The patch splits DMA preparatory code to dma_setup() callback. The change also
converts transfer_one() to program DMA whenever the transfer is DMA mapped. The
change is a follow up of the converion to use SPI core transfer_one_message().
Since the DMA mapped transfers can be interleaved with PIO ones the DMA related
configuration should respect that.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
DMAEngine has a specific type to be used for bus width. This patch converts the
code to use the values of the specific type when configure DMA transfer.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
All SPI drivers have been converted from legacy suspend/resume callbacks to
dev_pm_ops. So we can finally remove support for legacy PM from the SPI
core.
Since there aren't any special bus specific things to do during
suspend/resume and since the PM core will automatically fallback directly to
using the device's PM ops if no bus PM ops are specified there is no need to
have any special SPI bus PM ops.
Signed-off-by: Lars-Peter Clausen <lars@metafoo.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
The logic of DMA completion is broken now since test_and_clear_bit() never
returns the other bit is set. It means condition are always false and we have
spi_finalize_current_transfer() called per each DMA completion which is wrong.
The patch fixes logic by clearing BUSY bit first and then check for the other
one.
Fixes: 30c8eb52cc (spi: dw-mid: split rx and tx callbacks when DMA)
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
This patch adds DMA capabilities to the spi-qup driver. If DMA channels are
present, the QUP will use DMA instead of block mode for transfers to/from SPI
peripherals for transactions larger than the length of a block.
Signed-off-by: Andy Gross <agross@codeaurora.org>
Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
Reviewed-by: Ivan T. Ivanov <iivanov@mm-sol.com
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch removes a lot of duplicate code since SPI core provides a nice
message handling.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch refactors the code in pump_transfers() to reprogram the registers
immediately when we have a new configuration data. The behaviour is slightly
modified:
- chip is always disabled and reenabled
- CTRL0 is always reprogrammed
This change allows to do a further refactoring and simplier conversion to use
SPI core DMA routines in the future.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The error handling is partially broken since the controller is disabled on
error and is not re-enabled until condition occurs, i.e. mode (poll, PIO/DMA),
chip (cs_change), or speed (clk_div) is changed. In the result of these changes
we will have a predictable state of the SPi controller independently on how
successfull was a previous transfer.
The patch disables interrupts and re-enables the SPI controller wherever it
needs to be done. Thus most of the time the SPI controller is kept enabled. The
runtime PM, when it will be implemented, must take care of the controller
disabling and re-enabling.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Commit fd316941c ("spi/pl022: disable port when unused") introduced a race,
which leads to possible driver lock up (easily reproducible on SMP).
The problem happens in giveback() function where the completion of the transfer
is signalled to SPI subsystem and then the HW SPI controller is disabled. Another
transfer might be setup in between, which brings driver in locked-up state.
Exact event sequence on SMP:
core0 core1
=> pump_transfers()
/* message->state == STATE_DONE */
=> giveback()
=> spi_finalize_current_message()
=> pl022_unprepare_transfer_hardware()
=> pl022_transfer_one_message
=> flush()
=> do_interrupt_dma_transfer()
=> set_up_next_transfer()
/* Enable SSP, turn on interrupts */
writew((readw(SSP_CR1(pl022->virtbase)) |
SSP_CR1_MASK_SSE), SSP_CR1(pl022->virtbase));
...
=> pl022_interrupt_handler()
=> readwriter()
/* disable the SPI/SSP operation */
=> writew((readw(SSP_CR1(pl022->virtbase)) &
(~SSP_CR1_MASK_SSE)), SSP_CR1(pl022->virtbase));
Lockup! SPI controller is disabled and the data will never be received. Whole
SPI subsystem is waiting for transfer ACK and blocked.
So, only signal transfer completion after disabling the controller.
Fixes: fd316941c (spi/pl022: disable port when unused)
Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
PL022 Programmers model explicitely states "do not modify undefined register
bits". Correct the "all enable" interrupt mask so that it only enables defined
ones.
Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
According to PL022 specification, TNF bit states for "Transmit FIFO Not full".
So the logic here is inverted. But "Receive Overrun Interrupt", which is handled
here, is only triggered on Rx errors. So instead of fixing the if statement,
remove the whole message.
Signed-off-by: Alexander Sverdlin <alexander.sverdlin@nokia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This callback would be useful to handle an error that occurs in the generic
implementation of transfer_one_message(). The good candidate for this is to
drain FIFO and / or to terminate DMA transfers when timeout happened.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
When DMA descriptor allocation fails we should not try to assign any fields in
the bad descriptor. The patch adds the necessary checks for that.
Fixes: 7063c0d942 (spi/dw_spi: add DMA support)
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
Removing the udelay between setting and clearing the soft reset bit in
the spfi control register as it is not required.
