "$ofile" $map \ $platformo $tmp $object/wrapper.a rm $tmp fi
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
cover PPC: fixes for clang support
In reply to Bill WendlingThis series of patches include fixes for clang issues that arose. The "powerpc/64s" patch was "inspired" by a similar patch for ARM:
eb7c11ee3c5ce arm64: alternative: Work around .inst assembler bugs
Bill Wendling (3):
powerpc/wrapper: add "-z notext" flag to disable diagnostic powerpc/boot: Use clang when CC is clang powerpc/64s: feature: work around inline asm issues
arch/powerpc/boot/Makefile | 4 ++++ arch/powerpc/boot/wrapper | 4 +++- arch/powerpc/include/asm/feature-fixups.h | 19 ++++++++++++++----- 3 files changed, 21 insertions(+), 6 deletions(-)
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
v3 cover PPC: fixes for clang support
Note: This is a resend of PPC/clang patches I sent before. The previous series had a bad title, and one of the patches had a typo in it.
This series of patches include fixes for clang issues that arose. The "powerpc/64s" patch was "inspired" by a similar patch for ARM:
eb7c11ee3c5ce arm64: alternative: Work around .inst assembler bugs
Bill Wendling (3):
powerpc/wrapper: add "-z notext" flag to disable diagnostic powerpc/boot: Use clang when CC is clang powerpc/64s: feature: Work around inline asm issues
arch/powerpc/boot/Makefile | 4 ++++ arch/powerpc/boot/wrapper | 4 +++- arch/powerpc/include/asm/feature-fixups.h | 17 +++++++++++++---- 3 files changed, 20 insertions(+), 5 deletions(-)
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
PATCH v3 2/3 powerpc/boot: Use clang when CC is clang
The gcc compiler may not be available if CC is clang.
--- arch/powerpc/boot/Makefile | 4 ++++ 1 file changed, 4 insertions(+)
arch/powerpc/boot/Makefile +4 -0
@@ -21,7 +21,11 @@ all: $(obj)/zImage ifdef CROSS32_COMPILE +ifdef CONFIG_CC_IS_CLANG + BOOTCC := $(CROSS32_COMPILE)clang +else BOOTCC := $(CROSS32_COMPILE)gcc +endif BOOTAR := $(CROSS32_COMPILE)ar else BOOTCC := $(CC)
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
PATCH v3 3/3 powerpc/64s: feature: Work around inline asm issues
In reply to Bill WendlingThe clang toolchain treats inline assembly a bit differently than straight assembly code. In particular, inline assembly doesn't have the complete context available to resolve expressions. This is intentional to avoid divergence in the resulting assembly code.
We can work around this issue by borrowing a workaround done for ARM, i.e. not directly testing the labels themselves, but by moving the current output pointer by a value that should always be zero. If this value is not null, then we will trigger a backward move, which is explicitly forbidden.
--- arch/powerpc/include/asm/feature-fixups.h | 17 +++++++++++++---- 1 file changed, 13 insertions(+), 4 deletions(-)
arch/powerpc/include/asm/feature-fixups.h +13 -4
@@ -36,6 +36,18 @@ label##2: \ .align 2; \ label##3: +/* +* If the .org directive fails, it means that the feature instructions +* are smaller than the alternate instructions. This used to be written +* as +* +* .ifgt (label##4b-label##3b) - (label##2b-label##1b) +* .error "Feature section else case larger than body" +* .endif +* +* but clang's assembler complains about the expression being non-absolute +* when the code appears in an inline assembly statement. +*/ #define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \
@@ -48,12 +60,9 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \ .popsection; -/* CPU feature dependent sections */ #define BEGIN_FTR_SECTION_NESTED (label) START_FTR_SECTION(label)#define BEGIN_FTR_SECTION START_FTR_SECTION(97)
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
Michael Ellerman
reviewerRe: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Hi Bill,
Quoted 40 lines · Bill Wendling
The clang toolchain treats inline assembly a bit differently than straight assembly code. In particular, inline assembly doesn't have the complete context available to resolve expressions. This is intentional to avoid divergence in the resulting assembly code.
