Hi Hans, On Tuesday 20 January 2015 14:03:00 Hans Verkuil wrote: > On 01/20/15 13:59, Laurent Pinchart wrote: > > On Tuesday 20 January 2015 10:30:07 Hans Verkuil wrote: > >> On 01/20/15 10:24, Florian Echtler wrote: > >>> On 19.01.2015 11:38, Hans Verkuil wrote: > >>>> Sorry for the delay. > >>> > >>> No problem, thanks for your feedback. > >>> > >>>>> Note: I'm intentionally using dma-contig instead of vmalloc, as the > >>>>> USB core apparently _will_ try to use DMA for larger bulk transfers. > >>>> > >>>> As far as I can tell from looking through the usb core code it supports > >>>> scatter-gather DMA, so you should at least use dma-sg rather than > >>>> dma-contig. Physically contiguous memory should always be avoided. > >>> > >>> OK, will this work transparently (i.e. just switch from *-contig-* to > >>> *-sg-*)? If not, can you suggest an example driver to use as template? > >> > >> Yes, that should pretty much be seamless. BTW, the more I think about it, > >> the more I am convinced that DMA will also be used by the USB core when > >> you use videobuf2-vmalloc. > >> > >> I've CC-ed Laurent, I think he knows a lot more about this than I do. > >> > >> Laurent, when does the USB core use DMA? What do you need to do on the > >> driver side to have USB use DMA when doing bulk transfers? > > > > How USB HCD drivers map buffers for DMA is HCD-specific, but all drivers > > exepct ehci-tegra, max3421-hcd and musb use the default implementation > > usb_hcd_map_urb_for_dma() (in drivers/usb/core/hcd.c). > > > > Unless the buffer has already been mapped by the USB driver (in which case > > the driver will have set the URB_NO_TRANSFER_DMA_MAP flag in > > urb->transfer_flags and initialized the urb->transfer_dma field), the > > function will use dma_map_sg(), dma_map_page() or dma_map_single() > > depending on the buffer type (controlled through urb->sg and > > urb->num_sgs). DMA will thus always be used *expect* if the platform uses > > bounce buffers when the buffer can't be mapped directly for DMA. > > So we can safely use videobuf2-vmalloc, right? That depends on the platform and whether it can DMA to vmalloc'ed memory :-) To be totally safe I think vb2-dma-sg would be better, but I'm not sure it's worth the trouble. uvcvideo uses vb2-vmalloc as it performs a memcpy anyway. > >>>> I'm also missing a patch for the Kconfig that adds a dependency on > >>>> MEDIA_USB_SUPPORT and that selects VIDEOBUF2_DMA_SG. > >>> > >>> Good point, will add that. > >>> > >>>>> +err_unreg_video: > >>>>> + video_unregister_device(&sur40->vdev); > >>>>> +err_unreg_v4l2: > >>>>> + v4l2_device_unregister(&sur40->v4l2); > >>>>> > >>>>> err_free_buffer: > >>>>> kfree(sur40->bulk_in_buffer); > >>>>> > >>>>> err_free_polldev: > >>>>> @@ -436,6 +604,10 @@ static void sur40_disconnect(struct usb_interface > >>>>> *interface)>> > >>>> > >>>> Is this a hardwired device or hotpluggable? If it is hardwired, then > >>>> this code is OK, but if it is hotpluggable, then this isn't good > >>>> enough. > >>> > >>> It's hardwired. Out of curiosity, what would I have to change for a > >>> hotpluggable one? > >> > >> In that case you can't clean everything up since some application might > >> still have a filehandle open. You have to wait until the very last > >> filehandle is closed. > >> > >>>>> + i->type = V4L2_INPUT_TYPE_CAMERA; > >>>>> + i->std = V4L2_STD_UNKNOWN; > >>>>> + strlcpy(i->name, "In-Cell Sensor", sizeof(i->name)); > >>>> > >>>> Perhaps just say "Sensor" here? I'm not sure what "In-Cell" means. > >>> > >>> In-cell is referring to the concept of integrating sensor pixels > >>> directly with LCD pixels, I think it's what Samsung calls it. -- Regards, Laurent Pinchart -- To unsubscribe from this list: send the line "unsubscribe linux-input" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
