]> git.karo-electronics.de Git - mv-sheeva.git/commitdiff
dmaengine: add CSR SiRFprimaII DMAC driver
authorRongjun Ying <Rongjun.Ying@csr.com>
Fri, 28 Oct 2011 02:22:39 +0000 (19:22 -0700)
committerVinod Koul <vinod.koul@linux.intel.com>
Fri, 18 Nov 2011 06:55:22 +0000 (12:25 +0530)
Cc: Jassi Brar <jaswinder.singh@linaro.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Rongjun Ying <rongjun.ying@csr.com>
Signed-off-by: Barry Song <Baohua.Song@csr.com>
[fixed direction enums and cyclic api based on changes
 already merged]
Signed-off-by: Vinod Koul <vinod.koul@linux.intel.com>
MAINTAINERS
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/sirf-dma.c [new file with mode: 0644]
include/linux/sirfsoc_dma.h [new file with mode: 0644]

index 4808256446f298437ea00cd5a229376806d868b9..1b141d71ea13214b381752a5889e5b9c9c580cd0 100644 (file)
@@ -749,6 +749,7 @@ M:  Barry Song <baohua.song@csr.com>
 L:     linux-arm-kernel@lists.infradead.org (moderated for non-subscribers)
 S:     Maintained
 F:     arch/arm/mach-prima2/
+F:     drivers/dma/sirf-dma*
 
 ARM/EBSA110 MACHINE SUPPORT
 M:     Russell King <linux@arm.linux.org.uk>
index 7ec0d6cef0c3b6fb96f4b27eb69b241a0f95c6d4..f1a274994bb1fbdc1c8e0e54b26aebe58b161c2b 100644 (file)
@@ -187,6 +187,13 @@ config TIMB_DMA
        help
          Enable support for the Timberdale FPGA DMA engine.
 
+config SIRF_DMA
+       tristate "CSR SiRFprimaII DMA support"
+       depends on ARCH_PRIMA2
+       select DMA_ENGINE
+       help
+         Enable support for the CSR SiRFprimaII DMA engine.
+
 config ARCH_HAS_ASYNC_TX_FIND_CHANNEL
        bool
 
index 30cf3b1f0c5ca4ff633d9e50474feef0a49fc565..009a222e8283cba3929fd8763ae9abe9b4d6d694 100644 (file)
@@ -21,6 +21,7 @@ obj-$(CONFIG_IMX_SDMA) += imx-sdma.o
 obj-$(CONFIG_IMX_DMA) += imx-dma.o
 obj-$(CONFIG_MXS_DMA) += mxs-dma.o
 obj-$(CONFIG_TIMB_DMA) += timb_dma.o
+obj-$(CONFIG_SIRF_DMA) += sirf-dma.o
 obj-$(CONFIG_STE_DMA40) += ste_dma40.o ste_dma40_ll.o
 obj-$(CONFIG_PL330_DMA) += pl330.o
 obj-$(CONFIG_PCH_DMA) += pch_dma.o
diff --git a/drivers/dma/sirf-dma.c b/drivers/dma/sirf-dma.c
new file mode 100644 (file)
index 0000000..55ec679
--- /dev/null
@@ -0,0 +1,717 @@
+/*
+ * DMA controller driver for CSR SiRFprimaII
+ *
+ * Copyright (c) 2011 Cambridge Silicon Radio Limited, a CSR plc group company.
+ *
+ * Licensed under GPLv2 or later.
