Some SoCs (as seen on A20) seem to misreport the MMC FIFO level if the
FIFO is completely full: the level size reads as zero, but the FIFO_FULL
bit is set. We won't do a single iteration of the read loop in this
case, so will be stuck forever.
Check for this situation and use a safe minimal FIFO size instead when
we hit this case.
This fixes MMC boot on A20 devices after the MMC FIFO optimisation
(9faae5457f).
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
At the moment the Allwinner MMC driver parses the bus-width and
non-removable DT properties itself, in the probe() routine.
There is actually a generic function provided by the MMC framework doing
this job, also it parses more generic properties like broken-cd and
advanced transfer modes.
Drop our own code and call mmc_of_parse() instead, to get all new
features for free.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
To avoid the complexity of DMA operations (with chained descriptors), we
use repeated MMIO reads and writes to the SD_FIFO_REG, which allows us
to drain or fill the MMC data buffer FIFO very easily.
However those MMIO accesses are somewhat costly, so this limits our MMC
performance, to between 17 and 22 MB/s, but down to 9.5 MB/s on the H6
(partly due to the lower AHB1 frequency).
As it turns out we read the FIFO status register after *every* word we
read or write, which effectively doubles the number of MMIO accesses,
thus effectively more than halving our performance.
To avoid this overhead, we can make use of the FIFO level bits, which are
in the very same FIFO status registers.
So for a read request, we now can collect as many words as the FIFO
level originally indicated, and only then need to update the status
register.
We don't know for sure the size of the FIFO (and it seems to differ
across SoCs anyway), so writing is more fragile, which is why we still
use the old method for that. If we find a minimum FIFO size available on
all SoCs, we could use that, in a later optimisation.
This patch increases the eMMC read speed on a Pine64-LTS from about
22MB/s to 44 MB/s. SD card reads don't gain that much, but with 23 MB/s
we now reach the practical limit for 3.3V SD cards.
On the H6 we double our transfer speed, from 9.5 MB/s to 19.7 MB/s.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Newer SoCs have a self calibration feature, which avoids us writing hard
coded phase delay values into the controller.
Consolidate the code by avoiding unnecessary #ifdefs, and also enabling
the feature for all those newer SoCs.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Among the SoCs using the "new timing mode", only the A83T needs to
explicitly switch to that mode.
By just defining the symbol for that one odd A83T bit to 0 for any other
SoCs, we can always OR that in, and save the confusing nested #ifdefs.
Clean up the also confusing new_mode setting on the way.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Most Allwinner SoCs which use the so called "new timing mode" in their
MMC controllers actually use the double-rate PLL6/PERIPH0 clock as their
parent input clock. This is interestingly enough compensated by a hidden
"by 2" post-divider in the mod clock, so the divider and actual output
rate stay the same.
Even though for the H6 and H616 (but only for them!) we use the doubled
input clock for the divider computation, we never accounted for the
implicit post-divider, so the clock was only half the speed on those SoCs.
This didn't really matter so far, as our slow MMIO routine limits the
transfer speed anyway, but we will fix this soon.
Clean up the code around that selection, to always use the normal PLL6
(PERIPH0(1x)) clock as an input. As the rate and divider are the same,
that makes no difference.
Explain the hardware differences in a comment.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
When enabling PHYS_64BIT on 32-bit platforms, we get two warnings about
pointer casts in sunxi_mmc.c. Those are related to MMIO addresses, which
are always below 1GB on all Allwinner SoCs, so there is no problem with
anything having more than 32 bits.
Add the proper casts to make it compile cleanly.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
The delay and bus-width setup are slightly different across the
Allwinner SoC generations, and we covered this so far with some
preprocessor conditionals.
Use the more readable IS_ENABLE() instead.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
So far the only difference between the various Allwinner MMC controller
we are concerned about is the mod clock register offset.
This is actually not directly related to the MMC controller IP, but an
integration choice, dependent on the SoC this appears in.
To avoid becoming trapped with some compatible fallback strings, let's
remove the whole struct sunxi_mmc_variant, and replace this with a SoC
based choice, which we can derive from the CONFIG_MACH_SUNx_y symbols.
This will later simplify H616 support.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
It turns out that several SoCs share same mmc configuration as H6. In
order to lower ifdef clutter replace H6 specific macro with common one.
Signed-off-by: Jernej Skrabec <jernej.skrabec@siol.net>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
We use 'priv' for private data but often use 'platdata' for platform data.
