While implementing SDIO/MMC SPL booting for the Marvell Armada 38x, the
following problem occured. The SPL runs in internal SRAM which is
the L2 cache locked to memory. When the MMC buffers now are located
on the stack (or bss), the SDIO controller (SDHCI) can't write into
this L2 cache memory.
This patch introduces a method to use a fixed buffer that will be
used for all transfers by defining CONFIG_FIXED_SDHCI_ALIGNED_BUFFER.
This way, the board can use this buffer address located in SDRAM
for all transfers. This solves this SPL problem on the A38x and
should only be used in the SPL U-Boot version.
Tested for SPL booting on Marvell Armada 38x DB-88F6820-GP board.
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Pantelis Antoniou <panto@antoniou-consulting.com>
Cc: Luka Perkov <luka.perkov@sartura.hr>
The loop counter based timeout detection does not work on the Armada
38x based board (DB-88F6820-GP). At least with dcache enabled a
timeout is detected. Without dcache enabled, the timeout does not
occur. Increasing the loop counter solves this issue. But a better
solution is to use a timer based timeout detection instead. This
patch now implements this timer based detection.
Signed-off-by: Stefan Roese <sr@denx.de>
Cc: Pantelis Antoniou <panto@antoniou-consulting.com>
Cc: Luka Perkov <luka.perkov@sartura.hr>
Timeout interrupt also work for response busy command(R1b) like
cmd38/cmd6. So need to set it accordingly. Current code only
set timeout for data command.
Signed-off-by: Kevin Liu <kliu5@marvell.com>
Signed-off-by: Rob Herring <robh@kernel.org>
Cc: Pantelis Antoniou <panto@antoniou-consulting.com>
Change addresses to unsigned long to be compatible with 64-bit builds.
Regardless of fixing warnings, the device is still only 32-bit capable.
Signed-off-by: Rob Herring <robh@kernel.org>
Cc: Pantelis Antoniou <panto@antoniou-consulting.com>
SDHCI_HOST_CONTROL is a byte-sized register, so don't write to it
as if it were a long, as that would result in clobbering the three
registers following.
Signed-off-by: Matt Reimer <mreimer@sdgsystems.com>
Fix bus width switching from 8-bit mode down to 4-bit or 1-bit modes on
Samsung SoCs using SDHCI_QUIRK_USE_WIDE8. These SoCs report controller
version 2.0 yet they support 8-bit bus widths. If 8-bit mode was
previously enabled and then an operation like "mmc dev" caused a switch
back down to 4-bit or 1-bit mode, WIDE8 was left set, causing failures.
This problem was manifested by "mmc dev" timing out.
Signed-off-by: Matt Reimer <mreimer@sdgsystems.com>
The way that struct mmc was implemented was a bit of a mess;
configuration and internal state all jumbled up in a single structure.
On top of that the way initialization is done with mmc_register leads
to a lot of duplicated code in drivers.
Typically the initialization got something like this in every driver.
struct mmc *mmc = malloc(sizeof(struct mmc));
memset(mmc, 0, sizeof(struct mmc);
/* fill in fields of mmc struct */
/* store private data pointer */
mmc_register(mmc);
By using the new mmc_create call one just passes an mmc config struct
and an optional private data pointer like this:
struct mmc = mmc_create(&cfg, priv);
All in tree drivers have been updated to the new form, and expect
mmc_register to go away before long.
Changes since v1:
* Use calloc instead of manually calling memset.
* Mark mmc_register as deprecated.
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Using an array is pointless; even more pointless (and scary) is using
sprintf to fill it without a format string.
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Remove the in-structure ops and put them in mmc_ops with
a constant pointer to it.
This makes the mmc structure smaller as well as conserving
code space (in theory).
All in-tree drivers are converted as well; this is done in a
single patch in order to not break git bisect.
Changes since V1:
Fix compilation b0rked issue on omap platforms where OMAP_GPIO was
not set.
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Fixup prints to show where the print is done from, and
a few minor formatting/grammar issues.