+ */
+
+#include <linux/module.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/slab.h>
+#include <linux/of_irq.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/of_platform.h>
+#include <linux/sirfsoc_dma.h>
+
+#define SIRFSOC_DMA_DESCRIPTORS                 16
+#define SIRFSOC_DMA_CHANNELS                    16
+
+#define SIRFSOC_DMA_CH_ADDR                     0x00
+#define SIRFSOC_DMA_CH_XLEN                     0x04
+#define SIRFSOC_DMA_CH_YLEN                     0x08
+#define SIRFSOC_DMA_CH_CTRL                     0x0C
+
+#define SIRFSOC_DMA_WIDTH_0                     0x100
+#define SIRFSOC_DMA_CH_VALID                    0x140
+#define SIRFSOC_DMA_CH_INT                      0x144
+#define SIRFSOC_DMA_INT_EN                      0x148
+#define SIRFSOC_DMA_CH_LOOP_CTRL                0x150
+
+#define SIRFSOC_DMA_MODE_CTRL_BIT               4
+#define SIRFSOC_DMA_DIR_CTRL_BIT                5
+
+/* xlen and dma_width register is in 4 bytes boundary */
+#define SIRFSOC_DMA_WORD_LEN                   4
+
+struct sirfsoc_dma_desc {
+       struct dma_async_tx_descriptor  desc;
+       struct list_head                node;
+
+       /* SiRFprimaII 2D-DMA parameters */
+
+       int             xlen;           /* DMA xlen */
+       int             ylen;           /* DMA ylen */
+       int             width;          /* DMA width */
+       int             dir;
+       bool            cyclic;         /* is loop DMA? */
+       u32             addr;           /* DMA buffer address */
+};
+
+struct sirfsoc_dma_chan {
+       struct dma_chan                 chan;
+       struct list_head                free;
+       struct list_head                prepared;
+       struct list_head                queued;
+       struct list_head                active;
+       struct list_head                completed;
+       dma_cookie_t                    completed_cookie;
+       unsigned long                   happened_cyclic;
+       unsigned long                   completed_cyclic;
+
+       /* Lock for this structure */
+       spinlock_t                      lock;
+
+       int                             mode;
+};
+
+struct sirfsoc_dma {
+       struct dma_device               dma;
+       struct tasklet_struct           tasklet;
+       struct sirfsoc_dma_chan         channels[SIRFSOC_DMA_CHANNELS];
+       void __iomem                    *base;
+       int                             irq;
+};
+
+#define DRV_NAME       "sirfsoc_dma"
+
+/* Convert struct dma_chan to struct sirfsoc_dma_chan */
+static inline
+struct sirfsoc_dma_chan *dma_chan_to_sirfsoc_dma_chan(struct dma_chan *c)
+{
+       return container_of(c, struct sirfsoc_dma_chan, chan);
+}
+
+/* Convert struct dma_chan to struct sirfsoc_dma */
+static inline struct sirfsoc_dma *dma_chan_to_sirfsoc_dma(struct dma_chan *c)
+{
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(c);
+       return container_of(schan, struct sirfsoc_dma, channels[c->chan_id]);
+}
+
+/* Execute all queued DMA descriptors */
+static void sirfsoc_dma_execute(struct sirfsoc_dma_chan *schan)
+{
+       struct sirfsoc_dma *sdma = dma_chan_to_sirfsoc_dma(&schan->chan);
+       int cid = schan->chan.chan_id;
+       struct sirfsoc_dma_desc *sdesc = NULL;
+
+       /*
+        * lock has been held by functions calling this, so we don't hold
+        * lock again
+        */
+
+       sdesc = list_first_entry(&schan->queued, struct sirfsoc_dma_desc,
+               node);
+       /* Move the first queued descriptor to active list */
+       list_move_tail(&schan->queued, &schan->active);
+
+       /* Start the DMA transfer */
+       writel_relaxed(sdesc->width, sdma->base + SIRFSOC_DMA_WIDTH_0 +
+               cid * 4);
+       writel_relaxed(cid | (schan->mode << SIRFSOC_DMA_MODE_CTRL_BIT) |
+               (sdesc->dir << SIRFSOC_DMA_DIR_CTRL_BIT),
+               sdma->base + cid * 0x10 + SIRFSOC_DMA_CH_CTRL);
+       writel_relaxed(sdesc->xlen, sdma->base + cid * 0x10 +
+               SIRFSOC_DMA_CH_XLEN);
+       writel_relaxed(sdesc->ylen, sdma->base + cid * 0x10 +
+               SIRFSOC_DMA_CH_YLEN);
+       writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_INT_EN) |
+               (1 << cid), sdma->base + SIRFSOC_DMA_INT_EN);
+
+       /*
+        * writel has an implict memory write barrier to make sure data is
+        * flushed into memory before starting DMA
