Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1734750 > unrolled thread

[PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as a helper

Back to article view | Back to linux.kernel

Contents

  [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as a helper Huacai Chen <chenhc@lemote.com> - 2017-09-19 11:00 +0200
    [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment() function Huacai Chen <chenhc@lemote.com> - 2017-09-19 11:00 +0200
      Re: [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment()         function Christoph Hellwig <hch@lst.de> - 2017-09-19 17:10 +0200
        Re: [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment()function "陈华才" <chenhc@lemote.com> - 2017-09-21 06:30 +0200
          Re: [PATCH V6 2/3] dma-mapping: Rework         dma_get_cache_alignment()function Christoph Hellwig <hch@lst.de> - 2017-09-21 16:40 +0200
    Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as a  helper Robin Murphy <robin.murphy@arm.com> - 2017-09-21 12:50 +0200
      Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as ahelper "陈华才" <chenhc@lemote.com> - 2017-09-22 04:20 +0200
        Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as  ahelper Robin Murphy <robin.murphy@arm.com> - 2017-09-22 15:50 +0200
          Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as         ahelper Christoph Hellwig <hch@lst.de> - 2017-09-22 15:50 +0200

#1734750 — [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as a helper

We will use device_is_coherent() as a helper function, which will be
used in the next patch.

There is a MIPS-specific plat_device_is_coherent(), but we need a more
generic solution, so add and use a new function pointer in dma_map_ops.

Cc: stable@vger.kernel.org
Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 arch/mips/cavium-octeon/dma-octeon.c               |  3 ++-
 arch/mips/include/asm/mach-generic/dma-coherence.h |  6 +++---
 arch/mips/loongson64/common/dma-swiotlb.c          |  1 +
 arch/mips/mm/dma-default.c                         |  3 ++-
 arch/mips/netlogic/common/nlm-dma.c                |  3 ++-
 include/linux/dma-mapping.h                        | 10 ++++++++++
 6 files changed, 20 insertions(+), 6 deletions(-)

diff --git a/arch/mips/cavium-octeon/dma-octeon.c b/arch/mips/cavium-octeon/dma-octeon.c
index c64bd87..cd1a133 100644
--- a/arch/mips/cavium-octeon/dma-octeon.c
+++ b/arch/mips/cavium-octeon/dma-octeon.c
@@ -324,7 +324,8 @@ static struct octeon_dma_map_ops _octeon_pci_dma_map_ops = {
 		.sync_sg_for_cpu = swiotlb_sync_sg_for_cpu,
 		.sync_sg_for_device = octeon_dma_sync_sg_for_device,
 		.mapping_error = swiotlb_dma_mapping_error,
-		.dma_supported = swiotlb_dma_supported
+		.dma_supported = swiotlb_dma_supported,
+		.device_is_coherent = plat_device_is_coherent
 	},
 };
 
diff --git a/arch/mips/loongson64/common/dma-swiotlb.c b/arch/mips/loongson64/common/dma-swiotlb.c
index 34486c1..c758d9b 100644
--- a/arch/mips/loongson64/common/dma-swiotlb.c
+++ b/arch/mips/loongson64/common/dma-swiotlb.c
@@ -119,6 +119,7 @@ static const struct dma_map_ops loongson_dma_map_ops = {
 	.sync_sg_for_device = loongson_dma_sync_sg_for_device,
 	.mapping_error = swiotlb_dma_mapping_error,
 	.dma_supported = loongson_dma_supported,
+	.device_is_coherent = plat_device_is_coherent
 };
 
 void __init plat_swiotlb_setup(void)
diff --git a/arch/mips/mm/dma-default.c b/arch/mips/mm/dma-default.c
index c01bd20..6e18301 100644
--- a/arch/mips/mm/dma-default.c
+++ b/arch/mips/mm/dma-default.c
@@ -407,7 +407,8 @@ static const struct dma_map_ops mips_default_dma_map_ops = {
 	.sync_sg_for_cpu = mips_dma_sync_sg_for_cpu,
 	.sync_sg_for_device = mips_dma_sync_sg_for_device,
 	.mapping_error = mips_dma_mapping_error,
-	.dma_supported = mips_dma_supported
+	.dma_supported = mips_dma_supported,
+	.device_is_coherent = plat_device_is_coherent
 };
 
