]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - drivers/spi/omap2_mcspi.c
SPI: omap2_mcspi driver
[linux-2.6-omap-h63xx.git] / drivers / spi / omap2_mcspi.c
diff --git a/drivers/spi/omap2_mcspi.c b/drivers/spi/omap2_mcspi.c
new file mode 100644 (file)
index 0000000..6b357cd
--- /dev/null
@@ -0,0 +1,1081 @@
+/*
+ * OMAP2 McSPI controller driver
+ *
+ * Copyright (C) 2005, 2006 Nokia Corporation
+ * Author:     Samuel Ortiz <samuel.ortiz@nokia.com> and
+ *             Juha Yrjölä <juha.yrjola@nokia.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/delay.h>
+#include <linux/dma-mapping.h>
+#include <linux/platform_device.h>
+#include <linux/err.h>
+#include <linux/clk.h>
+#include <linux/io.h>
+
+#include <linux/spi/spi.h>
+
+#include <asm/arch/dma.h>
+#include <asm/arch/clock.h>
+
+
+#define OMAP2_MCSPI_MAX_FREQ           48000000
+
+#define OMAP2_MCSPI_REVISION           0x00
+#define OMAP2_MCSPI_SYSCONFIG          0x10
+#define OMAP2_MCSPI_SYSSTATUS          0x14
+#define OMAP2_MCSPI_IRQSTATUS          0x18
+#define OMAP2_MCSPI_IRQENABLE          0x1c
+#define OMAP2_MCSPI_WAKEUPENABLE       0x20
+#define OMAP2_MCSPI_SYST               0x24
+#define OMAP2_MCSPI_MODULCTRL          0x28
+
+/* per-channel banks, 0x14 bytes each, first is: */
+#define OMAP2_MCSPI_CHCONF0            0x2c
+#define OMAP2_MCSPI_CHSTAT0            0x30
+#define OMAP2_MCSPI_CHCTRL0            0x34
+#define OMAP2_MCSPI_TX0                        0x38
+#define OMAP2_MCSPI_RX0                        0x3c
+
+/* per-register bitmasks: */
+
+#define OMAP2_MCSPI_SYSCONFIG_AUTOIDLE (1 << 0)
+#define OMAP2_MCSPI_SYSCONFIG_SOFTRESET        (1 << 1)
+
+#define OMAP2_MCSPI_SYSSTATUS_RESETDONE        (1 << 0)
+
+#define OMAP2_MCSPI_MODULCTRL_SINGLE   (1 << 0)
+#define OMAP2_MCSPI_MODULCTRL_MS       (1 << 2)
+#define OMAP2_MCSPI_MODULCTRL_STEST    (1 << 3)
+
+#define OMAP2_MCSPI_CHCONF_PHA         (1 << 0)
+#define OMAP2_MCSPI_CHCONF_POL         (1 << 1)
+#define OMAP2_MCSPI_CHCONF_CLKD_MASK   (0x0f << 2)
+#define OMAP2_MCSPI_CHCONF_EPOL                (1 << 6)
+#define OMAP2_MCSPI_CHCONF_WL_MASK     (0x1f << 7)
+#define OMAP2_MCSPI_CHCONF_TRM_RX_ONLY (0x01 << 12)
+#define OMAP2_MCSPI_CHCONF_TRM_TX_ONLY (0x02 << 12)
+#define OMAP2_MCSPI_CHCONF_TRM_MASK    (0x03 << 12)
+#define OMAP2_MCSPI_CHCONF_DMAW                (1 << 14)
+#define OMAP2_MCSPI_CHCONF_DMAR                (1 << 15)
+#define OMAP2_MCSPI_CHCONF_DPE0                (1 << 16)
+#define OMAP2_MCSPI_CHCONF_DPE1                (1 << 17)
+#define OMAP2_MCSPI_CHCONF_IS          (1 << 18)
+#define OMAP2_MCSPI_CHCONF_TURBO       (1 << 19)
+#define OMAP2_MCSPI_CHCONF_FORCE       (1 << 20)
+
+#define OMAP2_MCSPI_CHSTAT_RXS         (1 << 0)
+#define OMAP2_MCSPI_CHSTAT_TXS         (1 << 1)
+#define OMAP2_MCSPI_CHSTAT_EOT         (1 << 2)
+
+#define OMAP2_MCSPI_CHCTRL_EN          (1 << 0)
+
+
+/* We have 2 DMA channels per CS, one for RX and one for TX */
+struct omap2_mcspi_dma {
+       int dma_tx_channel;
+       int dma_rx_channel;
+
+       int dma_tx_sync_dev;
+       int dma_rx_sync_dev;
+
+       struct completion dma_tx_completion;
+       struct completion dma_rx_completion;
+};
+
+/* use PIO for small transfers, avoiding DMA setup/teardown overhead and
+ * cache operations; better heuristics consider wordsize and bitrate.
