• Eric Dumazet's avatar
    lib/extable.c: remove an expensive integer divide in search_extable() · 15ae02ba
    Eric Dumazet authored
    
    Actual code let compiler generates idiv instruction on x86.
    
    Using a right shift is OK here and readable as well.
    
    Before patch
       10:   57                      push   %edi
       11:   56                      push   %esi
       12:   89 d6                   mov    %edx,%esi
       14:   53                      push   %ebx
       15:   89 c3                   mov    %eax,%ebx
       17:   eb 22                   jmp    3b <search_extable+0x2b>
       19:   89 f0                   mov    %esi,%eax
       1b:   ba 02 00 00 00          mov    $0x2,%edx
       20:   29 d8                   sub    %ebx,%eax
       22:   89 d7                   mov    %edx,%edi
       24:   c1 f8 03                sar    $0x3,%eax
       27:   99                      cltd
       28:   f7 ff                   idiv   %edi
       2a:   8d 04 c3                lea    (%ebx,%eax,8),%eax
       2d:   39 08                   cmp    %ecx,(%eax)
    ...
    
    After patch
    
    00000010 <search_extable>:
       10:   53                      push   %ebx
       11:   89 c3                   mov    %eax,%ebx
       13:   eb 18                   jmp    2d <search_extable+0x1d>
       15:   89 d0                   mov    %edx,%eax
       17:   29 d8                   sub    %ebx,%eax
       19:   c1 f8 04                sar    $0x4,%eax
       1c:   8d 04 c3                lea    (%ebx,%eax,8),%eax
       1f:   39 08                   cmp    %ecx,(%eax)
    ...
    Signed-off-by: default avatarEric Dumazet <dada1@cosmosbay.com>
    Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
    Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
    15ae02ba
extable.c 1.93 KB
/*
 * Derived from arch/ppc/mm/extable.c and arch/i386/mm/extable.c.
 *
 * Copyright (C) 2004 Paul Mackerras, IBM Corp.
 *
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License
 * as published by the Free Software Foundation; either version
 * 2 of the License, or (at your option) any later version.
 */

#include <linux/module.h>
#include <linux/init.h>
#include <linux/sort.h>
#include <asm/uaccess.h>

#ifndef ARCH_HAS_SORT_EXTABLE
/*
 * The exception table needs to be sorted so that the binary
 * search that we use to find entries in it works properly.
 * This is used both for the kernel exception table and for
 * the exception tables of modules that get loaded.
 */
static int cmp_ex(const void *a, const void *b)
{
	const struct exception_table_entry *x = a, *y = b;

	/* avoid overflow */
	if (x->insn > y->insn)
		return 1;
	if (x->insn < y->insn)
		return -1;
	return 0;
}

void sort_extable(struct exception_table_entry *start,
		  struct exception_table_entry *finish)
{
	sort(start, finish - start, sizeof(struct exception_table_entry),
	     cmp_ex, NULL);
}
#endif

#ifndef ARCH_HAS_SEARCH_EXTABLE
/*
 * Search one exception table for an entry corresponding to the
 * given instruction address, and return the address of the entry,
 * or NULL if none is found.
 * We use a binary search, and thus we assume that the table is
 * already sorted.
 */
const struct exception_table_entry *
search_extable(const struct exception_table_entry *first,
	       const struct exception_table_entry *last,
	       unsigned long value)
{
	while (first <= last) {
		const struct exception_table_entry *mid;

		mid = ((last - first) >> 1) + first;
		/*
		 * careful, the distance between value and insn
		 * can be larger than MAX_LONG:
		 */
		if (mid->insn < value)
			first = mid + 1;
		else if (mid->insn > value)
			last = mid - 1;
		else
			return mid;
        }
        return NULL;
}
#endif