Signed-off-by: Sifan Naeem <sifan.naeem@imgtec.com>
Reviewed-by: Andrew Bresticker <abrestic@chromium.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
Maximum transfer length supported by SPFI is 65535, this is limited
by the number of bits available in SPFI TSize register to represent
the transfer size.
For transfer requests larger than the maximum supported the driver
will return an invalid argument error.
Signed-off-by: Sifan Naeem <sifan.naeem@imgtec.com>
Reviewed-by: Andrew Bresticker <abrestic@chromium.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
The fourth argument of dmaengine_prep_slave_sg needs to be of the
enumeration type dma_transfer_direction instead of dma_data_direction.
Since the used enumeration values actually stay the same, this is not
an actual issue at runtime.
Signed-off-by: Stefan Agner <stefan@agner.ch>
Signed-off-by: Mark Brown <broonie@kernel.org>
Additionally to the current DMA transfer the PDC allows to set up a next DMA
transfer. This is useful for larger SPI transfers.
The driver currently waits for ENDRX as end of the transfer. But ENDRX is set
when the current DMA transfer is done (RCR = 0), i.e. it doesn't include the
next DMA transfer.
Thus a subsequent SPI transfer could be started although there is currently a
transfer in progress. This can cause invalid accesses to the SPI slave devices
and to SPI transfer errors.
This issue has been observed on a hardware with a M25P128 SPI NOR flash.
So instead of ENDRX we should wait for RXBUFF. This flag is set if there is
no more DMA transfer in progress (RCR = RNCR = 0).
Signed-off-by: Torsten Fleischer <torfl6749@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
The commit d297933cc7 (spi: dw: Fix detecting FIFO depth) tries to fix the
logic of the FIFO detection based on the description on the comments. However,
there is a slight difference between numbers in TX Level and TX FIFO size.
So, by specification the FIFO size would be in a range 2-256 bytes. From TX
Level prospective it means we can set threshold in the range 0-(FIFO size - 1)
bytes. Hence there are currently two issues:
a) FIFO size 2 bytes is actually skipped since TX Level is 1 bit and could be
either 0 or 1 byte;
b) FIFO size is incorrectly decreased by 1 which already done by meaning of
TX Level register.
This patch fixes it eventually right.
Fixes: d297933cc7 (spi: dw: Fix detecting FIFO depth)
Reviewed-by: Axel Lin <axel.lin@ingics.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
Instead of an additional reading from the register let's update it even if the
value is kept the same.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
This patch does the following changes:
a) the calculation of clk_div is simplified to oneliner;
b) chip->clk_div is updated if clk_div is not zero, therefore the condition is
simplified by using chip->clk_div in both cases;
c) while here, the redundant parentheses are removed.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
There is no need to have an additional variable to get a TX level. The patch
refactors this piece of code.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
There is no sense to keep a member assignment in the internal structure inside
the condition which reprograms HW. It makes code readability better if kept
outside of the condition.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
The commit d58cf5ff65 brought a second controller to the list of supported
devices and changed a number of the chip selects. Besides the previous number
was wrong anyway the mentioned patch makes it wrong again meanwhile has a
proper numbers in the commit message. Indeed, SPI1 has 5 bits and SPI2 has 2
bits, but it does not mean to have power of two of this bits as a possible
number of the chip selects. So, this patch fixes it eventually.
Fixes: d58cf5ff65 (spi: dw-pci: describe Intel MID controllers better)
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
This patch changes the way space is reserved in spidev's pre-allocated
TX and RX bounce buffers to avoid wasting space in the buffers for an
SPI message consisting of multiple, half-duplex transfers in different
directions.
Background:
spidev data structures have separate, pre-allocated TX and RX bounce
buffers (`spidev->tx_buffer` and `spidev->rx_buffer`) of fixed size
(`bufsiz`). The `SPI_IOC_MESSAGE(N)` ioctl processing uses a kernel
copy of the N `struct spi_ioc_transfer` elements copied from the
userspace ioctl arg pointer. In these elements: `.len` is the length of
transfer in bytes; `.rx_buf` is either a userspace pointer to a buffer
to copy the RX data to or is set to 0 to discard the data; and `.tx_buf`
is either a userspace pointer to TX data supplied by the user or is set
to 0 to transmit zeros for this transfer.
`spidev_message()` uses the array of N `struct spi_ioc_transfer`
elements to construct a kernel SPI message consisting of a `struct
spi_message` containing a linked list (allocated as an array) of N
`struct spi_transfer` elements. This involves iterating through the
`struct spi_ioc_transfer` and `struct spi_transfer` elements (variables
`u_tmp` and `k_tmp` respectively). Before the first iteration,
variables `tx_buf` and `rx_buf` point to the start of the TX and RX
bounce buffers `spidev->tx_buffer` and `spidev->rx_buffer` and variable
`total` is set to 0. These variables keep track of the next available
space in the bounce buffers and the total length of the SPI message.
