[PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment() function
From: Huacai Chen
Date: Tue Sep 19 2017 - 04:52:30 EST
Make dma_get_cache_alignment() to accept a 'dev' argument. As a result,
it can return different alignments due to different devices' I/O cache
coherency. For compatibility, make all existing callers pass a NULL dev
argument.
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Huacai Chen <chenhc@xxxxxxxxxx>
---
drivers/infiniband/hw/mthca/mthca_main.c | 2 +-
drivers/media/v4l2-core/videobuf2-dma-contig.c | 2 +-
drivers/net/ethernet/broadcom/b44.c | 2 +-
drivers/net/ethernet/ibm/emac/core.h | 2 +-
drivers/net/ethernet/mellanox/mlx4/main.c | 2 +-
drivers/spi/spi-qup.c | 4 ++--
drivers/tty/serial/mpsc.c | 16 ++++++++--------
drivers/tty/serial/samsung.c | 14 +++++++-------
include/linux/dma-mapping.h | 14 +++++++++-----
9 files changed, 31 insertions(+), 27 deletions(-)
diff --git a/drivers/infiniband/hw/mthca/mthca_main.c b/drivers/infiniband/hw/mthca/mthca_main.c
index e36a9bc..cac5fac 100644
--- a/drivers/infiniband/hw/mthca/mthca_main.c
+++ b/drivers/infiniband/hw/mthca/mthca_main.c
@@ -416,7 +416,7 @@ static int mthca_init_icm(struct mthca_dev *mdev,
/* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
mdev->limits.reserved_mtts = ALIGN(mdev->limits.reserved_mtts * mdev->limits.mtt_seg_size,
- dma_get_cache_alignment()) / mdev->limits.mtt_seg_size;
+ dma_get_cache_alignment(NULL)) / mdev->limits.mtt_seg_size;
mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
mdev->limits.mtt_seg_size,
diff --git a/drivers/media/v4l2-core/videobuf2-dma-contig.c b/drivers/media/v4l2-core/videobuf2-dma-contig.c
index 9f389f3..7f54739 100644
--- a/drivers/media/v4l2-core/videobuf2-dma-contig.c
+++ b/drivers/media/v4l2-core/videobuf2-dma-contig.c
@@ -484,7 +484,7 @@ static void *vb2_dc_get_userptr(struct device *dev, unsigned long vaddr,
int ret = 0;
struct sg_table *sgt;
unsigned long contig_size;
- unsigned long dma_align = dma_get_cache_alignment();
+ unsigned long dma_align = dma_get_cache_alignment(NULL);
/* Only cache aligned DMA transfers are reliable */
if (!IS_ALIGNED(vaddr | size, dma_align)) {
diff --git a/drivers/net/ethernet/broadcom/b44.c b/drivers/net/ethernet/broadcom/b44.c
index a1125d1..291d6af 100644
--- a/drivers/net/ethernet/broadcom/b44.c
+++ b/drivers/net/ethernet/broadcom/b44.c
@@ -2587,7 +2587,7 @@ static inline void b44_pci_exit(void)
static int __init b44_init(void)
{
- unsigned int dma_desc_align_size = dma_get_cache_alignment();
+ unsigned int dma_desc_align_size = dma_get_cache_alignment(NULL);
int err;
/* Setup paramaters for syncing RX/TX DMA descriptors */
diff --git a/drivers/net/ethernet/ibm/emac/core.h b/drivers/net/ethernet/ibm/emac/core.h
index 369de2c..236bf37 100644
--- a/drivers/net/ethernet/ibm/emac/core.h
+++ b/drivers/net/ethernet/ibm/emac/core.h
@@ -68,7 +68,7 @@ static inline int emac_rx_size(int mtu)
return mal_rx_size(ETH_DATA_LEN + EMAC_MTU_OVERHEAD);
}
-#define EMAC_DMA_ALIGN(x) ALIGN((x), dma_get_cache_alignment())
+#define EMAC_DMA_ALIGN(x) ALIGN((x), dma_get_cache_alignment(NULL))
#define EMAC_RX_SKB_HEADROOM \
EMAC_DMA_ALIGN(CONFIG_IBM_EMAC_RX_SKB_HEADROOM)