 const struct dma_map_ops *mips_dma_map_ops = &mips_default_dma_map_ops;
diff --git a/arch/mips/netlogic/common/nlm-dma.c b/arch/mips/netlogic/common/nlm-dma.c
index 0ec9d9d..aa11b27 100644
--- a/arch/mips/netlogic/common/nlm-dma.c
+++ b/arch/mips/netlogic/common/nlm-dma.c
@@ -79,7 +79,8 @@ const struct dma_map_ops nlm_swiotlb_dma_ops = {
 	.sync_sg_for_cpu = swiotlb_sync_sg_for_cpu,
 	.sync_sg_for_device = swiotlb_sync_sg_for_device,
 	.mapping_error = swiotlb_dma_mapping_error,
-	.dma_supported = swiotlb_dma_supported
+	.dma_supported = swiotlb_dma_supported,
+	.device_is_coherent = plat_device_is_coherent
 };
 
 void __init plat_swiotlb_setup(void)
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 29ce981..08da227 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -131,6 +131,7 @@ struct dma_map_ops {
 #ifdef ARCH_HAS_DMA_GET_REQUIRED_MASK
 	u64 (*get_required_mask)(struct device *dev);
 #endif
+	int (*device_is_coherent)(struct device *dev);
 	int is_phys;
 };
 
@@ -697,6 +698,15 @@ static inline void *dma_zalloc_coherent(struct device *dev, size_t size,
 }
 
 #ifdef CONFIG_HAS_DMA
+static inline int device_is_coherent(struct device *dev)
+{
+	const struct dma_map_ops *ops = get_dma_ops(dev);
+	if (ops && ops->device_is_coherent)
+		return ops->device_is_coherent(dev);
+	else
+		return 1;    /* compatible behavior */
+}
+
 static inline int dma_get_cache_alignment(void)
 {
 #ifdef ARCH_DMA_MINALIGN
-- 
2.7.0

[toc] | [next] | [standalone]

#1734753 — [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment() function

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@vger.kernel.org
Signed-off-by: Huacai Chen <chenhc@lemote.com>
---
 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

[toc] | [prev] | [next] | [standalone]

#1734992 — Re: [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment() function

>  	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,

Please pass the actually relevant struct device for each call.

[toc] | [prev] | [next] | [standalone]

#1736350 — Re: [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment()function

Hi, Christoph,

I have changed dma_get_cache_alignment's return value, and I don't know whether those drivers want to return ARCH_DMA_MINALIGN unconditionally. So I pass a NULL for those drivers, in order to keep their old behavior.
 
Huacai
 
------------------ Original ------------------
From:  "Christoph Hellwig"<hch@lst.de>;
Date:  Tue, Sep 19, 2017 11:02 PM
To:  "Huacai Chen"<chenhc@lemote.com>; 
Cc:  "Christoph Hellwig"<hch@lst.de>; "Marek Szyprowski"<m.szyprowski@samsung.com>; "Robin Murphy"<robin.murphy@arm.com>; "Andrew Morton"<akpm@linux-foundation.org>; "Fuxin Zhang"<zhangfx@lemote.com>; "linux-kernel"<linux-kernel@vger.kernel.org>; "James E . J . Bottomley"<jejb@linux.vnet.ibm.com>; "Martin K . Petersen"<martin.petersen@oracle.com>; "linux-scsi"<linux-scsi@vger.kernel.org>; "stable"<stable@vger.kernel.org>; 
Subject:  Re: [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment()function

 
>  	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,

Please pass the actually relevant struct device for each call.

[toc] | [prev] | [next] | [standalone]

#1736717 — Re: [PATCH V6 2/3] dma-mapping: Rework dma_get_cache_alignment()function

On Thu, Sep 21, 2017 at 12:28:25PM +0800, 陈华才 wrote:
> Hi, Christoph,
> 
> I have changed dma_get_cache_alignment's return value, and I don't know whether those drivers want to return ARCH_DMA_MINALIGN unconditionally. So I pass a NULL for those drivers, in order to keep their old behavior.

