Re: [PATCH] usbvision: remove (broken) image format conversion
Hi, On 06/11/2011 02:13 PM, Mauro Carvalho Chehab wrote: Hi Hans de G., Em 26-04-2011 08:54, Hans de Goede escreveu: If you could give it a shot that would be great. I've some hardware to test this with (although I've never actually tested that hardware), so I can hopefully pick things up if you cannot finish things before you need to return the hardware. As Ondrej couldn't work on that while he was with the hardware, could you please try to address this issue? I've put it on my to do list, not sure when I'll get around to it though. Regards, Hans -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] usbvision: remove (broken) image format conversion
Hi Hans de G., Em 26-04-2011 08:54, Hans de Goede escreveu: If you could give it a shot that would be great. I've some hardware to test this with (although I've never actually tested that hardware), so I can hopefully pick things up if you cannot finish things before you need to return the hardware. As Ondrej couldn't work on that while he was with the hardware, could you please try to address this issue? Thanks, Mauro Em 03-05-2011 13:37, Ondrej Zary escreveu: On Tuesday 03 May 2011 12:29:45 Mauro Carvalho Chehab wrote: Em 26-04-2011 17:40, Ondrej Zary escreveu: On Tuesday 26 April 2011 14:33:20 Hans Verkuil wrote: After digging in the code for hours, I'm giving this up. It's not worth it. The ISOC_MODE_YUV422 mode works as V4L2_PIX_FMT_YVYU with VLC and mplayer+libv4lconvert, reducing the loop (and dropping strech_*) in usbvision_parse_lines_422() to: scratch_get(usbvision, frame-data + (frame-v4l2_linesize * frame-curline), 2 * frame-frmwidth); The ISOC_MODE_YUV420 is some weird custom format with 64-byte lines of YYUV. usbvision_parse_lines_420() is real mess with that scratch_* crap everywhere. ISOC_MODE_COMPRESS: There are callbacks to usbvision_request_intra() and also usbvision_adjust_compression(). This is not going to work outside the kernel. So I can redo the conversion removal patch to keep the RGB formats and also provide another one to remove the testpattern (it oopses too). But I'm not going to do any major changes in the driver. While in a perfect world, this should be moved to userspace, I'm ok on keeping it there, but the OOPS/Panic conditions should be fixed. Could you please work on a patch fixing the broken stuff, without removing the conversions? I've already returned the hardware so I can't test the driver anymore. Did the YUV422P conversion ever work? The initialization of u and v pointers is missing in usbvision_parse_compress() function for YUV422P case. The following oops was captured when trying to fix YVU420. Seems to be related to the u pointer too... [ 181.169233] usbvision_parse_compress before conversion [ 181.169233] usbvision_parse_compress idx=0, format=842094169 [ 181.169233] usbvision_parse_compress YVU420 f=e088d000, Y=e0a3e000 [ 181.169233] usbvision_parse_compress YVU420 frame=df016b68 [ 181.169233] usbvision_parse_compress YVU420 frame-curline=0 [ 181.169233] usbvision_parse_compress YVU420 u=e08a4700, U=e0a50c00, v=e089fc00, V=e0a55700 [ 181.169233] BUG: unable to handle kernel paging request at e08a4700 [ 181.169233] IP: [e097fd57] usbvision_parse_compress+0x599/0x76e [usbvision] [ 181.169233] *pde = 1f0c2067 *pte = [ 181.169233] Oops: 0002 [#1] SMP [ 181.169233] last sysfs file: /sys/devices/pci:00/:00:1d.0/usb2/2-1/i2c-0/uevent [ 181.169233] Modules linked in: savage drm loop usbvision v4l2_common snd_fm801 snd_tea575x_tuner videodev snd_intel8x0 e [ 181.169233] [ 181.169233] Pid: 