| Message ID | 20200510223100.11641-2-laurent.pinchart@ideasonboard.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | media: imx-pxp: Fix operation on i.MX7 | expand |
Hi Laurent, On Mon, 2020-05-11 at 01:30 +0300, Laurent Pinchart wrote: > The PXP hangs without producing any frame on the i.MX7. This is caused > by a mismatch between the routing configuration and the modules > configuration. Fix the routing configuration by > > - Selecting Composite Alpha Blending/Color Key 0 (main path) as the > input to CSC2 (MUX6). > - Connecting the Rotation 0 block in the pipeline by feeding it with > data (MUX12) and selecting its output (MUX14). > - Disabling all other muxes () by selecting the "no output" option. The > datasheet doesn't explicitly require this, but the PXP has been > noticed to hang after several hundreds frames when testing routing > with the LUT disabled and the LUT input mux (MUX9) set to MUX8. It's > thus safer to explicitly disable all unused paths. > > Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> > --- > drivers/media/platform/imx-pxp.c | 34 +++++++++++++++++++------------- > 1 file changed, 20 insertions(+), 14 deletions(-) > > diff --git a/drivers/media/platform/imx-pxp.c b/drivers/media/platform/imx-pxp.c > index 74e16011fc09..53e3f8c8435f 100644 > --- a/drivers/media/platform/imx-pxp.c > +++ b/drivers/media/platform/imx-pxp.c > @@ -911,26 +911,32 @@ static int pxp_start(struct pxp_ctx *ctx, struct vb2_v4l2_buffer *in_vb, > /* bypass LUT */ > pxp_write(dev, HW_PXP_LUT_CTRL, BM_PXP_LUT_CTRL_BYPASS); > > + /* > + * Configure routing, disabling all paths that are not used by > + * selecting the "no output" (3) option. The datasheet doesn't > + * explicitly require this, but the PXP has been seen to hand after > + * processing a few hundreds of frames otherwise. > + */ > pxp_write(dev, HW_PXP_DATA_PATH_CTRL0, Which commit should this patch be applied on? I have a writel() here. > - BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(1)| > - BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(0)| > + BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(3)| > + BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(0)| > + BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(3)| > + BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(1)| > BF_PXP_DATA_PATH_CTRL0_MUX11_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(0)| > + BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(3)| > BF_PXP_DATA_PATH_CTRL0_MUX9_SEL(1)| > BF_PXP_DATA_PATH_CTRL0_MUX8_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(0)| > + BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(3)| > + BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(1)| > + BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(3)| > + BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(3)| > BF_PXP_DATA_PATH_CTRL0_MUX3_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(0)| > - BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(0)); > + BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(3)| > + BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(3)| > + BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(3)); > pxp_write(dev, HW_PXP_DATA_PATH_CTRL1, > - BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(1) | > - BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(1)); > + BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(3) | > + BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(3)); I have tried this change on i.MX6ULL / v5.6, it does not work. The PXP interrupt never triggers. regards Philipp
Hi Philipp, On Mon, May 18, 2020 at 03:33:09PM +0200, Philipp Zabel wrote: > On Mon, 2020-05-11 at 01:30 +0300, Laurent Pinchart wrote: > > The PXP hangs without producing any frame on the i.MX7. This is caused > > by a mismatch between the routing configuration and the modules > > configuration. Fix the routing configuration by > > > > - Selecting Composite Alpha Blending/Color Key 0 (main path) as the > > input to CSC2 (MUX6). > > - Connecting the Rotation 0 block in the pipeline by feeding it with > > data (MUX12) and selecting its output (MUX14). > > - Disabling all other muxes () by selecting the "no output" option. The > > datasheet doesn't explicitly require this, but the PXP has been > > noticed to hang after several hundreds frames when testing routing > > with the LUT disabled