platform/kernel/linux-starfive.git
5 years agospi: tegra114: flush fifos
Sowjanya Komatineni [Wed, 27 Mar 2019 05:56:28 +0000 (22:56 -0700)]
spi: tegra114: flush fifos

Fixes: Flush TX and RX FIFOs before start of new transfer and on FIFO
overflow or underrun errors.

Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: tegra114: terminate dma and reset on transfer timeout
Sowjanya Komatineni [Wed, 27 Mar 2019 05:56:27 +0000 (22:56 -0700)]
spi: tegra114: terminate dma and reset on transfer timeout

Fixes: terminate DMA and perform controller reset on transfer timeout
to clear the FIFO's and errors.

Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: tegra114: fix for unpacked mode transfers
Sowjanya Komatineni [Wed, 27 Mar 2019 05:56:24 +0000 (22:56 -0700)]
spi: tegra114: fix for unpacked mode transfers

Fixes: computation of actual bytes to fill/receive in/from FIFO in unpacked
mode when transfer length is not a multiple of requested bits per word.

unpacked mode transfers fails when the transfer includes partial bytes in
the last word.

Total words to be written/read to/from FIFO is computed based on transfer
length and bits per word. Unpacked mode includes 0 padding bytes for partial
words to align with bits per word and these extra bytes are also accounted
for calculating bytes left to transfer in the current driver.

This causes extra bytes access of tx/rx buffers along with buffer index
position crossing actual length where remain_len becomes negative and due to
unsigned type, negative value is a 32 bit representation of signed value
and transferred bytes never meets the actual transfer length resulting in
transfer timeout and a hang.

This patch fixes this with proper computation of the actual bytes to fill in
FIFO during transmit and the actual bytes to read from FIFO during receive
ignoring 0 padded bytes.

Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: tegra114: clear packed bit for unpacked mode
Sowjanya Komatineni [Wed, 27 Mar 2019 05:56:23 +0000 (22:56 -0700)]
spi: tegra114: clear packed bit for unpacked mode

Fixes: Clear packed bit when not using packed mode.

Packed bit is not cleared when not using packed mode. This results
in transfer timeouts for the unpacked mode transfers followed by the
packed mode transfers.

Signed-off-by: Sowjanya Komatineni <skomatineni@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: pic32: fix dma channels termination
Cezary Gapinski [Tue, 26 Mar 2019 21:48:59 +0000 (22:48 +0100)]
spi: pic32: fix dma channels termination

When timeout occurs DMA TX and RX channels should be stopped
instead of stopping RX channel twice time.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: tegra20-slink: change chip select action order
Randolph Maaßen [Tue, 26 Mar 2019 14:30:50 +0000 (15:30 +0100)]
spi: tegra20-slink: change chip select action order

To transfer via SPI the tegra20-slink driver first sets the command
register, which contains the chip select value, and after that the
command2 register, which contains the chip select line. This leads to a
small spike in the chip selct 0 line between the set of the value and
the selection of the chip select line.

This commit changes the order of the register writes so that first the
chip select line is chosen and then the value is set, removing the
spike.

Signed-off-by: Randolph Maaßen <gaireg@gaireg.de>
Reviewed-by: Sowjanya Komatineni <skomatineni@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: lpspi: fix dataloss when SS is inactivated between every words
Clark Wang [Thu, 21 Mar 2019 09:57:12 +0000 (09:57 +0000)]
spi: lpspi: fix dataloss when SS is inactivated between every words

If we don't use CONT to keep SS activated or use DMA mode without
cs-gpio, SS will be inactivated between every words. The word here
means the data sent once which length can be set as 1/2/4 bytes.

In the isr function, we read the FSR_RXCOUNT just behind the
fsl_lpspi_read_rx_fifo. This causes the value of FSR_RXCOUNT cannot
reflect whether there is still data not sent timely. So do this
judgement by FSR_TXCOUNT.

Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi : spi-topcliff-pch: Fix to handle empty DMA buffers
Aditya Pakki [Wed, 13 Mar 2019 16:55:41 +0000 (11:55 -0500)]
spi : spi-topcliff-pch: Fix to handle empty DMA buffers

pch_alloc_dma_buf allocated tx, rx DMA buffers which can fail. Further,
these buffers are used without a check. The patch checks for these
failures and sends the error upstream.

Signed-off-by: Aditya Pakki <pakki001@umn.edu>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: stm32-qspi: avoid memory corruption at low frequency
Ludovic Barre [Fri, 8 Mar 2019 13:12:20 +0000 (14:12 +0100)]
spi: spi-mem: stm32-qspi: avoid memory corruption at low frequency

This patch solves a memory corruption seen at 8 MHz.
To avoid such issue, timeout counter is disabled.

Signed-off-by: Ludovic Barre <ludovic.barre@st.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: rspi: Fix sequencer reset during initialization
Geert Uytterhoeven [Tue, 12 Mar 2019 18:45:13 +0000 (19:45 +0100)]
spi: rspi: Fix sequencer reset during initialization

While the sequencer is reset after each SPI message since commit
880c6d114fd79a69 ("spi: rspi: Add support for Quad and Dual SPI
Transfers on QSPI"), it was never reset for the first message, thus
relying on reset state or bootloader settings.

Fix this by initializing it explicitly during configuration.

Fixes: 0b2182ddac4b8837 ("spi: add support for Renesas RSPI")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: rspi: Fix register initialization while runtime-suspended
Geert Uytterhoeven [Tue, 12 Mar 2019 18:43:31 +0000 (19:43 +0100)]
spi: rspi: Fix register initialization while runtime-suspended

The Renesas RSPI/QSPI driver performs SPI controller register
initialization in its spi_operations.setup() callback, without calling
pm_runtime_get_sync() first, which may cause spurious failures.

So far this went unnoticed, as this SPI controller is typically used
with a single SPI NOR FLASH containing the boot loader:
  1. If the device's module clock is still enabled (left enabled by the
     bootloader, and not yet disabled by the clk_disable_unused() late
     initcall), register initialization succeeds,
  2. If the device's module clock is disabled, register writes don't
     seem to cause lock-ups or crashes.
     Data received in the first SPI message may be corrupted, though.
     Subsequent SPI messages seem to be OK.
     E.g. on r8a7791/koelsch, one bit is lost while receiving the 6th
     byte of the JEDEC ID for the s25fl512s FLASH, corrupting that byte
     and all later bytes.  But until commit a2126b0a010905e5 ("mtd:
     spi-nor: refine Spansion S25FL512S ID"), the 6th byte was not
     considered for FLASH identification.

Fix this by moving all initialization from the .setup() to the
.prepare_message() callback.  The latter is always called after the
device has been runtime-resumed by the SPI core.

This also makes the driver follow the rule that .setup() must not change
global driver state or register values, as that might break a transfer
in progress.

Fixes: 490c97747d5dc77d ("spi: rspi: Add runtime PM support, using spi core auto_runtime_pm")
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: imx: stop buffer overflow in RX FIFO flush
Trent Piepho [Mon, 4 Mar 2019 20:18:49 +0000 (20:18 +0000)]
spi: imx: stop buffer overflow in RX FIFO flush

Commit 71abd29057cb ("spi: imx: Add support for SPI Slave mode") added
an RX FIFO flush before start of a transfer.  In slave mode, the master
may have sent more data than expected and this data will still be in the
RX FIFO at the start of the next transfer, and so needs to be flushed.

However, the code to do the flush was accidentally saving this data into
the previous transfer's RX buffer, clobbering the contents of whatever
followed that buffer.

Change it to empty the FIFO and throw away the data.  Every one of the
RX functions for the different eCSPI versions and modes reads the RX
FIFO data using the same readl() call, so just use that, rather than
using the spi_imx->rx function pointer and making sure all the different
rx functions have a working "throw away" mode.

There is another issue, which affects master mode when switching from
DMA to PIO.  There can be extra data in the RX FIFO which triggers this
flush code, causing memory corruption in the same manner.  I don't know
why this data is unexpectedly in the FIFO.  It's likely there is a
different bug or erratum responsible for that.  But regardless of that,
I think this is proper fix the for bug at hand here.

Fixes: 71abd29057cb ("spi: imx: Add support for SPI Slave mode")
Cc: Jiada Wang <jiada_wang@mentor.com>
Cc: Fabio Estevam <festevam@gmail.com>
Cc: Stefan Agner <stefan@agner.ch>
Cc: Shawn Guo <shawnguo@kernel.org>
Signed-off-by: Trent Piepho <tpiepho@impinj.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: Fix zero length xfer bug
Chris Lesiak [Thu, 7 Mar 2019 20:39:00 +0000 (20:39 +0000)]
spi: Fix zero length xfer bug

This fixes a bug for messages containing both zero length and
unidirectional xfers.