0, comm: swapper Not tainted 2.6.39-rc2 #2/848P-ICH5 [ 181.169233] EIP: 0060:[e097fd57] EFLAGS: 00010046 CPU: 0 [ 181.169233] EIP is at usbvision_parse_compress+0x599/0x76e [usbvision] [ 181.169233] EAX: EBX: df016b68 ECX: e08a4700 EDX: e0a50c81 [ 181.169233] ESI: EDI: e088d001 EBP: 00a0 ESP: df021cb4 [ 181.169233] DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 [ 181.169233] Process swapper (pid: 0, ti=df02 task=c12dde60 task.ti=c12b) [ 181.169233] Stack: [ 181.169233] e0986ed3 e09861a4 e08a4700 e0a50c00 e089fc00 e0a55700 0003 e089fc00 [ 181.169233] e08a4700 00c0 0140 df021f14 df021d52 0506 00060027 0009 [ 181.169233] 00c0 3293 00605a25 ff00605a fff0 55d501fc 82995568 a0060124 [ 181.169233] Call Trace: [ 181.169233] [e0c9c69d] ? uhci_urb_enqueue+0x712/0x725 [uhci_hcd] [ 181.169233] [e08f4155] ? usb_hcd_submit_urb+0x4be/0x53d [usbcore] [ 181.169233] [c11e9840] ? printk+0xe/0x16 [ 181.169233] [e0980486] ? usbvision_isoc_irq+0x55a/0x17a0 [usbvision] [ 181.169233] [e097f5f9] ? usbvision_write_reg_irq+0xd5/0x10f [usbvision] [ 181.169233] [e0981505] ? usbvision_isoc_irq+0x15d9/0x17a0 [usbvision] [ 181.169233] [c1026eb2] ? try_to_wake_up+0x13b/0x13b [ 181.169233] [c117057c] ? ata_scsi_qc_complete+0x2b4/0x2c2 [ 181.169233] [c101e899] ? __wake_up+0x2c/0x3b [ 181.169233] [e08f33e5] ? usb_hcd_giveback_urb+0x46/0x71 [usbcore] [ 181.169233] [e0c9a98d] ? uhci_giveback_urb+0xea/0x15d [uhci_hcd] [ 181.169233] [e0c9aff7] ? uhci_scan_schedule+0x526/0x777 [uhci_hcd] [ 181.169233] [c117951b] ? __ata_sff_port_intr+0x97/0xa2 [ 181.169233] [e0c9c9f8] ? uhci_irq+0xbf/0xcd [uhci_hcd] [ 181.169233] [e08f2d2e] ? usb_hcd_irq+0x1e/0x5f [usbcore] [ 181.169233] [c1058581] ? handle_irq_event_percpu+0x1e/0x106 [ 181.169233] [c1059edc] ? handle_edge_irq+0xa0/0xa0 [ 181.169233] [c105868a] ? handle_irq_event+0x21/0x37 [ 181.169233] [c1059edc] ?
Re: [PATCH] usbvision: remove (broken) image format conversion
Em 26-04-2011 17:40, Ondrej Zary escreveu: On Tuesday 26 April 2011 14:33:20 Hans Verkuil wrote: After digging in the code for hours, I'm giving this up. It's not worth it. The ISOC_MODE_YUV422 mode works as V4L2_PIX_FMT_YVYU with VLC and mplayer+libv4lconvert, reducing the loop (and dropping strech_*) in usbvision_parse_lines_422() to: scratch_get(usbvision, frame-data + (frame-v4l2_linesize * frame-curline), 2 * frame-frmwidth); The ISOC_MODE_YUV420 is some weird custom format with 64-byte lines of YYUV. usbvision_parse_lines_420() is real mess with that scratch_* crap everywhere. ISOC_MODE_COMPRESS: There are callbacks to usbvision_request_intra() and also usbvision_adjust_compression(). This is not going to work outside the kernel. So I can redo the conversion removal patch to keep the RGB formats and also provide another one to remove the testpattern (it oopses too). But I'm not going to do any major changes in the driver. While in a perfect world, this should be moved to userspace, I'm ok on keeping it there, but the OOPS/Panic conditions should be fixed. Could you please work on a patch fixing the broken stuff, without removing the conversions? Thanks, Mauro -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] usbvision: remove (broken) image format conversion
