[PATCH v5 0/7] drm: Reuse temporary memory for format conversion

[PATCH v5 0/7] drm: Reuse temporary memory for format conversion

[Thomas Zimmermann]

DRM's format-conversion helpers require temporary memory. Pass the
buffer from the caller to allow the caller to preallocate the buffer
memory.

The motivation for this patchset is the recent work on a DRM panic
handler. [1] The panic handler requires format conversion to display an
error to the screen. But allocating memory during a kernel panic is
fragile. The changes in this patchset enable the DRM panic handler to
preallocate buffer storage before the panic occurs.

Patch 1 adds struct drm_format_conv_state, a simple interface to pass
around the buffer storage. Patch 2 adds an instance of the struct to
the shadow-plane state. Patch 3 moves the buffer's memory management
from the format helpers into their callers within the DRM drivers. Most
of the affected drivers use the state instance stored in their shadow-
plane state. The shadow-plane code releases the buffer memory automatically.

Patches 4 to 7 update three drivers to pre-allocate the format-conversion
buffer in their plane's atomic_check function. The drivers thus detect OOM
errors before the display update begins.

Tested with simpledrm.

v5:
	* fix repaper patch (Noralf)
	* fix documentation (Javier, Noralf, kernel test robot)
v4:
	* rename struct to drm_format_conv_state (Javier)
	* replace ARRAY_SIZE() with sizeof() (Jani)
	* store buffer in shadow-plane state (Javier, Maxime)
	* prealloc in atomic_check in several drivers
v3:
	* no changes
v2:
	* reserve storage during probing in the drivers

[1] https://patchwork.freedesktop.org/series/122244/

Thomas Zimmermann (7):
  drm/format-helper: Cache buffers with struct drm_format_conv_state
  drm/atomic-helper: Add format-conversion state to shadow-plane state
  drm/format-helper: Pass format-conversion state to helpers
  drm/ofdrm: Preallocate format-conversion buffer in atomic_check
  drm/simpledrm: Preallocate format-conversion buffer in atomic_check
  drm/ssd130x: Fix atomic_check for disabled planes
  drm/ssd130x: Preallocate format-conversion buffer in atomic_check

 drivers/gpu/drm/drm_format_helper.c           | 215 +++++++++++++-----
 drivers/gpu/drm/drm_gem_atomic_helper.c       |   9 +
 drivers/gpu/drm/drm_mipi_dbi.c                |  19 +-
 drivers/gpu/drm/gud/gud_pipe.c                |  30 ++-
 drivers/gpu/drm/solomon/ssd130x.c             |  36 ++-
 .../gpu/drm/tests/drm_format_helper_test.c    |  72 +++---
 drivers/gpu/drm/tiny/cirrus.c                 |   3 +-
 drivers/gpu/drm/tiny/ili9225.c                |  10 +-
 drivers/gpu/drm/tiny/ofdrm.c                  |  16 +-
 drivers/gpu/drm/tiny/repaper.c                |  10 +-
 drivers/gpu/drm/tiny/simpledrm.c              |  43 +++-
 drivers/gpu/drm/tiny/st7586.c                 |  19 +-
 include/drm/drm_format_helper.h               |  81 +++++--
 include/drm/drm_gem_atomic_helper.h           |  10 +
 include/drm/drm_mipi_dbi.h                    |   4 +-
 15 files changed, 433 insertions(+), 144 deletions(-)


base-commit: fe09b9f4c1435d9ce59b4729994b86a856db852f
prerequisite-patch-id: 0aa359f6144c4015c140c8a6750be19099c676fb
prerequisite-patch-id: c67e5d886a47b7d0266d81100837557fda34cb24
prerequisite-patch-id: cbc453ee02fae02af22fbfdce56ab732c7a88c36
-- 
2.42.0

[PATCH v5 1/7] drm/format-helper: Cache buffers with struct drm_format_conv_state

[Thomas Zimmermann]

Hold temporary memory for format conversion in an instance of struct
drm_format_conv_state. Update internal helpers of DRM's format-conversion
code accordingly. Drivers will later be able to maintain this cache by
themselves.

Besides caching, struct drm_format_conv_state will be useful to hold
additional information for format conversion, such as palette data or
foreground/background colors. This will enable conversion from indexed
color formats to component-based formats.

v5:
	* improve documentation (Javier, Noralf)
v3:
	* rename struct drm_xfrm_buf to struct drm_format_conv_state
	  (Javier)
	* remove managed cleanup
	* add drm_format_conv_state_copy() for shadow-plane support

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Acked-by: Noralf Trønnes <noralf@tronnes.org>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
---
 drivers/gpu/drm/drm_format_helper.c | 118 +++++++++++++++++++++++++---
 include/drm/drm_format_helper.h     |  51 ++++++++++++
 2 files changed, 158 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/drm_format_helper.c b/drivers/gpu/drm/drm_format_helper.c
index f93a4efcee909..28d013d427788 100644
--- a/drivers/gpu/drm/drm_format_helper.c
+++ b/drivers/gpu/drm/drm_format_helper.c
@@ -20,6 +20,97 @@
 #include <drm/drm_print.h>
 #include <drm/drm_rect.h>
 
+/**
+ * drm_format_conv_state_init - Initialize format-conversion state
+ * @state: The state to initialize
+ *
+ * Clears all fields in struct drm_format_conv_state. The state will
+ * be empty with no preallocated resources.
+ */
+void drm_format_conv_state_init(struct drm_format_conv_state *state)
+{
+	state->tmp.mem = NULL;
+	state->tmp.size = 0;
+	state->tmp.preallocated = false;
+}
+EXPORT_SYMBOL(drm_format_conv_state_init);
+
+/**
+ * drm_format_conv_state_copy - Copy format-conversion state
+ * @state: Destination state
+ * @old_state: Source state
+ *
+ * Copies format-conversion state from @old_state to @state; except for
+ * temporary storage.
+ */
+void drm_format_conv_state_copy(struct drm_format_conv_state *state,
+				const struct drm_format_conv_state *old_state)
+{
+	/*
+	 * So far, there's only temporary storage here, which we don't
+	 * duplicate. Just clear the fields.
+	 */
+	state->tmp.mem = NULL;
+	state->tmp.size = 0;
+	state->tmp.preallocated = false;
+}
+EXPORT_SYMBOL(drm_format_conv_state_copy);
+
+/**
+ * drm_format_conv_state_reserve - Allocates storage for format conversion
+ * @state: The format-conversion state
+ * @new_size: The minimum allocation size
+ * @flags: Flags for kmalloc()
+ *
+ * Allocates at least @new_size bytes and returns a pointer to the memory
+ * range. After calling this function, previously returned memory blocks
+ * are invalid. It's best to collect all memory requirements of a format
+ * conversion and call this function once to allocate the range.
+ *
+ * Returns:
+ * A pointer to the allocated memory range, or NULL otherwise.
+ */
+void *drm_format_conv_state_reserve(struct drm_format_conv_state *state,
+				    size_t new_size, gfp_t flags)
+{
+	void *mem;
+
+	if (new_size <= state->tmp.size)
+		goto out;
+	else if (state->tmp.preallocated)
+		return NULL;
+
+	mem = krealloc(state->tmp.mem, new_size, flags);
+	if (!mem)
+		return NULL;
+
+	state->tmp.mem = mem;
+	state->tmp.size = new_size;
+
+out:
+	return state->tmp.mem;
+}
+EXPORT_SYMBOL(drm_format_conv_state_reserve);
+
+/**
+ * drm_format_conv_state_release - Releases an format-conversion storage
+ * @state: The format-conversion state
+ *
+ * Releases the memory range references by the format-conversion state.
+ * After this call, all pointers to the memory are invalid. Prefer
+ * drm_format_conv_state_init() for cleaning up and unloading a driver.
+ */
+void drm_format_conv_state_release(struct drm_format_conv_state *state)
+{
+	if (state->tmp.preallocated)
+		return;
+
+	kfree(state->tmp.mem);
+	state->tmp.mem = NULL;
+	state->tmp.size = 0;
+}
+EXPORT_SYMBOL(drm_format_conv_state_release);
+
 static unsigned int clip_offset(const struct drm_rect *clip, unsigned int pitch, unsigned int cpp)
 {
 	return clip->y1 * pitch + clip->x1 * cpp;
@@ -45,6 +136,7 @@ EXPORT_SYMBOL(drm_fb_clip_offset);
 static int __drm_fb_xfrm(void *dst, unsigned long dst_pitch, unsigned long dst_pixsize,
 			 const void *vaddr, const struct drm_framebuffer *fb,
 			 const struct drm_rect *clip, bool vaddr_cached_hint,
+			 struct drm_format_conv_state *state,
 			 void (*xfrm_line)(void *dbuf, const void *sbuf, unsigned int npixels))
 {
 	unsigned long linepixels = drm_rect_width(clip);
@@ -60,7 +152,7 @@ static int __drm_fb_xfrm(void *dst, unsigned long dst_pitch, unsigned long dst_p
 	 * one line at a time.
 	 */
 	if (!vaddr_cached_hint) {
-		stmp = kmalloc(sbuf_len, GFP_KERNEL);
+		stmp = drm_format_conv_state_reserve(state, sbuf_len, GFP_KERNEL);
 		if (!stmp)
 			return -ENOMEM;
 	}
@@ -79,8 +171,6 @@ static int __drm_fb_xfrm(void *dst, unsigned long dst_pitch, unsigned long dst_p
 		dst += dst_pitch;
 	}
 
-	kfree(stmp);
-
 	return 0;
 }
 
@@ -88,6 +178,7 @@ static int __drm_fb_xfrm(void *dst, unsigned long dst_pitch, unsigned long dst_p
 static int __drm_fb_xfrm_toio(void __iomem *dst, unsigned long dst_pitch, unsigned long dst_pixsize,
 			      const void *vaddr, const struct drm_framebuffer *fb,
 			      const struct drm_rect *clip, bool vaddr_cached_hint,
+			      struct drm_format_conv_state *state,
 			      void (*xfrm_line)(void *dbuf, const void *sbuf, unsigned int npixels))
 {
 	unsigned long linepixels = drm_rect_width(clip);
@@ -101,9 +192,9 @@ static int __drm_fb_xfrm_toio(void __iomem *dst, unsigned long dst_pitch, unsign
 	void *dbuf;
 
 	if (vaddr_cached_hint) {
-		dbuf = kmalloc(dbuf_len, GFP_KERNEL);
+		dbuf = drm_format_conv_state_reserve(state, dbuf_len, GFP_KERNEL);
 	} else {
-		dbuf = kmalloc(stmp_off + sbuf_len, GFP_KERNEL);
+		dbuf = drm_format_conv_state_reserve(state, stmp_off + sbuf_len, GFP_KERNEL);
 		stmp = dbuf + stmp_off;
 	}
 	if (!dbuf)
@@ -124,8 +215,6 @@ static int __drm_fb_xfrm_toio(void __iomem *dst, unsigned long dst_pitch, unsign
 		dst += dst_pitch;
 	}
 
-	kfree(dbuf);
-
 	return 0;
 }
 
@@ -139,17 +228,24 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 	static const unsigned int default_dst_pitch[DRM_FORMAT_MAX_PLANES] = {
 		0, 0, 0, 0
 	};
+	struct drm_format_conv_state fmtcnv_state = DRM_FORMAT_CONV_STATE_INIT;
+	int ret;
 
 	if (!dst_pitch)
 		dst_pitch = default_dst_pitch;
 
 	/* TODO: handle src in I/O memory here */
 	if (dst[0].is_iomem)
-		return __drm_fb_xfrm_toio(dst[0].vaddr_iomem, dst_pitch[0], dst_pixsize[0],
-					  src[0].vaddr, fb, clip, vaddr_cached_hint, xfrm_line);
+		ret = __drm_fb_xfrm_toio(dst[0].vaddr_iomem, dst_pitch[0], dst_pixsize[0],
+					 src[0].vaddr, fb, clip, vaddr_cached_hint, &fmtcnv_state,
+					 xfrm_line);
 	else
-		return __drm_fb_xfrm(dst[0].vaddr, dst_pitch[0], dst_pixsize[0],
-				     src[0].vaddr, fb, clip, vaddr_cached_hint, xfrm_line);
+		ret = __drm_fb_xfrm(dst[0].vaddr, dst_pitch[0], dst_pixsize[0],
+				    src[0].vaddr, fb, clip, vaddr_cached_hint, &fmtcnv_state,
+				    xfrm_line);
+	drm_format_conv_state_release(&fmtcnv_state);
+
+	return ret;
 }
 
