On Tue, Oct 11, 2016 at 07:09:17AM -0300, Mauro Carvalho Chehab wrote:
> --- a/drivers/media/usb/dvb-usb/cinergyT2-core.c
> +++ b/drivers/media/usb/dvb-usb/cinergyT2-core.c
> @@ -41,6 +41,8 @@ DVB_DEFINE_MOD_OPT_ADAPTER_NR(adapter_nr);
>  
>  struct cinergyt2_state {
>       u8 rc_counter;
> +     unsigned char data[64];
> +     struct mutex data_mutex;
>  };

Sometimes my thinking is slow but it just occured to me
that this creates a potential issue with cache line sharing.
On an architecture which manages cache coherence in software
(ARM, MIPS etc.) a write to e.g. rc_counter in this example
would dirty the cache line, and a later writeback from the
cache could overwrite parts of data[] which was received via DMA.
In contrast, if the DMA buffer is allocated seperately via
kmalloc it is guaranteed to be safe wrt cache line sharing.
(see bottom of Documentation/DMA-API-HOWTO.txt).

But of course DMA on stack also had the same issue
and no one ever noticed so it's apparently not critical...


Johannes
--
To unsubscribe from this list: send the line "unsubscribe linux-media" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Reply via email to