and the LUT input mux (MUX9) set to MUX8. It's > > thus safer to explicitly disable all unused paths. > > > > Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> > > --- > > drivers/media/platform/imx-pxp.c | 34 +++++++++++++++++++------------- > > 1 file changed, 20 insertions(+), 14 deletions(-) > > > > diff --git a/drivers/media/platform/imx-pxp.c b/drivers/media/platform/imx-pxp.c > > index 74e16011fc09..53e3f8c8435f 100644 > > --- a/drivers/media/platform/imx-pxp.c > > +++ b/drivers/media/platform/imx-pxp.c > > @@ -911,26 +911,32 @@ static int pxp_start(struct pxp_ctx *ctx, struct vb2_v4l2_buffer *in_vb, > > /* bypass LUT */ > > pxp_write(dev, HW_PXP_LUT_CTRL, BM_PXP_LUT_CTRL_BYPASS); > > > > + /* > > + * Configure routing, disabling all paths that are not used by > > + * selecting the "no output" (3) option. The datasheet doesn't > > + * explicitly require this, but the PXP has been seen to hand after > > + * processing a few hundreds of frames otherwise. > > + */ > > pxp_write(dev, HW_PXP_DATA_PATH_CTRL0, > > Which commit should this patch be applied on? I have a writel() here. My bad, I have a commit in my tree that wraps writel() and readl() into inline accessors, as it made it easier during debugging to add printk() calls there to log all register accesses. Would you accept that ? If so I'll include it in v2. > > - BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(1)| > > - BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(0)| > > + BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(3)| > > + BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(0)| > > + BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(3)| > > + BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(1)| > > BF_PXP_DATA_PATH_CTRL0_MUX11_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(0)| > > + BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(3)| > > BF_PXP_DATA_PATH_CTRL0_MUX9_SEL(1)| > > BF_PXP_DATA_PATH_CTRL0_MUX8_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(0)| > > + BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(3)| > > + BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(1)| > > + BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(3)| > > + BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(3)| > > BF_PXP_DATA_PATH_CTRL0_MUX3_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(0)| > > - BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(0)); > > + BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(3)| > > + BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(3)| > > + BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(3)); > > pxp_write(dev, HW_PXP_DATA_PATH_CTRL1, > > - BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(1) | > > - BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(1)); > > + BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(3) | > > + BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(3)); > > I have tried this change on i.MX6ULL / v5.6, it does not work. > The PXP interrupt never triggers. :-( Could you try the combination of 1/2 and 2/2 ?
Hi Laurent, On Mon, 2020-05-18 at 16:54 +0300, Laurent Pinchart wrote: [...] > > > + /* > > > + * Configure routing, disabling all paths that are not used by > > > + * selecting the "no output" (3) option. The datasheet doesn't > > > + * explicitly require this, but the PXP has been seen to hand after > > > + * processing a few hundreds of frames otherwise. > > > + */ > > > pxp_write(dev, HW_PXP_DATA_PATH_CTRL0, > > > > Which commit should this patch be applied on? I have a writel() here. > > My bad, I have a commit in my tree that wraps writel() and readl() into > inline accessors, as it made it easier during debugging to add printk() > calls there to log all register accesses. Would you accept that ? If so > I'll include it in v2. Sure, go