The function spi_map_msg will allocate dummy tx and/or rx buffers
for use with unidirectional transfers when the hardware can only do
a bidirectional transfer.  That dummy buffer will be used in place
of a NULL buffer even when the xfer length is 0.

Then in the function __spi_map_msg, if he hardware can dma,
the zero length xfer will have spi_map_buf called on the dummy
buffer.

Eventually, __sg_alloc_table is called and returns -EINVAL
because nents == 0.

This fix prevents the error by not using the dummy buffer when
the xfer length is zero.

Signed-off-by: Chris Lesiak <chris.lesiak@licor.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sh-msiof: Restrict bits per word to 8/16/24/32 on R-Car Gen2/3
Geert Uytterhoeven [Thu, 28 Feb 2019 11:05:13 +0000 (12:05 +0100)]
spi: sh-msiof: Restrict bits per word to 8/16/24/32 on R-Car Gen2/3

While the MSIOF variants in older SuperH and SH/R-Mobile SoCs support
bits-per-word values in the full range 8..32, the variants present in
R-Car Gen2 and Gen3 SoCs are restricted to 8, 16, 24, or 32.

Obtain the value from family-specific sh_msiof_chipdata to fix this.

Reported-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sifive: Remove redundant dev_err call in sifive_spi_probe()
Wei Yongjun [Fri, 22 Feb 2019 05:46:32 +0000 (05:46 +0000)]
spi: sifive: Remove redundant dev_err call in sifive_spi_probe()

There is a error message within devm_ioremap_resource
already, so remove the dev_err call to avoid redundant
error message.

Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sifive: Remove spi_master_put in sifive_spi_remove()
Wei Yongjun [Fri, 22 Feb 2019 05:46:16 +0000 (05:46 +0000)]
spi: sifive: Remove spi_master_put in sifive_spi_remove()

The call to spi_master_put() in sifive_spi_remove() is redundant since
the master is registered using devm_spi_register_master() and no
reference hold by using spi_master_get() in sifive_spi_remove().

This is detected by Coccinelle semantic patch.

Fixes: 484a9a68d669 ("spi: sifive: Add driver for the SiFive SPI controller")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agoMerge branch 'for-5.0' of https://git.kernel.org/pub/scm/linux/kernel/git/broonie...
Mark Brown [Wed, 20 Feb 2019 17:58:18 +0000 (17:58 +0000)]
Merge branch 'for-5.0' of https://git./linux/kernel/git/broonie/spi into spi-5.1

5 years agospi: pxa2xx: Setup maximum supported DMA transfer length
Andy Shevchenko [Tue, 19 Feb 2019 20:21:28 +0000 (23:21 +0300)]
spi: pxa2xx: Setup maximum supported DMA transfer length

When the commit b6ced294fb61

   ("spi: pxa2xx: Switch to SPI core DMA mapping functionality")

switches to SPI core provided DMA helpers, it missed to setup maximum
supported DMA transfer length for the controller and thus users
mistakenly try to send more data than supported with the following
warning:

  ili9341 spi-PRP0001:01: DMA disabled for transfer length 153600 greater than 65536

Setup maximum supported DMA transfer length in order to make users know
the limit.

Fixes: b6ced294fb61 ("spi: pxa2xx: Switch to SPI core DMA mapping functionality")
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: stable@vger.kernel.org
5 years agospi: sifive: Add driver for the SiFive SPI controller
Yash Shah [Tue, 19 Feb 2019 11:40:07 +0000 (17:10 +0530)]
spi: sifive: Add driver for the SiFive SPI controller

Add driver for the SiFive SPI controller
on the HiFive Unleashed board.

Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
Signed-off-by: Yash Shah <yash.shah@sifive.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sifive: Add DT documentation for SiFive SPI controller
Yash Shah [Tue, 19 Feb 2019 11:40:06 +0000 (17:10 +0530)]
spi: sifive: Add DT documentation for SiFive SPI controller

DT documentation for SPI controller added.

Signed-off-by: Palmer Dabbelt <palmer@sifive.com>
Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
Signed-off-by: Yash Shah <yash.shah@sifive.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sprd: Add a prefix for SPI DMA channel macros
Baolin Wang [Wed, 13 Feb 2019 13:08:16 +0000 (21:08 +0800)]
spi: sprd: Add a prefix for SPI DMA channel macros

Add a prefix for SPI DMA channel macros to avoid namespace conflicts,
and no functional changes.

Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sprd: spi: sprd: Add DMA mode support
Lanqing Liu [Wed, 13 Feb 2019 07:36:11 +0000 (15:36 +0800)]
spi: sprd: spi: sprd: Add DMA mode support

Add DMA mode support for the Spreadtrum SPI controller, and we will enable
SPI interrupt to help to complete the SPI transfer work in DMA mode.

Signed-off-by: Lanqing Liu <lanqing.liu@unisoc.com>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agodt-bindings: spi: Add the DMA properties for the SPI dma mode
Lanqing Liu [Wed, 13 Feb 2019 07:36:10 +0000 (15:36 +0800)]
dt-bindings: spi: Add the DMA properties for the SPI dma mode

Add the DMA properties for the SPI dma mode.

Signed-off-by: Lanqing Liu <lanqing.liu@unisoc.com>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sprd: Add the SPI irq function for the SPI DMA mode
Lanqing Liu [Wed, 13 Feb 2019 07:36:09 +0000 (15:36 +0800)]
spi: sprd: Add the SPI irq function for the SPI DMA mode

The SPI irq event will use to complete the SPI work in the SPI DMA mode,
so this patch is a preparation for the following DMA mode support.

Moreover the SPI interrupt can be fired when removing the SPI controller,
so we should make sure the SPI controller has stopped the queue in
remove function before freeing the SPI irq.

Signed-off-by: Lanqing Liu <lanqing.liu@unisoc.com>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agodt-bindings: spi: imx: Add an entry for the i.MX8QM compatible
Fabio Estevam [Sat, 26 Jan 2019 21:30:11 +0000 (19:30 -0200)]
dt-bindings: spi: imx: Add an entry for the i.MX8QM compatible

Add an entry for the "fsl,imx8mq-ecspi" compatible to describe
the ECSPI version present on i.MX8M.

Signed-off-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: use gpio[d]_set_value_cansleep for setting chipselect GPIO
Felix Fietkau [Sat, 9 Feb 2019 23:38:25 +0000 (00:38 +0100)]
spi: use gpio[d]_set_value_cansleep for setting chipselect GPIO

Sleeping is safe inside spi_transfer_one_message, and some
GPIO chips are running on slow busses (such as I2C GPIO
expanders) and need to sleep for setting values.

Signed-off-by: Felix Fietkau <nbd@nbd.name>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: gpio: Advertise support for SPI_CS_HIGH
Jonathan Neuschäfer [Sun, 10 Feb 2019 17:10:08 +0000 (18:10 +0100)]
spi: gpio: Advertise support for SPI_CS_HIGH

The spi-gpio driver already handles different chip select polarities,
but so far this was not advertised in master->mode_bits.

This patch fixes mmc_spi on top of spi_gpio, which is useful in some
testing scenarios.

Signed-off-by: Jonathan Neuschäfer <j.neuschaefer@gmx.net>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sh-msiof: Replace spi_master by spi_controller
Geert Uytterhoeven [Fri, 8 Feb 2019 09:09:09 +0000 (10:09 +0100)]
spi: sh-msiof: Replace spi_master by spi_controller

As of commit 8caab75fd2c2a926 ('spi: Generalize SPI "master" to
"controller"'), the old master-centric names are compatibility wrappers
for the new controller-centric names.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sh-hspi: Replace spi_master by spi_controller
Geert Uytterhoeven [Fri, 8 Feb 2019 09:09:08 +0000 (10:09 +0100)]
spi: sh-hspi: Replace spi_master by spi_controller

As of commit 8caab75fd2c2a926 ('spi: Generalize SPI "master" to
"controller"'), the old master-centric names are compatibility wrappers
for the new controller-centric names.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: rspi: Replace spi_master by spi_controller
Geert Uytterhoeven [Fri, 8 Feb 2019 09:09:07 +0000 (10:09 +0100)]
spi: rspi: Replace spi_master by spi_controller

As of commit 8caab75fd2c2a926 ('spi: Generalize SPI "master" to
"controller"'), the old master-centric names are compatibility wrappers
for the new controller-centric names.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: add support for sam9x60 qspi controller
Tudor Ambarus [Tue, 5 Feb 2019 17:33:38 +0000 (17:33 +0000)]
spi: atmel-quadspi: add support for sam9x60 qspi controller

The sam9x60 qspi controller uses 2 clocks, one for the peripheral register
access, the other for the qspi core and phy. Both are mandatory. It uses
different transfer type bits in IFR register. It has dedicated registers
to specify a read or a write instruction: Read Instruction Code Register
(RICR) and Write Instruction Code Register (WICR). ICR/RICR/WICR have
identical fields.

