linux/drivers/spi/spi-dw-dma.c

657 lines
16 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
spi: dw: Move Non-DMA code to the DW PCIe-SPI driver This is a preparation patch before adding the DW DMA support into the DW SPI MMIO driver. We need to unpin the Non-DMA-specific code from the intended to be generic DW APB SSI DMA code. This isn't that hard, since the most part of the spi-dw-mid.c driver in fact implements a generic DMA interface for the DW SPI controller driver. The only Intel MID specifics concern getting the max frequency from the MRST Clock Control Unit and fetching the DMA controller channels from corresponding PCIe DMA controller. Since first one is related with the SPI interface configuration we moved it' implementation into the DW PCIe-SPI driver module. After that former spi-dw-mid.c file can be just renamed to be the DW SPI DMA module optionally compiled in to the DW APB SSI core driver. Co-developed-by: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Co-developed-by: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Signed-off-by: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Signed-off-by: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-11-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:59 +00:00
* Special handling for DW DMA core
*
* Copyright (c) 2009, 2014 Intel Corporation.
*/
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
#include <linux/completion.h>
#include <linux/dma-mapping.h>
#include <linux/dmaengine.h>
#include <linux/irqreturn.h>
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
#include <linux/jiffies.h>
#include <linux/pci.h>
#include <linux/platform_data/dma-dw.h>
spi: dw: Move Non-DMA code to the DW PCIe-SPI driver This is a preparation patch before adding the DW DMA support into the DW SPI MMIO driver. We need to unpin the Non-DMA-specific code from the intended to be generic DW APB SSI DMA code. This isn't that hard, since the most part of the spi-dw-mid.c driver in fact implements a generic DMA interface for the DW SPI controller driver. The only Intel MID specifics concern getting the max frequency from the MRST Clock Control Unit and fetching the DMA controller channels from corresponding PCIe DMA controller. Since first one is related with the SPI interface configuration we moved it' implementation into the DW PCIe-SPI driver module. After that former spi-dw-mid.c file can be just renamed to be the DW SPI DMA module optionally compiled in to the DW APB SSI core driver. Co-developed-by: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Co-developed-by: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Signed-off-by: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Signed-off-by: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-11-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:59 +00:00
#include <linux/spi/spi.h>
#include <linux/types.h>
#include "spi-dw.h"
#define RX_BUSY 0
#define RX_BURST_LEVEL 16
#define TX_BUSY 1
#define TX_BURST_LEVEL 16
static bool dw_spi_dma_chan_filter(struct dma_chan *chan, void *param)
{
struct dw_dma_slave *s = param;
if (s->dma_dev != chan->device->dev)
return false;
chan->private = s;
return true;
}
static void dw_spi_dma_maxburst_init(struct dw_spi *dws)
{
struct dma_slave_caps caps;
u32 max_burst, def_burst;
int ret;
def_burst = dws->fifo_len / 2;
ret = dma_get_slave_caps(dws->rxchan, &caps);
if (!ret && caps.max_burst)
max_burst = caps.max_burst;
else
max_burst = RX_BURST_LEVEL;
dws->rxburst = min(max_burst, def_burst);
dw_writel(dws, DW_SPI_DMARDLR, dws->rxburst - 1);
ret = dma_get_slave_caps(dws->txchan, &caps);
if (!ret && caps.max_burst)
max_burst = caps.max_burst;
else
max_burst = TX_BURST_LEVEL;
/*
* Having a Rx DMA channel serviced with higher priority than a Tx DMA
* channel might not be enough to provide a well balanced DMA-based
* SPI transfer interface. There might still be moments when the Tx DMA
* channel is occasionally handled faster than the Rx DMA channel.
* That in its turn will eventually cause the SPI Rx FIFO overflow if
* SPI bus speed is high enough to fill the SPI Rx FIFO in before it's
* cleared by the Rx DMA channel. In order to fix the problem the Tx
* DMA activity is intentionally slowed down by limiting the SPI Tx
* FIFO depth with a value twice bigger than the Tx burst length.