 /**
diff --git a/include/drm/drm_format_helper.h b/include/drm/drm_format_helper.h
index 291deb09475bb..724a9baf7315d 100644
--- a/include/drm/drm_format_helper.h
+++ b/include/drm/drm_format_helper.h
@@ -15,6 +15,57 @@ struct drm_rect;
 
 struct iosys_map;
 
+/**
+ * struct drm_format_conv_state - Stores format-conversion state
+ *
+ * DRM helpers for format conversion store temporary state in
+ * struct drm_xfrm_buf. The buffer's resources can be reused
+ * among multiple conversion operations.
+ *
+ * All fields are considered private.
+ */
+struct drm_format_conv_state {
+	struct {
+		void *mem;
+		size_t size;
+		bool preallocated;
+	} tmp;
+};
+
+#define __DRM_FORMAT_CONV_STATE_INIT(_mem, _size, _preallocated) { \
+		.tmp = { \
+			.mem = (_mem), \
+			.size = (_size), \
+			.preallocated = (_preallocated), \
+		} \
+	}
+
+/**
+ * DRM_FORMAT_CONV_STATE_INIT - Initializer for struct drm_format_conv_state
+ *
+ * Initializes an instance of struct drm_format_conv_state to default values.
+ */
+#define DRM_FORMAT_CONV_STATE_INIT \
+	__DRM_FORMAT_CONV_STATE_INIT(NULL, 0, false)
+
+/**
+ * DRM_FORMAT_CONV_STATE_INIT_PREALLOCATED - Initializer for struct drm_format_conv_state
+ * @_mem: The preallocated memory area
+ * @_size: The number of bytes in _mem
+ *
+ * Initializes an instance of struct drm_format_conv_state to preallocated
+ * storage. The caller is responsible for releasing the provided memory range.
+ */
+#define DRM_FORMAT_CONV_STATE_INIT_PREALLOCATED(_mem, _size) \
+	__DRM_FORMAT_CONV_STATE_INIT(_mem, _size, true)
+
+void drm_format_conv_state_init(struct drm_format_conv_state *state);
+void drm_format_conv_state_copy(struct drm_format_conv_state *state,
+				const struct drm_format_conv_state *old_state);
+void *drm_format_conv_state_reserve(struct drm_format_conv_state *state,
+				    size_t new_size, gfp_t flags);
+void drm_format_conv_state_release(struct drm_format_conv_state *state);
+
 unsigned int drm_fb_clip_offset(unsigned int pitch, const struct drm_format_info *format,
 				const struct drm_rect *clip);
 
-- 
2.42.0

[PATCH v5 2/7] drm/atomic-helper: Add format-conversion state to shadow-plane state

[Thomas Zimmermann]

Store an instance of struct drm_format_conv_state in the shadow-plane
state struct drm_shadow_plane_state. Many drivers with shadow planes
use DRM's format helpers to copy or convert the framebuffer data to
backing storage in the scanout buffer. The shadow plane provides the
necessary state and manages the conversion's intermediate buffer memory.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Acked-by: Noralf Trønnes <noralf@tronnes.org>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
---
 drivers/gpu/drm/drm_gem_atomic_helper.c |  9 +++++++++
 include/drm/drm_gem_atomic_helper.h     | 10 ++++++++++
 2 files changed, 19 insertions(+)

diff --git a/drivers/gpu/drm/drm_gem_atomic_helper.c b/drivers/gpu/drm/drm_gem_atomic_helper.c
index 5d4b9cd077f7a..e440f458b6633 100644
--- a/drivers/gpu/drm/drm_gem_atomic_helper.c
+++ b/drivers/gpu/drm/drm_gem_atomic_helper.c
@@ -218,7 +218,14 @@ void
 __drm_gem_duplicate_shadow_plane_state(struct drm_plane *plane,
 				       struct drm_shadow_plane_state *new_shadow_plane_state)
 {
+	struct drm_plane_state *plane_state = plane->state;
+	struct drm_shadow_plane_state *shadow_plane_state =
+		to_drm_shadow_plane_state(plane_state);
+
 	__drm_atomic_helper_plane_duplicate_state(plane, &new_shadow_plane_state->base);
+
+	drm_format_conv_state_copy(&shadow_plane_state->fmtcnv_state,
+				   &new_shadow_plane_state->fmtcnv_state);
 }
 EXPORT_SYMBOL(__drm_gem_duplicate_shadow_plane_state);
 
@@ -266,6 +273,7 @@ EXPORT_SYMBOL(drm_gem_duplicate_shadow_plane_state);
  */
 void __drm_gem_destroy_shadow_plane_state(struct drm_shadow_plane_state *shadow_plane_state)
 {
+	drm_format_conv_state_release(&shadow_plane_state->fmtcnv_state);
 	__drm_atomic_helper_plane_destroy_state(&shadow_plane_state->base);
 }
 EXPORT_SYMBOL(__drm_gem_destroy_shadow_plane_state);
@@ -302,6 +310,7 @@ void __drm_gem_reset_shadow_plane(struct drm_plane *plane,
 				  struct drm_shadow_plane_state *shadow_plane_state)
 {
 	__drm_atomic_helper_plane_reset(plane, &shadow_plane_state->base);
+	drm_format_conv_state_init(&shadow_plane_state->fmtcnv_state);
 }
 EXPORT_SYMBOL(__drm_gem_reset_shadow_plane);
 
diff --git a/include/drm/drm_gem_atomic_helper.h b/include/drm/drm_gem_atomic_helper.h
index 40b8b039518e0..3e01c619a25e0 100644
--- a/include/drm/drm_gem_atomic_helper.h
+++ b/include/drm/drm_gem_atomic_helper.h
@@ -5,6 +5,7 @@
 
 #include <linux/iosys-map.h>
 
+#include <drm/drm_format_helper.h>
 #include <drm/drm_fourcc.h>
 #include <drm/drm_plane.h>
 
@@ -49,6 +50,15 @@ struct drm_shadow_plane_state {
 	/** @base: plane state */
 	struct drm_plane_state base;
 
+	/**
+	 * @fmtcnv_state: Format-conversion state
+	 *
+	 * Per-plane state for format conversion.
+	 * Flags for copying shadow buffers into backend storage. Also holds
+	 * temporary storage for format conversion.
+	 */
+	struct drm_format_conv_state fmtcnv_state;
+
 	/* Transitional state - do not export or duplicate */
 
 	/**
-- 
2.42.0

[PATCH v5 3/7] drm/format-helper: Pass format-conversion state to helpers

[Thomas Zimmermann]

Pass an instance of struct drm_format_conv_state to DRM's format
conversion helpers. Update all callers.

Most drivers can use the format-conversion state from their shadow-
plane state. The shadow plane's destroy function releases the
allocated buffer. Drivers will later be able to allocate a buffer
of appropriate size in their plane's atomic_check code.

The gud driver uses a separate thread for committing updates. For
now, the update worker contains its own format-conversion state.

Images in the format-helper tests are small. The tests preallocate
a static page for the temporary buffer. Unloading the module releases
the memory.

v5:
	* avoid using unusupported shadow-plane state in repaper (Noralf)
	* fix documentation (Noralf, kernel test robot)
v3:
	* store buffer in shadow-plane state (Javier, Maxime)
	* replace ARRAY_SIZE() with sizeof() (Jani)

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Javier Martinez Canillas <javierm@redhat.com> # ssd130x
Cc: Noralf Trønnes <noralf@tronnes.org>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: David Lechner <david@lechnology.com>
---
 drivers/gpu/drm/drm_format_helper.c           | 123 ++++++++++--------
 drivers/gpu/drm/drm_mipi_dbi.c                |  19 ++-
 drivers/gpu/drm/gud/gud_pipe.c                |  30 +++--
 drivers/gpu/drm/solomon/ssd130x.c             |   8 +-
 .../gpu/drm/tests/drm_format_helper_test.c    |  72 ++++++----
 drivers/gpu/drm/tiny/cirrus.c                 |   3 +-
 drivers/gpu/drm/tiny/ili9225.c                |  10 +-
 drivers/gpu/drm/tiny/ofdrm.c                  |   2 +-
 drivers/gpu/drm/tiny/repaper.c                |  10 +-
 drivers/gpu/drm/tiny/simpledrm.c              |   2 +-
 drivers/gpu/drm/tiny/st7586.c                 |  19 +--
 include/drm/drm_format_helper.h               |  30 +++--
 include/drm/drm_mipi_dbi.h                    |   4 +-
 13 files changed, 195 insertions(+), 137 deletions(-)

diff --git a/drivers/gpu/drm/drm_format_helper.c b/drivers/gpu/drm/drm_format_helper.c
index 28d013d427788..b1be458ed4dda 100644
--- a/drivers/gpu/drm/drm_format_helper.c
+++ b/drivers/gpu/drm/drm_format_helper.c
@@ -223,29 +223,25 @@ static int drm_fb_xfrm(struct iosys_map *dst,
 		       const unsigned int *dst_pitch, const u8 *dst_pixsize,
 		       const struct iosys_map *src, const struct drm_framebuffer *fb,
 		       const struct drm_rect *clip, bool vaddr_cached_hint,
+		       struct drm_format_conv_state *state,
 		       void (*xfrm_line)(void *dbuf, const void *sbuf, unsigned int npixels))
 {
 	static const unsigned int default_dst_pitch[DRM_FORMAT_MAX_PLANES] = {
 		0, 0, 0, 0
 	};
-	struct drm_format_conv_state fmtcnv_state = DRM_FORMAT_CONV_STATE_INIT;
-	int ret;
 
 	if (!dst_pitch)
 		dst_pitch = default_dst_pitch;
 
 	/* TODO: handle src in I/O memory here */
 	if (dst[0].is_iomem)
-		ret = __drm_fb_xfrm_toio(dst[0].vaddr_iomem, dst_pitch[0], dst_pixsize[0],
-					 src[0].vaddr, fb, clip, vaddr_cached_hint, &fmtcnv_state,
-					 xfrm_line);
+		return __drm_fb_xfrm_toio(dst[0].vaddr_iomem, dst_pitch[0], dst_pixsize[0],
+					  src[0].vaddr, fb, clip, vaddr_cached_hint, state,
+					  xfrm_line);
 	else
-		ret = __drm_fb_xfrm(dst[0].vaddr, dst_pitch[0], dst_pixsize[0],
-				    src[0].vaddr, fb, clip, vaddr_cached_hint, &fmtcnv_state,
-				    xfrm_line);
-	drm_format_conv_state_release(&fmtcnv_state);
-
-	return ret;
+		return __drm_fb_xfrm(dst[0].vaddr, dst_pitch[0], dst_pixsize[0],
+				     src[0].vaddr, fb, clip, vaddr_cached_hint, state,
+				     xfrm_line);
 }
 
 /**
@@ -331,6 +327,7 @@ static void drm_fb_swab32_line(void *dbuf, const void *sbuf, unsigned int pixels
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
  * @cached: Source buffer is mapped cached (eg. not write-combined)
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and swaps per-pixel
  * bytes during the process. Destination and framebuffer formats must match. The
@@ -345,7 +342,8 @@ static void drm_fb_swab32_line(void *dbuf, const void *sbuf, unsigned int pixels
  */
 void drm_fb_swab(struct iosys_map *dst, const unsigned int *dst_pitch,
 		 const struct iosys_map *src, const struct drm_framebuffer *fb,
-		 const struct drm_rect *clip, bool cached)
+		 const struct drm_rect *clip, bool cached,
+		 struct drm_format_conv_state *state)
 {
 	const struct drm_format_info *format = fb->format;
 	u8 cpp = DIV_ROUND_UP(drm_format_info_bpp(format, 0), 8);
@@ -364,7 +362,7 @@ void drm_fb_swab(struct iosys_map *dst, const unsigned int *dst_pitch,
 		return;
 	}
 
-	drm_fb_xfrm(dst, dst_pitch, &cpp, src, fb, clip, cached, swab_line);
+	drm_fb_xfrm(dst, dst_pitch, &cpp, src, fb, clip, cached, state, swab_line);
 }
 EXPORT_SYMBOL(drm_fb_swab);
 