+        */
+       writel(sdesc->addr >> 2, sdma->base + cid * 0x10 + SIRFSOC_DMA_CH_ADDR);
+
+       if (sdesc->cyclic) {
+               writel((1 << cid) | 1 << (cid + 16) |
+                       readl_relaxed(sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL),
+                       sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+               schan->happened_cyclic = schan->completed_cyclic = 0;
+       }
+}
+
+/* Interrupt handler */
+static irqreturn_t sirfsoc_dma_irq(int irq, void *data)
+{
+       struct sirfsoc_dma *sdma = data;
+       struct sirfsoc_dma_chan *schan;
+       struct sirfsoc_dma_desc *sdesc = NULL;
+       u32 is;
+       int ch;
+
+       is = readl(sdma->base + SIRFSOC_DMA_CH_INT);
+       while ((ch = fls(is) - 1) >= 0) {
+               is &= ~(1 << ch);
+               writel_relaxed(1 << ch, sdma->base + SIRFSOC_DMA_CH_INT);
+               schan = &sdma->channels[ch];
+
+               spin_lock(&schan->lock);
+
+               sdesc = list_first_entry(&schan->active, struct sirfsoc_dma_desc,
+                       node);
+               if (!sdesc->cyclic) {
+                       /* Execute queued descriptors */
+                       list_splice_tail_init(&schan->active, &schan->completed);
+                       if (!list_empty(&schan->queued))
+                               sirfsoc_dma_execute(schan);
+               } else
+                       schan->happened_cyclic++;
+
+               spin_unlock(&schan->lock);
+       }
+
+       /* Schedule tasklet */
+       tasklet_schedule(&sdma->tasklet);
+
+       return IRQ_HANDLED;
+}
+
+/* process completed descriptors */
+static void sirfsoc_dma_process_completed(struct sirfsoc_dma *sdma)
+{
+       dma_cookie_t last_cookie = 0;
+       struct sirfsoc_dma_chan *schan;
+       struct sirfsoc_dma_desc *sdesc;
+       struct dma_async_tx_descriptor *desc;
+       unsigned long flags;
+       unsigned long happened_cyclic;
+       LIST_HEAD(list);
+       int i;
+
+       for (i = 0; i < sdma->dma.chancnt; i++) {
+               schan = &sdma->channels[i];
+
+               /* Get all completed descriptors */
+               spin_lock_irqsave(&schan->lock, flags);
+               if (!list_empty(&schan->completed)) {
+                       list_splice_tail_init(&schan->completed, &list);
+                       spin_unlock_irqrestore(&schan->lock, flags);
+
+                       /* Execute callbacks and run dependencies */
+                       list_for_each_entry(sdesc, &list, node) {
+                               desc = &sdesc->desc;
+
+                               if (desc->callback)
+                                       desc->callback(desc->callback_param);
+
+                               last_cookie = desc->cookie;
+                               dma_run_dependencies(desc);
+                       }
+
+                       /* Free descriptors */
+                       spin_lock_irqsave(&schan->lock, flags);
+                       list_splice_tail_init(&list, &schan->free);
+                       schan->completed_cookie = last_cookie;
+                       spin_unlock_irqrestore(&schan->lock, flags);
+               } else {
+                       /* for cyclic channel, desc is always in active list */
+                       sdesc = list_first_entry(&schan->active, struct sirfsoc_dma_desc,
+                               node);
+
+                       if (!sdesc || (sdesc && !sdesc->cyclic)) {
+                               /* without active cyclic DMA */
+                               spin_unlock_irqrestore(&schan->lock, flags);
+                               continue;
+                       }
+
+                       /* cyclic DMA */
+                       happened_cyclic = schan->happened_cyclic;
+                       spin_unlock_irqrestore(&schan->lock, flags);
+
+                       desc = &sdesc->desc;
+                       while (happened_cyclic != schan->completed_cyclic) {
+                               if (desc->callback)
+                                       desc->callback(desc->callback_param);
+                               schan->completed_cyclic++;
+                       }
+               }
+       }
+}
+
+/* DMA Tasklet */
+static void sirfsoc_dma_tasklet(unsigned long data)
+{
+       struct sirfsoc_dma *sdma = (void *)data;