+ */
+#define DMA_MIN_BYTES                  8
+
+
+struct omap2_mcspi {
+       struct work_struct      work;
+       /* lock protects queue and registers */
+       spinlock_t              lock;
+       struct list_head        msg_queue;
+       struct spi_master       *master;
+       struct clk              *ick;
+       struct clk              *fck;
+       /* Virtual base address of the controller */
+       void __iomem            *base;
+       /* SPI1 has 4 channels, while SPI2 has 2 */
+       struct omap2_mcspi_dma  *dma_channels;
+};
+
+struct omap2_mcspi_cs {
+       void __iomem            *base;
+       int                     word_len;
+};
+
+static struct workqueue_struct *omap2_mcspi_wq;
+
+#define MOD_REG_BIT(val, mask, set) do { \
+       if (set) \
+               val |= mask; \
+       else \
+               val &= ~mask; \
+} while (0)
+
+static inline void mcspi_write_reg(struct spi_master *master,
+               int idx, u32 val)
+{
+       struct omap2_mcspi *mcspi = spi_master_get_devdata(master);
+
+       __raw_writel(val, mcspi->base + idx);
+}
+
+static inline u32 mcspi_read_reg(struct spi_master *master, int idx)
+{
+       struct omap2_mcspi *mcspi = spi_master_get_devdata(master);
+
+       return __raw_readl(mcspi->base + idx);
+}
+
+static inline void mcspi_write_cs_reg(const struct spi_device *spi,
+               int idx, u32 val)
+{
+       struct omap2_mcspi_cs   *cs = spi->controller_state;
+
+       __raw_writel(val, cs->base +  idx);
+}
+
+static inline u32 mcspi_read_cs_reg(const struct spi_device *spi, int idx)
+{
+       struct omap2_mcspi_cs   *cs = spi->controller_state;
+
+       return __raw_readl(cs->base + idx);
+}
+
+static void omap2_mcspi_set_dma_req(const struct spi_device *spi,
+               int is_read, int enable)
+{
+       u32 l, rw;
+
+       l = mcspi_read_cs_reg(spi, OMAP2_MCSPI_CHCONF0);
+
+       if (is_read) /* 1 is read, 0 write */
+               rw = OMAP2_MCSPI_CHCONF_DMAR;
+       else
+               rw = OMAP2_MCSPI_CHCONF_DMAW;
+
+       MOD_REG_BIT(l, rw, enable);
+       mcspi_write_cs_reg(spi, OMAP2_MCSPI_CHCONF0, l);
+}
+
+static void omap2_mcspi_set_enable(const struct spi_device *spi, int enable)
+{
+       u32 l;
+
+       l = enable ? OMAP2_MCSPI_CHCTRL_EN : 0;
+       mcspi_write_cs_reg(spi, OMAP2_MCSPI_CHCTRL0, l);
+}
+
+static void omap2_mcspi_force_cs(struct spi_device *spi, int cs_active)
+{
+       u32 l;
+
+       l = mcspi_read_cs_reg(spi, OMAP2_MCSPI_CHCONF0);
+       MOD_REG_BIT(l, OMAP2_MCSPI_CHCONF_FORCE, cs_active);
+       mcspi_write_cs_reg(spi, OMAP2_MCSPI_CHCONF0, l);
+}
+
+static void omap2_mcspi_set_master_mode(struct spi_master *master)
+{
+       u32 l;
+
+       /* setup when switching from (reset default) slave mode
+        * to single-channel master mode
+        */
+       l = mcspi_read_reg(master, OMAP2_MCSPI_MODULCTRL);
+       MOD_REG_BIT(l, OMAP2_MCSPI_MODULCTRL_STEST, 0);
+       MOD_REG_BIT(l, OMAP2_MCSPI_MODULCTRL_MS, 0);
+       MOD_REG_BIT(l, OMAP2_MCSPI_MODULCTRL_SINGLE, 1);
+       mcspi_write_reg(master, OMAP2_MCSPI_MODULCTRL, l);
+}
+
+static unsigned
+omap2_mcspi_txrx_dma(struct spi_device *spi, struct spi_transfer *xfer)
+{
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_cs   *cs = spi->controller_state;
+       struct omap2_mcspi_dma  *mcspi_dma;
+       unsigned int            count, c;
+       unsigned long           base, tx_reg, rx_reg;
+       int                     word_len, data_type, element_count;
+       u8                      * rx;
+       const u8                * tx;
+
+       mcspi = spi_master_get_devdata(spi->master);
+       mcspi_dma = &mcspi->dma_channels[spi->chip_select];
+
+       count = xfer->len;
+       c = count;
+       word_len = cs->word_len;
+
+       base = (unsigned long) io_v2p(cs->base);
+       tx_reg = base + OMAP2_MCSPI_TX0;
+       rx_reg = base + OMAP2_MCSPI_RX0;
+       rx = xfer->rx_buf;
+       tx = xfer->tx_buf;
+
+       if (word_len <= 8) {
+               data_type = OMAP_DMA_DATA_TYPE_S8;
+               element_count = count;
+       } else if (word_len <= 16) {
+               data_type = OMAP_DMA_DATA_TYPE_S16;
+               element_count = count >> 1;
+       } else /* word_len <= 32 */ {
+               data_type = OMAP_DMA_DATA_TYPE_S32;
+               element_count = count >> 2;
+       }
+
+       if (tx != NULL) {
+               omap_set_dma_transfer_params(mcspi_dma->dma_tx_channel,
+                               data_type, element_count, 1,
+                               OMAP_DMA_SYNC_ELEMENT,
+                               mcspi_dma->dma_tx_sync_dev, 0);
+
+               omap_set_dma_dest_params(mcspi_dma->dma_tx_channel, 0,
+                               OMAP_DMA_AMODE_CONSTANT,
+                               tx_reg, 0, 0);
+
+               omap_set_dma_src_params(mcspi_dma->dma_tx_channel, 0,
