diff mbox

[v2,11/13] gpu: ipu-ic: Add complete image conversion support with tiling

Message ID 1468977071-29240-12-git-send-email-steve_longerbeam@mentor.com (mailing list archive)
State New, archived
Headers show

Commit Message

Steve Longerbeam July 20, 2016, 1:11 a.m. UTC
This patch implements complete image conversion support to ipu-ic,
with tiling to support scaling to and from images up to 4096x4096.
Image rotation is also supported.

The internal API is subsystem agnostic (no V4L2 dependency except
for the use of V4L2 fourcc pixel formats).

Callers prepare for image conversion by calling
ipu_image_convert_prepare(), which initializes the parameters of
the conversion. The caller passes in the ipu_ic task to use for
the conversion, the input and output image formats, a rotation mode,
and a completion callback and completion context pointer:

struct image_converter_ctx *
ipu_image_convert_prepare(struct ipu_ic *ic,
                          struct ipu_image *in, struct ipu_image *out,
                          enum ipu_rotate_mode rot_mode,
                          image_converter_cb_t complete,
                          void *complete_context);

The caller is given a new conversion context that must be passed to
the further APIs:

struct image_converter_run *
ipu_image_convert_run(struct image_converter_ctx *ctx,
                      dma_addr_t in_phys, dma_addr_t out_phys);

This queues a new image conversion request to a run queue, and
starts the conversion immediately if the run queue is empty. Only
the physaddr's of the input and output image buffers are needed,
since the conversion context was created previously with
ipu_image_convert_prepare(). Returns a new run object pointer. When
the conversion completes, the run pointer is returned to the
completion callback.

void image_convert_abort(struct image_converter_ctx *ctx);

This will abort any active or pending conversions for this context.
Any currently active or pending runs belonging to this context are
returned via the completion callback with an error status.

void ipu_image_convert_unprepare(struct image_converter_ctx *ctx);

Unprepares the conversion context. Any active or pending runs will
be aborted by calling image_convert_abort().

Signed-off-by: Steve Longerbeam <steve_longerbeam@mentor.com>
---
 drivers/gpu/ipu-v3/ipu-ic.c | 1691 ++++++++++++++++++++++++++++++++++++++++++-
 include/video/imx-ipu-v3.h  |   57 +-
 2 files changed, 1736 insertions(+), 12 deletions(-)

Comments

Philipp Zabel July 26, 2016, 10:08 a.m. UTC | #1
Am Dienstag, den 19.07.2016, 18:11 -0700 schrieb Steve Longerbeam:
> This patch implements complete image conversion support to ipu-ic,
> with tiling to support scaling to and from images up to 4096x4096.
> Image rotation is also supported.
> 
> The internal API is subsystem agnostic (no V4L2 dependency except
> for the use of V4L2 fourcc pixel formats).
> 
> Callers prepare for image conversion by calling
> ipu_image_convert_prepare(), which initializes the parameters of
> the conversion. The caller passes in the ipu_ic task to use for
> the conversion, the input and output image formats, a rotation mode,
> and a completion callback and completion context pointer:
> 
> struct image_converter_ctx *
> ipu_image_convert_prepare(struct ipu_ic *ic,
>                           struct ipu_image *in, struct ipu_image *out,
>                           enum ipu_rotate_mode rot_mode,
>                           image_converter_cb_t complete,
>                           void *complete_context);
> 
> The caller is given a new conversion context that must be passed to
> the further APIs:
> 
> struct image_converter_run *
> ipu_image_convert_run(struct image_converter_ctx *ctx,
>                       dma_addr_t in_phys, dma_addr_t out_phys);
> 
> This queues a new image conversion request to a run queue, and
> starts the conversion immediately if the run queue is empty. Only
> the physaddr's of the input and output image buffers are needed,
> since the conversion context was created previously with
> ipu_image_convert_prepare(). Returns a new run object pointer. When
> the conversion completes, the run pointer is returned to the
> completion callback.
> 
> void image_convert_abort(struct image_converter_ctx *ctx);
> 
> This will abort any active or pending conversions for this context.
> Any currently active or pending runs belonging to this context are
> returned via the completion callback with an error status.
> 
> void ipu_image_convert_unprepare(struct image_converter_ctx *ctx);
> 
> Unprepares the conversion context. Any active or pending runs will
> be aborted by calling image_convert_abort().
> 
> Signed-off-by: Steve Longerbeam <steve_longerbeam@mentor.com>
> ---
>  drivers/gpu/ipu-v3/ipu-ic.c | 1691 ++++++++++++++++++++++++++++++++++++++++++-
>  include/video/imx-ipu-v3.h  |   57 +-
>  2 files changed, 1736 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/gpu/ipu-v3/ipu-ic.c b/drivers/gpu/ipu-v3/ipu-ic.c
> index 1a37afc..5471b72 100644
> --- a/drivers/gpu/ipu-v3/ipu-ic.c
> +++ b/drivers/gpu/ipu-v3/ipu-ic.c
> @@ -17,6 +17,8 @@
>  #include <linux/bitrev.h>
>  #include <linux/io.h>
>  #include <linux/err.h>
> +#include <linux/interrupt.h>
> +#include <linux/dma-mapping.h>
>  #include "ipu-prv.h"
>  
>  /* IC Register Offsets */
> @@ -82,6 +84,40 @@
>  #define IC_IDMAC_3_PP_WIDTH_MASK        (0x3ff << 20)
>  #define IC_IDMAC_3_PP_WIDTH_OFFSET      20
>  
> +/*
> + * The IC Resizer has a restriction that the output frame from the
> + * resizer must be 1024 or less in both width (pixels) and height
> + * (lines).
> + *
> + * The image conversion support attempts to split up a conversion when
> + * the desired output (converted) frame resolution exceeds the IC resizer
> + * limit of 1024 in either dimension.
> + *
> + * If either dimension of the output frame exceeds the limit, the
> + * dimension is split into 1, 2, or 4 equal stripes, 

Imagine converting a 320x200 image up to 1280x800, and consider only the
x coordinate. The IC upscaler is a simple bilinear scaler.

We want target pixel x' = 1279 to sample the source pixel x = 319, so
the scaling factor rsc is calculated to:

x = x' * (320-1)/(1280-1) = x' * 8192/rsc, with rsc = 32846

That means that the target pixels x' = 639 and x' = 640 should be
sampled (bilinearly) from x = 639 * 8192/32846. = 159.37 and x = 640 *
8192/32846. = 159.62, respectively.

Now split the frame in half and suddenly pixel x' = 640 is the start of
a new tile, so it is sampled at x = 160, and pixel x' = 1279 will be
sampled at x = 160 + (1279 - 640) * 8192/32846. = 319.37, reading over
the edge of the source image.
This problem gets worse if you start using arbitrary frame sizes and YUV
planar images and consider that tile start addresses are (currently)
limited to 8 byte boundaries, to the point that there are very visible
seams in the center of the image, depending on scaling factor and image
sizes.

I wonder how much effort it would be to remove the tiling code for now
and add it back in a second step once it is fixed? Otherwise we could
just disallow scaled tiling for now, but I'd like this to be prepared
for tiles with different sizes at least, before merging.

regards
Philipp
Steve Longerbeam July 28, 2016, 11:09 p.m. UTC | #2
Hi Philipp,

On 07/26/2016 03:08 AM, Philipp Zabel wrote:
>
>>   
>> +/*
>> + * The IC Resizer has a restriction that the output frame from the
>> + * resizer must be 1024 or less in both width (pixels) and height
>> + * (lines).
>> + *
>> + * The image conversion support attempts to split up a conversion when
>> + * the desired output (converted) frame resolution exceeds the IC resizer
>> + * limit of 1024 in either dimension.
>> + *
>> + * If either dimension of the output frame exceeds the limit, the
>> + * dimension is split into 1, 2, or 4 equal stripes,
> Imagine converting a 320x200 image up to 1280x800, and consider only the
> x coordinate. The IC upscaler is a simple bilinear scaler.

Right, the upscaling is a simple linear interpolation between two
adjacent input pixels, just paraphrasing you.

> We want target pixel x' = 1279 to sample the source pixel x = 319, so
> the scaling factor rsc is calculated to:
>
> x = x' * (320-1)/(1280-1) = x' * 8192/rsc, with rsc = 32846
>
> That means that the target pixels x' = 639 and x' = 640 should be
> sampled (bilinearly) from x = 639 * 8192/32846. = 159.37 and x = 640 *
> 8192/32846. = 159.62, respectively.

I'm with you so far.

>
> Now split the frame in half and suddenly pixel x' = 640 is the start of
> a new tile, so it is sampled at x = 160, and pixel x' = 1279 will be
> sampled at x = 160 + (1279 - 640) * 8192/32846. = 319.37, reading over
> the edge of the source image.

Here's where we part.

The 320x200 --> 1280x800 conversion is split into two 160x200 -->
640x800 conversions. The DMA controller and ipu_ic_task_init() are given
those width/height dimensions, not the dimensions of the original images.
So this is simply two separate 160x200 --> 640x800 conversions. The only
difference from a true 160x200 --> 640x800 image conversion is that the DMA
controller must be given the stride lengths of the original 320x200 and 
1280x800
images.

The rsc for the 160x200 --> 640x800 conversions is

x = x' * (160-1)/(640-1) = x' * 8192/rsc, so rsc = 32923


So original horizontal position 640 is really x' = 0 of the second 
conversion,
which is sampled at x = 0 of the second conversion. And the pixel at x' 
= 1279
is really x' = 639 of the second conversion, which is sampled at x = 639 
* 8192/32923
= 158.98, which does not read over the edge of the source tile.


> This problem gets worse if you start using arbitrary frame sizes and YUV
> planar images and consider that tile start addresses are (currently)
> limited to 8 byte boundaries, to the point that there are very visible
> seams in the center of the image, depending on scaling factor and image
> sizes.

Indeed there could be other parameters that would cause the resizer to
read past the edge of the source tiles, I will need to try and find such 
cases.
But not in the above case.

That said, I _have_ noticed seams, but I have always attributed them to the
fact that we have a discontinuity in color-space conversion and/or resize
interpolation at the boundary between tiles.

I've also found that the seams are quite noticeable when rendered to a
display overlay, but become significantly less pronounced if the images are
converted to a back buffer, and then page-flipped to front buffer when the
conversion (all tiles) completes.