+
+       sirfsoc_dma_process_completed(sdma);
+}
+
+/* Submit descriptor to hardware */
+static dma_cookie_t sirfsoc_dma_tx_submit(struct dma_async_tx_descriptor *txd)
+{
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(txd->chan);
+       struct sirfsoc_dma_desc *sdesc;
+       unsigned long flags;
+       dma_cookie_t cookie;
+
+       sdesc = container_of(txd, struct sirfsoc_dma_desc, desc);
+
+       spin_lock_irqsave(&schan->lock, flags);
+
+       /* Move descriptor to queue */
+       list_move_tail(&sdesc->node, &schan->queued);
+
+       /* Update cookie */
+       cookie = schan->chan.cookie + 1;
+       if (cookie <= 0)
+               cookie = 1;
+
+       schan->chan.cookie = cookie;
+       sdesc->desc.cookie = cookie;
+
+       spin_unlock_irqrestore(&schan->lock, flags);
+
+       return cookie;
+}
+
+static int sirfsoc_dma_slave_config(struct sirfsoc_dma_chan *schan,
+       struct dma_slave_config *config)
+{
+       unsigned long flags;
+
+       if ((config->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES) ||
+               (config->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES))
+               return -EINVAL;
+
+       spin_lock_irqsave(&schan->lock, flags);
+       schan->mode = (config->src_maxburst == 4 ? 1 : 0);
+       spin_unlock_irqrestore(&schan->lock, flags);
+
+       return 0;
+}
+
+static int sirfsoc_dma_terminate_all(struct sirfsoc_dma_chan *schan)
+{
+       struct sirfsoc_dma *sdma = dma_chan_to_sirfsoc_dma(&schan->chan);
+       int cid = schan->chan.chan_id;
+       unsigned long flags;
+
+       writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_INT_EN) &
+               ~(1 << cid), sdma->base + SIRFSOC_DMA_INT_EN);
+       writel_relaxed(1 << cid, sdma->base + SIRFSOC_DMA_CH_VALID);
+
+       writel_relaxed(readl_relaxed(sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL)
+               & ~((1 << cid) | 1 << (cid + 16)),
+                       sdma->base + SIRFSOC_DMA_CH_LOOP_CTRL);
+
+       spin_lock_irqsave(&schan->lock, flags);
+       list_splice_tail_init(&schan->active, &schan->free);
+       list_splice_tail_init(&schan->queued, &schan->free);
+       spin_unlock_irqrestore(&schan->lock, flags);
+
+       return 0;
+}
+
+static int sirfsoc_dma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
+       unsigned long arg)
+{
+       struct dma_slave_config *config;
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+
+       switch (cmd) {
+       case DMA_TERMINATE_ALL:
+               return sirfsoc_dma_terminate_all(schan);
+       case DMA_SLAVE_CONFIG:
+               config = (struct dma_slave_config *)arg;
+               return sirfsoc_dma_slave_config(schan, config);
+
+       default:
+               break;
+       }
+
+       return -ENOSYS;
+}
+
+/* Alloc channel resources */
+static int sirfsoc_dma_alloc_chan_resources(struct dma_chan *chan)
+{
+       struct sirfsoc_dma *sdma = dma_chan_to_sirfsoc_dma(chan);
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+       struct sirfsoc_dma_desc *sdesc;
+       unsigned long flags;
+       LIST_HEAD(descs);
+       int i;
+
+       /* Alloc descriptors for this channel */
+       for (i = 0; i < SIRFSOC_DMA_DESCRIPTORS; i++) {
+               sdesc = kzalloc(sizeof(*sdesc), GFP_KERNEL);
+               if (!sdesc) {
+                       dev_notice(sdma->dma.dev, "Memory allocation error. "
+                               "Allocated only %u descriptors\n", i);
+                       break;
+               }
+
+               dma_async_tx_descriptor_init(&sdesc->desc, chan);
+               sdesc->desc.flags = DMA_CTRL_ACK;
+               sdesc->desc.tx_submit = sirfsoc_dma_tx_submit;
+
+               list_add_tail(&sdesc->node, &descs);
+       }
+
+       /* Return error only if no descriptors were allocated */
+       if (i == 0)
+               return -ENOMEM;
+
+       spin_lock_irqsave(&schan->lock, flags);
+
+       list_splice_tail_init(&descs, &schan->free);
+       spin_unlock_irqrestore(&schan->lock, flags);
+
+       return i;
+}
+
+/* Free channel resources */
+static void sirfsoc_dma_free_chan_resources(struct dma_chan *chan)
+{
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+       struct sirfsoc_dma_desc *sdesc, *tmp;
+       unsigned long flags;
+       LIST_HEAD(descs);
+
+       spin_lock_irqsave(&schan->lock, flags);
