The implementation of dma_map_single() and dma_unmap_single() is
exactly the same for all the architectures that support them.

Factor them out to <linux/dma-mapping.h>, and make all drivers to
include <linux/dma-mapping.h> instead of <asm/dma-mapping.h>.

If we need to differentiate them for some architectures, we can
move the generic definitions to <asm-generic/dma-mapping.h>.

Add some comments to the helpers. The concept is quite similar to
the DMA-API of Linux kernel. Drivers are agnostic about what is
going on behind the scene. Just call dma_map_single() before the
DMA, and dma_unmap_single() after it.

Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>

At present dm/device.h includes the linux-compatible features. This
requires including linux/compat.h which in turn includes a lot of headers.
One of these is malloc.h which we thus end up including in every file in
U-Boot. Apart from the inefficiency of this, it is problematic for sandbox
which needs to use the system malloc() in some files.

Move the compatibility features into a separate header file.

Signed-off-by: Simon Glass <sjg@chromium.org>

At present devres.h is included in all files that include dm.h but few
make use of it. Also this pulls in linux/compat which adds several more
headers. Drop the automatic inclusion and require files to include devres
themselves. This provides a good indication of which files use devres.

Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Anatolij Gustschin <agust@denx.de>

This function name conflicts with our desire to #define free() to
something else on sandbox. Since it deals with resources, rename it to
rfree().

Signed-off-by: Simon Glass <sjg@chromium.org>

Cast pointers properly so as to avoid warnings when driver is built for
32 bit platforms

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>

UDMA always expects 64 bit address pointer of the transfer descriptor in
the Ring. But on 32 bit cores like R5, pointer is always 32 bit in size.
Therefore copy over 32 bit pointer value to 64 bit variable before
pushing it over to the ring, so that upper 32 bits are 0s.

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>

Exposed ring mode works well with 32 bit and 64 bit cores without need
for Proxies for 32 bit cores. Therefore switch to exposed ring mode.

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>

Fix up the debug prints that were dumping state of TCHAN RT registers to
use tchan for MEM_TO_DEV transfers.

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>

Remove redundant coherency checks before calling cache ops in UDMA
driver. This is now handled in arch specific cache operation
implementation based on Kconfig option

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>

On K3 SoCs, DMA channels are shared across multiple entities, therefore
U-Boot DMA driver needs to query resource range from centralised
resource management controller i.e SystemFirmware and use DMA channels
allocated for A72 host. Add support for the same.

Signed-off-by: Vignesh Raghavendra <vigneshr@ti.com>
Reviewed-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>