Per our documentation yes, they do want ARCH_DMA_MINALIGN.  Please
Cc all the driver maintainers on your updated patch so that they can
review it, though.

[toc] | [prev] | [next] | [standalone]

#1736530 — Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as a helper

On 19/09/17 09:52, Huacai Chen wrote:
> We will use device_is_coherent() as a helper function, which will be
> used in the next patch.
> 
> There is a MIPS-specific plat_device_is_coherent(), but we need a more
> generic solution, so add and use a new function pointer in dma_map_ops.

I think we're heading in the right direction with the series, but I
still don't like this patch. I can pretty much guarantee that driver
authors *will* abuse a generic device_is_coherent() API to mean "I can
skip other DMA API calls and just use virt_to_phys()".

I think it would be far better to allow architectures to provide their
own override of dma_get_cache_alignment(), and let the coherency detail
remain internal to the relevant arch implementations.

[...]
> @@ -697,6 +698,15 @@ static inline void *dma_zalloc_coherent(struct device *dev, size_t size,
>  }
>  
>  #ifdef CONFIG_HAS_DMA
> +static inline int device_is_coherent(struct device *dev)
> +{
> +	const struct dma_map_ops *ops = get_dma_ops(dev);
> +	if (ops && ops->device_is_coherent)
> +		return ops->device_is_coherent(dev);
> +	else
> +		return 1;    /* compatible behavior */

That is also quite scary - if someone now adds a new
dma_get_cache_alignemnt() call and dutifully passes a non-NULL device,
they will now get back an alignment of 1 on all non-coherent platforms
except MIPS: hello data corruption.

Robin.

> +}
> +
>  static inline int dma_get_cache_alignment(void)
>  {
>  #ifdef ARCH_DMA_MINALIGN
> 

[toc] | [prev] | [next] | [standalone]

#1737136 — Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as ahelper

Hi, Robin,

Before 2.6.36 dma_get_cache_alignment is arch-dependent, and it is unified in commit 4565f0170dfc849b3629c27d7 ("dma-mapping: unify dma_get_cache_alignment implementations"). Should we revert to the old implementation?
 
Huacai
 
------------------ Original ------------------
From:  "Robin Murphy"<robin.murphy@arm.com>;
Date:  Thu, Sep 21, 2017 06:47 PM
To:  "Huacai Chen"<chenhc@lemote.com>; "Christoph Hellwig"<hch@lst.de>; 
Cc:  "Marek Szyprowski"<m.szyprowski@samsung.com>; "Andrew Morton"<akpm@linux-foundation.org>; "Fuxin Zhang"<zhangfx@lemote.com>; "linux-kernel"<linux-kernel@vger.kernel.org>; "James E . J . Bottomley"<jejb@linux.vnet.ibm.com>; "Martin K . Petersen"<martin.petersen@oracle.com>; "linux-scsi"<linux-scsi@vger.kernel.org>; "stable"<stable@vger.kernel.org>; 
Subject:  Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as ahelper

 
On 19/09/17 09:52, Huacai Chen wrote:
> We will use device_is_coherent() as a helper function, which will be
> used in the next patch.
> 
> There is a MIPS-specific plat_device_is_coherent(), but we need a more
> generic solution, so add and use a new function pointer in dma_map_ops.

I think we're heading in the right direction with the series, but I
still don't like this patch. I can pretty much guarantee that driver
authors *will* abuse a generic device_is_coherent() API to mean "I can
skip other DMA API calls and just use virt_to_phys()".

I think it would be far better to allow architectures to provide their
own override of dma_get_cache_alignment(), and let the coherency detail
remain internal to the relevant arch implementations.

[...]
> @@ -697,6 +698,15 @@ static inline void *dma_zalloc_coherent(struct device *dev, size_t size,
>  }
>  
>  #ifdef CONFIG_HAS_DMA
> +static inline int device_is_coherent(struct device *dev)
> +{
> +	const struct dma_map_ops *ops = get_dma_ops(dev);
> +	if (ops && ops->device_is_coherent)
> +		return ops->device_is_coherent(dev);
> +	else
> +		return 1;    /* compatible behavior */

That is also quite scary - if someone now adds a new
dma_get_cache_alignemnt() call and dutifully passes a non-NULL device,
they will now get back an alignment of 1 on all non-coherent platforms
except MIPS: hello data corruption.