We can't really use 'pdata' since that is ambiguous (it could mean private
or platform data).
Rename some of the latter variables to end with 'plat' for consistency.
Signed-off-by: Simon Glass <sjg@chromium.org>
This construct is quite long-winded. In earlier days it made some sense
since auto-allocation was a strange concept. But with driver model now
used pretty universally, we can shorten this to 'auto'. This reduces
verbosity and makes it easier to read.
Coincidentally it also ensures that every declaration is on one line,
thus making dtoc's job easier.
Signed-off-by: Simon Glass <sjg@chromium.org>
If a board DT describes a cd-gpios property, but also marks the storage
as non-removable, we must ignore the GPIO (as Linux does).
Teach the DM_MMC part of the Allwinner MMC driver about the
non-removable DT property, to fix DM_MMC access on the SoPine and
Pine64-LTS board.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Acked-by: Jagan Teki <jagan@openedev.com>
Now that we have the gate clocks and the reset gates in our new
Allwinner clock driver, let's make use of them in the MMC driver, when
DM_MMC is defined.
We treat the reset device as optional now, as the older SoCs don't
implement it.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
A80 gates clock already be part of CLK framework, so just
add mod_clk offset with A80 compatible string.
Cc: Rask Ingemann Lambertsen <rask@formelder.dk>
Cc: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Unlike other Allwinner SoC's, H6 uses a different MMC mod clock offset.
Connect that with the respective compatible string.
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Add MMC compatible strings for A83T, A64, H5.
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Existing clock configure code has been followed based on the
legacy MMC dt node definitions and it cannot work with recent
dts(i) sync from Linux.
So, add clock configure code for Allwinner platforms which support
DM_MMC and eventually this will drop once CLK support is in Mainline.
Fixes: 3c92cca3cd ("ARM: dts: sun4i: Update A10 dts(i) files from Linux-v4.18-rc3")
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Tested-by: Priit Laes <plaes@plaes.org> # Gemei G9 A10 Tablet
Tested-by: Marek Kraus <gamelasterv2@gmail.com> # A10-OLinuXino-Lime
Using new mode improves stability of eMMC and SD cards. Without
it SPL fails to load u-boot from SD on Pinebook.
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Tested-by: Jagan Teki <jagan@amarulasolutions.com> # Amarula A64-Relic
Reviewed-by: Jagan Teki <jagan@openedev.com>
Comment in Linux driver says that clock needs to be doubled only
if we use DDR modes, moreover divider has to be set accordingly.
U-boot driver doesn't declare support for any DDR modes and doesn't
set internal clock divider in CLKCR, so it doubles clock
unconditionally when new mode is used.
Some cards can't handle that and as result SPL fails to load u-boot.
Fixes: de9b1771c3 ("mmc: sunxi: Support new mode")
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Tested-by: Jagan Teki <jagan@amarulasolutions.com> # Amarula A64-Relic
Allwinner A64 has new mode but doesn't have a mode switch in CCM,
and CCM_MMC_CTRL_MODE_SEL_NEW is not defined, so compilation fails
if MMC_SUNXI_HAS_NEW_MODE is enabled
Introduce new MMC_SUNXI_HAS_MODE_SWITCH option to be able to ifdef usage
of CCM_MMC_CTRL_MODE_SEL_NEW
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
[jagan: update commit message]
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Tested-by: Jagan Teki <jagan@amarulasolutions.com> # Amarula A64-Relic
A64 and H6 support automatic delay calibration and Linux driver uses it
instead of hardcoded delays. Add support for it to u-boot driver.
Fixes eMMC instability on Pinebook
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Tested-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Cc: Vagrant Cascadian <vagrant@debian.org>
Reviewed-by: Jagan Teki <jagan@openedev.com>
The Allwinner H6 SoC has 3 MMC controllers like the ones in A64, with
the MMC2 come with the capability to do crypto by EMCE.
Add MMC support for H6. EMCE support is not added yet.
Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Tested-by: Jagan Teki <jagan@amarulasolutions.com>
Commit dd27918c22 ("dm: mmc: sunxi: Add support for driver model")
only added the allwinner,sun5i-a13-mmc compatible string for this
driver. The DM initialisation code here also works with (at least) A10
and A20, so add the appropriate compatible strings as per Linux 4.17's
driver.
Tested on A10 Cubieboard and A20 pcDuino3 Nano with CONFIG_DM_MMC.
(A20 worked already, because sun7i-a20.dtsi specifies both the A13 and
A20 strings.)
