* Ingo Molnar <mingo@kernel.org> wrote:

> > And then *if* there are any places that want to out-line the asm 
> > (why would you do that? At that point you'd be better off just 
> > writing assembler!), they could use an explicit __asm__ instead 
> > with a comment.
> > 
> > Sadly, I think doing just a mindless
> > 
> >     #define asm(...) __asm__ __inline__(__VA_ARGS__)
> > 
> > doesn't work, because we also have
> > 
> >         register void *tos asm("r11");
> > 
> > kind of patterns.

Plus we also have a lot of plain asm() statements in .c files in file 
scope that GCC doesn't accept with an __inline keyword:

    CC      kernel/configs.o
  kernel/configs.c:23:5: error: expected ‘(’ before ‘__inline’
     23 | asm __inline (
        |     ^~~~~~~~
        |     (
  make[3]: *** [scripts/Makefile.build:207: kernel/configs.o] Error 1

Because allowing that would have been way too simple. :-/

And a lot of helper macros that are used in such a scope are affected 
as well.

> > So first we'd have to change those to use __asm__(), and *then* we
> > could do the "asm() is always __asm__ __inline__()" thing.
> 
> Yeah, I'll try this out. Should this be done for all architectures, 
> or just for x86 for the time being?
> 
> I'm not sure an arch opt-in is worth it, it will only end up in the 
> active architectures picking it, but none of the others, and we'll 
> live with the duality forever.

So the attached patch builds on x86-[64|32]{defconfig|allmodconfig}
and boots on x86-64 defconfig.

The central part is:

 /* Make regular asm() templates inline too, on compilers that support it: */
 
 #ifdef CONFIG_CC_HAS_ASM_INLINE
 # define asm_inline __asm__ __inline
 # define asm(...) asm_inline(__VA_ARGS__)
 #else
 # define asm_inline asm
 #endif

And I fixed up the places where this isn't syntactically correct:

 35 files changed, 82 insertions(+), 79 deletions(-)

I haven't looked at code generation much yet, but text size changes are 
minimal:

      text	   data	    bss	     dec	    hex	filename
  29429076	7931870	1401196	38762142	24f769e	vmlinux.before
  29429631	7931870	1401200	38762701	24f78cd	vmlinux.after

Which is promising, assuming I haven't messed up anywhere.

Tested on GCC 14.2.0.

... and obviously this will break the build on all other architectures 
that have CONFIG_CC_HAS_ASM_INLINE, given that two dozen files had to 
be fixed on x86 alone.

Thanks,

	Ingo

===============>
From: Ingo Molnar <mingo@kernel.org>
Date: Tue, 18 Mar 2025 20:32:13 +0100
Subject: [PATCH] compiler/gcc: Make asm() templates asm __inline__() by default

Not-Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 arch/x86/coco/sev/shared.c                  | 10 +++++-----
 arch/x86/crypto/curve25519-x86_64.c         |  2 +-
 arch/x86/include/asm/alternative.h          |  2 +-
 arch/x86/include/asm/asm.h                  |  2 +-
 arch/x86/include/asm/cfi.h                  |  2 +-
 arch/x86/include/asm/irq_stack.h            |  2 +-
 arch/x86/include/asm/linkage.h              |  2 +-
 arch/x86/include/asm/paravirt.h             |  2 +-
 arch/x86/include/asm/static_call.h          |  4 ++--
 arch/x86/include/asm/uaccess.h              |  4 ++--
 arch/x86/include/asm/xen/hypercall.h        | 12 ++++++------
 arch/x86/kernel/alternative.c               |  6 +++---
 arch/x86/kernel/callthunks.c                |  2 +-
 arch/x86/kernel/cpu/resctrl/pseudo_lock.c   |  6 +++---
 arch/x86/kernel/kprobes/opt.c               |  2 +-
 arch/x86/kernel/rethook.c                   |  2 +-
 arch/x86/kernel/static_call.c               |  2 +-
 arch/x86/kernel/uprobes.c                   |  2 +-
 arch/x86/kvm/emulate.c                      |  4 ++--
 arch/x86/lib/error-inject.c                 |  2 +-
 include/linux/compiler_types.h              |  7 +++++--
 include/linux/export-internal.h             |  4 ++--
 include/linux/export.h                      |  2 +-
 include/linux/init.h                        |  2 +-
 include/linux/linkage.h                     |  4 ++--
 include/linux/pci.h                         |  2 +-
 include/linux/tracepoint.h                  |  2 +-
 include/vdso/math64.h                       |  2 +-
 kernel/configs.c                            |  2 +-
 kernel/kheaders.c                           |  2 +-
 samples/ftrace/ftrace-direct-modify.c       | 12 ++++++------
 samples/ftrace/ftrace-direct-multi-modify.c | 12 ++++++------
 samples/ftrace/ftrace-direct-multi.c        | 12 ++++++------
 samples/ftrace/ftrace-direct-too.c          | 12 ++++++------
 samples/ftrace/ftrace-direct.c              | 12 ++++++------
 35 files changed, 82 insertions(+), 79 deletions(-)

diff --git a/arch/x86/coco/sev/shared.c b/arch/x86/coco/sev/shared.c
index 2e4122f8aa6b..ac8b898e1d26 100644
--- a/arch/x86/coco/sev/shared.c
+++ b/arch/x86/coco/sev/shared.c
@@ -281,11 +281,11 @@ static inline int svsm_process_result_codes(struct svsm_call *call)
  */
 static __always_inline void svsm_issue_call(struct svsm_call *call, u8 *pending)
 {
-	register unsigned long rax asm("rax") = call->rax;
-	register unsigned long rcx asm("rcx") = call->rcx;
-	register unsigned long rdx asm("rdx") = call->rdx;
-	register unsigned long r8  asm("r8")  = call->r8;
-	register unsigned long r9  asm("r9")  = call->r9;
+	register unsigned long rax __asm__("rax") = call->rax;
+	register unsigned long rcx __asm__("rcx") = call->rcx;
+	register unsigned long rdx __asm__("rdx") = call->rdx;
+	register unsigned long r8  __asm__("r8")  = call->r8;
+	register unsigned long r9  __asm__("r9")  = call->r9;
 
 	call->caa->call_pending = 1;
 
diff --git a/arch/x86/crypto/curve25519-x86_64.c b/arch/x86/crypto/curve25519-x86_64.c
index dcfc0de333de..7d664bc8b2fd 100644
--- a/arch/x86/crypto/curve25519-x86_64.c
+++ b/arch/x86/crypto/curve25519-x86_64.c
@@ -542,7 +542,7 @@ static inline void fmul2(u64 *out, const u64 *f1, const u64 *f2, u64 *tmp)
  * Requires f2 to be smaller than 2^17 */
 static inline void fmul_scalar(u64 *out, const u64 *f1, u64 f2)
 {
-	register u64 f2_r asm("rdx") = f2;
+	register u64 f2_r __asm__("rdx") = f2;
 
 	asm volatile(
 		/* Compute the raw multiplication of f1*f2 */
diff --git a/arch/x86/include/asm/alternative.h b/arch/x86/include/asm/alternative.h
index e3903b731305..ab6c80eaf7f5 100644
--- a/arch/x86/include/asm/alternative.h
+++ b/arch/x86/include/asm/alternative.h
@@ -272,7 +272,7 @@ static inline int alternatives_text_reserved(void *start, void *end)
 
