Blackfin: blackfin on-chip SPI controller driver
authorWu, Bryan <bryan.wu@analog.com>
Sun, 6 May 2007 21:50:34 +0000 (14:50 -0700)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Mon, 7 May 2007 19:12:58 +0000 (12:12 -0700)
This patch implements the driver necessary use the Analog Devices Blackfin
processor's SPI Port.

Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
drivers/spi/Kconfig
drivers/spi/Makefile
drivers/spi/spi_bfin5xx.c [new file with mode: 0644]

index 7e54e48efd5cb6af7c9451811871f303bef7c854..4a012d9acbfffb2832bb8d8ddb6e5b919b2dda3b 100644 (file)
@@ -58,6 +58,12 @@ config SPI_ATMEL
          This selects a driver for the Atmel SPI Controller, present on
          many AT32 (AVR32) and AT91 (ARM) chips.
 
+config SPI_BFIN
+       tristate "SPI controller driver for ADI Blackfin5xx"
+       depends on SPI_MASTER && BFIN
+       help
+         This is the SPI controller master driver for Blackfin 5xx processor.
+
 config SPI_BITBANG
        tristate "Bitbanging SPI master"
        depends on SPI_MASTER && EXPERIMENTAL
@@ -157,7 +163,6 @@ config SPI_AT25
 # Add new SPI protocol masters in alphabetical order above this line
 #
 
-
 # (slave support would go here)
 
 endmenu # "SPI support"
index 3c280ad892026bedacdb966c3c98a726a7ad4157..a95ade857a2f328eb859cd66c6f924ea73ec4d34 100644 (file)
@@ -11,8 +11,9 @@ endif
 obj-$(CONFIG_SPI_MASTER)               += spi.o
 
 # SPI master controller drivers (bus)
-obj-$(CONFIG_SPI_BITBANG)              += spi_bitbang.o
 obj-$(CONFIG_SPI_ATMEL)                        += atmel_spi.o
+obj-$(CONFIG_SPI_BFIN)                 += spi_bfin5xx.o
+obj-$(CONFIG_SPI_BITBANG)              += spi_bitbang.o
 obj-$(CONFIG_SPI_BUTTERFLY)            += spi_butterfly.o
 obj-$(CONFIG_SPI_IMX)                  += spi_imx.o
 obj-$(CONFIG_SPI_PXA2XX)               += pxa2xx_spi.o
diff --git a/drivers/spi/spi_bfin5xx.c b/drivers/spi/spi_bfin5xx.c
new file mode 100644 (file)
index 0000000..ce3c0ce
--- /dev/null
@@ -0,0 +1,1313 @@
+/*
+ * File:         drivers/spi/bfin5xx_spi.c
+ * Based on:     N/A
+ * Author:       Luke Yang (Analog Devices Inc.)
+ *
+ * Created:      March. 10th 2006
+ * Description:  SPI controller driver for Blackfin 5xx
+ * Bugs:         Enter bugs at http://blackfin.uclinux.org/
+ *
+ * Modified:
+ *     March 10, 2006  bfin5xx_spi.c Created. (Luke Yang)
+ *      August 7, 2006  added full duplex mode (Axel Weiss & Luke Yang)
+ *
+ * Copyright 2004-2006 Analog Devices Inc.
+ *
+ * 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, 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 ;  see the file COPYING.
+ * If not, write to the Free Software Foundation,
+ * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ */
+
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/device.h>
+#include <linux/ioport.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/platform_device.h>
+#include <linux/dma-mapping.h>
+#include <linux/spi/spi.h>
+#include <linux/workqueue.h>
+#include <linux/errno.h>
+#include <linux/delay.h>
+
+#include <asm/io.h>
+#include <asm/irq.h>
+#include <asm/delay.h>
+#include <asm/dma.h>
+
+#include <asm/bfin5xx_spi.h>
+
+MODULE_AUTHOR("Luke Yang");
+MODULE_DESCRIPTION("Blackfin 5xx SPI Contoller");
+MODULE_LICENSE("GPL");
+
+#define IS_DMA_ALIGNED(x) (((u32)(x)&0x07)==0)
+
+#define DEFINE_SPI_REG(reg, off) \
+static inline u16 read_##reg(void) \
+            { return *(volatile unsigned short*)(SPI0_REGBASE + off); } \
+static inline void write_##reg(u16 v) \
+            {*(volatile unsigned short*)(SPI0_REGBASE + off) = v;\
+             SSYNC();}
+
+DEFINE_SPI_REG(CTRL, 0x00)
+DEFINE_SPI_REG(FLAG, 0x04)
+DEFINE_SPI_REG(STAT, 0x08)
+DEFINE_SPI_REG(TDBR, 0x0C)
+DEFINE_SPI_REG(RDBR, 0x10)
+DEFINE_SPI_REG(BAUD, 0x14)
+DEFINE_SPI_REG(SHAW, 0x18)
+#define START_STATE ((void*)0)
+#define RUNNING_STATE ((void*)1)
+#define DONE_STATE ((void*)2)
+#define ERROR_STATE ((void*)-1)
+#define QUEUE_RUNNING 0
+#define QUEUE_STOPPED 1
+int dma_requested;
+
+struct driver_data {
+       /* Driver model hookup */
+       struct platform_device *pdev;
+
+       /* SPI framework hookup */
+       struct spi_master *master;
+
+       /* BFIN hookup */
+       struct bfin5xx_spi_master *master_info;
+
+       /* Driver message queue */
+       struct workqueue_struct *workqueue;
+       struct work_struct pump_messages;
+       spinlock_t lock;
+       struct list_head queue;
+       int busy;
+       int run;
+
+       /* Message Transfer pump */
+       struct tasklet_struct pump_transfers;
+
+       /* Current message transfer state info */
+       struct spi_message *cur_msg;
+       struct spi_transfer *cur_transfer;
+       struct chip_data *cur_chip;
+       size_t len_in_bytes;
+       size_t len;
+       void *tx;
+       void *tx_end;
+       void *rx;
+       void *rx_end;
+       int dma_mapped;
+       dma_addr_t rx_dma;
+       dma_addr_t tx_dma;
+       size_t rx_map_len;
+       size_t tx_map_len;
+       u8 n_bytes;
+       void (*write) (struct driver_data *);
+       void (*read) (struct driver_data *);
+       void (*duplex) (struct driver_data *);
+};
+
+struct chip_data {
+       u16 ctl_reg;
+       u16 baud;
+       u16 flag;
+
+       u8 chip_select_num;
+       u8 n_bytes;
+       u32 width;              /* 0 or 1 */
+       u8 enable_dma;
+       u8 bits_per_word;       /* 8 or 16 */
+       u8 cs_change_per_word;
+       u8 cs_chg_udelay;
+       void (*write) (struct driver_data *);
+       void (*read) (struct driver_data *);
+       void (*duplex) (struct driver_data *);
+};
+
+void bfin_spi_enable(struct driver_data *drv_data)
+{
+       u16 cr;
+
+       cr = read_CTRL();
+       write_CTRL(cr | BIT_CTL_ENABLE);
+       SSYNC();
+}
+
+void bfin_spi_disable(struct driver_data *drv_data)
+{
+       u16 cr;
+
+       cr = read_CTRL();
+       write_CTRL(cr & (~BIT_CTL_ENABLE));
+       SSYNC();
+}
+
+/* Caculate the SPI_BAUD register value based on input HZ */
+static u16 hz_to_spi_baud(u32 speed_hz)
+{
+       u_long sclk = get_sclk();
+       u16 spi_baud = (sclk / (2 * speed_hz));
+
+       if ((sclk % (2 * speed_hz)) > 0)
+               spi_baud++;
+
+       pr_debug("sclk = %ld, speed_hz = %d, spi_baud = %d\n", sclk, speed_hz,
+                spi_baud);
+
+       return spi_baud;
+}
+
+static int flush(struct driver_data *drv_data)
+{
+       unsigned long limit = loops_per_jiffy << 1;
+
+       /* wait for stop and clear stat */
+       while (!(read_STAT() & BIT_STAT_SPIF) && limit--)
+               continue;
+
+       write_STAT(BIT_STAT_CLR);
+
+       return limit;
+}
+
+/* stop controller and re-config current chip*/
+static void restore_state(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       /* Clear status and disable clock */
+       write_STAT(BIT_STAT_CLR);
+       bfin_spi_disable(drv_data);
+       pr_debug("restoring spi ctl state\n");
+
+#if defined(CONFIG_BF534) || defined(CONFIG_BF536) || defined(CONFIG_BF537)
+       pr_debug("chip select number is %d\n", chip->chip_select_num);
+
+       switch (chip->chip_select_num) {
+       case 1:
+               bfin_write_PORTF_FER(bfin_read_PORTF_FER() | 0x3c00);
+               SSYNC();
+               break;
+
+       case 2:
+       case 3:
+               bfin_write_PORT_MUX(bfin_read_PORT_MUX() | PJSE_SPI);
+               SSYNC();
+               bfin_write_PORTF_FER(bfin_read_PORTF_FER() | 0x3800);
+               SSYNC();
+               break;
+
+       case 4:
+               bfin_write_PORT_MUX(bfin_read_PORT_MUX() | PFS4E_SPI);
+               SSYNC();
+               bfin_write_PORTF_FER(bfin_read_PORTF_FER() | 0x3840);
+               SSYNC();
+               break;
+
+       case 5:
+               bfin_write_PORT_MUX(bfin_read_PORT_MUX() | PFS5E_SPI);
+               SSYNC();
+               bfin_write_PORTF_FER(bfin_read_PORTF_FER() | 0x3820);
+               SSYNC();
+               break;
+
+       case 6:
+               bfin_write_PORT_MUX(bfin_read_PORT_MUX() | PFS6E_SPI);
+               SSYNC();
+               bfin_write_PORTF_FER(bfin_read_PORTF_FER() | 0x3810);
+               SSYNC();
+               break;
+
+       case 7:
+               bfin_write_PORT_MUX(bfin_read_PORT_MUX() | PJCE_SPI);
+               SSYNC();
+               bfin_write_PORTF_FER(bfin_read_PORTF_FER() | 0x3800);
+               SSYNC();
+               break;
+       }
+#endif
+
+       /* Load the registers */
+       write_CTRL(chip->ctl_reg);
+       write_BAUD(chip->baud);
+       write_FLAG(chip->flag);
+}
+
+/* used to kick off transfer in rx mode */
+static unsigned short dummy_read(void)
+{
+       unsigned short tmp;
+       tmp = read_RDBR();
+       return tmp;
+}
+
+static void null_writer(struct driver_data *drv_data)
+{
+       u8 n_bytes = drv_data->n_bytes;
+
+       while (drv_data->tx < drv_data->tx_end) {
+               write_TDBR(0);
+               while ((read_STAT() & BIT_STAT_TXS))
+                       continue;
+               drv_data->tx += n_bytes;
+       }
+}
+
+static void null_reader(struct driver_data *drv_data)
+{
+       u8 n_bytes = drv_data->n_bytes;
+       dummy_read();
+
+       while (drv_data->rx < drv_data->rx_end) {
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               dummy_read();
+               drv_data->rx += n_bytes;
+       }
+}
+
+static void u8_writer(struct driver_data *drv_data)
+{
+       pr_debug("cr8-s is 0x%x\n", read_STAT());
+       while (drv_data->tx < drv_data->tx_end) {
+               write_TDBR(*(u8 *) (drv_data->tx));
+               while (read_STAT() & BIT_STAT_TXS)
+                       continue;
+               ++drv_data->tx;
+       }
+
+       /* poll for SPI completion before returning */
+       while (!(read_STAT() & BIT_STAT_SPIF))
+               continue;
+}
+
+static void u8_cs_chg_writer(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       while (drv_data->tx < drv_data->tx_end) {
+               write_FLAG(chip->flag);
+               SSYNC();
+
+               write_TDBR(*(u8 *) (drv_data->tx));
+               while (read_STAT() & BIT_STAT_TXS)
+                       continue;
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               write_FLAG(0xFF00 | chip->flag);
+               SSYNC();
+               if (chip->cs_chg_udelay)
+                       udelay(chip->cs_chg_udelay);
+               ++drv_data->tx;
+       }
+       write_FLAG(0xFF00);
+       SSYNC();
+}
+
+static void u8_reader(struct driver_data *drv_data)
+{
+       pr_debug("cr-8 is 0x%x\n", read_STAT());
+
+       /* clear TDBR buffer before read(else it will be shifted out) */
+       write_TDBR(0xFFFF);
+
+       dummy_read();
+
+       while (drv_data->rx < drv_data->rx_end - 1) {
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               *(u8 *) (drv_data->rx) = read_RDBR();
+               ++drv_data->rx;
+       }
+
+       while (!(read_STAT() & BIT_STAT_RXS))
+               continue;
+       *(u8 *) (drv_data->rx) = read_SHAW();
+       ++drv_data->rx;
+}
+
+static void u8_cs_chg_reader(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       while (drv_data->rx < drv_data->rx_end) {
+               write_FLAG(chip->flag);
+               SSYNC();
+
+               read_RDBR();    /* kick off */
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               *(u8 *) (drv_data->rx) = read_SHAW();
+               write_FLAG(0xFF00 | chip->flag);
+               SSYNC();
+               if (chip->cs_chg_udelay)
+                       udelay(chip->cs_chg_udelay);
+               ++drv_data->rx;
+       }
+       write_FLAG(0xFF00);
+       SSYNC();
+}
+
+static void u8_duplex(struct driver_data *drv_data)
+{
+       /* in duplex mode, clk is triggered by writing of TDBR */
+       while (drv_data->rx < drv_data->rx_end) {
+               write_TDBR(*(u8 *) (drv_data->tx));
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               *(u8 *) (drv_data->rx) = read_RDBR();
+               ++drv_data->rx;
+               ++drv_data->tx;
+       }
+}
+
+static void u8_cs_chg_duplex(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       while (drv_data->rx < drv_data->rx_end) {
+               write_FLAG(chip->flag);
+               SSYNC();
+
+               write_TDBR(*(u8 *) (drv_data->tx));
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               *(u8 *) (drv_data->rx) = read_RDBR();
+               write_FLAG(0xFF00 | chip->flag);
+               SSYNC();
+               if (chip->cs_chg_udelay)
+                       udelay(chip->cs_chg_udelay);
+               ++drv_data->rx;
+               ++drv_data->tx;
+       }
+       write_FLAG(0xFF00);
+       SSYNC();
+}
+
+static void u16_writer(struct driver_data *drv_data)
+{
+       pr_debug("cr16 is 0x%x\n", read_STAT());
+       while (drv_data->tx < drv_data->tx_end) {
+               write_TDBR(*(u16 *) (drv_data->tx));
+               while ((read_STAT() & BIT_STAT_TXS))
+                       continue;
+               drv_data->tx += 2;
+       }
+
+       /* poll for SPI completion before returning */
+       while (!(read_STAT() & BIT_STAT_SPIF))
+               continue;
+}
+
+static void u16_cs_chg_writer(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       while (drv_data->tx < drv_data->tx_end) {
+               write_FLAG(chip->flag);
+               SSYNC();
+
+               write_TDBR(*(u16 *) (drv_data->tx));
+               while ((read_STAT() & BIT_STAT_TXS))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               write_FLAG(0xFF00 | chip->flag);
+               SSYNC();
+               if (chip->cs_chg_udelay)
+                       udelay(chip->cs_chg_udelay);
+               drv_data->tx += 2;
+       }
+       write_FLAG(0xFF00);
+       SSYNC();
+}
+
+static void u16_reader(struct driver_data *drv_data)
+{
+       pr_debug("cr-16 is 0x%x\n", read_STAT());
+       dummy_read();
+
+       while (drv_data->rx < (drv_data->rx_end - 2)) {
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               *(u16 *) (drv_data->rx) = read_RDBR();
+               drv_data->rx += 2;
+       }
+
+       while (!(read_STAT() & BIT_STAT_RXS))
+               continue;
+       *(u16 *) (drv_data->rx) = read_SHAW();
+       drv_data->rx += 2;
+}
+
+static void u16_cs_chg_reader(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       while (drv_data->rx < drv_data->rx_end) {
+               write_FLAG(chip->flag);
+               SSYNC();
+
+               read_RDBR();    /* kick off */
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               *(u16 *) (drv_data->rx) = read_SHAW();
+               write_FLAG(0xFF00 | chip->flag);
+               SSYNC();
+               if (chip->cs_chg_udelay)
+                       udelay(chip->cs_chg_udelay);
+               drv_data->rx += 2;
+       }
+       write_FLAG(0xFF00);
+       SSYNC();
+}
+
+static void u16_duplex(struct driver_data *drv_data)
+{
+       /* in duplex mode, clk is triggered by writing of TDBR */
+       while (drv_data->tx < drv_data->tx_end) {
+               write_TDBR(*(u16 *) (drv_data->tx));
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               *(u16 *) (drv_data->rx) = read_RDBR();
+               drv_data->rx += 2;
+               drv_data->tx += 2;
+       }
+}
+
+static void u16_cs_chg_duplex(struct driver_data *drv_data)
+{
+       struct chip_data *chip = drv_data->cur_chip;
+
+       while (drv_data->tx < drv_data->tx_end) {
+               write_FLAG(chip->flag);
+               SSYNC();
+
+               write_TDBR(*(u16 *) (drv_data->tx));
+               while (!(read_STAT() & BIT_STAT_SPIF))
+                       continue;
+               while (!(read_STAT() & BIT_STAT_RXS))
+                       continue;
+               *(u16 *) (drv_data->rx) = read_RDBR();
+               write_FLAG(0xFF00 | chip->flag);
+               SSYNC();
+               if (chip->cs_chg_udelay)
+                       udelay(chip->cs_chg_udelay);
+               drv_data->rx += 2;
+               drv_data->tx += 2;
+       }
+       write_FLAG(0xFF00);
+       SSYNC();
+}
+
+/* test if ther is more transfer to be done */
+static void *next_transfer(struct driver_data *drv_data)
+{
+       struct spi_message *msg = drv_data->cur_msg;
+       struct spi_transfer *trans = drv_data->cur_transfer;
+
+       /* Move to next transfer */
+       if (trans->transfer_list.next != &msg->transfers) {
+               drv_data->cur_transfer =
+                   list_entry(trans->transfer_list.next,
+                              struct spi_transfer, transfer_list);
+               return RUNNING_STATE;
+       } else
+               return DONE_STATE;
+}
+
+/*
+ * caller already set message->status;
+ * dma and pio irqs are blocked give finished message back
+ */
+static void giveback(struct driver_data *drv_data)
+{
+       struct spi_transfer *last_transfer;
+       unsigned long flags;
+       struct spi_message *msg;
+
+       spin_lock_irqsave(&drv_data->lock, flags);
+       msg = drv_data->cur_msg;
+       drv_data->cur_msg = NULL;
+       drv_data->cur_transfer = NULL;
+       drv_data->cur_chip = NULL;
+       queue_work(drv_data->workqueue, &drv_data->pump_messages);
+       spin_unlock_irqrestore(&drv_data->lock, flags);
+
+       last_transfer = list_entry(msg->transfers.prev,
+                                  struct spi_transfer, transfer_list);
+
+       msg->state = NULL;
+
+       /* disable chip select signal. And not stop spi in autobuffer mode */
+       if (drv_data->tx_dma != 0xFFFF) {
+               write_FLAG(0xFF00);
+               bfin_spi_disable(drv_data);
+       }
+
+       if (msg->complete)
+               msg->complete(msg->context);
+}
+
+static irqreturn_t dma_irq_handler(int irq, void *dev_id, struct pt_regs *regs)
+{
+       struct driver_data *drv_data = (struct driver_data *)dev_id;
+       struct spi_message *msg = drv_data->cur_msg;
+
+       pr_debug("in dma_irq_handler\n");
+       clear_dma_irqstat(CH_SPI);
+
+       /*
+        * wait for the last transaction shifted out.  yes, these two
+        * while loops are supposed to be the same (see the HRM).
+        */
+       if (drv_data->tx != NULL) {
+               while (bfin_read_SPI_STAT() & TXS)
+                       continue;
+               while (bfin_read_SPI_STAT() & TXS)
+                       continue;
+       }
+
+       while (!(bfin_read_SPI_STAT() & SPIF))
+               continue;
+
+       bfin_spi_disable(drv_data);
+
+       msg->actual_length += drv_data->len_in_bytes;
+
+       /* Move to next transfer */
+       msg->state = next_transfer(drv_data);
+
+       /* Schedule transfer tasklet */
+       tasklet_schedule(&drv_data->pump_transfers);
+
+       /* free the irq handler before next transfer */
+       pr_debug("disable dma channel irq%d\n", CH_SPI);
+       dma_disable_irq(CH_SPI);
+
+       return IRQ_HANDLED;
+}
+
+static void pump_transfers(unsigned long data)
+{
+       struct driver_data *drv_data = (struct driver_data *)data;
+       struct spi_message *message = NULL;
+       struct spi_transfer *transfer = NULL;
+       struct spi_transfer *previous = NULL;
+       struct chip_data *chip = NULL;
+       u16 cr, width, dma_width, dma_config;
+       u32 tranf_success = 1;
+
+       /* Get current state information */
+       message = drv_data->cur_msg;
+       transfer = drv_data->cur_transfer;
+       chip = drv_data->cur_chip;
+
+       /*
+        * if msg is error or done, report it back using complete() callback
+        */
+
+        /* Handle for abort */
+       if (message->state == ERROR_STATE) {
+               message->status = -EIO;
+               giveback(drv_data);
+               return;
+       }
+
+       /* Handle end of message */
+       if (message->state == DONE_STATE) {
+               message->status = 0;
+               giveback(drv_data);
+               return;
+       }
+
+       /* Delay if requested at end of transfer */
+       if (message->state == RUNNING_STATE) {
+               previous = list_entry(transfer->transfer_list.prev,
+                                     struct spi_transfer, transfer_list);
+               if (previous->delay_usecs)
+                       udelay(previous->delay_usecs);
+       }
+
+       /* Setup the transfer state based on the type of transfer */
+       if (flush(drv_data) == 0) {
+               dev_err(&drv_data->pdev->dev, "pump_transfers: flush failed\n");
+               message->status = -EIO;
+               giveback(drv_data);
+               return;
+       }
+
+       if (transfer->tx_buf != NULL) {
+               drv_data->tx = (void *)transfer->tx_buf;
+               drv_data->tx_end = drv_data->tx + transfer->len;
+               pr_debug("tx_buf is %p, tx_end is %p\n", transfer->tx_buf,
+                        drv_data->tx_end);
+       } else {
+               drv_data->tx = NULL;
+       }
+
+       if (transfer->rx_buf != NULL) {
+               drv_data->rx = transfer->rx_buf;
+               drv_data->rx_end = drv_data->rx + transfer->len;
+               pr_debug("rx_buf is %p, rx_end is %p\n", transfer->rx_buf,
+                        drv_data->rx_end);
+       } else {
+               drv_data->rx = NULL;
+       }
+
+       drv_data->rx_dma = transfer->rx_dma;
+       drv_data->tx_dma = transfer->tx_dma;
+       drv_data->len_in_bytes = transfer->len;
+
+       width = chip->width;
+       if (width == CFG_SPI_WORDSIZE16) {
+               drv_data->len = (transfer->len) >> 1;
+       } else {
+               drv_data->len = transfer->len;
+       }
+       drv_data->write = drv_data->tx ? chip->write : null_writer;
+       drv_data->read = drv_data->rx ? chip->read : null_reader;
+       drv_data->duplex = chip->duplex ? chip->duplex : null_writer;
+       pr_debug
+           ("transfer: drv_data->write is %p, chip->write is %p, null_wr is %p\n",
+            drv_data->write, chip->write, null_writer);
+
+       /* speed and width has been set on per message */
+       message->state = RUNNING_STATE;
+       dma_config = 0;
+
+       /* restore spi status for each spi transfer */
+       if (transfer->speed_hz) {
+               write_BAUD(hz_to_spi_baud(transfer->speed_hz));
+       } else {
+               write_BAUD(chip->baud);
+       }
+       write_FLAG(chip->flag);
+
+       pr_debug("now pumping a transfer: width is %d, len is %d\n", width,
+                transfer->len);
+
+       /*
+        * Try to map dma buffer and do a dma transfer if
+        * successful use different way to r/w according to
+        * drv_data->cur_chip->enable_dma
+        */
+       if (drv_data->cur_chip->enable_dma && drv_data->len > 6) {
+
+               write_STAT(BIT_STAT_CLR);
+               disable_dma(CH_SPI);
+               clear_dma_irqstat(CH_SPI);
+               bfin_spi_disable(drv_data);
+
+               /* config dma channel */
+               pr_debug("doing dma transfer\n");
+               if (width == CFG_SPI_WORDSIZE16) {
+                       set_dma_x_count(CH_SPI, drv_data->len);
+                       set_dma_x_modify(CH_SPI, 2);
+                       dma_width = WDSIZE_16;
+               } else {
+                       set_dma_x_count(CH_SPI, drv_data->len);
+                       set_dma_x_modify(CH_SPI, 1);
+                       dma_width = WDSIZE_8;
+               }
+
+               /* set transfer width,direction. And enable spi */
+               cr = (read_CTRL() & (~BIT_CTL_TIMOD));
+
+               /* dirty hack for autobuffer DMA mode */
+               if (drv_data->tx_dma == 0xFFFF) {
+                       pr_debug("doing autobuffer DMA out.\n");
+
+                       /* no irq in autobuffer mode */
+                       dma_config =
+                           (DMAFLOW_AUTO | RESTART | dma_width | DI_EN);
+                       set_dma_config(CH_SPI, dma_config);
+                       set_dma_start_addr(CH_SPI, (unsigned long)drv_data->tx);
+                       enable_dma(CH_SPI);
+                       write_CTRL(cr | CFG_SPI_DMAWRITE | (width << 8) |
+                                  (CFG_SPI_ENABLE << 14));
+
+                       /* just return here, there can only be one transfer in this mode */
+                       message->status = 0;
+                       giveback(drv_data);
+                       return;
+               }
+
+               /* In dma mode, rx or tx must be NULL in one transfer */
+               if (drv_data->rx != NULL) {
+                       /* set transfer mode, and enable SPI */
+                       pr_debug("doing DMA in.\n");
+
+                       /* disable SPI before write to TDBR */
+                       write_CTRL(cr & ~BIT_CTL_ENABLE);
+
+                       /* clear tx reg soformer data is not shifted out */
+                       write_TDBR(0xFF);
+
+                       set_dma_x_count(CH_SPI, drv_data->len);
+
+                       /* start dma */
+                       dma_enable_irq(CH_SPI);
+                       dma_config = (WNR | RESTART | dma_width | DI_EN);
+                       set_dma_config(CH_SPI, dma_config);
+                       set_dma_start_addr(CH_SPI, (unsigned long)drv_data->rx);
+                       enable_dma(CH_SPI);
+
+                       cr |=
+                           CFG_SPI_DMAREAD | (width << 8) | (CFG_SPI_ENABLE <<
+                                                             14);
+                       /* set transfer mode, and enable SPI */
+                       write_CTRL(cr);
+               } else if (drv_data->tx != NULL) {
+                       pr_debug("doing DMA out.\n");
+
+                       /* start dma */
+                       dma_enable_irq(CH_SPI);
+                       dma_config = (RESTART | dma_width | DI_EN);
+                       set_dma_config(CH_SPI, dma_config);
+                       set_dma_start_addr(CH_SPI, (unsigned long)drv_data->tx);
+                       enable_dma(CH_SPI);
+
+                       write_CTRL(cr | CFG_SPI_DMAWRITE | (width << 8) |
+                                  (CFG_SPI_ENABLE << 14));
+
+               }
+       } else {
+               /* IO mode write then read */
+               pr_debug("doing IO transfer\n");
+
+               write_STAT(BIT_STAT_CLR);
+
+               if (drv_data->tx != NULL && drv_data->rx != NULL) {
+                       /* full duplex mode */
+                       BUG_ON((drv_data->tx_end - drv_data->tx) !=
+                              (drv_data->rx_end - drv_data->rx));
+                       cr = (read_CTRL() & (~BIT_CTL_TIMOD));  /* clear the TIMOD bits */
+                       cr |=
+                           CFG_SPI_WRITE | (width << 8) | (CFG_SPI_ENABLE <<
+                                                           14);
+                       pr_debug("IO duplex: cr is 0x%x\n", cr);
+
+                       write_CTRL(cr);
+                       SSYNC();
+
+                       drv_data->duplex(drv_data);
+
+                       if (drv_data->tx != drv_data->tx_end)
+                               tranf_success = 0;
+               } else if (drv_data->tx != NULL) {
+                       /* write only half duplex */
+                       cr = (read_CTRL() & (~BIT_CTL_TIMOD));  /* clear the TIMOD bits */
+                       cr |=
+                           CFG_SPI_WRITE | (width << 8) | (CFG_SPI_ENABLE <<
+                                                           14);
+                       pr_debug("IO write: cr is 0x%x\n", cr);
+
+                       write_CTRL(cr);
+                       SSYNC();
+
+                       drv_data->write(drv_data);
+
+                       if (drv_data->tx != drv_data->tx_end)
+                               tranf_success = 0;
+               } else if (drv_data->rx != NULL) {
+                       /* read only half duplex */
+                       cr = (read_CTRL() & (~BIT_CTL_TIMOD));  /* cleare the TIMOD bits */
+                       cr |=
+                           CFG_SPI_READ | (width << 8) | (CFG_SPI_ENABLE <<
+                                                          14);
+                       pr_debug("IO read: cr is 0x%x\n", cr);
+
+                       write_CTRL(cr);
+                       SSYNC();
+
+                       drv_data->read(drv_data);
+                       if (drv_data->rx != drv_data->rx_end)
+                               tranf_success = 0;
+               }
+
+               if (!tranf_success) {
+                       pr_debug("IO write error!\n");
+                       message->state = ERROR_STATE;
+               } else {
+                       /* Update total byte transfered */
+                       message->actual_length += drv_data->len;
+
+                       /* Move to next transfer of this msg */
+                       message->state = next_transfer(drv_data);
+               }
+
+               /* Schedule next transfer tasklet */
+               tasklet_schedule(&drv_data->pump_transfers);
+
+       }
+}
+
+/* pop a msg from queue and kick off real transfer */
+static void pump_messages(struct work_struct *work)
+{
+       struct driver_data *drv_data = container_of(work, struct driver_data, pump_messages);
+       unsigned long flags;
+
+       /* Lock queue and check for queue work */
+       spin_lock_irqsave(&drv_data->lock, flags);
+       if (list_empty(&drv_data->queue) || drv_data->run == QUEUE_STOPPED) {
+               /* pumper kicked off but no work to do */
+               drv_data->busy = 0;
+               spin_unlock_irqrestore(&drv_data->lock, flags);
+               return;
+       }
+
+       /* Make sure we are not already running a message */
+       if (drv_data->cur_msg) {
+               spin_unlock_irqrestore(&drv_data->lock, flags);
+               return;
+       }
+
+       /* Extract head of queue */
+       drv_data->cur_msg = list_entry(drv_data->queue.next,
+                                      struct spi_message, queue);
+       list_del_init(&drv_data->cur_msg->queue);
+
+       /* Initial message state */
+       drv_data->cur_msg->state = START_STATE;
+       drv_data->cur_transfer = list_entry(drv_data->cur_msg->transfers.next,
+                                           struct spi_transfer, transfer_list);
+
+       /* Setup the SSP using the per chip configuration */
+       drv_data->cur_chip = spi_get_ctldata(drv_data->cur_msg->spi);
+       restore_state(drv_data);
+       pr_debug
+           ("got a message to pump, state is set to: baud %d, flag 0x%x, ctl 0x%x\n",
+            drv_data->cur_chip->baud, drv_data->cur_chip->flag,
+            drv_data->cur_chip->ctl_reg);
+       pr_debug("the first transfer len is %d\n", drv_data->cur_transfer->len);
+
+       /* Mark as busy and launch transfers */
+       tasklet_schedule(&drv_data->pump_transfers);
+
+       drv_data->busy = 1;
+       spin_unlock_irqrestore(&drv_data->lock, flags);
+}
+
+/*
+ * got a msg to transfer, queue it in drv_data->queue.
+ * And kick off message pumper
+ */
+static int transfer(struct spi_device *spi, struct spi_message *msg)
+{
+       struct driver_data *drv_data = spi_master_get_devdata(spi->master);
+       unsigned long flags;
+
+       spin_lock_irqsave(&drv_data->lock, flags);
+
+       if (drv_data->run == QUEUE_STOPPED) {
+               spin_unlock_irqrestore(&drv_data->lock, flags);
+               return -ESHUTDOWN;
+       }
+
+       msg->actual_length = 0;
+       msg->status = -EINPROGRESS;
+       msg->state = START_STATE;
+
+       pr_debug("adding an msg in transfer() \n");
+       list_add_tail(&msg->queue, &drv_data->queue);
+
+       if (drv_data->run == QUEUE_RUNNING && !drv_data->busy)
+               queue_work(drv_data->workqueue, &drv_data->pump_messages);
+
+       spin_unlock_irqrestore(&drv_data->lock, flags);
+
+       return 0;
+}
+
+/* first setup for new devices */
+static int setup(struct spi_device *spi)
+{
+       struct bfin5xx_spi_chip *chip_info = NULL;
+       struct chip_data *chip;
+       struct driver_data *drv_data = spi_master_get_devdata(spi->master);
+       u8 spi_flg;
+
+       /* Abort device setup if requested features are not supported */
+       if (spi->mode & ~(SPI_CPOL | SPI_CPHA | SPI_LSB_FIRST)) {
+               dev_err(&spi->dev, "requested mode not fully supported\n");
+               return -EINVAL;
+       }
+
+       /* Zero (the default) here means 8 bits */
+       if (!spi->bits_per_word)
+               spi->bits_per_word = 8;
+
+       if (spi->bits_per_word != 8 && spi->bits_per_word != 16)
+               return -EINVAL;
+
+       /* Only alloc (or use chip_info) on first setup */
+       chip = spi_get_ctldata(spi);
+       if (chip == NULL) {
+               chip = kzalloc(sizeof(struct chip_data), GFP_KERNEL);
+               if (!chip)
+                       return -ENOMEM;
+
+               chip->enable_dma = 0;
+               chip_info = spi->controller_data;
+       }
+
+       /* chip_info isn't always needed */
+       if (chip_info) {
+               chip->enable_dma = chip_info->enable_dma != 0
+                   && drv_data->master_info->enable_dma;
+               chip->ctl_reg = chip_info->ctl_reg;
+               chip->bits_per_word = chip_info->bits_per_word;
+               chip->cs_change_per_word = chip_info->cs_change_per_word;
+               chip->cs_chg_udelay = chip_info->cs_chg_udelay;
+       }
+
+       /* translate common spi framework into our register */
+       if (spi->mode & SPI_CPOL)
+               chip->ctl_reg |= CPOL;
+       if (spi->mode & SPI_CPHA)
+               chip->ctl_reg |= CPHA;
+       if (spi->mode & SPI_LSB_FIRST)
+               chip->ctl_reg |= LSBF;
+       /* we dont support running in slave mode (yet?) */
+       chip->ctl_reg |= MSTR;
+
+       /*
+        * if any one SPI chip is registered and wants DMA, request the
+        * DMA channel for it
+        */
+       if (chip->enable_dma && !dma_requested) {
+               /* register dma irq handler */
+               if (request_dma(CH_SPI, "BF53x_SPI_DMA") < 0) {
+                       pr_debug
+                           ("Unable to request BlackFin SPI DMA channel\n");
+                       return -ENODEV;
+               }
+               if (set_dma_callback(CH_SPI, (void *)dma_irq_handler, drv_data)
+                   < 0) {
+                       pr_debug("Unable to set dma callback\n");
+                       return -EPERM;
+               }
+               dma_disable_irq(CH_SPI);
+               dma_requested = 1;
+       }
+
+       /*
+        * Notice: for blackfin, the speed_hz is the value of register
+        * SPI_BAUD, not the real baudrate
+        */
+       chip->baud = hz_to_spi_baud(spi->max_speed_hz);
+       spi_flg = ~(1 << (spi->chip_select));
+       chip->flag = ((u16) spi_flg << 8) | (1 << (spi->chip_select));
+       chip->chip_select_num = spi->chip_select;
+
+       switch (chip->bits_per_word) {
+       case 8:
+               chip->n_bytes = 1;
+               chip->width = CFG_SPI_WORDSIZE8;
+               chip->read = chip->cs_change_per_word ?
+                       u8_cs_chg_reader : u8_reader;
+               chip->write = chip->cs_change_per_word ?
+                       u8_cs_chg_writer : u8_writer;
+               chip->duplex = chip->cs_change_per_word ?
+                       u8_cs_chg_duplex : u8_duplex;
+               break;
+
+       case 16:
+               chip->n_bytes = 2;
+               chip->width = CFG_SPI_WORDSIZE16;
+               chip->read = chip->cs_change_per_word ?
+                       u16_cs_chg_reader : u16_reader;
+               chip->write = chip->cs_change_per_word ?
+                       u16_cs_chg_writer : u16_writer;
+               chip->duplex = chip->cs_change_per_word ?
+                       u16_cs_chg_duplex : u16_duplex;
+               break;
+
+       default:
+               dev_err(&spi->dev, "%d bits_per_word is not supported\n",
+                               chip->bits_per_word);
+               kfree(chip);
+               return -ENODEV;
+       }
+
+       pr_debug("setup spi chip %s, width is %d, dma is %d,",
+                       spi->modalias, chip->width, chip->enable_dma);
+       pr_debug("ctl_reg is 0x%x, flag_reg is 0x%x\n",
+                       chip->ctl_reg, chip->flag);
+
+       spi_set_ctldata(spi, chip);
+
+       return 0;
+}
+
+/*
+ * callback for spi framework.
+ * clean driver specific data
+ */
+static void cleanup(const struct spi_device *spi)
+{
+       struct chip_data *chip = spi_get_ctldata((struct spi_device *)spi);
+
+       kfree(chip);
+}
+
+static inline int init_queue(struct driver_data *drv_data)
+{
+       INIT_LIST_HEAD(&drv_data->queue);
+       spin_lock_init(&drv_data->lock);
+
+       drv_data->run = QUEUE_STOPPED;
+       drv_data->busy = 0;
+
+       /* init transfer tasklet */
+       tasklet_init(&drv_data->pump_transfers,
+                    pump_transfers, (unsigned long)drv_data);
+
+       /* init messages workqueue */
+       INIT_WORK(&drv_data->pump_messages, pump_messages);
+       drv_data->workqueue =
+           create_singlethread_workqueue(drv_data->master->cdev.dev->bus_id);
+       if (drv_data->workqueue == NULL)
+               return -EBUSY;
+
+       return 0;
+}
+
+static inline int start_queue(struct driver_data *drv_data)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&drv_data->lock, flags);
+
+       if (drv_data->run == QUEUE_RUNNING || drv_data->busy) {
+               spin_unlock_irqrestore(&drv_data->lock, flags);
+               return -EBUSY;
+       }
+
+       drv_data->run = QUEUE_RUNNING;
+       drv_data->cur_msg = NULL;
+       drv_data->cur_transfer = NULL;
+       drv_data->cur_chip = NULL;
+       spin_unlock_irqrestore(&drv_data->lock, flags);
+
+       queue_work(drv_data->workqueue, &drv_data->pump_messages);
+
+       return 0;
+}
+
+static inline int stop_queue(struct driver_data *drv_data)
+{
+       unsigned long flags;
+       unsigned limit = 500;
+       int status = 0;
+
+       spin_lock_irqsave(&drv_data->lock, flags);
+
+       /*
+        * This is a bit lame, but is optimized for the common execution path.
+        * A wait_queue on the drv_data->busy could be used, but then the common
+        * execution path (pump_messages) would be required to call wake_up or
+        * friends on every SPI message. Do this instead
+        */
+       drv_data->run = QUEUE_STOPPED;
+       while (!list_empty(&drv_data->queue) && drv_data->busy && limit--) {
+               spin_unlock_irqrestore(&drv_data->lock, flags);
+               msleep(10);
+               spin_lock_irqsave(&drv_data->lock, flags);
+       }
+
+       if (!list_empty(&drv_data->queue) || drv_data->busy)
+               status = -EBUSY;
+
+       spin_unlock_irqrestore(&drv_data->lock, flags);
+
+       return status;
+}
+
+static inline int destroy_queue(struct driver_data *drv_data)
+{
+       int status;
+
+       status = stop_queue(drv_data);
+       if (status != 0)
+               return status;
+
+       destroy_workqueue(drv_data->workqueue);
+
+       return 0;
+}
+
+static int __init bfin5xx_spi_probe(struct platform_device *pdev)
+{
+       struct device *dev = &pdev->dev;
+       struct bfin5xx_spi_master *platform_info;
+       struct spi_master *master;
+       struct driver_data *drv_data = 0;
+       int status = 0;
+
+       platform_info = dev->platform_data;
+
+       /* Allocate master with space for drv_data */
+       master = spi_alloc_master(dev, sizeof(struct driver_data) + 16);
+       if (!master) {
+               dev_err(&pdev->dev, "can not alloc spi_master\n");
+               return -ENOMEM;
+       }
+       drv_data = spi_master_get_devdata(master);
+       drv_data->master = master;
+       drv_data->master_info = platform_info;
+       drv_data->pdev = pdev;
+
+       master->bus_num = pdev->id;
+       master->num_chipselect = platform_info->num_chipselect;
+       master->cleanup = cleanup;
+       master->setup = setup;
+       master->transfer = transfer;
+
+       /* Initial and start queue */
+       status = init_queue(drv_data);
+       if (status != 0) {
+               dev_err(&pdev->dev, "problem initializing queue\n");
+               goto out_error_queue_alloc;
+       }
+       status = start_queue(drv_data);
+       if (status != 0) {
+               dev_err(&pdev->dev, "problem starting queue\n");
+               goto out_error_queue_alloc;
+       }
+
+       /* Register with the SPI framework */
+       platform_set_drvdata(pdev, drv_data);
+       status = spi_register_master(master);
+       if (status != 0) {
+               dev_err(&pdev->dev, "problem registering spi master\n");
+               goto out_error_queue_alloc;
+       }
+       pr_debug("controller probe successfully\n");
+       return status;
+
+      out_error_queue_alloc:
+       destroy_queue(drv_data);
+       spi_master_put(master);
+       return status;
+}
+
+/* stop hardware and remove the driver */
+static int __devexit bfin5xx_spi_remove(struct platform_device *pdev)
+{
+       struct driver_data *drv_data = platform_get_drvdata(pdev);
+       int status = 0;
+
+       if (!drv_data)
+               return 0;
+
+       /* Remove the queue */
+       status = destroy_queue(drv_data);
+       if (status != 0)
+               return status;
+
+       /* Disable the SSP at the peripheral and SOC level */
+       bfin_spi_disable(drv_data);
+
+       /* Release DMA */
+       if (drv_data->master_info->enable_dma) {
+               if (dma_channel_active(CH_SPI))
+                       free_dma(CH_SPI);
+       }
+
+       /* Disconnect from the SPI framework */
+       spi_unregister_master(drv_data->master);
+
+       /* Prevent double remove */
+       platform_set_drvdata(pdev, NULL);
+
+       return 0;
+}
+
+#ifdef CONFIG_PM
+static int bfin5xx_spi_suspend(struct platform_device *pdev, pm_message_t state)
+{
+       struct driver_data *drv_data = platform_get_drvdata(pdev);
+       int status = 0;
+
+       status = stop_queue(drv_data);
+       if (status != 0)
+               return status;
+
+       /* stop hardware */
+       bfin_spi_disable(drv_data);
+
+       return 0;
+}
+
+static int bfin5xx_spi_resume(struct platform_device *pdev)
+{
+       struct driver_data *drv_data = platform_get_drvdata(pdev);
+       int status = 0;
+
+       /* Enable the SPI interface */
+       bfin_spi_enable(drv_data);
+
+       /* Start the queue running */
+       status = start_queue(drv_data);
+       if (status != 0) {
+               dev_err(&pdev->dev, "problem starting queue (%d)\n", status);
+               return status;
+       }
+
+       return 0;
+}
+#else
+#define bfin5xx_spi_suspend NULL
+#define bfin5xx_spi_resume NULL
+#endif                         /* CONFIG_PM */
+
+static struct platform_driver bfin5xx_spi_driver = {
+       .driver = {
+                  .name = "bfin-spi-master",
+                  .bus = &platform_bus_type,
+                  .owner = THIS_MODULE,
+                  },
+       .probe = bfin5xx_spi_probe,
+       .remove = __devexit_p(bfin5xx_spi_remove),
+       .suspend = bfin5xx_spi_suspend,
+       .resume = bfin5xx_spi_resume,
+};
+
+static int __init bfin5xx_spi_init(void)
+{
+       return platform_driver_register(&bfin5xx_spi_driver);
+}
+
+module_init(bfin5xx_spi_init);
+
+static void __exit bfin5xx_spi_exit(void)
+{
+       platform_driver_unregister(&bfin5xx_spi_driver);
+}
+
+module_exit(bfin5xx_spi_exit);