+                               OMAP_DMA_AMODE_POST_INC,
+                               xfer->tx_dma, 0, 0);
+       }
+
+       if (rx != NULL) {
+               omap_set_dma_transfer_params(mcspi_dma->dma_rx_channel,
+                               data_type, element_count, 1,
+                               OMAP_DMA_SYNC_ELEMENT,
+                               mcspi_dma->dma_rx_sync_dev, 1);
+
+               omap_set_dma_src_params(mcspi_dma->dma_rx_channel, 0,
+                               OMAP_DMA_AMODE_CONSTANT,
+                               rx_reg, 0, 0);
+
+               omap_set_dma_dest_params(mcspi_dma->dma_rx_channel, 0,
+                               OMAP_DMA_AMODE_POST_INC,
+                               xfer->rx_dma, 0, 0);
+       }
+
+       if (tx != NULL) {
+               omap_start_dma(mcspi_dma->dma_tx_channel);
+               omap2_mcspi_set_dma_req(spi, 0, 1);
+       }
+
+       if (rx != NULL) {
+               omap_start_dma(mcspi_dma->dma_rx_channel);
+               omap2_mcspi_set_dma_req(spi, 1, 1);
+       }
+
+       if (tx != NULL) {
+               wait_for_completion(&mcspi_dma->dma_tx_completion);
+               dma_unmap_single(NULL, xfer->tx_dma, count, DMA_TO_DEVICE);
+       }
+
+       if (rx != NULL) {
+               wait_for_completion(&mcspi_dma->dma_rx_completion);
+               dma_unmap_single(NULL, xfer->rx_dma, count, DMA_FROM_DEVICE);
+       }
+       return count;
+}
+
+static int mcspi_wait_for_reg_bit(void __iomem *reg, unsigned long bit)
+{
+       unsigned long timeout;
+
+       timeout = jiffies + msecs_to_jiffies(1000);
+       while (!(__raw_readl(reg) & bit)) {
+               if (time_after(jiffies, timeout))
+                       return -1;
+               cpu_relax();
+       }
+       return 0;
+}
+
+static unsigned
+omap2_mcspi_txrx_pio(struct spi_device *spi, struct spi_transfer *xfer)
+{
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_cs   *cs = spi->controller_state;
+       unsigned int            count, c;
+       u32                     l;
+       void __iomem            *base = cs->base;
+       void __iomem            *tx_reg;
+       void __iomem            *rx_reg;
+       void __iomem            *chstat_reg;
+       int                     word_len;
+
+       mcspi = spi_master_get_devdata(spi->master);
+       count = xfer->len;
+       c = count;
+       word_len = cs->word_len;
+
+       l = mcspi_read_cs_reg(spi, OMAP2_MCSPI_CHCONF0);
+       l &= ~OMAP2_MCSPI_CHCONF_TRM_MASK;
+
+       /* We store the pre-calculated register addresses on stack to speed
+        * up the transfer loop. */
+       tx_reg          = base + OMAP2_MCSPI_TX0;
+       rx_reg          = base + OMAP2_MCSPI_RX0;
+       chstat_reg      = base + OMAP2_MCSPI_CHSTAT0;
+
+       if (word_len <= 8) {
+               u8              *rx;
+               const u8        *tx;
+
+               rx = xfer->rx_buf;
+               tx = xfer->tx_buf;
+
+               do {
+                       if (tx != NULL) {
+                               if (mcspi_wait_for_reg_bit(chstat_reg,
+                                               OMAP2_MCSPI_CHSTAT_TXS) < 0) {
+                                       dev_err(&spi->dev, "TXS timed out\n");
+                                       goto out;
+                               }
+#ifdef VERBOSE
+                               dev_dbg(&spi->dev, "write-%d %02x\n",
+                                               word_len, *tx);
+#endif
+                               __raw_writel(*tx++, tx_reg);
+                       }
+                       if (rx != NULL) {
+                               if (mcspi_wait_for_reg_bit(chstat_reg,
+                                               OMAP2_MCSPI_CHSTAT_RXS) < 0) {
+                                       dev_err(&spi->dev, "RXS timed out\n");
+                                       goto out;
+                               }
+                               /* prevent last RX_ONLY read from triggering
+                                * more word i/o: switch to rx+tx
+                                */
+                               if (c == 0 && tx == NULL)
+                                       mcspi_write_cs_reg(spi,
+                                                       OMAP2_MCSPI_CHCONF0, l);
+                               *rx++ = __raw_readl(rx_reg);
+#ifdef VERBOSE
+                               dev_dbg(&spi->dev, "read-%d %02x\n",
+                                               word_len, *(rx - 1));
+#endif
+                       }
+                       c -= 1;
+               } while (c);
+       } else if (word_len <= 16) {
+               u16             *rx;
+               const u16       *tx;
+
+               rx = xfer->rx_buf;
+               tx = xfer->tx_buf;
+               do {
+                       if (tx != NULL) {