 /* Macro for creating assembler functions avoiding any C magic. */
 #define DEFINE_ASM_FUNC(func, instr, sec)		\
-	asm (".pushsection " #sec ", \"ax\"\n"		\
+	__asm__ (".pushsection " #sec ", \"ax\"\n"	\
 	     ".global " #func "\n\t"			\
 	     ".type " #func ", @function\n\t"		\
 	     ASM_FUNC_ALIGN "\n"			\
diff --git a/arch/x86/include/asm/asm.h b/arch/x86/include/asm/asm.h
index 2bec0c89a95c..5b88c875d512 100644
--- a/arch/x86/include/asm/asm.h
+++ b/arch/x86/include/asm/asm.h
@@ -219,7 +219,7 @@ static __always_inline __pure void *rip_rel_ptr(void *p)
  * gets set up by the containing function.  If you forget to do this, objtool
  * may print a "call without frame pointer save/setup" warning.
  */
-register unsigned long current_stack_pointer asm(_ASM_SP);
+register unsigned long current_stack_pointer __asm__(_ASM_SP);
 #define ASM_CALL_CONSTRAINT "+r" (current_stack_pointer)
 #endif /* __ASSEMBLY__ */
 
diff --git a/arch/x86/include/asm/cfi.h b/arch/x86/include/asm/cfi.h
index 31d19c815f99..5fd236a9ac25 100644
--- a/arch/x86/include/asm/cfi.h
+++ b/arch/x86/include/asm/cfi.h
@@ -140,7 +140,7 @@ static inline u32 cfi_get_func_hash(void *func)
 #endif /* CONFIG_CFI_CLANG */
 
 #if HAS_KERNEL_IBT == 1
-#define CFI_NOSEAL(x)	asm(IBT_NOSEAL(__stringify(x)))
+#define CFI_NOSEAL(x)	__asm__(IBT_NOSEAL(__stringify(x)))
 #endif
 
 #endif /* _ASM_X86_CFI_H */
diff --git a/arch/x86/include/asm/irq_stack.h b/arch/x86/include/asm/irq_stack.h
index 562a547c29a5..478291297ff2 100644
--- a/arch/x86/include/asm/irq_stack.h
+++ b/arch/x86/include/asm/irq_stack.h
@@ -80,7 +80,7 @@
  */
 #define call_on_stack(stack, func, asm_call, argconstr...)		\
 {									\
-	register void *tos asm("r11");					\
+	register void *tos __asm__("r11");				\
 									\
 	tos = ((void *)(stack));					\
 									\
diff --git a/arch/x86/include/asm/linkage.h b/arch/x86/include/asm/linkage.h
index dc31b13b87a0..8753b9f2e473 100644
--- a/arch/x86/include/asm/linkage.h
+++ b/arch/x86/include/asm/linkage.h
@@ -13,7 +13,7 @@
  * The generic version tends to create spurious ENDBR instructions under
  * certain conditions.
  */
-#define _THIS_IP_ ({ unsigned long __here; asm ("lea 0(%%rip), %0" : "=r" (__here)); __here; })
+#define _THIS_IP_ ({ unsigned long __here; __asm__ ("lea 0(%%rip), %0" : "=r" (__here)); __here; })
 #endif
 
 #ifdef CONFIG_X86_32
diff --git a/arch/x86/include/asm/paravirt.h b/arch/x86/include/asm/paravirt.h
index 041aff51eb50..7090d1478982 100644
--- a/arch/x86/include/asm/paravirt.h
+++ b/arch/x86/include/asm/paravirt.h
@@ -648,7 +648,7 @@ bool __raw_callee_save___native_vcpu_is_preempted(long cpu);
 #define __PV_CALLEE_SAVE_REGS_THUNK(func, section)			\
 	extern typeof(func) __raw_callee_save_##func;			\
 									\
-	asm(".pushsection " section ", \"ax\";"				\
+	__asm__(".pushsection " section ", \"ax\";"			\
 	    ".globl " PV_THUNK_NAME(func) ";"				\
 	    ".type " PV_THUNK_NAME(func) ", @function;"			\
 	    ASM_FUNC_ALIGN						\
diff --git a/arch/x86/include/asm/static_call.h b/arch/x86/include/asm/static_call.h
index 41502bd2afd6..23e6dca3dc63 100644
--- a/arch/x86/include/asm/static_call.h
+++ b/arch/x86/include/asm/static_call.h
@@ -32,7 +32,7 @@
  * and __static_call_fixup().
  */
 #define __ARCH_DEFINE_STATIC_CALL_TRAMP(name, insns)			\
-	asm(".pushsection .static_call.text, \"ax\"		\n"	\
+	__asm__(".pushsection .static_call.text, \"ax\"		\n"	\
 	    ".align 4						\n"	\
 	    ".globl " STATIC_CALL_TRAMP_STR(name) "		\n"	\
 	    STATIC_CALL_TRAMP_STR(name) ":			\n"	\
@@ -58,7 +58,7 @@
 	ARCH_DEFINE_STATIC_CALL_TRAMP(name, __static_call_return0)
 
 #define ARCH_ADD_TRAMP_KEY(name)					\
-	asm(".pushsection .static_call_tramp_key, \"a\"		\n"	\
+	__asm__(".pushsection .static_call_tramp_key, \"a\"	\n"	\
 	    ".long " STATIC_CALL_TRAMP_STR(name) " - .		\n"	\
 	    ".long " STATIC_CALL_KEY_STR(name) " - .		\n"	\
 	    ".popsection					\n")
diff --git a/arch/x86/include/asm/uaccess.h b/arch/x86/include/asm/uaccess.h
index 3a7755c1a441..32715343ebb1 100644
--- a/arch/x86/include/asm/uaccess.h
+++ b/arch/x86/include/asm/uaccess.h
@@ -76,7 +76,7 @@ extern int __get_user_bad(void);
 #define do_get_user_call(fn,x,ptr)					\
 ({									\
 	int __ret_gu;							\
-	register __inttype(*(ptr)) __val_gu asm("%"_ASM_DX);		\
+	register __inttype(*(ptr)) __val_gu __asm__("%"_ASM_DX);	\
 	__chk_user_ptr(ptr);						\
 	asm volatile("call __" #fn "_%c[size]"				\
 		     : "=a" (__ret_gu), "=r" (__val_gu),		\
@@ -171,7 +171,7 @@ extern void __put_user_nocheck_8(void);
 ({									\
 	int __ret_pu;							\
 	void __user *__ptr_pu;						\
-	register __typeof__(*(ptr)) __val_pu asm("%"_ASM_AX);		\
+	register __typeof__(*(ptr)) __val_pu __asm__("%"_ASM_AX);	\
 	__typeof__(*(ptr)) __x = (x); /* eval x once */			\
 	__typeof__(ptr) __ptr = (ptr); /* eval ptr once */		\
 	__chk_user_ptr(__ptr);						\
diff --git a/arch/x86/include/asm/xen/hypercall.h b/arch/x86/include/asm/xen/hypercall.h
index 97771b9d33af..0ecce35f287f 100644
--- a/arch/x86/include/asm/xen/hypercall.h
+++ b/arch/x86/include/asm/xen/hypercall.h
@@ -120,12 +120,12 @@ DECLARE_STATIC_CALL(xen_hypercall, xen_hypercall_func);
 #endif
 