We can work around this issue by borrowing a workaround done for ARM, i.e. not directly testing the labels themselves, but by moving the current output pointer by a value that should always be zero. If this value is not null, then we will trigger a backward move, which is explicitly forbidden.
Signed-off-by Bill Wendling <[email protected]>--- arch/powerpc/include/asm/feature-fixups.h | 17 +++++++++++++---- 1 file changed, 13 insertions(+), 4 deletions(-)arch/powerpc/include/asm/feature-fixups.h +12
-0@@ -36,6 +36,18 @@ label##2: \ .align 2; \ label##3: +/* +* If the .org directive fails, it means that the feature instructions +* are smaller than the alternate instructions. This used to be written +* as +* +* .ifgt (label##4b-label##3b) - (label##2b-label##1b) +* .error "Feature section else case larger than body" +* .endif +* +* but clang's assembler complains about the expression being non-absolute +* when the code appears in an inline assembly statement. +*/ #define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \
arch/powerpc/include/asm/feature-fixups.h label##5: \ · line 64
Quoted 8 lines
arch/powerpc/include/asm/feature-fixups.h +1
-3@@ -48,12 +60,9 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \ .popsection;
When I have an oversize alt section this doesn't seem to give me any error using binutils?
If I hard code:
.org . - (1);
It fails as expected.
But if I hard code:
.org . - (1 > 0);
It builds?
cheers
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 11 lines · Michael Ellerman
If I hard code:
.org . - (1);It fails as expected.
But if I hard code:
.org . - (1 > 0);It builds?
"true" (as a result of a comparison) in as is -1, not 1.
Segher
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
Quoted 18 lines · Segher Boessenkool
Quoted 11 lines · Michael Ellerman
If I hard code:
.org . - (1);It fails as expected.
But if I hard code:
.org . - (1 > 0);It builds?
"true" (as a result of a comparison) in as is -1, not 1.
Segher
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
After looking at this, I suspect that the correct change should be:
.org . + ((label##4b-label##3b) > (label##2b-label##1b));
I'm sorry about that. I can submit another version of the patch.
-bw
Quoted 24 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
Quoted 18 lines · Segher Boessenkool
Quoted 11 lines · Michael Ellerman
If I hard code:
.org . - (1);It fails as expected.
But if I hard code:
.org . - (1 > 0);It builds?
"true" (as a result of a comparison) in as is -1, not 1.
Segher
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
In reply to Bill Wendling(Please don't top-post.)
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \
- .error "Feature section else case larger than body"; \
- .endif; \
+ .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \
It should be a + in that last line, not a -. Was this tested?
Segher
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 16 lines · Segher Boessenkool
Quoted 4 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
arch/powerpc/include/asm/feature-fixups.h +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \
Quoted 1 line
It should be a + in that last line, not a -.
I said so in a follow up email.
Quoted 2 lines
Was this tested?
Please don't be insulting. Anyone can make an error.
-bw
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 12 lines · Bill Wendling
Quoted 10 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
Quoted 8 lines
Quoted 6 lines
arch/powerpc/include/asm/feature-fixups.h +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \It should be a + in that last line, not a -.
I said so in a follow up email.
Yeah, and that arrived a second after I pressed "send" :-)
Quoted 3 lines
Quoted 2 lines
Was this tested?
Please don't be insulting. Anyone can make an error.
Absolutely, but it is just a question. It seems you could improve that testing! It helps you yourself most of all ;-)
Segher
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 18 lines · Segher Boessenkool
Quoted 17 lines · Bill Wendling
Quoted 15 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
(quoted diff) +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \
Quoted 6 lines
Quoted 3 lines
Quoted 1 line
It should be a + in that last line, not a -.
I said so in a follow up email.
Yeah, and that arrived a second after I pressed "send" :-)
Michael, I apologize for the churn with these patches. I believe the policy is to resend the match as "v4", correct?
