| Message ID | 20250313114329.284104-11-acarmina@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | Add support for suppressing warning backtraces | expand |
On 3/13/25 04:43, Alessandro Carminati wrote: > From: Guenter Roeck <linux@roeck-us.net> > > Add name of functions triggering warning backtraces to the __bug_table > object section to enable support for suppressing WARNING backtraces. > > To limit image size impact, the pointer to the function name is only added > to the __bug_table section if both CONFIG_KUNIT_SUPPRESS_BACKTRACE and > CONFIG_DEBUG_BUGVERBOSE are enabled. Otherwise, the __func__ assembly > parameter is replaced with a (dummy) NULL parameter to avoid an image size > increase due to unused __func__ entries (this is necessary because > __func__ is not a define but a virtual variable). > > Tested-by: Linux Kernel Functional Testing <lkft@linaro.org> > Acked-by: Dan Carpenter <dan.carpenter@linaro.org> > Cc: Heiko Carstens <hca@linux.ibm.com> > Cc: Vasily Gorbik <gor@linux.ibm.com> > Cc: Alexander Gordeev <agordeev@linux.ibm.com> > Signed-off-by: Guenter Roeck <linux@roeck-us.net> > Signed-off-by: Alessandro Carminati <acarmina@redhat.com> > --- > arch/s390/include/asm/bug.h | 17 ++++++++++++++--- > 1 file changed, 14 insertions(+), 3 deletions(-) > > diff --git a/arch/s390/include/asm/bug.h b/arch/s390/include/asm/bug.h > index c500d45fb465..44d4e9f24ae0 100644 > --- a/arch/s390/include/asm/bug.h > +++ b/arch/s390/include/asm/bug.h > @@ -8,6 +8,15 @@ > > #ifdef CONFIG_DEBUG_BUGVERBOSE > > +#ifdef CONFIG_KUNIT_SUPPRESS_BACKTRACE > +# define HAVE_BUG_FUNCTION > +# define __BUG_FUNC_PTR " .long %0-.\n" > +# define __BUG_FUNC __func__ gcc 7.5.0 on s390 barfs; it doesn't like the use of "__func__" with "%0-." drivers/gpu/drm/bridge/analogix/analogix-i2c-dptx.c: In function 'anx_dp_aux_transfer': ././include/linux/compiler_types.h:492:20: warning: asm operand 0 probably doesn't match constraints I was unable to find an alternate constraint that the compiler would accept. I don't know if the same problem is seen with older compilers on other architectures, or if the problem is relevant in the first place. gcc 10.3.0 and later do not have this problem. I also tried s390 builds with gcc 9.4 and 9.5 but they both crash for unrelated reasons. If this is a concern, the best idea I have is to make KUNIT_SUPPRESS_BACKTRACE depend on, say, depends on CC_IS_CLANG || (CC_IS_GCC && GCC_VERSION >= 100300) A more complex solution might be to define an architecture flag such as HAVE_SUPPRESS_BACKTRACE, make that conditional on the gcc version for s390 only, and make KUNIT_SUPPRESS_BACKTRACE depend on it. Guenter
On Fri, Mar 21, 2025, at 18:05, Guenter Roeck wrote: > On 3/13/25 04:43, Alessandro Carminati wrote: > > gcc 10.3.0 and later do not have this problem. I also tried s390 builds > with gcc 9.4 > and 9.5 but they both crash for unrelated reasons. > > If this is a concern, the best idea I have is to make KUNIT_SUPPRESS_BACKTRACE > depend on, say, > depends on CC_IS_CLANG || (CC_IS_GCC && GCC_VERSION >= 100300) > > A more complex solution might be to define an architecture flag such > as HAVE_SUPPRESS_BACKTRACE, make that conditional on the gcc version > for s390 only, and make KUNIT_SUPPRESS_BACKTRACE depend on it. That is probably fine, there are very few users on s390 that build their own kernels, and they likely all have modern compilers anyway. I should still send a patch to raise the minimum compiler version to gcc-8.1, but unfortunately that is not enough here. Arnd