 #define __HYPERCALL_DECLS						\
-	register unsigned long __res  asm(__HYPERCALL_RETREG);		\
-	register unsigned long __arg1 asm(__HYPERCALL_ARG1REG) = __arg1; \
-	register unsigned long __arg2 asm(__HYPERCALL_ARG2REG) = __arg2; \
-	register unsigned long __arg3 asm(__HYPERCALL_ARG3REG) = __arg3; \
-	register unsigned long __arg4 asm(__HYPERCALL_ARG4REG) = __arg4; \
-	register unsigned long __arg5 asm(__HYPERCALL_ARG5REG) = __arg5;
+	register unsigned long __res  __asm__(__HYPERCALL_RETREG);		\
+	register unsigned long __arg1 __asm__(__HYPERCALL_ARG1REG) = __arg1; \
+	register unsigned long __arg2 __asm__(__HYPERCALL_ARG2REG) = __arg2; \
+	register unsigned long __arg3 __asm__(__HYPERCALL_ARG3REG) = __arg3; \
+	register unsigned long __arg4 __asm__(__HYPERCALL_ARG4REG) = __arg4; \
+	register unsigned long __arg5 __asm__(__HYPERCALL_ARG5REG) = __arg5;
 
 #define __HYPERCALL_0PARAM	"=r" (__res), ASM_CALL_CONSTRAINT
 #define __HYPERCALL_1PARAM	__HYPERCALL_0PARAM, "+r" (__arg1)
diff --git a/arch/x86/kernel/alternative.c b/arch/x86/kernel/alternative.c
index c71b575bf229..cde99c0e2502 100644
--- a/arch/x86/kernel/alternative.c
+++ b/arch/x86/kernel/alternative.c
@@ -942,7 +942,7 @@ extern unsigned int __bpf_prog_runX(const void *ctx,
 __ADDRESSABLE(__bpf_prog_runX);
 
 /* u32 __ro_after_init cfi_bpf_hash = __kcfi_typeid___bpf_prog_runX; */
-asm (
+__asm__ (
 "	.pushsection	.data..ro_after_init,\"aw\",@progbits	\n"
 "	.type	cfi_bpf_hash,@object				\n"
 "	.globl	cfi_bpf_hash					\n"
@@ -959,7 +959,7 @@ extern u64 __bpf_callback_fn(u64, u64, u64, u64, u64);
 __ADDRESSABLE(__bpf_callback_fn);
 
 /* u32 __ro_after_init cfi_bpf_subprog_hash = __kcfi_typeid___bpf_callback_fn; */
-asm (
+__asm__ (
 "	.pushsection	.data..ro_after_init,\"aw\",@progbits	\n"
 "	.type	cfi_bpf_subprog_hash,@object			\n"
 "	.globl	cfi_bpf_subprog_hash				\n"
@@ -1598,7 +1598,7 @@ int alternatives_text_reserved(void *start, void *end)
 
 extern void int3_magic(unsigned int *ptr); /* defined in asm */
 
-asm (
+__asm__ (
 "	.pushsection	.init.text, \"ax\", @progbits\n"
 "	.type		int3_magic, @function\n"
 "int3_magic:\n"
diff --git a/arch/x86/kernel/callthunks.c b/arch/x86/kernel/callthunks.c
index 8418a892d195..5d865c851b4f 100644
--- a/arch/x86/kernel/callthunks.c
+++ b/arch/x86/kernel/callthunks.c
@@ -64,7 +64,7 @@ static const struct core_text builtin_coretext = {
 	.name = "builtin",
 };
 
-asm (
+__asm__ (
 	".pushsection .rodata				\n"
 	".global skl_call_thunk_template		\n"
 	"skl_call_thunk_template:			\n"
diff --git a/arch/x86/kernel/cpu/resctrl/pseudo_lock.c b/arch/x86/kernel/cpu/resctrl/pseudo_lock.c
index 42cc162f7fc9..49381dc69837 100644
--- a/arch/x86/kernel/cpu/resctrl/pseudo_lock.c
+++ b/arch/x86/kernel/cpu/resctrl/pseudo_lock.c
@@ -446,9 +446,9 @@ static int pseudo_lock_fn(void *_rdtgrp)
 	unsigned int size;
 	void *mem_r;
 #else
-	register unsigned int line_size asm("esi");
-	register unsigned int size asm("edi");
-	register void *mem_r asm(_ASM_BX);
+	register unsigned int line_size __asm__("esi");
+	register unsigned int size __asm__("edi");
+	register void *mem_r __asm__(_ASM_BX);
 #endif /* CONFIG_KASAN */
 
 	/*
diff --git a/arch/x86/kernel/kprobes/opt.c b/arch/x86/kernel/kprobes/opt.c
index 36d6809c6c9e..eb394142a3b3 100644
--- a/arch/x86/kernel/kprobes/opt.c
+++ b/arch/x86/kernel/kprobes/opt.c
@@ -101,7 +101,7 @@ static void synthesize_set_arg1(kprobe_opcode_t *addr, unsigned long val)
 	*(unsigned long *)addr = val;
 }
 
-asm (
+__asm__ (
 			".pushsection .rodata\n"
 			"optprobe_template_func:\n"
 			".global optprobe_template_entry\n"
diff --git a/arch/x86/kernel/rethook.c b/arch/x86/kernel/rethook.c
index 8a1c0111ae79..bb3ebb6c7618 100644
--- a/arch/x86/kernel/rethook.c
+++ b/arch/x86/kernel/rethook.c
@@ -19,7 +19,7 @@ __visible void arch_rethook_trampoline_callback(struct pt_regs *regs);
  * When a target function returns, this code saves registers and calls
  * arch_rethook_trampoline_callback(), which calls the rethook handler.
  */
-asm(
+__asm__(
 	".text\n"
 	".global arch_rethook_trampoline\n"
 	".type arch_rethook_trampoline, @function\n"
diff --git a/arch/x86/kernel/static_call.c b/arch/x86/kernel/static_call.c
index 9e51242ed125..13984965b8a7 100644
--- a/arch/x86/kernel/static_call.c
+++ b/arch/x86/kernel/static_call.c
@@ -41,7 +41,7 @@ static u8 __is_Jcc(u8 *insn) /* Jcc.d32 */
 
 extern void __static_call_return(void);
 