I ran tests with the change above. It compiled with no error. If I switch the labels around to ".org . + ((label##2b-label##1b) > (label##4b-label##3b))", then it fails as expected.
-bw
Michael Ellerman
reviewerRe: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 24 lines · Bill Wendling
Quoted 22 lines · Segher Boessenkool
Quoted 20 lines · Bill Wendling
Quoted 15 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
(quoted diff) +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \Quoted 1 line
It should be a + in that last line, not a -.
I said so in a follow up email.
Quoted 8 lines
Quoted 2 lines
Yeah, and that arrived a second after I pressed "send" :-)
Michael, I apologize for the churn with these patches. I believe the policy is to resend the match as "v4", correct?
I ran tests with the change above. It compiled with no error. If I switch the labels around to ".org . + ((label##2b-label##1b) > (label##4b-label##3b))", then it fails as expected.
I wanted to retain the nicer error reporting for gcc builds, so I did it like this:
arch/powerpc/include/asm/feature-fixups.h +19 -3
@@ -36,6 +36,24 @@ label##2: \ .align 2; \ label##3: + +#ifndef CONFIG_CC_IS_CLANG +#define CHECK_ALT_SIZE (else_size, body_size) \ + .ifgt (else_size) - (body_size); \ + .error "Feature section else case larger than body"; \ + .endif; +#else +/* +* If we use the ifgt syntax above, clang's assembler complains about the +* expression being non-absolute when the code appears in an inline assembly +* statement. +* As a workaround use an .org directive that has no effect if the else case +* instructions are smaller than the body, but fails otherwise. +*/ +#define CHECK_ALT_SIZE (else_size, body_size) \ + .org . + ((else_size) > (body_size)); +#endif +#define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \
@@ -48,9 +66,7 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + CHECK_ALT_SIZE((label##4b-label##3b), (label##2b-label##1b)); \ .popsection;
I've pushed a branch with all your patches applied to:
https://github.com/linuxppc/linux/commits/next-test
Are you able to give that a quick test? It builds clean with clang for me, but we must be using different versions of clang because my branch already builds clean for me even without your patches.
cheers
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 78 lines · Michael Ellerman
Quoted 24 lines · Bill Wendling
Quoted 22 lines · Segher Boessenkool
Quoted 20 lines · Bill Wendling
Quoted 15 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
arch/powerpc/include/asm/feature-fixups.h +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \Quoted 1 line
It should be a + in that last line, not a -.
I said so in a follow up email.
Quoted 8 lines
Quoted 2 lines
Yeah, and that arrived a second after I pressed "send" :-)
Michael, I apologize for the churn with these patches. I believe the policy is to resend the match as "v4", correct?
I ran tests with the change above. It compiled with no error. If I switch the labels around to ".org . + ((label##2b-label##1b) > (label##4b-label##3b))", then it fails as expected.
I wanted to retain the nicer error reporting for gcc builds, so I did it like this:
arch/powerpc/include/asm/feature-fixups.h +19
-3@@ -36,6 +36,24 @@ label##2: \ .align 2; \ label##3: + +#ifndef CONFIG_CC_IS_CLANG +#define CHECK_ALT_SIZE (else_size, body_size) \ + .ifgt (else_size) - (body_size); \ + .error "Feature section else case larger than body"; \ + .endif; +#else +/* +* If we use the ifgt syntax above, clang's assembler complains about the +* expression being non-absolute when the code appears in an inline assembly +* statement. +* As a workaround use an .org directive that has no effect if the else case +* instructions are smaller than the body, but fails otherwise. +*/ +#define CHECK_ALT_SIZE (else_size, body_size) \ + .org . + ((else_size) > (body_size)); +#endif +#define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \@@ -48,9 +66,7 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + CHECK_ALT_SIZE((label##4b-label##3b), (label##2b-label##1b)); \ .popsection;
Quoted 4 lines
I've pushed a branch with all your patches applied to:
This works for me. Thanks!