Steve

> I wonder how much effort it would be to remove the tiling code for now
> and add it back in a second step once it is fixed? Otherwise we could
> just disallow scaled tiling for now, but I'd like this to be prepared
> for tiles with different sizes at least, before merging.
>
> regards
> Philipp
>
Philipp Zabel Aug. 1, 2016, 9:29 a.m. UTC | #3
Am Donnerstag, den 28.07.2016, 16:09 -0700 schrieb Steve Longerbeam:
> > Now split the frame in half and suddenly pixel x' = 640 is the start of
> > a new tile, so it is sampled at x = 160, and pixel x' = 1279 will be
> > sampled at x = 160 + (1279 - 640) * 8192/32846. = 319.37, reading over
> > the edge of the source image.
> 
> Here's where we part.
> 
> The 320x200 --> 1280x800 conversion is split into two 160x200 -->
> 640x800 conversions. The DMA controller and ipu_ic_task_init() are given
> those width/height dimensions, not the dimensions of the original images.
> So this is simply two separate 160x200 --> 640x800 conversions. The only
> difference from a true 160x200 --> 640x800 image conversion is that the DMA
> controller must be given the stride lengths of the original 320x200 and 
> 1280x800
> images.
> 
> The rsc for the 160x200 --> 640x800 conversions is
> 
> x = x' * (160-1)/(640-1) = x' * 8192/rsc, so rsc = 32923
> 
> 
> So original horizontal position 640 is really x' = 0 of the second 
> conversion,
> which is sampled at x = 0 of the second conversion. And the pixel at x' 
> = 1279
> is really x' = 639 of the second conversion, which is sampled at x = 639 
> * 8192/32923
> = 158.98, which does not read over the edge of the source tile.

My bad, I somehow thought that the scaling factor is calculated per
image (as it IMHO should be), not just per tile.

Of course in that case you won't ever read over the edge, but on the
other hand the visual problems are worse because you underestimate the
scaling factor and introduce a sharp edge at the center: even if the
source pixel step per target pixel step is a fraction, between pixels
width/2-1 and width/2 there's always a whole source pixel step.

Take the extreme example of scaling 32x32 to 1080x1080 pixels. The ideal
source pixels for x' = 519 and 520 should be x = 14.911 and 14.939,
respectively. Due to tiling they will be x = 15 and 16, introducing a
sharp seam in the otherwise blurry mess.

> > This problem gets worse if you start using arbitrary frame sizes and YUV
> > planar images and consider that tile start addresses are (currently)
> > limited to 8 byte boundaries, to the point that there are very visible
> > seams in the center of the image, depending on scaling factor and image
> > sizes.
> 
> Indeed there could be other parameters that would cause the resizer to
> read past the edge of the source tiles, I will need to try and find such 
> cases. 
> But not in the above case.

Ok.

> That said, I _have_ noticed seams, but I have always attributed them to the
> fact that we have a discontinuity in color-space conversion and/or resize
> interpolation at the boundary between tiles.
>
> I've also found that the seams are quite noticeable when rendered to a
> display overlay, but become significantly less pronounced if the images are
> converted to a back buffer, and then page-flipped to front buffer when the
> conversion (all tiles) completes.

I don't know what to make of this. Maybe it is a timing issue and what
you are actually seeing is tearing between tiles of different frames?

regards
Philipp
Steve Longerbeam Aug. 4, 2016, 12:18 a.m. UTC | #4
On 08/01/2016 02:29 AM, Philipp Zabel wrote:
> Am Donnerstag, den 28.07.2016, 16:09 -0700 schrieb Steve Longerbeam:
>>> Now split the frame in half and suddenly pixel x' = 640 is the start of
>>> a new tile, so it is sampled at x = 160, and pixel x' = 1279 will be
>>> sampled at x = 160 + (1279 - 640) * 8192/32846. = 319.37, reading over
>>> the edge of the source image.
>> Here's where we part.
>>
>> The 320x200 --> 1280x800 conversion is split into two 160x200 -->
>> 640x800 conversions. The DMA controller and ipu_ic_task_init() are given
>> those width/height dimensions, not the dimensions of the original images.
>> So this is simply two separate 160x200 --> 640x800 conversions. The only
>> difference from a true 160x200 --> 640x800 image conversion is that the DMA
>> controller must be given the stride lengths of the original 320x200 and 
>> 1280x800
>> images.
>>
>> The rsc for the 160x200 --> 640x800 conversions is
>>
>> x = x' * (160-1)/(640-1) = x' * 8192/rsc, so rsc = 32923
>>
>>
>> So original horizontal position 640 is really x' = 0 of the second 
>> conversion,
>> which is sampled at x = 0 of the second conversion. And the pixel at x' 
>> = 1279
>> is really x' = 639 of the second conversion, which is sampled at x = 639 
>> * 8192/32923
>> = 158.98, which does not read over the edge of the source tile.
> My bad, I somehow thought that the scaling factor is calculated per
> image (as it IMHO should be), not just per tile.
>
> Of course in that case you won't ever read over the edge, but on the
> other hand the visual problems are worse because you underestimate the
> scaling factor and introduce a sharp edge at the center: even if the
> source pixel step per target pixel step is a fraction, between pixels
> width/2-1 and width/2 there's always a whole source pixel step.
>
> Take the extreme example of scaling 32x32 to 1080x1080 pixels. The ideal
> source pixels for x' = 519 and 520 should be x = 14.911 and 14.939,
> respectively. Due to tiling they will be x = 15 and 16, introducing a
> sharp seam in the otherwise blurry mess.

I think you mean at x' = 539 and x' = 540.

But yes I agree. Due to tiling, at x' = 539, the input pixel is sampled at x = 15.
If the interpolation were to continue (no tiling), at x' = 540, the input pixel
would be sampled at (31/1079)*540 = 15.514. Instead, because of tiling,
there is a discontinuity in the interpolation (it is reset), beginning again at
x' = 0 (540), which is sampled at x = 0 (16).

The only way I can think of to resolve this problem is to add some width
to the output tiles such that the interpolation completes a full span between
input position w - 2 and w - 1. That is, add to w' until floor(F*w') increments
to the next whole integer, where F = (w-1)/(w'-1) is the scaling factor.

But that will likely cause the next tile DMA addrs to fail to fall on the IDMAC
8 byte alignment.


>
>
>> That said, I _have_ noticed seams, but I have always attributed them to the
>> fact that we have a discontinuity in color-space conversion and/or resize
>> interpolation at the boundary between tiles.
>>
>> I've also found that the seams are quite noticeable when rendered to a
>> display overlay, but become significantly less pronounced if the images are
>> converted to a back buffer, and then page-flipped to front buffer when the
>> conversion (all tiles) completes.
> I don't know what to make of this. Maybe it is a timing issue and what
> you are actually seeing is tearing between tiles of different frames?

Yes, that's always been my assumption, a scan-out contains a mix
of tiles from different frames, when page-flip is not used.

Steve
Philipp Zabel Aug. 12, 2016, 2:56 p.m. UTC | #5
Am Mittwoch, den 03.08.2016, 17:18 -0700 schrieb Steve Longerbeam:
> On 08/01/2016 02:29 AM, Philipp Zabel wrote:
> > Am Donnerstag, den 28.07.2016, 16:09 -0700 schrieb Steve Longerbeam:
> >>> Now split the frame in half and suddenly pixel x' = 640 is the start of
> >>> a new tile, so it is sampled at x = 160, and pixel x' = 1279 will be
> >>> sampled at x = 160 + (1279 - 640) * 8192/32846. = 319.37, reading over
> >>> the edge of the source image.
> >> Here's where we part.
> >>
> >> The 320x200 --> 1280x800 conversion is split into two 160x200 -->
> >> 640x800 conversions. The DMA controller and ipu_ic_task_init() are given
> >> those width/height dimensions, not the dimensions of the original images.
> >> So this is simply two separate 160x200 --> 640x800 conversions. The only
> >> difference from a true 160x200 --> 640x800 image conversion is that the DMA
> >> controller must be given the stride lengths of the original 320x200 and 
> >> 1280x800
> >> images.
> >>
> >> The rsc for the 160x200 --> 640x800 conversions is
> >>
> >> x = x' * (160-1)/(640-1) = x' * 8192/rsc, so rsc = 32923
> >>
> >>
> >> So original horizontal position 640 is really x' = 0 of the second 
> >> conversion,
> >> which is sampled at x = 0 of the second conversion. And the pixel at x' 
> >> = 1279
> >> is really x' = 639 of the second conversion, which is sampled at x = 639 
> >> * 8192/32923
> >> = 158.98, which does not read over the edge of the source tile.
> > My bad, I somehow thought that the scaling factor is calculated per
> > image (as it IMHO should be), not just per tile.
> >
> > Of course in that case you won't ever read over the edge, but on the
> > other hand the visual problems are worse because you underestimate the
> > scaling factor and introduce a sharp edge at the center: even if the
> > source pixel step per target pixel step is a fraction, between pixels
> > width/2-1 and width/2 there's always a whole source pixel step.
> >
> > Take the extreme example of scaling 32x32 to 1080x1080 pixels. The ideal
> > source pixels for x' = 519 and 520 should be x = 14.911 and 14.939,
> > respectively. Due to tiling they will be x = 15 and 16, introducing a
> > sharp seam in the otherwise blurry mess.
> 
> I think you mean at x' = 539 and x' = 540.
> 
> But yes I agree. Due to tiling, at x' = 539, the input pixel is sampled at x = 15.
> If the interpolation were to contnue (no tiling), at x' = 540, the input pixel
> would be sampled at (31/1079)*540 = 15.514. Instead, because of tiling,
> there is a discontinuity in the interpolation (it is reset), beginning again at
> x' = 0 (540), which is sampled at x = 0 (16).
> 
> The only way I can think of to resolve this problem is to add some width
> to the output tiles such that the interpolation completes a full span between
> input position w - 2 and w - 1. That is, add to w' until floor(F*w') increments
> to the next whole integer, where F = (w-1)/(w'-1) is the scaling factor.
> 
> But that will likely cause the next tile DMA addrs to fail to fall on the IDMAC
> 8 byte alignment.

