| Message ID | 20190409212517.7321-10-linux@rasmusvillemoes.dk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | implement DYNAMIC_DEBUG_RELATIVE_POINTERS | expand |
On Tue, Apr 9, 2019 at 11:26 PM Rasmus Villemoes <linux@rasmusvillemoes.dk> wrote: > > Similar to GENERIC_BUG_RELATIVE_POINTERS, one can now relativize the > four const char* members of struct _ddebug, thus saving 16 bytes per > instance (one for each pr_debug(), dev_debug() etc. in a > CONFIG_DYNAMIC_DEBUG kernel). The asm-generic implementation seems to > work out-of-the-box, though this is only compile-tested. > > Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk> This causes a build regression when compiling with clang, see http://staging-storage.kernelci.org/next-clang/master/next-20190424/arm64/allmodconfig/clang-8/build.log ../init/initramfs.c:477:3: error: invalid operand in inline asm: '.ifndef __UNIQUE_ID_ddebug18.pushsection __verbose,"aw".type __UNIQUE_ID_ddebug18, STT_OBJECT.size __UNIQUE_ID_ddebug18, ${5:c}1:__UNIQUE_ID_ddebug18: .int ${0:c} - 1b /* _ddebug::modname_disp */ .int ${1:c} - 1b /* _ddebug::function_disp */ .int ${2:c} - 1b /* _ddebug::filename_disp */ .int ${3:c} - 1b /* _ddebug::format_disp */ .int ${4:c} /* _ddebug::flags_lineno */ .org 1b+${5:c}.popsection.set __UNIQUE_ID_ddebug18.ddebug.once, ${6:c}.elseif __UNIQUE_ID_ddebug18.ddebug.once - ${6:c}.line 477 - 1.error "'__UNIQUE_ID_ddebug18' used as _ddebug identifier more than once".endif' pr_debug("Detected %s compressed data\n", compress_name); ^ ../include/linux/printk.h:336:2: note: expanded from macro 'pr_debug' dynamic_pr_debug(fmt, ##__VA_ARGS__) ^ ../include/linux/dynamic_debug.h:158:2: note: expanded from macro 'dynamic_pr_debug' _dynamic_func_call(fmt, __dynamic_pr_debug, \ ^ ../include/linux/dynamic_debug.h:148:2: note: expanded from macro '_dynamic_func_call' __dynamic_func_call(__UNIQUE_ID(ddebug), fmt, func, ##__VA_ARGS__) ^ ../include/linux/dynamic_debug.h:128:2: note: expanded from macro '__dynamic_func_call' DEFINE_DYNAMIC_DEBUG_METADATA(id, fmt); \ ^ ../include/asm-generic/dynamic_debug.h:92:15: note: expanded from macro 'DEFINE_DYNAMIC_DEBUG_METADATA' asm volatile(".ifndef " __stringify(name) "\n" \ ^ I assume the same thing happens on the other architectures as well. Arnd > --- > arch/arm64/Kconfig | 1 + > arch/arm64/include/asm/Kbuild | 1 + > 2 files changed, 2 insertions(+) > > diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig > index 7e34b9eba5de..d0871d523d5d 100644 > --- a/arch/arm64/Kconfig > +++ b/arch/arm64/Kconfig > @@ -83,6 +83,7 @@ config ARM64 > select CRC32 > select DCACHE_WORD_ACCESS > select DMA_DIRECT_REMAP > + select DYNAMIC_DEBUG_RELATIVE_POINTERS > select EDAC_SUPPORT > select FRAME_POINTER > select GENERIC_ALLOCATOR > diff --git a/arch/arm64/include/asm/Kbuild b/arch/arm64/include/asm/Kbuild > index 1e17ea5c372b..1ead0d645686 100644 > --- a/arch/arm64/include/asm/Kbuild > +++ b/arch/arm64/include/asm/Kbuild > @@ -3,6 +3,7 @@ generic-y += delay.h > generic-y += div64.h > generic-y += dma.h > generic-y += dma-contiguous.h > +generic-y += dynamic_debug.h > generic-y += early_ioremap.h > generic-y += emergency-restart.h > generic-y += hw_irq.h > -- > 2.20.1 > > > _______________________________________________ > linux-arm-kernel mailing list > linux-arm-kernel@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
On 26/04/2019 11.39, Arnd Bergmann wrote: > On Tue, Apr 9, 2019 at 11:26 PM Rasmus Villemoes > <linux@rasmusvillemoes.dk> wrote: >> >> Similar to GENERIC_BUG_RELATIVE_POINTERS, one can now relativize the >> four const char* members of struct _ddebug, thus saving 16 bytes per >> instance (one for each pr_debug(), dev_debug() etc. in a >> CONFIG_DYNAMIC_DEBUG kernel). The asm-generic implementation seems to >> work out-of-the-box, though this is only compile-tested. >> >> Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk> > > This causes a build regression when compiling with clang, > see http://staging-storage.kernelci.org/next-clang/master/next-20190424/arm64/allmodconfig/clang-8/build.log Yes, see also https://github.com/ClangBuiltLinux/linux/issues/456 . The quickest short-term fix is to append "if CC_IS_GCC" to the select statements. Then when a fix lands in clang one can change that to "if CC_IS_GCC || CLANG_VERSION >= something". It's probably best if we fix -next builds ASAP instead of waiting for knowing the proper value of "something". Nathan, Nick, WDYT? I had 0day verify my patches before sending them out officially, and thought it also did clang builds. But apparently not, or not with enough arch/.config combinations? Rasmus