Tested with sst26vf064b jedec,spi-nor flash. Backward compatibility test
done on sama5d2 qspi controller and mx25l25635e jedec,spi-nor flash.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agodt-bindings: spi: atmel-quadspi: QuadSPI driver for Microchip SAM9X60
Tudor Ambarus [Tue, 5 Feb 2019 17:33:35 +0000 (17:33 +0000)]
dt-bindings: spi: atmel-quadspi: QuadSPI driver for Microchip SAM9X60

The sam9x60 qspi controller uses 2 clocks, one for the peripheral register
access, the other for the qspi core and phy. Both are mandatory.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: add support for named peripheral clock
Tudor Ambarus [Tue, 5 Feb 2019 17:33:33 +0000 (17:33 +0000)]
spi: atmel-quadspi: add support for named peripheral clock

Naming clocks is a good practice. Keep supporting unnamed
peripheral clock, to be backward compatible with old DTs.
While here, rename clk to pclk, to indicate that it is a
peripheral clock.

Suggested-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agodt-bindings: spi: atmel-quadspi: make "pclk" mandatory
Tudor Ambarus [Tue, 5 Feb 2019 17:33:30 +0000 (17:33 +0000)]
dt-bindings: spi: atmel-quadspi: make "pclk" mandatory

Naming clocks is a good practice. Make "pclk" madatory even if
we support unnamed clock in the driver, to be backward compatible
with old DTs.

Suggested-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agodt-bindings: spi: atmel-quadspi: update example to new clock binding
Tudor Ambarus [Tue, 5 Feb 2019 17:33:27 +0000 (17:33 +0000)]
dt-bindings: spi: atmel-quadspi: update example to new clock binding

Introduced in:
commit b60557876849 ("ARM: dts: at91: sama5d2: switch to new clock binding")

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: rework transfer macros
Tudor Ambarus [Tue, 5 Feb 2019 17:33:25 +0000 (17:33 +0000)]
spi: atmel-quadspi: rework transfer macros

Split the TFRTYP_TRSFR_ bitfields in 2: one bit encoding the
mem/reg transfer type and one bit encoding the direction of
the transfer (read/write).

Remove NOP when setting read transfer type. Remove useless
setting of write transfer type when
op->data.dir == SPI_MEM_DATA_IN && !op->data.nbytes.

QSPI_IFR_TFRTYP_TRSFR_WRITE is specific just to sama5d2 qspi,
rename it to QSPI_IFR_SAMA5D2_WRITE_TRSFR.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: switch to SPDX license identifiers
Tudor Ambarus [Tue, 5 Feb 2019 17:33:22 +0000 (17:33 +0000)]
spi: atmel-quadspi: switch to SPDX license identifiers

Adopt the SPDX license identifiers to ease license compliance
management.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: return appropriate error code
Tudor Ambarus [Tue, 5 Feb 2019 17:33:19 +0000 (17:33 +0000)]
spi: atmel-quadspi: return appropriate error code

Return -ENOTSUPP when atmel_qspi_find_mode() fails. Propagate
the error in atmel_qspi_exec_op().

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: remove unnecessary cast
Tudor Ambarus [Tue, 5 Feb 2019 17:33:17 +0000 (17:33 +0000)]
spi: atmel-quadspi: remove unnecessary cast

The cast is done implicitly.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: fix naming scheme
Tudor Ambarus [Tue, 5 Feb 2019 17:33:14 +0000 (17:33 +0000)]
spi: atmel-quadspi: fix naming scheme

Let general names to core drivers.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: drop wrappers for iomem accesses
Tudor Ambarus [Tue, 5 Feb 2019 17:33:11 +0000 (17:33 +0000)]
spi: atmel-quadspi: drop wrappers for iomem accesses

The wrappers hid that the accesses are relaxed. Drop them.

Suggested-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: order header files inclusion alphabetically
Tudor Ambarus [Tue, 5 Feb 2019 17:33:08 +0000 (17:33 +0000)]
spi: atmel-quadspi: order header files inclusion alphabetically

Cosmetic change, no functional change.

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel-quadspi: cache MR value to avoid a write access
Tudor Ambarus [Tue, 5 Feb 2019 17:33:06 +0000 (17:33 +0000)]
spi: atmel-quadspi: cache MR value to avoid a write access

Set the controller by default in Serial Memory Mode (SMM) at probe.
Cache Mode Register (MR) value to avoid write access when setting
the controller in serial memory mode at exec_op().

Signed-off-by: Tudor Ambarus <tudor.ambarus@microchip.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-fsl-dspi: Provide support for DSPI slave mode operation (Vybryd vf610)
Lukasz Majewski [Tue, 5 Feb 2019 22:13:49 +0000 (23:13 +0100)]
spi: spi-fsl-dspi: Provide support for DSPI slave mode operation (Vybryd vf610)

The NXP's Vybryd vf610 can work as a SPI slave device (the CS and clock
signals are provided by master).

It is possible to specify a single device to work in that mode. As we do
use DMA for transferring data, the RX channel must be prepared for
incoming data.
Moreover, in slave mode we just set a subset of control fields in
configuration registers (CTAR0, PUSHR).

For testing the spidev_test program has been used.
Test script for this patch can be found here:
https://github.com/lmajewski/tests-spi/blob/master/tests/spi/spi_tests.sh

Signed-off-by: Lukasz Majewski <lukma@denx.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi-atmel: support inter-word delay
Jonas Bonn [Wed, 30 Jan 2019 08:40:05 +0000 (09:40 +0100)]
spi-atmel: support inter-word delay

If the SPI slave requires an inter-word delay, configure the DLYBCT
register accordingly.

Tested on a SAMA5D2 board (derived from SAMA5D2-Xplained reference
board).

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
Acked-by: Nicolas Ferre <nicolas.ferre@microchip.com>
CC: Nicolas Ferre <nicolas.ferre@microchip.com>
CC: Mark Brown <broonie@kernel.org>
CC: Alexandre Belloni <alexandre.belloni@bootlin.com>
CC: Ludovic Desroches <ludovic.desroches@microchip.com>
CC: linux-spi@vger.kernel.org
CC: linux-arm-kernel@lists.infradead.org
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: support inter-word delay requirement for devices
Jonas Bonn [Wed, 30 Jan 2019 08:40:04 +0000 (09:40 +0100)]
spi: support inter-word delay requirement for devices

Some devices are slow and cannot keep up with the SPI bus and therefore
require a short delay between words of the SPI transfer.

The example of this that I'm looking at is a SAMA5D2 with a minimum SPI
clock of 400kHz talking to an AVR-based SPI slave.  The AVR cannot put
bytes on the bus fast enough to keep up with the SoC's SPI controller
even at the lowest bus speed.

This patch introduces the ability to specify a required inter-word
delay for SPI devices.  It is up to the controller driver to configure
itself accordingly in order to introduce the requested delay.

Note that, for spi_transfer, there is already a field word_delay that
provides similar functionality.  This field, however, is specified in
clock cycles (and worse, SPI controller cycles, not SCK cycles); that
makes this value dependent on the master clock instead of the device
clock for which the delay is intended to provide some relief.  This
patch leaves this old word_delay in place and provides a time-based
word_delay_us alongside it; the new field fits in the struct padding
so struct size is constant.  There is only one in-kernel user of the
word_delay field and presumably that driver could be reworked to use
the time-based value instead.

The time-based delay is limited to 8 bits as these delays are intended
to be short.  The SAMA5D2 that I've tested this on limits delays to a
maximum of ~100us, which is already many word-transfer periods even at
the minimum transfer speed supported by the controller.

Signed-off-by: Jonas Bonn <jonas@norrbonn.se>
CC: Mark Brown <broonie@kernel.org>
CC: Rob Herring <robh+dt@kernel.org>
CC: Mark Rutland <mark.rutland@arm.com>
CC: linux-spi@vger.kernel.org
CC: devicetree@vger.kernel.org
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: mxs: add tracing to custom .transfer_one_message callback
Uwe Kleine-König [Tue, 29 Jan 2019 13:48:17 +0000 (14:48 +0100)]
spi: mxs: add tracing to custom .transfer_one_message callback

Driver specific implementations for .transfer_one_message need to call
the tracing stuff themself. This is necessary to make spi tracing
actually useful.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: spi-fsl-qspi: typo fix in author name
Yogesh Narayan Gaur [Tue, 29 Jan 2019 09:55:27 +0000 (09:55 +0000)]
spi: spi-mem: spi-fsl-qspi: typo fix in author name

Typo fix in Author Boris Brezillon last name and update with new
email address.