@@ -391,6 +389,7 @@ static void drm_fb_xrgb8888_to_rgb332_line(void *dbuf, const void *sbuf, unsigne
  * @src: Array of XRGB8888 source buffers
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts the
  * color format during the process. Destination and framebuffer formats must match. The
@@ -405,13 +404,13 @@ static void drm_fb_xrgb8888_to_rgb332_line(void *dbuf, const void *sbuf, unsigne
  */
 void drm_fb_xrgb8888_to_rgb332(struct iosys_map *dst, const unsigned int *dst_pitch,
 			       const struct iosys_map *src, const struct drm_framebuffer *fb,
-			       const struct drm_rect *clip)
+			       const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		1,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_rgb332_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgb332);
@@ -460,6 +459,7 @@ static void drm_fb_xrgb8888_to_rgb565_swab_line(void *dbuf, const void *sbuf,
  * @src: Array of XRGB8888 source buffer
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  * @swab: Swap bytes
  *
  * This function copies parts of a framebuffer to display memory and converts the
@@ -475,7 +475,8 @@ static void drm_fb_xrgb8888_to_rgb565_swab_line(void *dbuf, const void *sbuf,
  */
 void drm_fb_xrgb8888_to_rgb565(struct iosys_map *dst, const unsigned int *dst_pitch,
 			       const struct iosys_map *src, const struct drm_framebuffer *fb,
-			       const struct drm_rect *clip, bool swab)
+			       const struct drm_rect *clip, struct drm_format_conv_state *state,
+			       bool swab)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		2,
@@ -488,7 +489,7 @@ void drm_fb_xrgb8888_to_rgb565(struct iosys_map *dst, const unsigned int *dst_pi
 	else
 		xfrm_line = drm_fb_xrgb8888_to_rgb565_line;
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, xfrm_line);
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state, xfrm_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgb565);
 
@@ -517,6 +518,7 @@ static void drm_fb_xrgb8888_to_xrgb1555_line(void *dbuf, const void *sbuf, unsig
  * @src: Array of XRGB8888 source buffer
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts
  * the color format during the process. The parameters @dst, @dst_pitch and
@@ -532,13 +534,13 @@ static void drm_fb_xrgb8888_to_xrgb1555_line(void *dbuf, const void *sbuf, unsig
  */
 void drm_fb_xrgb8888_to_xrgb1555(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip)
+				 const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		2,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_xrgb1555_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_xrgb1555);
@@ -569,6 +571,7 @@ static void drm_fb_xrgb8888_to_argb1555_line(void *dbuf, const void *sbuf, unsig
  * @src: Array of XRGB8888 source buffer
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts
  * the color format during the process. The parameters @dst, @dst_pitch and
@@ -584,13 +587,13 @@ static void drm_fb_xrgb8888_to_argb1555_line(void *dbuf, const void *sbuf, unsig
  */
 void drm_fb_xrgb8888_to_argb1555(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip)
+				 const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		2,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_argb1555_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_argb1555);
@@ -621,6 +624,7 @@ static void drm_fb_xrgb8888_to_rgba5551_line(void *dbuf, const void *sbuf, unsig
  * @src: Array of XRGB8888 source buffer
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts
  * the color format during the process. The parameters @dst, @dst_pitch and
@@ -636,13 +640,13 @@ static void drm_fb_xrgb8888_to_rgba5551_line(void *dbuf, const void *sbuf, unsig
  */
 void drm_fb_xrgb8888_to_rgba5551(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip)
+				 const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		2,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_rgba5551_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgba5551);
@@ -671,6 +675,7 @@ static void drm_fb_xrgb8888_to_rgb888_line(void *dbuf, const void *sbuf, unsigne
  * @src: Array of XRGB8888 source buffers
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts the
  * color format during the process. Destination and framebuffer formats must match. The
@@ -686,13 +691,13 @@ static void drm_fb_xrgb8888_to_rgb888_line(void *dbuf, const void *sbuf, unsigne
  */
 void drm_fb_xrgb8888_to_rgb888(struct iosys_map *dst, const unsigned int *dst_pitch,
 			       const struct iosys_map *src, const struct drm_framebuffer *fb,
-			       const struct drm_rect *clip)
+			       const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		3,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_rgb888_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_rgb888);
@@ -719,6 +724,7 @@ static void drm_fb_xrgb8888_to_argb8888_line(void *dbuf, const void *sbuf, unsig
  * @src: Array of XRGB8888 source buffer
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts the
  * color format during the process. The parameters @dst, @dst_pitch and @src refer
@@ -734,13 +740,13 @@ static void drm_fb_xrgb8888_to_argb8888_line(void *dbuf, const void *sbuf, unsig
  */
 void drm_fb_xrgb8888_to_argb8888(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip)
+				 const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		4,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_argb8888_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_argb8888);
@@ -765,13 +771,14 @@ static void drm_fb_xrgb8888_to_abgr8888_line(void *dbuf, const void *sbuf, unsig
 static void drm_fb_xrgb8888_to_abgr8888(struct iosys_map *dst, const unsigned int *dst_pitch,
 					const struct iosys_map *src,
 					const struct drm_framebuffer *fb,
-					const struct drm_rect *clip)
+					const struct drm_rect *clip,
+					struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		4,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_abgr8888_line);
 }
 
@@ -795,13 +802,14 @@ static void drm_fb_xrgb8888_to_xbgr8888_line(void *dbuf, const void *sbuf, unsig
 static void drm_fb_xrgb8888_to_xbgr8888(struct iosys_map *dst, const unsigned int *dst_pitch,
 					const struct iosys_map *src,
 					const struct drm_framebuffer *fb,
-					const struct drm_rect *clip)
+					const struct drm_rect *clip,
+					struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		4,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_xbgr8888_line);
 }
 
@@ -831,6 +839,7 @@ static void drm_fb_xrgb8888_to_xrgb2101010_line(void *dbuf, const void *sbuf, un
  * @src: Array of XRGB8888 source buffers
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts the
  * color format during the process. Destination and framebuffer formats must match. The
@@ -846,13 +855,14 @@ static void drm_fb_xrgb8888_to_xrgb2101010_line(void *dbuf, const void *sbuf, un
  */
 void drm_fb_xrgb8888_to_xrgb2101010(struct iosys_map *dst, const unsigned int *dst_pitch,
 				    const struct iosys_map *src, const struct drm_framebuffer *fb,
-				    const struct drm_rect *clip)
+				    const struct drm_rect *clip,
+				    struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		4,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_xrgb2101010_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_xrgb2101010);
@@ -884,6 +894,7 @@ static void drm_fb_xrgb8888_to_argb2101010_line(void *dbuf, const void *sbuf, un
  * @src: Array of XRGB8888 source buffers
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts
  * the color format during the process. The parameters @dst, @dst_pitch and
@@ -899,13 +910,14 @@ static void drm_fb_xrgb8888_to_argb2101010_line(void *dbuf, const void *sbuf, un
  */
 void drm_fb_xrgb8888_to_argb2101010(struct iosys_map *dst, const unsigned int *dst_pitch,
 				    const struct iosys_map *src, const struct drm_framebuffer *fb,
-				    const struct drm_rect *clip)
+				    const struct drm_rect *clip,
+				    struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		4,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_argb2101010_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_argb2101010);
@@ -935,6 +947,7 @@ static void drm_fb_xrgb8888_to_gray8_line(void *dbuf, const void *sbuf, unsigned
  * @src: Array of XRGB8888 source buffers
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts the
  * color format during the process. Destination and framebuffer formats must match. The
@@ -954,13 +967,13 @@ static void drm_fb_xrgb8888_to_gray8_line(void *dbuf, const void *sbuf, unsigned
  */
 void drm_fb_xrgb8888_to_gray8(struct iosys_map *dst, const unsigned int *dst_pitch,
 			      const struct iosys_map *src, const struct drm_framebuffer *fb,
-			      const struct drm_rect *clip)
+			      const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const u8 dst_pixsize[DRM_FORMAT_MAX_PLANES] = {
 		1,
 	};
 
-	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false,
+	drm_fb_xfrm(dst, dst_pitch, dst_pixsize, src, fb, clip, false, state,
 		    drm_fb_xrgb8888_to_gray8_line);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_gray8);
@@ -974,6 +987,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_gray8);
  * @src:	The framebuffer memory to copy from
  * @fb:		The framebuffer to copy from
  * @clip:	Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory. If the
  * formats of the display and the framebuffer mismatch, the blit function
@@ -992,7 +1006,7 @@ EXPORT_SYMBOL(drm_fb_xrgb8888_to_gray8);
  */
 int drm_fb_blit(struct iosys_map *dst, const unsigned int *dst_pitch, uint32_t dst_format,
 		const struct iosys_map *src, const struct drm_framebuffer *fb,
-		const struct drm_rect *clip)
+		const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	uint32_t fb_format = fb->format->format;
 
@@ -1000,44 +1014,44 @@ int drm_fb_blit(struct iosys_map *dst, const unsigned int *dst_pitch, uint32_t d
 		drm_fb_memcpy(dst, dst_pitch, src, fb, clip);
 		return 0;
 	} else if (fb_format == (dst_format | DRM_FORMAT_BIG_ENDIAN)) {
-		drm_fb_swab(dst, dst_pitch, src, fb, clip, false);
+		drm_fb_swab(dst, dst_pitch, src, fb, clip, false, state);
 		return 0;
 	} else if (fb_format == (dst_format & ~DRM_FORMAT_BIG_ENDIAN)) {
-		drm_fb_swab(dst, dst_pitch, src, fb, clip, false);
+		drm_fb_swab(dst, dst_pitch, src, fb, clip, false, state);
 		return 0;
 	} else if (fb_format == DRM_FORMAT_XRGB8888) {
 		if (dst_format == DRM_FORMAT_RGB565) {
-			drm_fb_xrgb8888_to_rgb565(dst, dst_pitch, src, fb, clip, false);
+			drm_fb_xrgb8888_to_rgb565(dst, dst_pitch, src, fb, clip, state, false);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_XRGB1555) {
-			drm_fb_xrgb8888_to_xrgb1555(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_xrgb1555(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_ARGB1555) {
-			drm_fb_xrgb8888_to_argb1555(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_argb1555(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_RGBA5551) {
-			drm_fb_xrgb8888_to_rgba5551(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_rgba5551(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_RGB888) {
-			drm_fb_xrgb8888_to_rgb888(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_rgb888(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_ARGB8888) {
-			drm_fb_xrgb8888_to_argb8888(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_argb8888(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_XBGR8888) {
-			drm_fb_xrgb8888_to_xbgr8888(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_xbgr8888(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_ABGR8888) {
-			drm_fb_xrgb8888_to_abgr8888(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_abgr8888(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_XRGB2101010) {
-			drm_fb_xrgb8888_to_xrgb2101010(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_xrgb2101010(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_ARGB2101010) {
-			drm_fb_xrgb8888_to_argb2101010(dst, dst_pitch, src, fb, clip);
+			drm_fb_xrgb8888_to_argb2101010(dst, dst_pitch, src, fb, clip, state);
 			return 0;
 		} else if (dst_format == DRM_FORMAT_BGRX8888) {
-			drm_fb_swab(dst, dst_pitch, src, fb, clip, false);
+			drm_fb_swab(dst, dst_pitch, src, fb, clip, false, state);
 			return 0;
 		}
 	}
@@ -1074,6 +1088,7 @@ static void drm_fb_gray8_to_mono_line(void *dbuf, const void *sbuf, unsigned int
  * @src: Array of XRGB8888 source buffers
  * @fb: DRM framebuffer
  * @clip: Clip rectangle area to copy
+ * @state: Transform and conversion state
  *
  * This function copies parts of a framebuffer to display memory and converts the
  * color format during the process. Destination and framebuffer formats must match. The
@@ -1098,7 +1113,7 @@ static void drm_fb_gray8_to_mono_line(void *dbuf, const void *sbuf, unsigned int
  */
 void drm_fb_xrgb8888_to_mono(struct iosys_map *dst, const unsigned int *dst_pitch,
 			     const struct iosys_map *src, const struct drm_framebuffer *fb,
-			     const struct drm_rect *clip)
+			     const struct drm_rect *clip, struct drm_format_conv_state *state)
 {
 	static const unsigned int default_dst_pitch[DRM_FORMAT_MAX_PLANES] = {
 		0, 0, 0, 0
@@ -1138,7 +1153,7 @@ void drm_fb_xrgb8888_to_mono(struct iosys_map *dst, const unsigned int *dst_pitc
 	 * Allocate a buffer to be used for both copying from the cma
 	 * memory and to store the intermediate grayscale line pixels.
 	 */
-	src32 = kmalloc(len_src32 + linepixels, GFP_KERNEL);
+	src32 = drm_format_conv_state_reserve(state, len_src32 + linepixels, GFP_KERNEL);
 	if (!src32)
 		return;
 
