]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
atmel-mci: Add experimental DMA support
authorHaavard Skinnemoen <haavard.skinnemoen@atmel.com>
Wed, 30 Jul 2008 18:29:03 +0000 (20:29 +0200)
committerHaavard Skinnemoen <haavard.skinnemoen@atmel.com>
Sun, 5 Oct 2008 18:39:21 +0000 (20:39 +0200)
This adds support for DMA transfers through the generic DMA engine
framework with the DMA slave extensions.

The driver has been tested using mmc-block and ext3fs on several SD,
SDHC and MMC+ cards. Reads and writes work fine, with read transfer
rates up to 7.5 MiB/s on fast cards with debugging disabled.

Unfortunately, the driver has been known to lock up from time to time
with DMA enabled, so DMA support is currently optional and marked
EXPERIMENTAL. However, I didn't see any problems while testing 13
different cards (MMC, SD and SDHC of different brands and sizes), so I
suspect the "Initialize BLKR before sending data transfer command" fix
that was posted earlier fixed this as well.

Signed-off-by: Haavard Skinnemoen <haavard.skinnemoen@atmel.com>
arch/avr32/include/asm/atmel-mci.h
arch/avr32/mach-at32ap/at32ap700x.c
drivers/mmc/host/Kconfig
drivers/mmc/host/atmel-mci.c

index 5d5ae1295cfd239e233e969eb6be67e9597d6413..59f3fadd0b68a45e07353fb8d6e3d97d5e7aba90 100644 (file)
@@ -3,6 +3,8 @@
 
 #define ATMEL_MCI_MAX_NR_SLOTS 2
 
+struct dma_slave;
+
 /**
  * struct mci_slot_pdata - board-specific per-slot configuration
  * @bus_width: Number of data lines wired up the slot
@@ -26,9 +28,11 @@ struct mci_slot_pdata {
 
 /**
  * struct mci_platform_data - board-specific MMC/SDcard configuration
+ * @dma_slave: DMA slave interface to use in data transfers, or NULL.
  * @slot: Per-slot configuration data.
  */
 struct mci_platform_data {
+       struct dma_slave        *dma_slave;
        struct mci_slot_pdata   slot[ATMEL_MCI_MAX_NR_SLOTS];
 };
 
index 9967d5a3b6eb709d0bea4837221e2fb7f52f071a..f1b9a3ac27336fe92c9c15cbe114212b5f98d4f5 100644 (file)
@@ -1273,6 +1273,7 @@ struct platform_device *__init
 at32_add_device_mci(unsigned int id, struct mci_platform_data *data)
 {
        struct platform_device          *pdev;
+       struct dw_dma_slave             *dws;
 
        if (id != 0 || !data)
                return NULL;
@@ -1289,6 +1290,21 @@ at32_add_device_mci(unsigned int id, struct mci_platform_data *data)
                                ARRAY_SIZE(atmel_mci0_resource)))
                goto fail;
 
+       if (data->dma_slave)
+               dws = kmemdup(to_dw_dma_slave(data->dma_slave),
+                               sizeof(struct dw_dma_slave), GFP_KERNEL);
+       else
+               dws = kzalloc(sizeof(struct dw_dma_slave), GFP_KERNEL);
+
+       dws->slave.dev = &pdev->dev;
+       dws->slave.dma_dev = &dw_dmac0_device.dev;
+       dws->slave.reg_width = DMA_SLAVE_WIDTH_32BIT;
+       dws->cfg_hi = (DWC_CFGH_SRC_PER(0)
+                               | DWC_CFGH_DST_PER(1));
+       dws->cfg_lo &= ~(DWC_CFGL_HS_DST_POL
+                               | DWC_CFGL_HS_SRC_POL);
+
+       data->dma_slave = &dws->slave;
 
        if (platform_device_add_data(pdev, data,
                                sizeof(struct mci_platform_data)))
index ea8d7a3490d9b7ee575f23587721337ac8cbbc4a..1ce21d4c86086ef7427e5f83c10c2cad7a20f2b8 100644 (file)
@@ -114,6 +114,17 @@ config MMC_ATMELMCI
 
          If unsure, say N.
 
