The commit 0252924ac6 ("mmc: dw_mmc: Extract FIFO data transfer into a
separate routine") unintentionally changed behavior of the FIFO data
transfer routine.
When data is read and size reaches 0 the original loop would wait on
DWMCI_INTMSK_DTO or timeout. The remaining size to read and buf position
is no longer tracked across dwmci_data_transfer_fifo() calls and because
of this an extra call to fifo() and dwmci_fifo_ready() may now trigger a
FIFO underflow timeout error and slows down FIFO reading.
Buswidth = 4, clock: 50000000
Sending CMD16
Sending CMD17
dwmci_fifo_ready: FIFO underflow timeout
Sending CMD16
Sending CMD18
dwmci_fifo_ready: FIFO underflow timeout
Sending CMD12
## Checking hash(es) for config config-1 ... OK
This reverts commit 0252924ac6 to restore
the old working behavior.
Fixes: 0252924ac6 ("mmc: dw_mmc: Extract FIFO data transfer into a separate routine")
Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Tested-by: Quentin Schulz <quentin.schulz@cherry.de> # RK3588 Tiger
Reviewed-by: Simon Glass <sjg@chromium.org>
Fix most of checkpatch warnings and other obvious style issues.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Replace fifoth_val property with its fifo-depth counterpart in all DW
MMC drivers. fifo-depth is a common property used in upstream Linux
kernel. The FIFOTH register value will be calculated using fifo-depth
value in DW MMC core (dw_mmc.c). This change reduces code duplication in
platform drivers, and pulls common FIFOTH register value calculation
into core dw_mmc driver where it belongs.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Some DW MMC blocks (e.g. those on modern Exynos chips) support 64-bit
DMA addressing mode. 64-bit DW MMC variants differ from their 32-bit
counterparts:
- the register layout is a bit different (because there are additional
IDMAC registers present for storing upper part of 64-bit addresses)
- DMA descriptor structure is bigger and different from 32-bit one
Introduce all necessary changes to enable support for 64-bit DMA capable
DW MMC blocks. Next changes were made:
1. Check which DMA address mode is supported in current IP-core
version. HCON register (bit 27) indicates whether it's 32-bit or
64-bit addressing. Add boolean .dma_64bit_address field to struct
dwmci_host and store the result there. dwmci_init_dma() function is
introduced for doing so, which is called on driver's init.
2. Add 64-bit DMA descriptor (struct dwmci_idmac64) and use it in
dwmci_prepare_desc() in case if .dma_64bit_address field is true.
A new dwmci_set_idma_desc64() function was added for populating that
descriptor.
3. Add registers for 64-bit DMA capable blocks. To make the access to
IDMAC registers universal between 32-bit / 64-bit cases, a new
struct dwmci_idmac_regs (and corresponding host->regs field) was
introduced, which abstracts the hardware by being set to
appropriate offset constants on init. All direct calls to IDMAC
registers were correspondingly replaced by accessing host->regs.
4. Allocate and use 64-bit DMA descriptors buffer in case when IDMAC
is 64-bit capable. Extract all the code (except for the IDMAC
descriptors buffer allocation) from dwmci_send_cmd() to
dwmci_send_cmd_common(), so that it's possible to keep IDMAC
buffer (either 32-bit or 64-bit) on stack during send_cmd routine.
The insights for this implementation were taken from Linux kernel DW MMC
driver.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Prepare for adding 64-bit IDMAC descriptors by renaming current 32-bit
descriptor and its fields accordingly. While at it, make use of
virt_to_phys() to make it more obvious in which places the physical
addresses have to be used.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Make dwmci_prepare_data() function easier to read by extracting the
preparation of IDMAC descriptor into a dedicated function.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Make dwmci_send_cmd() easier to read by moving the DMA transfer handling
code into a dedicated function.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
FIFO data transfer is implemented as quite a massive chunk of code.
Extract it into a dedicated function to make dwmci_data_transfer()
easier to read and reduce the indentation level of the code.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Extract the clock divider update into dwmci_update_div() function. It's
a procedure recommended in TRM, so it's better to keep it in a dedicated
function to make the code clearer.
While at it also extract the clock control code into a separate routine
to avoid code duplication in dwmci_setup_bus().
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Move FIFO threshold initialization into a separate function to make
dwmci_init() more readable.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Waiting for data busy is a logically separate operation and should be
implemented as a separate routine. Follow Linux kernel example and
extract it from dwmci_send_cmd(). This way it doesn't clutter
dwmci_send_cmd() function, and can be reused later in other cases.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
struct idmac is only used in dw_mmc.c, so move it there from dwmmc.h to
avoid cluttering the interface in the header.
No functional change.