Quoted 4 lines
Are you able to give that a quick test? It builds clean with clang for me, but we must be using different versions of clang because my branch already builds clean for me even without your patches.
You may need to set LLVM_IAS=1 to get the behavior I'm seeing. That turns on clang's integrated assembler, which I think is disabled by default.
Note that with clang's integrated assembler, arch/powerpc/boot/util.S fails to compile. Alan Modra mentioned that he sent you a patch to "modernize" the file so that clang can compile it.
-bw
Michael Ellerman
reviewerRe: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 79 lines · Bill Wendling
Quoted 78 lines · Michael Ellerman
Quoted 24 lines · Bill Wendling
Quoted 22 lines · Segher Boessenkool
Quoted 20 lines · Bill Wendling
Quoted 15 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
arch/powerpc/include/asm/feature-fixups.h +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \Quoted 1 line
It should be a + in that last line, not a -.
I said so in a follow up email.
Quoted 8 lines
Quoted 2 lines
Yeah, and that arrived a second after I pressed "send" :-)
Michael, I apologize for the churn with these patches. I believe the policy is to resend the match as "v4", correct?
I ran tests with the change above. It compiled with no error. If I switch the labels around to ".org . + ((label##2b-label##1b) > (label##4b-label##3b))", then it fails as expected.
I wanted to retain the nicer error reporting for gcc builds, so I did it like this:
arch/powerpc/include/asm/feature-fixups.h +19
-3@@ -36,6 +36,24 @@ label##2: \ .align 2; \ label##3: + +#ifndef CONFIG_CC_IS_CLANG +#define CHECK_ALT_SIZE (else_size, body_size) \ + .ifgt (else_size) - (body_size); \ + .error "Feature section else case larger than body"; \ + .endif; +#else +/* +* If we use the ifgt syntax above, clang's assembler complains about the +* expression being non-absolute when the code appears in an inline assembly +* statement. +* As a workaround use an .org directive that has no effect if the else case +* instructions are smaller than the body, but fails otherwise. +*/ +#define CHECK_ALT_SIZE (else_size, body_size) \ + .org . + ((else_size) > (body_size)); +#endif +#define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \@@ -48,9 +66,7 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + CHECK_ALT_SIZE((label##4b-label##3b), (label##2b-label##1b)); \ .popsection;
Quoted 5 lines
Quoted 4 lines
I've pushed a branch with all your patches applied to:
This works for me. Thanks!
Great.
Quoted 7 lines
Quoted 4 lines
Are you able to give that a quick test? It builds clean with clang for me, but we must be using different versions of clang because my branch already builds clean for me even without your patches.
You may need to set LLVM_IAS=1 to get the behavior I'm seeing. That turns on clang's integrated assembler, which I think is disabled by default.
Yep that does it.
But then I get:
clang: error: unsupported argument '-mpower4' to option 'Wa,'
clang: error: unsupported argument '-many' to option 'Wa,'
So I guess I'm still missing something?
Quoted 3 lines
Note that with clang's integrated assembler,
arch/powerpc/boot/util.Sfails to compile. Alan Modra mentioned that he sent you a patch to "modernize" the file so that clang can compile it.
Ah you're right he did, it didn't go to patchwork so I missed it. Have grabbed it now.
cheers
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
Quoted 97 lines · Michael Ellerman
Quoted 79 lines · Bill Wendling
Quoted 78 lines · Michael Ellerman
Quoted 24 lines · Bill Wendling
Quoted 22 lines · Segher Boessenkool
Quoted 20 lines · Bill Wendling
Quoted 15 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 2 lines · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison, you'll get a build error.
But that means your patch is the wrong way around?
arch/powerpc/include/asm/feature-fixups.h +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b)); \Quoted 1 line
It should be a + in that last line, not a -.
I said so in a follow up email.
Quoted 8 lines
Quoted 2 lines
Yeah, and that arrived a second after I pressed "send" :-)
Michael, I apologize for the churn with these patches. I believe the policy is to resend the match as "v4", correct?