+                               if (mcspi_wait_for_reg_bit(chstat_reg,
+                                               OMAP2_MCSPI_CHSTAT_TXS) < 0) {
+                                       dev_err(&spi->dev, "TXS timed out\n");
+                                       goto out;
+                               }
+#ifdef VERBOSE
+                               dev_dbg(&spi->dev, "write-%d %04x\n",
+                                               word_len, *tx);
+#endif
+                               __raw_writel(*tx++, tx_reg);
+                       }
+                       if (rx != NULL) {
+                               if (mcspi_wait_for_reg_bit(chstat_reg,
+                                               OMAP2_MCSPI_CHSTAT_RXS) < 0) {
+                                       dev_err(&spi->dev, "RXS timed out\n");
+                                       goto out;
+                               }
+                               /* prevent last RX_ONLY read from triggering
+                                * more word i/o: switch to rx+tx
+                                */
+                               if (c == 0 && tx == NULL)
+                                       mcspi_write_cs_reg(spi,
+                                                       OMAP2_MCSPI_CHCONF0, l);
+                               *rx++ = __raw_readl(rx_reg);
+#ifdef VERBOSE
+                               dev_dbg(&spi->dev, "read-%d %04x\n",
+                                               word_len, *(rx - 1));
+#endif
+                       }
+                       c -= 2;
+               } while (c);
+       } else if (word_len <= 32) {
+               u32             *rx;
+               const u32       *tx;
+
+               rx = xfer->rx_buf;
+               tx = xfer->tx_buf;
+               do {
+                       if (tx != NULL) {
+                               if (mcspi_wait_for_reg_bit(chstat_reg,
+                                               OMAP2_MCSPI_CHSTAT_TXS) < 0) {
+                                       dev_err(&spi->dev, "TXS timed out\n");
+                                       goto out;
+                               }
+#ifdef VERBOSE
+                               dev_dbg(&spi->dev, "write-%d %04x\n",
+                                               word_len, *tx);
+#endif
+                               __raw_writel(*tx++, tx_reg);
+                       }
+                       if (rx != NULL) {
+                               if (mcspi_wait_for_reg_bit(chstat_reg,
+                                               OMAP2_MCSPI_CHSTAT_RXS) < 0) {
+                                       dev_err(&spi->dev, "RXS timed out\n");
+                                       goto out;
+                               }
+                               /* prevent last RX_ONLY read from triggering
+                                * more word i/o: switch to rx+tx
+                                */
+                               if (c == 0 && tx == NULL)
+                                       mcspi_write_cs_reg(spi,
+                                                       OMAP2_MCSPI_CHCONF0, l);
+                               *rx++ = __raw_readl(rx_reg);
+#ifdef VERBOSE
+                               dev_dbg(&spi->dev, "read-%d %04x\n",
+                                               word_len, *(rx - 1));
+#endif
+                       }
+                       c -= 4;
+               } while (c);
+       }
+
+       /* for TX_ONLY mode, be sure all words have shifted out */
+       if (xfer->rx_buf == NULL) {
+               if (mcspi_wait_for_reg_bit(chstat_reg,
+                               OMAP2_MCSPI_CHSTAT_TXS) < 0) {
+                       dev_err(&spi->dev, "TXS timed out\n");
+               } else if (mcspi_wait_for_reg_bit(chstat_reg,
+                               OMAP2_MCSPI_CHSTAT_EOT) < 0)
+                       dev_err(&spi->dev, "EOT timed out\n");
+       }
+out:
+       return count - c;
+}
+
+/* called only when no transfer is active to this device */
+static int omap2_mcspi_setup_transfer(struct spi_device *spi,
+               struct spi_transfer *t)
+{
+       struct omap2_mcspi_cs *cs = spi->controller_state;
+       struct omap2_mcspi *mcspi;
+       u32 l = 0, div = 0;
+       u8 word_len = spi->bits_per_word;
+
+       mcspi = spi_master_get_devdata(spi->master);
+
+       if (t != NULL && t->bits_per_word)
+               word_len = t->bits_per_word;
+
+       cs->word_len = word_len;
+
+       if (spi->max_speed_hz) {
+               while (div <= 15 && (OMAP2_MCSPI_MAX_FREQ / (1 << div))
+                                       > spi->max_speed_hz)
+                       div++;
+       } else
+               div = 15;
+
+       l = mcspi_read_cs_reg(spi, OMAP2_MCSPI_CHCONF0);
+
+       /* standard 4-wire master mode:  SCK, MOSI/out, MISO/in, nCS
+        * REVISIT: this controller could support SPI_3WIRE mode.