On Tuesday 03 May 2011 12:29:45 Mauro Carvalho Chehab wrote: Em 26-04-2011 17:40, Ondrej Zary escreveu: On Tuesday 26 April 2011 14:33:20 Hans Verkuil wrote: After digging in the code for hours, I'm giving this up. It's not worth it. The ISOC_MODE_YUV422 mode works as V4L2_PIX_FMT_YVYU with VLC and mplayer+libv4lconvert, reducing the loop (and dropping strech_*) in usbvision_parse_lines_422() to: scratch_get(usbvision, frame-data + (frame-v4l2_linesize * frame-curline), 2 * frame-frmwidth); The ISOC_MODE_YUV420 is some weird custom format with 64-byte lines of YYUV. usbvision_parse_lines_420() is real mess with that scratch_* crap everywhere. ISOC_MODE_COMPRESS: There are callbacks to usbvision_request_intra() and also usbvision_adjust_compression(). This is not going to work outside the kernel. So I can redo the conversion removal patch to keep the RGB formats and also provide another one to remove the testpattern (it oopses too). But I'm not going to do any major changes in the driver. While in a perfect world, this should be moved to userspace, I'm ok on keeping it there, but the OOPS/Panic conditions should be fixed. Could you please work on a patch fixing the broken stuff, without removing the conversions? I've already returned the hardware so I can't test the driver anymore. Did the YUV422P conversion ever work? The initialization of u and v pointers is missing in usbvision_parse_compress() function for YUV422P case. The following oops was captured when trying to fix YVU420. Seems to be related to the u pointer too... [ 181.169233] usbvision_parse_compress before conversion [ 181.169233] usbvision_parse_compress idx=0, format=842094169 [ 181.169233] usbvision_parse_compress YVU420 f=e088d000, Y=e0a3e000 [ 181.169233] usbvision_parse_compress YVU420 frame=df016b68 [ 181.169233] usbvision_parse_compress YVU420 frame-curline=0 [ 181.169233] usbvision_parse_compress YVU420 u=e08a4700, U=e0a50c00, v=e089fc00, V=e0a55700 [ 181.169233] BUG: unable to handle kernel paging request at e08a4700 [ 181.169233] IP: [e097fd57] usbvision_parse_compress+0x599/0x76e [usbvision] [ 181.169233] *pde = 1f0c2067 *pte = [ 181.169233] Oops: 0002 [#1] SMP [ 181.169233] last sysfs file: /sys/devices/pci:00/:00:1d.0/usb2/2-1/i2c-0/uevent [ 181.169233] Modules linked in: savage drm loop usbvision v4l2_common snd_fm801 snd_tea575x_tuner videodev snd_intel8x0 e [ 181.169233] [ 181.169233] Pid: 0, comm: swapper Not tainted 2.6.39-rc2 #2/848P-ICH5 [ 181.169233] EIP: 0060:[e097fd57] EFLAGS: 00010046 CPU: 0 [ 181.169233] EIP is at usbvision_parse_compress+0x599/0x76e [usbvision] [ 181.169233] EAX: EBX: df016b68 ECX: e08a4700 EDX: e0a50c81 [ 181.169233] ESI: EDI: e088d001 EBP: 00a0 ESP: df021cb4 [ 181.169233] DS: 007b ES: 007b FS: 00d8 GS: 00e0 SS: 0068 [ 181.169233] Process swapper (pid: 0, ti=df02 task=c12dde60 task.ti=c12b) [ 181.169233] Stack: [ 181.169233] e0986ed3 e09861a4 e08a4700 e0a50c00 e089fc00 e0a55700 0003 e089fc00 [ 181.169233] e08a4700 00c0 0140 df021f14 df021d52 0506 00060027 0009 [ 181.169233] 00c0 3293 00605a25 ff00605a fff0 55d501fc 82995568 a0060124 [ 181.169233] Call Trace: [ 181.169233] [e0c9c69d] ? uhci_urb_enqueue+0x712/0x725 [uhci_hcd] [ 181.169233] [e08f4155] ? usb_hcd_submit_urb+0x4be/0x53d [usbcore] [ 181.169233] [c11e9840] ? printk+0xe/0x16 [ 181.169233] [e0980486] ? usbvision_isoc_irq+0x55a/0x17a0 [usbvision] [ 181.169233] [e097f5f9] ? usbvision_write_reg_irq+0xd5/0x10f [usbvision] [ 181.169233] [e0981505] ? usbvision_isoc_irq+0x15d9/0x17a0 [usbvision] [ 181.169233] [c1026eb2] ? try_to_wake_up+0x13b/0x13b [ 181.169233] [c117057c] ? ata_scsi_qc_complete+0x2b4/0x2c2 [ 181.169233] [c101e899] ? __wake_up+0x2c/0x3b [ 181.169233] [e08f33e5] ? usb_hcd_giveback_urb+0x46/0x71 [usbcore] [ 181.169233] [e0c9a98d] ? uhci_giveback_urb+0xea/0x15d [uhci_hcd] [ 181.169233] [e0c9aff7] ? uhci_scan_schedule+0x526/0x777 [uhci_hcd] [ 181.169233] [c117951b] ? __ata_sff_port_intr+0x97/0xa2 [ 181.169233] [e0c9c9f8] ? uhci_irq+0xbf/0xcd [uhci_hcd] [ 