[PATCH] m68k: implement runtime consts

[PATCH] m68k: implement runtime consts

[Daniel Palmer]

This implements runtime consts for m68k which hopefully improves
performance a tiny bit, maybe, hopefully... going downhill with
the wind behind us.

Constant pointers are just a register load with an immediate that is
fixed up.

Constant shifts are a very small register load with an immediate that
is fixed up and then the val is shifted by the register. Putting the
shift amount in a register is needed as the shift instruction can only
encode a maximum of shift by 8 with an immediate.

Before:

0013c6ea <__d_lookup_rcu_op_compare>:
  ...
  13c6fc:       2003            movel %d3,%d0
  13c6fe:       2239 0069 a980  movel 69a980 <d_hash_shift>,%d1
  13c704:       e2a8            lsrl %d1,%d0
  13c706:       e588            lsll #2,%d0
  13c708:       d0b9 0069 a97c  addl 69a97c <dentry_hashtable>,%d0
  ...

After:

0013c6c0 <__d_lookup_rcu_op_compare>:
  ...
  13c6d2:       207c cafe f00d  moveal #-889262067,%a0
  13c6d8:       2003            movel %d3,%d0
  13c6da:       720c            moveq #12,%d1
  13c6dc:       e2a8            lsrl %d1,%d0
  13c6de:       e588            lsll #2,%d0
  13c6e0:       d1c0            addal %d0,%a0
  ...

Signed-off-by: Daniel Palmer <daniel@thingy.jp>
---
 arch/m68k/include/asm/runtime-const.h | 76 +++++++++++++++++++++++++++
 1 file changed, 76 insertions(+)
 create mode 100644 arch/m68k/include/asm/runtime-const.h

diff --git a/arch/m68k/include/asm/runtime-const.h b/arch/m68k/include/asm/runtime-const.h
new file mode 100644
index 000000000000..576031c5b808
--- /dev/null
+++ b/arch/m68k/include/asm/runtime-const.h
@@ -0,0 +1,76 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * runtime const implementation for m68k
+ * Copyright (C) 2025 Daniel Palmer<daniel@thingy.jp>
+ *
+ * Based on arm64 version.
+ */
+#ifndef _ASM_RUNTIME_CONST_H
+#define _ASM_RUNTIME_CONST_H
+
+#include <asm/cacheflush.h>
+#include <linux/bitfield.h>
+
+#define runtime_const_ptr(sym) ({				\
+	typeof(sym) __ret;					\
+	asm_inline("1:\t"					\
+		"mov.l #0xcafef00d,%0\n\t"			\
+		".pushsection runtime_ptr_" #sym ",\"a\"\n\t"	\
+		".long 1b - .\n\t"				\
+		".popsection"					\
+		: "=a" (__ret));				\
+	__ret; })
+
+static inline void __runtime_fixup_ptr(void *where, unsigned long val)
+{
+	u32 *value = where + 2;
+	const unsigned long start = ((unsigned long) where) + 2;
+	const unsigned long end = start + sizeof(*value);
+
+	*value = val;
+
+	flush_icache_range(start, end);
+}
+
+#define runtime_const_shift_right_32(val, sym) ({		\
+	u32 __tmp;						\
+	asm_inline("1:\t"					\
+		"moveq #12, %0\n\t"				\
+		"lsr.l %0, %1\n\t"				\
+		".pushsection runtime_shift_" #sym ",\"a\"\n\t"	\
+		".long 1b - .\n\t"				\
+		".popsection"					\
+		: "=&d" (__tmp), "+d" (val));			\
+	val; })
+
+#define MOVEQ_DATA GENMASK(7, 0)
+static inline void __runtime_fixup_shift(void *where, unsigned long val)
+{
+	u16 *insn = where;
+	const unsigned long start = (unsigned long) where;
+	const unsigned long end = start + sizeof(*insn);
+
+	FIELD_MODIFY(MOVEQ_DATA, insn, val);
+
+	flush_icache_range(start, end);
+}
+
+#define runtime_const_init(type, sym) do {		\
+	extern s32 __start_runtime_##type##_##sym[];	\
+	extern s32 __stop_runtime_##type##_##sym[];	\
+	runtime_const_fixup(__runtime_fixup_##type,	\
+		(unsigned long)(sym),			\
+		__start_runtime_##type##_##sym,		\
+		__stop_runtime_##type##_##sym);		\
+} while (0)
+
+static inline void runtime_const_fixup(void (*fn)(void *, unsigned long),
+	unsigned long val, s32 *start, s32 *end)
+{
+	while (start < end) {
+		fn(*start + (void *)start, val);
+		start++;
+	}
+}
+
+#endif
-- 
2.51.0