@@ -1152,8 +1167,6 @@ void drm_fb_xrgb8888_to_mono(struct iosys_map *dst, const unsigned int *dst_pitc
 		vaddr += fb->pitches[0];
 		mono += dst_pitch_0;
 	}
-
-	kfree(src32);
 }
 EXPORT_SYMBOL(drm_fb_xrgb8888_to_mono);
 
diff --git a/drivers/gpu/drm/drm_mipi_dbi.c b/drivers/gpu/drm/drm_mipi_dbi.c
index e90f0bf895b33..daac649aabdbe 100644
--- a/drivers/gpu/drm/drm_mipi_dbi.c
+++ b/drivers/gpu/drm/drm_mipi_dbi.c
@@ -197,12 +197,14 @@ EXPORT_SYMBOL(mipi_dbi_command_stackbuf);
  * @fb: The source framebuffer
  * @clip: Clipping rectangle of the area to be copied
  * @swap: When true, swap MSB/LSB of 16-bit values
+ * @fmtcnv_state: Format-conversion state
  *
  * Returns:
  * Zero on success, negative error code on failure.
  */
 int mipi_dbi_buf_copy(void *dst, struct iosys_map *src, struct drm_framebuffer *fb,
-		      struct drm_rect *clip, bool swap)
+		      struct drm_rect *clip, bool swap,
+		      struct drm_format_conv_state *fmtcnv_state)
 {
 	struct drm_gem_object *gem = drm_gem_fb_get_obj(fb, 0);
 	struct iosys_map dst_map = IOSYS_MAP_INIT_VADDR(dst);
@@ -215,12 +217,13 @@ int mipi_dbi_buf_copy(void *dst, struct iosys_map *src, struct drm_framebuffer *
 	switch (fb->format->format) {
 	case DRM_FORMAT_RGB565:
 		if (swap)
-			drm_fb_swab(&dst_map, NULL, src, fb, clip, !gem->import_attach);
+			drm_fb_swab(&dst_map, NULL, src, fb, clip, !gem->import_attach,
+				    fmtcnv_state);
 		else
 			drm_fb_memcpy(&dst_map, NULL, src, fb, clip);
 		break;
 	case DRM_FORMAT_XRGB8888:
-		drm_fb_xrgb8888_to_rgb565(&dst_map, NULL, src, fb, clip, swap);
+		drm_fb_xrgb8888_to_rgb565(&dst_map, NULL, src, fb, clip, fmtcnv_state, swap);
 		break;
 	default:
 		drm_err_once(fb->dev, "Format is not supported: %p4cc\n",
@@ -252,7 +255,7 @@ static void mipi_dbi_set_window_address(struct mipi_dbi_dev *dbidev,
 }
 
 static void mipi_dbi_fb_dirty(struct iosys_map *src, struct drm_framebuffer *fb,
-			      struct drm_rect *rect)
+			      struct drm_rect *rect, struct drm_format_conv_state *fmtcnv_state)
 {
 	struct mipi_dbi_dev *dbidev = drm_to_mipi_dbi_dev(fb->dev);
 	unsigned int height = rect->y2 - rect->y1;
@@ -270,7 +273,7 @@ static void mipi_dbi_fb_dirty(struct iosys_map *src, struct drm_framebuffer *fb,
 	if (!dbi->dc || !full || swap ||
 	    fb->format->format == DRM_FORMAT_XRGB8888) {
 		tr = dbidev->tx_buf;
-		ret = mipi_dbi_buf_copy(tr, src, fb, rect, swap);
+		ret = mipi_dbi_buf_copy(tr, src, fb, rect, swap, fmtcnv_state);
 		if (ret)
 			goto err_msg;
 	} else {
@@ -332,7 +335,8 @@ void mipi_dbi_pipe_update(struct drm_simple_display_pipe *pipe,
 		return;
 
 	if (drm_atomic_helper_damage_merged(old_state, state, &rect))
-		mipi_dbi_fb_dirty(&shadow_plane_state->data[0], fb, &rect);
+		mipi_dbi_fb_dirty(&shadow_plane_state->data[0], fb, &rect,
+				  &shadow_plane_state->fmtcnv_state);
 
 	drm_dev_exit(idx);
 }
@@ -368,7 +372,8 @@ void mipi_dbi_enable_flush(struct mipi_dbi_dev *dbidev,
 	if (!drm_dev_enter(&dbidev->drm, &idx))
 		return;
 
-	mipi_dbi_fb_dirty(&shadow_plane_state->data[0], fb, &rect);
+	mipi_dbi_fb_dirty(&shadow_plane_state->data[0], fb, &rect,
+			  &shadow_plane_state->fmtcnv_state);
 	backlight_enable(dbidev->backlight);
 
 	drm_dev_exit(idx);
diff --git a/drivers/gpu/drm/gud/gud_pipe.c b/drivers/gpu/drm/gud/gud_pipe.c
index d2f199ea3c111..28171de84718e 100644
--- a/drivers/gpu/drm/gud/gud_pipe.c
+++ b/drivers/gpu/drm/gud/gud_pipe.c
@@ -51,7 +51,8 @@ static bool gud_is_big_endian(void)
 
 static size_t gud_xrgb8888_to_r124(u8 *dst, const struct drm_format_info *format,
 				   void *src, struct drm_framebuffer *fb,
-				   struct drm_rect *rect)
+				   struct drm_rect *rect,
+				   struct drm_format_conv_state *fmtcnv_state)
 {
 	unsigned int block_width = drm_format_info_block_width(format, 0);
 	unsigned int bits_per_pixel = 8 / block_width;
@@ -75,7 +76,7 @@ static size_t gud_xrgb8888_to_r124(u8 *dst, const struct drm_format_info *format
 
 	iosys_map_set_vaddr(&dst_map, buf);
 	iosys_map_set_vaddr(&vmap, src);
-	drm_fb_xrgb8888_to_gray8(&dst_map, NULL, &vmap, fb, rect);
+	drm_fb_xrgb8888_to_gray8(&dst_map, NULL, &vmap, fb, rect, fmtcnv_state);
 	pix8 = buf;
 
 	for (y = 0; y < height; y++) {
@@ -152,7 +153,8 @@ static size_t gud_xrgb8888_to_color(u8 *dst, const struct drm_format_info *forma
 static int gud_prep_flush(struct gud_device *gdrm, struct drm_framebuffer *fb,
 			  const struct iosys_map *src, bool cached_reads,
 			  const struct drm_format_info *format, struct drm_rect *rect,
-			  struct gud_set_buffer_req *req)
+			  struct gud_set_buffer_req *req,
+			  struct drm_format_conv_state *fmtcnv_state)
 {
 	u8 compression = gdrm->compression;
 	struct iosys_map dst;
@@ -178,23 +180,23 @@ static int gud_prep_flush(struct gud_device *gdrm, struct drm_framebuffer *fb,
 	 */
 	if (format != fb->format) {
 		if (format->format == GUD_DRM_FORMAT_R1) {
-			len = gud_xrgb8888_to_r124(buf, format, vaddr, fb, rect);
+			len = gud_xrgb8888_to_r124(buf, format, vaddr, fb, rect, fmtcnv_state);
 			if (!len)
 				return -ENOMEM;
 		} else if (format->format == DRM_FORMAT_R8) {
-			drm_fb_xrgb8888_to_gray8(&dst, NULL, src, fb, rect);
+			drm_fb_xrgb8888_to_gray8(&dst, NULL, src, fb, rect, fmtcnv_state);
 		} else if (format->format == DRM_FORMAT_RGB332) {
-			drm_fb_xrgb8888_to_rgb332(&dst, NULL, src, fb, rect);
+			drm_fb_xrgb8888_to_rgb332(&dst, NULL, src, fb, rect, fmtcnv_state);
 		} else if (format->format == DRM_FORMAT_RGB565) {
-			drm_fb_xrgb8888_to_rgb565(&dst, NULL, src, fb, rect,
+			drm_fb_xrgb8888_to_rgb565(&dst, NULL, src, fb, rect, fmtcnv_state,
 						  gud_is_big_endian());
 		} else if (format->format == DRM_FORMAT_RGB888) {
-			drm_fb_xrgb8888_to_rgb888(&dst, NULL, src, fb, rect);
+			drm_fb_xrgb8888_to_rgb888(&dst, NULL, src, fb, rect, fmtcnv_state);
 		} else {
 			len = gud_xrgb8888_to_color(buf, format, vaddr, fb, rect);
 		}
 	} else if (gud_is_big_endian() && format->cpp[0] > 1) {
-		drm_fb_swab(&dst, NULL, src, fb, rect, cached_reads);
+		drm_fb_swab(&dst, NULL, src, fb, rect, cached_reads, fmtcnv_state);
 	} else if (compression && cached_reads && pitch == fb->pitches[0]) {
 		/* can compress directly from the framebuffer */
 		buf = vaddr + rect->y1 * pitch;
@@ -266,7 +268,8 @@ static int gud_usb_bulk(struct gud_device *gdrm, size_t len)
 
 static int gud_flush_rect(struct gud_device *gdrm, struct drm_framebuffer *fb,
 			  const struct iosys_map *src, bool cached_reads,
-			  const struct drm_format_info *format, struct drm_rect *rect)
+			  const struct drm_format_info *format, struct drm_rect *rect,
+			  struct drm_format_conv_state *fmtcnv_state)
 {
 	struct gud_set_buffer_req req;
 	size_t len, trlen;
@@ -274,7 +277,7 @@ static int gud_flush_rect(struct gud_device *gdrm, struct drm_framebuffer *fb,
 
 	drm_dbg(&gdrm->drm, "Flushing [FB:%d] " DRM_RECT_FMT "\n", fb->base.id, DRM_RECT_ARG(rect));
 
-	ret = gud_prep_flush(gdrm, fb, src, cached_reads, format, rect, &req);
+	ret = gud_prep_flush(gdrm, fb, src, cached_reads, format, rect, &req, fmtcnv_state);
 	if (ret)
 		return ret;
 
@@ -318,6 +321,7 @@ static void gud_flush_damage(struct gud_device *gdrm, struct drm_framebuffer *fb
 			     const struct iosys_map *src, bool cached_reads,
 			     struct drm_rect *damage)
 {
+	struct drm_format_conv_state fmtcnv_state = DRM_FORMAT_CONV_STATE_INIT;
 	const struct drm_format_info *format;
 	unsigned int i, lines;
 	size_t pitch;
@@ -340,7 +344,7 @@ static void gud_flush_damage(struct gud_device *gdrm, struct drm_framebuffer *fb
 		rect.y1 += i * lines;
 		rect.y2 = min_t(u32, rect.y1 + lines, damage->y2);
 
-		ret = gud_flush_rect(gdrm, fb, src, cached_reads, format, &rect);
+		ret = gud_flush_rect(gdrm, fb, src, cached_reads, format, &rect, &fmtcnv_state);
 		if (ret) {
 			if (ret != -ENODEV && ret != -ECONNRESET &&
 			    ret != -ESHUTDOWN && ret != -EPROTO)
@@ -350,6 +354,8 @@ static void gud_flush_damage(struct gud_device *gdrm, struct drm_framebuffer *fb
 			break;
 		}
 	}
+
+	drm_format_conv_state_release(&fmtcnv_state);
 }
 
 void gud_flush_work(struct work_struct *work)
diff --git a/drivers/gpu/drm/solomon/ssd130x.c b/drivers/gpu/drm/solomon/ssd130x.c
index 6dcf3e0411136..3dd8e8a444b6f 100644
--- a/drivers/gpu/drm/solomon/ssd130x.c
+++ b/drivers/gpu/drm/solomon/ssd130x.c
@@ -602,7 +602,8 @@ static void ssd130x_clear_screen(struct ssd130x_device *ssd130x, u8 *data_array)
 static int ssd130x_fb_blit_rect(struct drm_framebuffer *fb,
 				const struct iosys_map *vmap,
 				struct drm_rect *rect,
-				u8 *buf, u8 *data_array)
+				u8 *buf, u8 *data_array,
+				struct drm_format_conv_state *fmtcnv_state)
 {
 	struct ssd130x_device *ssd130x = drm_to_ssd130x(fb->dev);
 	unsigned int page_height = ssd130x->device_info->page_height;
@@ -621,7 +622,7 @@ static int ssd130x_fb_blit_rect(struct drm_framebuffer *fb,
 		return ret;
 
 	iosys_map_set_vaddr(&dst, buf);
-	drm_fb_xrgb8888_to_mono(&dst, &dst_pitch, vmap, fb, rect);
+	drm_fb_xrgb8888_to_mono(&dst, &dst_pitch, vmap, fb, rect, fmtcnv_state);
 
 	drm_gem_fb_end_cpu_access(fb, DMA_FROM_DEVICE);
 