181.169233] [e08f2d2e] ? usb_hcd_irq+0x1e/0x5f [usbcore] [ 181.169233] [c1058581] ? handle_irq_event_percpu+0x1e/0x106 [ 181.169233] [c1059edc] ? handle_edge_irq+0xa0/0xa0 [ 181.169233] [c105868a] ? handle_irq_event+0x21/0x37 [ 181.169233] [c1059edc] ? handle_edge_irq+0xa0/0xa0 [ 181.169233] [c1059f42] ? handle_fasteoi_irq+0x66/0x7e [ 181.169233] IRQ [ 181.169233] [c10035c7] ? do_IRQ+0x2e/0x84 [ 181.169233] [c11ec370] ? common_interrupt+0x30/0x38 [ 181.169233] [c1007699] ? mwait_idle+0x4f/0x54 [ 181.169233] [c1001a86] ? cpu_idle+0x91/0xab [ 181.169233] [c12f96d4] ? start_kernel+0x2a3/0x2a8 [ 181.169233] Code: 08 ff 35 8c a1 98 e0 ff 74 24 14 68 a4 61 98 e0 68 d3 6e 98 e0 e8 ec 9a 86 e0 8b 15 8c a1 98 e0 89 f0 [ 181.169233] 11 8b 15 88 a1 98 e0 41 89 4c 24
Re: [PATCH] usbvision: remove (broken) image format conversion
On Monday, April 25, 2011 23:23:17 Ondrej Zary wrote: The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). What really should happen is that the conversion is moved to libv4lconvert. I've never had the time to tackle that, but it would improve this driver a lot. Would you perhaps be interested in doing that work? The removal also reveals an off-by-one bug in enum_fmt ioctl - it misses the last format, so this patch fixes it too. Good. But why are the GREY/RGB formats also removed? Are those broken as well? Regards, Hans This allows the driver to work with mplayer without need to manually specify a format and also to work with VLC without causing kernel panic. Signed-off-by: Ondrej Zary li...@rainbow-software.org diff -up linux-2.6.39-rc2-/drivers/media/video/usbvision/usbvision-core.c linux-2.6.39-rc2/drivers/media/video/usbvision/usbvision-core.c --- linux-2.6.39-rc2-/drivers/media/video/usbvision/usbvision-core.c 2011-04-25 22:30:09.0 +0200 +++ linux-2.6.39-rc2/drivers/media/video/usbvision/usbvision-core.c 2011-04-25 23:07:46.0 +0200 @@ -586,9 +586,8 @@ static enum parse_state usbvision_parse_ int len; int i; unsigned char yuyv[4] = { 180, 128, 10, 128 }; /* YUV components */ - unsigned char rv, gv, bv; /* RGB components */ - int clipmask_index, bytes_per_pixel; - int stretch_bytes, clipmask_add; + int bytes_per_pixel; + int stretch_bytes; frame = usbvision-cur_frame; f = frame-data + (frame-v4l2_linesize * frame-curline); @@ -605,78 +604,16 @@ static enum parse_state usbvision_parse_ bytes_per_pixel = frame-v4l2_format.bytes_per_pixel; stretch_bytes = (usbvision-stretch_width - 1) * bytes_per_pixel; - clipmask_index = frame-curline * MAX_FRAME_WIDTH; - clipmask_add = usbvision-stretch_width; for (i = 0; i frame-frmwidth; i += (2 * usbvision-stretch_width)) { scratch_get(usbvision, yuyv[0], 4); - if (frame-v4l2_format.format == V4L2_PIX_FMT_YUYV) { - *f++ = yuyv[0]; /* Y */ - *f++ = yuyv[3]; /* U */ - } else { - YUV_TO_RGB_BY_THE_BOOK(yuyv[0], yuyv[1], yuyv[3], rv, gv, bv); - switch (frame-v4l2_format.format) { - case V4L2_PIX_FMT_RGB565: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x07 (gv 3)) | - (0xF8 bv); - break; - case V4L2_PIX_FMT_RGB24: - *f++ = rv; - *f++ = gv; - *f++ = bv; - break; - case V4L2_PIX_FMT_RGB32: - *f++ = rv; - *f++ = gv; - *f++ = bv; - f++; - break; - case V4L2_PIX_FMT_RGB555: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x03 (gv 3)) | - (0x7C (bv 2)); - break; - } - } - clipmask_index += clipmask_add; + *f++ = yuyv[0]; /* Y */ + *f++ = yuyv[3]; /* U */ f += stretch_bytes; - if (frame-v4l2_format.format == V4L2_PIX_FMT_YUYV) { - *f++ = yuyv[2]; /* Y */ - *f++ = yuyv[1]; /* V */ - } else { - YUV_TO_RGB_BY_THE_BOOK(yuyv[2], yuyv[1], yuyv[3], rv, gv, bv); - switch (frame-v4l2_format.format) { - case V4L2_PIX_FMT_RGB565: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x07 (gv 3)) | - (0xF8 bv); - break; - case V4L2_PIX_FMT_RGB24: - *f++ = rv; - *f++ = gv; - *f++ = bv; - break; - case V4L2_PIX_FMT_RGB32: - *f++ = rv; - *f++ = gv; - *f++ = bv; - f++; - break; -
