From: Gianluca Gennari Date: Tue, 14 Feb 2012 12:25:00 +0000 (-0300) Subject: [media] as102: map URB DMA addresses in the driver X-Git-Url: https://git.stricted.de/?a=commitdiff_plain;h=f8f7be019ac801950512cfb040949ccc380be301;p=GitHub%2Fmoto-9609%2Fandroid_kernel_motorola_exynos9610.git [media] as102: map URB DMA addresses in the driver On a set-top-box based on the Broadcom 7405 SoC (MIPS), the Abilis as102 driver causes a kernel oops while trying to map the URB stream buffers DMA addresses: CPU 0 Unable to handle kernel paging request at virtual address 007b9900, epc == 80010cc4, ra == 8039d108 Call Trace: [<80010cc4>] mips_dma_map_page+0x14/0x108 [<8039d108>] usb_hcd_map_urb_for_dma+0x338/0x4a8 [<8039d540>] usb_hcd_submit_urb+0x2c8/0x8cc [] as102_submit_urb_stream+0x64/0xc8 [dvb_as102] [] as102_usb_start_stream+0x44/0x80 [dvb_as102] [] as102_dvb_dmx_start_feed+0xb4/0x17c [dvb_as102] [<803e20f4>] dmx_ts_feed_start_filtering+0x5c/0x134 [<803de454>] dvb_dmxdev_start_feed+0xd4/0x158 [<803dff28>] dvb_dmxdev_filter_start+0x2b8/0x448 [<803e07ac>] dvb_demux_do_ioctl+0x2a0/0x654 [<803ddc8c>] dvb_usercopy+0x124/0x204 [<800d5284>] do_vfs_ioctl+0xa0/0x6c0 [<800d58e8>] sys_ioctl+0x44/0xa8 [<8000ecfc>] stack_done+0x20/0x40 On other boxes based on older SoCs (7401) this doesn't happen, so it looks like a bug in the kernel specific to MIPS SMP. This issue has been reproduced on several kernel versions from 2.6.18 to 3.1.0. Since the base DMA address and the offsets are known, it is possible to map the DMA addresses of the URB buffers directly in the driver. This workaround fixes the problem and has been tested on both MIPS and x86 CPUs with success. By the way, with this fix the driver works perfectly fine on the set-top-box: both UHF and VHF frequencies are tuned without problems, and zapping is quite fast. SNR and signal strength reports seems to work fine, too. The only remaining problem (on both the PC and the set-top-box) is that after a soft reboot the device is not recognized again by the kernel. It requires a power cycle (or a manual unplug/replug) to be recognized again. So probably the device state is not reset properly at shut-down. Signed-off-by: Gianluca Gennari Signed-off-by: Mauro Carvalho Chehab --- diff --git a/drivers/staging/media/as102/as102_usb_drv.c b/drivers/staging/media/as102/as102_usb_drv.c index d775be0173ea..8d2c84c36a5c 100644 --- a/drivers/staging/media/as102/as102_usb_drv.c +++ b/drivers/staging/media/as102/as102_usb_drv.c @@ -270,6 +270,8 @@ static int as102_alloc_usb_stream_buffer(struct as102_dev_t *dev) } urb->transfer_buffer = dev->stream + (i * AS102_USB_BUF_SIZE); + urb->transfer_dma = dev->dma_addr + (i * AS102_USB_BUF_SIZE); + urb->transfer_flags = URB_NO_TRANSFER_DMA_MAP; urb->transfer_buffer_length = AS102_USB_BUF_SIZE; dev->stream_urb[i] = urb;