I ran tests with the change above. It compiled with no error. If I switch the labels around to ".org . + ((label##2b-label##1b) > (label##4b-label##3b))", then it fails as expected.
I wanted to retain the nicer error reporting for gcc builds, so I did it like this:
arch/powerpc/include/asm/feature-fixups.h +19
-3@@ -36,6 +36,24 @@ label##2: \ .align 2; \ label##3: + +#ifndef CONFIG_CC_IS_CLANG +#define CHECK_ALT_SIZE (else_size, body_size) \ + .ifgt (else_size) - (body_size); \ + .error "Feature section else case larger than body"; \ + .endif; +#else +/* +* If we use the ifgt syntax above, clang's assembler complains about the +* expression being non-absolute when the code appears in an inline assembly +* statement. +* As a workaround use an .org directive that has no effect if the else case +* instructions are smaller than the body, but fails otherwise. +*/ +#define CHECK_ALT_SIZE (else_size, body_size) \ + .org . + ((else_size) > (body_size)); +#endif +#define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \@@ -48,9 +66,7 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + CHECK_ALT_SIZE((label##4b-label##3b), (label##2b-label##1b)); \ .popsection;Quoted 5 lines
Quoted 4 lines
I've pushed a branch with all your patches applied to:
This works for me. Thanks!
Great.
Quoted 7 lines
Quoted 4 lines
Are you able to give that a quick test? It builds clean with clang for me, but we must be using different versions of clang because my branch already builds clean for me even without your patches.
You may need to set LLVM_IAS=1 to get the behavior I'm seeing. That turns on clang's integrated assembler, which I think is disabled by default.
Quoted 8 lines
Yep that does it.
But then I get:
clang: error: unsupported argument '-mpower4' to option 'Wa,'
clang: error: unsupported argument '-many' to option 'Wa,'So I guess I'm still missing something?
[Resent, because my previous email went out as non-plain text.]
I've seen that too. I'm not entirely sure what's causing it, but I'll look into it. I've got a backlog of things to work on still. :-) It's probably a clang issue. There's another one that came up having to do with the format of some PPC instructions. I have a clang fix on review for those.
Quoted 7 lines
Quoted 3 lines
Note that with clang's integrated assembler,
arch/powerpc/boot/util.Sfails to compile. Alan Modra mentioned that he sent you a patch to "modernize" the file so that clang can compile it.Ah you're right he did, it didn't go to patchwork so I missed it. Have grabbed it now.
Thanks!
-bw
Re: [PATCH v3 3/3] powerpc/64s: feature: Work around inline asm issues
In reply to Michael EllermanQuoted 105 lines · Michael Ellerman
Quoted 1 line · Bill Wendling
On Mon, Nov 23, 2020 at 7:44 PM Michael Ellerman <[email protected]>
Quoted 4 lines · wrote:
Quoted 4 lines
Quoted 3 lines · Bill Wendling
Quoted 1 line · Segher Boessenkool
On Mon, Nov 23, 2020 at 12:01:01PM -0800, Bill Wendling wrote:
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines · Segher Boessenkool
Quoted 3 lines
Quoted 1 line · Segher Boessenkool
"true" (as a result of a comparison) in as is -1, not 1.
Quoted 1 line · Bill Wendling
What Segher said. :-) Also, if you reverse the comparison,
you'll get
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
arch/powerpc/include/asm/feature-fixups.h label##5: \
Quoted 8 lines
Quoted 1 line
a build error.
But that means your patch is the wrong way around?
arch/powerpc/include/asm/feature-fixups.h +1
-3- .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + .org . - ((label##4b-label##3b) >(label##2b-label##1b)); \
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
It should be a + in that last line, not a -.
I said so in a follow up email.
Yeah, and that arrived a second after I pressed "send" :-)
Michael, I apologize for the churn with these patches. I believe the policy is to resend the match as "v4", correct?
Quoted 8 lines
Review
Quoted 8 lines
Quoted 3 lines
I ran tests with the change above. It compiled with no error. If I switch the labels around to ".org . + ((label##2b-label##1b) > (label##4b-label##3b))", then it fails as expected.
