Add DMA support to the MSIOF driver using platform data.
As MSIOF DMA is limited to 32-bit words (requiring byte/wordswapping for
smaller wordsizes), and the group length is limited to 256 words, DMA is
performed on two fixed pages, allocated and mapped at driver initialization
time.
Performance figures (in Mbps) on r8a7791/koelsch at different SPI clock
frequencies for 1024-byte and 4096-byte transfers:
1024 bytes 4096 bytes
- 3.25 MHz: PIO 2.1, DMA 2.6 | PIO 2.8, DMA 3.1
- 6.5 MHz: PIO 3.2, DMA 4.4 | PIO 5.0, DMA 5.9
- 13 MHz: PIO 4.2, DMA 6.6 | PIO 8.2, DMA 10.7
- 26 MHz: PIO 5.9, DMA 10.4 | PIO 12.4, DMA 18.4
Note that DMA is only faster than PIO for transfers that exceed the FIFO
size (typically 64 words / 256 bytes).
Also note that large transfers (larger than the group length for DMA, or
larger than the FIFO size for PIO), should use cs-gpio (with the
appropriate pinmux setup), as the hardware chipselect will be deasserted in
between chunks.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
- Move buffer pointer and length setup to the top,
- Make unsigned values unsigned,
- Loop over words and increment pointers instead of recalculating them,
which allows to kill bytes_done.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
- Add a timeout when waiting for the transfer complete interrupt,
- If sh_msiof_spi_stop() fails, there's no need to clear IER, as the
interrupt handler has already done that,
- Propagate transfer failures in sh_msiof_transfer_one().
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Based on an old patch by Guennadi Liakhovetski <g.liakhovetski@gmx.de>
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Make of_device_id array const, because all OF functions handle
it as const.
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
Make of_device_id array const, because all OF functions handle
it as const.
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
Make of_device_id array const, because all OF functions handle
it as const.
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Acked-by: Harini Katakam <harinik@xilinx.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
Drop call to platform_set_drvdata() as driver data is not used
anywhere in the driver.
Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Signed-off-by: Mark Brown <broonie@linaro.org>
Drop call to platform_set_drvdata() as driver data is not used
anywhere in the driver.
Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Signed-off-by: Mark Brown <broonie@linaro.org>
Drop call to platform_set_drvdata() as driver data is not used
anywhere in the driver.
Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Signed-off-by: Mark Brown <broonie@linaro.org>
Drop call to platform_set_drvdata() as driver data is not used
anywhere in the driver.
Signed-off-by: Alexander Shiyan <shc_work@mail.ru>
Signed-off-by: Mark Brown <broonie@linaro.org>
Extract the common parts of rspi_transfer_one(), rspi_rz_transfer_one(),
and qspi_transfer_out_in() into the new function rspi_common_transfer().
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Enable DMA support for RSPI on r7s72100 (RZ/A1H).
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Enable DMA support for QSPI on R-Car Gen2, for Single, Dual, and Quad SPI
Transfers.
Performance figures for reading from a QSPI FLASH driven at 24.375 MHz
on r8a7791/koelsch:
- Single: 1.1 Mbps PIO, 23 Mbps DMA
- Dual : 12.7 Mbps PIO, 48 Mbps DMA
- Quad : 13 Mbps PIO, 70 Mbps DMA
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
rspi_send_dma() and rspi_send_receive_dma() are very similar. Consolidate
into a single function rspi_dma_transfer(), and add missing checks for
dmaengine_submit() failures.
Both sg_table pointer parameters can be NULL, as RSPI supports TX-only
mode, and unidirectional DMA transfers will also be needed later for
Dual/Quad DMA support.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
The DMA routines only need access to the scatter-gather tables inside the
spi_transfer structures, hence just pass those.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Refactor RSPI (on SH) DMA handling to make it reusable for other RSPI
implementations:
- Call the DMA routines after configuring the TX Mode bit and after
calling rspi_receive_init(), so these RSPI-specific operations can be
removed from the DMA routines,
- Absorb rspi_transfer_out_in() into rspi_transfer_one().
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Use the SPI core DMA mapping framework instead of our own.
If available, DMA is used for transfers larger than the FIFO size
(8 or 32 bytes).
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
The SPI DMA core framework needs both RX and TX DMA to function. As a
preparation for converting the driver to use this framework, fall back to
PIO if no DMA channel or only one DMA channel is available.
This affects only RSPI, which could do DMA transfers for TX-only before.
RSPI-RZ and QSPI (at least for Single SPI Transfers) will need both RX and
TX DMA anyway.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
The resource is know to exist, as rspi_probe() already mapped it.
Remove the test, and just pass the resource.
Pass the device pointer instead of the platform device pointer, as the
latter is no longer needed.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Setup of the receive and transmit DMA channels is very similar, so let's
consolidate.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Fall back to PIO if DMA configuration failed.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
The various PIO loops are very similar. Consolidate into a single
function rspi_pio_transfer().
Both buffer pointers can be NULL, as RSPI supports TX-only mode, and
Dual/Quad SPI Transfers are unidirectional.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
RSPI needs dummy transfers to generate the SPI clock on receive.
RSPI-RZ and QSPI always do both transmit and receive.
Use the SPI core SPI_MASTER_MUST_RX/SPI_MASTER_MUST_TX infrastructure
instead of checking for the presence of buffers and providing dummy data
ourselves (for PIO), or providing a dummy buffer (for DMA).
rspi_receive_dma() now provides full duplex DMA transfers on RSPI, and is
renamed to rspi_send_receive_dma().