-asm (".global __static_call_return\n\t"
+__asm__ (".global __static_call_return\n\t"
      ".type __static_call_return, @function\n\t"
      ASM_FUNC_ALIGN "\n\t"
      "__static_call_return:\n\t"
diff --git a/arch/x86/kernel/uprobes.c b/arch/x86/kernel/uprobes.c
index 5a952c5ea66b..2a56d8313cbb 100644
--- a/arch/x86/kernel/uprobes.c
+++ b/arch/x86/kernel/uprobes.c
@@ -310,7 +310,7 @@ static int uprobe_init_insn(struct arch_uprobe *auprobe, struct insn *insn, bool
 
 #ifdef CONFIG_X86_64
 
-asm (
+__asm__ (
 	".pushsection .rodata\n"
 	".global uretprobe_trampoline_entry\n"
 	"uretprobe_trampoline_entry:\n"
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index 60986f67c35a..80800901474d 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -311,7 +311,7 @@ static int fastop(struct x86_emulate_ctxt *ctxt, fastop_t fop);
 
 #define __FOP_START(op, align) \
 	extern void em_##op(struct fastop *fake); \
-	asm(".pushsection .text, \"ax\" \n\t" \
+	__asm__(".pushsection .text, \"ax\" \n\t" \
 	    ".global em_" #op " \n\t" \
 	    ".align " __stringify(align) " \n\t" \
 	    "em_" #op ":\n\t"
@@ -1069,7 +1069,7 @@ static __always_inline u8 test_cc(unsigned int condition, unsigned long flags)
 	void (*fop)(void) = (void *)em_setcc + FASTOP_SIZE * (condition & 0xf);
 
 	flags = (flags & EFLAGS_MASK) | X86_EFLAGS_IF;
-	asm("push %[flags]; popf; " CALL_NOSPEC
+	__asm__("push %[flags]; popf; " CALL_NOSPEC
 	    : "=a"(rc), ASM_CALL_CONSTRAINT : [thunk_target]"r"(fop), [flags]"r"(flags));
 	return rc;
 }
diff --git a/arch/x86/lib/error-inject.c b/arch/x86/lib/error-inject.c
index b5a6d83106bc..0fd65281078c 100644
--- a/arch/x86/lib/error-inject.c
+++ b/arch/x86/lib/error-inject.c
@@ -7,7 +7,7 @@
 
 asmlinkage void just_return_func(void);
 
-asm(
+__asm__(
 	".text\n"
 	".type just_return_func, @function\n"
 	".globl just_return_func\n"
diff --git a/include/linux/compiler_types.h b/include/linux/compiler_types.h
index 981cc3d7e3aa..4c59faf2c22f 100644
--- a/include/linux/compiler_types.h
+++ b/include/linux/compiler_types.h
@@ -473,10 +473,13 @@ struct ftrace_likely_data {
   #define ASM_INPUT_RM "rm"
 #endif
 
+/* Make regular asm() templates inline too, on compilers that support it: */
+
 #ifdef CONFIG_CC_HAS_ASM_INLINE
-#define asm_inline asm __inline
+# define asm_inline __asm__ __inline
+# define asm(...) asm_inline(__VA_ARGS__)
 #else
-#define asm_inline asm
+# define asm_inline asm
 #endif
 
 /* Are two types/vars the same type (ignoring qualifiers)? */
diff --git a/include/linux/export-internal.h b/include/linux/export-internal.h
index d445705ac13c..61dfbc986c9a 100644
--- a/include/linux/export-internal.h
+++ b/include/linux/export-internal.h
@@ -38,7 +38,7 @@
  * former apparently works on all arches according to the binutils source.
  */
 #define __KSYMTAB(name, sym, sec, ns)						\
-	asm("	.section \"__ksymtab_strings\",\"aMS\",%progbits,1"	"\n"	\
+	__asm__("	.section \"__ksymtab_strings\",\"aMS\",%progbits,1"	"\n"	\
 	    "__kstrtab_" #name ":"					"\n"	\
 	    "	.asciz \"" #name "\""					"\n"	\
 	    "__kstrtabns_" #name ":"					"\n"	\
@@ -63,7 +63,7 @@
 #define KSYMTAB_DATA(name, sec, ns)	__KSYMTAB(name, name, sec, ns)
 
 #define SYMBOL_CRC(sym, crc, sec)   \
-	asm(".section \"___kcrctab" sec "+" #sym "\",\"a\""	"\n" \
+	__asm__(".section \"___kcrctab" sec "+" #sym "\",\"a\""	"\n" \
 	    ".balign 4"						"\n" \
 	    "__crc_" #sym ":"					"\n" \
 	    ".long " #crc					"\n" \
diff --git a/include/linux/export.h b/include/linux/export.h
index a8c23d945634..ca2fc2cc5db8 100644
--- a/include/linux/export.h
+++ b/include/linux/export.h
@@ -70,7 +70,7 @@
 	extern typeof(sym) sym;					\
 	__ADDRESSABLE(sym)					\
 	__GENDWARFKSYMS_EXPORT(sym)				\
-	asm(__stringify(___EXPORT_SYMBOL(sym, license, ns)))
+	__asm__(__stringify(___EXPORT_SYMBOL(sym, license, ns)))
 
 #endif
 
diff --git a/include/linux/init.h b/include/linux/init.h
index ee1309473bc6..85bc49b7e34d 100644
--- a/include/linux/init.h
+++ b/include/linux/init.h
@@ -259,7 +259,7 @@ extern struct module __this_module;
 #ifdef CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
 #define ____define_initcall(fn, __stub, __name, __sec)		\
 	__define_initcall_stub(__stub, fn)			\
-	asm(".section	\"" __sec "\", \"a\"		\n"	\
+	__asm__(".section	\"" __sec "\", \"a\"	\n"	\
 	    __stringify(__name) ":			\n"	\
 	    ".long	" __stringify(__stub) " - .	\n"	\
 	    ".previous					\n");	\
diff --git a/include/linux/linkage.h b/include/linux/linkage.h
index 5c8865bb59d9..74bc0c88f8ca 100644
--- a/include/linux/linkage.h
+++ b/include/linux/linkage.h
@@ -23,14 +23,14 @@
 #endif
 