@@ -695,7 +696,8 @@ static void ssd130x_primary_plane_atomic_update(struct drm_plane *plane,
 
 		ssd130x_fb_blit_rect(fb, &shadow_plane_state->data[0], &dst_clip,
 				     ssd130x_plane_state->buffer,
-				     ssd130x_crtc_state->data_array);
+				     ssd130x_crtc_state->data_array,
+				     &shadow_plane_state->fmtcnv_state);
 	}
 
 	drm_dev_exit(idx);
diff --git a/drivers/gpu/drm/tests/drm_format_helper_test.c b/drivers/gpu/drm/tests/drm_format_helper_test.c
index f6408e56f7861..08992636ec05f 100644
--- a/drivers/gpu/drm/tests/drm_format_helper_test.c
+++ b/drivers/gpu/drm/tests/drm_format_helper_test.c
@@ -20,6 +20,10 @@
 
 #define TEST_USE_DEFAULT_PITCH 0
 
+static unsigned char fmtcnv_state_mem[PAGE_SIZE];
+static struct drm_format_conv_state fmtcnv_state =
+	DRM_FORMAT_CONV_STATE_INIT_PREALLOCATED(fmtcnv_state_mem, sizeof(fmtcnv_state_mem));
+
 struct convert_to_gray8_result {
 	unsigned int dst_pitch;
 	const u8 expected[TEST_BUF_SIZE];
@@ -630,8 +634,7 @@ static void drm_test_fb_xrgb8888_to_gray8(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_gray8(&dst, dst_pitch, &src, &fb, &params->clip);
-
+	drm_fb_xrgb8888_to_gray8(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 }
 
@@ -664,7 +667,7 @@ static void drm_test_fb_xrgb8888_to_rgb332(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_rgb332(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_rgb332(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 }
 
@@ -697,12 +700,14 @@ static void drm_test_fb_xrgb8888_to_rgb565(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_rgb565(&dst, dst_pitch, &src, &fb, &params->clip, false);
+	drm_fb_xrgb8888_to_rgb565(&dst, dst_pitch, &src, &fb, &params->clip,
+				  &fmtcnv_state, false);
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
 	buf = dst.vaddr; /* restore original value of buf */
-	drm_fb_xrgb8888_to_rgb565(&dst, &result->dst_pitch, &src, &fb, &params->clip, true);
+	drm_fb_xrgb8888_to_rgb565(&dst, &result->dst_pitch, &src, &fb, &params->clip,
+				  &fmtcnv_state, true);
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected_swab, dst_size);
 
@@ -711,7 +716,8 @@ static void drm_test_fb_xrgb8888_to_rgb565(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_RGB565, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_RGB565, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 
@@ -748,7 +754,7 @@ static void drm_test_fb_xrgb8888_to_xrgb1555(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_xrgb1555(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_xrgb1555(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -757,7 +763,8 @@ static void drm_test_fb_xrgb8888_to_xrgb1555(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XRGB1555, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XRGB1555, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 
@@ -794,7 +801,7 @@ static void drm_test_fb_xrgb8888_to_argb1555(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_argb1555(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_argb1555(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -803,7 +810,8 @@ static void drm_test_fb_xrgb8888_to_argb1555(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ARGB1555, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ARGB1555, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 
@@ -840,7 +848,7 @@ static void drm_test_fb_xrgb8888_to_rgba5551(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_rgba5551(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_rgba5551(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -849,7 +857,8 @@ static void drm_test_fb_xrgb8888_to_rgba5551(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_RGBA5551, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_RGBA5551, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le16buf_to_cpu(test, (__force const __le16 *)buf, dst_size / sizeof(__le16));
 
@@ -890,7 +899,7 @@ static void drm_test_fb_xrgb8888_to_rgb888(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_rgb888(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_rgb888(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
 	buf = dst.vaddr; /* restore original value of buf */
@@ -898,7 +907,8 @@ static void drm_test_fb_xrgb8888_to_rgb888(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_RGB888, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_RGB888, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	KUNIT_EXPECT_FALSE(test, blit_result);
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
@@ -933,7 +943,7 @@ static void drm_test_fb_xrgb8888_to_argb8888(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_argb8888(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_argb8888(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -942,7 +952,8 @@ static void drm_test_fb_xrgb8888_to_argb8888(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ARGB8888, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ARGB8888, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
@@ -979,7 +990,7 @@ static void drm_test_fb_xrgb8888_to_xrgb2101010(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_xrgb2101010(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_xrgb2101010(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le32buf_to_cpu(test, buf, dst_size / sizeof(u32));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -989,7 +1000,7 @@ static void drm_test_fb_xrgb8888_to_xrgb2101010(struct kunit *test)
 	int blit_result = 0;
 
 	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XRGB2101010, &src, &fb,
-				  &params->clip);
+				  &params->clip, &fmtcnv_state);
 
 	KUNIT_EXPECT_FALSE(test, blit_result);
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
@@ -1024,7 +1035,7 @@ static void drm_test_fb_xrgb8888_to_argb2101010(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_argb2101010(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_argb2101010(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -1034,7 +1045,7 @@ static void drm_test_fb_xrgb8888_to_argb2101010(struct kunit *test)
 	int blit_result = 0;
 
 	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ARGB2101010, &src, &fb,
-				  &params->clip);
+				  &params->clip, &fmtcnv_state);
 
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
@@ -1071,7 +1082,7 @@ static void drm_test_fb_xrgb8888_to_mono(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_xrgb8888_to_mono(&dst, dst_pitch, &src, &fb, &params->clip);
+	drm_fb_xrgb8888_to_mono(&dst, dst_pitch, &src, &fb, &params->clip, &fmtcnv_state);
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 }
 
@@ -1104,7 +1115,7 @@ static void drm_test_fb_swab(struct kunit *test)
 	const unsigned int *dst_pitch = (result->dst_pitch == TEST_USE_DEFAULT_PITCH) ?
 		NULL : &result->dst_pitch;
 
-	drm_fb_swab(&dst, dst_pitch, &src, &fb, &params->clip, false);
+	drm_fb_swab(&dst, dst_pitch, &src, &fb, &params->clip, false, &fmtcnv_state);
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 	KUNIT_EXPECT_MEMEQ(test, buf, result->expected, dst_size);
 
@@ -1114,7 +1125,7 @@ static void drm_test_fb_swab(struct kunit *test)
 	int blit_result;
 
 	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XRGB8888 | DRM_FORMAT_BIG_ENDIAN,
-				  &src, &fb, &params->clip);
+				  &src, &fb, &params->clip, &fmtcnv_state);
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
 	KUNIT_EXPECT_FALSE(test, blit_result);
@@ -1123,7 +1134,8 @@ static void drm_test_fb_swab(struct kunit *test)
 	buf = dst.vaddr;
 	memset(buf, 0, dst_size);
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_BGRX8888, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_BGRX8888, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
 	KUNIT_EXPECT_FALSE(test, blit_result);
@@ -1137,7 +1149,8 @@ static void drm_test_fb_swab(struct kunit *test)
 	mock_format.format |= DRM_FORMAT_BIG_ENDIAN;
 	fb.format = &mock_format;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XRGB8888, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XRGB8888, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
 	KUNIT_EXPECT_FALSE(test, blit_result);
@@ -1175,7 +1188,8 @@ static void drm_test_fb_xrgb8888_to_abgr8888(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ABGR8888, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_ABGR8888, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
@@ -1214,7 +1228,8 @@ static void drm_test_fb_xrgb8888_to_xbgr8888(struct kunit *test)
 
 	int blit_result = 0;
 
-	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XBGR8888, &src, &fb, &params->clip);
+	blit_result = drm_fb_blit(&dst, dst_pitch, DRM_FORMAT_XBGR8888, &src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	buf = le32buf_to_cpu(test, (__force const __le32 *)buf, dst_size / sizeof(u32));
 
@@ -1817,7 +1832,8 @@ static void drm_test_fb_memcpy(struct kunit *test)
 
 	int blit_result;
 
-	blit_result = drm_fb_blit(dst, dst_pitches, params->format, src, &fb, &params->clip);
+	blit_result = drm_fb_blit(dst, dst_pitches, params->format, src, &fb, &params->clip,
+				  &fmtcnv_state);
 
 	KUNIT_EXPECT_FALSE(test, blit_result);
 	for (size_t i = 0; i < fb.format->num_planes; i++) {
diff --git a/drivers/gpu/drm/tiny/cirrus.c b/drivers/gpu/drm/tiny/cirrus.c
index c5c34cd2edc1f..4e3a152f897ac 100644
--- a/drivers/gpu/drm/tiny/cirrus.c
+++ b/drivers/gpu/drm/tiny/cirrus.c
@@ -411,7 +411,8 @@ static void cirrus_primary_plane_helper_atomic_update(struct drm_plane *plane,
 		unsigned int offset = drm_fb_clip_offset(pitch, format, &damage);
 		struct iosys_map dst = IOSYS_MAP_INIT_OFFSET(&vaddr, offset);
 
-		drm_fb_blit(&dst, &pitch, format->format, shadow_plane_state->data, fb, &damage);
+		drm_fb_blit(&dst, &pitch, format->format, shadow_plane_state->data, fb,
+			    &damage, &shadow_plane_state->fmtcnv_state);
 	}
 
 	drm_dev_exit(idx);
diff --git a/drivers/gpu/drm/tiny/ili9225.c b/drivers/gpu/drm/tiny/ili9225.c
index 4ceb68ffac4be..dd8b0a181be94 100644
--- a/drivers/gpu/drm/tiny/ili9225.c
+++ b/drivers/gpu/drm/tiny/ili9225.c
@@ -78,7 +78,7 @@ static inline int ili9225_command(struct mipi_dbi *dbi, u8 cmd, u16 data)
 }
 
 static void ili9225_fb_dirty(struct iosys_map *src, struct drm_framebuffer *fb,
-			     struct drm_rect *rect)
+			     struct drm_rect *rect, struct drm_format_conv_state *fmtcnv_state)
 {
 	struct mipi_dbi_dev *dbidev = drm_to_mipi_dbi_dev(fb->dev);
 	unsigned int height = rect->y2 - rect->y1;
@@ -98,7 +98,7 @@ static void ili9225_fb_dirty(struct iosys_map *src, struct drm_framebuffer *fb,
 	if (!dbi->dc || !full || swap ||
 	    fb->format->format == DRM_FORMAT_XRGB8888) {
 		tr = dbidev->tx_buf;
-		ret = mipi_dbi_buf_copy(tr, src, fb, rect, swap);
+		ret = mipi_dbi_buf_copy(tr, src, fb, rect, swap, fmtcnv_state);
 		if (ret)
 			goto err_msg;
 	} else {
@@ -171,7 +171,8 @@ static void ili9225_pipe_update(struct drm_simple_display_pipe *pipe,
 		return;
 
 	if (drm_atomic_helper_damage_merged(old_state, state, &rect))
-		ili9225_fb_dirty(&shadow_plane_state->data[0], fb, &rect);
+		ili9225_fb_dirty(&shadow_plane_state->data[0], fb, &rect,
+				 &shadow_plane_state->fmtcnv_state);
 
 	drm_dev_exit(idx);
 }
@@ -281,7 +282,8 @@ static void ili9225_pipe_enable(struct drm_simple_display_pipe *pipe,
 
 	ili9225_command(dbi, ILI9225_DISPLAY_CONTROL_1, 0x1017);
 