As the SPI core will always provide a TX buffer, the logic to choose
between DMA send and DMA send/receive in rspi_transfer_one() now has to
check for the presence of an RX buffer. Likewise for the DMA availability
tests in rspi_is_dma().
The buffer tests in qspi_transfer_one() are now always true, so they're
removed.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
The 16-bit DMA support doesn't fit well within the SPI core DMA framework,
as it needs to manage its own double-sized temporary buffers, for handling
the interleaved data.
Remove it, as there is no in-tree board code that sets
rspi_plat_data.dma_width_16bit.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
Since commit 8449fd76de ("spi: rspi: Merge
rspi_send_pio() and rspi_receive_pio()"), rspi_receive_init() is called
for transmit-only transfers too, while this is not needed.
Only call rspi_receive_init() when receiving, to preserve behavior on
RSPI on SH.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
In commit 7dd6278733 (spi/pxa2xx: Convert
to core runtime PM) master->auto_runtime_pm was set to true.
In this case pm_runtime_enable() must be called *before*
spi_register_master(), otherwise the kernel hangs with this error
message:
spi_master spi0: Failed to power device: -13
A similar fix, but for spi/hspi, was applied in
268d76430d.
Signed-off-by: Antonio Ospite <ao2@ao2.it>
Signed-off-by: Mark Brown <broonie@linaro.org>
By default for every espi transfer, the rx_buf is placed right after the
tx_buf. This can lead to a buffer overflow when the size of both the TX
and RX data cumulated is higher than the allocated 64K buffer for the
transfer (this is the case when sending for instance a read command and
reading 64K back, please see:
http://article.gmane.org/gmane.linux.drivers.mtd/53411 )
This gets fixed by always setting the RX buffer pointer at the begining
of the transfer buffer.
[The driver shouldn't be doing the copy in the first place and instead
sending directly from the supplied buffer but this is at least not worse
than what's there -- broonie]
Signed-off-by: Valentin Longchamp <valentin.longchamp@keymile.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
Rejecting unsupported values of spi-tx-bus-width and spi-rx-bus-width
may break compatibility with future DTs. Just ignore them, falling back
to Single SPI Transfers.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
The calculation of the bit rate divider used a standard C division, which
rounds down the quotient. This may lead to a higher bitrate than requested.
Round up to avoid this.
E.g. on Koelsch, the SPI flash (configured for 30 MHz) was driven at 48.75
MHz. After this patch it's driven at a safe 24.375 MHz.
Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Mark Brown <broonie@linaro.org>
current PIO tranfer method be described as follows:
1. fill as much as bytes but no more than 256 bytes(fifo size)
2. enable oflow/uflow/txfifo_empty interrupt
3. isr process 3 interrupt signal, do complete works.
4. after isr done, if there are left bytes go into 1 else go into 5
5. transfer end
by current PIO transfer method:
1. reduce interrupt counts in spi interrupt line.
2. reduce interrupt latency because no do data fill/fetch in isr.
Signed-off-by: Qipan Li <Qipan.Li@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
The spi-topcliff-pch driver is for a companion chip to the Intel Atom
E600 series processors. These are 32-bit x86 processors so the driver
is only needed on X86_32. Add COMPILE_TEST as an alternative, so that
the driver can still be build-tested elsewhere.
Signed-off-by: Jean Delvare <jdelvare@suse.de>
Signed-off-by: Mark Brown <broonie@linaro.org>
This patch fixes the calculation for determining whether to use FIFO or BLOCK
mode.
Signed-off-by: Andy Gross <agross@codeaurora.org>
Signed-off-by: Mark Brown <broonie@linaro.org>
In case we are doing DMA transfer and the size of the buffer is not multiple
of 4 bytes the driver truncates that to 4-byte boundary and tries to handle
remaining bytes using PIO.
Or that is what it tried to do. What actually happens is that it calls
ALIGN() to the buffer size which aligns it to the next 4-byte boundary
(doesn't truncate). Doing this results 1-3 bytes extra to be transferred.
Furthermore we handle remaining bytes using PIO which results one extra
byte to be transferred. In worst case the driver transfers 4 extra bytes.
While investigating this it turned out that the DMA hardware doesn't even
have such limitation so we can solve this by dropping the code that tries
to handle unaligned bytes.
Reported-by: Chiau Ee Chew <chiau.ee.chew@intel.com>
Reported-by: Hock Leong Kweh <hock.leong.kweh@intel.com>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
The dw_spi_cleanup() function was removed by commit c63f5da008
"spi: dw: Don't call kfree for memory allocated by devm_kzalloc".
commit ec37e8e1f0 "spi: dw: migrate to generic queue infrastructure" added
dw_spi_cleanup() but never use it. So now I got below build warning:
CC [M] drivers/spi/spi-dw.o
drivers/spi/spi-dw.c:609:13: warning: 'dw_spi_cleanup' defined but not used [-Wunused-function]
Signed-off-by: Axel Lin <axel.lin@ingics.com>
Acked-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Mark Brown <broonie@linaro.org>
Make of_device_id array const, because all OF functions
handle it as const.
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
Make of_device_id array const, because all OF functions
handle it as const.
Signed-off-by: Jingoo Han <jg1.han@samsung.com>
Signed-off-by: Mark Brown <broonie@linaro.org>
