[PATCH v2] spi: atmel: Implements transfers with bounce buffer
From: Radu Pirea
Date: Tue Dec 12 2017 - 10:37:20 EST
This patch enables DMA transfers for Atmel SAM9 SoCs and implements a bounce
buffer for transfers which have vmalloc allocated buffers. Those buffers are
not cache coherent even if they have been transformed into sg lists. UBIFS
is affected by this cache coherency issue.
In this patch I also reverted "spi: atmel: fix corrupted data issue on SAM9
family SoCs"(7094576ccdc3acfe1e06a1e2ab547add375baf7f).
Signed-off-by: Radu Pirea <radu.pirea@xxxxxxxxxxxxx>
---
Please ignore the previous version. I messed up with file names.
drivers/spi/spi-atmel.c | 113 +++++++++++++++++++++++++++++++++++-------------
1 file changed, 84 insertions(+), 29 deletions(-)
diff --git a/drivers/spi/spi-atmel.c b/drivers/spi/spi-atmel.c
index f95da36..198b4cd 100644
--- a/drivers/spi/spi-atmel.c
+++ b/drivers/spi/spi-atmel.c
@@ -291,6 +291,10 @@ struct atmel_spi {
struct spi_transfer *current_transfer;
int current_remaining_bytes;
int done_status;
+ dma_addr_t dma_addr_rx_bbuf;
+ dma_addr_t dma_addr_tx_bbuf;
+ void *addr_rx_bbuf;
+ void *addr_tx_bbuf;
struct completion xfer_completion;
@@ -436,6 +440,11 @@ static void atmel_spi_unlock(struct atmel_spi *as) __releases(&as->lock)
spin_unlock_irqrestore(&as->lock, as->flags);
}
+static inline bool atmel_spi_is_vmalloc_xfer(struct spi_transfer *xfer)
+{
+ return is_vmalloc_addr(xfer->tx_buf) || is_vmalloc_addr(xfer->rx_buf);
+}
+
static inline bool atmel_spi_use_dma(struct atmel_spi *as,
struct spi_transfer *xfer)
{
@@ -448,7 +457,12 @@ static bool atmel_spi_can_dma(struct spi_master *master,
{
struct atmel_spi *as = spi_master_get_devdata(master);
- return atmel_spi_use_dma(as, xfer);
+ if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5))
+ return atmel_spi_use_dma(as, xfer) &&
+ !atmel_spi_is_vmalloc_xfer(xfer);
+ else
+ return atmel_spi_use_dma(as, xfer);
+
}
static int atmel_spi_dma_slave_config(struct atmel_spi *as,
@@ -594,6 +608,11 @@ static void dma_callback(void *data)
struct spi_master *master = data;
struct atmel_spi *as = spi_master_get_devdata(master);
+ if (is_vmalloc_addr(as->current_transfer->rx_buf) &&
+ IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
+ memcpy(as->current_transfer->rx_buf, as->addr_rx_bbuf,
+ as->current_transfer->len);
+ }
complete(&as->xfer_completion);
}
@@ -744,17 +763,41 @@ static int atmel_spi_next_xfer_dma_submit(struct spi_master *master,
goto err_exit;
/* Send both scatterlists */
- rxdesc = dmaengine_prep_slave_sg(rxchan,
- xfer->rx_sg.sgl, xfer->rx_sg.nents,
- DMA_FROM_DEVICE,
- DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (atmel_spi_is_vmalloc_xfer(xfer) &&
+ IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
+ rxdesc = dmaengine_prep_slave_single(rxchan,
+ as->dma_addr_rx_bbuf,
+ xfer->len,
+ DMA_FROM_DEVICE,
+ DMA_PREP_INTERRUPT |
+ DMA_CTRL_ACK);
+ } else {
+ rxdesc = dmaengine_prep_slave_sg(rxchan,
+ xfer->rx_sg.sgl,
+ xfer->rx_sg.nents,
+ DMA_FROM_DEVICE,
+ DMA_PREP_INTERRUPT |
+ DMA_CTRL_ACK);
+ }
if (!rxdesc)
goto err_dma;
- txdesc = dmaengine_prep_slave_sg(txchan,
