| Message ID | 20180410081910.858-1-vigneshr@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 04/10/2018 10:19 AM, Vignesh R wrote: > Add support to use DMA over memory mapped reads in direct mode. This > helps in reducing CPU usage from ~100% to ~10% when reading data from > flash. For non-DMA'able/vmalloc'd buffers, driver just falls back to CPU > based memcpy. > > Signed-off-by: Vignesh R <vigneshr@ti.com> Reviewed-by: Marek Vasut <marek.vasut@gmail.com>
On Tue, 10 Apr 2018 13:49:10 +0530 Vignesh R <vigneshr@ti.com> wrote: > Add support to use DMA over memory mapped reads in direct mode. This > helps in reducing CPU usage from ~100% to ~10% when reading data from > flash. For non-DMA'able/vmalloc'd buffers, driver just falls back to CPU > based memcpy. > > Signed-off-by: Vignesh R <vigneshr@ti.com> Applied to spi-nor/next. Thanks, Boris > --- > drivers/mtd/spi-nor/cadence-quadspi.c | 96 ++++++++++++++++++++++++++- > 1 file changed, 94 insertions(+), 2 deletions(-) > > diff --git a/drivers/mtd/spi-nor/cadence-quadspi.c b/drivers/mtd/spi-nor/cadence-quadspi.c > index 4b8e9183489a..2f3a4d4232b3 100644 > --- a/drivers/mtd/spi-nor/cadence-quadspi.c > +++ b/drivers/mtd/spi-nor/cadence-quadspi.c > @@ -18,6 +18,8 @@ > #include <linux/clk.h> > #include <linux/completion.h> > #include <linux/delay.h> > +#include <linux/dma-mapping.h> > +#include <linux/dmaengine.h> > #include <linux/err.h> > #include <linux/errno.h> > #include <linux/interrupt.h> > @@ -73,6 +75,10 @@ struct cqspi_st { > struct completion transfer_complete; > struct mutex bus_mutex; > > + struct dma_chan *rx_chan; > + struct completion rx_dma_complete; > + dma_addr_t mmap_phys_base; > + > int current_cs; > int current_page_size; > int current_erase_size; > @@ -915,11 +921,75 @@ static ssize_t cqspi_write(struct spi_nor *nor, loff_t to, > return len; > } > > +static void cqspi_rx_dma_callback(void *param) > +{ > + struct cqspi_st *cqspi = param; > + > + complete(&cqspi->rx_dma_complete); > +} > + > +static int cqspi_direct_read_execute(struct spi_nor *nor, u_char *buf, > + loff_t from, size_t len) > +{ > + struct cqspi_flash_pdata *f_pdata = nor->priv; > + struct cqspi_st *cqspi = f_pdata->cqspi; > + enum dma_ctrl_flags flags = DMA_CTRL_ACK | DMA_PREP_INTERRUPT; > + dma_addr_t dma_src = (dma_addr_t)cqspi->mmap_phys_base + from; > + int ret = 0; > + struct dma_async_tx_descriptor *tx; > + dma_cookie_t cookie; > + dma_addr_t dma_dst; > + > + if (!cqspi->rx_chan || !virt_addr_valid(buf)) { > + memcpy_fromio(buf, cqspi->ahb_base + from, len); > + return 0; > + } > + > + dma_dst = dma_map_single(nor->dev, buf, len, DMA_DEV_TO_MEM); > + if (dma_mapping_error(nor->dev, dma_dst)) { > + dev_err(nor->dev, "dma mapping failed\n"); > + return -ENOMEM; > + } > + tx = dmaengine_prep_dma_memcpy(cqspi->rx_chan, dma_dst, dma_src, > + len, flags); > + if (!tx) { > + dev_err(nor->dev, "device_prep_dma_memcpy error\n"); > + ret = -EIO; > + goto err_unmap; > + } > + > + tx->callback = cqspi_rx_dma_callback; > + tx->callback_param = cqspi; > + cookie = tx->tx_submit(tx); > + reinit_completion(&cqspi->rx_dma_complete); > + > + ret = dma_submit_error(cookie); > + if (ret) { > + dev_err(nor->dev, "dma_submit_error %d\n", cookie); > + ret = -EIO; > + goto err_unmap; > + } > + > + dma_async_issue_pending(cqspi->rx_chan); > + ret = wait_for_completion_timeout(&cqspi->rx_dma_complete, > + msecs_to_jiffies(len)); > + if (ret <= 0) { > + dmaengine_terminate_sync(cqspi->rx_chan); > + dev_err(nor->dev, "DMA