Robin.

> +}
> +
>  static inline int dma_get_cache_alignment(void)
>  {
>  #ifdef ARCH_DMA_MINALIGN
>

[toc] | [prev] | [next] | [standalone]

#1737512 — Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as ahelper

On 22/09/17 03:13, 陈华才 wrote:
> Hi, Robin,
> 
> Before 2.6.36 dma_get_cache_alignment is arch-dependent, and it is unified in commit 4565f0170dfc849b3629c27d7 ("dma-mapping: unify dma_get_cache_alignment implementations"). Should we revert to the old implementation?

Not quite - I mean instead of adding an ops->device_is_coherent callback
(which cannot really have a safe fallback value either way) and trying
to enforce that dma_get_cache_alignment() should be the only valid
caller, just add an ops->get_cache_alignment callback directly.

Robin.

>  
> Huacai
>  
> ------------------ Original ------------------
> From:  "Robin Murphy"<robin.murphy@arm.com>;
> Date:  Thu, Sep 21, 2017 06:47 PM
> To:  "Huacai Chen"<chenhc@lemote.com>; "Christoph Hellwig"<hch@lst.de>; 
> Cc:  "Marek Szyprowski"<m.szyprowski@samsung.com>; "Andrew Morton"<akpm@linux-foundation.org>; "Fuxin Zhang"<zhangfx@lemote.com>; "linux-kernel"<linux-kernel@vger.kernel.org>; "James E . J . Bottomley"<jejb@linux.vnet.ibm.com>; "Martin K . Petersen"<martin.petersen@oracle.com>; "linux-scsi"<linux-scsi@vger.kernel.org>; "stable"<stable@vger.kernel.org>; 
> Subject:  Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as ahelper
> 
>  
> On 19/09/17 09:52, Huacai Chen wrote:
>> We will use device_is_coherent() as a helper function, which will be
>> used in the next patch.
>>
>> There is a MIPS-specific plat_device_is_coherent(), but we need a more
>> generic solution, so add and use a new function pointer in dma_map_ops.
> 
> I think we're heading in the right direction with the series, but I
> still don't like this patch. I can pretty much guarantee that driver
> authors *will* abuse a generic device_is_coherent() API to mean "I can
> skip other DMA API calls and just use virt_to_phys()".
> 
> I think it would be far better to allow architectures to provide their
> own override of dma_get_cache_alignment(), and let the coherency detail
> remain internal to the relevant arch implementations.
> 
> [...]
>> @@ -697,6 +698,15 @@ static inline void *dma_zalloc_coherent(struct device *dev, size_t size,
>>  }
>>  
>>  #ifdef CONFIG_HAS_DMA
>> +static inline int device_is_coherent(struct device *dev)
>> +{
>> +	const struct dma_map_ops *ops = get_dma_ops(dev);
>> +	if (ops && ops->device_is_coherent)
>> +		return ops->device_is_coherent(dev);
>> +	else
>> +		return 1;    /* compatible behavior */
> 
> That is also quite scary - if someone now adds a new
> dma_get_cache_alignemnt() call and dutifully passes a non-NULL device,
> they will now get back an alignment of 1 on all non-coherent platforms
> except MIPS: hello data corruption.
> 
> Robin.
> 
>> +}
>> +
>>  static inline int dma_get_cache_alignment(void)
>>  {
>>  #ifdef ARCH_DMA_MINALIGN
>>
> 

[toc] | [prev] | [next] | [standalone]

#1737513 — Re: [PATCH V6 1/3] dma-mapping: Introduce device_is_coherent() as ahelper

On Fri, Sep 22, 2017 at 02:44:51PM +0100, Robin Murphy wrote:
> Not quite - I mean instead of adding an ops->device_is_coherent callback
> (which cannot really have a safe fallback value either way) and trying
> to enforce that dma_get_cache_alignment() should be the only valid
> caller, just add an ops->get_cache_alignment callback directly.

Exactly - and then fall back to ARCH_DMA_MINALIGN/1 if the ops vector
is not provided, to keep the existing behavior.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web