Re: do_div considered harmful

[Andrew Morton <akpm@osdl.org>]

Andries.Brouwer@cwi.nl wrote:
>
> Writing this ide capacity patch an hour ago or so
> I split off a helper sectors_to_MB() since Erik's recent
> patch uses this also.
> Now that I compare, he wrote
> 	nativeMb = do_div(nativeMb, 1000000);
> to divide nativeMb by 1000000.
> Similarly, I find in fs/cifs/inode.c
> 	inode->i_blocks = do_div(findData.NumOfBytes, inode->i_blksize);

This should be

	int blocksize = 1 << inode->i_blkbits;

	inode->i_blocks = (findData.NumOfBytes + blocksize - 1)
				>> inode->i_blkbits;
				

and inode.i_blksize should probably be removed from the kernel.

> So, it seems natural to expect that do_div() gives the quotient.
> But it gives the remainder.
> (Strange, Erik showed correct output.)
> 
> Since the semantics of this object are very unlike that of a C function,
> I wonder whether we should write DO_DIV instead, or DO_DIV_AND_REM
> to show that a remainder is returned.

Sometimes the slash-star operator comes in handy.


--- 25/include/asm-i386/div64.h~do_div-comment	2003-08-03 19:20:58.000000000 -0700
+++ 25-akpm/include/asm-i386/div64.h	2003-08-03 19:21:11.000000000 -0700
@@ -1,6 +1,16 @@
 #ifndef __I386_DIV64
 #define __I386_DIV64
 
+/*
+ * The semantics of do_div() are:
+ *
+ * uint32_t do_div(uint64_t *n, uint32_t base)
+ * {
+ * 	uint32_t remainder = *n % base;
+ * 	*n = *n / base;
+ * 	return remainder;
+ * }
+ */
 #define do_div(n,base) ({ \
 	unsigned long __upper, __low, __high, __mod; \
 	asm("":"=a" (__low), "=d" (__high):"A" (n)); \

_