Fixes: 84d043185dbe ("spi: Add a driver for the Freescale/NXP QuadSPI controller")
Signed-off-by: Yogesh Narayan Gaur <yogeshnarayan.gaur@nxp.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: ti-qspi: Fix mmap read when more than one CS in use
Vignesh R [Tue, 29 Jan 2019 07:44:22 +0000 (13:14 +0530)]
spi: ti-qspi: Fix mmap read when more than one CS in use

Commit 4dea6c9b0b64 ("spi: spi-ti-qspi: add mmap mode read support") has
has got order of parameter wrong when calling regmap_update_bits() to
select CS for mmap access. Mask and value arguments are interchanged.
Code will work on a system with single slave, but fails when more than
one CS is in use. Fix this by correcting the order of parameters when
calling regmap_update_bits().

Fixes: 4dea6c9b0b64 ("spi: spi-ti-qspi: add mmap mode read support")
Cc: stable@vger.kernel.org
Signed-off-by: Vignesh R <vigneshr@ti.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: spi-nxp-fspi: add module license info
Yogesh Narayan Gaur [Tue, 29 Jan 2019 09:49:22 +0000 (09:49 +0000)]
spi: spi-mem: spi-nxp-fspi: add module license info

Add MODULE_LICENSE info to fix below warning:
WARNING: modpost: missing MODULE_LICENSE() in drivers/spi/spi-nxp-fspi.o

Typo fix in Boris Brezillon last name.

Fixes: a5356aef6a90 ("spi: spi-mem: Add driver for NXP FlexSPI controller")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Yogesh Narayan Gaur <yogeshnarayan.gaur@nxp.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: pl022: add a message state STATE_TIMEOUT for timeout transfer
Jiwei Sun [Fri, 18 Jan 2019 03:32:29 +0000 (11:32 +0800)]
spi: pl022: add a message state STATE_TIMEOUT for timeout transfer

When transfer timeout, give -EAGAIN to the message's status, and it can
make the spi device driver choose repeated transimation or not. And if
transfer timeout, output some useful information for tracing the issue.

Signed-off-by: Jiwei Sun <jiwei.sun@windriver.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: Kconfig: imx: Update the help text to make it more generic
Fabio Estevam [Sat, 26 Jan 2019 21:32:07 +0000 (19:32 -0200)]
spi: Kconfig: imx: Update the help text to make it more generic

The spi-imx driver supports both master and slave modes, so update
the help text to make it more generic.

Signed-off-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: nxp-fspi: add octal mode flag bit for octal support
Yogesh Narayan Gaur [Tue, 15 Jan 2019 10:05:29 +0000 (10:05 +0000)]
spi: nxp-fspi: add octal mode flag bit for octal support

Add octal mode flags for octal I/O data transfer support.
NXP FlexSPI controller supports 8 lines Rx/Tx data transfer.

Signed-off-by: Yogesh Narayan Gaur <yogeshnarayan.gaur@nxp.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agoMAINTAINERS: add maintainers for the NXP FlexSPI driver
Yogesh Narayan Gaur [Tue, 15 Jan 2019 12:00:37 +0000 (12:00 +0000)]
MAINTAINERS: add maintainers for the NXP FlexSPI driver

Add maintainers for the NXP FlexSPI driver

Signed-off-by: Yogesh Narayan Gaur <yogeshnarayan.gaur@nxp.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agodt-bindings: spi: add binding file for NXP FlexSPI controller
Yogesh Narayan Gaur [Tue, 15 Jan 2019 12:00:20 +0000 (12:00 +0000)]
dt-bindings: spi: add binding file for NXP FlexSPI controller

Add binding file for NXP FlexSPI controller

Signed-off-by: Yogesh Narayan Gaur <yogeshnarayan.gaur@nxp.com>
Reviewed-by: Rob Herring <robh@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: Add driver for NXP FlexSPI controller
Yogesh Narayan Gaur [Tue, 15 Jan 2019 12:00:15 +0000 (12:00 +0000)]
spi: spi-mem: Add driver for NXP FlexSPI controller

- Add driver for NXP FlexSPI host controller

(0) What is the FlexSPI controller?
 FlexSPI is a flexsible SPI host controller which supports two SPI
 channels and up to 4 external devices. Each channel supports
 Single/Dual/Quad/Octal mode data transfer (1/2/4/8 bidirectional
 data lines) i.e. FlexSPI acts as an interface to external devices,
 maximum 4, each with up to 8 bidirectional data lines.

 It uses new SPI memory interface of the SPI framework to issue
 flash memory operations to up to four connected flash
 devices (2 buses with 2 CS each).

(1) Tested this driver with the mtd_debug and JFFS2 filesystem utility
 on NXP LX2160ARDB and LX2160AQDS targets.
 LX2160ARDB is having two NOR slave device connected on single bus A
 i.e. A0 and A1 (CS0 and CS1).
 LX2160AQDS is having two NOR slave device connected on separate buses
 one flash on A0 and second on B1 i.e. (CS0 and CS3).
 Verified this driver on following SPI NOR flashes:
    Micron, mt35xu512ab, [Read - 1 bit mode]
    Cypress, s25fl512s, [Read - 1/2/4 bit mode]

Signed-off-by: Yogesh Narayan Gaur <yogeshnarayan.gaur@nxp.com>
Reviewed-by: Frieder Schrempf <frieder.schrempf@kontron.de>
Reviewed-by: Boris Brezillon <bbrezillon@kernel.org>
Tested-by: Ashish Kumar <Ashish.Kumar@nxp.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agogpio: of: Fix logic inversion
Linus Walleij [Wed, 16 Jan 2019 08:21:10 +0000 (09:21 +0100)]
gpio: of: Fix logic inversion

The SPI chip selects were not properly inspected due to
a logic inversion. This made SPI GPIOs not work.

Cc: Jan Kotas <jank@cadence.com>
Reported-by: Jan Kotas <jank@cadence.com>
Tested-by: Jan Kotas <jank@cadence.com>
Fixes: f3186dd87669 ("spi: Optionally use GPIO descriptors for CS GPIOs")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: cadence: Fix default polarity of native chipselect
Linus Walleij [Wed, 16 Jan 2019 08:21:09 +0000 (09:21 +0100)]
spi: cadence: Fix default polarity of native chipselect

The Cadence controller also supports platforms specifying
native chipselects. When I enforce the use of high CS
for drivers opting in for using GPIO descriptors, I
inadvertedly switched the driver to also use active
high chip select for native chip selects.

Fix this by inverting the logic in the callback for the
native chip select. Rename the parameter from "is_high"
(which is interpreted as being high when 0, which is
confusing, I will not make any drug-related jokes here)
to "enabled" which is more intuitive, especially now that
it is true when CS is supposed to be enabled.

Cc: Wei Yongjun <weiyongjun1@huawei.com>
Fixes: cfeefa79dc37 ("spi: cadence: Convert to use CS GPIO descriptors")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: dw: Fix default polarity of native chipselect
Linus Walleij [Wed, 16 Jan 2019 08:21:08 +0000 (09:21 +0100)]
spi: dw: Fix default polarity of native chipselect

The DW controller also supports platforms specifying
native chipselects. When I enforce the use of high CS
for drivers opting in for using GPIO descriptors, I
inadvertedly switched the driver to also use active
high chip select for native chip selects.

As it turns out, the DW hardware driving chip selects
also thinks it is weird with active low chip selects
so all we need to do is remove an inversion in the
driver.

Cc: Jan Kotas <jank@cadence.com>
Reported-by: Jan Kotas <jank@cadence.com>
Tested-by: Jan Kotas <jank@cadence.com>
Fixes: 9400c41e77b8 ("spi: dw: Convert to use CS GPIO descriptors")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: Support high CS when using descriptors
Linus Walleij [Wed, 16 Jan 2019 08:21:07 +0000 (09:21 +0100)]
spi: Support high CS when using descriptors

All controllers using GPIO descriptors can by definition
support high CS connections, so just enforce this when
registering an SPI controller.

This fixes a regression where controllers were missing
SPI_CS_HIGH, the drivers would fail like this:

spi spi0.0: setup: unsupported mode bits 4
cdns-spi fd0b0000.spi: can't setup spi0.0, status -22

This is because as using descriptors moves the CS inversion
logic over to gpiolib, all such controllers are registered
with CS active high.