-	ili9225_fb_dirty(&shadow_plane_state->data[0], fb, &rect);
+	ili9225_fb_dirty(&shadow_plane_state->data[0], fb, &rect,
+			 &shadow_plane_state->fmtcnv_state);
 
 out_exit:
 	drm_dev_exit(idx);
diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
index 2d999a0facdee..404c83032cecc 100644
--- a/drivers/gpu/drm/tiny/ofdrm.c
+++ b/drivers/gpu/drm/tiny/ofdrm.c
@@ -817,7 +817,7 @@ static void ofdrm_primary_plane_helper_atomic_update(struct drm_plane *plane,
 
 		iosys_map_incr(&dst, drm_fb_clip_offset(dst_pitch, dst_format, &dst_clip));
 		drm_fb_blit(&dst, &dst_pitch, dst_format->format, shadow_plane_state->data, fb,
-			    &damage);
+			    &damage, &shadow_plane_state->fmtcnv_state);
 	}
 
 	drm_dev_exit(idx);
diff --git a/drivers/gpu/drm/tiny/repaper.c b/drivers/gpu/drm/tiny/repaper.c
index 73dd4f4289c20..8fd6758f5725f 100644
--- a/drivers/gpu/drm/tiny/repaper.c
+++ b/drivers/gpu/drm/tiny/repaper.c
@@ -509,7 +509,8 @@ static void repaper_get_temperature(struct repaper_epd *epd)
 	epd->factored_stage_time = epd->stage_time * factor10x / 10;
 }
 
-static int repaper_fb_dirty(struct drm_framebuffer *fb)
+static int repaper_fb_dirty(struct drm_framebuffer *fb,
+			    struct drm_format_conv_state *fmtcnv_state)
 {
 	struct drm_gem_dma_object *dma_obj = drm_fb_dma_get_gem_obj(fb, 0);
 	struct repaper_epd *epd = drm_to_epd(fb->dev);
@@ -545,7 +546,7 @@ static int repaper_fb_dirty(struct drm_framebuffer *fb)
 
 	iosys_map_set_vaddr(&dst, buf);
 	iosys_map_set_vaddr(&vmap, dma_obj->vaddr);
-	drm_fb_xrgb8888_to_mono(&dst, &dst_pitch, &vmap, fb, &clip);
+	drm_fb_xrgb8888_to_mono(&dst, &dst_pitch, &vmap, fb, &clip, fmtcnv_state);
 
 	drm_gem_fb_end_cpu_access(fb, DMA_FROM_DEVICE);
 
@@ -830,13 +831,16 @@ static void repaper_pipe_update(struct drm_simple_display_pipe *pipe,
 				struct drm_plane_state *old_state)
 {
 	struct drm_plane_state *state = pipe->plane.state;
+	struct drm_format_conv_state fmtcnv_state = DRM_FORMAT_CONV_STATE_INIT;
 	struct drm_rect rect;
 
 	if (!pipe->crtc.state->active)
 		return;
 
 	if (drm_atomic_helper_damage_merged(old_state, state, &rect))
-		repaper_fb_dirty(state->fb);
+		repaper_fb_dirty(state->fb, &fmtcnv_state);
+
+	drm_format_conv_state_release(&fmtcnv_state);
 }
 
 static const struct drm_simple_display_pipe_funcs repaper_pipe_funcs = {
diff --git a/drivers/gpu/drm/tiny/simpledrm.c b/drivers/gpu/drm/tiny/simpledrm.c
index 9c597461d1e2a..6788f465995b3 100644
--- a/drivers/gpu/drm/tiny/simpledrm.c
+++ b/drivers/gpu/drm/tiny/simpledrm.c
@@ -609,7 +609,7 @@ static void simpledrm_primary_plane_helper_atomic_update(struct drm_plane *plane
 
 		iosys_map_incr(&dst, drm_fb_clip_offset(sdev->pitch, sdev->format, &dst_clip));
 		drm_fb_blit(&dst, &sdev->pitch, sdev->format->format, shadow_plane_state->data,
-			    fb, &damage);
+			    fb, &damage, &shadow_plane_state->fmtcnv_state);
 	}
 
 	drm_dev_exit(idx);
diff --git a/drivers/gpu/drm/tiny/st7586.c b/drivers/gpu/drm/tiny/st7586.c
index 3cf4eec16a813..7336fa1ddaed1 100644
--- a/drivers/gpu/drm/tiny/st7586.c
+++ b/drivers/gpu/drm/tiny/st7586.c
@@ -64,7 +64,8 @@ static const u8 st7586_lookup[] = { 0x7, 0x4, 0x2, 0x0 };
 
 static void st7586_xrgb8888_to_gray332(u8 *dst, void *vaddr,
 				       struct drm_framebuffer *fb,
-				       struct drm_rect *clip)
+				       struct drm_rect *clip,
+				       struct drm_format_conv_state *fmtcnv_state)
 {
 	size_t len = (clip->x2 - clip->x1) * (clip->y2 - clip->y1);
 	unsigned int x, y;
@@ -77,7 +78,7 @@ static void st7586_xrgb8888_to_gray332(u8 *dst, void *vaddr,
 
 	iosys_map_set_vaddr(&dst_map, buf);
 	iosys_map_set_vaddr(&vmap, vaddr);
-	drm_fb_xrgb8888_to_gray8(&dst_map, NULL, &vmap, fb, clip);
+	drm_fb_xrgb8888_to_gray8(&dst_map, NULL, &vmap, fb, clip, fmtcnv_state);
 	src = buf;
 
 	for (y = clip->y1; y < clip->y2; y++) {
@@ -93,7 +94,7 @@ static void st7586_xrgb8888_to_gray332(u8 *dst, void *vaddr,
 }
 
 static int st7586_buf_copy(void *dst, struct iosys_map *src, struct drm_framebuffer *fb,
-			   struct drm_rect *clip)
+			   struct drm_rect *clip, struct drm_format_conv_state *fmtcnv_state)
 {
 	int ret;
 
@@ -101,7 +102,7 @@ static int st7586_buf_copy(void *dst, struct iosys_map *src, struct drm_framebuf
 	if (ret)
 		return ret;
 
-	st7586_xrgb8888_to_gray332(dst, src->vaddr, fb, clip);
+	st7586_xrgb8888_to_gray332(dst, src->vaddr, fb, clip, fmtcnv_state);
 
 	drm_gem_fb_end_cpu_access(fb, DMA_FROM_DEVICE);
 
@@ -109,7 +110,7 @@ static int st7586_buf_copy(void *dst, struct iosys_map *src, struct drm_framebuf
 }
 
 static void st7586_fb_dirty(struct iosys_map *src, struct drm_framebuffer *fb,
-			    struct drm_rect *rect)
+			    struct drm_rect *rect, struct drm_format_conv_state *fmtcnv_state)
 {
 	struct mipi_dbi_dev *dbidev = drm_to_mipi_dbi_dev(fb->dev);
 	struct mipi_dbi *dbi = &dbidev->dbi;
@@ -121,7 +122,7 @@ static void st7586_fb_dirty(struct iosys_map *src, struct drm_framebuffer *fb,
 
 	DRM_DEBUG_KMS("Flushing [FB:%d] " DRM_RECT_FMT "\n", fb->base.id, DRM_RECT_ARG(rect));
 
-	ret = st7586_buf_copy(dbidev->tx_buf, src, fb, rect);
+	ret = st7586_buf_copy(dbidev->tx_buf, src, fb, rect, fmtcnv_state);
 	if (ret)
 		goto err_msg;
 
@@ -160,7 +161,8 @@ static void st7586_pipe_update(struct drm_simple_display_pipe *pipe,
 		return;
 
 	if (drm_atomic_helper_damage_merged(old_state, state, &rect))
-		st7586_fb_dirty(&shadow_plane_state->data[0], fb, &rect);
+		st7586_fb_dirty(&shadow_plane_state->data[0], fb, &rect,
+				&shadow_plane_state->fmtcnv_state);
 
 	drm_dev_exit(idx);
 }
@@ -238,7 +240,8 @@ static void st7586_pipe_enable(struct drm_simple_display_pipe *pipe,
 
 	msleep(100);
 
-	st7586_fb_dirty(&shadow_plane_state->data[0], fb, &rect);
+	st7586_fb_dirty(&shadow_plane_state->data[0], fb, &rect,
+			&shadow_plane_state->fmtcnv_state);
 
 	mipi_dbi_command(dbi, MIPI_DCS_SET_DISPLAY_ON);
 out_exit:
diff --git a/include/drm/drm_format_helper.h b/include/drm/drm_format_helper.h
index 724a9baf7315d..f13b34e0b752b 100644
--- a/include/drm/drm_format_helper.h
+++ b/include/drm/drm_format_helper.h
@@ -74,45 +74,49 @@ void drm_fb_memcpy(struct iosys_map *dst, const unsigned int *dst_pitch,
 		   const struct drm_rect *clip);
 void drm_fb_swab(struct iosys_map *dst, const unsigned int *dst_pitch,
 		 const struct iosys_map *src, const struct drm_framebuffer *fb,
-		 const struct drm_rect *clip, bool cached);
+		 const struct drm_rect *clip, bool cached,
+		 struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_rgb332(struct iosys_map *dst, const unsigned int *dst_pitch,
 			       const struct iosys_map *src, const struct drm_framebuffer *fb,
-			       const struct drm_rect *clip);
+			       const struct drm_rect *clip, struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_rgb565(struct iosys_map *dst, const unsigned int *dst_pitch,
 			       const struct iosys_map *src, const struct drm_framebuffer *fb,
-			       const struct drm_rect *clip, bool swab);
+			       const struct drm_rect *clip, struct drm_format_conv_state *state,
+			       bool swab);
 void drm_fb_xrgb8888_to_xrgb1555(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip);
+				 const struct drm_rect *clip, struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_argb1555(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip);
+				 const struct drm_rect *clip, struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_rgba5551(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip);
+				 const struct drm_rect *clip, struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_rgb888(struct iosys_map *dst, const unsigned int *dst_pitch,
 			       const struct iosys_map *src, const struct drm_framebuffer *fb,
-			       const struct drm_rect *clip);
+			       const struct drm_rect *clip, struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_argb8888(struct iosys_map *dst, const unsigned int *dst_pitch,
 				 const struct iosys_map *src, const struct drm_framebuffer *fb,
-				 const struct drm_rect *clip);
+				 const struct drm_rect *clip, struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_xrgb2101010(struct iosys_map *dst, const unsigned int *dst_pitch,
 				    const struct iosys_map *src, const struct drm_framebuffer *fb,
-				    const struct drm_rect *clip);
+				    const struct drm_rect *clip,
+				    struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_argb2101010(struct iosys_map *dst, const unsigned int *dst_pitch,
 				    const struct iosys_map *src, const struct drm_framebuffer *fb,
-				    const struct drm_rect *clip);
+				    const struct drm_rect *clip,
+				    struct drm_format_conv_state *state);
 void drm_fb_xrgb8888_to_gray8(struct iosys_map *dst, const unsigned int *dst_pitch,
 			      const struct iosys_map *src, const struct drm_framebuffer *fb,
-			      const struct drm_rect *clip);
+			      const struct drm_rect *clip, struct drm_format_conv_state *state);
 
 int drm_fb_blit(struct iosys_map *dst, const unsigned int *dst_pitch, uint32_t dst_format,
 		const struct iosys_map *src, const struct drm_framebuffer *fb,
-		const struct drm_rect *rect);
+		const struct drm_rect *clip, struct drm_format_conv_state *state);
 
 void drm_fb_xrgb8888_to_mono(struct iosys_map *dst, const unsigned int *dst_pitch,
 			     const struct iosys_map *src, const struct drm_framebuffer *fb,
-			     const struct drm_rect *clip);
+			     const struct drm_rect *clip, struct drm_format_conv_state *state);
 
 size_t drm_fb_build_fourcc_list(struct drm_device *dev,
 				const u32 *native_fourccs, size_t native_nfourccs,
diff --git a/include/drm/drm_mipi_dbi.h b/include/drm/drm_mipi_dbi.h
index 816f196b3d4c5..e8e0f8d39f3a6 100644
--- a/include/drm/drm_mipi_dbi.h
+++ b/include/drm/drm_mipi_dbi.h
@@ -12,6 +12,7 @@
 #include <drm/drm_device.h>
 #include <drm/drm_simple_kms_helper.h>
 
