Re: RealTek 8139 driver, 2. revision

[Eric Doenges <Eric dot Doenges at DynaPel dot com>]

Michael Anburaj wrote:
> Hi Eric,
>
> Just curious :-)
>
> Line number 1173 of if_8139.c <revision 3>:
>
>
> #ifdef CYGPKG_DEVS_ETH_RLTK_8139_SOFTWARE_CACHE_COHERENCY
>      HAL_DCACHE_INVALIDATE(&rltk8139_info->rx_ring[rx_pos],
>                            sizeof(cyg_uint32));
> #endif
>
>      /*
>       * The 8139 prepends each packet with a 32 bit packet header that
>       * contains a 16 bit length and 16 bit status field, in little-endian
>       * byte order.
>       */
>      header = HAL_LE32TOC(*((volatile cyg_uint32 
> *)CYGARC_UNCACHED_ADDRESS(&rltk8139_info->rx_ring[rx_pos])));
>
>
>
> Here,
> 1st line of code:
> Invalidates the Dcache line for ?header? info <so that, if someone reads 
> it later on; A Dcache line fill will occur & then read out from the 
> Dcache>.
>
> 2nd line of code:
> Reads the ?header? info directly from the memory & not the cache. So, 
> even if Dcache line invalidate (1st line of code) is not there; this 
> would work fine. Am I right?

On platforms were CYGARC_UNCACHED_ADDRESS is supported, you are right.
In this case, defining
CYGPKG_DEVS_ETH_RLTK_8139_SOFTWARE_CACHE_COHERENCY is silly and will
most likely hurt performance. However, on my platform,
CYGARC_UNCACHED_ADDRESS does nothing (it just returns the value passed
in unchanged). Now, I could conditionalize the CYGARC macro and only use
it if software cache coherency is not selected, but it didn't seem
worth the trouble to do this.

> So, is the 1st line (Dcache line invalidate) intended to solve cache 
> coherency problem for later access to header info. from the Dcache. Or 
> am I missing something here?

Yes you are, see above.
--
--------------------------------------------------------------------
|     Eric Doenges              |     DynaPel Laboratories GmbH    |
|     Tel: +49 89 962428 23     |     Fraunhoferstrasse 9/2        |
|     Fax: +49 89 962428 90     |     D - 85737 Ismaning, Germany  |
--------------------------------------------------------------------