+config MMC_ATMELMCI_DMA
+       bool "Atmel MCI DMA support (EXPERIMENTAL)"
+       depends on MMC_ATMELMCI && DMA_ENGINE && EXPERIMENTAL
+       help
+         Say Y here to have the Atmel MCI driver use a DMA engine to
+         do data transfers and thus increase the throughput and
+         reduce the CPU utilization. Note that this is highly
+         experimental and may cause the driver to lock up.
+
+         If unsure, say N.
+
 config MMC_IMX
        tristate "Motorola i.MX Multimedia Card Interface support"
        depends on ARCH_IMX
index d8ab35175a538eaef70500775720d691e8e25c03..d45dfa259386fc934d1bdcc5294a875c69f76465 100644 (file)
@@ -11,6 +11,8 @@
 #include <linux/clk.h>
 #include <linux/debugfs.h>
 #include <linux/device.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/gpio.h>
 #include <linux/init.h>
@@ -33,6 +35,7 @@
 #include "atmel-mci-regs.h"
 
 #define ATMCI_DATA_ERROR_FLAGS (MCI_DCRCE | MCI_DTOE | MCI_OVRE | MCI_UNRE)
+#define ATMCI_DMA_THRESHOLD    16
 
 enum {
        EVENT_CMD_COMPLETE = 0,
@@ -50,6 +53,14 @@ enum atmel_mci_state {
        STATE_DATA_ERROR,
 };
 
+struct atmel_mci_dma {
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+       struct dma_client               client;
+       struct dma_chan                 *chan;
+       struct dma_async_tx_descriptor  *data_desc;
+#endif
+};
+
 /**
  * struct atmel_mci - MMC controller state shared between all slots
  * @lock: Spinlock protecting the queue and associated data.
@@ -62,6 +73,8 @@ enum atmel_mci_state {
  * @cmd: The command currently being sent to the card, or NULL.
  * @data: The data currently being transferred, or NULL if no data
  *     transfer is in progress.
+ * @dma: DMA client state.
+ * @data_chan: DMA channel being used for the current data transfer.
  * @cmd_status: Snapshot of SR taken upon completion of the current
  *     command. Only valid when EVENT_CMD_COMPLETE is pending.
  * @data_status: Snapshot of SR taken upon completion of the current
@@ -126,6 +139,9 @@ struct atmel_mci {
        struct mmc_command      *cmd;
        struct mmc_data         *data;
 
+       struct atmel_mci_dma    dma;
+       struct dma_chan         *data_chan;
+
        u32                     cmd_status;
        u32                     data_status;
        u32                     stop_cmdr;
@@ -485,6 +501,144 @@ static void send_stop_cmd(struct atmel_mci *host, struct mmc_data *data)
        mci_writel(host, IER, MCI_CMDRDY);
 }
 