*/
dws->txburst = min(max_burst, def_burst);
dw_writel(dws, DW_SPI_DMATDLR, dws->txburst);
}
spi: dw-dma: Add one-by-one SG list entries transfer In case if at least one of the requested DMA engine channels doesn't support the hardware accelerated SG list entries traverse, the DMA driver will most likely work that around by performing the IRQ-based SG list entries resubmission. That might and will cause a problem if the DMA Tx channel is recharged and re-executed before the Rx DMA channel. Due to non-deterministic IRQ-handler execution latency the DMA Tx channel will start pushing data to the SPI bus before the Rx DMA channel is even reinitialized with the next inbound SG list entry. By doing so the DMA Tx channel will implicitly start filling the DW APB SSI Rx FIFO up, which while the DMA Rx channel being recharged and re-executed will eventually be overflown. In order to solve the problem we have to feed the DMA engine with SG list entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs synchronized and prevent the Rx FIFO overflow. Since in general the SPI tx_sg and rx_sg lists may have different number of entries of different lengths (though total length should match) we virtually split the SG-lists to the set of DMA transfers, which length is a minimum of the ordered SG-entries lengths. The solution described above is only executed if a full-duplex SPI transfer is requested and the DMA engine hasn't provided channels with hardware accelerated SG list traverse capability to handle both SG lists at once. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Link: https://lore.kernel.org/r/20200920112322.24585-12-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:22 +00:00
static void dw_spi_dma_sg_burst_init(struct dw_spi *dws)
{
struct dma_slave_caps tx = {0}, rx = {0};
dma_get_slave_caps(dws->txchan, &tx);
dma_get_slave_caps(dws->rxchan, &rx);
if (tx.max_sg_burst > 0 && rx.max_sg_burst > 0)
dws->dma_sg_burst = min(tx.max_sg_burst, rx.max_sg_burst);
else if (tx.max_sg_burst > 0)
dws->dma_sg_burst = tx.max_sg_burst;
else if (rx.max_sg_burst > 0)
dws->dma_sg_burst = rx.max_sg_burst;
else
dws->dma_sg_burst = 0;
}
static int dw_spi_dma_init_mfld(struct device *dev, struct dw_spi *dws)
{
struct dw_dma_slave dma_tx = { .dst_id = 1 }, *tx = &dma_tx;
struct dw_dma_slave dma_rx = { .src_id = 0 }, *rx = &dma_rx;
struct pci_dev *dma_dev;
dma_cap_mask_t mask;
/*
* Get pci device for DMA controller, currently it could only
* be the DMA controller of Medfield
*/
dma_dev = pci_get_device(PCI_VENDOR_ID_INTEL, 0x0827, NULL);
if (!dma_dev)
return -ENODEV;
dma_cap_zero(mask);
dma_cap_set(DMA_SLAVE, mask);
/* 1. Init rx channel */
rx->dma_dev = &dma_dev->dev;
dws->rxchan = dma_request_channel(mask, dw_spi_dma_chan_filter, rx);
if (!dws->rxchan)
goto err_exit;
/* 2. Init tx channel */
tx->dma_dev = &dma_dev->dev;
dws->txchan = dma_request_channel(mask, dw_spi_dma_chan_filter, tx);
if (!dws->txchan)
goto free_rxchan;
dws->master->dma_rx = dws->rxchan;
dws->master->dma_tx = dws->txchan;
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
init_completion(&dws->dma_completion);
dw_spi_dma_maxburst_init(dws);
spi: dw-dma: Add one-by-one SG list entries transfer In case if at least one of the requested DMA engine channels doesn't support the hardware accelerated SG list entries traverse, the DMA driver will most likely work that around by performing the IRQ-based SG list entries resubmission. That might and will cause a problem if the DMA Tx channel is recharged and re-executed before the Rx DMA channel. Due to non-deterministic IRQ-handler execution latency the DMA Tx channel will start pushing data to the SPI bus before the Rx DMA channel is even reinitialized with the next inbound SG list entry. By doing so the DMA Tx channel will implicitly start filling the DW APB SSI Rx FIFO up, which while the DMA Rx channel being recharged and re-executed will eventually be overflown. In order to solve the problem we have to feed the DMA engine with SG list entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs synchronized and prevent the Rx FIFO overflow. Since in general the SPI tx_sg and rx_sg lists may have different number of entries of different lengths (though total length should match) we virtually split the SG-lists to the set of DMA transfers, which length is a minimum of the ordered SG-entries lengths. The solution described above is only executed if a full-duplex SPI transfer is requested and the DMA engine hasn't provided channels with hardware accelerated SG list traverse capability to handle both SG lists at once. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Link: https://lore.kernel.org/r/20200920112322.24585-12-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:22 +00:00
