Skip to contentSkip to conversation

lib: drop size parameter from sort()'s swap callback

PATCH In review Jan Beulich 3 participants 6 messages Jul 29, 2025 – Jul 30, 2025

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lib: drop size parameter from sort()'s swap callback

This was needed only for generic_swap(), which disappeared in 8cb0341a61fa ("xen/sort: Switch to an extern inline implementation").

xen/arch/arm/io.c +1 -1
@@ -100,7 +100,7 @@ static int cmp_mmio_handler(const void *
     return 0;
 }
 
-static void swap_mmio_handler(void *_a, void *_b, size_t size)
+static void swap_mmio_handler(void *_a, void *_b)
 {
     struct mmio_handler *a = _a, *b = _b;
 
xen/arch/x86/extable.c +1 -1
@@ -39,7 +39,7 @@ static int init_or_livepatch cf_check cm
 	return 0;
 }
 
-static void init_or_livepatch cf_check swap_ex(void *a, void *b, size_t size)
+static void init_or_livepatch cf_check swap_ex(void *a, void *b)
 {
 	struct exception_table_entry *l = a, *r = b, tmp;
 	long delta = b - a;
xen/common/device-tree/bootinfo-fdt.c +1 -1
@@ -449,7 +449,7 @@ static int __init cmp_memory_node(const
     return 0;
 }
 
-static void __init swap_memory_node(void *_a, void *_b, size_t size)
+static void __init swap_memory_node(void *_a, void *_b)
 {
     struct membank *a = _a, *b = _b;
 
xen/include/xen/sort.h +4 -4
@@ -24,7 +24,7 @@ extern gnu_inline
 #endif
 void sort(void *base, size_t num, size_t size,
           int (*cmp)(const void *a, const void *b),
-          void (*swap)(void *a, void *b, size_t size))
+          void (*swap)(void *a, void *b))
 {
     /* pre-scale counters for performance */
     size_t i = (num / 2) * size, n = num * size, c, r;
@@ -39,7 +39,7 @@ void sort(void *base, size_t num, size_t
                 c += size;
             if ( cmp(base + r, base + c) >= 0 )
                 break;
-            swap(base + r, base + c, size);
+            swap(base + r, base + c);
         }
     }
 
@@ -47,7 +47,7 @@ void sort(void *base, size_t num, size_t
     for ( i = n; i > 0; )
     {
         i -= size;
-        swap(base, base + i, size);
+        swap(base, base + i);
         for ( r = 0; r * 2 + size < i; r = c )
         {
             c = r * 2 + size;
@@ -55,7 +55,7 @@ void sort(void *base, size_t num, size_t
                 c += size;
             if ( cmp(base + r, base + c) >= 0 )
                 break;
-            swap(base + r, base + c, size);
+            swap(base + r, base + c);
         }
     }
 }
#2 lore raw
Quoted 9 lines · Andrew Cooper
Quoted 4 lines · Jan Beulich

This was needed only for generic_swap(), which disappeared in 8cb0341a61fa ("xen/sort: Switch to an extern inline implementation").

Oh, nice.

Thanks.

Quoted 2 lines

I'd expect there to be no change in generated code here, as everything gets inlined.

Not really, no. With the change in place, both gcc7 and gcc14 consider the inlining of swap_ex() (in x86'es extable.c) as less beneficial, and hence (like cmp_ex()) an out-of-line function appears, while overall code size reduces. I expect that's because inlining decisions are taken based on some intermediate internal representation rather than based on the code that would ultimately be generated. And that intermediate internal representation now changes, resulting in less of a win by doing the inlining.

Jan

#3 lore raw
Quoted 13 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
Quoted 4 lines · Jan Beulich

This was needed only for generic_swap(), which disappeared in 8cb0341a61fa ("xen/sort: Switch to an extern inline implementation").

Oh, nice.

Thanks.

Quoted 2 lines

I'd expect there to be no change in generated code here, as everything gets inlined.

Quoted 8 lines

Not really, no. With the change in place, both gcc7 and gcc14 consider the inlining of swap_ex() (in x86'es extable.c) as less beneficial, and hence (like cmp_ex()) an out-of-line function appears, while overall code size reduces. I expect that's because inlining decisions are taken based on some intermediate internal representation rather than based on the code that would ultimately be generated. And that intermediate internal representation now changes, resulting in less of a win by doing the inlining.

