On Mon, Aug 03, 2009 at 01:23:09PM +0200, Sven Neumann wrote: > Hi, > > On Mon, 2009-08-03 at 11:39 +0200, Denis Oliver Kropp wrote: > > Sven Neumann wrote: > > > Hi, > > > > > > On Mon, 2009-07-27 at 12:50 +0200, Denis Oliver Kropp wrote: > > > > > >>>> commit fbf15dbbba6398850b54112ace094d019351a44a > > >>>> Author: Claudio Ciccani <k...@directfb.org> > > >>>> Date: Fri Aug 12 11:00:56 2005 +0000 > > >>>> > > >>>> system_get_accelerator() returns -1 if mmio memory lenght is 0. > > >>> I've checked the changes Claudio was working on at the time he did that > > >>> change. Looks like he was working on the nvidia and the r200 drivers. > > >>> But I still can't figure out what this change is good for. > > >>> > > >>> What are our options now? I'd like to get rid of this questionable > > >>> check. But of course I could also introduce yet another hack in > > >>> system_get_accelerator() and special case the PXA3XX hardware. Any > > >>> opinions? > > >> I think he did that to avoid usage of the accelerated driver when the > > >> MMIO region is not exported. > > >> > > >> The check for the mmio length should move into the driver. > > > > > > Which drivers should have this check added? > > > > I think ATI and/or nVidia, but basically all with MMIO. > > As far as I can see no extra code should be needed. The drivers that I > checked call dfb_gfxcard_map_mmio() from driver_init_driver(). They pass > -1 as length argument. In system_map_mmio() in fbdev.c this value is > then replaced by dfb_fbdev->shared->fix.mmio_len. In case this is zero, > mmap() will be called with a size of zero and according to the manual > page, this is supposed to fail.
My manual page says linux < 2.6.12 will return addr if length == 0. But since system_map_mmio() will return NULL on failure too it shouldn't matter assuming that mmio_start is page aligned, but I guess it should generally be 0 if mmio_len is 0. > A failure will cause > driver_init_driver() to return DFB_IO. I think failure at that point will terminate the core init. So the user would be forced to use disable-module to get DFB up and running. -- Ville Syrjälä syrj...@sci.fi http://www.sci.fi/~syrjala/ _______________________________________________ directfb-dev mailing list directfb-dev@directfb.org http://mail.directfb.org/cgi-bin/mailman/listinfo/directfb-dev