dw_spi_dma_sg_burst_init(dws);
return 0;
free_rxchan:
dma_release_channel(dws->rxchan);
dws->rxchan = NULL;
err_exit:
return -EBUSY;
}
static int dw_spi_dma_init_generic(struct device *dev, struct dw_spi *dws)
{
dws->rxchan = dma_request_slave_channel(dev, "rx");
if (!dws->rxchan)
return -ENODEV;
dws->txchan = dma_request_slave_channel(dev, "tx");
if (!dws->txchan) {
dma_release_channel(dws->rxchan);
dws->rxchan = NULL;
return -ENODEV;
}
dws->master->dma_rx = dws->rxchan;
dws->master->dma_tx = dws->txchan;
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
init_completion(&dws->dma_completion);
dw_spi_dma_maxburst_init(dws);
spi: dw-dma: Add one-by-one SG list entries transfer In case if at least one of the requested DMA engine channels doesn't support the hardware accelerated SG list entries traverse, the DMA driver will most likely work that around by performing the IRQ-based SG list entries resubmission. That might and will cause a problem if the DMA Tx channel is recharged and re-executed before the Rx DMA channel. Due to non-deterministic IRQ-handler execution latency the DMA Tx channel will start pushing data to the SPI bus before the Rx DMA channel is even reinitialized with the next inbound SG list entry. By doing so the DMA Tx channel will implicitly start filling the DW APB SSI Rx FIFO up, which while the DMA Rx channel being recharged and re-executed will eventually be overflown. In order to solve the problem we have to feed the DMA engine with SG list entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs synchronized and prevent the Rx FIFO overflow. Since in general the SPI tx_sg and rx_sg lists may have different number of entries of different lengths (though total length should match) we virtually split the SG-lists to the set of DMA transfers, which length is a minimum of the ordered SG-entries lengths. The solution described above is only executed if a full-duplex SPI transfer is requested and the DMA engine hasn't provided channels with hardware accelerated SG list traverse capability to handle both SG lists at once. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Link: https://lore.kernel.org/r/20200920112322.24585-12-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:22 +00:00
dw_spi_dma_sg_burst_init(dws);
return 0;
}
static void dw_spi_dma_exit(struct dw_spi *dws)
{
if (dws->txchan) {
dmaengine_terminate_sync(dws->txchan);
dma_release_channel(dws->txchan);
}
if (dws->rxchan) {
dmaengine_terminate_sync(dws->rxchan);
dma_release_channel(dws->rxchan);
}
}
static irqreturn_t dw_spi_dma_transfer_handler(struct dw_spi *dws)
{
dw_spi_check_status(dws, false);
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
complete(&dws->dma_completion);
return IRQ_HANDLED;
}
static bool dw_spi_can_dma(struct spi_controller *master,
struct spi_device *spi, struct spi_transfer *xfer)
{
struct dw_spi *dws = spi_controller_get_devdata(master);
return xfer->len > dws->fifo_len;
}
static enum dma_slave_buswidth dw_spi_dma_convert_width(u8 n_bytes)
{
if (n_bytes == 1)
return DMA_SLAVE_BUSWIDTH_1_BYTE;
else if (n_bytes == 2)
return DMA_SLAVE_BUSWIDTH_2_BYTES;
return DMA_SLAVE_BUSWIDTH_UNDEFINED;
}
static int dw_spi_dma_wait(struct dw_spi *dws, unsigned int len, u32 speed)
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
{
unsigned long long ms;
ms = len * MSEC_PER_SEC * BITS_PER_BYTE;
do_div(ms, speed);
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
ms += ms + 200;
if (ms > UINT_MAX)
ms = UINT_MAX;
ms = wait_for_completion_timeout(&dws->dma_completion,
msecs_to_jiffies(ms));
if (ms == 0) {
dev_err(&dws->master->cur_msg->spi->dev,
"DMA transaction timed out\n");
return -ETIMEDOUT;
}
return 0;
}
static inline bool dw_spi_dma_tx_busy(struct dw_spi *dws)
{
return !(dw_readl(dws, DW_SPI_SR) & SR_TF_EMPT);
}
static int dw_spi_dma_wait_tx_done(struct dw_spi *dws,
struct spi_transfer *xfer)
{
int retry = SPI_WAIT_RETRIES;
struct spi_delay delay;
u32 nents;
nents = dw_readl(dws, DW_SPI_TXFLR);
delay.unit = SPI_DELAY_UNIT_SCK;
delay.value = nents * dws->n_bytes * BITS_PER_BYTE;
while (dw_spi_dma_tx_busy(dws) && retry--)
spi_delay_exec(&delay, xfer);
if (retry < 0) {
dev_err(&dws->master->dev, "Tx hanged up\n");
return -EIO;
}
return 0;
}
/*
* dws->dma_chan_busy is set before the dma transfer starts, callback for tx
* channel will clear a corresponding bit.