Hello Guenter, Sorry for being late to the party. On Fri, Mar 21, 2025 at 6:06 PM Guenter Roeck <linux@roeck-us.net> wrote: > > On 3/13/25 04:43, Alessandro Carminati wrote: > > From: Guenter Roeck <linux@roeck-us.net> > > > > Add name of functions triggering warning backtraces to the __bug_table > > object section to enable support for suppressing WARNING backtraces. > > > > To limit image size impact, the pointer to the function name is only added > > to the __bug_table section if both CONFIG_KUNIT_SUPPRESS_BACKTRACE and > > CONFIG_DEBUG_BUGVERBOSE are enabled. Otherwise, the __func__ assembly > > parameter is replaced with a (dummy) NULL parameter to avoid an image size > > increase due to unused __func__ entries (this is necessary because > > __func__ is not a define but a virtual variable). > > > > Tested-by: Linux Kernel Functional Testing <lkft@linaro.org> > > Acked-by: Dan Carpenter <dan.carpenter@linaro.org> > > Cc: Heiko Carstens <hca@linux.ibm.com> > > Cc: Vasily Gorbik <gor@linux.ibm.com> > > Cc: Alexander Gordeev <agordeev@linux.ibm.com> > > Signed-off-by: Guenter Roeck <linux@roeck-us.net> > > Signed-off-by: Alessandro Carminati <acarmina@redhat.com> > > --- > > arch/s390/include/asm/bug.h | 17 ++++++++++++++--- > > 1 file changed, 14 insertions(+), 3 deletions(-) > > > > diff --git a/arch/s390/include/asm/bug.h b/arch/s390/include/asm/bug.h > > index c500d45fb465..44d4e9f24ae0 100644 > > --- a/arch/s390/include/asm/bug.h > > +++ b/arch/s390/include/asm/bug.h > > @@ -8,6 +8,15 @@ > > > > #ifdef CONFIG_DEBUG_BUGVERBOSE > > > > +#ifdef CONFIG_KUNIT_SUPPRESS_BACKTRACE > > +# define HAVE_BUG_FUNCTION > > +# define __BUG_FUNC_PTR " .long %0-.\n" > > +# define __BUG_FUNC __func__ > > gcc 7.5.0 on s390 barfs; it doesn't like the use of "__func__" with "%0-." > > drivers/gpu/drm/bridge/analogix/analogix-i2c-dptx.c: In function 'anx_dp_aux_transfer': > ././include/linux/compiler_types.h:492:20: warning: asm operand 0 probably doesn't match constraints > > I was unable to find an alternate constraint that the compiler would accept. > > I don't know if the same problem is seen with older compilers on other architectures, > or if the problem is relevant in the first place. > > gcc 10.3.0 and later do not have this problem. I also tried s390 builds with gcc 9.4 > and 9.5 but they both crash for unrelated reasons. > > If this is a concern, the best idea I have is to make KUNIT_SUPPRESS_BACKTRACE > depend on, say, > depends on CC_IS_CLANG || (CC_IS_GCC && GCC_VERSION >= 100300) > > A more complex solution might be to define an architecture flag such > as HAVE_SUPPRESS_BACKTRACE, make that conditional on the gcc version > for s390 only, and make KUNIT_SUPPRESS_BACKTRACE depend on it. I've spent some time trying to better define the problem. Although it may seem trivial, the old compiler simply doesn't work—I believe the issue is a bit more complex. So, let me share some code and then comment on it. $ cat bug-s390.c #include "bug_entry.h" #define asm_inline asm __inline # define __BUG_FUNC_PTR " .long %0-.