diff --git a/drivers/net/ethernet/mellanox/mlx4/main.c b/drivers/net/ethernet/mellanox/mlx4/main.c
index e61c99e..56b1449 100644
--- a/drivers/net/ethernet/mellanox/mlx4/main.c
+++ b/drivers/net/ethernet/mellanox/mlx4/main.c
@@ -1660,7 +1660,7 @@ static int mlx4_init_icm(struct mlx4_dev *dev, struct mlx4_dev_cap *dev_cap,
*/
dev->caps.reserved_mtts =
ALIGN(dev->caps.reserved_mtts * dev->caps.mtt_entry_sz,
- dma_get_cache_alignment()) / dev->caps.mtt_entry_sz;
+ dma_get_cache_alignment(NULL)) / dev->caps.mtt_entry_sz;
err = mlx4_init_icm_table(dev, &priv->mr_table.mtt_table,
init_hca->mtt_base,
diff --git a/drivers/spi/spi-qup.c b/drivers/spi/spi-qup.c
index 974a8ce..0c698c3 100644
--- a/drivers/spi/spi-qup.c
+++ b/drivers/spi/spi-qup.c
@@ -862,7 +862,7 @@ static bool spi_qup_can_dma(struct spi_master *master, struct spi_device *spi,
struct spi_transfer *xfer)
{
struct spi_qup *qup = spi_master_get_devdata(master);
- size_t dma_align = dma_get_cache_alignment();
+ size_t dma_align = dma_get_cache_alignment(NULL);
int n_words;
if (xfer->rx_buf) {
@@ -1038,7 +1038,7 @@ static int spi_qup_probe(struct platform_device *pdev)
master->transfer_one = spi_qup_transfer_one;
master->dev.of_node = pdev->dev.of_node;
master->auto_runtime_pm = true;
- master->dma_alignment = dma_get_cache_alignment();
+ master->dma_alignment = dma_get_cache_alignment(NULL);
master->max_dma_len = SPI_MAX_XFER;
platform_set_drvdata(pdev, master);
diff --git a/drivers/tty/serial/mpsc.c b/drivers/tty/serial/mpsc.c
index 67ffecc..e2f792a 100644
--- a/drivers/tty/serial/mpsc.c
+++ b/drivers/tty/serial/mpsc.c
@@ -81,19 +81,19 @@
* Number of Tx & Rx descriptors must be powers of 2.
*/
#define MPSC_RXR_ENTRIES 32
-#define MPSC_RXRE_SIZE dma_get_cache_alignment()
+#define MPSC_RXRE_SIZE dma_get_cache_alignment(NULL)
#define MPSC_RXR_SIZE (MPSC_RXR_ENTRIES * MPSC_RXRE_SIZE)
-#define MPSC_RXBE_SIZE dma_get_cache_alignment()
+#define MPSC_RXBE_SIZE dma_get_cache_alignment(NULL)
#define MPSC_RXB_SIZE (MPSC_RXR_ENTRIES * MPSC_RXBE_SIZE)
#define MPSC_TXR_ENTRIES 32
-#define MPSC_TXRE_SIZE dma_get_cache_alignment()
+#define MPSC_TXRE_SIZE dma_get_cache_alignment(NULL)
#define MPSC_TXR_SIZE (MPSC_TXR_ENTRIES * MPSC_TXRE_SIZE)
-#define MPSC_TXBE_SIZE dma_get_cache_alignment()
+#define MPSC_TXBE_SIZE dma_get_cache_alignment(NULL)
#define MPSC_TXB_SIZE (MPSC_TXR_ENTRIES * MPSC_TXBE_SIZE)
#define MPSC_DMA_ALLOC_SIZE (MPSC_RXR_SIZE + MPSC_RXB_SIZE + MPSC_TXR_SIZE \
- + MPSC_TXB_SIZE + dma_get_cache_alignment() /* for alignment */)
+ + MPSC_TXB_SIZE + dma_get_cache_alignment(NULL) /* for alignment */)
/* Rx and Tx Ring entry descriptors -- assume entry size is <= cacheline size */
struct mpsc_rx_desc {
@@ -738,7 +738,7 @@ static void mpsc_init_hw(struct mpsc_port_info *pi)
mpsc_brg_init(pi, pi->brg_clk_src);
mpsc_brg_enable(pi);
- mpsc_sdma_init(pi, dma_get_cache_alignment()); /* burst a cacheline */
+ mpsc_sdma_init(pi, dma_get_cache_alignment(NULL)); /* burst a cacheline */
mpsc_sdma_stop(pi);
mpsc_hw_init(pi);
}
@@ -798,8 +798,8 @@ static void mpsc_init_rings(struct mpsc_port_info *pi)
* Descriptors & buffers are multiples of cacheline size and must be
* cacheline aligned.