*/
static void dw_spi_dma_tx_done(void *arg)
{
struct dw_spi *dws = arg;
clear_bit(TX_BUSY, &dws->dma_chan_busy);
if (test_bit(RX_BUSY, &dws->dma_chan_busy))
return;
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
complete(&dws->dma_completion);
}
static int dw_spi_dma_config_tx(struct dw_spi *dws)
{
struct dma_slave_config txconf;
memset(&txconf, 0, sizeof(txconf));
txconf.direction = DMA_MEM_TO_DEV;
txconf.dst_addr = dws->dma_addr;
txconf.dst_maxburst = dws->txburst;
txconf.src_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
txconf.dst_addr_width = dw_spi_dma_convert_width(dws->n_bytes);
txconf.device_fc = false;
return dmaengine_slave_config(dws->txchan, &txconf);
}
static int dw_spi_dma_submit_tx(struct dw_spi *dws, struct scatterlist *sgl,
unsigned int nents)
{
struct dma_async_tx_descriptor *txdesc;
dma_cookie_t cookie;
int ret;
txdesc = dmaengine_prep_slave_sg(dws->txchan, sgl, nents,
DMA_MEM_TO_DEV,
DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
if (!txdesc)
return -ENOMEM;
txdesc->callback = dw_spi_dma_tx_done;
txdesc->callback_param = dws;
cookie = dmaengine_submit(txdesc);
ret = dma_submit_error(cookie);
if (ret) {
dmaengine_terminate_sync(dws->txchan);
return ret;
}
set_bit(TX_BUSY, &dws->dma_chan_busy);
return 0;
}
static inline bool dw_spi_dma_rx_busy(struct dw_spi *dws)
{
return !!(dw_readl(dws, DW_SPI_SR) & SR_RF_NOT_EMPT);
}
static int dw_spi_dma_wait_rx_done(struct dw_spi *dws)
{
int retry = SPI_WAIT_RETRIES;
struct spi_delay delay;
unsigned long ns, us;
u32 nents;
/*
* It's unlikely that DMA engine is still doing the data fetching, but
* if it's let's give it some reasonable time. The timeout calculation
* is based on the synchronous APB/SSI reference clock rate, on a
* number of data entries left in the Rx FIFO, times a number of clock
* periods normally needed for a single APB read/write transaction
* without PREADY signal utilized (which is true for the DW APB SSI
* controller).
*/
nents = dw_readl(dws, DW_SPI_RXFLR);
ns = 4U * NSEC_PER_SEC / dws->max_freq * nents;
if (ns <= NSEC_PER_USEC) {
delay.unit = SPI_DELAY_UNIT_NSECS;
delay.value = ns;
} else {
us = DIV_ROUND_UP(ns, NSEC_PER_USEC);
delay.unit = SPI_DELAY_UNIT_USECS;
delay.value = clamp_val(us, 0, USHRT_MAX);
}
while (dw_spi_dma_rx_busy(dws) && retry--)
spi_delay_exec(&delay, NULL);
if (retry < 0) {
dev_err(&dws->master->dev, "Rx hanged up\n");
return -EIO;
}
return 0;
}
/*
* dws->dma_chan_busy is set before the dma transfer starts, callback for rx
* channel will clear a corresponding bit.
*/
static void dw_spi_dma_rx_done(void *arg)
{
struct dw_spi *dws = arg;
clear_bit(RX_BUSY, &dws->dma_chan_busy);
if (test_bit(TX_BUSY, &dws->dma_chan_busy))
return;
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
complete(&dws->dma_completion);
}
static int dw_spi_dma_config_rx(struct dw_spi *dws)
{
struct dma_slave_config rxconf;
memset(&rxconf, 0, sizeof(rxconf));
rxconf.direction = DMA_DEV_TO_MEM;
rxconf.src_addr = dws->dma_addr;
rxconf.src_maxburst = dws->rxburst;
rxconf.dst_addr_width = DMA_SLAVE_BUSWIDTH_4_BYTES;
rxconf.src_addr_width = dw_spi_dma_convert_width(dws->n_bytes);
rxconf.device_fc = false;
return dmaengine_slave_config(dws->rxchan, &rxconf);
}
static int dw_spi_dma_submit_rx(struct dw_spi *dws, struct scatterlist *sgl,
unsigned int nents)
{
struct dma_async_tx_descriptor *rxdesc;
dma_cookie_t cookie;
int ret;
rxdesc = dmaengine_prep_slave_sg(dws->rxchan, sgl, nents,
DMA_DEV_TO_MEM,
DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
if (!rxdesc)
return -ENOMEM;
rxdesc->callback = dw_spi_dma_rx_done;
rxdesc->callback_param = dws;
cookie = dmaengine_submit(rxdesc);
ret = dma_submit_error(cookie);
if (ret) {
dmaengine_terminate_sync(dws->rxchan);
return ret;
}
set_bit(RX_BUSY, &dws->dma_chan_busy);
return 0;
}
static int dw_spi_dma_setup(struct dw_spi *dws, struct spi_transfer *xfer)
{
u16 imr, dma_ctrl;
int ret;
if (!xfer->tx_buf)
return -EINVAL;
/* Setup DMA channels */
ret = dw_spi_dma_config_tx(dws);
if (ret)
return ret;
if (xfer->rx_buf) {
ret = dw_spi_dma_config_rx(dws);
if (ret)
return ret;
}
/* Set the DMA handshaking interface */
dma_ctrl = SPI_DMA_TDMAE;
if (xfer->rx_buf)
dma_ctrl |= SPI_DMA_RDMAE;
dw_writel(dws, DW_SPI_DMACR, dma_ctrl);
/* Set the interrupt mask */
imr = SPI_INT_TXOI;
if (xfer->rx_buf)
imr |= SPI_INT_RXUI | SPI_INT_RXOI;
spi_umask_intr(dws, imr);
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
reinit_completion(&dws->dma_completion);
dws->transfer_handler = dw_spi_dma_transfer_handler;
return 0;
}
static int dw_spi_dma_transfer_all(struct dw_spi *dws,
struct spi_transfer *xfer)
{
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
int ret;
/* Submit the DMA Tx transfer */
ret = dw_spi_dma_submit_tx(dws, xfer->tx_sg.sgl, xfer->tx_sg.nents);
