dmaengine: Add hisilicon k3 DMA engine driver
authorZhangfei Gao <zhangfei.gao@linaro.org>
Tue, 27 Aug 2013 02:20:10 +0000 (10:20 +0800)
committerVinod Koul <vinod.koul@intel.com>
Wed, 28 Aug 2013 05:53:40 +0000 (11:23 +0530)
Add dmaengine driver for hisilicon k3 platform based on virt_dma

Signed-off-by: Zhangfei Gao <zhangfei.gao@linaro.org>
Tested-by: Kai Yang <jean.yangkai@huawei.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Vinod Koul <vinod.koul@intel.com>
Documentation/devicetree/bindings/dma/k3dma.txt [new file with mode: 0644]
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/k3dma.c [new file with mode: 0644]

diff --git a/Documentation/devicetree/bindings/dma/k3dma.txt b/Documentation/devicetree/bindings/dma/k3dma.txt
new file mode 100644 (file)
index 0000000..23f8d71
--- /dev/null
@@ -0,0 +1,46 @@
+* Hisilicon K3 DMA controller
+
+See dma.txt first
+
+Required properties:
+- compatible: Should be "hisilicon,k3-dma-1.0"
+- reg: Should contain DMA registers location and length.
+- interrupts: Should contain one interrupt shared by all channel
+- #dma-cells: see dma.txt, should be 1, para number
+- dma-channels: physical channels supported
+- dma-requests: virtual channels supported, each virtual channel
+               have specific request line
+- clocks: clock required
+
+Example:
+
+Controller:
+               dma0: dma@fcd02000 {
+                       compatible = "hisilicon,k3-dma-1.0";
+                       reg = <0xfcd02000 0x1000>;
+                       #dma-cells = <1>;
+                       dma-channels = <16>;
+                       dma-requests = <27>;
+                       interrupts = <0 12 4>;
+                       clocks = <&pclk>;
+                       status = "disable";
+               };
+
+Client:
+Use specific request line passing from dmax
+For example, i2c0 read channel request line is 18, while write channel use 19
+
+               i2c0: i2c@fcb08000 {
+                       compatible = "snps,designware-i2c";
+                       dmas =  <&dma0 18          /* read channel */
+                                &dma0 19>;        /* write channel */
+                       dma-names = "rx", "tx";
+               };
+
+               i2c1: i2c@fcb09000 {
+                       compatible = "snps,designware-i2c";
+                       dmas =  <&dma0 20          /* read channel */
+                                &dma0 21>;        /* write channel */
+                       dma-names = "rx", "tx";
+               };
+
index 6825957c97fb9896dbfa944e6c5520928160a74c..2945ff084a048d849d3da75fc36008425ddcd361 100644 (file)
@@ -300,6 +300,15 @@ config DMA_JZ4740
        select DMA_ENGINE
        select DMA_VIRTUAL_CHANNELS
 
+config K3_DMA
+       tristate "Hisilicon K3 DMA support"
+       depends on ARCH_HI3xxx
+       select DMA_ENGINE
+       select DMA_VIRTUAL_CHANNELS
+       help
+         Support the DMA engine for Hisilicon K3 platform
+         devices.
+
 config DMA_ENGINE
        bool
 