- xfer->tx_sg.sgl, xfer->tx_sg.nents,
- DMA_TO_DEVICE,
- DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (atmel_spi_is_vmalloc_xfer(xfer) &&
+ IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
+ memcpy(as->addr_tx_bbuf, xfer->tx_buf, xfer->len);
+ txdesc = dmaengine_prep_slave_single(txchan,
+ as->dma_addr_tx_bbuf,
+ xfer->len, DMA_TO_DEVICE,
+ DMA_PREP_INTERRUPT |
+ DMA_CTRL_ACK);
+ } else {
+ txdesc = dmaengine_prep_slave_sg(txchan,
+ xfer->tx_sg.sgl,
+ xfer->tx_sg.nents,
+ DMA_TO_DEVICE,
+ DMA_PREP_INTERRUPT |
+ DMA_CTRL_ACK);
+ }
if (!txdesc)
goto err_dma;
@@ -1426,27 +1469,7 @@ static void atmel_get_caps(struct atmel_spi *as)
as->caps.is_spi2 = version > 0x121;
as->caps.has_wdrbt = version >= 0x210;
-#ifdef CONFIG_SOC_SAM_V4_V5
- /*
- * Atmel SoCs based on ARM9 (SAM9x) cores should not use spi_map_buf()
- * since this later function tries to map buffers with dma_map_sg()
- * even if they have not been allocated inside DMA-safe areas.
- * On SoCs based on Cortex A5 (SAMA5Dx), it works anyway because for
- * those ARM cores, the data cache follows the PIPT model.
- * Also the L2 cache controller of SAMA5D2 uses the PIPT model too.
- * In case of PIPT caches, there cannot be cache aliases.
- * However on ARM9 cores, the data cache follows the VIVT model, hence
- * the cache aliases issue can occur when buffers are allocated from
- * DMA-unsafe areas, by vmalloc() for instance, where cache coherency is
- * not taken into account or at least not handled completely (cache
- * lines of aliases are not invalidated).
- * This is not a theorical issue: it was reproduced when trying to mount
- * a UBI file-system on a at91sam9g35ek board.
- */
- as->caps.has_dma_support = false;
-#else
as->caps.has_dma_support = version >= 0x212;
-#endif
as->caps.has_pdc_support = version < 0x212;
}
@@ -1592,6 +1615,30 @@ static int atmel_spi_probe(struct platform_device *pdev)
as->use_pdc = true;
}
+ if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
+ as->addr_rx_bbuf = dma_alloc_coherent(&pdev->dev,
+ SPI_MAX_DMA_XFER,
+ &as->dma_addr_rx_bbuf,
+ GFP_KERNEL | GFP_DMA);
+ if (!as->addr_rx_bbuf) {
+ as->use_dma = false;
+ } else {
+ as->addr_tx_bbuf = dma_alloc_coherent(&pdev->dev,
+ SPI_MAX_DMA_XFER,
+ &as->dma_addr_tx_bbuf,
+ GFP_KERNEL | GFP_DMA);
+ if (!as->addr_tx_bbuf) {
+ as->use_dma = false;
+ dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
+ as->addr_rx_bbuf,
+ as->dma_addr_rx_bbuf);
+ }
+ }
+ if(!as->use_dma)
+ dev_info(master->dev.parent,
+ " can not allocate dma coherent memory\n");
+ }
+
if (as->caps.has_dma_support && !as->use_dma)
dev_info(&pdev->dev, "Atmel SPI Controller using PIO only\n");
@@ -1665,6 +1712,14 @@ static int atmel_spi_remove(struct platform_device *pdev)
if (as->use_dma) {
atmel_spi_stop_dma(master);
atmel_spi_release_dma(master);
+ if (IS_ENABLED(CONFIG_SOC_SAM_V4_V5)) {
+ dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
+ as->addr_tx_bbuf,
+ as->dma_addr_tx_bbuf);
+ dma_free_coherent(&pdev->dev, SPI_MAX_DMA_XFER,
+ as->addr_rx_bbuf,
+ as->dma_addr_rx_bbuf);
+ }
}
spi_writel(as, CR, SPI_BIT(SWRST));
--
2.7.4