I always wanted to have a look at the scroll feature, maybe SX can be
used to start at odd pixels?

regards
Philipp
diff mbox

Patch

diff --git a/drivers/gpu/ipu-v3/ipu-ic.c b/drivers/gpu/ipu-v3/ipu-ic.c
index 1a37afc..5471b72 100644
--- a/drivers/gpu/ipu-v3/ipu-ic.c
+++ b/drivers/gpu/ipu-v3/ipu-ic.c
@@ -17,6 +17,8 @@ 
 #include <linux/bitrev.h>
 #include <linux/io.h>
 #include <linux/err.h>
+#include <linux/interrupt.h>
+#include <linux/dma-mapping.h>
 #include "ipu-prv.h"
 
 /* IC Register Offsets */
@@ -82,6 +84,40 @@ 
 #define IC_IDMAC_3_PP_WIDTH_MASK        (0x3ff << 20)
 #define IC_IDMAC_3_PP_WIDTH_OFFSET      20
 
+/*
+ * The IC Resizer has a restriction that the output frame from the
+ * resizer must be 1024 or less in both width (pixels) and height
+ * (lines).
+ *
+ * The image conversion support attempts to split up a conversion when
+ * the desired output (converted) frame resolution exceeds the IC resizer
+ * limit of 1024 in either dimension.
+ *
+ * If either dimension of the output frame exceeds the limit, the
+ * dimension is split into 1, 2, or 4 equal stripes, for a maximum
+ * of 4*4 or 16 tiles. A conversion is then carried out for each
+ * tile (but taking care to pass the full frame stride length to
+ * the DMA channel's parameter memory!). IDMA double-buffering is used
+ * to convert each tile back-to-back when possible (see note below
+ * when double_buffering boolean is set).
+ *
+ * Note that the input frame must be split up into the same number
+ * of tiles as the output frame.
+ */
+#define MAX_STRIPES_W    4
+#define MAX_STRIPES_H    4
+#define MAX_TILES (MAX_STRIPES_W * MAX_STRIPES_H)
+
+#define MIN_W     128
+#define MIN_H     128
+#define MAX_W     4096
+#define MAX_H     4096
+
+enum image_convert_type {
+	IMAGE_CONVERT_IN = 0,
+	IMAGE_CONVERT_OUT,
+};
+
 struct ic_task_regoffs {
 	u32 rsc;
 	u32 tpmem_csc[2];
@@ -96,6 +132,16 @@  struct ic_task_bitfields {
 	u32 ic_cmb_galpha_bit;
 };
 
+struct ic_task_channels {
+	int in;
+	int out;
+	int rot_in;
+	int rot_out;
+	int vdi_in_p;
+	int vdi_in;
+	int vdi_in_n;
+};
+
 static const struct ic_task_regoffs ic_task_reg[IC_NUM_TASKS] = {
 	[IC_TASK_ENCODER] = {
 		.rsc = IC_PRP_ENC_RSC,
@@ -138,12 +184,159 @@  static const struct ic_task_bitfields ic_task_bit[IC_NUM_TASKS] = {
 	},
 };
 
+static const struct ic_task_channels ic_task_ch[IC_NUM_TASKS] = {
+	[IC_TASK_ENCODER] = {
+		.out = IPUV3_CHANNEL_IC_PRP_ENC_MEM,
+		.rot_in = IPUV3_CHANNEL_MEM_ROT_ENC,
+		.rot_out = IPUV3_CHANNEL_ROT_ENC_MEM,
+	},
+	[IC_TASK_VIEWFINDER] = {
+		.in = IPUV3_CHANNEL_MEM_IC_PRP_VF,
+		.out = IPUV3_CHANNEL_IC_PRP_VF_MEM,
+		.rot_in = IPUV3_CHANNEL_MEM_ROT_VF,
+		.rot_out = IPUV3_CHANNEL_ROT_VF_MEM,
+		.vdi_in_p = IPUV3_CHANNEL_MEM_VDI_PREV,
+		.vdi_in = IPUV3_CHANNEL_MEM_VDI_CUR,
+		.vdi_in_n = IPUV3_CHANNEL_MEM_VDI_NEXT,
+	},
+	[IC_TASK_POST_PROCESSOR] = {
+		.in = IPUV3_CHANNEL_MEM_IC_PP,
+		.out = IPUV3_CHANNEL_IC_PP_MEM,
+		.rot_in = IPUV3_CHANNEL_MEM_ROT_PP,
+		.rot_out = IPUV3_CHANNEL_ROT_PP_MEM,
+	},
+};
+
+struct ipu_ic_dma_buf {
+	void          *virt;
+	dma_addr_t    phys;
+	unsigned long len;
+};
+
+/* dimensions of one tile */
+struct ipu_ic_tile {
+	unsigned int width;
+	unsigned int height;
+	/* size and strides are in bytes */
+	unsigned int size;
+	unsigned int stride;
+	unsigned int rot_stride;
+};
+
+struct ipu_ic_tile_off {
+	/* start Y or packed offset of this tile */
+	u32     offset;
+	/* offset from start to tile in U plane, for planar formats */
+	u32     u_off;
+	/* offset from start to tile in V plane, for planar formats */
+	u32     v_off;
+};
+
+struct ipu_ic_pixfmt {
+	char	*name;
+	u32	fourcc;        /* V4L2 fourcc */
+	int     bpp;           /* total bpp */
+	int     y_depth;       /* depth of Y plane for planar formats */
+	int     uv_width_dec;  /* decimation in width for U/V planes */
+	int     uv_height_dec; /* decimation in height for U/V planes */
+	bool    uv_swapped;    /* U and V planes are swapped */
+	bool    uv_packed;     /* partial planar (U and V in same plane) */
+};
+
+struct ipu_ic_image {
+	struct ipu_image base;
+	enum image_convert_type type;
+
+	const struct ipu_ic_pixfmt *fmt;
+	unsigned int stride;
+
+	/* # of rows (horizontal stripes) if dest height is > 1024 */
+	unsigned int num_rows;
+	/* # of columns (vertical stripes) if dest width is > 1024 */
+	unsigned int num_cols;
+
+	struct ipu_ic_tile tile;
+	struct ipu_ic_tile_off tile_off[MAX_TILES];
+};
+
+struct image_converter_ctx;
+struct image_converter;
 struct ipu_ic_priv;
+struct ipu_ic;
+
+struct image_converter_run {
+	struct image_converter_ctx *ctx;
+
+	dma_addr_t in_phys;
+	dma_addr_t out_phys;
+
+	int status;
+
+	struct list_head list;
+};
+
+struct image_converter_ctx {
+	struct image_converter *cvt;
+
+	image_converter_cb_t complete;
+	void *complete_context;
+
+	/* Source/destination image data and rotation mode */
+	struct ipu_ic_image in;
+	struct ipu_ic_image out;
+	enum ipu_rotate_mode rot_mode;
+
+	/* intermediate buffer for rotation */
+	struct ipu_ic_dma_buf rot_intermediate[2];
+
+	/* current buffer number for double buffering */
+	int cur_buf_num;
+
+	bool aborting;
+	struct completion aborted;
+
+	/* can we use double-buffering for this conversion operation? */
+	bool double_buffering;
+	/* num_rows * num_cols */
+	unsigned int num_tiles;
+	/* next tile to process */
+	unsigned int next_tile;
+	/* where to place converted tile in dest image */
+	unsigned int out_tile_map[MAX_TILES];
+
+	struct list_head list;
+};
+
+struct image_converter {
+	struct ipu_ic *ic;
+
+	struct ipuv3_channel *in_chan;
+	struct ipuv3_channel *out_chan;
+	struct ipuv3_channel *rotation_in_chan;
+	struct ipuv3_channel *rotation_out_chan;
+
+	/* the IPU end-of-frame irqs */
+	int out_eof_irq;
+	int rot_out_eof_irq;
+
+	spinlock_t irqlock;
+
+	/* list of convert contexts */
+	struct list_head ctx_list;
+	/* queue of conversion runs */
+	struct list_head pending_q;
+	/* queue of completed runs */
+	struct list_head done_q;
+
+	/* the current conversion run */
+	struct image_converter_run *current_run;
+};
 
 struct ipu_ic {
 	enum ipu_ic_task task;
 	const struct ic_task_regoffs *reg;
 	const struct ic_task_bitfields *bit;
+	const struct ic_task_channels *ch;
 
 	enum ipu_color_space in_cs, g_in_cs;
 	enum ipu_color_space out_cs;
@@ -151,6 +344,8 @@  struct ipu_ic {
 	bool rotation;
 	bool in_use;
 
+	struct image_converter cvt;
+
 	struct ipu_ic_priv *priv;
 };
 
@@ -619,7 +814,7 @@  int ipu_ic_task_idma_init(struct ipu_ic *ic, struct ipuv3_channel *channel,
 	ipu_ic_write(ic, ic_idmac_2, IC_IDMAC_2);
 	ipu_ic_write(ic, ic_idmac_3, IC_IDMAC_3);
 
-	if (rot >= IPU_ROTATE_90_RIGHT)
+	if (ipu_rot_mode_is_irt(rot))
 		ic->rotation = true;
 
 unlock:
@@ -648,6 +843,1480 @@  static void ipu_irt_disable(struct ipu_ic *ic)
 	}
 }
 
