Re: [PATCH] x86/irqflags: Force a register output in native_save_fl()

[Eric Dumazet <edumazet@google.com>]

On Thu, Dec 18, 2025 at 11:22 AM Peter Zijlstra <peterz@infradead.org> wrote:
>
> On Thu, Dec 18, 2025 at 11:15:37AM +0100, Eric Dumazet wrote:
>
> > > >--- a/arch/x86/include/asm/irqflags.h
> > > >+++ b/arch/x86/include/asm/irqflags.h
> > > >@@ -18,14 +18,9 @@ extern __always_inline unsigned long native_save_fl(void)
> > > > {
> > > >       unsigned long flags;
> > > >
> > > >-      /*
> > > >-       * "=rm" is safe here, because "pop" adjusts the stack before
> > > >-       * it evaluates its effective address -- this is part of the
> > > >-       * documented behavior of the "pop" instruction.
> > > >-       */
> > > >       asm volatile("# __raw_save_flags\n\t"
> > > >                    "pushf ; pop %0"
> > > >-                   : "=rm" (flags)
> > > >+                   : "=r" (flags)
> > > >                    : /* no input */
> > > >                    : "memory");
> > > >
> > >
> > > Maybe report a bug to the clang team?
> >
> > This has been done 8 years ago. No progress so far...
> >
> > They claim we should use __builtin_ia32_readeflags_u64 /
> > __builtin_ia32_writeeflags_u64, which seems unlikely to please x86
> > maintainers ?
>
> Yeah, because the "rm" thing not being supported properly goes way
> further than just this one case. And it really is abysmal that clang
> hasn't managed to implement it in all that time :/

Let me know if you prefer a V2 with this instead. It saves ~190 bytes
more with clang ;)

Thanks !

diff --git a/arch/x86/include/asm/irqflags.h b/arch/x86/include/asm/irqflags.h
index b30e5474c18e1be63b7c69354c26ae6a6cb02731..0209ae149f2b5f9ddacc6c8066a6d3191696ec42
100644
--- a/arch/x86/include/asm/irqflags.h
+++ b/arch/x86/include/asm/irqflags.h
@@ -16,20 +16,11 @@
 extern inline unsigned long native_save_fl(void);
 extern __always_inline unsigned long native_save_fl(void)
 {
-       unsigned long flags;
-
-       /*
-        * "=rm" is safe here, because "pop" adjusts the stack before
-        * it evaluates its effective address -- this is part of the
-        * documented behavior of the "pop" instruction.
-        */
-       asm volatile("# __raw_save_flags\n\t"
-                    "pushf ; pop %0"
-                    : "=rm" (flags)
-                    : /* no input */
-                    : "memory");
-
-       return flags;
+#ifdef CONFIG_X86_64
+       return __builtin_ia32_readeflags_u64();
+#else
+       return __builtin_ia32_readeflags_u32();
+#endif
 }

 static __always_inline void native_irq_disable(void)