Cc: Jan Kotas <jank@cadence.com>
Reported-by: Jan Kotas <jank@cadence.com>
Tested-by: Jan Kotas <jank@cadence.com>
Fixes: f3186dd87669 ("spi: Optionally use GPIO descriptors for CS GPIOs")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: Go back to immediate teardown
Mark Brown [Wed, 23 Jan 2019 17:29:53 +0000 (17:29 +0000)]
spi: Go back to immediate teardown

Commit 412e6037324 ("spi: core: avoid waking pump thread from spi_sync
instead run teardown delayed") introduced regressions on some boards,
apparently connected to spi_mem not triggering shutdown properly any
more.  Since we've thus far been unable to figure out exactly where the
breakage is revert the optimisation for now.

Reported-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
Cc: kernel@martin.sperl.org
5 years agospi/topcliff_pch: Fix potential NULL dereference on allocation error
YueHaibing [Wed, 23 Jan 2019 12:00:22 +0000 (20:00 +0800)]
spi/topcliff_pch: Fix potential NULL dereference on allocation error

In pch_spi_handle_dma, it doesn't check for NULL returns of kcalloc
so it would result in an Oops.

Fixes: c37f3c2749b5 ("spi/topcliff_pch: DMA support")
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agopxa2xx: replace spi_master with spi_controller
Lubomir Rintel [Wed, 16 Jan 2019 15:13:31 +0000 (16:13 +0100)]
pxa2xx: replace spi_master with spi_controller

It's also a slave controller driver now, calling it "master" is slightly
misleading.

Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
Acked-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: bcm2835aux: remove unneeded NULL check of devm_clk_get
YueHaibing [Wed, 23 Jan 2019 07:05:07 +0000 (15:05 +0800)]
spi: bcm2835aux: remove unneeded NULL check of devm_clk_get

Fix a static code checker warning:
drivers/spi/spi-bcm2835aux.c:460
 bcm2835aux_spi_probe() warn: passing zero to 'PTR_ERR'

In case of error, the function devm_clk_get() returns ERR_PTR()
and not returns NULL.

Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: Add devm_spi_mem_dirmap_{create,destroy}()
Boris Brezillon [Sat, 19 Jan 2019 15:04:12 +0000 (16:04 +0100)]
spi: spi-mem: Add devm_spi_mem_dirmap_{create,destroy}()

Since direct mapping descriptors usually the same lifetime as the SPI
MEM device adding devm_ variants of the spi_mem_dirmap_{create,destroy}()
should greatly simplify error/remove path of spi-mem drivers making use
of the direct mapping API.

Signed-off-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: Fix a memory leak in spi_mem_dirmap_destroy()
Boris Brezillon [Sat, 19 Jan 2019 14:57:57 +0000 (15:57 +0100)]
spi: spi-mem: Fix a memory leak in spi_mem_dirmap_destroy()

The dirmap descriptor object allocated in spi_mem_dirmap_create is
never freed. Add a kfree(desc) in spi_mem_dirmap_destroy().

Fixes: aa167f3fed0c ("spi: spi-mem: Add a new API to support direct mapping")
Signed-off-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-mem: Fix spi_mem_dirmap_destroy() kerneldoc
Boris Brezillon [Sat, 19 Jan 2019 14:57:56 +0000 (15:57 +0100)]
spi: spi-mem: Fix spi_mem_dirmap_destroy() kerneldoc

spi_mem_dirmap_destroy() takes a single argument, remove the @info entry
in the doc.

Fixes: aa167f3fed0c ("spi: spi-mem: Add a new API to support direct mapping")
Signed-off-by: Boris Brezillon <bbrezillon@kernel.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sh-msiof: Use DMA if possible
Hoan Nguyen An [Fri, 18 Jan 2019 09:29:31 +0000 (18:29 +0900)]
spi: sh-msiof: Use DMA if possible

Currently, this driver only supports feature for DMA 32-bits.
In this case, only if the data length is divisible by 4 to use
DMA, otherwise PIO will be used. This patch will suggest use
the DMA 32-bits with 4bytes of words, then the remaining data
will be transmitted by PIO mode.

Signed-off-by: Hoan Nguyen An <na-hoan@jinso.co.jp>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sh-msiof: fix *info pointer in request_dma()
Hoan Nguyen An [Fri, 18 Jan 2019 09:29:30 +0000 (18:29 +0900)]
spi: sh-msiof: fix *info pointer in request_dma()

sh_msiof_spi_info *info struct pointer was initialized in the probe() function
no need to get back and keep consistency.

Signed-off-by: Hoan Nguyen An <na-hoan@jinso.co.jp>
Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sh-hspi: Remove error messages on out-of-memory conditions
Geert Uytterhoeven [Fri, 18 Jan 2019 14:11:05 +0000 (15:11 +0100)]
spi: sh-hspi: Remove error messages on out-of-memory conditions

There is no need to print an error message when memory allocations or
related operations fail, as the core will take care of that.

Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: ath79: Remove now useless code
Alban Bedel [Wed, 16 Jan 2019 18:55:47 +0000 (19:55 +0100)]
spi: ath79: Remove now useless code

The custom setup/cleanup routines included in the ath79 driver only
take care of setting the initial CS state. However that is already
handled by the bitbang code, so this code can be removed.

Signed-off-by: Alban Bedel <albeu@free.fr>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: ath79: Enable support for compile test
Alban Bedel [Wed, 16 Jan 2019 18:55:46 +0000 (19:55 +0100)]
spi: ath79: Enable support for compile test

To allow building this driver in compile test we need to remove all
dependency on headers from arch/mips/include. To allow this we
explicitly define all the registers locally instead of using
ar71xx_regs.h and we move the platform data struct definition to
include/linux/platform_data/spi-ath79.h.

Signed-off-by: Alban Bedel <albeu@free.fr>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: ath79: Simplify ath79_spi_chipselect()
Alban Bedel [Wed, 16 Jan 2019 18:55:45 +0000 (19:55 +0100)]
spi: ath79: Simplify ath79_spi_chipselect()

First of all this callback was slightly misused to setup the clock
polarity at the beginning of a transfer. Beside being at the wrong
place, it is also useless as only SPI mode 1 is supported. Instead
just make sure the base value used for IOC is suitable to start a
transfer by clearing the clock and data bits during the controller
setup.

This also remove the last direct usage of the GPIO API, so we can
remove the direct dependency on GPIOLIB.

Signed-off-by: Alban Bedel <albeu@free.fr>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: bitbang: Don't call chipselect() in spi_bitbang_setup()
Alban Bedel [Wed, 16 Jan 2019 18:55:44 +0000 (19:55 +0100)]
spi: bitbang: Don't call chipselect() in spi_bitbang_setup()

spi_setup() already call spi_set_cs() right after calling the
controller setup method, so there is no need for the bitbang driver to
do that. Because of this the chipselect() callback was confusingly
still called when CS is GPIO based.

Signed-off-by: Alban Bedel <albeu@free.fr>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: dw: change snprintf to scnprintf for possible overflow
Silvio Cesare [Sat, 12 Jan 2019 15:28:44 +0000 (16:28 +0100)]
spi: dw: change snprintf to scnprintf for possible overflow

Change snprintf to scnprintf. There are generally two cases where using
snprintf causes problems.

1) Uses of size += snprintf(buf, SIZE - size, fmt, ...)
In this case, if snprintf would have written more characters than what the
buffer size (SIZE) is, then size will end up larger than SIZE. In later
uses of snprintf, SIZE - size will result in a negative number, leading
to problems. Note that size might already be too large by using
size = snprintf before the code reaches a case of size += snprintf.

2) If size is ultimately used as a length parameter for a copy back to user
space, then it will potentially allow for a buffer overflow and information
disclosure when size is greater than SIZE. When the size is used to index
the buffer directly, we can have memory corruption. This also means when
size = snprintf... is used, it may also cause problems since size may become
large.  Copying to userspace is mitigated by the HARDENED_USERCOPY kernel
configuration.

The solution to these issues is to use scnprintf which returns the number of
characters actually written to the buffer, so the size variable will never
exceed SIZE.

Signed-off-by: Silvio Cesare <silvio.cesare@gmail.com>
Signed-off-by: Willy Tarreau <w@1wt.eu>
Reviewed-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: sprd: Fix the error data length in SPI read-only mode
Lanqing Liu [Tue, 15 Jan 2019 13:46:50 +0000 (21:46 +0800)]
spi: sprd: Fix the error data length in SPI read-only mode

In SPI read-only mode, we will always return the writing length,
which is always the power of "bits_per_word", but the length unit
using by users is byte.

Thus this patch fixes the returning length by getting from
read_bufs() function to get the correct length.

