diff options
| -rw-r--r-- | arch/x86_64/lib/bitops.c | 66 |
1 files changed, 50 insertions, 16 deletions
diff --git a/arch/x86_64/lib/bitops.c b/arch/x86_64/lib/bitops.c index a29fb75b33ac..95b6d9639fba 100644 --- a/arch/x86_64/lib/bitops.c +++ b/arch/x86_64/lib/bitops.c | |||
| @@ -5,19 +5,23 @@ | |||
| 5 | #undef find_first_bit | 5 | #undef find_first_bit |
| 6 | #undef find_next_bit | 6 | #undef find_next_bit |
| 7 | 7 | ||
| 8 | /** | 8 | static inline long |
| 9 | * find_first_zero_bit - find the first zero bit in a memory region | 9 | __find_first_zero_bit(const unsigned long * addr, unsigned long size) |
| 10 | * @addr: The address to start the search at | ||
| 11 | * @size: The maximum size to search | ||
| 12 | * | ||
| 13 | * Returns the bit-number of the first zero bit, not the number of the byte | ||
| 14 | * containing a bit. | ||
| 15 | */ | ||
| 16 | inline long find_first_zero_bit(const unsigned long * addr, unsigned long size) | ||
| 17 | { | 10 | { |
| 18 | long d0, d1, d2; | 11 | long d0, d1, d2; |
| 19 | long res; | 12 | long res; |
| 20 | 13 | ||
| 14 | /* | ||
| 15 | * We must test the size in words, not in bits, because | ||
| 16 | * otherwise incoming sizes in the range -63..-1 will not run | ||
| 17 | * any scasq instructions, and then the flags used by the je | ||
| 18 | * instruction will have whatever random value was in place | ||
| 19 | * before. Nobody should call us like that, but | ||
| 20 | * find_next_zero_bit() does when offset and size are at the | ||
| 21 | * same word and it fails to find a zero itself. | ||
| 22 | */ | ||
| 23 | size += 63; | ||
| 24 | size >>= 6; | ||
| 21 | if (!size) | 25 | if (!size) |
| 22 | return 0; | 26 | return 0; |
| 23 | asm volatile( | 27 | asm volatile( |
| @@ -30,12 +34,30 @@ inline long find_first_zero_bit(const unsigned long * addr, unsigned long size) | |||
| 30 | " shlq $3,%%rdi\n" | 34 | " shlq $3,%%rdi\n" |
| 31 | " addq %%rdi,%%rdx" | 35 | " addq %%rdi,%%rdx" |
| 32 | :"=d" (res), "=&c" (d0), "=&D" (d1), "=&a" (d2) | 36 | :"=d" (res), "=&c" (d0), "=&D" (d1), "=&a" (d2) |
| 33 | :"0" (0ULL), "1" ((size + 63) >> 6), "2" (addr), "3" (-1ULL), | 37 | :"0" (0ULL), "1" (size), "2" (addr), "3" (-1ULL), |
| 34 | [addr] "r" (addr) : "memory"); | 38 | [addr] "S" (addr) : "memory"); |
| 39 | /* | ||
| 40 | * Any register would do for [addr] above, but GCC tends to | ||
| 41 | * prefer rbx over rsi, even though rsi is readily available | ||
| 42 | * and doesn't have to be saved. | ||
| 43 | */ | ||
| 35 | return res; | 44 | return res; |
| 36 | } | 45 | } |
| 37 | 46 | ||
| 38 | /** | 47 | /** |
| 48 | * find_first_zero_bit - find the first zero bit in a memory region | ||
| 49 | * @addr: The address to start the search at | ||
| 50 | * @size: The maximum size to search | ||
| 51 | * | ||
| 52 | * Returns the bit-number of the first zero bit, not the number of the byte | ||
| 53 | * containing a bit. | ||
| 54 | */ | ||
| 55 | long find_first_zero_bit(const unsigned long * addr, unsigned long size) | ||
| 56 | { | ||
| 57 | return __find_first_zero_bit (addr, size); | ||