+/*
+ * Complete image conversion support follows
+ */
+
+static const struct ipu_ic_pixfmt ipu_ic_formats[] = {
+	{
+		.name	= "RGB565",
+		.fourcc	= V4L2_PIX_FMT_RGB565,
+		.bpp    = 16,
+	}, {
+		.name	= "RGB24",
+		.fourcc	= V4L2_PIX_FMT_RGB24,
+		.bpp    = 24,
+	}, {
+		.name	= "BGR24",
+		.fourcc	= V4L2_PIX_FMT_BGR24,
+		.bpp    = 24,
+	}, {
+		.name	= "RGB32",
+		.fourcc	= V4L2_PIX_FMT_RGB32,
+		.bpp    = 32,
+	}, {
+		.name	= "BGR32",
+		.fourcc	= V4L2_PIX_FMT_BGR32,
+		.bpp    = 32,
+	}, {
+		.name	= "4:2:2 packed, YUYV",
+		.fourcc	= V4L2_PIX_FMT_YUYV,
+		.bpp    = 16,
+		.uv_width_dec = 2,
+		.uv_height_dec = 1,
+	}, {
+		.name	= "4:2:2 packed, UYVY",
+		.fourcc	= V4L2_PIX_FMT_UYVY,
+		.bpp    = 16,
+		.uv_width_dec = 2,
+		.uv_height_dec = 1,
+	}, {
+		.name	= "4:2:0 planar, YUV",
+		.fourcc	= V4L2_PIX_FMT_YUV420,
+		.bpp    = 12,
+		.y_depth = 8,
+		.uv_width_dec = 2,
+		.uv_height_dec = 2,
+	}, {
+		.name	= "4:2:0 planar, YVU",
+		.fourcc	= V4L2_PIX_FMT_YVU420,
+		.bpp    = 12,
+		.y_depth = 8,
+		.uv_width_dec = 2,
+		.uv_height_dec = 2,
+		.uv_swapped = true,
+	}, {
+		.name   = "4:2:0 partial planar, NV12",
+		.fourcc = V4L2_PIX_FMT_NV12,
+		.bpp    = 12,
+		.y_depth = 8,
+		.uv_width_dec = 2,
+		.uv_height_dec = 2,
+		.uv_packed = true,
+	}, {
+		.name   = "4:2:2 planar, YUV",
+		.fourcc = V4L2_PIX_FMT_YUV422P,
+		.bpp    = 16,
+		.y_depth = 8,
+		.uv_width_dec = 2,
+		.uv_height_dec = 1,
+	}, {
+		.name   = "4:2:2 partial planar, NV16",
+		.fourcc = V4L2_PIX_FMT_NV16,
+		.bpp    = 16,
+		.y_depth = 8,
+		.uv_width_dec = 2,
+		.uv_height_dec = 1,
+		.uv_packed = true,
+	},
+};
+
+static const struct ipu_ic_pixfmt *ipu_ic_get_format(u32 fourcc)
+{
+	const struct ipu_ic_pixfmt *ret = NULL;
+	unsigned int i;
+
+	for (i = 0; i < ARRAY_SIZE(ipu_ic_formats); i++) {
+		if (ipu_ic_formats[i].fourcc == fourcc) {
+			ret = &ipu_ic_formats[i];
+			break;
+		}
+	}
+
+	return ret;
+}
+
+static void ipu_ic_dump_format(struct image_converter_ctx *ctx,
+			       struct ipu_ic_image *ic_image)
+{
+	struct ipu_ic_priv *priv = ctx->cvt->ic->priv;
+
+	dev_dbg(priv->ipu->dev,
+		"ctx %p: %s format: %dx%d (%dx%d tiles of size %dx%d), %c%c%c%c\n",
+		ctx,
+		ic_image->type == IMAGE_CONVERT_OUT ? "Output" : "Input",
+		ic_image->base.pix.width, ic_image->base.pix.height,
+		ic_image->num_cols, ic_image->num_rows,
+		ic_image->tile.width, ic_image->tile.height,
+		ic_image->fmt->fourcc & 0xff,
+		(ic_image->fmt->fourcc >> 8) & 0xff,
+		(ic_image->fmt->fourcc >> 16) & 0xff,
+		(ic_image->fmt->fourcc >> 24) & 0xff);
+}
+
+int ipu_image_convert_enum_format(int index, const char **desc, u32 *fourcc)
+{
+	const struct ipu_ic_pixfmt *fmt;
+
+	if (index >= (int)ARRAY_SIZE(ipu_ic_formats))
+		return -EINVAL;
+
+	/* Format found */
+	fmt = &ipu_ic_formats[index];
+	*desc = fmt->name;
+	*fourcc = fmt->fourcc;
+	return 0;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_enum_format);
+
+static void ipu_ic_free_dma_buf(struct ipu_ic_priv *priv,
+				struct ipu_ic_dma_buf *buf)
+{
+	if (buf->virt)
+		dma_free_coherent(priv->ipu->dev,
+				  buf->len, buf->virt, buf->phys);
+	buf->virt = NULL;
+	buf->phys = 0;
+}
+
+static int ipu_ic_alloc_dma_buf(struct ipu_ic_priv *priv,
+				struct ipu_ic_dma_buf *buf,
+				int size)
+{
+	unsigned long newlen = PAGE_ALIGN(size);
+
+	if (buf->virt) {
+		if (buf->len == newlen)
+			return 0;
+		ipu_ic_free_dma_buf(priv, buf);
+	}
+
+	buf->len = newlen;
+	buf->virt = dma_alloc_coherent(priv->ipu->dev, buf->len, &buf->phys,
+				       GFP_DMA | GFP_KERNEL);
+	if (!buf->virt) {
+		dev_err(priv->ipu->dev, "failed to alloc dma buffer\n");
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static inline int ipu_ic_num_stripes(int dim)
+{
+	if (dim <= 1024)
+		return 1;
+	else if (dim <= 2048)
+		return 2;
+	else
+		return 4;
+}
+
+static void ipu_ic_calc_tile_dimensions(struct image_converter_ctx *ctx,
+					struct ipu_ic_image *image)
+{
+	struct ipu_ic_tile *tile = &image->tile;
+
+	tile->height = image->base.pix.height / image->num_rows;
+	tile->width = image->base.pix.width / image->num_cols;
+	tile->size = ((tile->height * image->fmt->bpp) >> 3) * tile->width;
+
+	if (image->fmt->y_depth) {
+		tile->stride = (image->fmt->y_depth * tile->width) >> 3;
+		tile->rot_stride = (image->fmt->y_depth * tile->height) >> 3;
+	} else {
+		tile->stride = (image->fmt->bpp * tile->width) >> 3;
+		tile->rot_stride = (image->fmt->bpp * tile->height) >> 3;
+	}
+}
+
+/*
+ * Use the rotation transformation to find the tile coordinates
+ * (row, col) of a tile in the destination frame that corresponds
+ * to the given tile coordinates of a source frame. The destination
+ * coordinate is then converted to a tile index.
+ */
+static int ipu_ic_transform_tile_index(struct image_converter_ctx *ctx,
+				       int src_row, int src_col)
+{
+	struct ipu_ic_priv *priv = ctx->cvt->ic->priv;
+	struct ipu_ic_image *s_image = &ctx->in;
+	struct ipu_ic_image *d_image = &ctx->out;
+	int cos, sin, dst_row, dst_col;
+
+	/* with no rotation it's a 1:1 mapping */
+	if (ctx->rot_mode == IPU_ROTATE_NONE)
+		return src_row * s_image->num_cols + src_col;
+
+	if (ctx->rot_mode & IPU_ROT_BIT_90) {
+		cos = 0;
+		sin = 1;
+	} else {
+		cos = 1;
+		sin = 0;
+	}
+
+	/*
+	 * before doing the transform, first we have to translate
+	 * source row,col for an origin in the center of s_image
+	 */
+	src_row *= 2;
+	src_col *= 2;
+	src_row -= s_image->num_rows - 1;
+	src_col -= s_image->num_cols - 1;
+
+	/* do the rotation transform */
+	dst_col = src_col * cos - src_row * sin;
+	dst_row = src_col * sin + src_row * cos;
+
+	/* apply flip */
+	if (ctx->rot_mode & IPU_ROT_BIT_HFLIP)
+		dst_col = -dst_col;
+	if (ctx->rot_mode & IPU_ROT_BIT_VFLIP)
+		dst_row = -dst_row;
+
+	dev_dbg(priv->ipu->dev, "ctx %p: [%d,%d] --> [%d,%d]\n",
+		ctx, src_col, src_row, dst_col, dst_row);
+
+	/*
+	 * finally translate dest row,col using an origin in upper
+	 * left of d_image
+	 */
+	dst_row += d_image->num_rows - 1;
+	dst_col += d_image->num_cols - 1;
+	dst_row /= 2;
+	dst_col /= 2;
+
+	return dst_row * d_image->num_cols + dst_col;
+}
+
+/*
+ * Fill the out_tile_map[] with transformed destination tile indeces.
+ */
+static void ipu_ic_calc_out_tile_map(struct image_converter_ctx *ctx)
+{
+	struct ipu_ic_image *s_image = &ctx->in;
+	unsigned int row, col, tile = 0;
+
+	for (row = 0; row < s_image->num_rows; row++) {
+		for (col = 0; col < s_image->num_cols; col++) {
+			ctx->out_tile_map[tile] =
+				ipu_ic_transform_tile_index(ctx, row, col);
+			tile++;
+		}
+	}
+}
+
+static void ipu_ic_calc_tile_offsets_planar(struct image_converter_ctx *ctx,
+					    struct ipu_ic_image *image)
+{
+	struct ipu_ic_priv *priv = ctx->cvt->ic->priv;
+	const struct ipu_ic_pixfmt *fmt = image->fmt;
+	unsigned int row, col, tile = 0;
+	u32 H, w, h, y_depth, y_stride, uv_stride;
+	u32 uv_row_off, uv_col_off, uv_off, u_off, v_off, tmp;
+	u32 y_row_off, y_col_off, y_off;
+	u32 y_size, uv_size;
+
+	/* setup some convenience vars */
+	H = image->base.pix.height;
+	w = image->tile.width;
+	h = image->tile.height;
+
+	y_depth = fmt->y_depth;
+	y_stride = image->stride;
+	uv_stride = y_stride / fmt->uv_width_dec;
+	if (fmt->uv_packed)
+		uv_stride *= 2;
+
+	y_size = H * y_stride;
+	uv_size = y_size / (fmt->uv_width_dec * fmt->uv_height_dec);
+
+	for (row = 0; row < image->num_rows; row++) {
+		y_row_off = row * h * y_stride;
+		uv_row_off = (row * h * uv_stride) / fmt->uv_height_dec;
+
+		for (col = 0; col < image->num_cols; col++) {
+			y_col_off = (col * w * y_depth) >> 3;
+			uv_col_off = y_col_off / fmt->uv_width_dec;
+			if (fmt->uv_packed)
+				uv_col_off *= 2;
+
+			y_off = y_row_off + y_col_off;
+			uv_off = uv_row_off + uv_col_off;
+
+			u_off = y_size - y_off + uv_off;
+			v_off = (fmt->uv_packed) ? 0 : u_off + uv_size;
+			if (fmt->uv_swapped) {
+				tmp = u_off;
+				u_off = v_off;
+				v_off = tmp;
+			}
+
+			image->tile_off[tile].offset = y_off;
+			image->tile_off[tile].u_off = u_off;
+			image->tile_off[tile++].v_off = v_off;
+
+			dev_dbg(priv->ipu->dev,
+				"ctx %p: %s@[%d,%d]: y_off %08x, u_off %08x, v_off %08x\n",
+				ctx, image->type == IMAGE_CONVERT_IN ?
+				"Input" : "Output", row, col,
+				y_off, u_off, v_off);
+		}
+	}
+}
+
+static void ipu_ic_calc_tile_offsets_packed(struct image_converter_ctx *ctx,
+					    struct ipu_ic_image *image)
+{
+	struct ipu_ic_priv *priv = ctx->cvt->ic->priv;
+	const struct ipu_ic_pixfmt *fmt = image->fmt;
+	unsigned int row, col, tile = 0;
+	u32 w, h, bpp, stride;
+	u32 row_off, col_off;
+
+	/* setup some convenience vars */
+	w = image->tile.width;
+	h = image->tile.height;
+	stride = image->stride;
+	bpp = fmt->bpp;
+
+	for (row = 0; row < image->num_rows; row++) {
+		row_off = row * h * stride;
+
+		for (col = 0; col < image->num_cols; col++) {
+			col_off = (col * w * bpp) >> 3;
+
+			image->tile_off[tile].offset = row_off + col_off;
+			image->tile_off[tile].u_off = 0;
+			image->tile_off[tile++].v_off = 0;
+
+			dev_dbg(priv->ipu->dev,
+				"ctx %p: %s@[%d,%d]: phys %08x\n", ctx,
+				image->type == IMAGE_CONVERT_IN ?
+				"Input" : "Output", row, col,
+				row_off + col_off);
+		}
+	}
+}
+
+static void ipu_ic_calc_tile_offsets(struct image_converter_ctx *ctx,
+				     struct ipu_ic_image *image)
+{
+	memset(image->tile_off, 0, sizeof(image->tile_off));
+
+	if (image->fmt->y_depth)
+		ipu_ic_calc_tile_offsets_planar(ctx, image);
+	else
+		ipu_ic_calc_tile_offsets_packed(ctx, image);
+}
+
+/*
+ * return the number of runs in given queue (pending_q or done_q)
+ * for this context. hold irqlock when calling.
+ */
+static int ipu_ic_get_run_count(struct image_converter_ctx *ctx,
+				struct list_head *q)
+{
+	struct image_converter_run *run;
+	int count = 0;
+
+	list_for_each_entry(run, q, list) {
+		if (run->ctx == ctx)
+			count++;
+	}
+
+	return count;
+}
+
+/* hold irqlock when calling */
+static void ipu_ic_convert_stop(struct image_converter_run *run)
+{
+	struct image_converter_ctx *ctx = run->ctx;
+	struct image_converter *cvt = ctx->cvt;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+
+	dev_dbg(priv->ipu->dev, "%s: stopping ctx %p run %p\n",
+		__func__, ctx, run);
+
+	/* disable IC tasks and the channels */
+	ipu_ic_task_disable(cvt->ic);
+	ipu_idmac_disable_channel(cvt->in_chan);
+	ipu_idmac_disable_channel(cvt->out_chan);
+
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		ipu_idmac_disable_channel(cvt->rotation_in_chan);
+		ipu_idmac_disable_channel(cvt->rotation_out_chan);
+		ipu_idmac_unlink(cvt->out_chan, cvt->rotation_in_chan);
+	}
+
+	ipu_ic_disable(cvt->ic);
+}
+
+/* hold irqlock when calling */
+static void init_idmac_channel(struct image_converter_ctx *ctx,
+			       struct ipuv3_channel *channel,
+			       struct ipu_ic_image *image,
+			       enum ipu_rotate_mode rot_mode,
+			       bool rot_swap_width_height)
+{
+	struct image_converter *cvt = ctx->cvt;
+	unsigned int burst_size;
+	u32 width, height, stride;
+	dma_addr_t addr0, addr1 = 0;
+	struct ipu_image tile_image;
+	unsigned int tile_idx[2];
+
+	if (image->type == IMAGE_CONVERT_OUT) {
+		tile_idx[0] = ctx->out_tile_map[0];
+		tile_idx[1] = ctx->out_tile_map[1];
+	} else {
+		tile_idx[0] = 0;
+		tile_idx[1] = 1;
+	}
+
+	if (rot_swap_width_height) {
+		width = image->tile.height;
+		height = image->tile.width;
+		stride = image->tile.rot_stride;
+		addr0 = ctx->rot_intermediate[0].phys;
+		if (ctx->double_buffering)
+			addr1 = ctx->rot_intermediate[1].phys;
+	} else {
+		width = image->tile.width;
+		height = image->tile.height;
+		stride = image->stride;
+		addr0 = image->base.phys0 +
+			image->tile_off[tile_idx[0]].offset;
+		if (ctx->double_buffering)
+			addr1 = image->base.phys0 +
+				image->tile_off[tile_idx[1]].offset;
+	}
+
+	ipu_cpmem_zero(channel);
+
+	memset(&tile_image, 0, sizeof(tile_image));
+	tile_image.pix.width = tile_image.rect.width = width;
+	tile_image.pix.height = tile_image.rect.height = height;
+	tile_image.pix.bytesperline = stride;
+	tile_image.pix.pixelformat =  image->fmt->fourcc;
+	tile_image.phys0 = addr0;
+	tile_image.phys1 = addr1;
+	ipu_cpmem_set_image(channel, &tile_image);
+
+	if (image->fmt->y_depth && !rot_swap_width_height)
+		ipu_cpmem_set_uv_offset(channel,
+					image->tile_off[tile_idx[0]].u_off,
+					image->tile_off[tile_idx[0]].v_off);
+
+	if (rot_mode)
+		ipu_cpmem_set_rotation(channel, rot_mode);
+
+	if (channel == cvt->rotation_in_chan ||
+	    channel == cvt->rotation_out_chan) {
+		burst_size = 8;
+		ipu_cpmem_set_block_mode(channel);
+	} else
+		burst_size = (width % 16) ? 8 : 16;
+
+	ipu_cpmem_set_burstsize(channel, burst_size);
+
+	ipu_ic_task_idma_init(cvt->ic, channel, width, height,
+			      burst_size, rot_mode);
+
+	ipu_cpmem_set_axi_id(channel, 1);
+
+	ipu_idmac_set_double_buffer(channel, ctx->double_buffering);
+}
+
+/* hold irqlock when calling */
+static int ipu_ic_convert_start(struct image_converter_run *run)
+{
+	struct image_converter_ctx *ctx = run->ctx;
+	struct image_converter *cvt = ctx->cvt;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct ipu_ic_image *s_image = &ctx->in;
+	struct ipu_ic_image *d_image = &ctx->out;
+	enum ipu_color_space src_cs, dest_cs;
+	unsigned int dest_width, dest_height;
+	int ret;
+
+	dev_dbg(priv->ipu->dev, "%s: starting ctx %p run %p\n",
+		__func__, ctx, run);
+
+	src_cs = ipu_pixelformat_to_colorspace(s_image->fmt->fourcc);
+	dest_cs = ipu_pixelformat_to_colorspace(d_image->fmt->fourcc);
+
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		/* swap width/height for resizer */
+		dest_width = d_image->tile.height;
+		dest_height = d_image->tile.width;
+	} else {
+		dest_width = d_image->tile.width;
+		dest_height = d_image->tile.height;
+	}
+
+	/* setup the IC resizer and CSC */
+	ret = ipu_ic_task_init(cvt->ic,
+			       s_image->tile.width,
+			       s_image->tile.height,
+			       dest_width,
+			       dest_height,
+			       src_cs, dest_cs);
+	if (ret) {
+		dev_err(priv->ipu->dev, "ipu_ic_task_init failed, %d\n", ret);
+		return ret;
+	}
+
+	/* init the source MEM-->IC PP IDMAC channel */
+	init_idmac_channel(ctx, cvt->in_chan, s_image,
+			   IPU_ROTATE_NONE, false);
+
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		/* init the IC PP-->MEM IDMAC channel */
+		init_idmac_channel(ctx, cvt->out_chan, d_image,
+				   IPU_ROTATE_NONE, true);
+
+		/* init the MEM-->IC PP ROT IDMAC channel */
+		init_idmac_channel(ctx, cvt->rotation_in_chan, d_image,
+				   ctx->rot_mode, true);
+
+		/* init the destination IC PP ROT-->MEM IDMAC channel */
+		init_idmac_channel(ctx, cvt->rotation_out_chan, d_image,
+				   IPU_ROTATE_NONE, false);
+
+		/* now link IC PP-->MEM to MEM-->IC PP ROT */
+		ipu_idmac_link(cvt->out_chan, cvt->rotation_in_chan);
+	} else {
+		/* init the destination IC PP-->MEM IDMAC channel */
+		init_idmac_channel(ctx, cvt->out_chan, d_image,
+				   ctx->rot_mode, false);
+	}
+
+	/* enable the IC */
+	ipu_ic_enable(cvt->ic);
+
+	/* set buffers ready */
+	ipu_idmac_select_buffer(cvt->in_chan, 0);
+	ipu_idmac_select_buffer(cvt->out_chan, 0);
+	if (ipu_rot_mode_is_irt(ctx->rot_mode))
+		ipu_idmac_select_buffer(cvt->rotation_out_chan, 0);
+	if (ctx->double_buffering) {
+		ipu_idmac_select_buffer(cvt->in_chan, 1);
+		ipu_idmac_select_buffer(cvt->out_chan, 1);
+		if (ipu_rot_mode_is_irt(ctx->rot_mode))
+			ipu_idmac_select_buffer(cvt->rotation_out_chan, 1);
+	}
+
+	/* enable the channels! */
+	ipu_idmac_enable_channel(cvt->in_chan);
+	ipu_idmac_enable_channel(cvt->out_chan);
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		ipu_idmac_enable_channel(cvt->rotation_in_chan);
+		ipu_idmac_enable_channel(cvt->rotation_out_chan);
+	}
+
+	ipu_ic_task_enable(cvt->ic);
+
+	ipu_cpmem_dump(cvt->in_chan);
+	ipu_cpmem_dump(cvt->out_chan);
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		ipu_cpmem_dump(cvt->rotation_in_chan);
+		ipu_cpmem_dump(cvt->rotation_out_chan);
+	}
+
+	ipu_dump(priv->ipu);
+
+	return 0;
+}
+
+/* hold irqlock when calling */
+static int ipu_ic_run(struct image_converter_run *run)
+{
+	struct image_converter_ctx *ctx = run->ctx;
+	struct image_converter *cvt = ctx->cvt;
+
+	ctx->in.base.phys0 = run->in_phys;
+	ctx->out.base.phys0 = run->out_phys;
+
+	ctx->cur_buf_num = 0;
+	ctx->next_tile = 1;
+
+	/* remove run from pending_q and set as current */
+	list_del(&run->list);
+	cvt->current_run = run;
+
+	return ipu_ic_convert_start(run);
+}
+
+/* hold irqlock when calling */
+static void ipu_ic_run_next(struct image_converter *cvt)
+{
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct image_converter_run *run, *tmp;
+	int ret;
+
+	list_for_each_entry_safe(run, tmp, &cvt->pending_q, list) {
+		/* skip contexts that are aborting */
+		if (run->ctx->aborting) {
+			dev_dbg(priv->ipu->dev,
+				 "%s: skipping aborting ctx %p run %p\n",
+				 __func__, run->ctx, run);
+			continue;
+		}
+
+		ret = ipu_ic_run(run);
+		if (!ret)
+			break;
+
+		/*
+		 * something went wrong with start, add the run
+		 * to done q and continue to the next run in the
+		 * pending q.
+		 */
+		run->status = ret;
+		list_add_tail(&run->list, &cvt->done_q);
+		cvt->current_run = NULL;
+	}
+}
+
+static void ipu_ic_empty_done_q(struct image_converter *cvt)
+{
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct image_converter_run *run;
+	unsigned long flags;
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	while (!list_empty(&cvt->done_q)) {
+		run = list_entry(cvt->done_q.next,
+				 struct image_converter_run,
+				 list);
+
+		list_del(&run->list);
+
+		dev_dbg(priv->ipu->dev,
+			"%s: completing ctx %p run %p with %d\n",
+			__func__, run->ctx, run, run->status);
+
+		/* call the completion callback and free the run */
+		spin_unlock_irqrestore(&cvt->irqlock, flags);
+		run->ctx->complete(run->ctx->complete_context, run,
+				   run->status);
+		kfree(run);
+		spin_lock_irqsave(&cvt->irqlock, flags);
+	}
+
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+}
+
+/*
+ * the bottom half thread clears out the done_q, calling the
+ * completion handler for each.
+ */
+static irqreturn_t ipu_ic_bh(int irq, void *dev_id)
+{
+	struct image_converter *cvt = dev_id;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct image_converter_ctx *ctx;
+	unsigned long flags;
+
+	dev_dbg(priv->ipu->dev, "%s: enter\n", __func__);
+
+	ipu_ic_empty_done_q(cvt);
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	/*
+	 * the done_q is cleared out, signal any contexts
+	 * that are aborting that abort can complete.
+	 */
+	list_for_each_entry(ctx, &cvt->ctx_list, list) {
+		if (ctx->aborting) {
+			dev_dbg(priv->ipu->dev,
+				 "%s: signaling abort for ctx %p\n",
+				 __func__, ctx);
+			complete(&ctx->aborted);
+		}
+	}
+
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+
+	dev_dbg(priv->ipu->dev, "%s: exit\n", __func__);
+	return IRQ_HANDLED;
+}
+
+/* hold irqlock when calling */
+static irqreturn_t ipu_ic_doirq(struct image_converter_run *run)
+{
+	struct image_converter_ctx *ctx = run->ctx;
+	struct image_converter *cvt = ctx->cvt;
+	struct ipu_ic_tile_off *src_off, *dst_off;
+	struct ipu_ic_image *s_image = &ctx->in;
+	struct ipu_ic_image *d_image = &ctx->out;
+	struct ipuv3_channel *outch;
+	unsigned int dst_idx;
+
+	outch = ipu_rot_mode_is_irt(ctx->rot_mode) ?
+		cvt->rotation_out_chan : cvt->out_chan;
+
+	/*
+	 * It is difficult to stop the channel DMA before the channels
+	 * enter the paused state. Without double-buffering the channels
+	 * are always in a paused state when the EOF irq occurs, so it
+	 * is safe to stop the channels now. For double-buffering we
+	 * just ignore the abort until the operation completes, when it
+	 * is safe to shut down.
+	 */
+	if (ctx->aborting && !ctx->double_buffering) {
+		ipu_ic_convert_stop(run);
+		run->status = -EIO;
+		goto done;
+	}
+
+	if (ctx->next_tile == ctx->num_tiles) {
+		/*
+		 * the conversion is complete
+		 */
+		ipu_ic_convert_stop(run);
+		run->status = 0;
+		goto done;
+	}
+
+	/*
+	 * not done, place the next tile buffers.
+	 */
+	if (!ctx->double_buffering) {
+
+		src_off = &s_image->tile_off[ctx->next_tile];
+		dst_idx = ctx->out_tile_map[ctx->next_tile];
+		dst_off = &d_image->tile_off[dst_idx];
+
+		ipu_cpmem_set_buffer(cvt->in_chan, 0,
+				     s_image->base.phys0 + src_off->offset);
+		ipu_cpmem_set_buffer(outch, 0,
+				     d_image->base.phys0 + dst_off->offset);
+		if (s_image->fmt->y_depth)
