Jon Burgess wrote: > On Wed, 2007-05-02 at 02:32 +0200, Oliver Endriss wrote: > > Jon Burgess wrote: > > > Sorry for the delay. I wanted to track down the DMA sync bug before I > > > came back to look at this again. > > > > > > I have added saa7146_vmalloc_destroy_pgtable() which frees the resources > > > > Shouldn't it better be called saa7146_vfree_destroy_pgtable()? > > ^^^^^ > > Sounds good. Implemented in the attached patch. Thanks, commited to HG. > > > allocated by saa7146_vmalloc_build_pgtable() and updated the callers in > > > budget-core.c and av711o.c. I have also been through the updated > > > functions and updated the error paths to ensure they free all allocated > > > resources on error. > > > > > > Also included in this patch are the previous fixes for pci_unmap_sg() > > > and syncing the PCI streamed data to work with a SWIOTLB and match the > > > requirements documented in DMA-API.txt. > > > > Thanks, looks perfect. > > > > @all: > > Any comments? > > > > Btw, why did DMA work without pci_dma_sync_sg_for_cpu() > > on most machines? > > It is probably the CPU cache snooping which happens on x86 to ensure > cache coherency. Most non-x86 architectures require manual cache > writeback and invalidation. Having cache coherency costs transistors, > power, design complexity etc. so RISC architectures avoid it at the > expense of a slightly more complex driver. > > The 2003 OLS presentation "Integrating DMA Into the Generic Device > Model" has some more details and background. > http://licensing.steeleye.com/support/papers/ols_2003_dma.pdf Interesting article, thanks. CU Oliver -- -------------------------------------------------------- VDR Remote Plugin 0.3.9 available at http://www.escape-edv.de/endriss/vdr/ -------------------------------------------------------- _______________________________________________ linux-dvb mailing list linux-dvb@xxxxxxxxxxx http://www.linuxtv.org/cgi-bin/mailman/listinfo/linux-dvb