+struct drm_format_conv_state;
 struct drm_rect;
 struct gpio_desc;
 struct iosys_map;
@@ -192,7 +193,8 @@ int mipi_dbi_command_buf(struct mipi_dbi *dbi, u8 cmd, u8 *data, size_t len);
 int mipi_dbi_command_stackbuf(struct mipi_dbi *dbi, u8 cmd, const u8 *data,
 			      size_t len);
 int mipi_dbi_buf_copy(void *dst, struct iosys_map *src, struct drm_framebuffer *fb,
-		      struct drm_rect *clip, bool swap);
+		      struct drm_rect *clip, bool swap,
+		      struct drm_format_conv_state *fmtcnv_state);
 
 /**
  * mipi_dbi_command - MIPI DCS command with optional parameter(s)
-- 
2.42.0

[PATCH v5 4/7] drm/ofdrm: Preallocate format-conversion buffer in atomic_check

[Thomas Zimmermann]

Preallocate the format-conversion state's storage in the plane's
atomic_check function if a format conversion is necessary. Allows
the update to fail if no memory is available. Avoids the same
allocation within atomic_update, which may not fail.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
---
 drivers/gpu/drm/tiny/ofdrm.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/gpu/drm/tiny/ofdrm.c b/drivers/gpu/drm/tiny/ofdrm.c
index 404c83032cecc..05a72473cfc65 100644
--- a/drivers/gpu/drm/tiny/ofdrm.c
+++ b/drivers/gpu/drm/tiny/ofdrm.c
@@ -758,7 +758,11 @@ static const uint64_t ofdrm_primary_plane_format_modifiers[] = {
 static int ofdrm_primary_plane_helper_atomic_check(struct drm_plane *plane,
 						   struct drm_atomic_state *new_state)
 {
+	struct drm_device *dev = plane->dev;
+	struct ofdrm_device *odev = ofdrm_device_of_dev(dev);
 	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(new_state, plane);
+	struct drm_shadow_plane_state *new_shadow_plane_state =
+		to_drm_shadow_plane_state(new_plane_state);
 	struct drm_framebuffer *new_fb = new_plane_state->fb;
 	struct drm_crtc *new_crtc = new_plane_state->crtc;
 	struct drm_crtc_state *new_crtc_state = NULL;
@@ -777,6 +781,16 @@ static int ofdrm_primary_plane_helper_atomic_check(struct drm_plane *plane,
 	else if (!new_plane_state->visible)
 		return 0;
 
+	if (new_fb->format != odev->format) {
+		void *buf;
+
+		/* format conversion necessary; reserve buffer */
+		buf = drm_format_conv_state_reserve(&new_shadow_plane_state->fmtcnv_state,
+						    odev->pitch, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+	}
+
 	new_crtc_state = drm_atomic_get_new_crtc_state(new_state, new_plane_state->crtc);
 
 	new_ofdrm_crtc_state = to_ofdrm_crtc_state(new_crtc_state);
-- 
2.42.0

[PATCH v5 5/7] drm/simpledrm: Preallocate format-conversion buffer in atomic_check

[Thomas Zimmermann]

Preallocate the format-conversion state's storage in the plane's
atomic_check function if a format conversion is necessary. Allows
the update to fail if no memory is available. Avoids the same
allocation within atomic_update, which may not fail.

Also inline drm_plane_helper_atomic_check() into the driver and thus
return early for invisible planes. Avoids memory allocation entirely
in this case.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
---
 drivers/gpu/drm/tiny/simpledrm.c | 41 +++++++++++++++++++++++++++++++-
 1 file changed, 40 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/tiny/simpledrm.c b/drivers/gpu/drm/tiny/simpledrm.c
index 6788f465995b3..d2280c81cbaae 100644
--- a/drivers/gpu/drm/tiny/simpledrm.c
+++ b/drivers/gpu/drm/tiny/simpledrm.c
@@ -19,6 +19,7 @@
 #include <drm/drm_drv.h>
 #include <drm/drm_fbdev_generic.h>
 #include <drm/drm_format_helper.h>
+#include <drm/drm_framebuffer.h>
 #include <drm/drm_gem_atomic_helper.h>
 #include <drm/drm_gem_framebuffer_helper.h>
 #include <drm/drm_gem_shmem_helper.h>
@@ -579,6 +580,44 @@ static const uint64_t simpledrm_primary_plane_format_modifiers[] = {
 	DRM_FORMAT_MOD_INVALID
 };
 
+static int simpledrm_primary_plane_helper_atomic_check(struct drm_plane *plane,
+						       struct drm_atomic_state *state)
+{
+	struct drm_plane_state *new_plane_state = drm_atomic_get_new_plane_state(state, plane);
+	struct drm_shadow_plane_state *new_shadow_plane_state =
+		to_drm_shadow_plane_state(new_plane_state);
+	struct drm_framebuffer *new_fb = new_plane_state->fb;
+	struct drm_crtc *new_crtc = new_plane_state->crtc;
+	struct drm_crtc_state *new_crtc_state = NULL;
+	struct drm_device *dev = plane->dev;
+	struct simpledrm_device *sdev = simpledrm_device_of_dev(dev);
+	int ret;
+
+	if (new_crtc)
+		new_crtc_state = drm_atomic_get_new_crtc_state(state, new_crtc);
+
+	ret = drm_atomic_helper_check_plane_state(new_plane_state, new_crtc_state,
+						  DRM_PLANE_NO_SCALING,
+						  DRM_PLANE_NO_SCALING,
+						  false, false);
+	if (ret)
+		return ret;
+	else if (!new_plane_state->visible)
+		return 0;
+
+	if (new_fb->format != sdev->format) {
+		void *buf;
+
+		/* format conversion necessary; reserve buffer */
+		buf = drm_format_conv_state_reserve(&new_shadow_plane_state->fmtcnv_state,
+						    sdev->pitch, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+	}
+
+	return 0;
+}
+
 static void simpledrm_primary_plane_helper_atomic_update(struct drm_plane *plane,
 							 struct drm_atomic_state *state)
 {
@@ -635,7 +674,7 @@ static void simpledrm_primary_plane_helper_atomic_disable(struct drm_plane *plan
 
 static const struct drm_plane_helper_funcs simpledrm_primary_plane_helper_funcs = {
 	DRM_GEM_SHADOW_PLANE_HELPER_FUNCS,
-	.atomic_check = drm_plane_helper_atomic_check,
+	.atomic_check = simpledrm_primary_plane_helper_atomic_check,
 	.atomic_update = simpledrm_primary_plane_helper_atomic_update,
 	.atomic_disable = simpledrm_primary_plane_helper_atomic_disable,
 };
-- 
2.42.0

[PATCH v5 6/7] drm/ssd130x: Fix atomic_check for disabled planes

[Thomas Zimmermann]

The plane's atomic_check returns -EINVAL if the CRTC has not been
set. This is the case for disabled planes, for which atomic_check
should return 0. For disabled planes, it also omits the mandatory
call to drm_atomic_helper_check_plane_state().

Replace the test with the boiler-plate code that first invokes
drm_atomic_helper_check_plane_state() and then tests for the plane
to be visible. Return early for non-visible planes.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Fixes: d51f9fbd98b6 ("drm/ssd130x: Store the HW buffer in the driver-private CRTC state")
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Javier Martinez Canillas <javierm@redhat.com>
Cc: Maxime Ripard <mripard@kernel.org>
---
 drivers/gpu/drm/solomon/ssd130x.c | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/solomon/ssd130x.c b/drivers/gpu/drm/solomon/ssd130x.c
index 3dd8e8a444b6f..dccbfe33edb5e 100644
--- a/drivers/gpu/drm/solomon/ssd130x.c
+++ b/drivers/gpu/drm/solomon/ssd130x.c
@@ -639,21 +639,22 @@ static int ssd130x_primary_plane_atomic_check(struct drm_plane *plane,
 	struct drm_plane_state *plane_state = drm_atomic_get_new_plane_state(state, plane);
 	struct ssd130x_plane_state *ssd130x_state = to_ssd130x_plane_state(plane_state);
 	struct drm_crtc *crtc = plane_state->crtc;
-	struct drm_crtc_state *crtc_state;
+	struct drm_crtc_state *crtc_state = NULL;
 	const struct drm_format_info *fi;
 	unsigned int pitch;
 	int ret;
 
-	if (!crtc)
-		return -EINVAL;
-
-	crtc_state = drm_atomic_get_crtc_state(state, crtc);
-	if (IS_ERR(crtc_state))
-		return PTR_ERR(crtc_state);
+	if (crtc)
+		crtc_state = drm_atomic_get_new_crtc_state(state, crtc);
 
-	ret = drm_plane_helper_atomic_check(plane, state);
+	ret = drm_atomic_helper_check_plane_state(plane_state, crtc_state,
+						  DRM_PLANE_NO_SCALING,
+						  DRM_PLANE_NO_SCALING,
+						  false, false);
 	if (ret)
 		return ret;
+	else if (!plane_state->visible)
+		return 0;
 
 	fi = drm_format_info(DRM_FORMAT_R1);
 	if (!fi)
-- 
2.42.0

[PATCH v5 7/7] drm/ssd130x: Preallocate format-conversion buffer in atomic_check

[Thomas Zimmermann]

Preallocate the format-conversion state's storage in the plane's
atomic_check function if a format conversion is necessary. Allows
the update to fail if no memory is available. Avoids the same
allocation within atomic_update, which may not fail.

Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Javier Martinez Canillas <javierm@redhat.com>
---
 drivers/gpu/drm/solomon/ssd130x.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/drivers/gpu/drm/solomon/ssd130x.c b/drivers/gpu/drm/solomon/ssd130x.c
index dccbfe33edb5e..d3761e48bca12 100644
--- a/drivers/gpu/drm/solomon/ssd130x.c
+++ b/drivers/gpu/drm/solomon/ssd130x.c
@@ -638,6 +638,7 @@ static int ssd130x_primary_plane_atomic_check(struct drm_plane *plane,
 	struct ssd130x_device *ssd130x = drm_to_ssd130x(drm);
 	struct drm_plane_state *plane_state = drm_atomic_get_new_plane_state(state, plane);
 	struct ssd130x_plane_state *ssd130x_state = to_ssd130x_plane_state(plane_state);
+	struct drm_shadow_plane_state *shadow_plane_state = &ssd130x_state->base;
 	struct drm_crtc *crtc = plane_state->crtc;
 	struct drm_crtc_state *crtc_state = NULL;
 	const struct drm_format_info *fi;
@@ -662,6 +663,16 @@ static int ssd130x_primary_plane_atomic_check(struct drm_plane *plane,
 
 	pitch = drm_format_info_min_pitch(fi, 0, ssd130x->width);
 
+	if (plane_state->fb->format != fi) {
+		void *buf;
+
+		/* format conversion necessary; reserve buffer */
+		buf = drm_format_conv_state_reserve(&shadow_plane_state->fmtcnv_state,
+						    pitch, GFP_KERNEL);
+		if (!buf)
+			return -ENOMEM;
+	}
+
 	ssd130x_state->buffer = kcalloc(pitch, ssd130x->height, GFP_KERNEL);
 	if (!ssd130x_state->buffer)
 		return -ENOMEM;
-- 
2.42.0

Re: [PATCH v5 2/7] drm/atomic-helper: Add format-conversion state to shadow-plane state

[Javier Martinez Canillas]

Thomas Zimmermann <tzimmermann@suse.de> writes:

Hello Thomas,

> Store an instance of struct drm_format_conv_state in the shadow-plane
> state struct drm_shadow_plane_state. Many drivers with shadow planes
> use DRM's format helpers to copy or convert the framebuffer data to
> backing storage in the scanout buffer. The shadow plane provides the
> necessary state and manages the conversion's intermediate buffer memory.
>

I would argue this is one of the primary selling points for this series
(since Maxime was asking about the motivation). As you point out, many
drivers that use a shadow-buffer can also expose a different format than
the native one to user-space, so makes sense to have the buffer used for
conversion as a part of the shadow-plane state.

Just mentioning this because you are not including that in the paragraph
talking about the motivation for the patch series in its cover letter.