+			ipu_cpmem_set_uv_offset(cvt->in_chan,
+						src_off->u_off,
+						src_off->v_off);
+		if (d_image->fmt->y_depth)
+			ipu_cpmem_set_uv_offset(outch,
+						dst_off->u_off,
+						dst_off->v_off);
+
+		ipu_idmac_select_buffer(cvt->in_chan, 0);
+		ipu_idmac_select_buffer(outch, 0);
+
+	} else if (ctx->next_tile < ctx->num_tiles - 1) {
+
+		src_off = &s_image->tile_off[ctx->next_tile + 1];
+		dst_idx = ctx->out_tile_map[ctx->next_tile + 1];
+		dst_off = &d_image->tile_off[dst_idx];
+
+		ipu_cpmem_set_buffer(cvt->in_chan, ctx->cur_buf_num,
+				     s_image->base.phys0 + src_off->offset);
+		ipu_cpmem_set_buffer(outch, ctx->cur_buf_num,
+				     d_image->base.phys0 + dst_off->offset);
+
+		ipu_idmac_select_buffer(cvt->in_chan, ctx->cur_buf_num);
+		ipu_idmac_select_buffer(outch, ctx->cur_buf_num);
+
+		ctx->cur_buf_num ^= 1;
+	}
+
+	ctx->next_tile++;
+	return IRQ_HANDLED;
+done:
+	list_add_tail(&run->list, &cvt->done_q);
+	cvt->current_run = NULL;
+	ipu_ic_run_next(cvt);
+	return IRQ_WAKE_THREAD;
+}
+
+static irqreturn_t ipu_ic_norotate_irq(int irq, void *data)
+{
+	struct image_converter *cvt = data;
+	struct image_converter_ctx *ctx;
+	struct image_converter_run *run;
+	unsigned long flags;
+	irqreturn_t ret;
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	/* get current run and its context */
+	run = cvt->current_run;
+	if (!run) {
+		ret = IRQ_NONE;
+		goto out;
+	}
+
+	ctx = run->ctx;
+
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		/* this is a rotation operation, just ignore */
+		spin_unlock_irqrestore(&cvt->irqlock, flags);
+		return IRQ_HANDLED;
+	}
+
+	ret = ipu_ic_doirq(run);
+out:
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+	return ret;
+}
+
+static irqreturn_t ipu_ic_rotate_irq(int irq, void *data)
+{
+	struct image_converter *cvt = data;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct image_converter_ctx *ctx;
+	struct image_converter_run *run;
+	unsigned long flags;
+	irqreturn_t ret;
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	/* get current run and its context */
+	run = cvt->current_run;
+	if (!run) {
+		ret = IRQ_NONE;
+		goto out;
+	}
+
+	ctx = run->ctx;
+
+	if (!ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		/* this was NOT a rotation operation, shouldn't happen */
+		dev_err(priv->ipu->dev, "Unexpected rotation interrupt\n");
+		spin_unlock_irqrestore(&cvt->irqlock, flags);
+		return IRQ_HANDLED;
+	}
+
+	ret = ipu_ic_doirq(run);
+out:
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+	return ret;
+}
+
+/*
+ * try to force the completion of runs for this ctx. Called when
+ * abort wait times out in ipu_image_convert_abort().
+ */
+static void ipu_ic_force_abort(struct image_converter_ctx *ctx)
+{
+	struct image_converter *cvt = ctx->cvt;
+	struct image_converter_run *run;
+	unsigned long flags;
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	run = cvt->current_run;
+	if (run && run->ctx == ctx) {
+		ipu_ic_convert_stop(run);
+		run->status = -EIO;
+		list_add_tail(&run->list, &cvt->done_q);
+		cvt->current_run = NULL;
+		ipu_ic_run_next(cvt);
+	}
+
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+
+	ipu_ic_empty_done_q(cvt);
+}
+
+static void ipu_ic_release_ipu_resources(struct image_converter *cvt)
+{
+	if (cvt->out_eof_irq >= 0)
+		free_irq(cvt->out_eof_irq, cvt);
+	if (cvt->rot_out_eof_irq >= 0)
+		free_irq(cvt->rot_out_eof_irq, cvt);
+
+	if (!IS_ERR_OR_NULL(cvt->in_chan))
+		ipu_idmac_put(cvt->in_chan);
+	if (!IS_ERR_OR_NULL(cvt->out_chan))
+		ipu_idmac_put(cvt->out_chan);
+	if (!IS_ERR_OR_NULL(cvt->rotation_in_chan))
+		ipu_idmac_put(cvt->rotation_in_chan);
+	if (!IS_ERR_OR_NULL(cvt->rotation_out_chan))
+		ipu_idmac_put(cvt->rotation_out_chan);
+
+	cvt->in_chan = cvt->out_chan = cvt->rotation_in_chan =
+		cvt->rotation_out_chan = NULL;
+	cvt->out_eof_irq = cvt->rot_out_eof_irq = -1;
+}
+
+static int ipu_ic_get_ipu_resources(struct image_converter *cvt)
+{
+	const struct ic_task_channels *chan = cvt->ic->ch;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	int ret;
+
+	/* get IDMAC channels */
+	cvt->in_chan = ipu_idmac_get(priv->ipu, chan->in);
+	cvt->out_chan = ipu_idmac_get(priv->ipu, chan->out);
+	if (IS_ERR(cvt->in_chan) || IS_ERR(cvt->out_chan)) {
+		dev_err(priv->ipu->dev, "could not acquire idmac channels\n");
+		ret = -EBUSY;
+		goto err;
+	}
+
+	cvt->rotation_in_chan = ipu_idmac_get(priv->ipu, chan->rot_in);
+	cvt->rotation_out_chan = ipu_idmac_get(priv->ipu, chan->rot_out);
+	if (IS_ERR(cvt->rotation_in_chan) || IS_ERR(cvt->rotation_out_chan)) {
+		dev_err(priv->ipu->dev,
+			"could not acquire idmac rotation channels\n");
+		ret = -EBUSY;
+		goto err;
+	}
+
+	/* acquire the EOF interrupts */
+	cvt->out_eof_irq = ipu_idmac_channel_irq(priv->ipu,
+						cvt->out_chan,
+						IPU_IRQ_EOF);
+
+	ret = request_threaded_irq(cvt->out_eof_irq,
+				   ipu_ic_norotate_irq, ipu_ic_bh,
+				   0, "ipu-ic", cvt);
+	if (ret < 0) {
+		dev_err(priv->ipu->dev, "could not acquire irq %d\n",
+			 cvt->out_eof_irq);
+		cvt->out_eof_irq = -1;
+		goto err;
+	}
+
+	cvt->rot_out_eof_irq = ipu_idmac_channel_irq(priv->ipu,
+						     cvt->rotation_out_chan,
+						     IPU_IRQ_EOF);
+
+	ret = request_threaded_irq(cvt->rot_out_eof_irq,
+				   ipu_ic_rotate_irq, ipu_ic_bh,
+				   0, "ipu-ic", cvt);
+	if (ret < 0) {
+		dev_err(priv->ipu->dev, "could not acquire irq %d\n",
+			cvt->rot_out_eof_irq);
+		cvt->rot_out_eof_irq = -1;
+		goto err;
+	}
+
+	return 0;
+err:
+	ipu_ic_release_ipu_resources(cvt);
+	return ret;
+}
+
+static int ipu_ic_fill_image(struct image_converter_ctx *ctx,
+			     struct ipu_ic_image *ic_image,
+			     struct ipu_image *image,
+			     enum image_convert_type type)
+{
+	struct ipu_ic_priv *priv = ctx->cvt->ic->priv;
+
+	ic_image->base = *image;
+	ic_image->type = type;
+
+	ic_image->fmt = ipu_ic_get_format(image->pix.pixelformat);
+	if (!ic_image->fmt) {
+		dev_err(priv->ipu->dev, "pixelformat not supported for %s\n",
+			type == IMAGE_CONVERT_OUT ? "Output" : "Input");
+		return -EINVAL;
+	}
+
+	if (ic_image->fmt->y_depth)
+		ic_image->stride = (ic_image->fmt->y_depth *
+				    ic_image->base.pix.width) >> 3;
+	else
+		ic_image->stride  = ic_image->base.pix.bytesperline;
+
+	ipu_ic_calc_tile_dimensions(ctx, ic_image);
+	ipu_ic_calc_tile_offsets(ctx, ic_image);
+
+	return 0;
+}
+
+/* borrowed from drivers/media/v4l2-core/v4l2-common.c */
+static unsigned int clamp_align(unsigned int x, unsigned int min,
+				unsigned int max, unsigned int align)
+{
+	/* Bits that must be zero to be aligned */
+	unsigned int mask = ~((1 << align) - 1);
+
+	/* Clamp to aligned min and max */
+	x = clamp(x, (min + ~mask) & mask, max & mask);
+
+	/* Round to nearest aligned value */
+	if (align)
+		x = (x + (1 << (align - 1))) & mask;
+
+	return x;
+}
+
+/*
+ * We have to adjust the tile width such that the tile physaddrs and
+ * U and V plane offsets are multiples of 8 bytes as required by
+ * the IPU DMA Controller. For the planar formats, this corresponds
+ * to a pixel alignment of 16 (but use a more formal equation since
+ * the variables are available). For all the packed formats, 8 is
+ * good enough.
+ */
+static inline u32 tile_width_align(const struct ipu_ic_pixfmt *fmt)
+{
+	return fmt->y_depth ? (64 * fmt->uv_width_dec) / fmt->y_depth : 8;
+}
+
+/*
+ * For tile height alignment, we have to ensure that the output tile
+ * heights are multiples of 8 lines if the IRT is required by the
+ * given rotation mode (the IRT performs rotations on 8x8 blocks
+ * at a time). If the IRT is not used, or for input image tiles,
+ * 2 lines are good enough.
+ */
+static inline u32 tile_height_align(enum image_convert_type type,
+				    enum ipu_rotate_mode rot_mode)
+{
+	return (type == IMAGE_CONVERT_OUT &&
+		ipu_rot_mode_is_irt(rot_mode)) ? 8 : 2;
+}
+
+/* Adjusts input/output images to IPU restrictions */
+int ipu_image_convert_adjust(struct ipu_image *in, struct ipu_image *out,
+			     enum ipu_rotate_mode rot_mode)
+{
+	const struct ipu_ic_pixfmt *infmt, *outfmt;
+	unsigned int num_in_rows, num_in_cols;
+	unsigned int num_out_rows, num_out_cols;
+	u32 w_align, h_align;
+
+	infmt = ipu_ic_get_format(in->pix.pixelformat);
+	outfmt = ipu_ic_get_format(out->pix.pixelformat);
+
+	/* set some defaults if needed */
+	if (!infmt) {
+		in->pix.pixelformat = V4L2_PIX_FMT_RGB24;
+		infmt = ipu_ic_get_format(V4L2_PIX_FMT_RGB24);
+	}
+	if (!outfmt) {
+		out->pix.pixelformat = V4L2_PIX_FMT_RGB24;
+		outfmt = ipu_ic_get_format(V4L2_PIX_FMT_RGB24);
+	}
+
+	if (!in->pix.width || !in->pix.height) {
+		in->pix.width = 640;
+		in->pix.height = 480;
+	}
+	if (!out->pix.width || !out->pix.height) {
+		out->pix.width = 640;
+		out->pix.height = 480;
+	}
+
+	/* image converter does not handle fields */
+	in->pix.field = out->pix.field = V4L2_FIELD_NONE;
+
+	/* resizer cannot downsize more than 4:1 */
+	if (ipu_rot_mode_is_irt(rot_mode)) {
+		out->pix.height = max_t(__u32, out->pix.height,
+					in->pix.width / 4);
+		out->pix.width = max_t(__u32, out->pix.width,
+				       in->pix.height / 4);
+	} else {
+		out->pix.width = max_t(__u32, out->pix.width,
+				       in->pix.width / 4);
+		out->pix.height = max_t(__u32, out->pix.height,
+					in->pix.height / 4);
+	}
+
+	/* get tiling rows/cols from output format */
+	num_out_rows = ipu_ic_num_stripes(out->pix.height);
+	num_out_cols = ipu_ic_num_stripes(out->pix.width);
+	if (ipu_rot_mode_is_irt(rot_mode)) {
+		num_in_rows = num_out_cols;
+		num_in_cols = num_out_rows;
+	} else {
+		num_in_rows = num_out_rows;
+		num_in_cols = num_out_cols;
+	}
+
+	/* align input width/height */
+	w_align = ilog2(tile_width_align(infmt) * num_in_cols);
+	h_align = ilog2(tile_height_align(IMAGE_CONVERT_IN, rot_mode) *
+			num_in_rows);
+	in->pix.width = clamp_align(in->pix.width, MIN_W, MAX_W, w_align);
+	in->pix.height = clamp_align(in->pix.height, MIN_H, MAX_H, h_align);
+
+	/* align output width/height */
+	w_align = ilog2(tile_width_align(outfmt) * num_out_cols);
+	h_align = ilog2(tile_height_align(IMAGE_CONVERT_OUT, rot_mode) *
+			num_out_rows);
+	out->pix.width = clamp_align(out->pix.width, MIN_W, MAX_W, w_align);
+	out->pix.height = clamp_align(out->pix.height, MIN_H, MAX_H, h_align);
+
+	/* set input/output strides and image sizes */
+	in->pix.bytesperline = (in->pix.width * infmt->bpp) >> 3;
+	in->pix.sizeimage = in->pix.height * in->pix.bytesperline;
+	out->pix.bytesperline = (out->pix.width * outfmt->bpp) >> 3;
+	out->pix.sizeimage = out->pix.height * out->pix.bytesperline;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_adjust);
+
+/*
+ * this is used by ipu_image_convert_prepare() to verify set input and
+ * output images are valid before starting the conversion. Clients can
+ * also call it before calling ipu_image_convert_prepare().
+ */
+int ipu_image_convert_verify(struct ipu_image *in, struct ipu_image *out,
+			     enum ipu_rotate_mode rot_mode)
+{
+	struct ipu_image testin, testout;
+	int ret;
+
+	testin = *in;
+	testout = *out;
+
+	ret = ipu_image_convert_adjust(&testin, &testout, rot_mode);
+	if (ret)
+		return ret;
+
+	if (testin.pix.width != in->pix.width ||
+	    testin.pix.height != in->pix.height ||
+	    testout.pix.width != out->pix.width ||
+	    testout.pix.height != out->pix.height)
+		return -EINVAL;
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_verify);
+
+/*
+ * Call ipu_image_convert_prepare() to prepare for the conversion of
+ * given images and rotation mode. Returns a new conversion context.
+ */
+struct image_converter_ctx *
+ipu_image_convert_prepare(struct ipu_ic *ic,
+			  struct ipu_image *in, struct ipu_image *out,
+			  enum ipu_rotate_mode rot_mode,
+			  image_converter_cb_t complete,
+			  void *complete_context)
+{
+	struct ipu_ic_priv *priv = ic->priv;
+	struct image_converter *cvt = &ic->cvt;
+	struct ipu_ic_image *s_image, *d_image;
+	struct image_converter_ctx *ctx;
+	unsigned long flags;
+	bool get_res;
+	int ret;
+
+	if (!ic || !in || !out || !complete)
+		return ERR_PTR(-EINVAL);
+
+	/* verify the in/out images before continuing */
+	ret = ipu_image_convert_verify(in, out, rot_mode);
+	if (ret) {
+		dev_err(priv->ipu->dev, "%s: in/out formats invalid\n",
+			__func__);
+		return ERR_PTR(ret);
+	}
+
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return ERR_PTR(-ENOMEM);
+
+	dev_dbg(priv->ipu->dev, "%s: ctx %p\n", __func__, ctx);
+
+	ctx->cvt = cvt;
+	init_completion(&ctx->aborted);
+
+	s_image = &ctx->in;
+	d_image = &ctx->out;
+
+	/* set tiling and rotation */
+	d_image->num_rows = ipu_ic_num_stripes(out->pix.height);
+	d_image->num_cols = ipu_ic_num_stripes(out->pix.width);
+	if (ipu_rot_mode_is_irt(rot_mode)) {
+		s_image->num_rows = d_image->num_cols;
+		s_image->num_cols = d_image->num_rows;
+	} else {
+		s_image->num_rows = d_image->num_rows;
+		s_image->num_cols = d_image->num_cols;
+	}
+
+	ctx->num_tiles = d_image->num_cols * d_image->num_rows;
+	ctx->rot_mode = rot_mode;
+
+	ret = ipu_ic_fill_image(ctx, s_image, in, IMAGE_CONVERT_IN);
+	if (ret)
+		goto out_free;
+	ret = ipu_ic_fill_image(ctx, d_image, out, IMAGE_CONVERT_OUT);
+	if (ret)
+		goto out_free;
+
+	ipu_ic_calc_out_tile_map(ctx);
+
+	ipu_ic_dump_format(ctx, s_image);
+	ipu_ic_dump_format(ctx, d_image);
+
+	ctx->complete = complete;
+	ctx->complete_context = complete_context;
+
+	/*
+	 * Can we use double-buffering for this operation? If there is
+	 * only one tile (the whole image can be converted in a single
+	 * operation) there's no point in using double-buffering. Also,
+	 * the IPU's IDMAC channels allow only a single U and V plane
+	 * offset shared between both buffers, but these offsets change
+	 * for every tile, and therefore would have to be updated for
+	 * each buffer which is not possible. So double-buffering is
+	 * impossible when either the source or destination images are
+	 * a planar format (YUV420, YUV422P, etc.).
+	 */
+	ctx->double_buffering = (ctx->num_tiles > 1 &&
+				 !s_image->fmt->y_depth &&
+				 !d_image->fmt->y_depth);
+
+	if (ipu_rot_mode_is_irt(ctx->rot_mode)) {
+		ret = ipu_ic_alloc_dma_buf(priv, &ctx->rot_intermediate[0],
+					   d_image->tile.size);
+		if (ret)
+			goto out_free;
+		if (ctx->double_buffering) {
+			ret = ipu_ic_alloc_dma_buf(priv,
+						   &ctx->rot_intermediate[1],
+						   d_image->tile.size);
+			if (ret)
+				goto out_free_dmabuf0;
+		}
+	}
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	get_res = list_empty(&cvt->ctx_list);
+
+	list_add_tail(&ctx->list, &cvt->ctx_list);
+
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+
+	if (get_res) {
+		ret = ipu_ic_get_ipu_resources(cvt);
+		if (ret)
+			goto out_free_dmabuf1;
+	}
+
+	return ctx;
+
+out_free_dmabuf1:
+	ipu_ic_free_dma_buf(priv, &ctx->rot_intermediate[1]);
+	spin_lock_irqsave(&cvt->irqlock, flags);
+	list_del(&ctx->list);
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+out_free_dmabuf0:
+	ipu_ic_free_dma_buf(priv, &ctx->rot_intermediate[0]);
+out_free:
+	kfree(ctx);
+	return ERR_PTR(ret);
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_prepare);
+
+/*
+ * Carry out a single image conversion. Only the physaddr's of the input
+ * and output image buffers are needed. The conversion context must have
+ * been created previously with ipu_image_convert_prepare(). Returns the
+ * new run object.
+ */
+struct image_converter_run *
+ipu_image_convert_run(struct image_converter_ctx *ctx,
+		      dma_addr_t in_phys, dma_addr_t out_phys)
+{
+	struct image_converter *cvt = ctx->cvt;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct image_converter_run *run;
+	unsigned long flags;
+	int ret = 0;
+
+	run = kzalloc(sizeof(*run), GFP_KERNEL);
+	if (!run)
+		return ERR_PTR(-ENOMEM);
+
+	run->ctx = ctx;
+	run->in_phys = in_phys;
+	run->out_phys = out_phys;
+
+	dev_dbg(priv->ipu->dev, "%s: ctx %p run %p\n", __func__,
+		ctx, run);
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	if (ctx->aborting) {
+		ret = -EIO;
+		goto unlock;
+	}
+
+	list_add_tail(&run->list, &cvt->pending_q);
+
+	if (!cvt->current_run) {
+		ret = ipu_ic_run(run);
+		if (ret)
+			cvt->current_run = NULL;
+	}
+unlock:
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+
+	if (ret) {
+		kfree(run);
+		run = ERR_PTR(ret);
+	}
+
+	return run;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_run);
+
+/* Abort any active or pending conversions for this context */
+void ipu_image_convert_abort(struct image_converter_ctx *ctx)
+{
+	struct image_converter *cvt = ctx->cvt;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	struct image_converter_run *run, *active_run, *tmp;
+	unsigned long flags;
+	int run_count, ret;
+	bool need_abort;
+
+	reinit_completion(&ctx->aborted);
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	/* move all remaining pending runs in this context to done_q */
+	list_for_each_entry_safe(run, tmp, &cvt->pending_q, list) {
+		if (run->ctx != ctx)
+			continue;
+		run->status = -EIO;
+		list_move_tail(&run->list, &cvt->done_q);
+	}
+
+	run_count = ipu_ic_get_run_count(ctx, &cvt->done_q);
+	active_run = (cvt->current_run && cvt->current_run->ctx == ctx) ?
+		cvt->current_run : NULL;
+
+	need_abort = (run_count || active_run);
+
+	ctx->aborting = need_abort;
+
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+
+	if (!need_abort) {
+		dev_dbg(priv->ipu->dev, "%s: no abort needed for ctx %p\n",
+			__func__, ctx);
+		return;
+	}
+
+	dev_dbg(priv->ipu->dev,
+		 "%s: wait for completion: %d runs, active run %p\n",
+		 __func__, run_count, active_run);
+
+	ret = wait_for_completion_timeout(&ctx->aborted,
+					  msecs_to_jiffies(10000));
+	if (ret == 0) {
+		dev_warn(priv->ipu->dev, "%s: timeout\n", __func__);
+		ipu_ic_force_abort(ctx);
+	}
+
+	ctx->aborting = false;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_abort);
+
+/* Unprepare image conversion context */
+void ipu_image_convert_unprepare(struct image_converter_ctx *ctx)
+{
+	struct image_converter *cvt = ctx->cvt;
+	struct ipu_ic_priv *priv = cvt->ic->priv;
+	unsigned long flags;
+	bool put_res;
+
+	/* make sure no runs are hanging around */
+	ipu_image_convert_abort(ctx);
+
+	dev_dbg(priv->ipu->dev, "%s: removing ctx %p\n", __func__, ctx);
+
+	spin_lock_irqsave(&cvt->irqlock, flags);
+
+	list_del(&ctx->list);
+
+	put_res = list_empty(&cvt->ctx_list);
+
+	spin_unlock_irqrestore(&cvt->irqlock, flags);
+
+	if (put_res)
+		ipu_ic_release_ipu_resources(cvt);
+
+	ipu_ic_free_dma_buf(priv, &ctx->rot_intermediate[1]);
+	ipu_ic_free_dma_buf(priv, &ctx->rot_intermediate[0]);
+
+	kfree(ctx);
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_unprepare);
+
+/*
+ * "Canned" asynchronous single image conversion. On successful return
+ * caller must call ipu_image_convert_unprepare() after conversion completes.
+ * Returns the new conversion context.
+ */
+struct image_converter_ctx *
+ipu_image_convert(struct ipu_ic *ic,
+		  struct ipu_image *in, struct ipu_image *out,
+		  enum ipu_rotate_mode rot_mode,
+		  image_converter_cb_t complete,
+		  void *complete_context)
+{
+	struct image_converter_ctx *ctx;
+	struct image_converter_run *run;
+
+	ctx = ipu_image_convert_prepare(ic, in, out, rot_mode,
+					complete, complete_context);
+	if (IS_ERR(ctx))
+		return ctx;
+
+	run = ipu_image_convert_run(ctx, in->phys0, out->phys0);
+	if (IS_ERR(run)) {
+		ipu_image_convert_unprepare(ctx);
+		return ERR_PTR(PTR_ERR(run));
+	}
+
+	return ctx;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert);
+
+/* "Canned" synchronous single image conversion */
+static void image_convert_sync_complete(void *data,
+					struct image_converter_run *run,
+					int err)
+{
+	struct completion *comp = data;
+
+	complete(comp);
+}
+
+int ipu_image_convert_sync(struct ipu_ic *ic,
+			   struct ipu_image *in, struct ipu_image *out,
+			   enum ipu_rotate_mode rot_mode)
+{
+	struct image_converter_ctx *ctx;
+	struct completion comp;
+	int ret;
+
+	init_completion(&comp);
+
+	ctx = ipu_image_convert(ic, in, out, rot_mode,
+				image_convert_sync_complete, &comp);
+	if (IS_ERR(ctx))
+		return PTR_ERR(ctx);
+
+	ret = wait_for_completion_timeout(&comp, msecs_to_jiffies(10000));
+	ret = (ret == 0) ? -ETIMEDOUT : 0;
+
+	ipu_image_convert_unprepare(ctx);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(ipu_image_convert_sync);
+
 int ipu_ic_enable(struct ipu_ic *ic)
 {
 	struct ipu_ic_priv *priv = ic->priv;
@@ -746,6 +2415,7 @@  int ipu_ic_init(struct ipu_soc *ipu, struct device *dev,
 	ipu->ic_priv = priv;
 