+        */
+       l &= ~(OMAP2_MCSPI_CHCONF_IS|OMAP2_MCSPI_CHCONF_DPE1);
+       l |= OMAP2_MCSPI_CHCONF_DPE0;
+
+       /* wordlength */
+       l &= ~OMAP2_MCSPI_CHCONF_WL_MASK;
+       l |= (word_len - 1) << 7;
+
+       /* set chipselect polarity; manage with FORCE */
+       if (!(spi->mode & SPI_CS_HIGH))
+               l |= OMAP2_MCSPI_CHCONF_EPOL;   /* active-low; normal */
+       else
+               l &= ~OMAP2_MCSPI_CHCONF_EPOL;
+
+       /* set clock divisor */
+       l &= ~OMAP2_MCSPI_CHCONF_CLKD_MASK;
+       l |= div << 2;
+
+       /* set SPI mode 0..3 */
+       if (spi->mode & SPI_CPOL)
+               l |= OMAP2_MCSPI_CHCONF_POL;
+       else
+               l &= ~OMAP2_MCSPI_CHCONF_POL;
+       if (spi->mode & SPI_CPHA)
+               l |= OMAP2_MCSPI_CHCONF_PHA;
+       else
+               l &= ~OMAP2_MCSPI_CHCONF_PHA;
+
+       mcspi_write_cs_reg(spi, OMAP2_MCSPI_CHCONF0, l);
+
+       dev_dbg(&spi->dev, "setup: speed %d, sample %s edge, clk %s\n",
+                       OMAP2_MCSPI_MAX_FREQ / (1 << div),
+                       (spi->mode & SPI_CPHA) ? "trailing" : "leading",
+                       (spi->mode & SPI_CPOL) ? "inverted" : "normal");
+
+       return 0;
+}
+
+static void omap2_mcspi_dma_rx_callback(int lch, u16 ch_status, void *data)
+{
+       struct spi_device       *spi = data;
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_dma  *mcspi_dma;
+
+       mcspi = spi_master_get_devdata(spi->master);
+       mcspi_dma = &(mcspi->dma_channels[spi->chip_select]);
+
+       complete(&mcspi_dma->dma_rx_completion);
+
+       /* We must disable the DMA RX request */
+       omap2_mcspi_set_dma_req(spi, 1, 0);
+}
+
+static void omap2_mcspi_dma_tx_callback(int lch, u16 ch_status, void *data)
+{
+       struct spi_device       *spi = data;
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_dma  *mcspi_dma;
+
+       mcspi = spi_master_get_devdata(spi->master);
+       mcspi_dma = &(mcspi->dma_channels[spi->chip_select]);
+
+       complete(&mcspi_dma->dma_tx_completion);
+
+       /* We must disable the DMA TX request */
+       omap2_mcspi_set_dma_req(spi, 0, 0);
+}
+
+static int omap2_mcspi_request_dma(struct spi_device *spi)
+{
+       struct spi_master       *master = spi->master;
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_dma  *mcspi_dma;
+
+       mcspi = spi_master_get_devdata(master);
+       mcspi_dma = mcspi->dma_channels + spi->chip_select;
+
+       if (omap_request_dma(mcspi_dma->dma_rx_sync_dev, "McSPI RX",
+                       omap2_mcspi_dma_rx_callback, spi,
+                       &mcspi_dma->dma_rx_channel)) {
+               dev_err(&spi->dev, "no RX DMA channel for McSPI\n");
+               return -EAGAIN;
+       }
+
+       if (omap_request_dma(mcspi_dma->dma_tx_sync_dev, "McSPI TX",
+                       omap2_mcspi_dma_tx_callback, spi,
+                       &mcspi_dma->dma_tx_channel)) {
+               omap_free_dma(mcspi_dma->dma_rx_channel);
+               mcspi_dma->dma_rx_channel = -1;
+               dev_err(&spi->dev, "no TX DMA channel for McSPI\n");
+               return -EAGAIN;
+       }
+
+       init_completion(&mcspi_dma->dma_rx_completion);
+       init_completion(&mcspi_dma->dma_tx_completion);
+
+       return 0;
+}
+
+/* the spi->mode bits understood by this driver: */
+#define MODEBITS (SPI_CPOL | SPI_CPHA | SPI_CS_HIGH)
+
+static int omap2_mcspi_setup(struct spi_device *spi)
+{
+       int                     ret;
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_dma  *mcspi_dma;
+       struct omap2_mcspi_cs   *cs = spi->controller_state;
+
+       if (spi->mode & ~MODEBITS) {
+               dev_dbg(&spi->dev, "setup: unsupported mode bits %x\n",
+                       spi->mode & ~MODEBITS);
+               return -EINVAL;
+       }
+
+       if (spi->bits_per_word == 0)
+               spi->bits_per_word = 8;
+       else if (spi->bits_per_word < 4 || spi->bits_per_word > 32) {
+               dev_dbg(&spi->dev, "setup: unsupported %d bit words\n",
+                       spi->bits_per_word);
+               return -EINVAL;
+       }
+
+       mcspi = spi_master_get_devdata(spi->master);
+       mcspi_dma = &mcspi->dma_channels[spi->chip_select];
+
+       if (!cs) {
+               cs = kzalloc(sizeof *cs, GFP_KERNEL);
+               if (!cs)
+                       return -ENOMEM;
+               cs->base = mcspi->base + spi->chip_select * 0x14;
+               spi->controller_state = cs;
+       }
+
+       if (mcspi_dma->dma_rx_channel == -1
+                       || mcspi_dma->dma_tx_channel == -1) {
+               ret = omap2_mcspi_request_dma(spi);
+               if (ret < 0)
+                       return ret;
+       }
+
+       clk_enable(mcspi->ick);
+       clk_enable(mcspi->fck);
+       ret =  omap2_mcspi_setup_transfer(spi, NULL);
+       clk_disable(mcspi->fck);
+       clk_disable(mcspi->ick);
+
+       return ret;
+}
+
+static void omap2_mcspi_cleanup(struct spi_device *spi)
+{
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_dma  *mcspi_dma;
+
+       mcspi = spi_master_get_devdata(spi->master);
+       mcspi_dma = &mcspi->dma_channels[spi->chip_select];
+
+       kfree(spi->controller_state);
+
+       if (mcspi_dma->dma_rx_channel != -1) {
+               omap_free_dma(mcspi_dma->dma_rx_channel);
+               mcspi_dma->dma_rx_channel = -1;
+       }
+       if (mcspi_dma->dma_tx_channel != -1) {
+               omap_free_dma(mcspi_dma->dma_tx_channel);
+               mcspi_dma->dma_tx_channel = -1;
+       }
+}
+
+static void omap2_mcspi_work(struct work_struct *work)
+{
+       struct omap2_mcspi      *mcspi;
+
+       mcspi = container_of(work, struct omap2_mcspi, work);
+       spin_lock_irq(&mcspi->lock);
+
+       clk_enable(mcspi->ick);
+       clk_enable(mcspi->fck);
+
+       /* We only enable one channel at a time -- the one whose message is
+        * at the head of the queue -- although this controller would gladly
+        * arbitrate among multiple channels.  This corresponds to "single
+        * channel" master mode.  As a side effect, we need to manage the
+        * chipselect with the FORCE bit ... CS != channel enable.