On Fri, Apr 26, 2019 at 12:05:57PM +0200, Rasmus Villemoes wrote: > On 26/04/2019 11.39, Arnd Bergmann wrote: > > On Tue, Apr 9, 2019 at 11:26 PM Rasmus Villemoes > > <linux@rasmusvillemoes.dk> wrote: > >> > >> Similar to GENERIC_BUG_RELATIVE_POINTERS, one can now relativize the > >> four const char* members of struct _ddebug, thus saving 16 bytes per > >> instance (one for each pr_debug(), dev_debug() etc. in a > >> CONFIG_DYNAMIC_DEBUG kernel). The asm-generic implementation seems to > >> work out-of-the-box, though this is only compile-tested. > >> > >> Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk> > > > > This causes a build regression when compiling with clang, > > see http://staging-storage.kernelci.org/next-clang/master/next-20190424/arm64/allmodconfig/clang-8/build.log > > Yes, see also https://github.com/ClangBuiltLinux/linux/issues/456 . > > The quickest short-term fix is to append "if CC_IS_GCC" to the select > statements. Then when a fix lands in clang one can change that to "if > CC_IS_GCC || CLANG_VERSION >= something". It's probably best if we fix > -next builds ASAP instead of waiting for knowing the proper value of > "something". Nathan, Nick, WDYT? Those select statements have to be added regardless, we might as well do it now. It should unbreak the auto builders because they use the latest clang stable version, which is 8.0.0. For the record, the fix in Clang is https://reviews.llvm.org/D60887 and should land shortly unless there are any further objections, meaning this will be fixed in the 9.0.0 release. 'if CC_IS_GCC || CLANG_VERSION >= 90000' should do it on the arm64 and powerpc select statements (x86 works fine because %c support has always been present). > > I had 0day verify my patches before sending them out officially, and > thought it also did clang builds. But apparently not, or not with enough > arch/.config combinations? They do not do clang builds unfortunately. Nick was in contact with them before forced asm-goto on x86 happened and derailed that. We try to do our own tests and let people know when stuff breaks but that is usually after it hits -next, rather than the mailing list. > > Rasmus Cheers, Nathan
diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig index 7e34b9eba5de..d0871d523d5d 100644 --- a/arch/arm64/Kconfig +++ b/arch/arm64/Kconfig @@ -83,6 +83,7 @@ config ARM64 select CRC32 select DCACHE_WORD_ACCESS select DMA_DIRECT_REMAP + select DYNAMIC_DEBUG_RELATIVE_POINTERS select EDAC_SUPPORT select FRAME_POINTER select GENERIC_ALLOCATOR diff --git a/arch/arm64/include/asm/Kbuild b/arch/arm64/include/asm/Kbuild index 1e17ea5c372b..1ead0d645686 100644 --- a/arch/arm64/include/asm/Kbuild +++ b/arch/arm64/include/asm/Kbuild @@ -3,6 +3,7 @@ generic-y += delay.h generic-y += div64.h generic-y += dma.h generic-y += dma-contiguous.h +generic-y += dynamic_debug.h generic-y += early_ioremap.h generic-y += emergency-restart.h generic-y += hw_irq.h
Similar to GENERIC_BUG_RELATIVE_POINTERS, one can now relativize the four const char* members of struct _ddebug, thus saving 16 bytes per instance (one for each pr_debug(), dev_debug() etc. in a CONFIG_DYNAMIC_DEBUG kernel). The asm-generic implementation seems to work out-of-the-box, though this is only compile-tested. Signed-off-by: Rasmus Villemoes <linux@rasmusvillemoes.dk> --- arch/arm64/Kconfig | 1 + arch/arm64/include/asm/Kbuild | 1 + 2 files changed, 2 insertions(+)