*/
- dp = ALIGN((u32)pi->dma_region, dma_get_cache_alignment());
- dp_p = ALIGN((u32)pi->dma_region_p, dma_get_cache_alignment());
+ dp = ALIGN((u32)pi->dma_region, dma_get_cache_alignment(NULL));
+ dp_p = ALIGN((u32)pi->dma_region_p, dma_get_cache_alignment(NULL));
/*
* Partition dma region into rx ring descriptor, rx buffers,
diff --git a/drivers/tty/serial/samsung.c b/drivers/tty/serial/samsung.c
index 8aca18c..b40a681 100644
--- a/drivers/tty/serial/samsung.c
+++ b/drivers/tty/serial/samsung.c
@@ -241,7 +241,7 @@ static void enable_tx_dma(struct s3c24xx_uart_port *ourport)
/* Enable tx dma mode */
ucon = rd_regl(port, S3C2410_UCON);
ucon &= ~(S3C64XX_UCON_TXBURST_MASK | S3C64XX_UCON_TXMODE_MASK);
- ucon |= (dma_get_cache_alignment() >= 16) ?
+ ucon |= (dma_get_cache_alignment(NULL) >= 16) ?
S3C64XX_UCON_TXBURST_16 : S3C64XX_UCON_TXBURST_1;
ucon |= S3C64XX_UCON_TXMODE_DMA;
wr_regl(port, S3C2410_UCON, ucon);
@@ -292,7 +292,7 @@ static int s3c24xx_serial_start_tx_dma(struct s3c24xx_uart_port *ourport,
if (ourport->tx_mode != S3C24XX_TX_DMA)
enable_tx_dma(ourport);
- dma->tx_size = count & ~(dma_get_cache_alignment() - 1);
+ dma->tx_size = count & ~(dma_get_cache_alignment(NULL) - 1);
dma->tx_transfer_addr = dma->tx_addr + xmit->tail;
dma_sync_single_for_device(ourport->port.dev, dma->tx_transfer_addr,
@@ -332,7 +332,7 @@ static void s3c24xx_serial_start_next_tx(struct s3c24xx_uart_port *ourport)
if (!ourport->dma || !ourport->dma->tx_chan ||
count < ourport->min_dma_size ||
- xmit->tail & (dma_get_cache_alignment() - 1))
+ xmit->tail & (dma_get_cache_alignment(NULL) - 1))
s3c24xx_serial_start_tx_pio(ourport);
else
s3c24xx_serial_start_tx_dma(ourport, count);
@@ -718,8 +718,8 @@ static irqreturn_t s3c24xx_serial_tx_chars(int irq, void *id)
if (ourport->dma && ourport->dma->tx_chan &&
count >= ourport->min_dma_size) {
- int align = dma_get_cache_alignment() -
- (xmit->tail & (dma_get_cache_alignment() - 1));
+ int align = dma_get_cache_alignment(NULL) -
+ (xmit->tail & (dma_get_cache_alignment(NULL) - 1));
if (count-align >= ourport->min_dma_size) {
dma_count = count-align;
count = align;
@@ -870,7 +870,7 @@ static int s3c24xx_serial_request_dma(struct s3c24xx_uart_port *p)
dma->tx_conf.direction = DMA_MEM_TO_DEV;
dma->tx_conf.dst_addr_width = DMA_SLAVE_BUSWIDTH_1_BYTE;
dma->tx_conf.dst_addr = p->port.mapbase + S3C2410_UTXH;
- if (dma_get_cache_alignment() >= 16)
+ if (dma_get_cache_alignment(NULL) >= 16)
dma->tx_conf.dst_maxburst = 16;
else
dma->tx_conf.dst_maxburst = 1;
@@ -1849,7 +1849,7 @@ static int s3c24xx_serial_probe(struct platform_device *pdev)
* so find minimal transfer size suitable for DMA mode
*/
ourport->min_dma_size = max_t(int, ourport->port.fifosize,
- dma_get_cache_alignment());
+ dma_get_cache_alignment(NULL));
dbg("%s: initialising port %p...\n", __func__, ourport);
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 08da227..ef70b0d 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -707,12 +707,16 @@ static inline int device_is_coherent(struct device *dev)
return 1; /* compatible behavior */
}
-static inline int dma_get_cache_alignment(void)
-{
-#ifdef ARCH_DMA_MINALIGN
- return ARCH_DMA_MINALIGN;
+#ifndef ARCH_DMA_MINALIGN
+#define ARCH_DMA_MINALIGN 1
#endif
- return 1;
+
+static inline int dma_get_cache_alignment(struct device *dev)
+{
+ if (dev && device_is_coherent(dev))
+ return 1;
+ else /* dev is NULL or noncoherent */
+ return ARCH_DMA_MINALIGN;
}
#endif
--
2.7.0