+
+       /* Channel must be idle */
+       BUG_ON(!list_empty(&schan->prepared));
+       BUG_ON(!list_empty(&schan->queued));
+       BUG_ON(!list_empty(&schan->active));
+       BUG_ON(!list_empty(&schan->completed));
+
+       /* Move data */
+       list_splice_tail_init(&schan->free, &descs);
+
+       spin_unlock_irqrestore(&schan->lock, flags);
+
+       /* Free descriptors */
+       list_for_each_entry_safe(sdesc, tmp, &descs, node)
+               kfree(sdesc);
+}
+
+/* Send pending descriptor to hardware */
+static void sirfsoc_dma_issue_pending(struct dma_chan *chan)
+{
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+       unsigned long flags;
+
+       spin_lock_irqsave(&schan->lock, flags);
+
+       if (list_empty(&schan->active) && !list_empty(&schan->queued))
+               sirfsoc_dma_execute(schan);
+
+       spin_unlock_irqrestore(&schan->lock, flags);
+}
+
+/* Check request completion status */
+static enum dma_status
+sirfsoc_dma_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
+       struct dma_tx_state *txstate)
+{
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+       unsigned long flags;
+       dma_cookie_t last_used;
+       dma_cookie_t last_complete;
+
+       spin_lock_irqsave(&schan->lock, flags);
+       last_used = schan->chan.cookie;
+       last_complete = schan->completed_cookie;
+       spin_unlock_irqrestore(&schan->lock, flags);
+
+       dma_set_tx_state(txstate, last_complete, last_used, 0);
+       return dma_async_is_complete(cookie, last_complete, last_used);
+}
+
+static struct dma_async_tx_descriptor *sirfsoc_dma_prep_interleaved(
+       struct dma_chan *chan, struct dma_interleaved_template *xt,
+       unsigned long flags)
+{
+       struct sirfsoc_dma *sdma = dma_chan_to_sirfsoc_dma(chan);
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+       struct sirfsoc_dma_desc *sdesc = NULL;
+       unsigned long iflags;
+       int ret;
+
+       if ((xt->dir != DMA_MEM_TO_DEV) || (xt->dir != DMA_DEV_TO_MEM)) {
+               ret = -EINVAL;
+               goto err_dir;
+       }
+
+       /* Get free descriptor */
+       spin_lock_irqsave(&schan->lock, iflags);
+       if (!list_empty(&schan->free)) {
+               sdesc = list_first_entry(&schan->free, struct sirfsoc_dma_desc,
+                       node);
+               list_del(&sdesc->node);
+       }
+       spin_unlock_irqrestore(&schan->lock, iflags);
+
+       if (!sdesc) {
+               /* try to free completed descriptors */
+               sirfsoc_dma_process_completed(sdma);
+               ret = 0;
+               goto no_desc;
+       }
+
+       /* Place descriptor in prepared list */
+       spin_lock_irqsave(&schan->lock, iflags);
+
+       /*
+        * Number of chunks in a frame can only be 1 for prima2
+        * and ylen (number of frame - 1) must be at least 0
+        */
+       if ((xt->frame_size == 1) && (xt->numf > 0)) {
+               sdesc->cyclic = 0;
+               sdesc->xlen = xt->sgl[0].size / SIRFSOC_DMA_WORD_LEN;
+               sdesc->width = (xt->sgl[0].size + xt->sgl[0].icg) /
+                               SIRFSOC_DMA_WORD_LEN;
+               sdesc->ylen = xt->numf - 1;
+               if (xt->dir == DMA_MEM_TO_DEV) {
+                       sdesc->addr = xt->src_start;
+                       sdesc->dir = 1;
+               } else {
+                       sdesc->addr = xt->dst_start;
+                       sdesc->dir = 0;
+               }
+
+               list_add_tail(&sdesc->node, &schan->prepared);
+       } else {
+               pr_err("sirfsoc DMA Invalid xfer\n");
+               ret = -EINVAL;
+               goto err_xfer;
+       }
+       spin_unlock_irqrestore(&schan->lock, iflags);
+
+       return &sdesc->desc;
+err_xfer:
+       spin_unlock_irqrestore(&schan->lock, iflags);
+no_desc:
+err_dir:
+       return ERR_PTR(ret);
+}
+
+static struct dma_async_tx_descriptor *
+sirfsoc_dma_prep_cyclic(struct dma_chan *chan, dma_addr_t addr,
+       size_t buf_len, size_t period_len,
+       enum dma_transfer_direction direction)
+{
+       struct sirfsoc_dma_chan *schan = dma_chan_to_sirfsoc_dma_chan(chan);
+       struct sirfsoc_dma_desc *sdesc = NULL;
+       unsigned long iflags;
+
+       /*
+        * we only support cycle transfer with 2 period
+        * If the X-length is set to 0, it would be the loop mode.