Re: [PATCH] usbvision: remove (broken) image format conversion
On Tuesday 26 April 2011, you wrote: On Monday, April 25, 2011 23:23:17 Ondrej Zary wrote: The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). What really should happen is that the conversion is moved to libv4lconvert. I've never had the time to tackle that, but it would improve this driver a lot. Depending on isoc_mode module parameter, the device uses different image formats: YUV 4:2:2 interleaved, YUV 4:2:0 planar or compressed format. Maybe the parameter should go away and these three formats exposed to userspace? Hopefully the non-compressed formats could be used directly without any conversion. But the compressed format (with new V4L2_PIX_FMT_ assigned?) should be preferred (as it provides much higher frame rates). The code moved into libv4lconvert would decompress the format and convert into something standard (YUV420?). Would you perhaps be interested in doing that work? I can try it. But the hardware isn't mine so my time is limited. The removal also reveals an off-by-one bug in enum_fmt ioctl - it misses the last format, so this patch fixes it too. Good. But why are the GREY/RGB formats also removed? Are those broken as well? GREY, RGB24 and RGB32 seem to work (at least with mplayer). RGB565 and RGB555 have wrong colors. GREY is implemented only in compressed mode but can be selected in other modes too. Can't userspace do the conversion better? Regards, Hans -- Ondrej Zary -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] usbvision: remove (broken) image format conversion
Hi, On 04/26/2011 08:32 AM, Hans Verkuil wrote: On Monday, April 25, 2011 23:23:17 Ondrej Zary wrote: The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). What really should happen is that the conversion is moved to libv4lconvert. I've never had the time to tackle that, but it would improve this driver a lot. Would you perhaps be interested in doing that work? Seconded! I would be more then happy to help with the libv4l part (in the form of review + merging + advice). Regards, Hans -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] usbvision: remove (broken) image format conversion
Hi, On 04/26/2011 10:30 AM, Ondrej Zary wrote: On Tuesday 26 April 2011, you wrote: On Monday, April 25, 2011 23:23:17 Ondrej Zary wrote: The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). What really should happen is that the conversion is moved to libv4lconvert. I've never had the time to tackle that, but it would improve this driver a lot. Depending on isoc_mode module parameter, the device uses different image formats: YUV 4:2:2 interleaved, YUV 4:2:0 planar or compressed format. Maybe the parameter should go away and these three formats exposed to userspace? That sounds right, Hopefully the non-compressed formats could be used directly without any conversion. But the compressed format (with new V4L2_PIX_FMT_ assigned?) should be preferred (as it provides much higher frame rates). The code moved into libv4lconvert would decompress the format and convert into something standard (YUV420?). Correct. Would you perhaps be interested in doing that work? I can try it. But the hardware isn't mine so my time is limited. If you could give it a shot that would be great. I've some hardware to test this with (although I've never actually tested that hardware), so I can hopefully pick things up if you cannot finish things before you need to return the hardware. Thanks Regards, Hans -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] usbvision: remove (broken) image format conversion