Signed-off-by: Darwin Rambo <drambo@broadcom.com>
Acked-by: Pantelis Antoniou <panto@antoniou-consulting.com>
Old command timeout value was too small and it caused I/O errors which
led to uncompleted read/write/erase operations and filesystem errors.
Timeout adaptation fixes this issue.
Changes in sdhci_send_command() function:
- change timeout variable to static
- increase default command timeout to 100 ms
- add definition of max command timeout value,
which can be redefined in each board config file
- wait for card ready state for max defined time
if it doesn't exceed defined maximum or return COMM_ERR
Once successfully increased timeout value will be used in next function
call. This fix was tested on Goni, Trats, Trats2 boards by testing UMS
on MMC storage.
Changes v2:
- move global variable cmd_timeout into function sdhci_send_command()
- change condition "==" to ">=" when comparing time with timeout
- print information about timeout increasing and card busy timeout
Signed-off-by: Przemyslaw Marczak <p.marczak@samsung.com>
Cc: Pantelis Antoniou <panto@antoniou-consulting.com>
The patch fixes the improper read and write of sdhci
host control register for sdma transfer.
The problem comes when reading and writing 1 byte long
host control register with the sdhci_readl() and
sdhci_writel(). The misuse of these functions overwrite
the value of the next registers which are in 4 bytes boundary.
This patch replaces four byte register read/write functions
with one byte read/write ones. Beside, it eliminates
unnecessary bit operation. i.e. or-ing zero against a variable.
Signed-off-by: Juhyun (Justin) Oh <Juhyun_Oh@sigmadesigns.com>
Samsung SoC is supported the WIDE8, even if Controller version is v2.0.
So add the SDHCI_QUIRK_USE_WIDE8 for Samsung-SoC.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Pantelis Antoniou <panto@antoniou-consulting.com>
CAP register don't have any information for 8-bit buswidth support
on 2.0 sdhci spec, only from 3.0 onwards bit[18] got this information.
Due to this misassignment in sdhci, mmc is setting 8-bit buswidth using
mmc_set_bus_width even if controller doesn't support.
Below change has code information.
"mmc: Properly determine maximum supported bus width"
(sha1: 7798f6dbd5)
Bug log: <mmc plus and emmc cards)
-------
zynq-uboot> mmcinfo
Error detected in status(0x208100)!
Device: zynq_sdhci
Manufacturer ID: fe
.....
So enable 8-bit support only for 3.0 spec using CAP and for below 3.0
assign mmc->host_caps = MMC_MODE_8BIT on respective platform driver
if host have a support.
Signed-off-by: Jagannadha Sutradharudu Teki <jaganna@xilinx.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
This patch changes sdhci_init()'s behavior to NOT enable all interrupt
sources by default. Moreover interrupt signaling has been disabled.
This patch do not enable interrupts which aren't served in u-boot
(they are defined at sdhci.h but NOT used elsewhere):
- SDHCI_INT_CARD_INSERT, SDHCI_INT_CARD_REMOVE, SDHCI_BUS_POWER,
SDHCI_INT_CARD_REMOVE, SDHCI_INT_CARD_INT
Special care shall be put on SDHCI_INT_CARD_INT, which indicates
interrupt generated by SD card.
According to "SD Host Controller Simplified Spec. ver 3.00" when bit 8
(Card Interrupt Status Enable) at "Normal Interrupt Status Enable
Register" (offset 0x34) is set, the card interrupt detection is started.
Then eMMC card may cause the SD controller to set this bit and then this
interrupt is passed to booted OS and might cause kernel crash.
To sum up:
- Only enable interrupts, which are served at u-boot
- This cleanup as a side effect fixes SDHCI's CARD INTERRUPT problem at
Linux kernel (versions 3.6+, sdhci controller)
- Keep masked bits at "Normal Interrupt Signal Enable Register" (0x38h)
Signed-off-by: Lukasz Majewski <l.majewski@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Cc: Lei Wen <leiwen@marvell.com>
Cc: Andy Fleming <afleming@freescale.com>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Bring in the code from Linux kernel.