+        * The DMA address keeps increasing until reaching the end of a loop
+        * area whose size is defined by (DMA_WIDTH x (Y_LENGTH + 1)). Then
+        * the DMA address goes back to the beginning of this area.
+        * In loop mode, the DMA data region is divided into two parts, BUFA
+        * and BUFB. DMA controller generates interrupts twice in each loop:
+        * when the DMA address reaches the end of BUFA or the end of the
+        * BUFB
+        */
+       if (buf_len !=  2 * period_len)
+               return ERR_PTR(-EINVAL);
+
+       /* Get free descriptor */
+       spin_lock_irqsave(&schan->lock, iflags);
+       if (!list_empty(&schan->free)) {
+               sdesc = list_first_entry(&schan->free, struct sirfsoc_dma_desc,
+                       node);
+               list_del(&sdesc->node);
+       }
+       spin_unlock_irqrestore(&schan->lock, iflags);
+
+       if (!sdesc)
+               return 0;
+
+       /* Place descriptor in prepared list */
+       spin_lock_irqsave(&schan->lock, iflags);
+       sdesc->addr = addr;
+       sdesc->cyclic = 1;
+       sdesc->xlen = 0;
+       sdesc->ylen = buf_len / SIRFSOC_DMA_WORD_LEN - 1;
+       sdesc->width = 1;
+       list_add_tail(&sdesc->node, &schan->prepared);
+       spin_unlock_irqrestore(&schan->lock, iflags);
+
+       return &sdesc->desc;
+}
+
+/*
+ * The DMA controller consists of 16 independent DMA channels.
+ * Each channel is allocated to a different function
+ */
+bool sirfsoc_dma_filter_id(struct dma_chan *chan, void *chan_id)
+{
+       unsigned int ch_nr = (unsigned int) chan_id;
+
+       if (ch_nr == chan->chan_id +
+               chan->device->dev_id * SIRFSOC_DMA_CHANNELS)
+               return true;
+
+       return false;
+}
+EXPORT_SYMBOL(sirfsoc_dma_filter_id);
+
+static int __devinit sirfsoc_dma_probe(struct platform_device *op)
+{
+       struct device_node *dn = op->dev.of_node;
+       struct device *dev = &op->dev;
+       struct dma_device *dma;
+       struct sirfsoc_dma *sdma;
+       struct sirfsoc_dma_chan *schan;
+       struct resource res;
+       ulong regs_start, regs_size;
+       u32 id;
+       int ret, i;
+
+       sdma = devm_kzalloc(dev, sizeof(*sdma), GFP_KERNEL);
+       if (!sdma) {
+               dev_err(dev, "Memory exhausted!\n");
+               return -ENOMEM;
+       }
+
+       if (of_property_read_u32(dn, "cell-index", &id)) {
+               dev_err(dev, "Fail to get DMAC index\n");
+               ret = -ENODEV;
+               goto free_mem;
+       }
+
+       sdma->irq = irq_of_parse_and_map(dn, 0);
+       if (sdma->irq == NO_IRQ) {
+               dev_err(dev, "Error mapping IRQ!\n");
+               ret = -EINVAL;
+               goto free_mem;
+       }
+
+       ret = of_address_to_resource(dn, 0, &res);
+       if (ret) {
+               dev_err(dev, "Error parsing memory region!\n");
+               goto free_mem;
+       }
+
+       regs_start = res.start;
+       regs_size = resource_size(&res);
+
+       sdma->base = devm_ioremap(dev, regs_start, regs_size);
+       if (!sdma->base) {
+               dev_err(dev, "Error mapping memory region!\n");
+               ret = -ENOMEM;
+               goto irq_dispose;
+       }
+
+       ret = devm_request_irq(dev, sdma->irq, &sirfsoc_dma_irq, 0, DRV_NAME,
+               sdma);
+       if (ret) {
+               dev_err(dev, "Error requesting IRQ!\n");
+               ret = -EINVAL;
+               goto unmap_mem;
+       }
+
+       dma = &sdma->dma;