\n" # define __BUG_FUNC __func__ #define __EMIT_BUG(x) do { \ asm_inline volatile( \ "0: mc 0,0\n" \ ".section .rodata.str,\"aMS\",@progbits,1\n" \ "1: .asciz \""__FILE__"\"\n" \ ".previous\n" \ ".section __bug_table,\"aw\"\n" \ "2: .long 0b-.\n" \ " .long 1b-.\n" \ __BUG_FUNC_PTR \ " .short %1,%2\n" \ " .org 2b+%3\n" \ ".previous\n" \ : : "i" (__BUG_FUNC), \ "i" (__LINE__), \ "i" (x), \ "i" (sizeof(struct bug_entry))); \ } while (0) #define BUG() do { \ __EMIT_BUG(0); \ } while (0) void f1(){ BUG(); } void f2(){ BUG(); } int main() { BUG(); f1(); f2(); return 0; } $ # This is a stripped version of the s390x code for bug $ ~/x-tools/s390x-ibm-linux-gnu_14/bin/s390x-ibm-linux-gnu-gcc -v Using built-in specs. COLLECT_GCC=/home/alessandro/x-tools/s390x-ibm-linux-gnu_14/bin/s390x-ibm-linux-gnu-gcc COLLECT_LTO_WRAPPER=/home/alessandro/x-tools/s390x-ibm-linux-gnu_14/bin/../libexec/gcc/s390x-ibm-linux-gnu/14.2.0/lto-wrapper Target: s390x-ibm-linux-gnu Configured with: /home/alessandro/src/s390x-toolchain/.build/s390x-ibm-linux-gnu/src/gcc/configure --build=x86_64-build_pc-linux-gnu --host=x86_64-build_pc-linux-gnu --target=s390x-ibm-linux-gnu --prefix=/home/alessandro/x-tools/s390x-ibm-linux-gnu --exec_prefix=/home/alessandro/x-tools/s390x-ibm-linux-gnu --with-sysroot=/home/alessandro/x-tools/s390x-ibm-linux-gnu/s390x-ibm-linux-gnu/sysroot --enable-languages=c,c++ --with-pkgversion='crosstool-NG 1.27.0.18_7458341' --enable-__cxa_atexit --disable-libmudflap --disable-libgomp --disable-libssp --disable-libquadmath --disable-libquadmath-support --disable-libsanitizer --disable-libmpx --with-gmp=/home/alessandro/src/s390x-toolchain/.build/s390x-ibm-linux-gnu/buildtools --with-mpfr=/home/alessandro/src/s390x-toolchain/.build/s390x-ibm-linux-gnu/buildtools --with-mpc=/home/alessandro/src/s390x-toolchain/.build/s390x-ibm-linux-gnu/buildtools --with-isl=/home/alessandro/src/s390x-toolchain/.build/s390x-ibm-linux-gnu/buildtools --enable-lto --enable-threads=posix --enable-target-optspace --disable-plugin --disable-nls --disable-multilib --with-local-prefix=/home/alessandro/x-tools/s390x-ibm-linux-gnu/s390x-ibm-linux-gnu/sysroot --enable-long-long Thread model: posix Supported LTO compression algorithms: zlib zstd gcc version 14.2.0 (crosstool-NG 1.27.0.18_7458341) $ ~/x-tools/s390x-ibm-linux-gnu_14/bin/s390x-ibm-linux-gnu-gcc -S -m64 bug-s390.c $ ~/x-tools/s390x-ibm-linux-gnu_14/bin/s390x-ibm-linux-gnu-gcc -S -m64 -fPIC bug-s390.c $ ~/x-tools/s390x-ibm-linux-gnu/bin/s390x-ibm-linux-gnu-gcc -v Using built-in specs. COLLECT_GCC=/home/alessandro/x-tools/s390x-ibm-linux-gnu/bin/s390x-ibm-linux-gnu-gcc COLLECT_LTO_WRAPPER=/home/alessandro/x-tools/s390x-ibm-linux-gnu/libexec/gcc/s390x-ibm-linux-gnu/7.5.0/lto-wrapper Target: s390x-ibm-linux-gnu Configured with: /home/alessandro/src/cross-s390/.build/s390x-ibm-linux-gnu/src/gcc/configure --build=x86_64-build_pc-linux-gnu --host=x86_64-build_pc-linux-gnu --target=s390x-ibm-linux-gnu --prefix=/home/alessandro/x-tools/s390x-ibm-linux-gnu --exec_prefix=/home/alessandro/x-tools/s390x-ibm-linux-gnu --with-sysroot=/home/alessandro/x-tools/s390x-ibm-linux-gnu/s390x-ibm-linux-gnu/sysroot --enable-languages=c --with-pkgversion='crosstool-NG 1.27.0.18_7458341' --enable-__cxa_atexit --disable-tm-clone-registry --disable-libmudflap --disable-libgomp --disable-libssp --disable-libquadmath --disable-libquadmath-support --disable-libsanitizer --disable-libmpx --disable-libstdcxx-verbose --with-gmp=/home/alessandro/src/cross-s390/.build/s390x-ibm-linux-gnu/buildtools --with-mpfr=/home/alessandro/src/cross-s390/.build/s390x-ibm-linux-gnu/buildtools --with-mpc=/home/alessandro/src/cross-s390/.build/s390x-ibm-linux-gnu/buildtools --with-isl=/home/alessandro/src/cross-s390/.build/s390x-ibm-linux-gnu/buildtools --enable-lto --without-zstd --enable-threads=posix --enable-target-optspace --disable-plugin --disable-nls --disable-multilib --with-local-prefix=/home/alessandro/x-tools/s390x-ibm-linux-gnu/s390x-ibm-linux-gnu/sysroot --enable-long-long Thread model: posix gcc version 7.5.0 (crosstool-NG 1.27.0.18_7458341) $ ~/x-tools/s390x-ibm-linux-gnu/bin/s390x-ibm-linux-gnu-gcc -S -m64 bug-s390.c $ ~/x-tools/s390x-ibm-linux-gnu/bin/s390x-ibm-linux-gnu-gcc -S -m64 -fPIC bug-s390.c bug-s390.c: In function 'f1': bug-s390.c:2:20: warning: asm operand 0 probably doesn't match constraints #define asm_inline asm __inline ^ bug-s390.c:6:2: note: in expansion of macro 'asm_inline' asm_inline volatile( \ ^~~~~~~~~~ bug-s390.c:25:2: note: in expansion of macro '__EMIT_BUG' __EMIT_BUG(0); \ ^~~~~~~~~~ bug-s390.c:29:2: note: in expansion of macro 'BUG' BUG(); ^~~ bug-s390.c:2:20: error: impossible constraint in 'asm' #define asm_inline asm __inline ^ bug-s390.c:6:2: note: in expansion of macro 'asm_inline' asm_inline volatile( \ ^~~~~~~~~~ bug-s390.c:25:2: note: in expansion of macro '__EMIT_BUG' __EMIT_BUG(0); \ ^~~~~~~~~~ bug-s390.c:29:2: note: in expansion of macro 'BUG' BUG(); ^~~ bug-s390.c: In function 'f2': bug-s390.c:2:20: warning: asm operand 0 probably doesn't match constraints #define asm_inline asm __inline ^ bug-s390.c:6:2: note: in expansion of macro 'asm_inline' asm_inline volatile( \ ^~~~~~~~~~ bug-s390.c:25:2: note: in expansion of macro '__EMIT_BUG' __EMIT_BUG(0); \ ^~~~~~~~~~ bug-s390.c:32:2: note: in expansion of macro 'BUG' BUG(); ^~~ bug-s390.c: In function 'main': bug-s390.c:2:20: warning: asm operand 0 probably doesn't match constraints #define asm_inline asm __inline ^ bug-s390.c:6:2: note: in expansion of macro 'asm_inline' asm_inline volatile( \ ^~~~~~~~~~ bug-s390.c:25:2: note: in expansion of macro '__EMIT_BUG' __EMIT_BUG(0); \ ^~~~~~~~~~ bug-s390.c:35:2: note: in expansion of macro 'BUG' BUG(); ^~~ $ cat linux-6.14-rc7/arch/s390/Makefile| grep "fPIC" KBUILD_AFLAGS_MODULE += -fPIC KBUILD_CFLAGS_MODULE += -fPIC KBUILD_CFLAGS += -fPIC As you can see, the problem is not that the compiler itself doesn't work, but rather that -fPIC introduces some complications. __func__ is a compile-time constant, but this holds true only for traditionally linked code. When compiling position-independent code, this assumption no longer applies. GCC makes significant efforts to handle this, and