 #ifndef cond_syscall
-#define cond_syscall(x)	asm(				\
+#define cond_syscall(x)	__asm__(			\
 	".weak " __stringify(x) "\n\t"			\
 	".set  " __stringify(x) ","			\
 		 __stringify(sys_ni_syscall))
 #endif
 
 #ifndef SYSCALL_ALIAS
-#define SYSCALL_ALIAS(alias, name) asm(			\
+#define SYSCALL_ALIAS(alias, name) __asm__(		\
 	".globl " __stringify(alias) "\n\t"		\
 	".set   " __stringify(alias) ","		\
 		  __stringify(name))
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 47b31ad724fa..0904d24c9e17 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -2206,7 +2206,7 @@ enum pci_fixup_pass {
 #define ___DECLARE_PCI_FIXUP_SECTION(sec, name, vendor, device, class,	\
 				    class_shift, hook)			\
 	__ADDRESSABLE(hook)						\
-	asm(".section "	#sec ", \"a\"				\n"	\
+	__asm__(".section "	#sec ", \"a\"			\n"	\
 	    ".balign	16					\n"	\
 	    ".short "	#vendor ", " #device "			\n"	\
 	    ".long "	#class ", " #class_shift "		\n"	\
diff --git a/include/linux/tracepoint.h b/include/linux/tracepoint.h
index a351763e6965..4f0f7cca7a62 100644
--- a/include/linux/tracepoint.h
+++ b/include/linux/tracepoint.h
@@ -149,7 +149,7 @@ static inline struct tracepoint *tracepoint_ptr_deref(tracepoint_ptr_t *p)
 }
 
 #define __TRACEPOINT_ENTRY(name)					\
-	asm("	.section \"__tracepoints_ptrs\", \"a\"		\n"	\
+	__asm__("	.section \"__tracepoints_ptrs\", \"a\"	\n"	\
 	    "	.balign 4					\n"	\
 	    "	.long 	__tracepoint_" #name " - .		\n"	\
 	    "	.previous					\n")
diff --git a/include/vdso/math64.h b/include/vdso/math64.h
index 22ae212f8b28..d391d9622414 100644
--- a/include/vdso/math64.h
+++ b/include/vdso/math64.h
@@ -10,7 +10,7 @@ __iter_div_u64_rem(u64 dividend, u32 divisor, u64 *remainder)
 	while (dividend >= divisor) {
 		/* The following asm() prevents the compiler from
 		   optimising this loop into a modulo operation.  */
-		asm("" : "+rm"(dividend));
+		__asm__("" : "+rm"(dividend));
 
 		dividend -= divisor;
 		ret++;
diff --git a/kernel/configs.c b/kernel/configs.c
index a28c79c5f713..076342773995 100644
--- a/kernel/configs.c
+++ b/kernel/configs.c
@@ -20,7 +20,7 @@
  * "IKCFG_ST" and "IKCFG_ED" are used to extract the config data from
  * a binary kernel image or a module. See scripts/extract-ikconfig.
  */
-asm (
+__asm__ (
 "	.pushsection .rodata, \"a\"		\n"
 "	.ascii \"IKCFG_ST\"			\n"
 "	.global kernel_config_data		\n"
diff --git a/kernel/kheaders.c b/kernel/kheaders.c
index 378088b07f46..1468c46806a3 100644
--- a/kernel/kheaders.c
+++ b/kernel/kheaders.c
@@ -16,7 +16,7 @@
  * compressed kernel headers are stored. The file is first compressed with xz.
  */
 
-asm (
+__asm__ (
 "	.pushsection .rodata, \"a\"		\n"
 "	.global kernel_headers_data		\n"
 "kernel_headers_data:				\n"
diff --git a/samples/ftrace/ftrace-direct-modify.c b/samples/ftrace/ftrace-direct-modify.c
index cfea7a38befb..a59d68845a58 100644
--- a/samples/ftrace/ftrace-direct-modify.c
+++ b/samples/ftrace/ftrace-direct-modify.c
@@ -27,7 +27,7 @@ static unsigned long my_ip = (unsigned long)schedule;
 #ifdef CONFIG_RISCV
 #include <asm/asm.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -64,7 +64,7 @@ asm (
 #include <asm/ibt.h>
 #include <asm/nospec-branch.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -96,7 +96,7 @@ asm (
 
 #ifdef CONFIG_S390
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -135,7 +135,7 @@ asm (
 
 #ifdef CONFIG_ARM64
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -167,7 +167,7 @@ asm (
 
 #ifdef CONFIG_LOONGARCH
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -239,7 +239,7 @@ asm (
 "	bctr\n"
 #endif
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
diff --git a/samples/ftrace/ftrace-direct-multi-modify.c b/samples/ftrace/ftrace-direct-multi-modify.c
index 8f7986d698d8..0f45e47e5137 100644
--- a/samples/ftrace/ftrace-direct-multi-modify.c
+++ b/samples/ftrace/ftrace-direct-multi-modify.c
@@ -25,7 +25,7 @@ extern void my_tramp2(void *);
 #ifdef CONFIG_RISCV
 #include <asm/asm.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -68,7 +68,7 @@ asm (
 #include <asm/ibt.h>
 #include <asm/nospec-branch.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -106,7 +106,7 @@ asm (
 
 #ifdef CONFIG_S390
 
-asm (
+__asm__ (
 "       .pushsection    .text, \"ax\", @progbits\n"
 "       .type           my_tramp1, @function\n"
 "       .globl          my_tramp1\n"
@@ -148,7 +148,7 @@ asm (
 
 #ifdef CONFIG_ARM64
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -187,7 +187,7 @@ asm (
 #ifdef CONFIG_LOONGARCH
 #include <asm/asm.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
@@ -273,7 +273,7 @@ asm (
 #define PPC_FTRACE_RECOVER_IP ""
 #endif
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp1, @function\n"
 "	.globl		my_tramp1\n"
diff --git a/samples/ftrace/ftrace-direct-multi.c b/samples/ftrace/ftrace-direct-multi.c
index db326c81a27d..c687682780fa 100644
--- a/samples/ftrace/ftrace-direct-multi.c
+++ b/samples/ftrace/ftrace-direct-multi.c
@@ -20,7 +20,7 @@ extern void my_tramp(void *);
 #ifdef CONFIG_RISCV
 #include <asm/asm.h>
 
-asm (
+__asm__ (
 "       .pushsection    .text, \"ax\", @progbits\n"
 "       .type           my_tramp, @function\n"
 "       .globl          my_tramp\n"
@@ -47,7 +47,7 @@ asm (
 #include <asm/ibt.h>
 #include <asm/nospec-branch.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -70,7 +70,7 @@ asm (
 
 #ifdef CONFIG_S390
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -95,7 +95,7 @@ asm (
 