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+static void atmci_dma_cleanup(struct atmel_mci *host)
+{
+       struct mmc_data                 *data = host->data;
+
+       dma_unmap_sg(&host->pdev->dev, data->sg, data->sg_len,
+                    ((data->flags & MMC_DATA_WRITE)
+                     ? DMA_TO_DEVICE : DMA_FROM_DEVICE));
+}
+
+static void atmci_stop_dma(struct atmel_mci *host)
+{
+       struct dma_chan *chan = host->data_chan;
+
+       if (chan) {
+               chan->device->device_terminate_all(chan);
+               atmci_dma_cleanup(host);
+       } else {
+               /* Data transfer was stopped by the interrupt handler */
+               atmci_set_pending(host, EVENT_XFER_COMPLETE);
+               mci_writel(host, IER, MCI_NOTBUSY);
+       }
+}
+
+/* This function is called by the DMA driver from tasklet context. */
+static void atmci_dma_complete(void *arg)
+{
+       struct atmel_mci        *host = arg;
+       struct mmc_data         *data = host->data;
+
+       dev_vdbg(&host->pdev->dev, "DMA complete\n");
+
+       atmci_dma_cleanup(host);
+
+       /*
+        * If the card was removed, data will be NULL. No point trying
+        * to send the stop command or waiting for NBUSY in this case.
+        */
+       if (data) {
+               atmci_set_pending(host, EVENT_XFER_COMPLETE);
+               tasklet_schedule(&host->tasklet);
+
+               /*
+                * Regardless of what the documentation says, we have
+                * to wait for NOTBUSY even after block read
+                * operations.
+                *
+                * When the DMA transfer is complete, the controller
+                * may still be reading the CRC from the card, i.e.
+                * the data transfer is still in progress and we
+                * haven't seen all the potential error bits yet.
+                *
+                * The interrupt handler will schedule a different
+                * tasklet to finish things up when the data transfer
+                * is completely done.
+                *
+                * We may not complete the mmc request here anyway
+                * because the mmc layer may call back and cause us to
+                * violate the "don't submit new operations from the
+                * completion callback" rule of the dma engine
+                * framework.
+                */
+               mci_writel(host, IER, MCI_NOTBUSY);
+       }
+}
+
+static int
+atmci_submit_data_dma(struct atmel_mci *host, struct mmc_data *data)
+{
+       struct dma_chan                 *chan;
+       struct dma_async_tx_descriptor  *desc;
+       struct scatterlist              *sg;
+       unsigned int                    i;
+       enum dma_data_direction         direction;
+
+       /*
+        * We don't do DMA on "complex" transfers, i.e. with
+        * non-word-aligned buffers or lengths. Also, we don't bother
+        * with all the DMA setup overhead for short transfers.
+        */
+       if (data->blocks * data->blksz < ATMCI_DMA_THRESHOLD)
+               return -EINVAL;
+       if (data->blksz & 3)
+               return -EINVAL;
+
+       for_each_sg(data->sg, sg, data->sg_len, i) {
+               if (sg->offset & 3 || sg->length & 3)
+                       return -EINVAL;
+       }
+
+       /* If we don't have a channel, we can't do DMA */
+       chan = host->dma.chan;
+       if (chan) {
+               dma_chan_get(chan);
+               host->data_chan = chan;
+       }
+
+       if (!chan)
+               return -ENODEV;
+
+       if (data->flags & MMC_DATA_READ)
+               direction = DMA_FROM_DEVICE;
+       else
+               direction = DMA_TO_DEVICE;
+
+       desc = chan->device->device_prep_slave_sg(chan,
+                       data->sg, data->sg_len, direction,
+                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+       if (!desc)
+               return -ENOMEM;
+
+       host->dma.data_desc = desc;
+       desc->callback = atmci_dma_complete;
+       desc->callback_param = host;
+       desc->tx_submit(desc);
+
+       /* Go! */
+       chan->device->device_issue_pending(chan);
+
+       return 0;
+}
+
+#else /* CONFIG_MMC_ATMELMCI_DMA */
+
+static int atmci_submit_data_dma(struct atmel_mci *host, struct mmc_data *data)
+{
+       return -ENOSYS;
+}
+
+static void atmci_stop_dma(struct atmel_mci *host)
+{
+       /* Data transfer was stopped by the interrupt handler */
+       atmci_set_pending(host, EVENT_XFER_COMPLETE);
+       mci_writel(host, IER, MCI_NOTBUSY);
+}
+
+#endif /* CONFIG_MMC_ATMELMCI_DMA */
+
 /*
  * Returns a mask of interrupt flags to be enabled after the whole
  * request has been prepared.
@@ -500,24 +654,27 @@ static u32 atmci_submit_data(struct atmel_mci *host, struct mmc_data *data)
        host->data = data;
 
        iflags = ATMCI_DATA_ERROR_FLAGS;
+       if (atmci_submit_data_dma(host, data)) {
+               host->data_chan = NULL;
 
-       /*
-        * Errata: MMC data write operation with less than 12
-        * bytes is impossible.
-        *
-        * Errata: MCI Transmit Data Register (TDR) FIFO
-        * corruption when length is not multiple of 4.
-        */
-       if (data->blocks * data->blksz < 12
-                       || (data->blocks * data->blksz) & 3)
-               host->need_reset = true;
+               /*
+                * Errata: MMC data write operation with less than 12
+                * bytes is impossible.
+                *
+                * Errata: MCI Transmit Data Register (TDR) FIFO
+                * corruption when length is not multiple of 4.
+                */
+               if (data->blocks * data->blksz < 12
+                               || (data->blocks * data->blksz) & 3)
+                       host->need_reset = true;
 
-       host->sg = data->sg;
-       host->pio_offset = 0;
-       if (data->flags & MMC_DATA_READ)
-               iflags |= MCI_RXRDY;
-       else
-               iflags |= MCI_TXRDY;
+               host->sg = data->sg;
+               host->pio_offset = 0;
+               if (data->flags & MMC_DATA_READ)
+                       iflags |= MCI_RXRDY;
+               else
+                       iflags |= MCI_TXRDY;
+       }
 
        return iflags;
 }
@@ -848,6 +1005,7 @@ static void atmci_command_complete(struct atmel_mci *host,
 