Signed-off-by: Lanqing Liu <lanqing.liu@spreadtrum.com>
Signed-off-by: Baolin Wang <baolin.wang@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: omap2-mcspi: Fix DMA and FIFO event trigger size mismatch
Vignesh R [Tue, 15 Jan 2019 06:58:32 +0000 (12:28 +0530)]
spi: omap2-mcspi: Fix DMA and FIFO event trigger size mismatch

Commit b682cffa3ac6 ("spi: omap2-mcspi: Set FIFO DMA trigger level to word length")
broke SPI transfers where bits_per_word != 8. This is because of
mimsatch between McSPI FIFO level event trigger size (SPI word length) and
DMA request size(word length * maxburst). This leads to data
corruption, lockup and errors like:

spi1.0: EOW timed out

Fix this by setting DMA maxburst size to 1 so that
McSPI FIFO level event trigger size matches DMA request size.

Fixes: b682cffa3ac6 ("spi: omap2-mcspi: Set FIFO DMA trigger level to word length")
Cc: stable@vger.kernel.org
Reported-by: David Lechner <david@lechnology.com>
Tested-by: David Lechner <david@lechnology.com>
Signed-off-by: Vignesh R <vigneshr@ti.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: cadence: Correct initialisation of runtime PM
Charles Keepax [Fri, 4 Jan 2019 18:08:09 +0000 (18:08 +0000)]
spi: cadence: Correct initialisation of runtime PM

Currently the driver calls pm_runtime_put_autosuspend but without ever
having done a pm_runtime_get, this causes the reference count in the pm
runtime core to become -1. The bad reference count causes the core to
sometimes suspend whilst an active SPI transfer is in progress.

arizona spi0.1: SPI transfer timed out
spi_master spi0: failed to transfer one message from queue

The correct proceedure is to do all the initialisation that requires the
hardware to be powered up before enabling the PM runtime, then enable
the PM runtime having called pm_runtime_set_active to inform it that the
hardware is currently powered up. The core will then power it down at
it's leisure and no explicit pm_runtime_put is required.

Fixes: d36ccd9f7ea4 ("spi: cadence: Runtime pm adaptation")
Signed-off-by: Charles Keepax <ckeepax@opensource.cirrus.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-geni-qcom: Get rid of forward declaration
Stephen Boyd [Thu, 10 Jan 2019 21:02:53 +0000 (13:02 -0800)]
spi: spi-geni-qcom: Get rid of forward declaration

We don't need this forward declaration. Move the function to where it
needed so we can drop it and shave some lines of code.

CC: Girish Mahadevan <girishm@codeaurora.org>
CC: Dilip Kota <dkota@codeaurora.org>
CC: Alok Chauhan <alokc@codeaurora.org>
Cc: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: spi-geni-qcom: Don't initialize completion for normal message
Stephen Boyd [Thu, 10 Jan 2019 21:02:52 +0000 (13:02 -0800)]
spi: spi-geni-qcom: Don't initialize completion for normal message

We only use this completion when we're doing something that isn't a
message transfer. For example, changing CS or aborting/canceling a
command. All of those situations properly reinitialize the completion
before sending the GENI the special command to change CS or cancel, etc.
Given that, let's remove the initialization here.

Cc: Girish Mahadevan <girishm@codeaurora.org>
Cc: Dilip Kota <dkota@codeaurora.org>
Cc: Alok Chauhan <alokc@codeaurora.org>
Cc: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: davinci: Get rid of dangling variable
Linus Walleij [Thu, 10 Jan 2019 11:48:42 +0000 (12:48 +0100)]
spi: davinci: Get rid of dangling variable

The previous commit left a variable unused, my bad.
Clean it up.

Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: David Lechner <david@lechnology.com>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Fixes: 101a68e74fe0 ("spi: davinci: Convert to use CS GPIO descriptors")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: dw: fix warning unused variable 'ret'
Anders Roxell [Thu, 10 Jan 2019 10:14:07 +0000 (11:14 +0100)]
spi: dw: fix warning unused variable 'ret'

When CONFIG_SPI_DESIGNWARE are enabled we see the unused variable
warning in dw_spi_setup.

../drivers/spi/spi-dw.c: In function ‘dw_spi_setup’:
../drivers/spi/spi-dw.c:400:6: warning: unused variable ‘ret’ [-Wunused-variable]
  int ret;
      ^~~

Remove the unused varable.

Fixes: 9400c41e77b8 ("spi: dw: Convert to use CS GPIO descriptors")
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Anders Roxell <anders.roxell@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: core: avoid waking pump thread from spi_sync instead run teardown delayed
Martin Sperl [Tue, 8 Jan 2019 12:13:45 +0000 (12:13 +0000)]
spi: core: avoid waking pump thread from spi_sync instead run teardown delayed

When spi_sync is running alone with no other spi devices connected
to the bus the worker thread is woken during spi_finalize_current_message
to run the teardown code every time.

This is totally unnecessary in the case that there is no message queued.

On a multi-core system this results in one wakeup of the thread for each
spi_message processed via spi_sync where in most cases the teardown does
not happen as the hw is already in use.

This patch now delays the teardown by 1 second by using a separate
kthread_delayed_work for the teardown.

This avoids waking the kthread too often.

For spi_sync transfers in a tight loop (say 40k messages/s) this
avoids the penalty of waking the worker thread 40k times/s.
On a rasperry pi 3 with 4 cores the results in 32% of a single core
only to find out that there is nothing in the queue and it can go back
to sleep.

With this patch applied the spi-worker is woken exactly once: after
the load finishes and the spi bus is idle for 1 second.

I believe I have also seen situations where during a spi_sync loop
the worker thread (triggered by the last message finished) is slightly
faster and _wins_ the race to process the message, so we are actually
running the kthread and letting it do some work...

This is also no longer observed with this patch applied as.

Tested with a new CAN controller driver for the mcp2517fd which
uses spi_sync for interrupt handling and spi_async for scheduling
of can frames for transmission (in a different thread)

Some statistics when receiving 100000 CAN frames with the mcp25xxfd driver
on a Raspberry pi 3:

without the patch:
------------------
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0)                    5
(irq/94-mcp25xxf)         0
root@raspcm3:~# vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa st
 1  0      0 821960  13592  50848    0    0    80     2 1986  105  1  2 97  0  0
 0  0      0 821968  13592  50876    0    0     0     0 8046   30  0  0 100  0  0
 0  0      0 821936  13592  50876    0    0     0     0 8032   24  0  0 100  0  0
 0  0      0 821936  13592  50876    0    0     0     0 8035   30  0  0 100  0  0
 0  0      0 821936  13592  50876    0    0     0     0 8033   22  0  0 100  0  0
 2  0      0 821936  13592  50876    0    0     0     0 11598 7129  0  3 97  0  0
 1  0      0 821872  13592  50876    0    0     0     0 37741 59003  0 31 69  0  0
 2  0      0 821840  13592  50876    0    0     0     0 37762 59078  0 29 71  0  0
 2  0      0 821776  13592  50876    0    0     0     0 37593 58792  0 28 72  0  0
 1  0      0 821744  13592  50876    0    0     0     0 37642 58881  0 30 70  0  0
 2  0      0 821680  13592  50876    0    0     0     0 37490 58602  0 27 73  0  0
 1  0      0 821648  13592  50876    0    0     0     0 37412 58418  0 29 71  0  0
 1  0      0 821584  13592  50876    0    0     0     0 37337 58288  0 27 73  0  0
 1  0      0 821552  13592  50876    0    0     0     0 37584 58774  0 27 73  0  0
 0  0      0 821520  13592  50876    0    0     0     0 18363 20566  0  9 91  0  0
 0  0      0 821520  13592  50876    0    0     0     0 8037   32  0  0 100  0  0
 0  0      0 821520  13592  50876    0    0     0     0 8031   23  0  0 100  0  0
 0  0      0 821520  13592  50876    0    0     0     0 8034   26  0  0 100  0  0
 0  0      0 821520  13592  50876    0    0     0     0 8033   24  0  0 100  0  0
^C
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0)                  228
(irq/94-mcp25xxf)       794
root@raspcm3:~# cat /proc/interrupts
           CPU0       CPU1       CPU2       CPU3
 17:         34          0          0          0  ARMCTRL-level   1 Edge      3f00b880.mailbox
 27:          1          0          0          0  ARMCTRL-level  35 Edge      timer
 33:    1416870          0          0          0  ARMCTRL-level  41 Edge      3f980000.usb, dwc2_hsotg:usb1
 34:          1          0          0          0  ARMCTRL-level  42 Edge      vc4
 35:          0          0          0          0  ARMCTRL-level  43 Edge      3f004000.txp
 40:       1753          0          0          0  ARMCTRL-level  48 Edge      DMA IRQ
 42:         11          0          0          0  ARMCTRL-level  50 Edge      DMA IRQ
 44:         11          0          0          0  ARMCTRL-level  52 Edge      DMA IRQ
 45:          0          0          0          0  ARMCTRL-level  53 Edge      DMA IRQ
 66:          0          0          0          0  ARMCTRL-level  74 Edge      vc4 crtc
 69:          0          0          0          0  ARMCTRL-level  77 Edge      vc4 crtc
 70:          0          0          0          0  ARMCTRL-level  78 Edge      vc4 crtc
 77:         20          0          0          0  ARMCTRL-level  85 Edge      3f205000.i2c, 3f804000.i2c, 3f805000.i2c
 78:       6346          0          0          0  ARMCTRL-level  86 Edge      3f204000.spi
 80:        205          0          0          0  ARMCTRL-level  88 Edge      mmc0
 81:        493          0          0          0  ARMCTRL-level  89 Edge      uart-pl011
 89:          0          0          0          0  bcm2836-timer   0 Edge      arch_timer
 90:       4291       3821       2180       1649  bcm2836-timer   1 Edge      arch_timer
 94:      14289          0          0          0  pinctrl-bcm2835  16 Level     mcp25xxfd
IPI0:          0          0          0          0  CPU wakeup interrupts
IPI1:          0          0          0          0  Timer broadcast interrupts
IPI2:       3645     242371       7919       1328  Rescheduling interrupts
IPI3:        112        543        273        194  Function call interrupts
IPI4:          0          0          0          0  CPU stop interrupts
IPI5:          1          0          0          0  IRQ work interrupts
IPI6:          0          0          0          0  completion interrupts
Err:          0

