media: imx: imx7-media-csi: Sync frames to start of frame for MIPI
PATCH 1/1 media: imx: imx7-media-csi: Sync frames to start of frame for MIPI
The default behavior for a base address change is to do it automatically after a DMA completion. This can lead to the situation, were one corrupted frame, with less lines than configured, results in all following frames being corrupted as well, due to a missing re-synchronization to the beginning of the next frame.
Fix this by configuring the base address switch to be synced with the start of frame event.
Currently this is already implemented for the parallel interface. To have it with MIPI as well, simply configure it unconditionally.
Tested on i.MX 8MM.
--- drivers/media/platform/nxp/imx7-media-csi.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
drivers/media/platform/nxp/imx7-media-csi.c +3 -3
@@ -529,13 +529,13 @@ static void imx7_csi_configure(struct imx7_csi *csi, stride = out_pix->width; } + cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL | +BIT_BASEADDR_CHG_ERR_EN ; +if (!csi->is_csi2) { cr1 =BIT_SOF_POL |BIT_REDGE |BIT_GCLK_MODE |BIT_HSYNC_POL |BIT_FCC |BIT_MCLKDIV (1 ) |BIT_MCLKEN ; - cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL | -BIT_BASEADDR_CHG_ERR_EN ; -if (out_pix->pixelformat ==V4L2_PIX_FMT_UYVY || out_pix->pixelformat ==V4L2_PIX_FMT_YUYV ) width *=2 ;
Signature · 1 line
2.25.1
Fabio Estevam
reviewerHi Stefan,
Quoted 7 lines · Stefan Riedmüller
The default behavior for a base address change is to do it automatically after a DMA completion. This can lead to the situation, were one corrupted frame, with less lines than configured, results in all following frames being corrupted as well, due to a missing re-synchronization to the beginning of the next frame.
Quoted 7 lines
Fix this by configuring the base address switch to be synced with the start of frame event.
Currently this is already implemented for the parallel interface. To have it with MIPI as well, simply configure it unconditionally.
Tested on i.MX 8MM.
Should this contain a Fixes tag? Thanks
Hi Fabio,
Quoted 14 lines · Fabio Estevam
Hi Stefan,
Quoted 10 lines · Stefan Riedmüller
The default behavior for a base address change is to do it automatically after a DMA completion. This can lead to the situation, were one corrupted frame, with less lines than configured, results in all following frames being corrupted as well, due to a missing re-synchronization to the beginning of the next frame.
Fix this by configuring the base address switch to be synced with the start of frame event.
Quoted 6 lines
Quoted 4 lines
Currently this is already implemented for the parallel interface. To have it with MIPI as well, simply configure it unconditionally.
Tested on i.MX 8MM.
Should this contain a Fixes tag? Thanks
Now that you mention it, I guess it could. I'm just not sure which commit to select for the fixes tag. I think this issue is there since day one of the driver, but I doubt that it would apply on every version of it. The fixes tag is to trigger a stable fix, right? Or is it just meant as a note?
Do you have any suggestions?
Thanks,
Stefan
Laurent Pinchart
maintainerHi Stefan,
Thank you for the patch.
Quoted 6 lines · Stefan Riedmüller
The default behavior for a base address change is to do it automatically after a DMA completion. This can lead to the situation, were one corrupted frame, with less lines than configured, results in all following frames being corrupted as well, due to a missing re-synchronization to the beginning of the next frame.
Quoted 7 lines
Fix this by configuring the base address switch to be synced with the start of frame event.
Currently this is already implemented for the parallel interface. To have it with MIPI as well, simply configure it unconditionally.
Tested on i.MX 8MM.
I recall not doing this unconditionally as it didn't work on some of the platforms I was testing, but I'm not sure of the details anymore. I'll retest on i.MX7.
Do we have any buffer overflow protection in this mode ? If the sensor happens to send more lines than expected, will extra lines be dropped, or overflow the buffer ?
Quoted 26 lines
Signed-off-by Stefan Riedmüller <[email protected]>--- drivers/media/platform/nxp/imx7-media-csi.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)drivers/media/platform/nxp/imx7-media-csi.c +3
-3@@ -529,13 +529,13 @@ static void imx7_csi_configure(struct imx7_csi *csi, stride = out_pix->width; } + cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL | +BIT_BASEADDR_CHG_ERR_EN ; +if (!csi->is_csi2) { cr1 =BIT_SOF_POL |BIT_REDGE |BIT_GCLK_MODE |BIT_HSYNC_POL |BIT_FCC |BIT_MCLKDIV (1 ) |BIT_MCLKEN ; - cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL | -BIT_BASEADDR_CHG_ERR_EN ; -if (out_pix->pixelformat ==V4L2_PIX_FMT_UYVY || out_pix->pixelformat ==V4L2_PIX_FMT_YUYV ) width *=2 ;
Signature · 3 lines
Regards, Laurent Pinchart
Hi Laurant,
Quoted 14 lines · Laurent Pinchart
Hi Stefan,
Thank you for the patch.
Quoted 9 lines · Stefan Riedmüller
The default behavior for a base address change is to do it automatically after a DMA completion. This can lead to the situation, were one corrupted frame, with less lines than configured, results in all following frames being corrupted as well, due to a missing re-synchronization to the beginning of the next frame.
Fix this by configuring the base address switch to be synced with the start of frame event.
