This symbol was causing a build failure with the new toolchain. It looks like it has always been wrong.
The main issue was checking for PIC rather than __PIC__.
Remove all PSEUDO_* macros and ther SYSCALL_ERROR_NAME macro as they are not needed by uclibc-ng, they are used in glibc for building up syscalls there, but not here.
Fixes error: /opt/shorne/software/or1k-linux/bin/../lib/gcc/or1k-linux/9.0.1/../../../../or1k-linux/bin/ld: libc/libc_so.a(or1k_clone.os): pc-relative relocation against dynamic symbol __syscall_error /opt/shorne/software/or1k-linux/bin/../lib/gcc/or1k-linux/9.0.1/../../../../or1k-linux/bin/ld: final link failed: bad value
Signed-off-by: Stafford Horne shorne@gmail.com --- libc/sysdeps/linux/or1k/or1k_clone.S | 8 +++++-- libc/sysdeps/linux/or1k/sysdep.h | 32 ---------------------------- 2 files changed, 6 insertions(+), 34 deletions(-)
diff --git a/libc/sysdeps/linux/or1k/or1k_clone.S b/libc/sysdeps/linux/or1k/or1k_clone.S index a2c16ac9e..2de873a02 100644 --- a/libc/sysdeps/linux/or1k/or1k_clone.S +++ b/libc/sysdeps/linux/or1k/or1k_clone.S @@ -71,7 +71,11 @@ L(oldpid): l.ori r3, r11, 0
L(error): - l.j SYSCALL_ERROR_NAME +#ifdef __PIC__ + l.j plt(__syscall_error) +#else + l.j __syscall_error +#endif l.ori r3,r11,0
-PSEUDO_END (__or1k_clone) +END (__or1k_clone) diff --git a/libc/sysdeps/linux/or1k/sysdep.h b/libc/sysdeps/linux/or1k/sysdep.h index 4dba3f2a0..782981c55 100644 --- a/libc/sysdeps/linux/or1k/sysdep.h +++ b/libc/sysdeps/linux/or1k/sysdep.h @@ -43,38 +43,6 @@ License along with the GNU C Library; if not, see l.sys 1; \ l.nop
-#define PSEUDO(name, syscall_name, args) \ - ENTRY (name); \ - DO_CALL(syscall_name); \ - /* if -4096 < ret < 0 holds, it's an error */ \ - l.sfgeui r11, 0xf001; \ - l.bf L(pseudo_end); \ - l.nop - -#define PSEUDO_NOERRNO(name, syscall_name, args) \ - ENTRY (name); \ - DO_CALL(syscall_name) - -#define PSEUDO_END(name) \ -L(pseudo_end): \ - l.j SYSCALL_ERROR_NAME; \ - l.ori r3,r11,0; \ - END (name) - -#define PSEUDO_END_NOERRNO(name) \ - END (name) - -#ifndef PIC -/* For static code, on error jump to __syscall_error directly. */ -# define SYSCALL_ERROR_NAME __syscall_error -#elif NOT_IN_libc -/* Use the internal name for libc/libpthread shared objects. */ -# define SYSCALL_ERROR_NAME __GI___syscall_error -#else -/* Otherwise, on error do a full PLT jump. */ -# define SYSCALL_ERROR_NAME plt(__syscall_error) -#endif - /* Make use of .size directive. */ #define ASM_SIZE_DIRECTIVE(name) .size name,.-name;
Hi Stafford,
Le 26/08/2019 à 22:58, Stafford Horne a écrit :
This symbol was causing a build failure with the new toolchain. It looks like it has always been wrong.
The main issue was checking for PIC rather than __PIC__.
Indeed, I also noticed this. I've made a similar/simpler patch but the system doesn't boot (init program is stuck after "Run /init as init process")
Sadly, the system doesn't boot even with your patch. Have you successfully runtime tested uClibc-ng build with your patch?
Best regards, Romain
Remove all PSEUDO_* macros and ther SYSCALL_ERROR_NAME macro as they are not needed by uclibc-ng, they are used in glibc for building up syscalls there, but not here.