 #ifdef CONFIG_ARM64
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -119,7 +119,7 @@ asm (
 #ifdef CONFIG_LOONGARCH
 
 #include <asm/asm.h>
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -189,7 +189,7 @@ asm (
 #define PPC_FTRACE_RECOVER_IP ""
 #endif
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
diff --git a/samples/ftrace/ftrace-direct-too.c b/samples/ftrace/ftrace-direct-too.c
index 3d0fa260332d..63b778b7285b 100644
--- a/samples/ftrace/ftrace-direct-too.c
+++ b/samples/ftrace/ftrace-direct-too.c
@@ -22,7 +22,7 @@ extern void my_tramp(void *);
 #ifdef CONFIG_RISCV
 #include <asm/asm.h>
 
-asm (
+__asm__ (
 "       .pushsection    .text, \"ax\", @progbits\n"
 "       .type           my_tramp, @function\n"
 "       .globl          my_tramp\n"
@@ -52,7 +52,7 @@ asm (
 #include <asm/ibt.h>
 #include <asm/nospec-branch.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -80,7 +80,7 @@ asm (
 
 #ifdef CONFIG_S390
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -104,7 +104,7 @@ asm (
 
 #ifdef CONFIG_ARM64
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -128,7 +128,7 @@ asm (
 
 #ifdef CONFIG_LOONGARCH
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -201,7 +201,7 @@ asm (
 "	bctr\n"
 #endif
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
diff --git a/samples/ftrace/ftrace-direct.c b/samples/ftrace/ftrace-direct.c
index 956834b0d19a..7ad16e5c4a5b 100644
--- a/samples/ftrace/ftrace-direct.c
+++ b/samples/ftrace/ftrace-direct.c
@@ -19,7 +19,7 @@ extern void my_tramp(void *);
 #ifdef CONFIG_RISCV
 #include <asm/asm.h>
 
-asm (
+__asm__ (
 "       .pushsection    .text, \"ax\", @progbits\n"
 "       .type           my_tramp, @function\n"
 "       .globl          my_tramp\n"
@@ -45,7 +45,7 @@ asm (
 #include <asm/ibt.h>
 #include <asm/nospec-branch.h>
 
-asm (
+__asm__ (
 "	.pushsection    .text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -67,7 +67,7 @@ asm (
 
 #ifdef CONFIG_S390
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -91,7 +91,7 @@ asm (
 
 #ifdef CONFIG_ARM64
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -113,7 +113,7 @@ asm (
 
 #ifdef CONFIG_LOONGARCH
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"
@@ -174,7 +174,7 @@ asm (
 "	bctr\n"
 #endif
 
-asm (
+__asm__ (
 "	.pushsection	.text, \"ax\", @progbits\n"
 "	.type		my_tramp, @function\n"
 "	.globl		my_tramp\n"

On Tue, Mar 18, 2025 at 9:11 PM Ingo Molnar <mingo@kernel.org> wrote:

>  #ifdef CONFIG_CC_HAS_ASM_INLINE
>  # define asm_inline __asm__ __inline
>  # define asm(...) asm_inline(__VA_ARGS__)
>  #else
>  # define asm_inline asm
>  #endif
>
> And I fixed up the places where this isn't syntactically correct:
>
>  35 files changed, 82 insertions(+), 79 deletions(-)
>
> I haven't looked at code generation much yet, but text size changes are
> minimal:
>
>       text         data     bss      dec            hex filename
>   29429076      7931870 1401196 38762142        24f769e vmlinux.before
>   29429631      7931870 1401200 38762701        24f78cd vmlinux.after
>
> Which is promising, assuming I haven't messed up anywhere.

Please use bloat-o-meter, it is more precise.

Actually, functions with the most impact (x86 locking functions and
__arch_hweight) were recently converted to asm_inline, so besides
__untagged_addr, the remaining have very little impact, if at all
(c.f. amd_clear_divider() ). There is also no need to convert asm()
without directives inside.

My proposal would be to convert the remaining few cases (the remaining
asms involving ALTERNATIVE and exceptions) "by hand" to asm_inline()
and stick a rule in checkpatch to use asm_inline() in the code
involving asm(), like we have the rule with asm volatile.

I don't think redefining an important C keyword is a good approach, it
obfuscates its meaning too much. And as has been shown by Ingo's
experiment, there is a substantial effort to fix false positives.
Instead of fixing these, we can trivially convert the remaining cases
to asm_volatile() as well, without obfuscating asm(). Checkpatch can
take care of future cases.

Thanks,
Uros.

* Uros Bizjak <ubizjak@gmail.com> wrote:

> On Tue, Mar 18, 2025 at 9:11 PM Ingo Molnar <mingo@kernel.org> wrote:
> 
> >  #ifdef CONFIG_CC_HAS_ASM_INLINE
> >  # define asm_inline __asm__ __inline
> >  # define asm(...) asm_inline(__VA_ARGS__)
> >  #else
> >  # define asm_inline asm
> >  #endif
> >
> > And I fixed up the places where this isn't syntactically correct:
> >
> >  35 files changed, 82 insertions(+), 79 deletions(-)
> >
> > I haven't looked at code generation much yet, but text size changes are
> > minimal:
> >
> >       text         data     bss      dec            hex filename
> >   29429076      7931870 1401196 38762142        24f769e vmlinux.before
> >   29429631      7931870 1401200 38762701        24f78cd vmlinux.after
> >
> > Which is promising, assuming I haven't messed up anywhere.
> 
> Please use bloat-o-meter, it is more precise.

Here's the bloat-o-meter output between vanilla and patched vmlinux:

add/remove: 6/20 grow/shrink: 43/13 up/down: 4245/-2812 (1433)
Function                                     old     new   delta
__ia32_sys_pidfd_send_signal                  21     818    +797
__x64_sys_pidfd_send_signal                   22     818    +796
nl80211_send_wiphy                         11189   11867    +678
icl_update_topdown_event                     473     691    +218
intel_joiner_adjust_timings.isra               -     145    +145
deactivate_locked_super                      148     249    +101
tcp_v4_send_synack                           301     389     +88
kill_anon_super                               53     137     +84
xa_destroy                                   291     371     +80
__xa_set_mark                                135     209     +74
ip_fraglist_prepare                          204     269     +65
store_hwp_dynamic_boost                      136     200     +64
csum_partial                                 239     302     +63
ip_fraglist_init                             155     217     +62
store_max_perf_pct                           252     311     +59
ip_frag_next                                 377     434     +57
ip_do_fragment                              1644    1701     +57
__ip_local_out                               283     340     +57
store_min_perf_pct                           273     329     +56
__udp4_lib_rcv                              2917    2970     +53
freeze_super                                1124    1175     +51
super_wake                                     -      47     +47
ieee80211_parse_tx_radiotap                 1267    1311     +44
ic_bootp_recv                               1333    1371     +38
vlv_compute_watermarks                      2267    2299     +32
intel_atomic_commit_tail                    5423    5455     +32
_g4x_compute_pipe_wm                         397     429     +32
__ieee80211_xmit_fast                       2611    2643     +32
intel_format_info_is_yuv_semiplanar           43      72     +29
skl_main_to_aux_plane                        113     136     +23
ip_auto_config                              4029    4050     +21
__memcpy_flushcache                          369     386     +17
validate_beacon_head                         240     256     +16
tcp_v4_rcv                                  4850    4866     +16
intel_enable_transcoder                     1443    1459     +16
intel_cx0pll_state_verify                   1859    1875     +16
intel_cx0_phy_check_hdmi_link_rate           181     197     +16
inet_gro_receive                             569     585     +16
__pfx_super_wake                               -      16     +16
__pfx_intel_joiner_adjust_timings.isra         -      16     +16
__pfx_intel_cx0_read.constprop                 -      16     +16
intel_fb_is_gen12_ccs_aux_plane.isra          95     107     +12
_intel_modeset_primary_pipes                  79      91     +12
intel_cx0_read.constprop                       -      10     +10
intel_fb_is_ccs_aux_plane                    110     117      +7
intel_crtc_readout_derived_state             513     516      +3
wq_update_node_max_active                    538     540      +2
intel_set_cdclk_pre_plane_update             838     840      +2
intel_sseu_subslice_total                     67      68      +1
intel_crtc_compute_config                    890     889      -1
_intel_modeset_secondary_pipes                50      47      -3
intel_mtl_pll_enable                        7472    7460     -12
bxt_set_cdclk                               1328    1316     -12
vfs_get_tree                                 205     189     -16
intel_fb_modifier_to_tiling                  186     170     -16
__pfx_nl80211_send_iftype_data                16       -     -16
__pfx_lane_mask_to_lane                       16       -     -16
__pfx_kill_super_notify.part                  16       -     -16
__pfx_ip_fast_csum                            16       -     -16
__pfx_intel_pstate_update_policies            16       -     -16
__pfx_intel_joiner_adjust_timings             16       -     -16
__pfx_format_is_yuv_semiplanar.part.isra      16       -     -16
__pfx_cdclk_divider                           16       -     -16
__pfx___icl_update_topdown_event              16       -     -16
__pfx___do_sys_pidfd_send_signal              16       -     -16
intel_c20_sram_read.constprop                173     149     -24
xas_split_alloc                              302     270     -32
nl80211_parse_sched_scan                    3311    3279     -32
kill_litter_super                             70      31     -39
format_is_yuv_semiplanar.part.isra            41       -     -41
lane_mask_to_lane                             44       -     -44
ip_fast_csum                                  48       -     -48
intel_cx0pll_readout_hw_state               1085    1037     -48
intel_pstate_update_policies                  66       -     -66
cdclk_divider                                 69       -     -69
kill_super_notify.part                       111       -    -111
__icl_update_topdown_event                   112       -    -112
intel_joiner_adjust_timings                  140       -    -140
intel_fill_fb_info                          2998    2813    -185
intel_tile_width_bytes                       747     531    -216
nl80211_send_iftype_data                     574       -    -574
__do_sys_pidfd_send_signal                   811       -    -811
Total: Before=22547058, After=22548491, chg +0.01%

A lot fewer functions are affected than I expected from such a 
large-scope change.

> Actually, functions with the most impact (x86 locking functions and
> __arch_hweight) were recently converted to asm_inline, so besides
> __untagged_addr, the remaining have very little impact, if at all
> (c.f. amd_clear_divider() ). There is also no need to convert asm()
> without directives inside.

I did the test with Linus-vanilla (81e4f8d68c66) to maximize the 
potential effect, which doesn't have those changes yet.

See tip:WIP.x86/core.

> My proposal would be to convert the remaining few cases (the remaining
> asms involving ALTERNATIVE and exceptions) "by hand" to asm_inline()
> and stick a rule in checkpatch to use asm_inline() in the code
> involving asm(), like we have the rule with asm volatile.
> 
> I don't think redefining an important C keyword is a good approach, it
> obfuscates its meaning too much. And as has been shown by Ingo's
> experiment, there is a substantial effort to fix false positives.
> Instead of fixing these, we can trivially convert the remaining cases
> to asm_volatile() as well, without obfuscating asm(). Checkpatch can
> take care of future cases.

This would work for me too. The cross-arch impact and churn seems 
substantial.

Thanks,

	Ingo

On Wed, Mar 19, 2025 at 11:34 PM Ingo Molnar <mingo@kernel.org> wrote:
>
>
> * Uros Bizjak <ubizjak@gmail.com> wrote:
>
> > On Tue, Mar 18, 2025 at 9:11 PM Ingo Molnar <mingo@kernel.org> wrote:
> >
> > >  #ifdef CONFIG_CC_HAS_ASM_INLINE
> > >  # define asm_inline __asm__ __inline
> > >  # define asm(...) asm_inline(__VA_ARGS__)
> > >  #else
> > >  # define asm_inline asm
> > >  #endif
> > >
> > > And I fixed up the places where this isn't syntactically correct:
> > >
> > >  35 files changed, 82 insertions(+), 79 deletions(-)
> > >
> > > I haven't looked at code generation much yet, but text size changes are
> > > minimal:
> > >
> > >       text         data     bss      dec            hex filename
> > >   29429076      7931870 1401196 38762142        24f769e vmlinux.before
> > >   29429631      7931870 1401200 38762701        24f78cd vmlinux.after
> > >
> > > Which is promising, assuming I haven't messed up anywhere.
> >
> > Please use bloat-o-meter, it is more precise.
>
> Here's the bloat-o-meter output between vanilla and patched vmlinux:

[...]

> A lot fewer functions are affected than I expected from such a
> large-scope change.

Interestingly, I got *many* more changes just from converting atomic
locking functions to asm_inline, as reported in [1].

[1] https://lore.kernel.org/lkml/CAFULd4YBcG45bigHBox2pu+To+Y5BzbRxG+pUr42AVOWSnfKsg@mail.gmail.com/

> > Actually, functions with the most impact (x86 locking functions and
> > __arch_hweight) were recently converted to asm_inline, so besides
> > __untagged_addr, the remaining have very little impact, if at all
> > (c.f. amd_clear_divider() ). There is also no need to convert asm()
> > without directives inside.
>
> I did the test with Linus-vanilla (81e4f8d68c66) to maximize the
> potential effect, which doesn't have those changes yet.
>
> See tip:WIP.x86/core.
>
> > My proposal would be to convert the remaining few cases (the remaining
> > asms involving ALTERNATIVE and exceptions) "by hand" to asm_inline()
> > and stick a rule in checkpatch to use asm_inline() in the code
> > involving asm(), like we have the rule with asm volatile.
> >
> > I don't think redefining an important C keyword is a good approach, it
> > obfuscates its meaning too much. And as has been shown by Ingo's
> > experiment, there is a substantial effort to fix false positives.
> > Instead of fixing these, we can trivially convert the remaining cases
> > to asm_volatile() as well, without obfuscating asm(). Checkpatch can
> > take care of future cases.
>
> This would work for me too. The cross-arch impact and churn seems
> substantial.