I wanted to retain the nicer error reporting for gcc builds, so I did it like this:
(quoted diff) +0
-0
b/arch/powerpc/include/asm/feature-fixups.hQuoted 4 lines
arch/powerpc/include/asm/feature-fixups.h label##2:
Quoted 4 lines
index
b0af97add751..c4ad33074df5100644arch/powerpc/include/asm/feature-fixups.h +0
-0@@ -36,6 +36,24 @@ label##2:\
Quoted 12 lines
arch/powerpc/include/asm/feature-fixups.h label##2: \
Quoted 12 lines
arch/powerpc/include/asm/feature-fixups.h +9
-0.align 2; \ label##3: + +#ifndef CONFIG_CC_IS_CLANG +#define CHECK_ALT_SIZE (else_size, body_size) \ + .ifgt (else_size) - (body_size); \ + .error "Feature section else case larger than body"; \ + .endif; +#else +/* +* If we use the ifgt syntax above, clang's assembler complains about the
Quoted 1 line
arch/powerpc/include/asm/feature-fixups.h label##2: · line 48
Quoted 1 line
arch/powerpc/include/asm/feature-fixups.h +1
-0+* expression being non-absolute when the code appears in an inline assembly
Quoted 2 lines
arch/powerpc/include/asm/feature-fixups.h label##2: · line 50
Quoted 2 lines
arch/powerpc/include/asm/feature-fixups.h +2
-0+* statement. +* As a workaround use an .org directive that has no effect if the else case
Quoted 10 lines
arch/powerpc/include/asm/feature-fixups.h label##5:
Quoted 10 lines
arch/powerpc/include/asm/feature-fixups.h +6
-0+* instructions are smaller than the body, but fails otherwise. +*/ +#define CHECK_ALT_SIZE (else_size, body_size) \ + .org . + ((else_size) > (body_size)); +#endif +#define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \@@ -48,9 +66,7 @@ label##5:\Quoted 11 lines
Quoted 11 lines
arch/powerpc/include/asm/feature-fixups.h +1
-3FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ + CHECK_ALT_SIZE((label##4b-label##3b), (label##2b-label##1b)); \ .popsection;Quoted 5 lines
Quoted 4 lines
I've pushed a branch with all your patches applied to:
This works for me. Thanks!
Great.
Quoted 7 lines
Quoted 4 lines
Are you able to give that a quick test? It builds clean with clang for me, but we must be using different versions of clang because my branch already builds clean for me even without your patches.
You may need to set LLVM_IAS=1 to get the behavior I'm seeing. That turns on clang's integrated assembler, which I think is disabled by default.
Quoted 8 lines
Yep that does it.
But then I get:
clang: error: unsupported argument '-mpower4' to option 'Wa,'
clang: error: unsupported argument '-many' to option 'Wa,'So I guess I'm still missing something?
I've seen that too. I'm not entirely sure what's causing it, but I'll look into it. I've got a backlog of things to work on still. :-) It's probably a clang issue. There's another one that came up having to do with the format of some PPC instructions. I have a clang fix on review for those.
Quoted 7 lines
Note that with clang's integrated assembler,
arch/powerpc/boot/util.SQuoted 2 lines
fails to compile. Alan Modra mentioned that he sent you a patch to "modernize" the file so that clang can compile it.
Ah you're right he did, it didn't go to patchwork so I missed it. Have grabbed it now.
Thanks!
-bw
Quoted 1 line
PATCH 1/3 powerpc/wrapper: add "-z notext" flag to disable diagnostic
In reply to Bill WendlingThe "-z notext" flag disables reporting an error if DT_TEXTREL is set.
ld.lld: error: can't create dynamic relocation R_PPC64_ADDR64 against
symbol: _start in readonly segment; recompile object files with -fPIC or pass '-Wl,-z,notext' to allow text relocations in the output
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
defined in
referenced by crt0.o:(.text+0x8) in archivearch/powerpc/boot/wrapper.a
The BFD linker disables this by default (though it's configurable in current versions). LLD enables this by default. So we add the flag to keep LLD from emitting the error.