Signed-off-by: Sam Protsenko <semen.protsenko@linaro.org>
Reviewed-by: Quentin Schulz <quentin.schulz@cherry.de>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
As part of bringing the master branch back in to next, we need to allow
for all of these changes to exist here.
Reported-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Tom Rini <trini@konsulko.com>
When bringing in the series 'arm: dts: am62-beagleplay: Fix Beagleplay
Ethernet"' I failed to notice that b4 noticed it was based on next and
so took that as the base commit and merged that part of next to master.
This reverts commit c8ffd1356d, reversing
changes made to 2ee6f3a5f7.
Reported-by: Jonas Karlman <jonas@kwiboo.se>
Signed-off-by: Tom Rini <trini@konsulko.com>
Remove <common.h> from this driver directory and when needed
add missing include files directly.
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Peter Robinson <pbrobinson@gmail.com>
Signed-off-by: Tom Rini <trini@konsulko.com>
With the commit 4fcba5d556 ("regulator: implement basic reference
counter"), keeping regulator enablement in balance become more important.
Disable vqmmc-supply before signal voltage is changed to keep regulator
enable counter in balance.
Signed-off-by: Jonas Karlman <jonas@kwiboo.se>
Unconditionally clearing DTO when RXDR is set leads to spurious timeouts
in FIFO mode transfers if events occur in the following order:
mask = dwmci_readl(host, DWMCI_RINTSTS);
// Hardware asserts DWMCI_INTMSK_DTO here
dwmci_writel(host, DWMCI_RINTSTS, DWMCI_INTMSK_DTO);
if (mask & DWMCI_INTMSK_DTO) {
// Unreachable as DTO is cleared without being handled!
return 0;
}
Only clear interrupts that we have seen and are handling so that DTO is
not missed.
Signed-off-by: John Keeping <john@metanate.com>
Tested-by: Jerome Forissier <jerome.forissier@linaro.org> (Rock PI 4B)
Tested-by: Quentin Schulz <quentin.schulz@theobroma-systems.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
When failing to send a command because the hardware is busy, return
EBUSY to indicate the cause instead of just -1.
Signed-off-by: John Keeping <john@metanate.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
The data read is not working when using FIFO mode.
From DesignWare databook, when a Data_Transfer_Over (DTO) interrupt is
received, the software should read the remaining data from FIFO.
Add DTO interrupt checking on data read path and clear interrupts before
start reading from FIFO. So, it doesn't clear the next pending
interrupts unintentionally after read from FIFO.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Change 'clksel' callback function to allow the code to return a
status.
This patch is a preparation for enabling Arm-Trusted-Firmware (ATF)
in Intel SoC FPGA. This patch does not change functionality.
When using Arm-Trusted-Firmware (ATF) in Intel SoC FPGA, the MMC clock
related register is secure register which is required to be written
via SMC/PCSI call. It is possible that U-Boot fail to write the
register if there is unexpected error between U-Boot and ATF.
As a result, there maybe signal integrity on MMC connection due to
clock. So, the code should reports error to user when 'clksel' fail.
Signed-off-by: Siew Chin Lim <elly.siew.chin.lim@intel.com>
Move this header out of the common header. Network support is used in
quite a few places but it still does not warrant blanket inclusion.
Note that this net.h header itself has quite a lot in it. It could be
split into the driver-mode support, functions, structures, checksumming,
etc.
Signed-off-by: Simon Glass <sjg@chromium.org>
These functions are CPU-related and do not use driver model. Move them to
cpu_func.h
Signed-off-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Daniel Schwierzeck <daniel.schwierzeck@gmail.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
There are two cases not been considered:
- use uint for timeout, it will overflow when size bigger than 512KB for
it *8*1000 at the beginning, but we may use size up to 32MB; The
'timeout' will overflow if size bigger than 51.2MB after this fix, which
should be enough for U-Boot;
- The timeout is using clock speed for data rate, but the device may not
have such high speed, eg. clock is 52MHz while the device write speed may
be less than 10MB/s, and we may use up to 150MHz clock.
Fix them in this patch, the max timeout is about 6500 when size is 32MB
after fix.
Signed-off-by: Kever Yang <kever.yang@rock-chips.com>
Fixes the microSD slot on the ASUS C201.