Re: [PATCH] m68k: implement runtime consts

[kernel test robot]

Hi Daniel,

kernel test robot noticed the following build warnings:

[auto build test WARNING on geert-m68k/for-next]
[also build test WARNING on geert-m68k/for-linus linus/master v6.18-rc7 next-20251126]
[If your patch is applied to the wrong git tree, kindly drop us a note.
And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]

url:    https://github.com/intel-lab-lkp/linux/commits/Daniel-Palmer/m68k-implement-runtime-consts/20251127-080652
base:   https://git.kernel.org/pub/scm/linux/kernel/git/geert/linux-m68k.git for-next
patch link:    https://lore.kernel.org/r/20251127000505.2117956-1-daniel%40thingy.jp
patch subject: [PATCH] m68k: implement runtime consts
config: m68k-allnoconfig (https://download.01.org/0day-ci/archive/20251127/202511270956.hzGiPzdZ-lkp@intel.com/config)
compiler: m68k-linux-gcc (GCC) 15.1.0
reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20251127/202511270956.hzGiPzdZ-lkp@intel.com/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of
the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <lkp@intel.com>
| Closes: https://lore.kernel.org/oe-kbuild-all/202511270956.hzGiPzdZ-lkp@intel.com/

All warnings (new ones prefixed by >>):

   In file included from fs/dcache.c:38:
   arch/m68k/include/asm/runtime-const.h: In function '__runtime_fixup_ptr':
>> arch/m68k/include/asm/runtime-const.h:28:29: warning: unused variable 'end' [-Wunused-variable]
      28 |         const unsigned long end = start + sizeof(*value);
         |                             ^~~
   arch/m68k/include/asm/runtime-const.h: In function '__runtime_fixup_shift':
   arch/m68k/include/asm/runtime-const.h:51:29: warning: unused variable 'end' [-Wunused-variable]
      51 |         const unsigned long end = start + sizeof(*insn);
         |                             ^~~


vim +/end +28 arch/m68k/include/asm/runtime-const.h

    13	
    14	#define runtime_const_ptr(sym) ({				\
    15		typeof(sym) __ret;					\
    16		asm_inline("1:\t"					\
    17			"mov.l #0xcafef00d,%0\n\t"			\
    18			".pushsection runtime_ptr_" #sym ",\"a\"\n\t"	\
    19			".long 1b - .\n\t"				\
    20			".popsection"					\
    21			: "=a" (__ret));				\
    22		__ret; })
    23	
    24	static inline void __runtime_fixup_ptr(void *where, unsigned long val)
    25	{
    26		u32 *value = where + 2;
    27		const unsigned long start = ((unsigned long) where) + 2;
  > 28		const unsigned long end = start + sizeof(*value);
    29	
    30		*value = val;
    31	
    32		flush_icache_range(start, end);
    33	}
    34	

-- 
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki

Re: [PATCH] m68k: implement runtime consts

[Daniel Palmer]

Hi test robot,

On Thu, 27 Nov 2025 at 10:59, kernel test robot <lkp@intel.com> wrote:
> All warnings (new ones prefixed by >>):
>
>    In file included from fs/dcache.c:38:
>    arch/m68k/include/asm/runtime-const.h: In function '__runtime_fixup_ptr':
> >> arch/m68k/include/asm/runtime-const.h:28:29: warning: unused variable 'end' [-Wunused-variable]
>       28 |         const unsigned long end = start + sizeof(*value);
>          |                             ^~~
>    arch/m68k/include/asm/runtime-const.h: In function '__runtime_fixup_shift':
>    arch/m68k/include/asm/runtime-const.h:51:29: warning: unused variable 'end' [-Wunused-variable]
>       51 |         const unsigned long end = start + sizeof(*insn);
>          |                             ^~~

I guess this is because allnoconfig results in CONFIG_MMU=n and that means
that the nommu defines for cache operations are used and those look like:
#define flush_icache_range(start, len) __flush_icache_all()

And the arguments getting passed aren't used. Also noticed that it's
start and len
there not start and end like everywhere else seems to be...