wait_for_completion_timeout\n"); > + ret = -ETIMEDOUT; > + goto err_unmap; > + } > + > +err_unmap: > + dma_unmap_single(nor->dev, dma_dst, len, DMA_DEV_TO_MEM); > + > + return 0; > +} > + > static ssize_t cqspi_read(struct spi_nor *nor, loff_t from, > size_t len, u_char *buf) > { > struct cqspi_flash_pdata *f_pdata = nor->priv; > - struct cqspi_st *cqspi = f_pdata->cqspi; > int ret; > > ret = cqspi_set_protocol(nor, 1); > @@ -931,7 +1001,7 @@ static ssize_t cqspi_read(struct spi_nor *nor, loff_t from, > return ret; > > if (f_pdata->use_direct_mode) > - memcpy_fromio(buf, cqspi->ahb_base + from, len); > + ret = cqspi_direct_read_execute(nor, buf, from, len); > else > ret = cqspi_indirect_read_execute(nor, buf, from, len); > if (ret) > @@ -1100,6 +1170,21 @@ static void cqspi_controller_init(struct cqspi_st *cqspi) > cqspi_controller_enable(cqspi, 1); > } > > +static void cqspi_request_mmap_dma(struct cqspi_st *cqspi) > +{ > + dma_cap_mask_t mask; > + > + dma_cap_zero(mask); > + dma_cap_set(DMA_MEMCPY, mask); > + > + cqspi->rx_chan = dma_request_chan_by_mask(&mask); > + if (IS_ERR(cqspi->rx_chan)) { > + dev_err(&cqspi->pdev->dev, "No Rx DMA available\n"); > + cqspi->rx_chan = NULL; > + } > + init_completion(&cqspi->rx_dma_complete); > +} > + > static int cqspi_setup_flash(struct cqspi_st *cqspi, struct device_node *np) > { > const struct spi_nor_hwcaps hwcaps = { > @@ -1177,6 +1262,9 @@ static int cqspi_setup_flash(struct cqspi_st *cqspi, struct device_node *np) > f_pdata->use_direct_mode = true; > dev_dbg(nor->dev, "using direct mode for %s\n", > mtd->name); > + > + if (!cqspi->rx_chan) > + cqspi_request_mmap_dma(cqspi); > } > } > > @@ -1237,6 +1325,7 @@ static int cqspi_probe(struct platform_device *pdev) > dev_err(dev, "Cannot remap AHB address.\n"); > return PTR_ERR(cqspi->ahb_base); > } > + cqspi->mmap_phys_base = (dma_addr_t)res_ahb->start; > cqspi->ahb_size = resource_size(res_ahb); > > init_completion(&cqspi->transfer_complete); > @@ -1307,6 +1396,9 @@ static int cqspi_remove(struct platform_device *pdev) > > cqspi_controller_enable(cqspi, 0); > > + if (cqspi->rx_chan) > + dma_release_channel(cqspi->rx_chan); > + > clk_disable_unprepare(cqspi->clk); > > pm_runtime_put_sync(&pdev->dev);
diff --git a/drivers/mtd/spi-nor/cadence-quadspi.c b/drivers/mtd/spi-nor/cadence-quadspi.c index 4b8e9183489a..2f3a4d4232b3 100644 --- a/drivers/mtd/spi-nor/cadence-quadspi.c +++ b/drivers/mtd/spi-nor/cadence-quadspi.c @@ -18,6 +18,8 @@ #include <linux/clk.h> #include <linux/completion.h> #include <linux/delay.h> +#include <linux/dma-mapping.h> +#include <linux/dmaengine.h> #include <linux/err.h> #include <linux/errno.h> #include <linux/interrupt.h> @@ -73,6 +75,10 @@ struct cqspi_st { struct completion transfer_complete; struct mutex bus_mutex; + struct dma_chan *rx_chan; + struct completion rx_dma_complete; + dma_addr_t mmap_phys_base; + int current_cs; int current_page_size; int current_erase_size; @@ -915,11 +921,75 @@ static ssize_t cqspi_write(struct spi_nor *nor, loff_t to, return len; } +static void cqspi_rx_dma_callback(void *param) +{ + struct cqspi_st *cqspi = param; + + complete(&cqspi->rx_dma_complete); +} + +static int cqspi_direct_read_execute(struct spi_nor *nor, u_char *buf, + loff_t from, size_t len) +{ + struct cqspi_flash_pdata *f_pdata = nor->priv; + struct cqspi_st *cqspi = f_pdata->cqspi; + enum dma_ctrl_flags flags = DMA_CTRL_ACK | DMA_PREP_INTERRUPT; + dma_addr_t dma_src = (dma_addr_t)cqspi->mmap_phys_base + from; + int ret = 0; + struct dma_async_tx_descriptor *tx; + dma_cookie_t cookie; + dma_addr_t dma_dst; + + if (!cqspi->rx_chan || !virt_addr_valid(buf)) { + memcpy_fromio(buf, cqspi->ahb_base + from, len); + return 0; + } + + dma_dst = dma_map_single(nor->dev, buf, len, DMA_DEV_TO_MEM); + if (dma_mapping_error(nor->dev, dma_dst)) { + dev_err(nor->dev, "dma mapping failed\n"); + return -ENOMEM; + } + tx = dmaengine_prep_dma_memcpy(cqspi->rx_chan, dma_dst, dma_src, + len, flags); + if (!tx) { + dev_err(nor->dev, "device_prep_dma_memcpy error\n"); + ret = -EIO; + goto err_unmap; + } + + tx->callback = cqspi_rx_dma_callback; + tx->callback_param = cqspi; + cookie = tx->tx_submit(tx); + reinit_completion(&cqspi->rx_dma_complete); + + ret = dma_submit_error(cookie); + if (ret) { + dev_err(nor->dev, "dma_submit_error %d\n", cookie); + ret = -EIO; + goto err_unmap; + } + + dma_async_issue_pending(cqspi->rx_chan); + ret = wait_for_completion_timeout(&cqspi->rx_dma_complete, + msecs_to_jiffies(len)); + if (ret <= 0) { + dmaengine_terminate_sync(cqspi->rx_chan); + dev_err(nor->dev, "DMA wait_for_completion_timeout\n"); + ret = -ETIMEDOUT; + goto err_unmap; + } + +err_unmap: + dma_unmap_single(nor->dev, dma_dst, len, DMA_DEV_TO_MEM); + + return 0; +} + static ssize_t cqspi_read(struct spi_nor *nor, loff_t from, size_t len, u_char *buf) { struct cqspi_flash_pdata *f_pdata = nor->priv; - struct cqspi_st *cqspi = f_pdata->cqspi; int ret; ret = cqspi_set_protocol(nor, 1); @@ -931,7 +1001,7 @@ static ssize_t cqspi_read(struct spi_nor *nor, loff_t from, return ret; if (f_pdata->use_direct_mode) - memcpy_fromio(buf, cqspi->ahb_base + from, len); + ret = cqspi_direct_read_execute(nor, buf, from, len); else ret = cqspi_indirect_read_execute(nor, buf, from, len); if (ret) @@ -1100,6 +1170,21 @@ static void cqspi_controller_init(struct cqspi_st *cqspi) cqspi_controller_enable(cqspi, 1); } +static void cqspi_request_mmap_dma(struct cqspi_st *cqspi) +{ + dma_cap_mask_t mask; + + dma_cap_zero(mask); + dma_cap_set(DMA_MEMCPY, mask); + + cqspi->rx_chan = dma_request_chan_by_mask(&mask); + if (IS_ERR(cqspi->rx_chan)) { + dev_err(&cqspi->pdev->dev, "No Rx DMA available\n"); + cqspi->rx_chan = NULL; + } + init_completion(&cqspi->rx_dma_complete); +} + static int cqspi_setup_flash(struct cqspi_st *cqspi, struct device_node *np) { const struct spi_nor_hwcaps hwcaps = { @@ -1177,6 +1262,9 @@ static int cqspi_setup_flash(struct cqspi_st *cqspi, struct device_node *np) f_pdata->use_direct_mode = true; dev_dbg(nor->dev, "using direct mode for %s\n", mtd->name); + + if (!cqspi->rx_chan) + cqspi_request_mmap_dma(cqspi); } } @@ -1237,6 +1325,7 @@ static int cqspi_probe(struct platform_device *pdev) dev_err(dev, "Cannot remap AHB address.\n"); return PTR_ERR(cqspi->ahb_base); } + cqspi->mmap_phys_base = (dma_addr_t)res_ahb->start; cqspi->ahb_size = resource_size(res_ahb); init_completion(&cqspi->transfer_complete); @@ -1307,6 +1396,9 @@ static int cqspi_remove(struct platform_device *pdev) cqspi_controller_enable(cqspi, 0); + if (cqspi->rx_chan) + dma_release_channel(cqspi->rx_chan); + clk_disable_unprepare(cqspi->clk); pm_runtime_put_sync(&pdev->dev);
Add support to use DMA over memory mapped reads in direct mode. This helps in reducing CPU usage from ~100% to ~10% when reading data from flash. For non-DMA'able/vmalloc'd buffers, driver just falls back to CPU based memcpy. Signed-off-by: Vignesh R <vigneshr@ti.com> --- drivers/mtd/spi-nor/cadence-quadspi.c | 96 ++++++++++++++++++++++++++- 1 file changed, 94 insertions(+), 2 deletions(-)