if (ret)
spi: dw-dma: Move DMAC register cleanup to DMA transfer method DW APB SSI DMA driver doesn't use the native SPI core wait API since commit bdbdf0f06337 ("spi: dw: Locally wait for the DMA transfers completion"). Due to that the driver can now clear the DMAC register in a single place synchronously with the DMA transactions completion or failure. After that all the possible code paths are still covered: 1) DMA completion callbacks are executed in case if the corresponding DMA transactions are finished. When they are, one of them will eventually wake the SPI messages pump kernel thread and dw_spi_dma_transfer_all() method will clean the DMAC register as implied by this patch. 2) dma_stop is called when the SPI core detects an error either returned from the transfer_one() callback or set in the SPI message status field. Both types of errors will be noticed by the dw_spi_dma_transfer_all() method. 3) dma_exit is called when either SPI controller driver or the corresponding device is removed. In any case the SPI core will first flush the SPI messages pump kernel thread, so any pending or in-fly SPI transfers will be finished before that. Due to all of that let's simplify the DW APB SSI DMA driver a bit and move the DMAC register cleanup to a single place in the dw_spi_dma_transfer_all() method. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Link: https://lore.kernel.org/r/20200920112322.24585-10-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:20 +00:00
goto err_clear_dmac;
/* Submit the DMA Rx transfer if required */
if (xfer->rx_buf) {
ret = dw_spi_dma_submit_rx(dws, xfer->rx_sg.sgl,
xfer->rx_sg.nents);
if (ret)
spi: dw-dma: Move DMAC register cleanup to DMA transfer method DW APB SSI DMA driver doesn't use the native SPI core wait API since commit bdbdf0f06337 ("spi: dw: Locally wait for the DMA transfers completion"). Due to that the driver can now clear the DMAC register in a single place synchronously with the DMA transactions completion or failure. After that all the possible code paths are still covered: 1) DMA completion callbacks are executed in case if the corresponding DMA transactions are finished. When they are, one of them will eventually wake the SPI messages pump kernel thread and dw_spi_dma_transfer_all() method will clean the DMAC register as implied by this patch. 2) dma_stop is called when the SPI core detects an error either returned from the transfer_one() callback or set in the SPI message status field. Both types of errors will be noticed by the dw_spi_dma_transfer_all() method. 3) dma_exit is called when either SPI controller driver or the corresponding device is removed. In any case the SPI core will first flush the SPI messages pump kernel thread, so any pending or in-fly SPI transfers will be finished before that. Due to all of that let's simplify the DW APB SSI DMA driver a bit and move the DMAC register cleanup to a single place in the dw_spi_dma_transfer_all() method. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Link: https://lore.kernel.org/r/20200920112322.24585-10-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:20 +00:00
goto err_clear_dmac;
/* rx must be started before tx due to spi instinct */
dma_async_issue_pending(dws->rxchan);
}
dma_async_issue_pending(dws->txchan);
ret = dw_spi_dma_wait(dws, xfer->len, xfer->effective_speed_hz);
spi: dw-dma: Move DMAC register cleanup to DMA transfer method DW APB SSI DMA driver doesn't use the native SPI core wait API since commit bdbdf0f06337 ("spi: dw: Locally wait for the DMA transfers completion"). Due to that the driver can now clear the DMAC register in a single place synchronously with the DMA transactions completion or failure. After that all the possible code paths are still covered: 1) DMA completion callbacks are executed in case if the corresponding DMA transactions are finished. When they are, one of them will eventually wake the SPI messages pump kernel thread and dw_spi_dma_transfer_all() method will clean the DMAC register as implied by this patch. 2) dma_stop is called when the SPI core detects an error either returned from the transfer_one() callback or set in the SPI message status field. Both types of errors will be noticed by the dw_spi_dma_transfer_all() method. 3) dma_exit is called when either SPI controller driver or the corresponding device is removed. In any case the SPI core will first flush the SPI messages pump kernel thread, so any pending or in-fly SPI transfers will be finished before that. Due to all of that let's simplify the DW APB SSI DMA driver a bit and move the DMAC register cleanup to a single place in the dw_spi_dma_transfer_all() method. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Link: https://lore.kernel.org/r/20200920112322.24585-10-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:20 +00:00