ahead. > > > - BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(1)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(0)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(3)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(0)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(3)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(1)| > > > BF_PXP_DATA_PATH_CTRL0_MUX11_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(0)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(3)| > > > BF_PXP_DATA_PATH_CTRL0_MUX9_SEL(1)| > > > BF_PXP_DATA_PATH_CTRL0_MUX8_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(0)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(3)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(1)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(3)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(3)| > > > BF_PXP_DATA_PATH_CTRL0_MUX3_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(0)| > > > - BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(0)); > > > + BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(3)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(3)| > > > + BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(3)); > > > pxp_write(dev, HW_PXP_DATA_PATH_CTRL1, > > > - BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(1) | > > > - BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(1)); > > > + BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(3) | > > > + BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(3)); > > > > I have tried this change on i.MX6ULL / v5.6, it does not work. > > The PXP interrupt never triggers. > > :-( Could you try the combination of 1/2 and 2/2 ? With both patches applied, it still hangs. Looking at the "PXP Architecture" Figures 41-1 in the i.MX6ULL Reference Manual and 13-87 in the i.MX7D Reference Manual, there are quite a few muxes dropped on i.MX6ULL, especially around the rotation engines, and the routing is slightly different. Muxes 2, 4-7, 10, 13, and 15 are all documented as "reserved, read-only, always 0". We'll have to use different path control settings per compatible. regards Philipp
diff --git a/drivers/media/platform/imx-pxp.c b/drivers/media/platform/imx-pxp.c index 74e16011fc09..53e3f8c8435f 100644 --- a/drivers/media/platform/imx-pxp.c +++ b/drivers/media/platform/imx-pxp.c @@ -911,26 +911,32 @@ static int pxp_start(struct pxp_ctx *ctx, struct vb2_v4l2_buffer *in_vb, /* bypass LUT */ pxp_write(dev, HW_PXP_LUT_CTRL, BM_PXP_LUT_CTRL_BYPASS); + /* + * Configure routing, disabling all paths that are not used by + * selecting the "no output" (3) option. The datasheet doesn't + * explicitly require this, but the PXP has been seen to hand after + * processing a few hundreds of frames otherwise. + */ pxp_write(dev, HW_PXP_DATA_PATH_CTRL0, - BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(1)| - BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(0)| + BF_PXP_DATA_PATH_CTRL0_MUX15_SEL(3)| + BF_PXP_DATA_PATH_CTRL0_MUX14_SEL(0)| + BF_PXP_DATA_PATH_CTRL0_MUX13_SEL(3)| + BF_PXP_DATA_PATH_CTRL0_MUX12_SEL(1)| BF_PXP_DATA_PATH_CTRL0_MUX11_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(0)| + BF_PXP_DATA_PATH_CTRL0_MUX10_SEL(3)| BF_PXP_DATA_PATH_CTRL0_MUX9_SEL(1)| BF_PXP_DATA_PATH_CTRL0_MUX8_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(0)| + BF_PXP_DATA_PATH_CTRL0_MUX7_SEL(3)| + BF_PXP_DATA_PATH_CTRL0_MUX6_SEL(1)| + BF_PXP_DATA_PATH_CTRL0_MUX5_SEL(3)| + BF_PXP_DATA_PATH_CTRL0_MUX4_SEL(3)| BF_PXP_DATA_PATH_CTRL0_MUX3_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(0)| - BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(0)); + BF_PXP_DATA_PATH_CTRL0_MUX2_SEL(3)| + BF_PXP_DATA_PATH_CTRL0_MUX1_SEL(3)| + BF_PXP_DATA_PATH_CTRL0_MUX0_SEL(3)); pxp_write(dev, HW_PXP_DATA_PATH_CTRL1, - BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(1) | - BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(1)); + BF_PXP_DATA_PATH_CTRL1_MUX17_SEL(3) | + BF_PXP_DATA_PATH_CTRL1_MUX16_SEL(3)); pxp_write(dev, HW_PXP_IRQ_MASK, 0xffff);
The PXP hangs without producing any frame on the i.MX7. This is caused by a mismatch between the routing configuration and the modules configuration. Fix the routing configuration by - Selecting Composite Alpha Blending/Color Key 0 (main path) as the input to CSC2 (MUX6). - Connecting the Rotation 0 block in the pipeline by feeding it with data (MUX12) and selecting its output (MUX14). - Disabling all other muxes () by selecting the "no output" option. The datasheet doesn't explicitly require this, but the PXP has been noticed to hang after several hundreds frames when testing routing with the LUT disabled and the LUT input mux (MUX9) set to MUX8. It's thus safer to explicitly disable all unused paths. Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com> --- drivers/media/platform/imx-pxp.c | 34 +++++++++++++++++++------------- 1 file changed, 20 insertions(+), 14 deletions(-)