Skip to contentSkip to conversation

__ffs64()

RFC In review Steven Whitehouse 4 participants 5 messages Apr 22, 2009 – Apr 23, 2009 Bitops API

lore
Share & tools
lore
#0 lore raw

RFC 1/1 __ffs64()

Hi,

I'd like to add a new bitop, __ffs64() which I need in order to fix a bug in GFS2. The question is, where should it go?

On 64 bit arches, __ffs64() would be a synonym for __ffs(), but on 32 bit arches it degenerates to a conditional plus a call to __ffs(). I'm assuming that there would not be a lot of point in optimising this operation on 32 bit arches even if such an instruction was available, so that I should do something like the below patch.

Does that seem reasonable, or should I give it a separate header file under asm-generic/bitops/ like some of the similar operations? It looks like I'd have to touch a lot of other files if I were to go that route,

Steve.

include/linux/bitops.h +17 -0
index 6182913..d15255f 100644
@@ -112,6 +112,23 @@ static inline unsigned fls_long(unsigned long l)
 	return fls64(l);
 }
 
+/**
+ * __ffs64 - find first set bit in a 64 bit word
+ * @word: The 64 bit word
+ *
+ * On 64 bit arches this is a synomyn for __ffs
+ */
+static inline unsigned long __ffs64(u64 word)
+{
+#if BITS_PER_LONG == 32
+	if (((u32)word) == 0UL)
+		return __ffs((u32)(word >> 32)) + 32;
+#elif BITS_PER_LONG != 64
+#error BITS_PER_LONG not 32 or 64
+#endif
+	return __ffs((unsigned long)word);
+}
+
 #ifdef __KERNEL__
 #ifdef CONFIG_GENERIC_FIND_FIRST_BIT
 
#1 lore raw

Hi Steve,

Quoted 11 lines · Steven Whitehouse

Hi,

I'd like to add a new bitop, __ffs64() which I need in order to fix a bug in GFS2. The question is, where should it go?

On 64 bit arches, __ffs64() would be a synonym for __ffs(), but on 32 bit arches it degenerates to a conditional plus a call to __ffs(). I'm assuming that there would not be a lot of point in optimising this operation on 32 bit arches even if such an instruction was available, so that I should do something like the below patch.

Quoted 3 lines

Does that seem reasonable, or should I give it a separate header file under asm-generic/bitops/ like some of the similar operations? It looks like I'd have to touch a lot of other files if I were to go that route,

While I have no reply to your questions above, I have one minor comment :

Quoted 6 lines

+/**
+ * __ffs64 - find first set bit in a 64 bit word
+ * @word: The 64 bit word
+ *
+ * On 64 bit arches this is a synomyn for __ffs
+ */

IMHO you should remind here that the result is undefined when word==0 and the caller must check against it first.

include/linux/bitops.h static inline unsigned fls_long(unsigned long l) · line 133

Quoted 13 lines
include/linux/bitops.h +11 -0
+static inline unsigned long __ffs64(u64 word)
+{
+#if BITS_PER_LONG == 32
+	if (((u32)word) == 0UL)
+		return __ffs((u32)(word >> 32)) + 32;
+#elif BITS_PER_LONG != 64
+#error BITS_PER_LONG not 32 or 64
+#endif
+	return __ffs((unsigned long)word);
+}
+
 #ifdef __KERNEL__
 #ifdef CONFIG_GENERIC_FIND_FIRST_BIT

Regards,
Willy

#2 lore raw
In reply to Steven Whitehouse
Quoted 2 lines · Steven Whitehouse

I'd like to add a new bitop, __ffs64() which I need in order to fix a bug in GFS2. The question is, where should it go?

I think the location is right.

Quoted 9 lines

On 64 bit arches, __ffs64() would be a synonym for __ffs(), but on 32 bit arches it degenerates to a conditional plus a call to __ffs(). I'm assuming that there would not be a lot of point in optimising this operation on 32 bit arches even if such an instruction was available, so that I should do something like the below patch.

Does that seem reasonable, or should I give it a separate header file under asm-generic/bitops/ like some of the similar operations? It looks like I'd have to touch a lot of other files if I were to go that route,

One issue may be that some 32 bit architectures have a better way of doing 64 bit ffs.

#3 lore raw

Hi,

Quoted 13 lines · Christoph Lameter
Quoted 2 lines · Steven Whitehouse

I'd like to add a new bitop, __ffs64() which I need in order to fix a bug in GFS2. The question is, where should it go?

I think the location is right.

Quoted 6 lines

On 64 bit arches, __ffs64() would be a synonym for __ffs(), but on 32 bit arches it degenerates to a conditional plus a call to __ffs(). I'm assuming that there would not be a lot of point in optimising this operation on 32 bit arches even if such an instruction was available, so that I should do something like the below patch.

Quoted 7 lines
Quoted 3 lines

Does that seem reasonable, or should I give it a separate header file under asm-generic/bitops/ like some of the similar operations? It looks like I'd have to touch a lot of other files if I were to go that route,

One issue may be that some 32 bit architectures have a better way of doing 64 bit ffs.

Yes, thats what I was worried about. I don't have a wide enough knowledge of the different architectures to make a judgement about whether this is likely or not.

I guess maybe the right thing to do is to leave it as I did it in the patch and if an arch wants to create its own implementation, then it could be moved at that stage.

Steve.

#4 lore raw
Quoted 21 lines · Steven Whitehouse

Hi,

Quoted 12 lines · Christoph Lameter
Quoted 2 lines · Steven Whitehouse

I'd like to add a new bitop, __ffs64() which I need in order to fix a bug in GFS2. The question is, where should it go?

I think the location is right.

Quoted 6 lines

On 64 bit arches, __ffs64() would be a synonym for __ffs(), but on 32 bit arches it degenerates to a conditional plus a call to __ffs(). I'm assuming that there would not be a lot of point in optimising this operation on 32 bit arches even if such an instruction was available, so that I should do something like the below patch.

Quoted 6 lines
Quoted 3 lines

Does that seem reasonable, or should I give it a separate header file under asm-generic/bitops/ like some of the similar operations? It looks like I'd have to touch a lot of other files if I were to go that route,

One issue may be that some 32 bit architectures have a better way of doing 64 bit ffs.

Quoted 7 lines

Yes, thats what I was worried about. I don't have a wide enough knowledge of the different architectures to make a judgement about whether this is likely or not.

I guess maybe the right thing to do is to leave it as I did it in the patch and if an arch wants to create its own implementation, then it could be moved at that stage.

Agreed.

Benny

Quoted 9 lines

Steve.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/