err_clear_dmac:
dw_writel(dws, DW_SPI_DMACR, 0);
return ret;
}
spi: dw-dma: Add one-by-one SG list entries transfer In case if at least one of the requested DMA engine channels doesn't support the hardware accelerated SG list entries traverse, the DMA driver will most likely work that around by performing the IRQ-based SG list entries resubmission. That might and will cause a problem if the DMA Tx channel is recharged and re-executed before the Rx DMA channel. Due to non-deterministic IRQ-handler execution latency the DMA Tx channel will start pushing data to the SPI bus before the Rx DMA channel is even reinitialized with the next inbound SG list entry. By doing so the DMA Tx channel will implicitly start filling the DW APB SSI Rx FIFO up, which while the DMA Rx channel being recharged and re-executed will eventually be overflown. In order to solve the problem we have to feed the DMA engine with SG list entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs synchronized and prevent the Rx FIFO overflow. Since in general the SPI tx_sg and rx_sg lists may have different number of entries of different lengths (though total length should match) we virtually split the SG-lists to the set of DMA transfers, which length is a minimum of the ordered SG-entries lengths. The solution described above is only executed if a full-duplex SPI transfer is requested and the DMA engine hasn't provided channels with hardware accelerated SG list traverse capability to handle both SG lists at once. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Link: https://lore.kernel.org/r/20200920112322.24585-12-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:22 +00:00
/*
* In case if at least one of the requested DMA channels doesn't support the
* hardware accelerated SG list entries traverse, the DMA driver will most
* likely work that around by performing the IRQ-based SG list entries
* resubmission. That might and will cause a problem if the DMA Tx channel is
* recharged and re-executed before the Rx DMA channel. Due to
* non-deterministic IRQ-handler execution latency the DMA Tx channel will
* start pushing data to the SPI bus before the Rx DMA channel is even
* reinitialized with the next inbound SG list entry. By doing so the DMA Tx
* channel will implicitly start filling the DW APB SSI Rx FIFO up, which while
* the DMA Rx channel being recharged and re-executed will eventually be
* overflown.
*
* In order to solve the problem we have to feed the DMA engine with SG list
* entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs
* synchronized and prevent the Rx FIFO overflow. Since in general the tx_sg
* and rx_sg lists may have different number of entries of different lengths
* (though total length should match) let's virtually split the SG-lists to the
* set of DMA transfers, which length is a minimum of the ordered SG-entries
* lengths. An ASCII-sketch of the implemented algo is following:
* xfer->len
* |___________|
* tx_sg list: |___|____|__|
* rx_sg list: |_|____|____|
* DMA transfers: |_|_|__|_|__|
*
* Note in order to have this workaround solving the denoted problem the DMA
* engine driver should properly initialize the max_sg_burst capability and set
* the DMA device max segment size parameter with maximum data block size the
* DMA engine supports.
*/
static int dw_spi_dma_transfer_one(struct dw_spi *dws,
struct spi_transfer *xfer)
{
struct scatterlist *tx_sg = NULL, *rx_sg = NULL, tx_tmp, rx_tmp;
unsigned int tx_len = 0, rx_len = 0;
unsigned int base, len;
int ret;
sg_init_table(&tx_tmp, 1);
sg_init_table(&rx_tmp, 1);
for (base = 0, len = 0; base < xfer->len; base += len) {
/* Fetch next Tx DMA data chunk */
if (!tx_len) {
tx_sg = !tx_sg ? &xfer->tx_sg.sgl[0] : sg_next(tx_sg);
sg_dma_address(&tx_tmp) = sg_dma_address(tx_sg);
tx_len = sg_dma_len(tx_sg);
}
/* Fetch next Rx DMA data chunk */
if (!rx_len) {
rx_sg = !rx_sg ? &xfer->rx_sg.sgl[0] : sg_next(rx_sg);
sg_dma_address(&rx_tmp) = sg_dma_address(rx_sg);
rx_len = sg_dma_len(rx_sg);
}
len = min(tx_len, rx_len);
sg_dma_len(&tx_tmp) = len;
sg_dma_len(&rx_tmp) = len;
/* Submit DMA Tx transfer */
ret = dw_spi_dma_submit_tx(dws, &tx_tmp, 1);
if (ret)
break;
/* Submit DMA Rx transfer */
ret = dw_spi_dma_submit_rx(dws, &rx_tmp, 1);
if (ret)
break;
/* Rx must be started before Tx due to SPI instinct */
dma_async_issue_pending(dws->rxchan);
dma_async_issue_pending(dws->txchan);
/*
* Here we only need to wait for the DMA transfer to be
* finished since SPI controller is kept enabled during the
* procedure this loop implements and there is no risk to lose
* data left in the Tx/Rx FIFOs.
