The usage of the C keyword "inline" seems to be a common
misunderstanding: it's a *hint* only, and modern compilers will inline
(or not) functions based on their own judgement and provided compiler
options.
So while marking functions as "inline" does not do much, missing the
"static" keyword will force to compiler to spell out a version of the
function for potential external callers, which actually increases the
code size (though hopefully the linker will drop the function).
Change the "inline" attribute for the mbus_configure_port() functions in
some Allwinner DRAM drivers to "static", so that the explicit version
can actually be dropped from the object file, reducing the code size.
"static inline" has a use case in header files, where it avoids a warning
if a .c file including this header does not use the particular function.
In a .c file itself "static inline" is not useful otherwise, so just use
static here as well.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jernej Skrabec <jernej.skrabec@gmail.com>
H6 and H616 SPL code has a few writes to unknown PRCM registers. Now
that we know what they are, let's replace magic offsets with proper
register names.
Signed-off-by: Jernej Skrabec <jernej.skrabec@gmail.com>
Reviewed-by: Samuel Holland <samuel@sholland.org>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Most clock factors and dividers in the H6 PLLs use a "+1 encoding",
which we were missing on two occasions.
This fixes the MMC clock setup on the H6, which could be slightly off due
to the wrong parent frequency:
mmc 2 set mod-clk req 52000000 parent 1176000000 n 2 m 12 rate 49000000
Also the CPU frequency (PLL1) was a tad too high before.
For PLL5 (DRAM) we already accounted for this +1, but in the DRAM code
itself, not in the bit field macro. Move this there to be aligned with
what the other SoCs and other PLLs do.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jernej Skrabec <jernej.skrabec@gmail.com>
It turns out that in rare cases, current analytical approach to detect
correct DRAM bus width and rank on H6 doesn't work. On some TV boxes
with DDR3, incorrect DRAM configuration triggers write leveling error
which immediately stops initialization process. Exact reason why this
error appears isn't known. However, if correct configuration is used,
initalization works without problem.
In order to fix this issue, simply try another configuration when any
kind of error appears during initialization, not just those related to
rank and bus width.
Signed-off-by: Jernej Skrabec <jernej.skrabec@siol.net>
Tested-by: Thomas Graichen <thomas.graichen@googlemail.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Half DQ configuration seems to be very rare for H6 based boards/STBs,
but exists nevertheless. Currently the only known product which needs
this support is Tanix TX6 mini.
This commit adds support for half DQ configuration. Code was tested
for regressions on other configurations (OrangePi 3 1 GiB/LPDDR3, Tanix
TX6 4 GiB/DDR3) and none were found.
Thanks to Icenowy Zheng for help with this code.
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: thomas graichen <thomas.graichen@gmail.com>
Signed-off-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Maxime Ripard <mripard@kernel.org>
Add some basic line delay values to be used with DDR3 DRAM chips on
some H6 TV boxes.
Taken from a register dump after boot0 initialised the DRAM.
Put them as the default delay values for DDR3 DRAM until we know better.
Signed-off-by: Jernej Skrabec <jernej.skrabec@siol.net>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
At the moment the H6 DRAM driver only supports LPDDR3 DRAM.
Extend the driver to cover DDR3 DRAM as well.
The changes are partly motivated by looking at the ZynqMP register
documentation, partly by looking at register dumps after boot0/libdram
has initialised the controller.
Many thanks to Jernej for contributing some fixes!
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Currently the H6 DRAM driver only supports one kind of LPDDR3 DRAM.
Split the timing parameters for this LPDDR3 configuration into a
separate file, to allow selecting an alternative later at compile time
(as the sunxi-dw driver does).
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
The DRAM controller manual suggests to first program the PHY
initialisation parameters to the PHY_PIR register, and then set bit 0 to
trigger the initialisation. This is also used in boot0.
Follow this recommendation by setting bit 0 in a separate step.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Using memcpy() is, however tempting, not a good idea: It depends on the
specific implementation of memcpy, also lacks barriers. In this
particular case the first registers were written using 64-bit writes,
and the last register using four separate single-byte writes.
Replace the memcpy with a proper loop using the writel() accessor.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Tested-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
Currently the DRAM bus gate and reset is changed at the same time in
H6 DRAM initialization code, which disobeys the user manual's
programming guide.
Fix the sequence by follow the sequence suggested by the user manual
(ungate the bus clock after release the reset signal).
By some experiments it seems to fix the DRAM size detection failure that
rarely happens.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Jagan Teki <jagan@amarulasolutions.com>
The Allwinner H6 SoC comes with a set of new DRAM controller+PHY combo.
Both the controller and the PHY seem to be originate from DesignWare,
and are similar to the ones in ZynqMP SoCs.
This commit introduces an initial DRAM driver for H6, which contains
only LPDDR3 support. The currently known SBCs with H6 all come with
LPDDR3 memory, including Pine H64 and several Orange Pi's.
The BSP DRAM initialization code is closed source and violates GPL. Code
in this commit is written by experimenting, referring the code/document
of other users of the IPs (mainly the ZynqMP, as it's the only found PHY
reference) and disassebling the BSP blob.
Thanks for Jernej Skrabec for review and fix some issues in this driver
(including the most critical one which made it to work), and rewrite
some code from register dump!
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Acked-by: Jagan Teki <jagan@openedev.com>
Tested-by: Jagan Teki <jagan@amarulasolutions.com>