 	spin_lock_init(&priv->lock);
+
 	priv->base = devm_ioremap(dev, base, PAGE_SIZE);
 	if (!priv->base)
 		return -ENOMEM;
@@ -758,10 +2428,21 @@  int ipu_ic_init(struct ipu_soc *ipu, struct device *dev,
 	priv->ipu = ipu;
 
 	for (i = 0; i < IC_NUM_TASKS; i++) {
-		priv->task[i].task = i;
-		priv->task[i].priv = priv;
-		priv->task[i].reg = &ic_task_reg[i];
-		priv->task[i].bit = &ic_task_bit[i];
+		struct ipu_ic *ic = &priv->task[i];
+		struct image_converter *cvt = &ic->cvt;
+
+		ic->task = i;
+		ic->priv = priv;
+		ic->reg = &ic_task_reg[i];
+		ic->bit = &ic_task_bit[i];
+		ic->ch = &ic_task_ch[i];
+
+		cvt->ic = ic;
+		spin_lock_init(&cvt->irqlock);
+		INIT_LIST_HEAD(&cvt->ctx_list);
+		INIT_LIST_HEAD(&cvt->pending_q);
+		INIT_LIST_HEAD(&cvt->done_q);
+		cvt->out_eof_irq = cvt->rot_out_eof_irq = -1;
 	}
 