top shows 93% for the mcp25xxfd interrupt handler, 31% for spi0.

with the patch:
---------------
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0)                    0
(irq/94-mcp25xxf)         0
root@raspcm3:~# vmstat 1
procs -----------memory---------- ---swap-- -----io---- -system-- ------cpu-----
 0  0      0 804768  13584  62628    0    0     0     0 8038   24  0  0 100  0  0
 0  0      0 804768  13584  62628    0    0     0     0 8042   25  0  0 100  0  0
 1  0      0 804704  13584  62628    0    0     0     0 9603 2967  0 20 80  0  0
 1  0      0 804672  13584  62628    0    0     0     0 9828 3380  0 24 76  0  0
 1  0      0 804608  13584  62628    0    0     0     0 9823 3375  0 23 77  0  0
 1  0      0 804608  13584  62628    0    0     0    12 9829 3394  0 23 77  0  0
 1  0      0 804544  13584  62628    0    0     0     0 9816 3362  0 22 78  0  0
 1  0      0 804512  13584  62628    0    0     0     0 9817 3367  0 23 77  0  0
 1  0      0 804448  13584  62628    0    0     0     0 9822 3370  0 22 78  0  0
 1  0      0 804416  13584  62628    0    0     0     0 9815 3367  0 23 77  0  0
 0  0      0 804352  13584  62628    0    0     0    84 9222 2250  0 14 86  0  0
 0  0      0 804352  13592  62620    0    0     0    24 8131  209  0  0 93  7  0
 0  0      0 804320  13592  62628    0    0     0     0 8041   27  0  0 100  0  0
 0  0      0 804352  13592  62628    0    0     0     0 8040   26  0  0 100  0  0
root@raspcm3:~# for x in $(pgrep spi0) $(pgrep irq/94-mcp25xxf) ; do awk '{printf "%-20s %6i\n", $2,$15}' /proc/$x/stat; done
(spi0)                    0
(irq/94-mcp25xxf)       767
root@raspcm3:~# cat /proc/interrupts
           CPU0       CPU1       CPU2       CPU3
 17:         29          0          0          0  ARMCTRL-level   1 Edge      3f00b880.mailbox
 27:          1          0          0          0  ARMCTRL-level  35 Edge      timer
 33:    1024412          0          0          0  ARMCTRL-level  41 Edge      3f980000.usb, dwc2_hsotg:usb1
 34:          1          0          0          0  ARMCTRL-level  42 Edge      vc4
 35:          0          0          0          0  ARMCTRL-level  43 Edge      3f004000.txp
 40:       1773          0          0          0  ARMCTRL-level  48 Edge      DMA IRQ
 42:         11          0          0          0  ARMCTRL-level  50 Edge      DMA IRQ
 44:         11          0          0          0  ARMCTRL-level  52 Edge      DMA IRQ
 45:          0          0          0          0  ARMCTRL-level  53 Edge      DMA IRQ
 66:          0          0          0          0  ARMCTRL-level  74 Edge      vc4 crtc
 69:          0          0          0          0  ARMCTRL-level  77 Edge      vc4 crtc
 70:          0          0          0          0  ARMCTRL-level  78 Edge      vc4 crtc
 77:         20          0          0          0  ARMCTRL-level  85 Edge      3f205000.i2c, 3f804000.i2c, 3f805000.i2c
 78:       6417          0          0          0  ARMCTRL-level  86 Edge      3f204000.spi
 80:        237          0          0          0  ARMCTRL-level  88 Edge      mmc0
 81:        489          0          0          0  ARMCTRL-level  89 Edge      uart-pl011
 89:          0          0          0          0  bcm2836-timer   0 Edge      arch_timer
 90:       4048       3704       2383       1892  bcm2836-timer   1 Edge      arch_timer
 94:      14287          0          0          0  pinctrl-bcm2835  16 Level     mcp25xxfd
IPI0:          0          0          0          0  CPU wakeup interrupts
IPI1:          0          0          0          0  Timer broadcast interrupts
IPI2:       2361       2948       7890       1616  Rescheduling interrupts
IPI3:         65        617        301        166  Function call interrupts
IPI4:          0          0          0          0  CPU stop interrupts
IPI5:          1          0          0          0  IRQ work interrupts
IPI6:          0          0          0          0  completion interrupts
Err:          0
top shows 91% for the mcp25xxfd interrupt handler, 0% for spi0

So we see that spi0 is no longer getting scheduled wasting CPU cycles
There are a lot less context switches and corresponding Rescheduling interrupts
All of these show that this improves efficiency of the system and reduces
CPU utilization.

Signed-off-by: Martin Sperl <kernel@martin.sperl.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: dw: Convert to use CS GPIO descriptors
Linus Walleij [Mon, 7 Jan 2019 15:51:56 +0000 (16:51 +0100)]
spi: dw: Convert to use CS GPIO descriptors

This converts the DesignWare (dw) SPI master driver to
use GPIO descriptors for chip select handling.

This driver has a duplicate DT parser in addition to the
one in the core, sets up the line as non-asserted and
relies on the core to drive the GPIOs.

It is a pretty straight-forward conversion.

Cc: Talel Shenhar <talel@amazon.com>
Cc: Simon Goldschmidt <simon.k.r.goldschmidt@gmail.com>
Cc: Alexandre Belloni <alexandre.belloni@bootlin.com>
Cc: Linuxarm <linuxarm@huawei.com>
Tested-by: Jay Fang <f.fangjian@huawei.com>
Reviewed-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: davinci: Convert to use CS GPIO descriptors
Linus Walleij [Mon, 7 Jan 2019 15:51:55 +0000 (16:51 +0100)]
spi: davinci: Convert to use CS GPIO descriptors

This converts the DaVinci SPI master driver to use GPIO
descriptors for chip select handling.

DaVinci parses the device tree a second time for the chip
select GPIOs (no relying on the parsing already happening
in the SPI core) and handles inversion semantics locally.

We simply drop the extra parsing and set up and move the
CS handling to the core and gpiolib. The fact that the
driver is actively driving the GPIO in the
davinci_spi_chipselect() callback is confusing since the
host does not set SPI_MASTER_GPIO_SS so this should not
ever get called when using GPIO CS. I put in a comment
about this.

This driver also supports instantiation from board files,
but these are all using native chip selects so no problem
with GPIO lines here.

Cc: David Lechner <david@lechnology.com>
Cc: Bartosz Golaszewski <bgolaszewski@baylibre.com>
Cc: Linuxarm <linuxarm@huawei.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: clps711x: Convert to use CS GPIO descriptors
Linus Walleij [Mon, 7 Jan 2019 15:51:54 +0000 (16:51 +0100)]
spi: clps711x: Convert to use CS GPIO descriptors

This converts the CLPS711x SPI master driver to use GPIO
descriptors for chip select handling.

The CLPS711x driver was merely requesting the GPIO and
setting the CS line non-asserted so this was a pretty
straight-forward conversion. The setup callback goes away.

Cc: Alexander Shiyan <shc_work@mail.ru>
Cc: Linuxarm <linuxarm@huawei.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: cadence: Convert to use CS GPIO descriptors
Linus Walleij [Mon, 7 Jan 2019 15:51:53 +0000 (16:51 +0100)]
spi: cadence: Convert to use CS GPIO descriptors

This converts the Cadence SPI master driver to use GPIO
descriptors for chip select handling.