Each iteration checks that there is enough room left in the buffers for
the transfer. If `u_tmp->rx_buf` is non-zero, `k_tmp->rx_buf` is set to
`rx_buf`, otherwise it remains set to NULL. If `u_tmp->tx_buf` is
non-zero, `k_tmp->tx_buf` is set to `tx_buf` and the userspace TX data
copied there, otherwise it remains set to NULL. The variables `total`,
`rx_buf` and `tx_buf` are advanced by the length of the transfer.
The "problem":
While iterating through the transfers, the local bounce buffer "free
space" pointer variables `tx_buf` and `rx_buf` are always advanced by
the length of the transfer. If `u_tmp->rx_buf` is 0 (so `k_tmp->rx_buf`
is NULL), then `rx_buf` is advanced unnecessarily and that part of
`spidev->rx_buffer` is wasted. Similarly, if `u_tmp->tx_buf` is 0 (so
`k_tmp->tx_buf` is NULL), part of `spidev->tx_buffer` is wasted.
What this patch does:
To avoid wasting space unnecessarily in the RX bounce buffer, only
advance `rx_buf` by the transfer length if `u_tmp->rx_buf` is non-zero.
Similarly, to avoid wasting space unnecessarily in the TX bounce buffer,
only advance `tx_buf` if `u_tmp->tx_buf is non-zero. To avoid pointer
subtraction, use new variables `rx_total` and `tx_total` to keep track
of the amount of space allocated in each of the bounce buffers. If
these exceed the available space, a `-EMSGSIZE` error will be returned.
Limit the total length of the transfers (tracked by variable `total`) to
`INT_MAX` instead of `bufsiz`, returning an `-EMSGSIZE` error if
exceeded. The total length is returned by `spidev_message()` on success
and we want that to be non-negative. The message size limits for the
`SPI_IOC_MESSAGE(N)` ioctl are now as follows:
(a) total length of transfers is <= INTMAX;
(b) total length of transfers with non-NULL rx_buf is <= bufsiz;
(c) total length of transfers with non-NULL tx_buf is <= bufsiz.
Some transfers may have NULL rx_buf and NULL tx_buf.
If the transfer is completed successfully by the SPI core,
`spidev_message()` iterates through the transfers to copy any RX data
from the bounce buffer back to userspace on those transfers where
`u_tmp->rx_buf` is non-zero. The variable `rx_buf` is again used to
keep track of the corresponding positions in the bounce buffer. Now it
is only advanced for those transfers that use the RX bounce buffer.
Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Signed-off-by: Mark Brown <broonie@kernel.org>
The Exynos 7 arm64 support now allows the S3C64xx SPI driver to be
compiled into an ARM64 kernel, so the cast from the [rt]x_dmach int
variable to a void* in this driver now triggers a warning.
Add a long cast to silence the compiler.
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Data corruption is seen while reading/writing large data from/to qspi
device because the data register is over written or read before data
is ready which is denoted by busy bit in status register. SO adding
a busy bit check before writing/reading data to/from qspi device.
Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Pull kconfig updates from Michal Marek:
"Yann E Morin was supposed to take over kconfig maintainership, but
this hasn't happened. So I'm sending a few kconfig patches that I
collected:
- Fix for missing va_end in kconfig
- merge_config.sh displays used if given too few arguments
- s/boolean/bool/ in Kconfig files for consistency, with the plan to
only support bool in the future"
* 'kconfig' of git://git.kernel.org/pub/scm/linux/kernel/git/mmarek/kbuild:
kconfig: use va_end to match corresponding va_start
merge_config.sh: Display usage if given too few arguments
kconfig: use bool instead of boolean for type definition attributes
Use the normal {} instead of a macro to terminate an array.
Remove the macro too.
Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The driver currently only supports a single device per I2C bus since it uses
the I2C bus number to set the SPI bus number. This makes it impossible to
connect more than one chip to a single I2C bus.
We don't want to use dynamic bus numbers unconditionally since this would
result in every instantiation getting a different bus number starting with
65,535 counting down unless devicetree is configured. If devicetree is
configured, however, the SPI bus number is obtained from devicetree
data. So we can use dynamic SPI bus numbers in this case.
Reported-and-Tested-by: Marco Menchise <marco.menchise@gmail.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Mark Brown <broonie@kernel.org>
Converting milliseconds to jiffies by "val * HZ / 1000" is technically
ok but msecs_to_jiffies(val) is the cleaner solution and handles all
corner cases correctly.
This is only an API consolidation and should make things more readable
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
return type of wait_for_completion_timeout is unsigned long not int, this
patch uses the return value of wait_for_completion_timeout in the condition
directly rather than adding a additional appropriately typed variable.
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
return type of wait_for_completion_timeout is unsigned long not int, this
patch uses the return value of wait_for_completion_timeout in the condition
directly rather than assigning it to an incorrect type variable.
Signed-off-by: Nicholas Mc Guire <hofrat@osadl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