Hmm.  We might consider __always_inline, although it seems like gcc12 does decide to inline them with this patch as-is.

Either way, that's something for later.

~Andrew

#4 lore raw
Quoted 23 lines · Andrew Cooper
Quoted 13 lines · Jan Beulich
Quoted 8 lines · Andrew Cooper
Quoted 4 lines · Jan Beulich

This was needed only for generic_swap(), which disappeared in 8cb0341a61fa ("xen/sort: Switch to an extern inline implementation").

Oh, nice.

Thanks.

Quoted 2 lines

I'd expect there to be no change in generated code here, as everything gets inlined.

Quoted 8 lines

Not really, no. With the change in place, both gcc7 and gcc14 consider the inlining of swap_ex() (in x86'es extable.c) as less beneficial, and hence (like cmp_ex()) an out-of-line function appears, while overall code size reduces. I expect that's because inlining decisions are taken based on some intermediate internal representation rather than based on the code that would ultimately be generated. And that intermediate internal representation now changes, resulting in less of a win by doing the inlining.

Quoted 1 line

Hmm.  We might consider __always_inline,

I would actually expect a compiler to oppose to an always-inline function to have its address taken.

Quoted 2 lines

although it seems like gcc12
does decide to inline them with this patch as-is.

Interesting how things can move back and forth between versions ...

Jan

Acked-by on v1 1/1
#5 lore raw
In reply to Jan Beulich

Hi,

Quoted 6 lines · Jan Beulich

This was needed only for generic_swap(), which disappeared in 8cb0341a61fa ("xen/sort: Switch to an extern inline implementation").

For arm part:

Cheers
Bertrand

Quoted 72 lines
xen/arch/arm/io.c +1 -1
@@ -100,7 +100,7 @@ static int cmp_mmio_handler(const void *
    return 0;
 }
 
-static void swap_mmio_handler(void *_a, void *_b, size_t size)
+static void swap_mmio_handler(void *_a, void *_b)
 {
    struct mmio_handler *a = _a, *b = _b;
 
xen/arch/x86/extable.c +1 -1
@@ -39,7 +39,7 @@ static int init_or_livepatch cf_check cm
 return 0;
 }
 
-static void init_or_livepatch cf_check swap_ex(void *a, void *b, size_t size)
+static void init_or_livepatch cf_check swap_ex(void *a, void *b)
 {
 struct exception_table_entry *l = a, *r = b, tmp;
 long delta = b - a;
xen/common/device-tree/bootinfo-fdt.c +1 -1
@@ -449,7 +449,7 @@ static int __init cmp_memory_node(const
    return 0;
 }
 
-static void __init swap_memory_node(void *_a, void *_b, size_t size)
+static void __init swap_memory_node(void *_a, void *_b)
 {
    struct membank *a = _a, *b = _b;
 
xen/include/xen/sort.h +4 -4
@@ -24,7 +24,7 @@ extern gnu_inline
 #endif
 void sort(void *base, size_t num, size_t size,
          int (*cmp)(const void *a, const void *b),
-          void (*swap)(void *a, void *b, size_t size))
+          void (*swap)(void *a, void *b))
 {
    /* pre-scale counters for performance */
    size_t i = (num / 2) * size, n = num * size, c, r;
@@ -39,7 +39,7 @@ void sort(void *base, size_t num, size_t
                c += size;
            if ( cmp(base + r, base + c) >= 0 )
                break;
-            swap(base + r, base + c, size);
+            swap(base + r, base + c);
        }
    }
 
@@ -47,7 +47,7 @@ void sort(void *base, size_t num, size_t
    for ( i = n; i > 0; )
    {
        i -= size;
-        swap(base, base + i, size);
+        swap(base, base + i);
        for ( r = 0; r * 2 + size < i; r = c )
        {
            c = r * 2 + size;
@@ -55,7 +55,7 @@ void sort(void *base, size_t num, size_t
                c += size;
            if ( cmp(base + r, base + c) >= 0 )
                break;
-            swap(base + r, base + c, size);
+            swap(base + r, base + c);
        }
    }
 }