for several architectures, it manages to solve the problem. However, this is not universally the case. Additionally, -fPIC is not widely used in kernel code... I have only seen it used for VDSO, the x86 boot piggyback decompressor, PowerPC boot, and the s390x architecture. That said, GCC has been mitigating this issue, allowing us to treat a non-compile-time constant as if it were one. A proof of this is that, at least since GCC 11, the s390x version of GCC is able to build this code. Before that... certainly in GCC 7.5 it couldn't. A simple fix would be to restrict usage to GCC versions greater than 11 for s390. The real concern is that we have a latent issue that could be triggered by changes in build settings, as "feature" support varies across GCC versions and architectures. For example, while x86_64 at version 14 seems to work both with and without -fPIC, this is not the case for AArch64, where enabling -fPIC causes failures even in version 14. I'm currently working on a long-term fix for this. > > Guenter >
On Fri, Mar 21, 2025 at 10:05:42PM +0100, Alessandro Carminati wrote: > > > +#ifdef CONFIG_KUNIT_SUPPRESS_BACKTRACE > > > +# define HAVE_BUG_FUNCTION > > > +# define __BUG_FUNC_PTR " .long %0-.\n" > > > +# define __BUG_FUNC __func__ > > > > gcc 7.5.0 on s390 barfs; it doesn't like the use of "__func__" with "%0-." ... > GCC makes significant efforts to handle this, and for several > architectures, it manages to solve the problem. > However, this is not universally the case. > Additionally, -fPIC is not widely used in kernel code... I have only > seen it used for VDSO, the x86 boot piggyback decompressor, PowerPC > boot, and the s390x architecture. > > That said, GCC has been mitigating this issue, allowing us to treat a > non-compile-time constant as if it were one. > A proof of this is that, at least since GCC 11, the s390x version of > GCC is able to build this code. > Before that... certainly in GCC 7.5 it couldn't. > > A simple fix would be to restrict usage to GCC versions greater than > 11 for s390. But please add that dependency only for this new feature for the time being. Right now I would not like to see that s390 is the only architecture (besides parisc) which requires a much higher minimum gcc level than every other architecture. Unless there are specific reasons.
diff --git a/arch/s390/include/asm/bug.h b/arch/s390/include/asm/bug.h index c500d45fb465..44d4e9f24ae0 100644 --- a/arch/s390/include/asm/bug.h +++ b/arch/s390/include/asm/bug.h @@ -8,6 +8,15 @@ #ifdef CONFIG_DEBUG_BUGVERBOSE +#ifdef CONFIG_KUNIT_SUPPRESS_BACKTRACE +# define HAVE_BUG_FUNCTION +# define __BUG_FUNC_PTR " .long %0-.\n" +# define __BUG_FUNC __func__ +#else +# define __BUG_FUNC_PTR +# define __BUG_FUNC NULL +#endif /* CONFIG_KUNIT_SUPPRESS_BACKTRACE */ + #define __EMIT_BUG(x) do { \ asm_inline volatile( \ "0: mc 0,0\n" \ @@ -17,10 +26,12 @@ ".section __bug_table,\"aw\"\n" \ "2: .long 0b-.\n" \ " .long 1b-.\n" \ - " .short %0,%1\n" \ - " .org 2b+%2\n" \ + __BUG_FUNC_PTR \ + " .short %1,%2\n" \ + " .org 2b+%3\n" \ ".previous\n" \ - : : "i" (__LINE__), \ + : : "i" (__BUG_FUNC), \ + "i" (__LINE__), \ "i" (x), \ "i" (sizeof(struct bug_entry))); \ } while (0)