 	return 0;
diff --git a/include/video/imx-ipu-v3.h b/include/video/imx-ipu-v3.h
index afbc89d..25c40b3 100644
--- a/include/video/imx-ipu-v3.h
+++ b/include/video/imx-ipu-v3.h
@@ -63,17 +63,25 @@  enum ipu_csi_dest {
 /*
  * Enumeration of IPU rotation modes
  */
+#define IPU_ROT_BIT_VFLIP (1 << 0)
+#define IPU_ROT_BIT_HFLIP (1 << 1)
+#define IPU_ROT_BIT_90    (1 << 2)
+
 enum ipu_rotate_mode {
 	IPU_ROTATE_NONE = 0,
-	IPU_ROTATE_VERT_FLIP,
-	IPU_ROTATE_HORIZ_FLIP,
-	IPU_ROTATE_180,
-	IPU_ROTATE_90_RIGHT,
-	IPU_ROTATE_90_RIGHT_VFLIP,
-	IPU_ROTATE_90_RIGHT_HFLIP,
-	IPU_ROTATE_90_LEFT,
+	IPU_ROTATE_VERT_FLIP = IPU_ROT_BIT_VFLIP,
+	IPU_ROTATE_HORIZ_FLIP = IPU_ROT_BIT_HFLIP,
+	IPU_ROTATE_180 = (IPU_ROT_BIT_VFLIP | IPU_ROT_BIT_HFLIP),
+	IPU_ROTATE_90_RIGHT = IPU_ROT_BIT_90,
+	IPU_ROTATE_90_RIGHT_VFLIP = (IPU_ROT_BIT_90 | IPU_ROT_BIT_VFLIP),
+	IPU_ROTATE_90_RIGHT_HFLIP = (IPU_ROT_BIT_90 | IPU_ROT_BIT_HFLIP),
+	IPU_ROTATE_90_LEFT = (IPU_ROT_BIT_90 |
+			      IPU_ROT_BIT_VFLIP | IPU_ROT_BIT_HFLIP),
 };
 