Quoted 8 lines
Quoted 4 lines
Currently this is already implemented for the parallel interface. To have it with MIPI as well, simply configure it unconditionally.
Tested on i.MX 8MM.
I recall not doing this unconditionally as it didn't work on some of the platforms I was testing, but I'm not sure of the details anymore. I'll retest on i.MX7.
Thanks!
Quoted 3 lines
Do we have any buffer overflow protection in this mode ? If the sensor happens to send more lines than expected, will extra lines be dropped, or overflow the buffer ?
As far as I understand the imx8mm reference manual, I think there is a buffer overflow protection, in the form that the DMA will only write the amount of data which was configured in CSI_IMAG_PARA. If additional data arrives on the bus, without a new start of frame event, a new DMA transfer will be triggered eventually. Since there was no base address change, due to the missing start frame event, the new DMA transfer will use the same buffer as before and corrupt that data. But all subsequent frames will be ok since a new frame will trigger the start of frame event and thus a base address change.
Regards,
Stefan
Quoted 31 lines
Quoted 30 lines
Signed-off-by Stefan Riedmüller <[email protected]>---
drivers/media/platform/nxp/imx7-media-csi.c | 6 +++---1 file changed, 3 insertions(+), 3 deletions(-)drivers/media/platform/nxp/imx7-media-csi.c +0
-0@@ -529,13 +529,13 @@ static void imx7_csi_configure(struct imx7_csi *csi, stride = out_pix->width; }drivers/media/platform/nxp/imx7-media-csi.c +3
-0+ cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL | +BIT_BASEADDR_CHG_ERR_EN ; +if (!csi->is_csi2) {
cr1 =BIT_SOF_POL|BIT_REDGE|BIT_GCLK_MODE|BIT_HSYNC_POL
|BIT_FCC|BIT_MCLKDIV(1)|BIT_MCLKEN;drivers/media/platform/nxp/imx7-media-csi.c +0
-1- cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL |
drivers/media/platform/nxp/imx7-media-csi.c +0
-2-BIT_BASEADDR_CHG_ERR_EN ; -if (
out_pix->pixelformat==V4L2_PIX_FMT_UYVY||out_pix->pixelformat==V4L2_PIX_FMT_YUYV) width *= 2;
Hi Laurent,
Quoted 20 lines · Stefan Riedmüller
Hi Laurant,
Quoted 17 lines · Laurent Pinchart
Hi Stefan,
Thank you for the patch.
Quoted 12 lines · Stefan Riedmüller
The default behavior for a base address change is to do it automatically after a DMA completion. This can lead to the situation, were one corrupted frame, with less lines than configured, results in all following frames being corrupted as well, due to a missing re-synchronization to the beginning of the next frame.
Fix this by configuring the base address switch to be synced with the start of frame event.
Currently this is already implemented for the parallel interface. To have it with MIPI as well, simply configure it unconditionally.
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
Tested on i.MX 8MM.
I recall not doing this unconditionally as it didn't work on some of the platforms I was testing, but I'm not sure of the details anymore. I'll retest on i.MX7.
Thanks!
Any update yet on the i.MX7 test?
Regards,
Stefan
Quoted 59 lines
Quoted 4 lines
Do we have any buffer overflow protection in this mode ? If the sensor happens to send more lines than expected, will extra lines be dropped, or overflow the buffer ?
As far as I understand the imx8mm reference manual, I think there is a buffer
overflow protection, in the form that the DMA will only write the amount of data which was configured inCSI_IMAG_PARA. If additional data arrives on the
bus, without a new start of frame event, a new DMA transfer will be triggered
eventually. Since there was no base address change, due to the missing start frame event, the new DMA transfer will use the same buffer as before and corrupt that data. But all subsequent frames will be ok since a new frame will
trigger the start of frame event and thus a base address change.Regards,
Stefan tQuoted 27 lines
Quoted 27 lines
Signed-off-by Stefan Riedmüller <[email protected]>---
drivers/media/platform/nxp/imx7-media-csi.c | 6 +++---1 file changed, 3 insertions(+), 3 deletions(-)drivers/media/platform/nxp/imx7-media-csi.c +0
-0@@ -529,13 +529,13 @@ static void imx7_csi_configure(struct imx7_csi *csi, stride = out_pix->width; }drivers/media/platform/nxp/imx7-media-csi.c +3
-0+ cr18 |=BIT_BASEADDR_SWITCH_EN |BIT_BASEADDR_SWITCH_SEL | +BIT_BASEADDR_CHG_ERR_EN ; +if (!csi->is_csi2) {
cr1 =BIT_SOF_POL|BIT_REDGE|BIT_GCLK_MODE|BIT_HSYNC_POL
|BIT_FCC|BIT_MCLKDIV(1)|BIT_MCLKEN;drivers/media/platform/nxp/imx7-media-csi.c +0
-1- cr18 |= BIT_BASEADDR_SWITCH_EN |
BIT_BASEADDR_SWITCH_SELQuoted 1 line
Quoted 6 lines
Quoted 5 lines
drivers/media/platform/nxp/imx7-media-csi.c +0
-2-BIT_BASEADDR_CHG_ERR_EN ; -if (
out_pix->pixelformat==V4L2_PIX_FMT_UYVY||out_pix->pixelformat==V4L2_PIX_FMT_YUYV) width *= 2;_______________________________________________
upstream mailing list
[email protected]
http://lists.phytec.de/cgi-bin/mailman/listinfo/upstream