                if (cmd->data) {
                        host->data = NULL;
+                       atmci_stop_dma(host);
                        mci_writel(host, IDR, MCI_NOTBUSY
                                        | MCI_TXRDY | MCI_RXRDY
                                        | ATMCI_DATA_ERROR_FLAGS);
@@ -917,6 +1075,7 @@ static void atmci_detect_change(unsigned long data)
                                        /* fall through */
                                case STATE_SENDING_DATA:
                                        mrq->data->error = -ENOMEDIUM;
+                                       atmci_stop_dma(host);
                                        break;
                                case STATE_DATA_BUSY:
                                case STATE_DATA_ERROR:
@@ -995,6 +1154,7 @@ static void atmci_tasklet_func(unsigned long priv)
                case STATE_SENDING_DATA:
                        if (atmci_test_and_clear_pending(host,
                                                EVENT_DATA_ERROR)) {
+                               atmci_stop_dma(host);
                                if (data->stop)
                                        send_stop_cmd(host, data);
                                state = STATE_DATA_ERROR;
@@ -1280,6 +1440,60 @@ static irqreturn_t atmci_detect_interrupt(int irq, void *dev_id)
        return IRQ_HANDLED;
 }
 
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+
+static inline struct atmel_mci *
+dma_client_to_atmel_mci(struct dma_client *client)
+{
+       return container_of(client, struct atmel_mci, dma.client);
+}
+
+static enum dma_state_client atmci_dma_event(struct dma_client *client,
+               struct dma_chan *chan, enum dma_state state)
+{
+       struct atmel_mci        *host;
+       enum dma_state_client   ret = DMA_NAK;
+
+       host = dma_client_to_atmel_mci(client);
+
+       switch (state) {
+       case DMA_RESOURCE_AVAILABLE:
+               spin_lock_bh(&host->lock);
+               if (!host->dma.chan) {
+                       host->dma.chan = chan;
+                       ret = DMA_ACK;
+               }
+               spin_unlock_bh(&host->lock);
+
+               if (ret == DMA_ACK)
+                       dev_info(&host->pdev->dev,
+                                       "Using %s for DMA transfers\n",
+                                       chan->dev.bus_id);
+               break;
+
+       case DMA_RESOURCE_REMOVED:
+               spin_lock_bh(&host->lock);
+               if (host->dma.chan == chan) {
+                       host->dma.chan = NULL;
+                       ret = DMA_ACK;
+               }
+               spin_unlock_bh(&host->lock);
+
+               if (ret == DMA_ACK)
+                       dev_info(&host->pdev->dev,
+                                       "Lost %s, falling back to PIO\n",
+                                       chan->dev.bus_id);
+               break;
+
+       default:
+               break;
+       }
+
+
+       return ret;
+}
+#endif /* CONFIG_MMC_ATMELMCI_DMA */
+
 static int __init atmci_init_slot(struct atmel_mci *host,
                struct mci_slot_pdata *slot_data, unsigned int id,
                u32 sdc_reg)
@@ -1434,6 +1648,25 @@ static int __init atmci_probe(struct platform_device *pdev)
        if (ret)
                goto err_request_irq;
 
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+       if (pdata->dma_slave) {
+               struct dma_slave *slave = pdata->dma_slave;
+
+               slave->tx_reg = regs->start + MCI_TDR;
+               slave->rx_reg = regs->start + MCI_RDR;
+
+               /* Try to grab a DMA channel */
+               host->dma.client.event_callback = atmci_dma_event;
+               dma_cap_set(DMA_SLAVE, host->dma.client.cap_mask);
+               host->dma.client.slave = slave;
+
+               dma_async_client_register(&host->dma.client);
+               dma_async_client_chan_request(&host->dma.client);
+       } else {
+               dev_notice(&pdev->dev, "DMA not available, using PIO\n");
+       }
+#endif /* CONFIG_MMC_ATMELMCI_DMA */
+
        platform_set_drvdata(pdev, host);
 
        /* We need at least one slot to succeed */
@@ -1462,6 +1695,10 @@ static int __init atmci_probe(struct platform_device *pdev)
        return 0;
 
 err_init_slot:
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+       if (pdata->dma_slave)
+               dma_async_client_unregister(&host->dma.client);
+#endif
        free_irq(irq, host);
 err_request_irq:
        iounmap(host->regs);
@@ -1490,6 +1727,11 @@ static int __exit atmci_remove(struct platform_device *pdev)
        mci_readl(host, SR);
        clk_disable(host->mck);
 
+#ifdef CONFIG_MMC_ATMELMCI_DMA
+       if (host->dma.client.slave)
+               dma_async_client_unregister(&host->dma.client);
+#endif
+
        free_irq(platform_get_irq(pdev, 0), host);
        iounmap(host->regs);