Fixes error: /opt/shorne/software/or1k-linux/bin/../lib/gcc/or1k-linux/9.0.1/../../../../or1k-linux/bin/ld: libc/libc_so.a(or1k_clone.os): pc-relative relocation against dynamic symbol __syscall_error /opt/shorne/software/or1k-linux/bin/../lib/gcc/or1k-linux/9.0.1/../../../../or1k-linux/bin/ld: final link failed: bad value
Signed-off-by: Stafford Horne shorne@gmail.com
libc/sysdeps/linux/or1k/or1k_clone.S | 8 +++++-- libc/sysdeps/linux/or1k/sysdep.h | 32 ---------------------------- 2 files changed, 6 insertions(+), 34 deletions(-)
diff --git a/libc/sysdeps/linux/or1k/or1k_clone.S b/libc/sysdeps/linux/or1k/or1k_clone.S index a2c16ac9e..2de873a02 100644 --- a/libc/sysdeps/linux/or1k/or1k_clone.S +++ b/libc/sysdeps/linux/or1k/or1k_clone.S @@ -71,7 +71,11 @@ L(oldpid): l.ori r3, r11, 0
L(error):
- l.j SYSCALL_ERROR_NAME
+#ifdef __PIC__
- l.j plt(__syscall_error)
+#else
- l.j __syscall_error
+#endif l.ori r3,r11,0
-PSEUDO_END (__or1k_clone) +END (__or1k_clone) diff --git a/libc/sysdeps/linux/or1k/sysdep.h b/libc/sysdeps/linux/or1k/sysdep.h index 4dba3f2a0..782981c55 100644 --- a/libc/sysdeps/linux/or1k/sysdep.h +++ b/libc/sysdeps/linux/or1k/sysdep.h @@ -43,38 +43,6 @@ License along with the GNU C Library; if not, see l.sys 1; \ l.nop
-#define PSEUDO(name, syscall_name, args) \
- ENTRY (name); \
- DO_CALL(syscall_name); \
- /* if -4096 < ret < 0 holds, it's an error */ \
- l.sfgeui r11, 0xf001; \
- l.bf L(pseudo_end); \
- l.nop
-#define PSEUDO_NOERRNO(name, syscall_name, args) \
- ENTRY (name); \
- DO_CALL(syscall_name)
-#define PSEUDO_END(name) \ -L(pseudo_end): \
- l.j SYSCALL_ERROR_NAME; \
- l.ori r3,r11,0; \
- END (name)
-#define PSEUDO_END_NOERRNO(name) \
- END (name)
-#ifndef PIC -/* For static code, on error jump to __syscall_error directly. */ -# define SYSCALL_ERROR_NAME __syscall_error -#elif NOT_IN_libc -/* Use the internal name for libc/libpthread shared objects. */ -# define SYSCALL_ERROR_NAME __GI___syscall_error -#else -/* Otherwise, on error do a full PLT jump. */ -# define SYSCALL_ERROR_NAME plt(__syscall_error) -#endif
/* Make use of .size directive. */ #define ASM_SIZE_DIRECTIVE(name) .size name,.-name;
On Tue, Aug 27, 2019 at 09:37:32PM +0200, Romain Naour wrote:
Hi Stafford,
Le 26/08/2019 à 22:58, Stafford Horne a écrit :
This symbol was causing a build failure with the new toolchain. It looks like it has always been wrong.
The main issue was checking for PIC rather than __PIC__.
Indeed, I also noticed this. I've made a similar/simpler patch but the system doesn't boot (init program is stuck after "Run /init as init process")
Sadly, the system doesn't boot even with your patch. Have you successfully runtime tested uClibc-ng build with your patch?
Hello,
No, I haven't got it booting.
I have only been able to successfully compile using this. I don't have an environment setup right now to boot a uclibc-ng target. What are you using qemu, or1ksim, fpga?
In terms of possible issues, I did just fix 2 bugs that caused glibc to not build: - gcc bug - could not read GOT due to clobbering r9 https://gcc.gnu.org/ml/gcc-patches/2019-08/msg01549.html - binutils bug - issue with PLT entries and lazy symbol resolution (Already upstream) https://sourceware.org/ml/binutils/2019-08/msg00214.html
-Stafford
Hi, Stafford Horne wrote,
On Tue, Aug 27, 2019 at 09:37:32PM +0200, Romain Naour wrote:
Hi Stafford,
Le 26/08/2019 à 22:58, Stafford Horne a écrit :
This symbol was causing a build failure with the new toolchain. It looks like it has always been wrong.
The main issue was checking for PIC rather than __PIC__.
Indeed, I also noticed this. I've made a similar/simpler patch but the system doesn't boot (init program is stuck after "Run /init as init process")
Sadly, the system doesn't boot even with your patch. Have you successfully runtime tested uClibc-ng build with your patch?
Hello,
No, I haven't got it booting.
I have only been able to successfully compile using this. I don't have an environment setup right now to boot a uclibc-ng target. What are you using qemu, or1ksim, fpga?
In terms of possible issues, I did just fix 2 bugs that caused glibc to not build:
- gcc bug - could not read GOT due to clobbering r9 https://gcc.gnu.org/ml/gcc-patches/2019-08/msg01549.html
- binutils bug - issue with PLT entries and lazy symbol resolution (Already upstream) https://sourceware.org/ml/binutils/2019-08/msg00214.html
Patch applied and pushed. With the binutils/gcc patch the system boots fine in Qemu. The testsuite does show also good results, just 6-7 test failures.
best regards Waldemar