*/
ret = dw_spi_dma_wait(dws, len, xfer->effective_speed_hz);
if (ret)
break;
reinit_completion(&dws->dma_completion);
sg_dma_address(&tx_tmp) += len;
sg_dma_address(&rx_tmp) += len;
tx_len -= len;
rx_len -= len;
}
dw_writel(dws, DW_SPI_DMACR, 0);
return ret;
}
static int dw_spi_dma_transfer(struct dw_spi *dws, struct spi_transfer *xfer)
{
spi: dw-dma: Add one-by-one SG list entries transfer In case if at least one of the requested DMA engine channels doesn't support the hardware accelerated SG list entries traverse, the DMA driver will most likely work that around by performing the IRQ-based SG list entries resubmission. That might and will cause a problem if the DMA Tx channel is recharged and re-executed before the Rx DMA channel. Due to non-deterministic IRQ-handler execution latency the DMA Tx channel will start pushing data to the SPI bus before the Rx DMA channel is even reinitialized with the next inbound SG list entry. By doing so the DMA Tx channel will implicitly start filling the DW APB SSI Rx FIFO up, which while the DMA Rx channel being recharged and re-executed will eventually be overflown. In order to solve the problem we have to feed the DMA engine with SG list entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs synchronized and prevent the Rx FIFO overflow. Since in general the SPI tx_sg and rx_sg lists may have different number of entries of different lengths (though total length should match) we virtually split the SG-lists to the set of DMA transfers, which length is a minimum of the ordered SG-entries lengths. The solution described above is only executed if a full-duplex SPI transfer is requested and the DMA engine hasn't provided channels with hardware accelerated SG list traverse capability to handle both SG lists at once. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Link: https://lore.kernel.org/r/20200920112322.24585-12-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:22 +00:00
unsigned int nents;
int ret;
spi: dw-dma: Add one-by-one SG list entries transfer In case if at least one of the requested DMA engine channels doesn't support the hardware accelerated SG list entries traverse, the DMA driver will most likely work that around by performing the IRQ-based SG list entries resubmission. That might and will cause a problem if the DMA Tx channel is recharged and re-executed before the Rx DMA channel. Due to non-deterministic IRQ-handler execution latency the DMA Tx channel will start pushing data to the SPI bus before the Rx DMA channel is even reinitialized with the next inbound SG list entry. By doing so the DMA Tx channel will implicitly start filling the DW APB SSI Rx FIFO up, which while the DMA Rx channel being recharged and re-executed will eventually be overflown. In order to solve the problem we have to feed the DMA engine with SG list entries one-by-one. It shall keep the DW APB SSI Tx and Rx FIFOs synchronized and prevent the Rx FIFO overflow. Since in general the SPI tx_sg and rx_sg lists may have different number of entries of different lengths (though total length should match) we virtually split the SG-lists to the set of DMA transfers, which length is a minimum of the ordered SG-entries lengths. The solution described above is only executed if a full-duplex SPI transfer is requested and the DMA engine hasn't provided channels with hardware accelerated SG list traverse capability to handle both SG lists at once. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Link: https://lore.kernel.org/r/20200920112322.24585-12-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-09-20 11:23:22 +00:00
nents = max(xfer->tx_sg.nents, xfer->rx_sg.nents);
/*
* Execute normal DMA-based transfer (which submits the Rx and Tx SG
* lists directly to the DMA engine at once) if either full hardware
* accelerated SG list traverse is supported by both channels, or the
* Tx-only SPI transfer is requested, or the DMA engine is capable to
* handle both SG lists on hardware accelerated basis.