The Cadence driver was allocating a state container just
to hold the requested GPIO line and contained lots of
polarity inversion code. As this is all handled by gpiolib
and a simple devm_* request in the core, and as the driver
is fully device tree only, most of this code chunk goes
away in favour of central handling. The setup/cleanup
callbacks goes away.

This driver does NOT drive the CS line by setting the
value of the GPIO so it relies on the SPI core to do
this, which should work just fine with the descriptors.

Cc: Wei Yongjun <weiyongjun1@huawei.com>
Cc: Janek Kotas <jank@cadence.com>
Cc: Linuxarm <linuxarm@huawei.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: atmel: Convert to use CS GPIO descriptors
Linus Walleij [Mon, 7 Jan 2019 15:51:52 +0000 (16:51 +0100)]
spi: atmel: Convert to use CS GPIO descriptors

This converts the Atmel SPI master driver to use GPIO descriptors
for chip select handling.

The Atmel driver has duplicate code to look up and initialize CS
GPIOs from the device tree, so this is removed. It further has code
to retrieve a CS GPIO from .controller_data but this seems to be
completely unused in the kernel (legacy codepath?) so I deleted
this support. It keeps track of polarity when switching the CS, but
this is not needed anymore since we moved this over to the gpiolib.

The local handling of the "npcs_pin" (I guess this might mean
"negative polarity chip select pin") is preserved, but I strongly
suspect this can be switched over to handling by the core and
using the SPI_MASTER_GPIO_SS flag on the master to assure that
the additional CS handling in the driver is also done.

Cc: Eugen Hristev <eugen.hristev@microchip.com>
Cc: Nicolas Ferre <nicolas.ferre@microchip.com>
Cc: Radu Pirea <radu.pirea@microchip.com>
Cc: Linuxarm <linuxarm@huawei.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: ath79: Convert to use CS GPIO descriptors
Linus Walleij [Mon, 7 Jan 2019 15:51:51 +0000 (16:51 +0100)]
spi: ath79: Convert to use CS GPIO descriptors

This converts the ATH79 SPI master driver to use GPIO descriptors
for chip select handling.

The ATH79 driver was requesting the GPIO and driving it from the
bitbang .chipselect callback. Do not request it anymore as the SPI
core will request it, remove the line inversion semantics for the
GPIO case (handled by gpiolib) and let the SPI core deal with
requesting the GPIO line from the device tree node of the controller.

This driver can be instantiated from a board file (no device tree)
but the board files only use native CS (no GPIO lines) so we should
be fine just letting the SPI core grab the GPIO from the device.

The fact that the driver is actively driving the GPIO in the
ath79_spi_chipselect() callback is confusing since the host does
not set SPI_MASTER_GPIO_SS so this should not ever get called when
using GPIO CS. I put in a comment about this.

Cc: Felix Fietkau <nbd@nbd.name>
Cc: Alban Bedel <albeu@free.fr>
Cc: Linuxarm <linuxarm@huawei.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: Optionally use GPIO descriptors for CS GPIOs
Linus Walleij [Mon, 7 Jan 2019 15:51:50 +0000 (16:51 +0100)]
spi: Optionally use GPIO descriptors for CS GPIOs

This augments the SPI core to optionally use GPIO descriptors
for chip select on a per-master-driver opt-in basis.

Drivers using this will rely on the SPI core to look up
GPIO descriptors associated with the device, such as
when using device tree or board files with GPIO descriptor
tables.

When getting descriptors from the device tree, this will in
turn activate the code in gpiolib that was
added in commit 6953c57ab172
("gpio: of: Handle SPI chipselect legacy bindings")
which means that these descriptors are aware of the active
low semantics that is the default for SPI CS GPIO lines
and we can assume that all of these are "active high" and
thus assign SPI_CS_HIGH to all CS lines on the DT path.

The previously used gpio_set_value() would call down into
gpiod_set_raw_value() and ignore the polarity inversion
semantics.

It seems like many drivers go to great lengths to set up the
CS GPIO line as non-asserted, respecting SPI_CS_HIGH. We pull
this out of the SPI drivers and into the core, and by simply
requesting the line as GPIOD_OUT_LOW when retrieveing it from
the device and relying on the gpiolib to handle any inversion
semantics. This way a lot of code can be simplified and
removed in each converted driver.

The end goal after dealing with each driver in turn, is to
delete the non-descriptor path (of_spi_register_master() for
example) and let the core deal with only descriptors.

The different SPI drivers have complex interactions with the
core so we cannot simply change them all over, we need to use
a stepwise, bisectable approach so that each driver can be
converted and fixed in isolation.

This patch has the intended side effect of adding support for
ACPI GPIOs as it starts relying on gpiod_get_*() to get
the GPIO handle associated with the device.

Cc: Linuxarm <linuxarm@huawei.com>
Acked-by: Jonathan Cameron <jonathan.cameron@huawei.com>
Tested-by: Fangjian (Turing) <f.fangjian@huawei.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi/trace: include buffer contents in traces
Uwe Kleine-König [Fri, 21 Dec 2018 10:35:36 +0000 (11:35 +0100)]
spi/trace: include buffer contents in traces

It highly improves usability when the buffer contents are inspecable via
tracing.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi/trace: drop useless and wrong (but harmless) casts
Uwe Kleine-König [Fri, 21 Dec 2018 10:35:35 +0000 (11:35 +0100)]
spi/trace: drop useless and wrong (but harmless) casts

bus_num, chip_select and len are already ints, so there is no gain in
casting them to int. xfer is a pointer to a struct spi_transfer. Casting
that to struct spi_message * is wrong but as only the pointer value is
used for the %p format specifier no harm is done.

Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: add description about STM32F4 bindings
Cezary Gapinski [Mon, 24 Dec 2018 22:00:40 +0000 (23:00 +0100)]
spi: stm32: add description about STM32F4 bindings

Add description that STM32F4 can be used in compatible property.
Master Inter-Data Idleness optional property cannot be used in STM32F4.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: add support for STM32F4
Cezary Gapinski [Mon, 24 Dec 2018 22:00:38 +0000 (23:00 +0100)]
spi: stm32: add support for STM32F4

Add routines, registers & bitfield definition. Also baud rate divisor
definitions for STM32F4 SPI. This version supports full-duplex,
simplex TX and half-duplex TX communication with 8 or 16-bit per word.
DMA capability is optionally supported for transfer longer than 16 bytes.
For transfer less than 16 bytes frames can be send in discontinuous mode.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: introduce compatible data cfg
Cezary Gapinski [Mon, 24 Dec 2018 22:00:37 +0000 (23:00 +0100)]
spi: stm32: introduce compatible data cfg

Prepare support for STM32F4 spi variant by introducing compatible
configuration data.
Move STM32H7 specific stuff to compatible data structure:
 - registers & bit fields
 - routines to control driver
 - baud rate divisor definitions
 - fifo availability
 - split IRQ functions to parts to be called when the IRQ occurs
   and for threaded interrupt what helps to provide less discontinuous
   mode for drivers without FIFO.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: add start dma transfer function
Cezary Gapinski [Mon, 24 Dec 2018 22:00:36 +0000 (23:00 +0100)]
spi: stm32: add start dma transfer function

Add transfer_one_dma_start function to be more generic for other
stm32 SPI family drivers.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: split transfer one setup function
Cezary Gapinski [Mon, 24 Dec 2018 22:00:35 +0000 (23:00 +0100)]
spi: stm32: split transfer one setup function

Split stm32_spi_transfer_one_setup function into smaller chunks
to be more generic for other stm32 SPI family drivers.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: rename interrupt function
Cezary Gapinski [Mon, 24 Dec 2018 22:00:34 +0000 (23:00 +0100)]
spi: stm32: rename interrupt function

Interrupt function is used as a thread so rename it to express
meaning directly by more clear function name.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: rename STM32 SPI registers to STM32H7
Cezary Gapinski [Mon, 24 Dec 2018 22:00:33 +0000 (23:00 +0100)]
spi: stm32: rename STM32 SPI registers to STM32H7

Rename STM32 SPI registers to be related to STM32H7 SPI driver
and not STM32 generally.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
5 years agospi: stm32: remove SPI LOOP mode
Cezary Gapinski [Mon, 24 Dec 2018 22:00:32 +0000 (23:00 +0100)]
spi: stm32: remove SPI LOOP mode

This driver does not support SPI LOOP mode by configuration
in registers.

Signed-off-by: Cezary Gapinski <cezary.gapinski@gmail.com>
Signed-off-by: Mark Brown <broonie@kernel.org>