I guess the  flush_icache_range macro needs to be tweaked to not
trigger the unused
warning but I'm not sure how right now.

Cheers,

Daniel

Re: [PATCH] m68k: implement runtime consts

[Geert Uytterhoeven]

Hi Daniel,

On Thu, 27 Nov 2025 at 11:13, Daniel Palmer <daniel@thingy.jp> wrote:
> On Thu, 27 Nov 2025 at 10:59, kernel test robot <lkp@intel.com> wrote:
> > All warnings (new ones prefixed by >>):
> >
> >    In file included from fs/dcache.c:38:
> >    arch/m68k/include/asm/runtime-const.h: In function '__runtime_fixup_ptr':
> > >> arch/m68k/include/asm/runtime-const.h:28:29: warning: unused variable 'end' [-Wunused-variable]
> >       28 |         const unsigned long end = start + sizeof(*value);
> >          |                             ^~~
> >    arch/m68k/include/asm/runtime-const.h: In function '__runtime_fixup_shift':
> >    arch/m68k/include/asm/runtime-const.h:51:29: warning: unused variable 'end' [-Wunused-variable]
> >       51 |         const unsigned long end = start + sizeof(*insn);
> >          |                             ^~~
>
> I guess this is because allnoconfig results in CONFIG_MMU=n and that means
> that the nommu defines for cache operations are used and those look like:
> #define flush_icache_range(start, len) __flush_icache_all()
>
> And the arguments getting passed aren't used. Also noticed that it's
> start and len
> there not start and end like everywhere else seems to be...
>
> I guess the  flush_icache_range macro needs to be tweaked to not
> trigger the unused
> warning but I'm not sure how right now.

Replacing the macros by static inline functions should fix that.

Gr{oetje,eeting}s,

                        Geert

-- 
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds

Re: [PATCH] m68k: implement runtime consts

[Daniel Palmer]

Hi Geert,

On Thu, 27 Nov 2025 at 19:22, Geert Uytterhoeven <geert@linux-m68k.org> wrote:

>
> Replacing the macros by static inline functions should fix that.

Thanks. I've done that now. Any comments on the rest of this?
I have a few things like implementing jump labels, a vDSO to get
around the sys_get_thread_area() bottleneck to come after this.

Cheers,

Daniel

Re: [PATCH] m68k: implement runtime consts

[Jean-Michel Hautbois]

On Wed, Dec 03, 2025 at 05:06:52PM +0900, Daniel Palmer wrote:
> Hi Geert,
> 
> On Thu, 27 Nov 2025 at 19:22, Geert Uytterhoeven <geert@linux-m68k.org> wrote:
> 
> >
> > Replacing the macros by static inline functions should fix that.
> 
> Thanks. I've done that now. Any comments on the rest of this?
> I have a few things like implementing jump labels, a vDSO to get
> around the sys_get_thread_area() bottleneck to come after this.

Hi Daniel,

It would be great to have vDSO at least for sys_get_thread_area() ! I
implemented something, but if you already have it, please, submit a RFC
I will test on coldfire !

Thanks,
JM

> Cheers,
> 
> Daniel

Re: [PATCH] m68k: implement runtime consts

[Daniel Palmer]

Hi Jean-Michel,

On Thu, 4 Dec 2025 at 18:57, Jean-Michel Hautbois
<jeanmichel.hautbois@yoseli.org> wrote:
> It would be great to have vDSO at least for sys_get_thread_area() ! I
> implemented something, but if you already have it, please, submit a RFC
> I will test on coldfire !

I will try to clean it up and send it. A very untidy version along
with other stuff I'm messing with is on github[0].
You will also need a patch for your libc to use it. So far I have
worked out how to get glibc[1] and uclibng[2] to work but I haven't
managed to get musl to work yet[3].
Note that I have no idea if it's safe to do what it's doing. ;)

0 - https://github.com/fifteenhex/linux/tree/m68kvdso-20251201
1 - https://github.com/fifteenhex/glibc/tree/m68kgetthreadarea-20251203
2 - https://github.com/fifteenhex/uclibc-ng/tree/m68k-vdsogetthreadarea
3 - https://github.com/fifteenhex/musl/tree/m68kgetthreadarea-20251203

Cheers,

Daniel