+        */
+       while (!list_empty(&mcspi->msg_queue)) {
+               struct spi_message              *m;
+               struct spi_device               *spi;
+               struct spi_transfer             *t = NULL;
+               int                             cs_active = 0;
+               struct omap2_mcspi_device_config *conf;
+               struct omap2_mcspi_cs           *cs;
+               int                             par_override = 0;
+               int                             status = 0;
+               u32                             chconf;
+
+               m = container_of(mcspi->msg_queue.next, struct spi_message,
+                                queue);
+
+               list_del_init(&m->queue);
+               spin_unlock_irq(&mcspi->lock);
+
+               spi = m->spi;
+               conf = spi->controller_data;
+               cs = spi->controller_state;
+
+               omap2_mcspi_set_enable(spi, 1);
+               list_for_each_entry(t, &m->transfers, transfer_list) {
+                       if (t->tx_buf == NULL && t->rx_buf == NULL && t->len) {
+                               status = -EINVAL;
+                               break;
+                       }
+                       if (par_override || t->speed_hz || t->bits_per_word) {
+                               par_override = 1;
+                               status = omap2_mcspi_setup_transfer(spi, t);
+                               if (status < 0)
+                                       break;
+                               if (!t->speed_hz && !t->bits_per_word)
+                                       par_override = 0;
+                       }
+
+                       if (!cs_active) {
+                               omap2_mcspi_force_cs(spi, 1);
+                               cs_active = 1;
+                       }
+
+                       chconf = mcspi_read_cs_reg(spi, OMAP2_MCSPI_CHCONF0);
+                       chconf &= ~OMAP2_MCSPI_CHCONF_TRM_MASK;
+                       if (t->tx_buf == NULL)
+                               chconf |= OMAP2_MCSPI_CHCONF_TRM_RX_ONLY;
+                       else if (t->rx_buf == NULL)
+                               chconf |= OMAP2_MCSPI_CHCONF_TRM_TX_ONLY;
+                       mcspi_write_cs_reg(spi, OMAP2_MCSPI_CHCONF0, chconf);
+
+                       if (t->len) {
+                               unsigned        count;
+
+                               /* RX_ONLY mode needs dummy data in TX reg */
+                               if (t->tx_buf == NULL)
+                                       __raw_writel(0, cs->base
+                                                       + OMAP2_MCSPI_TX0);
+
+                               if (m->is_dma_mapped || t->len >= DMA_MIN_BYTES)
+                                       count = omap2_mcspi_txrx_dma(spi, t);
+                               else
+                                       count = omap2_mcspi_txrx_pio(spi, t);
+                               m->actual_length += count;
+
+                               if (count != t->len) {
+                                       status = -EIO;
+                                       break;
+                               }
+                       }
+
+                       if (t->delay_usecs)
+                               udelay(t->delay_usecs);
+
+                       /* ignore the "leave it on after last xfer" hint */
+                       if (t->cs_change) {
+                               omap2_mcspi_force_cs(spi, 0);
+                               cs_active = 0;
+                       }
+               }
+
+               /* Restore defaults if they were overriden */
+               if (par_override) {
+                       par_override = 0;
+                       status = omap2_mcspi_setup_transfer(spi, NULL);
+               }
+
+               if (cs_active)
+                       omap2_mcspi_force_cs(spi, 0);
+
+               omap2_mcspi_set_enable(spi, 0);
+
+               m->status = status;
+               m->complete(m->context);
+
+               spin_lock_irq(&mcspi->lock);
+       }
+
+       clk_disable(mcspi->fck);
+       clk_disable(mcspi->ick);
+
+       spin_unlock_irq(&mcspi->lock);
+}
+
+static int omap2_mcspi_transfer(struct spi_device *spi, struct spi_message *m)
+{
+       struct omap2_mcspi      *mcspi;
+       unsigned long           flags;
+       struct spi_transfer     *t;
+
+       m->actual_length = 0;
+       m->status = 0;
+
+       /* reject invalid messages and transfers */
+       if (list_empty(&m->transfers) || !m->complete)
+               return -EINVAL;
+       list_for_each_entry(t, &m->transfers, transfer_list) {
+               const void      *tx_buf = t->tx_buf;
+               void            *rx_buf = t->rx_buf;
+               unsigned        len = t->len;
+
+               if (t->speed_hz > OMAP2_MCSPI_MAX_FREQ
+                               || (len && !(rx_buf || tx_buf))
+                               || (t->bits_per_word &&
+                                       (  t->bits_per_word < 4
+                                       || t->bits_per_word > 32))) {
+                       dev_dbg(&spi->dev, "transfer: %d Hz, %d %s%s, %d bpw\n",
+                                       t->speed_hz,
+                                       len,
+                                       tx_buf ? "tx" : "",
+                                       rx_buf ? "rx" : "",
+                                       t->bits_per_word);
+                       return -EINVAL;
+               }
+               if (t->speed_hz && t->speed_hz < OMAP2_MCSPI_MAX_FREQ/(1<<16)) {
+                       dev_dbg(&spi->dev, "%d Hz max exceeds %d\n",
+                                       t->speed_hz,
+                                       OMAP2_MCSPI_MAX_FREQ/(1<<16));
+                       return -EINVAL;
+               }
+
+               if (m->is_dma_mapped || len < DMA_MIN_BYTES)
+                       continue;
+
+               /* Do DMA mapping "early" for better error reporting and
+                * dcache use.  Note that if dma_unmap_single() ever starts
+                * to do real work on ARM, we'd need to clean up mappings
+                * for previous transfers on *ALL* exits of this loop...
+                */
+               if (tx_buf != NULL) {
+                       t->tx_dma = dma_map_single(&spi->dev, (void *) tx_buf,
+                                       len, DMA_TO_DEVICE);
+                       if (dma_mapping_error(t->tx_dma)) {
+                               dev_dbg(&spi->dev, "dma %cX %d bytes error\n",
+                                               'T', len);
+                               return -EINVAL;
+                       }
+               }
+               if (rx_buf != NULL) {
+                       t->rx_dma = dma_map_single(&spi->dev, rx_buf, t->len,
+                                       DMA_FROM_DEVICE);
+                       if (dma_mapping_error(t->rx_dma)) {
+                               dev_dbg(&spi->dev, "dma %cX %d bytes error\n",
+                                               'R', len);
+                               if (tx_buf != NULL)
+                                       dma_unmap_single(NULL, t->tx_dma,
+                                                       len, DMA_TO_DEVICE);
+                               return -EINVAL;
+                       }
+               }
+       }
+
+       mcspi = spi_master_get_devdata(spi->master);
+
+       spin_lock_irqsave(&mcspi->lock, flags);
+       list_add_tail(&m->queue, &mcspi->msg_queue);
+       queue_work(omap2_mcspi_wq, &mcspi->work);
+       spin_unlock_irqrestore(&mcspi->lock, flags);
+
+       return 0;
+}
+
+static int __init omap2_mcspi_reset(struct omap2_mcspi *mcspi)
+{
+       struct spi_master       *master = mcspi->master;
+       u32                     tmp;
+
+       clk_enable(mcspi->ick);
+       clk_enable(mcspi->fck);
+
+       mcspi_write_reg(master, OMAP2_MCSPI_SYSCONFIG,
+                       OMAP2_MCSPI_SYSCONFIG_SOFTRESET);
+       do {
+               tmp = mcspi_read_reg(master, OMAP2_MCSPI_SYSSTATUS);
+       } while (!(tmp & OMAP2_MCSPI_SYSSTATUS_RESETDONE));
+
+       mcspi_write_reg(master, OMAP2_MCSPI_SYSCONFIG,
+                       /* (3 << 8) | (2 << 3) | */
+                       OMAP2_MCSPI_SYSCONFIG_AUTOIDLE);
+
+       omap2_mcspi_set_master_mode(master);
+
+       clk_disable(mcspi->fck);
+       clk_disable(mcspi->ick);
+       return 0;
+}
+
+static u8 __initdata spi1_rxdma_id [] = {
+       OMAP24XX_DMA_SPI1_RX0,
+       OMAP24XX_DMA_SPI1_RX1,
+       OMAP24XX_DMA_SPI1_RX2,
+       OMAP24XX_DMA_SPI1_RX3,
+};
+
+static u8 __initdata spi1_txdma_id [] = {
+       OMAP24XX_DMA_SPI1_TX0,
+       OMAP24XX_DMA_SPI1_TX1,
+       OMAP24XX_DMA_SPI1_TX2,
+       OMAP24XX_DMA_SPI1_TX3,
+};
+
+static u8 __initdata spi2_rxdma_id[] = {
+       OMAP24XX_DMA_SPI2_RX0,
+       OMAP24XX_DMA_SPI2_RX1,
+};
+
+static u8 __initdata spi2_txdma_id[] = {
+       OMAP24XX_DMA_SPI2_TX0,
+       OMAP24XX_DMA_SPI2_TX1,
+};
+
+static int __init omap2_mcspi_probe(struct platform_device *pdev)
+{
+       struct spi_master       *master;
+       struct omap2_mcspi      *mcspi;
+       struct resource         *r;
+       int                     status = 0, i;
+       const u8                *rxdma_id, *txdma_id;
+       unsigned                num_chipselect;
+
+       switch (pdev->id) {
+       case 1:
+               rxdma_id = spi1_rxdma_id;
+               txdma_id = spi1_txdma_id;
+               num_chipselect = 4;
+               break;
+       case 2:
+               rxdma_id = spi2_rxdma_id;
+               txdma_id = spi2_txdma_id;
+               num_chipselect = 2;
+               break;
+       /* REVISIT omap2430 has a third McSPI ... */
+       default:
+               return -EINVAL;
+       }
+
+       master = spi_alloc_master(&pdev->dev, sizeof *mcspi);
+       if (master == NULL) {
+               dev_dbg(&pdev->dev, "master allocation failed\n");
+               return -ENOMEM;
+       }
+
+       if (pdev->id != -1)
+               master->bus_num = pdev->id;
+
+       master->setup = omap2_mcspi_setup;
+       master->transfer = omap2_mcspi_transfer;
+       master->cleanup = omap2_mcspi_cleanup;
+       master->num_chipselect = num_chipselect;
+
+       dev_set_drvdata(&pdev->dev, master);
+
+       mcspi = spi_master_get_devdata(master);
+       mcspi->master = master;
+
+       r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       if (r == NULL) {
+               status = -ENODEV;
+               goto err1;
+       }
+       if (!request_mem_region(r->start, (r->end - r->start) + 1,
+                       pdev->dev.bus_id)) {
+               status = -EBUSY;
+               goto err1;
+       }
+
+       mcspi->base = (void __iomem *) io_p2v(r->start);
+
+       INIT_WORK(&mcspi->work, omap2_mcspi_work);
+
+       spin_lock_init(&mcspi->lock);
+       INIT_LIST_HEAD(&mcspi->msg_queue);
+
+       mcspi->ick = clk_get(&pdev->dev, "mcspi_ick");
+       if (IS_ERR(mcspi->ick)) {
+               dev_dbg(&pdev->dev, "can't get mcspi_ick\n");
+               status = PTR_ERR(mcspi->ick);
+               goto err1a;
+       }
+       mcspi->fck = clk_get(&pdev->dev, "mcspi_fck");
+       if (IS_ERR(mcspi->fck)) {
+               dev_dbg(&pdev->dev, "can't get mcspi_fck\n");
+               status = PTR_ERR(mcspi->fck);
+               goto err2;
+       }
+
+       mcspi->dma_channels = kcalloc(master->num_chipselect,
+                       sizeof(struct omap2_mcspi_dma),
+                       GFP_KERNEL);
+
+       if (mcspi->dma_channels == NULL)
+               goto err3;
+
+       for (i = 0; i < num_chipselect; i++) {
+               mcspi->dma_channels[i].dma_rx_channel = -1;
+               mcspi->dma_channels[i].dma_rx_sync_dev = rxdma_id[i];
+               mcspi->dma_channels[i].dma_tx_channel = -1;
+               mcspi->dma_channels[i].dma_tx_sync_dev = txdma_id[i];
+       }
+
+       if (omap2_mcspi_reset(mcspi) < 0)
+               goto err4;
+
+       status = spi_register_master(master);
+       if (status < 0)
+               goto err4;
+
+       return status;
+
+err4:
+       kfree(mcspi->dma_channels);
+err3:
+       clk_put(mcspi->fck);
+err2:
+       clk_put(mcspi->ick);
+err1a:
+       release_mem_region(r->start, (r->end - r->start) + 1);
+err1:
+       spi_master_put(master);
+       return status;
+}
+
+static int __exit omap2_mcspi_remove(struct platform_device *pdev)
+{
+       struct spi_master       *master;
+       struct omap2_mcspi      *mcspi;
+       struct omap2_mcspi_dma  *dma_channels;
+       struct resource         *r;
+
+       master = dev_get_drvdata(&pdev->dev);
+       mcspi = spi_master_get_devdata(master);
+       dma_channels = mcspi->dma_channels;
+
+       clk_put(mcspi->fck);
+       clk_put(mcspi->ick);
+
+       r = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+       release_mem_region(r->start, (r->end - r->start) + 1);
+
+       spi_unregister_master(master);
+       kfree(dma_channels);
+
+       return 0;
+}
+
+static struct platform_driver omap2_mcspi_driver = {
+       .driver = {
+               .name =         "omap2_mcspi",
+               .owner =        THIS_MODULE,
+       },
+       .remove =       __exit_p(omap2_mcspi_remove),
+};
+
+
+static int __init omap2_mcspi_init(void)
+{
+       omap2_mcspi_wq = create_singlethread_workqueue(
+                               omap2_mcspi_driver.driver.name);
+       if (omap2_mcspi_wq == NULL)
+               return -1;
+       return platform_driver_probe(&omap2_mcspi_driver, omap2_mcspi_probe);
+}
+subsys_initcall(omap2_mcspi_init);
+
+static void __exit omap2_mcspi_exit(void)
+{
+       platform_driver_unregister(&omap2_mcspi_driver);
+
+       destroy_workqueue(omap2_mcspi_wq);
+}
+module_exit(omap2_mcspi_exit);
+
+MODULE_LICENSE("GPL");