Re: [PATCH net-next 08/11] ath9k: work around false-positive gcc warning

[Johannes Berg]

On Mon, 2020-11-02 at 18:26 +0200, Kalle Valo wrote:
> Arnd Bergmann <arnd@xxxxxxxxxx> writes:
> 
> > From: Arnd Bergmann <arnd@xxxxxxxx>
> > 
> > gcc-10 shows a false-positive warning with CONFIG_KASAN:
> > 
> > drivers/net/wireless/ath/ath9k/dynack.c: In function 'ath_dynack_sample_tx_ts':
> > include/linux/etherdevice.h:290:14: warning: writing 4 bytes into a region of size 0 [-Wstringop-overflow=]
> >   290 |  *(u32 *)dst = *(const u32 *)src;
> >       |  ~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~
> > 
> > Until gcc is fixed, work around this by using memcpy() in place
> > of ether_addr_copy(). Hopefully gcc-11 will not have this problem.
> > 
> > Link: https://godbolt.org/z/sab1MK
> > Link: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97490
> > Signed-off-by: Arnd Bergmann <arnd@xxxxxxxx>
> > ---
> >  drivers/net/wireless/ath/ath9k/dynack.c | 6 ++++++
> >  1 file changed, 6 insertions(+)
> > 
> > diff --git a/drivers/net/wireless/ath/ath9k/dynack.c b/drivers/net/wireless/ath/ath9k/dynack.c
> > index fbeb4a739d32..e4eb96b26ca4 100644
> > --- a/drivers/net/wireless/ath/ath9k/dynack.c
> > +++ b/drivers/net/wireless/ath/ath9k/dynack.c
> > @@ -247,8 +247,14 @@ void ath_dynack_sample_tx_ts(struct ath_hw *ah, struct sk_buff *skb,
> >  	ridx = ts->ts_rateindex;
> >  
> >  	da->st_rbf.ts[da->st_rbf.t_rb].tstamp = ts->ts_tstamp;
> > +#if defined(CONFIG_KASAN) && (CONFIG_GCC_VERSION >= 100000) && (CONFIG_GCC_VERSION < 110000)
> > +	/* https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97490 */
> > +	memcpy(da->st_rbf.addr[da->st_rbf.t_rb].h_dest, hdr->addr1, ETH_ALEN);
> > +	memcpy(da->st_rbf.addr[da->st_rbf.t_rb].h_src, hdr->addr2, ETH_ALEN);
> > +#else
> >  	ether_addr_copy(da->st_rbf.addr[da->st_rbf.t_rb].h_dest, hdr->addr1);
> >  	ether_addr_copy(da->st_rbf.addr[da->st_rbf.t_rb].h_src, hdr->addr2);
> > +#endif
> 
> Isn't there a better way to handle this? I really would not want
> checking for GCC versions become a common approach in drivers.
> 
> I even think that using memcpy() always is better than the ugly ifdef.

If you put memcpy() always somebody will surely go and clean it up to
use ether_addr_copy() soon ...

That said, if there's a gcc issue with ether_addr_copy() then how come
it's specific to this place?

johannes