Hi, On 04/26/2011 10:30 AM, Ondrej Zary wrote: On Tuesday 26 April 2011, you wrote: On Monday, April 25, 2011 23:23:17 Ondrej Zary wrote: The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). What really should happen is that the conversion is moved to libv4lconvert. I've never had the time to tackle that, but it would improve this driver a lot. Depending on isoc_mode module parameter, the device uses different image formats: YUV 4:2:2 interleaved, YUV 4:2:0 planar or compressed format. Maybe the parameter should go away and these three formats exposed to userspace? That sounds right, Hopefully the non-compressed formats could be used directly without any conversion. But the compressed format (with new V4L2_PIX_FMT_ assigned?) should be preferred (as it provides much higher frame rates). The code moved into libv4lconvert would decompress the format and convert into something standard (YUV420?). Correct. Would you perhaps be interested in doing that work? I can try it. But the hardware isn't mine so my time is limited. If you could give it a shot that would be great. I've some hardware to test this with (although I've never actually tested that hardware), so I can hopefully pick things up if you cannot finish things before you need to return the hardware. I can also test this. Regards, Hans Thanks Regards, Hans -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] usbvision: remove (broken) image format conversion
On Tuesday 26 April 2011 14:33:20 Hans Verkuil wrote: Hi, On 04/26/2011 10:30 AM, Ondrej Zary wrote: On Tuesday 26 April 2011, you wrote: On Monday, April 25, 2011 23:23:17 Ondrej Zary wrote: The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). What really should happen is that the conversion is moved to libv4lconvert. I've never had the time to tackle that, but it would improve this driver a lot. Depending on isoc_mode module parameter, the device uses different image formats: YUV 4:2:2 interleaved, YUV 4:2:0 planar or compressed format. Maybe the parameter should go away and these three formats exposed to userspace? That sounds right, Hopefully the non-compressed formats could be used directly without any conversion. But the compressed format (with new V4L2_PIX_FMT_ assigned?) should be preferred (as it provides much higher frame rates). The code moved into libv4lconvert would decompress the format and convert into something standard (YUV420?). Correct. Would you perhaps be interested in doing that work? I can try it. But the hardware isn't mine so my time is limited. If you could give it a shot that would be great. I've some hardware to test this with (although I've never actually tested that hardware), so I can hopefully pick things up if you cannot finish things before you need to return the hardware. I can also test this. After digging in the code for hours, I'm giving this up. It's not worth it. The ISOC_MODE_YUV422 mode works as V4L2_PIX_FMT_YVYU with VLC and mplayer+libv4lconvert, reducing the loop (and dropping strech_*) in usbvision_parse_lines_422() to: scratch_get(usbvision, frame-data + (frame-v4l2_linesize * frame-curline), 2 * frame-frmwidth); The ISOC_MODE_YUV420 is some weird custom format with 64-byte lines of YYUV. usbvision_parse_lines_420() is real mess with that scratch_* crap everywhere. ISOC_MODE_COMPRESS: There are callbacks to usbvision_request_intra() and also usbvision_adjust_compression(). This is not going to work outside the kernel. So I can redo the conversion removal patch to keep the RGB formats and also provide another one to remove the testpattern (it oopses too). But I'm not going to do any major changes in the driver. -- Ondrej Zary -- To unsubscribe from this list: send the line unsubscribe linux-media in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] usbvision: remove (broken) image format conversion
