linux-kernel - Re: [PATCH]staging: fbtft: Fix sparse warnings about endianness

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Tue, 14 Feb 2017 18:57:20 +0000
From:   Al Viro <viro@...IV.linux.org.uk>
To:     Greg KH <gregkh@...uxfoundation.org>
Cc:     maomao xu <albert008.xu@...il.com>,
        thomas.petazzoni@...e-electrons.com, noralf@...nnes.org,
        devel@...verdev.osuosl.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH]staging: fbtft: Fix sparse warnings about endianness

On Tue, Feb 14, 2017 at 09:18:25AM -0800, Greg KH wrote:
> On Tue, Feb 14, 2017 at 04:53:10PM +0800, maomao xu wrote:
> >     Fixed sparse warnings about endianness. E.g.:
> > 
> >     warning: incorrect type in assignment (different base types)
> 
> Why are these lines indented?
> 
> > Signed-off-by: maomao xu <albert008.xu@...il.com>
> > 
> > diff --git a/drivers/staging/fbtft/fbtft-io.c b/drivers/staging/fbtft/fbtft-io.c
> > index d868405..ffb9a3b 100644
> > --- a/drivers/staging/fbtft/fbtft-io.c
> > +++ b/drivers/staging/fbtft/fbtft-io.c
> > @@ -71,7 +71,7 @@ int fbtft_write_spi_emulate_9(struct fbtft_par *par, void *buf, size_t len)
> >  			src++;
> >  		}
> >  		tmp |= ((*src & 0x0100) ? 1 : 0);
> > -		*(u64 *)dst = cpu_to_be64(tmp);
> > +		*(__be64 *)dst = cpu_to_be64(tmp);
> 
> Really?  That seems very odd.  I need a maintainer's ack for this before
> I can take it...

If anything, I would turn the inner loop into
                tmp = 0;
                for (j = 0; j < 7; j++) {
			tmp <<= 9;
			tmp |= *src++ & 0x1ff;
                }
		tmp <<= 1;
there - the whole thing looks like an obfuscated "take an array of 9-bit
values, each stored in host-endian 16bit, repack them into octets".
Which architecture is that supposed to run on?  Anything that doesn't like
unaligned stores won't be happy with that code...

Another interesting part is that we have size = len / 2 and verify that len
is a multiple of 8.  I.e. that size is only guaranteed to be a multiple
of 4.  Each pass through the outer loop consumes 8 16bit values and decrements
size by 8, so if len is e.g. 24 we'll end up accepting that, get size equal
to 12, pass through the outer loop twice and chew through 32 bytes of buf...