Perhaps we should also coordinate the effort with Josh, so we won't
step on each other's toes. From my analysis, gains from converting the
remaining asm() that involve ALTERNATIVE would be minor. It is also
questionable if asm() involving exceptions is worth converting at all,
I have to analyse them case-by-case some more during the next
development window.

Thanks,
Uros.

* Uros Bizjak <ubizjak@gmail.com> wrote:

> > > > I haven't looked at code generation much yet, but text size changes are
> > > > minimal:
> > > >
> > > >       text         data     bss      dec            hex filename
> > > >   29429076      7931870 1401196 38762142        24f769e vmlinux.before
> > > >   29429631      7931870 1401200 38762701        24f78cd vmlinux.after
> > > >
> > > > Which is promising, assuming I haven't messed up anywhere.
> > >
> > > Please use bloat-o-meter, it is more precise.
> >
> > Here's the bloat-o-meter output between vanilla and patched vmlinux:
> 
> [...]
> 
> > A lot fewer functions are affected than I expected from such a
> > large-scope change.
> 
> Interestingly, I got *many* more changes just from converting atomic
> locking functions to asm_inline, as reported in [1].
> 
> [1] https://lore.kernel.org/lkml/CAFULd4YBcG45bigHBox2pu+To+Y5BzbRxG+pUr42AVOWSnfKsg@mail.gmail.com/

Have you used a pristine x86-64 defconfig for your build tests?

Could you perhaps check my patch against your patch in your build 
environment and figure out why there's such unexpected differences?

As you noted my patch should be a blunt-instrument superset of your 
changes so if then it should affect *more* functions, not fewer.

Thanks,

	Ingo

On Thu, Mar 20, 2025 at 9:59 AM Ingo Molnar <mingo@kernel.org> wrote:
>
>
> * Uros Bizjak <ubizjak@gmail.com> wrote:
>
> > > > > I haven't looked at code generation much yet, but text size changes are
> > > > > minimal:
> > > > >
> > > > >       text         data     bss      dec            hex filename
> > > > >   29429076      7931870 1401196 38762142        24f769e vmlinux.before
> > > > >   29429631      7931870 1401200 38762701        24f78cd vmlinux.after
> > > > >
> > > > > Which is promising, assuming I haven't messed up anywhere.
> > > >
> > > > Please use bloat-o-meter, it is more precise.
> > >
> > > Here's the bloat-o-meter output between vanilla and patched vmlinux:
> >
> > [...]
> >
> > > A lot fewer functions are affected than I expected from such a
> > > large-scope change.
> >
> > Interestingly, I got *many* more changes just from converting atomic
> > locking functions to asm_inline, as reported in [1].
> >
> > [1] https://lore.kernel.org/lkml/CAFULd4YBcG45bigHBox2pu+To+Y5BzbRxG+pUr42AVOWSnfKsg@mail.gmail.com/
>
> Have you used a pristine x86-64 defconfig for your build tests?

Yes, it was vanilla x86-64 defconfig, but in -tip tree.

> Could you perhaps check my patch against your patch in your build
> environment and figure out why there's such unexpected differences?
>
> As you noted my patch should be a blunt-instrument superset of your
> changes so if then it should affect *more* functions, not fewer.

Yes, I'll try this experiment.

Thanks,
Uros.

On Thu, Mar 20, 2025 at 11:30 AM Uros Bizjak <ubizjak@gmail.com> wrote:

> > > Interestingly, I got *many* more changes just from converting atomic
> > > locking functions to asm_inline, as reported in [1].
> > >
> > > [1] https://lore.kernel.org/lkml/CAFULd4YBcG45bigHBox2pu+To+Y5BzbRxG+pUr42AVOWSnfKsg@mail.gmail.com/
> >
> > Have you used a pristine x86-64 defconfig for your build tests?
>
> Yes, it was vanilla x86-64 defconfig, but in -tip tree.
>
> > Could you perhaps check my patch against your patch in your build
> > environment and figure out why there's such unexpected differences?
> >
> > As you noted my patch should be a blunt-instrument superset of your
> > changes so if then it should affect *more* functions, not fewer.
>
> Yes, I'll try this experiment.

So, trying your patch on vanilla mainline kernel with gcc-14.2.1
resulted in (approx) the list you reported. However, adding my patch
that converted locking instructions [1] over your patch resulted in
(expected) hundreds (~1700) of changed functions.

[1] https://lore.kernel.org/lkml/20250309170955.48919-1-ubizjak@gmail.com/

The takeout from the experiment is that your proposed change is not
enough to convert "everything" and misses many cases (at least locking
instructions, as confirmed by my experiment). We should use a more
precise instrument and use it case-by-case. This will be some more
work, but on the first look, there are relatively few interesting
cases remaining.

Uros.

On Tue, Mar 18, 2025 at 09:11:40PM +0100, Ingo Molnar wrote:
>  #ifdef CONFIG_CC_HAS_ASM_INLINE
>  # define asm_inline __asm__ __inline
>  # define asm(...) asm_inline(__VA_ARGS__)
>  #else
>  # define asm_inline asm
>  #endif

Nice, I'd been wanting to do that for a while.

> And I fixed up the places where this isn't syntactically correct:
> 
>  35 files changed, 82 insertions(+), 79 deletions(-)
> 
> I haven't looked at code generation much yet, but text size changes are 
> minimal:
> 
>       text	   data	    bss	     dec	    hex	filename
>   29429076	7931870	1401196	38762142	24f769e	vmlinux.before
>   29429631	7931870	1401200	38762701	24f78cd	vmlinux.after
> 
> Which is promising, assuming I haven't messed up anywhere.

Unfortunately "size" can be misleading:

  - Inexplicably, "text" includes a lot of non-executable sections, many
    of which have sizes which are directly affected by actual .text
    changes.

  - CONFIG_MITIGATION_SRSO adds ~2MB padding between entry code and the
    rest of the runtime text, though this is not much of a concern for
    comparing apples to apples as it only hides the size of the entry
    text which is tiny.

This was discussed before (and yes "objtool size" would be nice
someday).

  https://lore.kernel.org/20231012013507.jrqnm35p7az6atov@treble

I think this should work decently enough:

  $ readelf -WS vmlinux | grep " .text" | sed 's/\[ //' | awk '{printf("0x%s\n", $6)}'
  0x5588b08

-- 
Josh