*/
if (!dws->dma_sg_burst || !xfer->rx_buf || nents <= dws->dma_sg_burst)
ret = dw_spi_dma_transfer_all(dws, xfer);
else
ret = dw_spi_dma_transfer_one(dws, xfer);
spi: dw: Locally wait for the DMA transfers completion In general each DMA-based SPI transfer can be split up into two stages: DMA data transmission/reception and SPI-bus transmission/reception. DMA asynchronous transactions completion can be tracked by means of the DMA async Tx-descriptor completion callback. But that callback being called indicates that the DMA transfer has been finished, it doesn't mean that SPI data transmission is also done. Moreover in fact it isn't for at least Tx-only SPI transfers. Upon DMA transfer completion some data is left in the Tx FIFO and being pushed out by the SPI controller. So in order to make sure that an SPI transfer is completely pushed to the SPI-bus, the driver has to wait for both DMA transaction and the SPI-bus transmission/reception are finished. Note if there is a way to asynchronously track the former event by means of the DMA async Tx callback, there isn't easy one for the later (IRQ-based solution won't work since SPI controller doesn't notify about Rx FIFO being empty). The DMA transfer completion callback isn't suitable to wait for the SPI controller activity finish either. The callback might (in case of DW DMAC it will) be called in the tasklet context. Waiting for the SPI controller to complete the transfer might take a considerable amount of time since SPI-bus might be pretty slow. In this case delaying the execution in the tasklet atomic context might cause significant system performance drop. So to speak the best option we've got to solve the problem is to consequently wait for both stages being finished in the locally implemented SPI transfer execution procedure even if it costs us of the local wait-function re-implementation. In this case we don't need to use the SPI-core transfer-wait functionality, but we'll make sure that all DMA and SPI-bus transactions are completely finished before the SPI-core transfer_one callback returns. In this commit we provide an implementation of the DMA-transfers completion wait functionality. The DW APB SSI DMA-specific SPI transfer_one function waits for both Tx and Rx DMA transfers being finished, and only then exits with zero returned signalling to the SPI core that the SPI transfer is finished. This implementation is fully equivalent to the currently used DMA-execution-SPI-core-wait algorithm. The SPI-bus transmission/reception wait methods will be added in the follow-up commits. Signed-off-by: Serge Semin <Sergey.Semin@baikalelectronics.ru> Cc: Georgy Vlasov <Georgy.Vlasov@baikalelectronics.ru> Cc: Ramil Zaripov <Ramil.Zaripov@baikalelectronics.ru> Cc: Alexey Malahov <Alexey.Malahov@baikalelectronics.ru> Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de> Cc: Arnd Bergmann <arnd@arndb.de> Cc: Feng Tang <feng.tang@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Rob Herring <robh+dt@kernel.org> Cc: linux-mips@vger.kernel.org Cc: devicetree@vger.kernel.org Link: https://lore.kernel.org/r/20200529131205.31838-4-Sergey.Semin@baikalelectronics.ru Signed-off-by: Mark Brown <broonie@kernel.org>
2020-05-29 13:11:52 +00:00
if (ret)
return ret;
if (dws->master->cur_msg->status == -EINPROGRESS) {
ret = dw_spi_dma_wait_tx_done(dws, xfer);
if (ret)
return ret;
}
if (xfer->rx_buf && dws->master->cur_msg->status == -EINPROGRESS)
ret = dw_spi_dma_wait_rx_done(dws);
return ret;
}
static void dw_spi_dma_stop(struct dw_spi *dws)
{
if (test_bit(TX_BUSY, &dws->dma_chan_busy)) {
dmaengine_terminate_sync(dws->txchan);
clear_bit(TX_BUSY, &dws->dma_chan_busy);
}
if (test_bit(RX_BUSY, &dws->dma_chan_busy)) {
dmaengine_terminate_sync(dws->rxchan);
clear_bit(RX_BUSY, &dws->dma_chan_busy);
}
}
static const struct dw_spi_dma_ops dw_spi_dma_mfld_ops = {
.dma_init = dw_spi_dma_init_mfld,
.dma_exit = dw_spi_dma_exit,
.dma_setup = dw_spi_dma_setup,
.can_dma = dw_spi_can_dma,
.dma_transfer = dw_spi_dma_transfer,
.dma_stop = dw_spi_dma_stop,
};
void dw_spi_dma_setup_mfld(struct dw_spi *dws)
{
dws->dma_ops = &dw_spi_dma_mfld_ops;
}
EXPORT_SYMBOL_GPL(dw_spi_dma_setup_mfld);
static const struct dw_spi_dma_ops dw_spi_dma_generic_ops = {
.dma_init = dw_spi_dma_init_generic,
.dma_exit = dw_spi_dma_exit,
.dma_setup = dw_spi_dma_setup,
.can_dma = dw_spi_can_dma,
.dma_transfer = dw_spi_dma_transfer,
.dma_stop = dw_spi_dma_stop,
};
void dw_spi_dma_setup_generic(struct dw_spi *dws)
{
dws->dma_ops = &dw_spi_dma_generic_ops;
}
EXPORT_SYMBOL_GPL(dw_spi_dma_setup_generic);