Signed-off-by: Adam Sampson <ats@offog.org>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Jagan Teki <jteki@openedev.com>
When U-Boot started using SPDX tags we were among the early adopters and
there weren't a lot of other examples to borrow from. So we picked the
area of the file that usually had a full license text and replaced it
with an appropriate SPDX-License-Identifier: entry. Since then, the
Linux Kernel has adopted SPDX tags and they place it as the very first
line in a file (except where shebangs are used, then it's second line)
and with slightly different comment styles than us.
In part due to community overlap, in part due to better tag visibility
and in part for other minor reasons, switch over to that style.
This commit changes all instances where we have a single declared
license in the tag as both the before and after are identical in tag
contents. There's also a few places where I found we did not have a tag
and have introduced one.
Signed-off-by: Tom Rini <trini@konsulko.com>
Throughput tests have shown the sunxi_mmc driver to take over 10s to
read 10MB from a fast eMMC device due to excessive delays in polling
loops.
This commit restructures the main polling loops to use get_timer(...)
to determine whether a (millisecond) timeout has expired. We choose
not to use the wait_bit function, as we don't need interruptability
with ctrl-c and have at least one case where two bits (one for an
error condition and another one for completion) need to be read and
using wait_bit would have not added to the clarity.
The observed speedup in testing on a A31 is greater than 10x (e.g. a
10MB write decreases from 9.302s to 0.884s).
Signed-off-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Signed-off-by: Maxime Ripard <maxime.ripard@bootlin.com>
Tested-by: Mylène Josserand <mylene.josserand@bootlin.com>
Acked-by: Jagan Teki <jagan@openedev.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
U-boot driver for sunxi-mmc uses PLL6, unlike linux kernel where
PLL5 is used, with clock rates respectively 600MHz and 768MHz.
Thus there are different phase degree steps - 24 for the kernel and
30 for u-boot.
In the kernel driver the phase is set 90 deg for output and 120 for
sample. Dividing by 30 will result values 3 and 4. Those are the
values set in the u-boot driver.
However, the condition defining delays is wrong. MMC core driver
requests clock of 52MHz, sunxi-driver sets clock of 50MHz, but
phase is set 30 deg for output and 120 deg for sample.
Apparently this works for most cards.
On A20-SOM204-EVB-eMMC there is eMMC card (KLMAG2GEND) which complains
about it. Maybe there is other boards with similar problem?
So the fix is to match delays for both u-boot and kernel.
Signed-off-by: Stefan Mavrodiev <stefan@olimex.com>
Acked-by: Maxime Ripard <maxime.ripard@bootlin.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
With CONFIG_DM_MMC the BananaPi does not detect SD cards.
The sunxi device trees use the cd-inverted property to indicate that
the card detect is inverted.
This property is documented in Linux kernel devicetree/bindings/mmc/mmc.txt
The property is not marked as deprecated.
A similar patch was posted by Tuomas but is in status "Changes Requested".
https://patchwork.ozlabs.org/patch/850377/
This patch is a stripped down version of his patch.
Suggested-by: Tuomas Tynkkynen <tuomas@tuxera.com>
Signed-off-by: Heinrich Schuchardt <xypron.glpk@gmx.de>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
When enabling the new mmc timing mode, we inadvertently clear all the
remaining bits in the new timing mode register. The bits cleared
include a default phase delay on the output clock. The BSP kernel
states that the default values are supposed to be used. Clearing them
results in decreased performance or transfer errors on some boards.
Fixes: de9b1771c3 ("mmc: sunxi: Support new mode")
Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
The driver-model rework changed, among other things, the way the private
data were moved around. It now uses the private field in the struct mmc.
However, the mmc_create argument was changed in the process to always pass
the array we used to have to store our private structures.
The basically means that all the MMC driver instances will now have the
private data of the first instance, which obviously doesn't work very well.
Pass the proper pointer to mmc_create.
Fixes: 034e226bc7 ("dm: mmc: sunxi: Pass private data around explicitly")
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Tested-by: Chen-Yu Tsai <wens@csie.org>
Tested-by: Jagan Teki <jagan@openedev.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Almost all of the newer Allwinner SoCs have a new operating mode for the
eMMC clocks that needs to be enabled in both the clock and the MMC
controller.
Details about that mode are sparse, and the name itself (new mode vs old
mode) doesn't give much details, but it seems that the it changes the
sampling of the MMC clock. One side effect is also that it divides the
parent clock rate by 2.
