Re: blackfin + dmaengine: conflicting define/enum "DMA_COMPLETE"

[Vinod Koul <vinod.koul@intel.com>]

On Thu, Feb 27, 2014 at 10:59:47AM -0500, Oleksandr G Zhadan wrote:
> Hello,
> 
> It's good to "... matched the manual ...", and in this case we can
> match the manual more pedantically, maybe with prefix "HOST".
> 
> In the case of Host DMA port STATUS register:
> 
> From manual :
> .... HOST_STATUS register bits include:
> • DMA Ready (DMA_RDY)
> • FIFO Full (FIFOFULL)
> • FIFO Empty (FIFOEMPTY)
> • DMA Complete (DMA_CMPLT)
> • HOSTDP Handshake (HSHK)
> • HOSTDP Timeout (HOSTDP_TOUT)
> • HOSTDP Interrupt Request (HIRQ).
> • Allow Configurations (ALLOW_CNFG)
> • DMA Direction (DMA_DIR)
> • Bus Timeout Enabled (BTE)
> 
> We could change definitions to something like:
> 
> #define DMA_CMPLT    0x08        /* DMA Complete */
> or
> #define HOST_DMA_CMPLT  0x08    /* DMA Complete */
> 
> And make the similar for other bits/registers.
> 
> Oleks
> 
> On 01/18/2014 02:02 AM, Mike Frysinger wrote:
> >On Saturday 11 January 2014 13:55:15 Marc Kleine-Budde wrote:
> >>On 01/11/2014 07:31 PM, Randy Dunlap wrote:
> >>>On 01/11/2014 10:09 AM, Marc Kleine-Budde wrote:
> >>>>Hello,
> >>>>
> >>>>in current linux-next (and net-next) the compilation of the CAN
> >>>>
> >>>>drivers[1] with ARCH=blackfin fails with:
> >>>>>   CC [M]  drivers/net/can/c_can/c_can.o
> >>>>>
> >>>>>In file included from linux/include/linux/netdevice.h:38:0,
> >>>>>
> >>>>>                  from linux/drivers/net/can/c_can/c_can.c:32:
> >>>>>linux/include/linux/dmaengine.h:55:2: error: expected identifier before
> >>>>>numeric constant linux/include/linux/dmaengine.h: In function
> >>>>>'dma_async_is_complete': linux/include/linux/dmaengine.h:1023:9:
> >>>>>error: 'DMA_IN_PROGRESS' undeclared (first use in this function)
> >>>>>linux/include/linux/dmaengine.h:1023:9: note: each undeclared
> >>>>>identifier is reported only once for each function it appears in
> >>>>There are two locations where DMA_COMPLETE is defined:
> >>>>>arch/blackfin/mach-bf548/include/mach/defBF547.h:602:#define
> >>>>>   DMA_COMPLETE  0x8        /* DMA Complete */
> >>>>>arch/blackfin/mach-bf548/include/mach/defBF544.h:622:#define
> >>>>>    DMA_COMPLETE  0x8        /* DMA Complete */
> >>>>and
> >>>>
> >>>>>include/linux/dmaengine.h-enum dma_status {
> >>>>>include/linux/dmaengine.h:      DMA_COMPLETE,
> >>>>>include/linux/dmaengine.h-      DMA_IN_PROGRESS,
> >>>>>include/linux/dmaengine.h-      DMA_PAUSED,
> >>>>>include/linux/dmaengine.h-      DMA_ERROR,
> >>>>>include/linux/dmaengine.h-};
> >>>>What's the appropriate fix for the problem?
> >>>arch/blackfin/mach-bf548/ needs a less generic name for its macro.
> >>Mike, is there a in tree user of blacksfin's DMA_COMPLETE? I cannot find
> >>anyone.
> >looks like those are defines for the host port peripheral on the BF54x.
> >typically for peripherals we didn't have proper drivers for (like CAN and UART
> >and SPI and such), we left the defines in the headers.  those in turn matched
> >the manual so people coming from other Blackfin environments (and reading the
> >manuals) didn't have to figure out what name the Linux headers used.
> >
> >unfortunately, it leads to cases like this where the names are pretty bad.
> >considering the host peripheral most likely never saw any serious use, it
> >should be fine to delete all the bit defines in those headers related to those
> >registers (i see HOST_{STATUS,CONTROL,TIMEOUT}.
IMHO BFN/BLACKFIN_HOST_{} would be more apt! HOST_{} is too generic and can
again clash with something else!

-- 
~Vinod