Added to Linux kernel by:
commit e08c1694d9e2138204f2b79b73f0f159074ce2f5
Author: Andres Salomon <dilinger@queued.net>
Date: Fri Jul 4 10:00:03 2008 -0700
Some HW balks when writing both voltage setting and power up at the same
time to SDHCI_POWER_CONTROL register.
Signed-off-by: Rommel G Custodio <sessyargc@gmail.com>
CC: Andy Fleming <afleming@freescale.com>
v2: fix attribution and SOB
Signed-off-by: Andy Fleming <afleming@freescale.com>
MMC host controller requires a delay between every sdhci_send_cmd()
execution. In s5p_mmc driver (s5p_sdhci replaces this driver), a delay
of 1000us was provided after every mmc_send_cmd() call. Adding a quirk
in current sdhci driver to replicate the behaviour.
Without this delay, MMC initialization on Origen board fails with
following error messages.
Timeout for status update!
mmc fail to send stop cmd
Signed-off-by: Tushar Behera <tushar.behera@linaro.org>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
In host-control register, DMA select bit field is present.
BUt in sdhci.c, didn't select for DMA.
if set CONFIG_MMC_SDMA, we need to set SDMA-select bit.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
Timeout value is tunable.
When run read/write operation, sometime returned the timeout error.
Because the timeout value is too short.
So increased the enough timeout value.
(This timeout value is used to prevent the infinite loop.)
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
Samsung SoC use the cmu control to set clock.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
Some boards have no Card Detect wired. In that case, set the CD test
bits in the standard interface.
Signed-off-by: Joe Hershberger <joe.hershberger@ni.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
When setting up the clocks in the sdhci driver, the "spec version"
must be masked off. Otherwise any time the vendor version is not 0,
the check will allways assume the interface is version 3. This breaks
when the interface is actually version 1 or 2.
Signed-off-by: Joe Hershberger <joe.hershberger@ni.com>
Signed-off-by: Andy Fleming <afleming@freescale.com>
To support the Samsung-SoC, added the basically functions.
Samsung-SoC didn't used the SDHCI_CTRL_HISPD.
And added set_control_reg callback for s3c64xx.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Acked-by: Lei Wen<leiwen@marvell.com>
When response type is R1b, mask value is added the SDHCI_INT_DAT_END.
but in while(), didn't check that flag.
So sdhci controller didn't work fine.
CMD6 didn't always complete.
So add the quirks for broken r1b response
and add the timeout value to prevent the infinite loop.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Kyungmin Park <kyungmin.park@samsung.com>
Acked-by: Lei Wen<leiwen@marvell.com>
Check for card detect each time an MMC/SD device is initialized. If card
detection is not implemented, this code behaves as before and continues
assuming a card is present. If no card is detected, has_init is reset
for the MMC/SD device (to force initialization next time) and an error
is returned.
Signed-off-by: Thierry Reding <thierry.reding@avionic-design.de>
Tested-by: Jason Liu <jason.hui@linaro.org>
SDHCI spec need to reset the sdma base address while the software
try to accorss the 512k bytes address boundary. When meet such
accross behavior, sdhci controller would generate a interrupt
automatically, and software need handle this.
Signed-off-by: Lei Wen <leiwen@marvell.com>
Nowdays, there are plenty of mmc driver in uboot adopt the sd standard
host design, aka as sdhci. It is better to centralize the common logic
together to better maintenance.
Signed-off-by: Lei Wen <leiwen@marvell.com>
Acked-by: Andy Fleming <afleming@freescale.com>
Stefan Roese
Kevin Liu
Rob Herring
Matt Reimer
Jeroen Hofstee
Pantelis Antoniou
Darwin Rambo
Przemyslaw Marczak
Juhyun \(Justin\) Oh
Jaehoon Chung
Wolfgang Denk
Jagannadha Sutradharudu Teki
Albert ARIBAUD
Łukasz Majewski
Nikita Kiryanov
Mela Custodio
Tushar Behera
Joe Hershberger
Thierry Reding
Lei Wen