Signed-off-by: Urja Rannikko <urjaman@gmail.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Peng Fan <peng.fan@nxp.com>
Reviewed-by: Peng Fan <peng.fan@nxp.com<mailto:peng.fan@nxp.com>>
The driver currently calculates the end address of cache flush operation
for the DMA descriptors by adding cacheline size to the start address of
the last DMA descriptor. This is not safe, as the cacheline size may be,
in some unlikely cases, smaller than the DMA descriptor size. Replace the
addition with roundup() applied on the end address of the last DMA
descriptor to round it up to the nearest cacheline size multiple.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Jaehoon Chung <jh80.chung@samsung.com>
Cc: Simon Glass <sjg@chromium.org>
The bounce_buffer_start() can return -ENOMEM in case memory allocation
failed. However, in that case, the bounce buffer address is the same as
the possibly unaligned input address, and the cache maintenance operations
were not applied to this address. This could cause subtle problems. Add
handling for the bounce_buffer_start() return value to prevent such a
problem from happening.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Jaehoon Chung <jh80.chung@samsung.com>
Cc: Simon Glass <sjg@chromium.org>
The current 4-minute data transfer timeout is misleading and broken.
Instead of such a long wait, calculate the timeout duration based on
the length of the data transfer. The current formula is the transfer
length in bits, divided by a multiplication of bus frequency in Hz,
bus width, DDR mode and converted the mSec. The value is bounded from
the bottom to 1000 mSec.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Jaehoon Chung <jh80.chung@samsung.com>
Cc: Simon Glass <sjg@chromium.org>
Poll for iDMAC TX/RX interrupt before disable DMA.
This to prevent disable DMA before data is transfer
completed.
Signed-off-by: Ley Foon Tan <ley.foon.tan@intel.com>
This patch adds check for command response CRC failure. The driver
is currently ignoring CRC check failure on command resposes which
have CRC atteched to it, which can be potentially dangerous. Even
more grueling problem happens when the command response is followed
by data transfer though, as in that case, the dwmci_data_transfer()
function will spin until it reaches the 240s timeout.
Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Heiko Stuebner <heiko@sntech.de>
Cc: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
While trying to enable the dw_mmc on rk3188 I managed to confuse
and hang the dw_mmc controller into not delivering further data.
The fifo state never became ready and the driver was iterating in
the while loop reading 0-byte packets forever.
So inspired by how other implementations handle this, check the fifo-
state beforhand and add a timeout to catch any glaring fifo issues
without hanging uboot altogether.
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Reviewed-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Acked-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.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>
All boards which use DM_MMC have now been converted to use DM_MMC_OPS.
Drop the option and good riddance.
Signed-off-by: Simon Glass <sjg@chromium.org>
Some arguments don't need to pass to dwmci_setup_cfg.
They are already included in dwmci_host structure.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
When DTO interrupt occurred, there are any remaining data still in FIFO
due to RX FIFO threshold is larger than remaining data. It also
causes that dwmmc didn't trigger RXDR interrupt, so is TX.
It's responsibility of driver to read remaining bytes on seeing DTO
interrupt.
Signed-off-by: Jacob Chen <jacob2.chen@rock-chips.com>
Signed-off-by: Ziyuan Xu <xzy.xu@rock-chips.com>
This patch fixes data starvation by host timeout(HTO) error interrupt
which occurred under FIFO mode transfer on rk3036 board.
The former implement, the actual bytes were transmitted may be less than
should be. The size will still subtract value of len in case of there is
no receive/transmit FIFO data request interrupt.
Signed-off-by: Ziyuan Xu <xzy.xu@rock-chips.com>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
The former implement, dw_mmc will push and pop the redundant data to
FIFO, we should transfer it according to the real size.
Signed-off-by: Ziyuan Xu <xzy.xu@rock-chips.com>
Acked-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Reviewed-by: Shawn Lin <shawn.lin@rock-chips.com>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.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>
<asm-generic/errno.h> is already included in <errno.h>.
It can use <errno.h> instead of <asm-generic/errno.h>
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
It's no need to speed 10 seconds to wait the mmc device out from busy
status. 500 milliseconds enough.
Signed-off-by: Ziyuan Xu <xzy.xu@rock-chips.com>
Reviewed-by: Jaehoon Chung <jh80.chung@samsung.com>
Tested-by: Jaehoon Chung <jh80.chung@samsung.com>
There are no places to call these functions.
It should be used the callback function.
Then it can be used as static functions.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
These condition checking are wrong.
Original Author's intention might be "&" instead of "&&".
It can know whether receive or transmit data request with
BIT[4]/BIT[5] of RINTSTS register.
Signed-off-by: Jaehoon Chung <jh80.chung@samsung.com>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
When enable dcache on HiKey, we're running into MMC command timeouts
because our retry loop is now faster than the eMMC (or an external SD
card) can answer.
Increase the retry count to the same as the timeout value for status
reports.
The real fix is obviously to not base this whole thing on a cycle counter
but on real wall time, but that would be slightly more intrusive.
Signed-off-by: Alexander Graf <agraf@suse.de>
some soc(rk3036 etc) use dw_mmc but do not have internal dma,
so we implement fifo mode to read and write data.
Signed-off-by: Lin Huang <hl@rock-chips.com>
Acked-by: Simon Glass <sjg@chromium.org>