--- arch/powerpc/boot/wrapper | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
arch/powerpc/boot/wrapper +3 -1
@@ -46,6 +46,7 @@ compression=.gz uboot_comp=gzip pie= format= +notext= rodynamic= # cross-compilation prefix
@@ -354,6 +355,7 @@ epapr) platformo="$object/pseries-head.o $object/epapr.o $object/epapr-wrapper.o" link_address='0x20000000' pie=-pie + notext='-z notext' rodynamic=$(if ${CROSS}ld -V 2>&1 | grep -q LLD ; then echo "-z rodynamic"; fi) ;; mvme5100)
@@ -495,7 +497,7 @@ if [ "$platform" != "miboot" ]; then text_start="-Ttext $link_address" fi #link everything - ${CROSS}ld -m $format -T $lds $text_start $pie $nodl $rodynamic -o "$ofile" $map \ + ${CROSS}ld -m $format -T $lds $text_start $pie $nodl $rodynamic $notext -o "$ofile" $map \ $platformo $tmp $object/wrapper.a rm $tmp fi
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
PATCH 2/3 powerpc/boot: Use clang when CC is clang
In reply to Bill WendlingThe gcc compiler may not be available if CC is clang.
--- arch/powerpc/boot/Makefile | 4 ++++ 1 file changed, 4 insertions(+)
arch/powerpc/boot/Makefile +4 -0
@@ -21,7 +21,11 @@ all: $(obj)/zImage ifdef CROSS32_COMPILE +ifdef CONFIG_CC_IS_CLANG + BOOTCC := $(CROSS32_COMPILE)clang +else BOOTCC := $(CROSS32_COMPILE)gcc +endif BOOTAR := $(CROSS32_COMPILE)ar else BOOTCC := $(CC)
Signature · 1 line
2.29.2.454.gaff20da3a2-goog
PATCH 3/3 powerpc/64s: feature: work around inline asm issues
In reply to Bill WendlingThe clang toolchain treats inline assembly a bit differently than straight assembly code. In particular, inline assembly doesn't have the complete context available to resolve expressions. This is intentional to avoid divergence in the resulting assembly code.
We can work around this issue by borrowing a workaround done for ARM, i.e. not directly testing the labels themselves, but by moving the current output pointer by a value that should always be zero. If this value is not null, then we will trigger a backward move, which is explicitly forbidden.
--- arch/powerpc/include/asm/feature-fixups.h | 19 ++++++++++++++----- 1 file changed, 14 insertions(+), 5 deletions(-)
arch/powerpc/include/asm/feature-fixups.h +14 -5
@@ -36,6 +36,18 @@ label##2: \ .align 2; \ label##3: +/* +* If the .org directive fails, it means that the feature instructions +* are smaller than the alternate instructions. This used to be written +* as +* +* .ifgt (label##4b-label##3b) - (label##2b-label##1b) +* .error "Feature section else case larger than body" +* .endif +* +* but clang's assembler complains about the expression being non-absolute +* when the code appears in an inline assembly statement. +*/ #define MAKE_FTR_SECTION_ENTRY (msk, val, label, sect) \ label##4: \ .popsection; \
@@ -48,11 +60,8 @@ label##5: \ FTR_ENTRY_OFFSET label##2b-label##5b; \ FTR_ENTRY_OFFSET label##3b-label##5b; \ FTR_ENTRY_OFFSET label##4b-label##5b; \ - .ifgt (label##4b- label##3b)-(label##2b- label##1b); \ - .error "Feature section else case larger than body"; \ - .endif; \ - .popsection; - + .popsection; \ + .org . - ((label##4b-label##3b) > (label##2b-label##1b));/* CPU feature dependent sections */ #define BEGIN_FTR_SECTION_NESTED (label) START_FTR_SECTION(label)
Signature · 1 line
2.29.2.454.gaff20da3a2-goog