-- 
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat

Re: [PATCH v5 2/7] drm/atomic-helper: Add format-conversion state to shadow-plane state

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1001 bytes --]

On Mon, Oct 09, 2023 at 04:22:29PM +0200, Javier Martinez Canillas wrote:
> Thomas Zimmermann <tzimmermann@suse.de> writes:
> > Store an instance of struct drm_format_conv_state in the shadow-plane
> > state struct drm_shadow_plane_state. Many drivers with shadow planes
> > use DRM's format helpers to copy or convert the framebuffer data to
> > backing storage in the scanout buffer. The shadow plane provides the
> > necessary state and manages the conversion's intermediate buffer memory.
> >
> 
> I would argue this is one of the primary selling points for this series
> (since Maxime was asking about the motivation). As you point out, many
> drivers that use a shadow-buffer can also expose a different format than
> the native one to user-space, so makes sense to have the buffer used for
> conversion as a part of the shadow-plane state.

I'm confused, can we convert between formats or not? Or is it about
XRGB8888 vs whatever native format is used by the device?

Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH v5 2/7] drm/atomic-helper: Add format-conversion state to shadow-plane state

[Javier Martinez Canillas]

Maxime Ripard <mripard@kernel.org> writes:

> On Mon, Oct 09, 2023 at 04:22:29PM +0200, Javier Martinez Canillas wrote:
>> Thomas Zimmermann <tzimmermann@suse.de> writes:
>> > Store an instance of struct drm_format_conv_state in the shadow-plane
>> > state struct drm_shadow_plane_state. Many drivers with shadow planes
>> > use DRM's format helpers to copy or convert the framebuffer data to
>> > backing storage in the scanout buffer. The shadow plane provides the
>> > necessary state and manages the conversion's intermediate buffer memory.
>> >
>> 
>> I would argue this is one of the primary selling points for this series
>> (since Maxime was asking about the motivation). As you point out, many
>> drivers that use a shadow-buffer can also expose a different format than
>> the native one to user-space, so makes sense to have the buffer used for
>> conversion as a part of the shadow-plane state.
>
> I'm confused, can we convert between formats or not? Or is it about
> XRGB8888 vs whatever native format is used by the device?
>

It is about the latter, converting XRGB8888 to the native format.

-- 
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat

Re: [PATCH v5 6/7] drm/ssd130x: Fix atomic_check for disabled planes

[Javier Martinez Canillas]

Thomas Zimmermann <tzimmermann@suse.de> writes:

Hello Thomas,

> The plane's atomic_check returns -EINVAL if the CRTC has not been
> set. This is the case for disabled planes, for which atomic_check
> should return 0. For disabled planes, it also omits the mandatory
> call to drm_atomic_helper_check_plane_state().
>
> Replace the test with the boiler-plate code that first invokes
> drm_atomic_helper_check_plane_state() and then tests for the plane
> to be visible. Return early for non-visible planes.
>
> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
> Fixes: d51f9fbd98b6 ("drm/ssd130x: Store the HW buffer in the driver-private CRTC state")
> Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
> Tested-by: Javier Martinez Canillas <javierm@redhat.com>
> Cc: Geert Uytterhoeven <geert@linux-m68k.org>
> Cc: Javier Martinez Canillas <javierm@redhat.com>
> Cc: Maxime Ripard <mripard@kernel.org>
> ---

I've picked this since it's an unrelated fix to the rest of the series.

Pushed to drm-misc (drm-misc-next) because the offending commit is not
in mainline yet. Thanks!

-- 
Best regards,

Javier Martinez Canillas
Core Platforms
Red Hat

Re: [PATCH v5 6/7] drm/ssd130x: Fix atomic_check for disabled planes

[Thomas Zimmermann]

[-- Attachment #1.1: Type: text/plain, Size: 1504 bytes --]

Hi

Am 13.10.23 um 19:16 schrieb Javier Martinez Canillas:
> Thomas Zimmermann <tzimmermann@suse.de> writes:
> 
> Hello Thomas,
> 
>> The plane's atomic_check returns -EINVAL if the CRTC has not been
>> set. This is the case for disabled planes, for which atomic_check
>> should return 0. For disabled planes, it also omits the mandatory
>> call to drm_atomic_helper_check_plane_state().
>>
>> Replace the test with the boiler-plate code that first invokes
>> drm_atomic_helper_check_plane_state() and then tests for the plane
>> to be visible. Return early for non-visible planes.
>>
>> Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
>> Fixes: d51f9fbd98b6 ("drm/ssd130x: Store the HW buffer in the driver-private CRTC state")
>> Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
>> Tested-by: Javier Martinez Canillas <javierm@redhat.com>
>> Cc: Geert Uytterhoeven <geert@linux-m68k.org>
>> Cc: Javier Martinez Canillas <javierm@redhat.com>
>> Cc: Maxime Ripard <mripard@kernel.org>
>> ---
> 
> I've picked this since it's an unrelated fix to the rest of the series.
> 
> Pushed to drm-misc (drm-misc-next) because the offending commit is not
> in mainline yet. Thanks!

Makes sense, thanks a lot.

Best regards
Thomas

> 

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstrasse 146, 90461 Nuernberg, Germany
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman
HRB 36809 (AG Nuernberg)

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [PATCH v5 0/7] drm: Reuse temporary memory for format conversion

[Maxime Ripard]

[-- Attachment #1: Type: text/plain, Size: 1657 bytes --]

On Mon, Oct 09, 2023 at 04:06:29PM +0200, Thomas Zimmermann wrote:
> DRM's format-conversion helpers require temporary memory. Pass the
> buffer from the caller to allow the caller to preallocate the buffer
> memory.
> 
> The motivation for this patchset is the recent work on a DRM panic
> handler. [1] The panic handler requires format conversion to display an
> error to the screen. But allocating memory during a kernel panic is
> fragile. The changes in this patchset enable the DRM panic handler to
> preallocate buffer storage before the panic occurs.
> 
> Patch 1 adds struct drm_format_conv_state, a simple interface to pass
> around the buffer storage. Patch 2 adds an instance of the struct to
> the shadow-plane state. Patch 3 moves the buffer's memory management
> from the format helpers into their callers within the DRM drivers. Most
> of the affected drivers use the state instance stored in their shadow-
> plane state. The shadow-plane code releases the buffer memory automatically.
> 
> Patches 4 to 7 update three drivers to pre-allocate the format-conversion
> buffer in their plane's atomic_check function. The drivers thus detect OOM
> errors before the display update begins.
> 
> Tested with simpledrm.

So, I just discovered that you merged that series.

You've complained before about "sneaking patches in", and while I was
disagreeing with you then, this particular instance is definitely a
strong case for it. You've merged it without telling anyone, and despite
our ongoing conversation on the v4 that was active more recently than
the v5. And that you never responded to.

Awesome.

Maxime

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]

Re: [PATCH v5 0/7] drm: Reuse temporary memory for format conversion

[Thomas Zimmermann]

[-- Attachment #1.1: Type: text/plain, Size: 2486 bytes --]

Hi

Am 17.11.23 um 10:34 schrieb Maxime Ripard:
> On Mon, Oct 09, 2023 at 04:06:29PM +0200, Thomas Zimmermann wrote:
>> DRM's format-conversion helpers require temporary memory. Pass the
>> buffer from the caller to allow the caller to preallocate the buffer
>> memory.
>>
>> The motivation for this patchset is the recent work on a DRM panic
>> handler. [1] The panic handler requires format conversion to display an
>> error to the screen. But allocating memory during a kernel panic is
>> fragile. The changes in this patchset enable the DRM panic handler to
>> preallocate buffer storage before the panic occurs.
>>
>> Patch 1 adds struct drm_format_conv_state, a simple interface to pass
>> around the buffer storage. Patch 2 adds an instance of the struct to
>> the shadow-plane state. Patch 3 moves the buffer's memory management
>> from the format helpers into their callers within the DRM drivers. Most
>> of the affected drivers use the state instance stored in their shadow-
>> plane state. The shadow-plane code releases the buffer memory automatically.
>>
>> Patches 4 to 7 update three drivers to pre-allocate the format-conversion
>> buffer in their plane's atomic_check function. The drivers thus detect OOM
>> errors before the display update begins.
>>
>> Tested with simpledrm.
> 
> So, I just discovered that you merged that series.
> 
> You've complained before about "sneaking patches in", and while I was
> disagreeing with you then, this particular instance is definitely a
> strong case for it. You've merged it without telling anyone, and despite
> our ongoing conversation on the v4 that was active more recently than
> the v5. And that you never responded to.
> 
> Awesome.

My apologies. From my point of view, that conversion had ended. I left 
the patch set for a while to wait for further comments or questions, but 
nothing happened. So I merged it.

Revert it if you cannot live with the changes. IIRC you found the 
reduced number of alloc/free cycles to be irrelevant. But even then, the 
patches allow us to move the allocation from atomic_update to 
atomic_check, thus detecting allocation failures early. That's an 
improvement to me.

Best regards
Thomas

> 
> Maxime

-- 
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstrasse 146, 90461 Nuernberg, Germany
GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman
HRB 36809 (AG Nuernberg)

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: [PATCH v5 0/7] drm: Reuse temporary memory for format conversion

[Daniel Vetter]

On Fri, 17 Nov 2023 at 11:07, Thomas Zimmermann <tzimmermann@suse.de> wrote:
>
> Hi
>
> Am 17.11.23 um 10:34 schrieb Maxime Ripard:
> > On Mon, Oct 09, 2023 at 04:06:29PM +0200, Thomas Zimmermann wrote:
> >> DRM's format-conversion helpers require temporary memory. Pass the
> >> buffer from the caller to allow the caller to preallocate the buffer
> >> memory.
> >>
> >> The motivation for this patchset is the recent work on a DRM panic
> >> handler. [1] The panic handler requires format conversion to display an
> >> error to the screen. But allocating memory during a kernel panic is
> >> fragile. The changes in this patchset enable the DRM panic handler to
> >> preallocate buffer storage before the panic occurs.
> >>
> >> Patch 1 adds struct drm_format_conv_state, a simple interface to pass
> >> around the buffer storage. Patch 2 adds an instance of the struct to
> >> the shadow-plane state. Patch 3 moves the buffer's memory management
> >> from the format helpers into their callers within the DRM drivers. Most
> >> of the affected drivers use the state instance stored in their shadow-
> >> plane state. The shadow-plane code releases the buffer memory automatically.
> >>
> >> Patches 4 to 7 update three drivers to pre-allocate the format-conversion
> >> buffer in their plane's atomic_check function. The drivers thus detect OOM
> >> errors before the display update begins.
> >>
> >> Tested with simpledrm.
> >
> > So, I just discovered that you merged that series.
> >
> > You've complained before about "sneaking patches in", and while I was
> > disagreeing with you then, this particular instance is definitely a
> > strong case for it. You've merged it without telling anyone, and despite
> > our ongoing conversation on the v4 that was active more recently than
> > the v5. And that you never responded to.
> >
> > Awesome.
>
> My apologies. From my point of view, that conversion had ended. I left
> the patch set for a while to wait for further comments or questions, but
> nothing happened. So I merged it.
>
> Revert it if you cannot live with the changes. IIRC you found the
> reduced number of alloc/free cycles to be irrelevant. But even then, the
> patches allow us to move the allocation from atomic_update to
> atomic_check, thus detecting allocation failures early. That's an
> improvement to me.

Just a small comment on this, I didn't read up the full discussion:

Yeah allocating memory in atomic_commit after the point of no return
is no-go. Usually the best spot is the prepare_fb hooks since that
avoids doing expensive allocations for TEST_ONLY commits, but since
the allocation is fairly small it probably doesn't matter overall.
Well the new-ish begin/end_fb_access helpers would be even better I
guess for the atomic flow, but break the panic handler use-case I
think.

Oh and a bikeshed, _copy() feels a bit like a funny postfix for
something that does a state duplication, everywhere else in atomic we
put _duplicate into the name for these functions.

Cheers, Sima

>
> Best regards
> Thomas
>
> >
> > Maxime
>
> --
> Thomas Zimmermann
> Graphics Driver Developer
> SUSE Software Solutions Germany GmbH
> Frankenstrasse 146, 90461 Nuernberg, Germany
> GF: Ivo Totev, Andrew Myers, Andrew McDonald, Boudien Moerman
> HRB 36809 (AG Nuernberg)



-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch