spi/topcliff_pch: DMA support
Tomoya MORINAGA [Tue, 7 Jun 2011 05:50:11 +0000 (14:50 +0900)]
This patch enables this SPI driver works with DMA mode.

Signed-off-by: Tomoya MORINAGA <tomoya-linux@dsn.okisemi.com>
Signed-off-by: Grant Likely <grant.likely@secretlab.ca>

drivers/spi/spi-topcliff-pch.c

index be84e3a..1b8030b 100644 (file)
@@ -28,6 +28,9 @@
 #include <linux/device.h>
 #include <linux/platform_device.h>
 
+#include <linux/dmaengine.h>
+#include <linux/pch_dma.h>
+
 /* Register offsets */
 #define PCH_SPCR               0x00    /* SPI control register */
 #define PCH_SPBRR              0x04    /* SPI baud rate register */
@@ -36,7 +39,7 @@
 #define PCH_SPDRR              0x10    /* SPI read data register */
 #define PCH_SSNXCR             0x18    /* SSN Expand Control Register */
 #define PCH_SRST               0x1C    /* SPI reset register */
-#define PCH_SPI_ADDRESS_SIZE   0x20
+#define PCH_ADDRESS_SIZE       0x20
 
 #define PCH_SPSR_TFD           0x000007C0
 #define PCH_SPSR_RFD           0x0000F800
@@ -54,8 +57,6 @@
 #define STATUS_EXITING         2
 #define PCH_SLEEP_TIME         10
 
-#define PCH_ADDRESS_SIZE       0x20
-
 #define SSN_LOW                        0x02U
 #define SSN_NO_CONTROL         0x00U
 #define PCH_MAX_CS             0xFF
@@ -75,6 +76,7 @@
 #define SPSR_TFI_BIT           (1 << 0)
 #define SPSR_RFI_BIT           (1 << 1)
 #define SPSR_FI_BIT            (1 << 2)
+#define SPSR_ORF_BIT           (1 << 3)
 #define SPBRR_SIZE_BIT         (1 << 10)
 
 #define PCH_ALL                        (SPCR_TFIE_BIT|SPCR_RFIE_BIT|SPCR_FIE_BIT|\
 #define SPCR_RFIC_FIELD                20
 #define SPCR_TFIC_FIELD                16
 
-#define SPSR_INT_BITS          0x1F
-#define MASK_SPBRR_SPBR_BITS   (~((1 << 10) - 1))
-#define MASK_RFIC_SPCR_BITS    (~(0xf << 20))
-#define MASK_TFIC_SPCR_BITS    (~(0xf000f << 12))
+#define MASK_SPBRR_SPBR_BITS   ((1 << 10) - 1)
+#define MASK_RFIC_SPCR_BITS    (0xf << SPCR_RFIC_FIELD)
+#define MASK_TFIC_SPCR_BITS    (0xf << SPCR_TFIC_FIELD)
 
 #define PCH_CLOCK_HZ           50000000
 #define PCH_MAX_SPBR           1023
 */
 #define PCH_SPI_MAX_DEV                        2
 
+#define PCH_BUF_SIZE           4096
+#define PCH_DMA_TRANS_SIZE     12
+
+static int use_dma = 1;
+
+struct pch_spi_dma_ctrl {
+       struct dma_async_tx_descriptor  *desc_tx;
+       struct dma_async_tx_descriptor  *desc_rx;
+       struct pch_dma_slave            param_tx;
+       struct pch_dma_slave            param_rx;
+       struct dma_chan         *chan_tx;
+       struct dma_chan         *chan_rx;
+       struct scatterlist              *sg_tx_p;
+       struct scatterlist              *sg_rx_p;
+       struct scatterlist              sg_tx;
+       struct scatterlist              sg_rx;
+       int                             nent;
+       void                            *tx_buf_virt;
+       void                            *rx_buf_virt;
+       dma_addr_t                      tx_buf_dma;
+       dma_addr_t                      rx_buf_dma;
+};
 /**
  * struct pch_spi_data - Holds the SPI channel specific details
  * @io_remap_addr:             The remapped PCI base address
  */
 struct pch_spi_data {
        void __iomem *io_remap_addr;
+       unsigned long io_base_addr;
        struct spi_master *master;
        struct work_struct work;
        struct workqueue_struct *wk;
@@ -162,6 +186,8 @@ struct pch_spi_data {
        struct pch_spi_board_data *board_dat;
        struct platform_device  *plat_dev;
        int ch;
+       struct pch_spi_dma_ctrl dma;
+       int use_dma;
        u8 irq_reg_sts;
 };
 
@@ -273,10 +299,10 @@ static void pch_spi_handler_sub(struct pch_spi_data *data, u32 reg_spsr_val,
                        reg_spcr_val &= ~SPCR_RFIE_BIT; /* disable RFI */
 
                        /* reset rx threshold */
-                       reg_spcr_val &= MASK_RFIC_SPCR_BITS;
+                       reg_spcr_val &= ~MASK_RFIC_SPCR_BITS;
                        reg_spcr_val |= (PCH_RX_THOLD_MAX << SPCR_RFIC_FIELD);
-                       iowrite32(((reg_spcr_val) &= (~(SPCR_RFIE_BIT))),
-                                (io_remap_addr + PCH_SPCR));
+
+                       iowrite32(reg_spcr_val, (io_remap_addr + PCH_SPCR));
                }
 
                /* update counts */
@@ -287,12 +313,15 @@ static void pch_spi_handler_sub(struct pch_spi_data *data, u32 reg_spsr_val,
 
        /* if transfer complete interrupt */
        if (reg_spsr_val & SPSR_FI_BIT) {
-               /* disable FI & RFI interrupts */
-               pch_spi_setclr_reg(data->master, PCH_SPCR, 0,
-                                  SPCR_FIE_BIT | SPCR_RFIE_BIT);
+               if (tx_index < bpw_len)
+                       dev_err(&data->master->dev,
+                               "%s : Transfer is not completed", __func__);
+               /* disable interrupts */
+               pch_spi_setclr_reg(data->master, PCH_SPCR, 0, PCH_ALL);
 
                /* transfer is completed;inform pch_spi_process_messages */
                data->transfer_complete = true;
+               data->transfer_active = false;
                wake_up(&data->wait);
        }
 }
@@ -316,12 +345,17 @@ static irqreturn_t pch_spi_handler(int irq, void *dev_id)
                        "%s returning due to suspend\n", __func__);
                return IRQ_NONE;
        }
+       if (data->use_dma)
+               return IRQ_NONE;
 
        io_remap_addr = data->io_remap_addr;
        spsr = io_remap_addr + PCH_SPSR;
 
        reg_spsr_val = ioread32(spsr);
 
+       if (reg_spsr_val & SPSR_ORF_BIT)
+               dev_err(&board_dat->pdev->dev, "%s Over run error", __func__);
+
        /* Check if the interrupt is for SPI device */
        if (reg_spsr_val & (SPSR_FI_BIT | SPSR_RFI_BIT)) {
                pch_spi_handler_sub(data, reg_spsr_val, io_remap_addr);
@@ -347,7 +381,7 @@ static void pch_spi_set_baud_rate(struct spi_master *master, u32 speed_hz)
        if (n_spbr > PCH_MAX_SPBR)
                n_spbr = PCH_MAX_SPBR;
 
-       pch_spi_setclr_reg(master, PCH_SPBRR, n_spbr, ~MASK_SPBRR_SPBR_BITS);
+       pch_spi_setclr_reg(master, PCH_SPBRR, n_spbr, MASK_SPBRR_SPBR_BITS);
 }
 
 /**
@@ -456,26 +490,27 @@ static int pch_spi_transfer(struct spi_device *pspi, struct spi_message *pmsg)
        dev_dbg(&pspi->dev, "%s Transfer List not empty. "
                "Transfer Speed is set.\n", __func__);
 
+       spin_lock_irqsave(&data->lock, flags);
        /* validate Tx/Rx buffers and Transfer length */
        list_for_each_entry(transfer, &pmsg->transfers, transfer_list) {
                if (!transfer->tx_buf && !transfer->rx_buf) {
                        dev_err(&pspi->dev,
                                "%s Tx and Rx buffer NULL\n", __func__);
                        retval = -EINVAL;
-                       goto err_out;
+                       goto err_return_spinlock;
                }
 
                if (!transfer->len) {
                        dev_err(&pspi->dev, "%s Transfer length invalid\n",
                                __func__);
                        retval = -EINVAL;
-                       goto err_out;
+                       goto err_return_spinlock;
                }
 
                dev_dbg(&pspi->dev, "%s Tx/Rx buffer valid. Transfer length"
                        " valid\n", __func__);
 
-               /* if baud rate hs been specified validate the same */
+               /* if baud rate has been specified validate the same */
                if (transfer->speed_hz > PCH_MAX_BAUDRATE)
                        transfer->speed_hz = PCH_MAX_BAUDRATE;
 
@@ -486,25 +521,24 @@ static int pch_spi_transfer(struct spi_device *pspi, struct spi_message *pmsg)
                                retval = -EINVAL;
                                dev_err(&pspi->dev,
                                        "%s Invalid bits per word\n", __func__);
-                               goto err_out;
+                               goto err_return_spinlock;
                        }
                }
        }
-
-       spin_lock_irqsave(&data->lock, flags);
+       spin_unlock_irqrestore(&data->lock, flags);
 
        /* We won't process any messages if we have been asked to terminate */
        if (data->status == STATUS_EXITING) {
                dev_err(&pspi->dev, "%s status = STATUS_EXITING.\n", __func__);
                retval = -ESHUTDOWN;
-               goto err_return_spinlock;
+               goto err_out;
        }
 
        /* If suspended ,return -EINVAL */
        if (data->board_dat->suspend_sts) {
                dev_err(&pspi->dev, "%s suspend; returning EINVAL\n", __func__);
                retval = -EINVAL;
-               goto err_return_spinlock;
+               goto err_out;
        }
 
        /* set status of message */
@@ -512,9 +546,11 @@ static int pch_spi_transfer(struct spi_device *pspi, struct spi_message *pmsg)
        dev_dbg(&pspi->dev, "%s - pmsg->status =%d\n", __func__, pmsg->status);
 
        pmsg->status = -EINPROGRESS;
-
+       spin_lock_irqsave(&data->lock, flags);
        /* add message to queue */
        list_add_tail(&pmsg->queue, &data->queue);
+       spin_unlock_irqrestore(&data->lock, flags);
+
        dev_dbg(&pspi->dev, "%s - Invoked list_add_tail\n", __func__);
 
        /* schedule work queue to run */
@@ -523,11 +559,13 @@ static int pch_spi_transfer(struct spi_device *pspi, struct spi_message *pmsg)
 
        retval = 0;
 
-err_return_spinlock:
-       spin_unlock_irqrestore(&data->lock, flags);
 err_out:
        dev_dbg(&pspi->dev, "%s RETURN=%d\n", __func__, retval);
        return retval;
+err_return_spinlock:
+       dev_dbg(&pspi->dev, "%s RETURN=%d\n", __func__, retval);
+       spin_unlock_irqrestore(&data->lock, flags);
+       return retval;
 }
 
 static inline void pch_spi_select_chip(struct pch_spi_data *data,
@@ -548,8 +586,7 @@ static inline void pch_spi_select_chip(struct pch_spi_data *data,
        pch_spi_setup_transfer(pspi);
 }
 
-static void pch_spi_set_tx(struct pch_spi_data *data, int *bpw,
-                          struct spi_message **ppmsg)
+static void pch_spi_set_tx(struct pch_spi_data *data, int *bpw)
 {
        int size;
        u32 n_writes;
@@ -558,8 +595,6 @@ static void pch_spi_set_tx(struct pch_spi_data *data, int *bpw,
        const u8 *tx_buf;
        const u16 *tx_sbuf;
 
-       pmsg = *ppmsg;
-
        /* set baud rate if needed */
        if (data->cur_trans->speed_hz) {
                dev_dbg(&data->master->dev, "%s:setting baud rate\n", __func__);
@@ -642,10 +677,9 @@ static void pch_spi_set_tx(struct pch_spi_data *data, int *bpw,
        data->transfer_active = true;
 }
 
-
-static void pch_spi_nomore_transfer(struct pch_spi_data *data,
-                                               struct spi_message *pmsg)
+static void pch_spi_nomore_transfer(struct pch_spi_data *data)
 {
+       struct spi_message *pmsg;
        dev_dbg(&data->master->dev, "%s called\n", __func__);
        /* Invoke complete callback
         * [To the spi core..indicating end of transfer] */
@@ -696,29 +730,21 @@ static void pch_spi_nomore_transfer(struct pch_spi_data *data,
 
 static void pch_spi_set_ir(struct pch_spi_data *data)
 {
-       /* enable interrupts */
-       if ((data->bpw_len) > PCH_MAX_FIFO_DEPTH) {
+       /* enable interrupts, set threshold, enable SPI */
+       if ((data->bpw_len) > PCH_MAX_FIFO_DEPTH)
                /* set receive threshold to PCH_RX_THOLD */
                pch_spi_setclr_reg(data->master, PCH_SPCR,
-                                  PCH_RX_THOLD << SPCR_RFIC_FIELD,
-                                  ~MASK_RFIC_SPCR_BITS);
-               /* enable FI and RFI interrupts */
-               pch_spi_setclr_reg(data->master, PCH_SPCR,
-                                  SPCR_RFIE_BIT | SPCR_FIE_BIT, 0);
-       } else {
+                                  PCH_RX_THOLD << SPCR_RFIC_FIELD |
+                                  SPCR_FIE_BIT | SPCR_RFIE_BIT |
+                                  SPCR_ORIE_BIT | SPCR_SPE_BIT,
+                                  MASK_RFIC_SPCR_BITS | PCH_ALL);
+       else
                /* set receive threshold to maximum */
                pch_spi_setclr_reg(data->master, PCH_SPCR,
-                                  PCH_RX_THOLD_MAX << SPCR_TFIC_FIELD,
-                                  ~MASK_TFIC_SPCR_BITS);
-               /* enable FI interrupt */
-               pch_spi_setclr_reg(data->master, PCH_SPCR, SPCR_FIE_BIT, 0);
-       }
-
-       dev_dbg(&data->master->dev,
-               "%s:invoking pch_spi_set_enable to enable SPI\n", __func__);
-
-       /* SPI set enable */
-       pch_spi_setclr_reg(data->current_chip->master, PCH_SPCR, SPCR_SPE_BIT, 0);
+                                  PCH_RX_THOLD_MAX << SPCR_RFIC_FIELD |
+                                  SPCR_FIE_BIT | SPCR_ORIE_BIT |
+                                  SPCR_SPE_BIT,
+                                  MASK_RFIC_SPCR_BITS | PCH_ALL);
 
        /* Wait until the transfer completes; go to sleep after
                                 initiating the transfer. */
@@ -731,15 +757,13 @@ static void pch_spi_set_ir(struct pch_spi_data *data)
        dev_dbg(&data->master->dev,
                "%s:no more control over SSN-writing 0 to SSNXCR.", __func__);
 
-       data->transfer_active = false;
-       dev_dbg(&data->master->dev,
-               "%s set data->transfer_active = false\n", __func__);
-
        /* clear all interrupts */
        pch_spi_writereg(data->master, PCH_SPSR,
                         pch_spi_readreg(data->master, PCH_SPSR));
-       /* disable interrupts */
-       pch_spi_setclr_reg(data->master, PCH_SPCR, 0, PCH_ALL);
+       /* Disable interrupts and SPI transfer */
+       pch_spi_setclr_reg(data->master, PCH_SPCR, 0, PCH_ALL | SPCR_SPE_BIT);
+       /* clear FIFO */
+       pch_spi_clear_fifo(data->master);
 }
 
 static void pch_spi_copy_rx_data(struct pch_spi_data *data, int bpw)
@@ -763,6 +787,327 @@ static void pch_spi_copy_rx_data(struct pch_spi_data *data, int bpw)
        }
 }
 
+static void pch_spi_copy_rx_data_for_dma(struct pch_spi_data *data, int bpw)
+{
+       int j;
+       u8 *rx_buf;
+       u16 *rx_sbuf;
+       const u8 *rx_dma_buf;
+       const u16 *rx_dma_sbuf;
+
+       /* copy Rx Data */
+       if (!data->cur_trans->rx_buf)
+               return;
+
+       if (bpw == 8) {
+               rx_buf = data->cur_trans->rx_buf;
+               rx_dma_buf = data->dma.rx_buf_virt;
+               for (j = 0; j < data->bpw_len; j++)
+                       *rx_buf++ = *rx_dma_buf++ & 0xFF;
+       } else {
+               rx_sbuf = data->cur_trans->rx_buf;
+               rx_dma_sbuf = data->dma.rx_buf_virt;
+               for (j = 0; j < data->bpw_len; j++)
+                       *rx_sbuf++ = *rx_dma_sbuf++;
+       }
+}
+
+static void pch_spi_start_transfer(struct pch_spi_data *data)
+{
+       struct pch_spi_dma_ctrl *dma;
+       unsigned long flags;
+
+       dma = &data->dma;
+
+       spin_lock_irqsave(&data->lock, flags);
+
+       /* disable interrupts, SPI set enable */
+       pch_spi_setclr_reg(data->master, PCH_SPCR, SPCR_SPE_BIT, PCH_ALL);
+
+       spin_unlock_irqrestore(&data->lock, flags);
+
+       /* Wait until the transfer completes; go to sleep after
+                                initiating the transfer. */
+       dev_dbg(&data->master->dev,
+               "%s:waiting for transfer to get over\n", __func__);
+       wait_event_interruptible(data->wait, data->transfer_complete);
+
+       dma_sync_sg_for_cpu(&data->master->dev, dma->sg_rx_p, dma->nent,
+                           DMA_FROM_DEVICE);
+       async_tx_ack(dma->desc_rx);
+       async_tx_ack(dma->desc_tx);
+       kfree(dma->sg_tx_p);
+       kfree(dma->sg_rx_p);
+
+       spin_lock_irqsave(&data->lock, flags);
+       pch_spi_writereg(data->master, PCH_SSNXCR, SSN_NO_CONTROL);
+       dev_dbg(&data->master->dev,
+               "%s:no more control over SSN-writing 0 to SSNXCR.", __func__);
+
+       /* clear fifo threshold, disable interrupts, disable SPI transfer */
+       pch_spi_setclr_reg(data->master, PCH_SPCR, 0,
+                          MASK_RFIC_SPCR_BITS | MASK_TFIC_SPCR_BITS | PCH_ALL |
+                          SPCR_SPE_BIT);
+       /* clear all interrupts */
+       pch_spi_writereg(data->master, PCH_SPSR,
+                        pch_spi_readreg(data->master, PCH_SPSR));
+       /* clear FIFO */
+       pch_spi_clear_fifo(data->master);
+
+       spin_unlock_irqrestore(&data->lock, flags);
+}
+
+static void pch_dma_rx_complete(void *arg)
+{
+       struct pch_spi_data *data = arg;
+
+       /* transfer is completed;inform pch_spi_process_messages_dma */
+       data->transfer_complete = true;
+       wake_up_interruptible(&data->wait);
+}
+
+static bool pch_spi_filter(struct dma_chan *chan, void *slave)
+{
+       struct pch_dma_slave *param = slave;
+
+       if ((chan->chan_id == param->chan_id) &&
+           (param->dma_dev == chan->device->dev)) {
+               chan->private = param;
+               return true;
+       } else {
+               return false;
+       }
+}
+
+static void pch_spi_request_dma(struct pch_spi_data *data, int bpw)
+{
+       dma_cap_mask_t mask;
+       struct dma_chan *chan;
+       struct pci_dev *dma_dev;
+       struct pch_dma_slave *param;
+       struct pch_spi_dma_ctrl *dma;
+       unsigned int width;
+
+       if (bpw == 8)
+               width = PCH_DMA_WIDTH_1_BYTE;
+       else
+               width = PCH_DMA_WIDTH_2_BYTES;
+
+       dma = &data->dma;
+       dma_cap_zero(mask);
+       dma_cap_set(DMA_SLAVE, mask);
+
+       /* Get DMA's dev information */
+       dma_dev = pci_get_bus_and_slot(2, PCI_DEVFN(12, 0));
+
+       /* Set Tx DMA */
+       param = &dma->param_tx;
+       param->dma_dev = &dma_dev->dev;
+       param->chan_id = data->master->bus_num * 2; /* Tx = 0, 2 */
+       param->tx_reg = data->io_base_addr + PCH_SPDWR;
+       param->width = width;
+       chan = dma_request_channel(mask, pch_spi_filter, param);
+       if (!chan) {
+               dev_err(&data->master->dev,
+                       "ERROR: dma_request_channel FAILS(Tx)\n");
+               data->use_dma = 0;
+               return;
+       }
+       dma->chan_tx = chan;
+
+       /* Set Rx DMA */
+       param = &dma->param_rx;
+       param->dma_dev = &dma_dev->dev;
+       param->chan_id = data->master->bus_num * 2 + 1; /* Rx = Tx + 1 */
+       param->rx_reg = data->io_base_addr + PCH_SPDRR;
+       param->width = width;
+       chan = dma_request_channel(mask, pch_spi_filter, param);
+       if (!chan) {
+               dev_err(&data->master->dev,
+                       "ERROR: dma_request_channel FAILS(Rx)\n");
+               dma_release_channel(dma->chan_tx);
+               dma->chan_tx = NULL;
+               data->use_dma = 0;
+               return;
+       }
+       dma->chan_rx = chan;
+}
+
+static void pch_spi_release_dma(struct pch_spi_data *data)
+{
+       struct pch_spi_dma_ctrl *dma;
+
+       dma = &data->dma;
+       if (dma->chan_tx) {
+               dma_release_channel(dma->chan_tx);
+               dma->chan_tx = NULL;
+       }
+       if (dma->chan_rx) {
+               dma_release_channel(dma->chan_rx);
+               dma->chan_rx = NULL;
+       }
+       return;
+}
+
+static void pch_spi_handle_dma(struct pch_spi_data *data, int *bpw)
+{
+       const u8 *tx_buf;
+       const u16 *tx_sbuf;
+       u8 *tx_dma_buf;
+       u16 *tx_dma_sbuf;
+       struct scatterlist *sg;
+       struct dma_async_tx_descriptor *desc_tx;
+       struct dma_async_tx_descriptor *desc_rx;
+       int num;
+       int i;
+       int size;
+       int rem;
+       unsigned long flags;
+       struct pch_spi_dma_ctrl *dma;
+
+       dma = &data->dma;
+
+       /* set baud rate if needed */
+       if (data->cur_trans->speed_hz) {
+               dev_dbg(&data->master->dev, "%s:setting baud rate\n", __func__);
+               spin_lock_irqsave(&data->lock, flags);
+               pch_spi_set_baud_rate(data->master, data->cur_trans->speed_hz);
+               spin_unlock_irqrestore(&data->lock, flags);
+       }
+
+       /* set bits per word if needed */
+       if (data->cur_trans->bits_per_word &&
+           (data->current_msg->spi->bits_per_word !=
+            data->cur_trans->bits_per_word)) {
+               dev_dbg(&data->master->dev, "%s:set bits per word\n", __func__);
+               spin_lock_irqsave(&data->lock, flags);
+               pch_spi_set_bits_per_word(data->master,
+                                         data->cur_trans->bits_per_word);
+               spin_unlock_irqrestore(&data->lock, flags);
+               *bpw = data->cur_trans->bits_per_word;
+       } else {
+               *bpw = data->current_msg->spi->bits_per_word;
+       }
+       data->bpw_len = data->cur_trans->len / (*bpw / 8);
+
+       /* copy Tx Data */
+       if (data->cur_trans->tx_buf != NULL) {
+               if (*bpw == 8) {
+                       tx_buf = data->cur_trans->tx_buf;
+                       tx_dma_buf = dma->tx_buf_virt;
+                       for (i = 0; i < data->bpw_len; i++)
+                               *tx_dma_buf++ = *tx_buf++;
+               } else {
+                       tx_sbuf = data->cur_trans->tx_buf;
+                       tx_dma_sbuf = dma->tx_buf_virt;
+                       for (i = 0; i < data->bpw_len; i++)
+                               *tx_dma_sbuf++ = *tx_sbuf++;
+               }
+       }
+       if (data->bpw_len > PCH_DMA_TRANS_SIZE) {
+               num = data->bpw_len / PCH_DMA_TRANS_SIZE + 1;
+               size = PCH_DMA_TRANS_SIZE;
+               rem = data->bpw_len % PCH_DMA_TRANS_SIZE;
+       } else {
+               num = 1;
+               size = data->bpw_len;
+               rem = data->bpw_len;
+       }
+       dev_dbg(&data->master->dev, "%s num=%d size=%d rem=%d\n",
+               __func__, num, size, rem);
+       spin_lock_irqsave(&data->lock, flags);
+
+       /* set receive fifo threshold and transmit fifo threshold */
+       pch_spi_setclr_reg(data->master, PCH_SPCR,
+                          ((size - 1) << SPCR_RFIC_FIELD) |
+                          ((PCH_MAX_FIFO_DEPTH - PCH_DMA_TRANS_SIZE) <<
+                           SPCR_TFIC_FIELD),
+                          MASK_RFIC_SPCR_BITS | MASK_TFIC_SPCR_BITS);
+
+       spin_unlock_irqrestore(&data->lock, flags);
+
+       /* RX */
+       dma->sg_rx_p = kzalloc(sizeof(struct scatterlist)*num, GFP_ATOMIC);
+       sg_init_table(dma->sg_rx_p, num); /* Initialize SG table */
+       /* offset, length setting */
+       sg = dma->sg_rx_p;
+       for (i = 0; i < num; i++, sg++) {
+               if (i == 0) {
+                       sg->offset = 0;
+                       sg_set_page(sg, virt_to_page(dma->rx_buf_virt), rem,
+                                   sg->offset);
+                       sg_dma_len(sg) = rem;
+               } else {
+                       sg->offset = rem + size * (i - 1);
+                       sg->offset = sg->offset * (*bpw / 8);
+                       sg_set_page(sg, virt_to_page(dma->rx_buf_virt), size,
+                                   sg->offset);
+                       sg_dma_len(sg) = size;
+               }
+               sg_dma_address(sg) = dma->rx_buf_dma + sg->offset;
+       }
+       sg = dma->sg_rx_p;
+       desc_rx = dma->chan_rx->device->device_prep_slave_sg(dma->chan_rx, sg,
+                                       num, DMA_FROM_DEVICE,
+                                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+       if (!desc_rx) {
+               dev_err(&data->master->dev, "%s:device_prep_slave_sg Failed\n",
+                       __func__);
+               return;
+       }
+       dma_sync_sg_for_device(&data->master->dev, sg, num, DMA_FROM_DEVICE);
+       desc_rx->callback = pch_dma_rx_complete;
+       desc_rx->callback_param = data;
+       dma->nent = num;
+       dma->desc_rx = desc_rx;
+
+       /* TX */
+       dma->sg_tx_p = kzalloc(sizeof(struct scatterlist)*num, GFP_ATOMIC);
+       sg_init_table(dma->sg_tx_p, num); /* Initialize SG table */
+       /* offset, length setting */
+       sg = dma->sg_tx_p;
+       for (i = 0; i < num; i++, sg++) {
+               if (i == 0) {
+                       sg->offset = 0;
+                       sg_set_page(sg, virt_to_page(dma->tx_buf_virt), rem,
+                                   sg->offset);
+                       sg_dma_len(sg) = rem;
+               } else {
+                       sg->offset = rem + size * (i - 1);
+                       sg->offset = sg->offset * (*bpw / 8);
+                       sg_set_page(sg, virt_to_page(dma->tx_buf_virt), size,
+                                   sg->offset);
+                       sg_dma_len(sg) = size;
+               }
+               sg_dma_address(sg) = dma->tx_buf_dma + sg->offset;
+       }
+       sg = dma->sg_tx_p;
+       desc_tx = dma->chan_tx->device->device_prep_slave_sg(dma->chan_tx,
+                                       sg, num, DMA_TO_DEVICE,
+                                       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+       if (!desc_tx) {
+               dev_err(&data->master->dev, "%s:device_prep_slave_sg Failed\n",
+                       __func__);
+               return;
+       }
+       dma_sync_sg_for_device(&data->master->dev, sg, num, DMA_TO_DEVICE);
+       desc_tx->callback = NULL;
+       desc_tx->callback_param = data;
+       dma->nent = num;
+       dma->desc_tx = desc_tx;
+
+       dev_dbg(&data->master->dev, "\n%s:Pulling down SSN low - writing "
+               "0x2 to SSNXCR\n", __func__);
+
+       spin_lock_irqsave(&data->lock, flags);
+       pch_spi_writereg(data->master, PCH_SSNXCR, SSN_LOW);
+       desc_rx->tx_submit(desc_rx);
+       desc_tx->tx_submit(desc_tx);
+       spin_unlock_irqrestore(&data->lock, flags);
+
+       /* reset transfer complete flag */
+       data->transfer_complete = false;
+}
 
 static void pch_spi_process_messages(struct work_struct *pwork)
 {
@@ -774,13 +1119,10 @@ static void pch_spi_process_messages(struct work_struct *pwork)
        dev_dbg(&data->master->dev, "%s data initialized\n", __func__);
 
        spin_lock(&data->lock);
-
        /* check if suspend has been initiated;if yes flush queue */
        if (data->board_dat->suspend_sts || (data->status == STATUS_EXITING)) {
-               dev_dbg(&data->master->dev,
-                       "%s suspend/remove initiated,flushing queue\n",
-                       __func__);
-
+               dev_dbg(&data->master->dev, "%s suspend/remove initiated,"
+                       "flushing queue\n", __func__);
                list_for_each_entry(pmsg, data->queue.next, queue) {
                        pmsg->status = -EIO;
 
@@ -814,53 +1156,42 @@ static void pch_spi_process_messages(struct work_struct *pwork)
 
        spin_unlock(&data->lock);
 
+       if (data->use_dma)
+               pch_spi_request_dma(data,
+                                   data->current_msg->spi->bits_per_word);
        do {
                /* If we are already processing a message get the next
                transfer structure from the message otherwise retrieve
                the 1st transfer request from the message. */
                spin_lock(&data->lock);
-
                if (data->cur_trans == NULL) {
                        data->cur_trans =
-                           list_entry(data->current_msg->transfers.
-                                      next, struct spi_transfer,
-                                      transfer_list);
-                       dev_dbg(&data->master->dev,
-                               "%s :Getting 1st transfer message\n", __func__);
+                               list_entry(data->current_msg->transfers.next,
+                                          struct spi_transfer, transfer_list);
+                       dev_dbg(&data->master->dev, "%s "
+                               ":Getting 1st transfer message\n", __func__);
                } else {
                        data->cur_trans =
-                           list_entry(data->cur_trans->transfer_list.next,
-                                      struct spi_transfer,
-                                      transfer_list);
-                       dev_dbg(&data->master->dev,
-                               "%s :Getting next transfer message\n",
-                               __func__);
+                               list_entry(data->cur_trans->transfer_list.next,
+                                          struct spi_transfer, transfer_list);
+                       dev_dbg(&data->master->dev, "%s "
+                               ":Getting next transfer message\n", __func__);
                }
-
                spin_unlock(&data->lock);
 
-               pch_spi_set_tx(data, &bpw, &pmsg);
-
-               /* Control interrupt*/
-               pch_spi_set_ir(data);
-
-               /* Disable SPI transfer */
-               pch_spi_setclr_reg(data->current_chip->master, PCH_SPCR, 0,
-                                  SPCR_SPE_BIT);
-
-               /* clear FIFO */
-               pch_spi_clear_fifo(data->master);
-
-               /* copy Rx Data */
-               pch_spi_copy_rx_data(data, bpw);
-
-               /* free memory */
-               kfree(data->pkt_rx_buff);
-               data->pkt_rx_buff = NULL;
-
-               kfree(data->pkt_tx_buff);
-               data->pkt_tx_buff = NULL;
-
+               if (data->use_dma) {
+                       pch_spi_handle_dma(data, &bpw);
+                       pch_spi_start_transfer(data);
+                       pch_spi_copy_rx_data_for_dma(data, bpw);
+               } else {
+                       pch_spi_set_tx(data, &bpw);
+                       pch_spi_set_ir(data);
+                       pch_spi_copy_rx_data(data, bpw);
+                       kfree(data->pkt_rx_buff);
+                       data->pkt_rx_buff = NULL;
+                       kfree(data->pkt_tx_buff);
+                       data->pkt_tx_buff = NULL;
+               }
                /* increment message count */
                data->current_msg->actual_length += data->cur_trans->len;
 
@@ -881,12 +1212,15 @@ static void pch_spi_process_messages(struct work_struct *pwork)
                /* No more transfer in this message. */
                if ((data->cur_trans->transfer_list.next) ==
                    &(data->current_msg->transfers)) {
-                       pch_spi_nomore_transfer(data, pmsg);
+                       pch_spi_nomore_transfer(data);
                }
 
                spin_unlock(&data->lock);
 
        } while (data->cur_trans != NULL);
+
+       if (data->use_dma)
+               pch_spi_release_dma(data);
 }
 
 static void pch_spi_free_resources(struct pch_spi_board_data *board_dat,
@@ -939,6 +1273,35 @@ err_return:
        return retval;
 }
 
+static void pch_free_dma_buf(struct pch_spi_board_data *board_dat,
+                            struct pch_spi_data *data)
+{
+       struct pch_spi_dma_ctrl *dma;
+
+       dma = &data->dma;
+       if (dma->tx_buf_dma)
+               dma_free_coherent(&board_dat->pdev->dev, PCH_BUF_SIZE,
+                                 dma->tx_buf_virt, dma->tx_buf_dma);
+       if (dma->rx_buf_dma)
+               dma_free_coherent(&board_dat->pdev->dev, PCH_BUF_SIZE,
+                                 dma->rx_buf_virt, dma->rx_buf_dma);
+       return;
+}
+
+static void pch_alloc_dma_buf(struct pch_spi_board_data *board_dat,
+                             struct pch_spi_data *data)
+{
+       struct pch_spi_dma_ctrl *dma;
+
+       dma = &data->dma;
+       /* Get Consistent memory for Tx DMA */
+       dma->tx_buf_virt = dma_alloc_coherent(&board_dat->pdev->dev,
+                               PCH_BUF_SIZE, &dma->tx_buf_dma, GFP_KERNEL);
+       /* Get Consistent memory for Rx DMA */
+       dma->rx_buf_virt = dma_alloc_coherent(&board_dat->pdev->dev,
+                               PCH_BUF_SIZE, &dma->rx_buf_dma, GFP_KERNEL);
+}
+
 static int __devinit pch_spi_pd_probe(struct platform_device *plat_dev)
 {
        int ret;
@@ -946,6 +1309,8 @@ static int __devinit pch_spi_pd_probe(struct platform_device *plat_dev)
        struct pch_spi_board_data *board_dat = dev_get_platdata(&plat_dev->dev);
        struct pch_spi_data *data;
 
+       dev_dbg(&plat_dev->dev, "%s:debug\n", __func__);
+
        master = spi_alloc_master(&board_dat->pdev->dev,
                                  sizeof(struct pch_spi_data));
        if (!master) {
@@ -959,9 +1324,11 @@ static int __devinit pch_spi_pd_probe(struct platform_device *plat_dev)
 
        platform_set_drvdata(plat_dev, data);
 
-       /* baseaddress + 0x20(offset) */
+       /* baseaddress + address offset) */
+       data->io_base_addr = pci_resource_start(board_dat->pdev, 1) +
+                                        PCH_ADDRESS_SIZE * plat_dev->id;
        data->io_remap_addr = pci_iomap(board_dat->pdev, 1, 0) +
-                                                  0x20 * plat_dev->id;
+                                        PCH_ADDRESS_SIZE * plat_dev->id;
        if (!data->io_remap_addr) {
                dev_err(&plat_dev->dev, "%s pci_iomap failed\n", __func__);
                ret = -ENOMEM;
@@ -982,6 +1349,7 @@ static int __devinit pch_spi_pd_probe(struct platform_device *plat_dev)
        data->n_curnt_chip = 255;
        data->status = STATUS_RUNNING;
        data->ch = plat_dev->id;
+       data->use_dma = use_dma;
 
        INIT_LIST_HEAD(&data->queue);
        spin_lock_init(&data->lock);
@@ -1012,6 +1380,11 @@ static int __devinit pch_spi_pd_probe(struct platform_device *plat_dev)
                goto err_spi_register_master;
        }
 
+       if (use_dma) {
+               dev_info(&plat_dev->dev, "Use DMA for data transfers\n");
+               pch_alloc_dma_buf(board_dat, data);
+       }
+
        return 0;
 
 err_spi_register_master:
@@ -1031,22 +1404,27 @@ static int __devexit pch_spi_pd_remove(struct platform_device *plat_dev)
        struct pch_spi_board_data *board_dat = dev_get_platdata(&plat_dev->dev);
        struct pch_spi_data *data = platform_get_drvdata(plat_dev);
        int count;
+       unsigned long flags;
 
        dev_dbg(&plat_dev->dev, "%s:[ch%d] irq=%d\n",
                __func__, plat_dev->id, board_dat->pdev->irq);
+
+       if (use_dma)
+               pch_free_dma_buf(board_dat, data);
+
        /* check for any pending messages; no action is taken if the queue
         * is still full; but at least we tried.  Unload anyway */
        count = 500;
-       spin_lock(&data->lock);
+       spin_lock_irqsave(&data->lock, flags);
        data->status = STATUS_EXITING;
        while ((list_empty(&data->queue) == 0) && --count) {
                dev_dbg(&board_dat->pdev->dev, "%s :queue not empty\n",
                        __func__);
-               spin_unlock(&data->lock);
+               spin_unlock_irqrestore(&data->lock, flags);
                msleep(PCH_SLEEP_TIME);
-               spin_lock(&data->lock);
+               spin_lock_irqsave(&data->lock, flags);
        }
-       spin_unlock(&data->lock);
+       spin_unlock_irqrestore(&data->lock, flags);
 
        pch_spi_free_resources(board_dat, data);
        /* disable interrupts & free IRQ */
@@ -1083,8 +1461,8 @@ static int pch_spi_pd_suspend(struct platform_device *pd_dev,
        /* check if the current message is processed:
           Only after thats done the transfer will be suspended */
        count = 255;
-       while ((--count) > 0)
-               if (!(data->bcurrent_msg_processing)) {
+       while ((--count) > 0) {
+               if (!(data->bcurrent_msg_processing))
                        break;
                msleep(PCH_SLEEP_TIME);
        }
@@ -1328,5 +1706,9 @@ static void __exit pch_spi_exit(void)
 }
 module_exit(pch_spi_exit);
 
+module_param(use_dma, int, 0644);
+MODULE_PARM_DESC(use_dma,
+                "to use DMA for data transfers pass 1 else 0; default 1");
+
 MODULE_LICENSE("GPL");
 MODULE_DESCRIPTION("Intel EG20T PCH/OKI SEMICONDUCTOR ML7213 IOH SPI Driver");