Add support for it through a Kconfig option.
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Reviewed-by: Jagan Teki <jagan@openedev.com>
Add a driver-model version of this driver which mostly uses the existing
code. The old code can be removed once all boards are switched over.
Signed-off-by: Simon Glass <sjg@chromium.org>
This function has #ifdefs in it which we want to avoid for driver model.
Instead we should use different compatible strings and the .data field.
It also uses the MMC device number which is not available in driver
model except through aliases.
Move the function's into its caller so that the driver-model version can
do things its own way.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
At present the driver-private data is obtained in various functions by
various means. With driver model this is provided automatically. Without
driver model it comes from a C array declared at the top of the file.
Adjust internal functions so that they are passed the private data as
a parameter, allowing the caller to obtain it using either means.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Use the driver-model naming convention for this structure. It is data
private to the driver so the local variable should be called 'priv'.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Use the driver-model naming convention for this structure. It is data
private to the driver.
Signed-off-by: Simon Glass <sjg@chromium.org>
Acked-by: Maxime Ripard <maxime.ripard@free-electrons.com>
The sun8i SoCs also have a 8 bits capable MMC2 controller. Enable the
support for those too.
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Use the generic error number instead of specific error number.
If use the generic error number, it can debug more easier.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Minkyu Kang <mk7.kang@samsung.com>
Now that we know that the BROM stores a value indicating the boot-source
at the beginning of SRAM, use that instead of trying to recreate the
BROM's boot probing.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Ian Campbell <ijc@hellion.org.uk>
With a recent bunch of SD3.0 cards in our A20-based board we
experienced data transfer rates of about 250 KiB/s instead of 10 MiB/s
with previous cards from the same vendor (both 4 GB/class 10). By
increasing status register polling rate from 1 kHz to 1 MHz we were
able to reach the original transfer rates again. With the old cards
we now even reach about 16 MiB/s.
Signed-off-by: Tobias Doerffel <tobias.doerffel@ed-chemnitz.de>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
The Allwinner A64 SoC is used in the Pine64. This patch adds
all bits necessary to compile U-Boot for it running in AArch64
mode.
Unfortunately SPL is not ready yet due to legal problems, so
we need to boot using the binary boot0 for now.
Signed-off-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>
[agraf: remove SPL code, move to AArch64]
Signed-off-by: Alexander Graf <agraf@suse.de>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Some parts of the sunxi code cast explicitly between u32 values and pointers.
This is not a problem in practice, because all 64bit SoCs today only use the
lower 32 bits for their phyical address space. But we need to make sure that
the compiler is sure this is not an accident as well.
Signed-off-by: Alexander Graf <agraf@suse.de>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
This will allow the implementation to make use of data in the block_dev
structure beyond the base device number. This will be useful so that eMMC
block devices can encompass the HW partition ID rather than treating this
out-of-band. Equally, the existence of the priv field is crying out for
this patch to exist.
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
In recent allwinner kernel sources the mmc/sdio clk-delay settings have
been slightly tweaked, and for sun9i they are completely different then
what we are using.
This commit brings us in sync with what allwinner does, fixing problems
accessing sdcards on some A33 devices (and likely others).
For pre sun9i hardware this makes the following changes:
-At 400Khz change the sample delay from 7 to 0 (first introduced in A31 sdk)
-At 50 Mhz change the sample delay from 5 to 4 (first introduced in A23 sdk)
-Above 50 MHz change the out delay from 2 to 1 (first introduced in A20 sdk)
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Ian Campbell <ijc@hellion.org.uk>
Originally a timeout value of 2 seconds was used regardless of the size
of data to be transfered. This prevented slow devices from working
correctly while there was no much gain for faster devices, e.g. it takes
3708ms for a transfer of uImage of size 1899008 bytes.
Signed-off-by: Yousong Zhou <yszhou4tech@gmail.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
This patch extracts checking for valid SD card "eGON.BT0" signature from
`board_mmc_init` into function `sunxi_mmc_has_egon_boot_signature`.
Buffer for mmc sector is allocated and freed at runtime. `panic` is
triggered on malloc failure.
Signed-off-by: Daniel Kochmański <dkochmanski@turtle-solutions.eu>
CC: Roy Spliet <r.spliet@ultimaker.com>
Cc: Ian Campbell <ijc@hellion.org.uk>
[hdegoede@redhat.com: Small bugfix to make it work for devs other then mmc0]
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