+/* 90-degree rotations require the IRT unit */
+#define ipu_rot_mode_is_irt(m) ((m) >= IPU_ROTATE_90_RIGHT)
+
 enum ipu_color_space {
 	IPUV3_COLORSPACE_RGB,
 	IPUV3_COLORSPACE_YUV,
@@ -316,6 +324,7 @@  enum ipu_ic_task {
 };
 
 struct ipu_ic;
+
 int ipu_ic_task_init(struct ipu_ic *ic,
 		     int in_width, int in_height,
 		     int out_width, int out_height,
@@ -330,6 +339,40 @@  void ipu_ic_task_disable(struct ipu_ic *ic);
 int ipu_ic_task_idma_init(struct ipu_ic *ic, struct ipuv3_channel *channel,
 			  u32 width, u32 height, int burst_size,
 			  enum ipu_rotate_mode rot);
+
+struct image_converter_ctx;
+struct image_converter_run;
+
+typedef void (*image_converter_cb_t)(void *ctx,
+				     struct image_converter_run *run,
+				     int err);
+
+int ipu_image_convert_enum_format(int index, const char **desc, u32 *fourcc);
+int ipu_image_convert_adjust(struct ipu_image *in, struct ipu_image *out,
+			     enum ipu_rotate_mode rot_mode);
+int ipu_image_convert_verify(struct ipu_image *in, struct ipu_image *out,
+			     enum ipu_rotate_mode rot_mode);
+struct image_converter_ctx *
+ipu_image_convert_prepare(struct ipu_ic *ic,
+			  struct ipu_image *in, struct ipu_image *out,
+			  enum ipu_rotate_mode rot_mode,
+			  image_converter_cb_t complete,
+			  void *complete_context);
+void ipu_image_convert_unprepare(struct image_converter_ctx *ctx);
+struct image_converter_run *
+ipu_image_convert_run(struct image_converter_ctx *ctx,
+		      dma_addr_t in_phys, dma_addr_t out_phys);
+void ipu_image_convert_abort(struct image_converter_ctx *ctx);
+struct image_converter_ctx *
+ipu_image_convert(struct ipu_ic *ic,
+		  struct ipu_image *in, struct ipu_image *out,
+		  enum ipu_rotate_mode rot_mode,
+		  image_converter_cb_t complete,
+		  void *complete_context);
+int ipu_image_convert_sync(struct ipu_ic *ic,
+			   struct ipu_image *in, struct ipu_image *out,
+			   enum ipu_rotate_mode rot_mode);
+
 int ipu_ic_enable(struct ipu_ic *ic);
 int ipu_ic_disable(struct ipu_ic *ic);
 struct ipu_ic *ipu_ic_get(struct ipu_soc *ipu, enum ipu_ic_task task);