The YVU420 and YUV422P formats are broken and cause kernel panic on use. (YVU420 does not work and sometimes causes unable to handle paging request panic, YUV422P always causes NULL pointer dereference). As V4L2 spec says that drivers shouldn't do any in-kernel image format conversion, remove it completely (except YUYV). The removal also reveals an off-by-one bug in enum_fmt ioctl - it misses the last format, so this patch fixes it too. This allows the driver to work with mplayer without need to manually specify a format and also to work with VLC without causing kernel panic. Signed-off-by: Ondrej Zary li...@rainbow-software.org diff -up linux-2.6.39-rc2-/drivers/media/video/usbvision/usbvision-core.c linux-2.6.39-rc2/drivers/media/video/usbvision/usbvision-core.c --- linux-2.6.39-rc2-/drivers/media/video/usbvision/usbvision-core.c 2011-04-25 22:30:09.0 +0200 +++ linux-2.6.39-rc2/drivers/media/video/usbvision/usbvision-core.c 2011-04-25 23:07:46.0 +0200 @@ -586,9 +586,8 @@ static enum parse_state usbvision_parse_ int len; int i; unsigned char yuyv[4] = { 180, 128, 10, 128 }; /* YUV components */ - unsigned char rv, gv, bv; /* RGB components */ - int clipmask_index, bytes_per_pixel; - int stretch_bytes, clipmask_add; + int bytes_per_pixel; + int stretch_bytes; frame = usbvision-cur_frame; f = frame-data + (frame-v4l2_linesize * frame-curline); @@ -605,78 +604,16 @@ static enum parse_state usbvision_parse_ bytes_per_pixel = frame-v4l2_format.bytes_per_pixel; stretch_bytes = (usbvision-stretch_width - 1) * bytes_per_pixel; - clipmask_index = frame-curline * MAX_FRAME_WIDTH; - clipmask_add = usbvision-stretch_width; for (i = 0; i frame-frmwidth; i += (2 * usbvision-stretch_width)) { scratch_get(usbvision, yuyv[0], 4); - if (frame-v4l2_format.format == V4L2_PIX_FMT_YUYV) { - *f++ = yuyv[0]; /* Y */ - *f++ = yuyv[3]; /* U */ - } else { - YUV_TO_RGB_BY_THE_BOOK(yuyv[0], yuyv[1], yuyv[3], rv, gv, bv); - switch (frame-v4l2_format.format) { - case V4L2_PIX_FMT_RGB565: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x07 (gv 3)) | - (0xF8 bv); - break; - case V4L2_PIX_FMT_RGB24: - *f++ = rv; - *f++ = gv; - *f++ = bv; - break; - case V4L2_PIX_FMT_RGB32: - *f++ = rv; - *f++ = gv; - *f++ = bv; - f++; - break; - case V4L2_PIX_FMT_RGB555: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x03 (gv 3)) | - (0x7C (bv 2)); - break; - } - } - clipmask_index += clipmask_add; + *f++ = yuyv[0]; /* Y */ + *f++ = yuyv[3]; /* U */ f += stretch_bytes; - if (frame-v4l2_format.format == V4L2_PIX_FMT_YUYV) { - *f++ = yuyv[2]; /* Y */ - *f++ = yuyv[1]; /* V */ - } else { - YUV_TO_RGB_BY_THE_BOOK(yuyv[2], yuyv[1], yuyv[3], rv, gv, bv); - switch (frame-v4l2_format.format) { - case V4L2_PIX_FMT_RGB565: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x07 (gv 3)) | - (0xF8 bv); - break; - case V4L2_PIX_FMT_RGB24: - *f++ = rv; - *f++ = gv; - *f++ = bv; - break; - case V4L2_PIX_FMT_RGB32: - *f++ = rv; - *f++ = gv; - *f++ = bv; - f++; - break; - case V4L2_PIX_FMT_RGB555: - *f++ = (0x1F rv) | - (0xE0 (gv 5)); - *f++ = (0x03 (gv 3)) | - (0x7C (bv 2)); - break; -