+       dma->dev = dev;
+       dma->chancnt = SIRFSOC_DMA_CHANNELS;
+
+       dma->device_alloc_chan_resources = sirfsoc_dma_alloc_chan_resources;
+       dma->device_free_chan_resources = sirfsoc_dma_free_chan_resources;
+       dma->device_issue_pending = sirfsoc_dma_issue_pending;
+       dma->device_control = sirfsoc_dma_control;
+       dma->device_tx_status = sirfsoc_dma_tx_status;
+       dma->device_prep_interleaved_dma = sirfsoc_dma_prep_interleaved;
+       dma->device_prep_dma_cyclic = sirfsoc_dma_prep_cyclic;
+
+       INIT_LIST_HEAD(&dma->channels);
+       dma_cap_set(DMA_SLAVE, dma->cap_mask);
+       dma_cap_set(DMA_CYCLIC, dma->cap_mask);
+       dma_cap_set(DMA_INTERLEAVE, dma->cap_mask);
+       dma_cap_set(DMA_PRIVATE, dma->cap_mask);
+
+       for (i = 0; i < dma->chancnt; i++) {
+               schan = &sdma->channels[i];
+
+               schan->chan.device = dma;
+               schan->chan.cookie = 1;
+               schan->completed_cookie = schan->chan.cookie;
+
+               INIT_LIST_HEAD(&schan->free);
+               INIT_LIST_HEAD(&schan->prepared);
+               INIT_LIST_HEAD(&schan->queued);
+               INIT_LIST_HEAD(&schan->active);
+               INIT_LIST_HEAD(&schan->completed);
+
+               spin_lock_init(&schan->lock);
+               list_add_tail(&schan->chan.device_node, &dma->channels);
+       }
+
+       tasklet_init(&sdma->tasklet, sirfsoc_dma_tasklet, (unsigned long)sdma);
+
+       /* Register DMA engine */
+       dev_set_drvdata(dev, sdma);
+       ret = dma_async_device_register(dma);
+       if (ret)
+               goto free_irq;
+
+       dev_info(dev, "initialized SIRFSOC DMAC driver\n");
+
+       return 0;
+
+free_irq:
+       devm_free_irq(dev, sdma->irq, sdma);
+irq_dispose:
+       irq_dispose_mapping(sdma->irq);
+unmap_mem:
+       iounmap(sdma->base);
+free_mem:
+       devm_kfree(dev, sdma);
+       return ret;
+}
+
+static int __devexit sirfsoc_dma_remove(struct platform_device *op)
+{
+       struct device *dev = &op->dev;
+       struct sirfsoc_dma *sdma = dev_get_drvdata(dev);
+
+       dma_async_device_unregister(&sdma->dma);
+       devm_free_irq(dev, sdma->irq, sdma);
+       irq_dispose_mapping(sdma->irq);
+       iounmap(sdma->base);
+       devm_kfree(dev, sdma);
+       return 0;
+}
+
+static struct of_device_id sirfsoc_dma_match[] = {
+       { .compatible = "sirf,prima2-dmac", },
+       {},
+};
+
+static struct platform_driver sirfsoc_dma_driver = {
+       .probe          = sirfsoc_dma_probe,
+       .remove         = __devexit_p(sirfsoc_dma_remove),
+       .driver = {
+               .name = DRV_NAME,
+               .owner = THIS_MODULE,
+               .of_match_table = sirfsoc_dma_match,
+       },
+};
+
+static int __init sirfsoc_dma_init(void)
+{
+       return platform_driver_register(&sirfsoc_dma_driver);
+}
+module_init(sirfsoc_dma_init);
+
+static void __exit sirfsoc_dma_exit(void)
+{
+       platform_driver_unregister(&sirfsoc_dma_driver);
+}
+module_exit(sirfsoc_dma_exit);
+
+MODULE_AUTHOR("Rongjun Ying <rongjun.ying@csr.com>, "
+       "Barry Song <baohua.song@csr.com>");
+MODULE_DESCRIPTION("SIRFSOC DMA control driver");
+MODULE_LICENSE("GPL v2");
diff --git a/include/linux/sirfsoc_dma.h b/include/linux/sirfsoc_dma.h
new file mode 100644 (file)
index 0000000..29d9593
--- /dev/null
@@ -0,0 +1,6 @@
+#ifndef _SIRFSOC_DMA_H_
+#define _SIRFSOC_DMA_H_
+
+bool sirfsoc_dma_filter_id(struct dma_chan *chan, void *chan_id);
+
+#endif