| 58 | } | ||
| 59 | |||
| 60 | /** | ||
| 39 | * find_next_zero_bit - find the first zero bit in a memory region | 61 | * find_next_zero_bit - find the first zero bit in a memory region |
| 40 | * @addr: The address to base the search on | 62 | * @addr: The address to base the search on |
| 41 | * @offset: The bitnumber to start searching at | 63 | * @offset: The bitnumber to start searching at |
| @@ -43,7 +65,7 @@ inline long find_first_zero_bit(const unsigned long * addr, unsigned long size) | |||
| 43 | */ | 65 | */ |
| 44 | long find_next_zero_bit (const unsigned long * addr, long size, long offset) | 66 | long find_next_zero_bit (const unsigned long * addr, long size, long offset) |
| 45 | { | 67 | { |
| 46 | unsigned long * p = ((unsigned long *) addr) + (offset >> 6); | 68 | const unsigned long * p = addr + (offset >> 6); |
| 47 | unsigned long set = 0; | 69 | unsigned long set = 0; |
| 48 | unsigned long res, bit = offset&63; | 70 | unsigned long res, bit = offset&63; |
| 49 | 71 | ||
| @@ -63,8 +85,8 @@ long find_next_zero_bit (const unsigned long * addr, long size, long offset) | |||
| 63 | /* | 85 | /* |
| 64 | * No zero yet, search remaining full words for a zero | 86 | * No zero yet, search remaining full words for a zero |
| 65 | */ | 87 | */ |
| 66 | res = find_first_zero_bit ((const unsigned long *)p, | 88 | res = __find_first_zero_bit (p, size - 64 * (p - addr)); |
| 67 | size - 64 * (p - (unsigned long *) addr)); | 89 | |
| 68 | return (offset + set + res); | 90 | return (offset + set + res); |
| 69 | } | 91 | } |
| 70 | 92 | ||
| @@ -74,6 +96,19 @@ __find_first_bit(const unsigned long * addr, unsigned long size) | |||
| 74 | long d0, d1; | 96 | long d0, d1; |
| 75 | long res; | 97 | long res; |
| 76 | 98 | ||
| 99 | /* | ||
| 100 | * We must test the size in words, not in bits, because | ||
| 101 | * otherwise incoming sizes in the range -63..-1 will not run | ||
| 102 | * any scasq instructions, and then the flags used by the jz | ||
| 103 | * instruction will have whatever random value was in place | ||
| 104 | * before. Nobody should call us like that, but | ||
| 105 | * find_next_bit() does when offset and size are at the same | ||
| 106 | * word and it fails to find a one itself. | ||
| 107 | */ | ||
| 108 | size += 63; | ||
| 109 | size >>= 6; | ||
| 110 | if (!size) | ||
| 111 | return 0; | ||
| 77 | asm volatile( | 112 | asm volatile( |
| 78 | " repe; scasq\n" | 113 | " repe; scasq\n" |
| 79 | " jz 1f\n" | 114 | " jz 1f\n" |
| @@ -83,8 +118,7 @@ __find_first_bit(const unsigned long * addr, unsigned long size) | |||
| 83 | " shlq $3,%%rdi\n" | 118 | " shlq $3,%%rdi\n" |
| 84 | " addq %%rdi,%%rax" | 119 | " addq %%rdi,%%rax" |
| 85 | :"=a" (res), "=&c" (d0), "=&D" (d1) | 120 | :"=a" (res), "=&c" (d0), "=&D" (d1) |
| 86 | :"0" (0ULL), | 121 | :"0" (0ULL), "1" (size), "2" (addr), |
| 87 | "1" ((size + 63) >> 6), "2" (addr), | ||
| 88 | [addr] "r" (addr) : "memory"); | 122 | [addr] "r" (addr) : "memory"); |
| 89 | return res; | 123 | return res; |
| 90 | } | 124 | } |
