[PATCH] fs/squashfs: use do_div function for math

[Sean Nyekjaer]

On Thu, May 12, 2022 at 08:41:15PM +0200, Pali Rohár wrote:
> On Thursday 12 May 2022 20:02:33 Sean Nyekjaer wrote:
> > On Thu, May 12, 2022 at 12:24:01PM +0200, Pali Rohár wrote:
> > > On Thursday 12 May 2022 12:19:01 Pali Rohár wrote:
> > > > On Monday 09 May 2022 15:08:15 Miquel Raynal wrote:
> > > > > Hello,
> > > > > 
> > > > > trini at konsulko.com wrote on Thu, 5 May 2022 09:15:00 -0400:
> > > > > 
> > > > > > On Thu, May 05, 2022 at 11:26:37AM +0200, Sean Nyekjaer wrote:
> > > > > > 
> > > > > > > From: Sean Nyekjaer <sean.nyekjaer.ext at siemensgamesa.com>
> > > > > > > 
> > > > > > > When compling for x86:
> > > > > > > ld.bfd: fs/squashfs/sqfs.o: in function `sqfs_read':
> > > > > > > u-boot/fs/squashfs/sqfs.c:1443: undefined reference to `__udivmoddi4'
> > > > > > > ld.bfd: u-boot/fs/squashfs/sqfs.c:1521: undefined reference to `__udivmoddi4'
> > > > > > > 
> > > > > > > Signed-off-by: Sean Nyekjaer <sean.nyekjaer.ext at siemensgamesa.com>
> > > > > > > ---
> > > > > > >  fs/squashfs/sqfs.c | 5 +++--
> > > > > > >  1 file changed, 3 insertions(+), 2 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/fs/squashfs/sqfs.c b/fs/squashfs/sqfs.c
> > > > > > > index 5d9c52af80..6405db4ff3 100644
> > > > > > > --- a/fs/squashfs/sqfs.c
> > > > > > > +++ b/fs/squashfs/sqfs.c
> > > > > > > @@ -8,6 +8,7 @@
> > > > > > >   */
> > > > > > >  
> > > > > > >  #include <asm/unaligned.h>
> > > > > > > +#include <div64.h>
> > > > > > >  #include <errno.h>
> > > > > > >  #include <fs.h>
> > > > > > >  #include <linux/types.h>
> > > > > > > @@ -1440,7 +1441,7 @@ int sqfs_read(const char *filename, void *buf, loff_t offset, loff_t len,
> > > > > > >  	}
> > > > > > >  
> > > > > > >  	for (j = 0; j < datablk_count; j++) {
> > > > > > > -		start = data_offset / ctxt.cur_dev->blksz;
> > > > > > > +		start = do_div(data_offset, ctxt.cur_dev->blksz);
> > > > > > >  		table_size = SQFS_BLOCK_SIZE(finfo.blk_sizes[j]);
> > > > > > >  		table_offset = data_offset - (start * ctxt.cur_dev->blksz);
> > > > > > >  		n_blks = DIV_ROUND_UP(table_size + table_offset,
> > > > > > > @@ -1516,7 +1517,7 @@ int sqfs_read(const char *filename, void *buf, loff_t offset, loff_t len,
> > > > > > >  		goto out;
> > > > > > >  	}
> > > > > > >  
> > > > > > > -	start = frag_entry.start / ctxt.cur_dev->blksz;
> > > > > > > +	start = do_div(frag_entry.start, ctxt.cur_dev->blksz);
> > > > > > >  	table_size = SQFS_BLOCK_SIZE(frag_entry.size);
> > > > > > >  	table_offset = frag_entry.start - (start * ctxt.cur_dev->blksz);
> > > > > > >  	n_blks = DIV_ROUND_UP(table_size + table_offset, ctxt.cur_dev->blksz);  
> > > > > > 
> > > > > > Adding maintainers...
> > > > > 
> > > > > I guess that's a correct fix, so
> > > > > 
> > > > > Reviewed-by: Miquel Raynal <miquel.raynal at bootlin.com>
> > > > > 
> > > > > Thanks,
> > > > > Miquèl
> > > > 
> > > > Hello! do_div() macro modifies its first argument. It is correct?
> > > > Because previous code did not modify anything.
> > > 
> > >  * The semantics of do_div() are:
> > >  *
> > >  * u32 do_div(u64 *n, u32 base)
> > >  * {
> > >  *	u32 remainder = *n % base;
> > >  *	*n = *n / base;
> > >  *	return remainder;
> > >  * }
> > > 
> > > And this patch replaced:
> > > 
> > >   start = data_offset / ctxt.cur_dev->blksz;
> > > 
> > > by:
> > > 
> > >   start = do_div(data_offset, ctxt.cur_dev->blksz);
> > > 
> > > which expands to:
> > > 
> > >   start = data_offset % ctxt.cur_dev->blksz;
> > >   data_offset = data_offset / ctxt.cur_dev->blksz;
> > > 
> > > Which is wrong as it calculates something totally different.
> > > 
> > > So this patch is incorrect.
> > 
> > Hi Pali,
> > 
> > Yes I agree, I got tricked by some wrong cache from Yocto :/
> > I have tried lldiv(), and that works on our x86 target.
> > 
> > Do you have an opinion on that?
> > Or do you have another idea to avoid the
> > "undefined reference to `__udivmoddi4'" error?
> > 
> > /Sean
> 
> Hello! You really should not use C division operator for 64-bit integers
> as gcc still is not able to generate assembler code for such division
> on many 32-bit platforms.
> 
> You should use some function or macro from linux/math64.h or div64.h
> header file. For every combination of types, pairs and required
> operation there is specific function/macro (e.g. u64/u32, u64%u32,
> u64/u64, etc...).
> 
> I think that lldiv() is the correct call here. This function expects u64
> as dividend, u32 as divisor and returns u64 result.

Thank you for the explanation :)
Updated patch should be on the list.

/Sean