index 5e0f2ef8561485f9a75d7f7362e06c99fffdbcac..8c97941df2738f1d1f6973a747507d7dffbbc1db 100644 (file)
@@ -39,3 +39,4 @@ obj-$(CONFIG_MMP_TDMA) += mmp_tdma.o
 obj-$(CONFIG_DMA_OMAP) += omap-dma.o
 obj-$(CONFIG_MMP_PDMA) += mmp_pdma.o
 obj-$(CONFIG_DMA_JZ4740) += dma-jz4740.o
+obj-$(CONFIG_K3_DMA) += k3dma.o
diff --git a/drivers/dma/k3dma.c b/drivers/dma/k3dma.c
new file mode 100644 (file)
index 0000000..ef7bc85
--- /dev/null
@@ -0,0 +1,837 @@
+/*
+ * Copyright (c) 2013 Linaro Ltd.
+ * Copyright (c) 2013 Hisilicon Limited.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+#include <linux/sched.h>
+#include <linux/device.h>
+#include <linux/dmaengine.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+#include <linux/of_device.h>
+#include <linux/of.h>
+#include <linux/clk.h>
+#include <linux/of_dma.h>
+
+#include "virt-dma.h"
+
+#define DRIVER_NAME            "k3-dma"
+#define DMA_ALIGN              3
+#define DMA_MAX_SIZE           0x1ffc
+
+#define INT_STAT               0x00
+#define INT_TC1                        0x04
+#define INT_ERR1               0x0c
+#define INT_ERR2               0x10
+#define INT_TC1_MASK           0x18
+#define INT_ERR1_MASK          0x20
+#define INT_ERR2_MASK          0x24
+#define INT_TC1_RAW            0x600
+#define INT_ERR1_RAW           0x608
+#define INT_ERR2_RAW           0x610
+#define CH_PRI                 0x688
+#define CH_STAT                        0x690
+#define CX_CUR_CNT             0x704
+#define CX_LLI                 0x800
+#define CX_CNT                 0x810
+#define CX_SRC                 0x814
+#define CX_DST                 0x818
+#define CX_CFG                 0x81c
+#define AXI_CFG                        0x820
+#define AXI_CFG_DEFAULT                0x201201
+
+#define CX_LLI_CHAIN_EN                0x2
+#define CX_CFG_EN              0x1
+#define CX_CFG_MEM2PER         (0x1 << 2)
+#define CX_CFG_PER2MEM         (0x2 << 2)
+#define CX_CFG_SRCINCR         (0x1 << 31)
+#define CX_CFG_DSTINCR         (0x1 << 30)
+
+struct k3_desc_hw {
+       u32 lli;
+       u32 reserved[3];
+       u32 count;
+       u32 saddr;
+       u32 daddr;
+       u32 config;
+} __aligned(32);
+
+struct k3_dma_desc_sw {
+       struct virt_dma_desc    vd;
+       dma_addr_t              desc_hw_lli;
+       size_t                  desc_num;
+       size_t                  size;
+       struct k3_desc_hw       desc_hw[0];
+};
+
+struct k3_dma_phy;
+
+struct k3_dma_chan {
+       u32                     ccfg;
+       struct virt_dma_chan    vc;
+       struct k3_dma_phy       *phy;
+       struct list_head        node;
+       enum dma_transfer_direction dir;
+       dma_addr_t              dev_addr;
+       enum dma_status         status;
+};
+
+struct k3_dma_phy {
+       u32                     idx;
+       void __iomem            *base;
+       struct k3_dma_chan      *vchan;
+       struct k3_dma_desc_sw   *ds_run;
+       struct k3_dma_desc_sw   *ds_done;
+};
+
+struct k3_dma_dev {
+       struct dma_device       slave;
+       void __iomem            *base;
+       struct tasklet_struct   task;
+       spinlock_t              lock;
+       struct list_head        chan_pending;
+       struct k3_dma_phy       *phy;
+       struct k3_dma_chan      *chans;
+       struct clk              *clk;
+       u32                     dma_channels;
+       u32                     dma_requests;
+};
+
+#define to_k3_dma(dmadev) container_of(dmadev, struct k3_dma_dev, slave)
+
+static struct k3_dma_chan *to_k3_chan(struct dma_chan *chan)
+{
+       return container_of(chan, struct k3_dma_chan, vc.chan);
+}
+
+static void k3_dma_pause_dma(struct k3_dma_phy *phy, bool on)
+{
+       u32 val = 0;
+
+       if (on) {
+               val = readl_relaxed(phy->base + CX_CFG);
+               val |= CX_CFG_EN;
+               writel_relaxed(val, phy->base + CX_CFG);
+       } else {
+               val = readl_relaxed(phy->base + CX_CFG);
+               val &= ~CX_CFG_EN;
+               writel_relaxed(val, phy->base + CX_CFG);
+       }
+}
+
+static void k3_dma_terminate_chan(struct k3_dma_phy *phy, struct k3_dma_dev *d)
+{
+       u32 val = 0;
+
+       k3_dma_pause_dma(phy, false);
+
+       val = 0x1 << phy->idx;
+       writel_relaxed(val, d->base + INT_TC1_RAW);
+       writel_relaxed(val, d->base + INT_ERR1_RAW);
+       writel_relaxed(val, d->base + INT_ERR2_RAW);
+}
+
+static void k3_dma_set_desc(struct k3_dma_phy *phy, struct k3_desc_hw *hw)
+{
+       writel_relaxed(hw->lli, phy->base + CX_LLI);
+       writel_relaxed(hw->count, phy->base + CX_CNT);
+       writel_relaxed(hw->saddr, phy->base + CX_SRC);
+       writel_relaxed(hw->daddr, phy->base + CX_DST);
+       writel_relaxed(AXI_CFG_DEFAULT, phy->base + AXI_CFG);
+       writel_relaxed(hw->config, phy->base + CX_CFG);
+}
+
+static u32 k3_dma_get_curr_cnt(struct k3_dma_dev *d, struct k3_dma_phy *phy)
+{
+       u32 cnt = 0;
+
+       cnt = readl_relaxed(d->base + CX_CUR_CNT + phy->idx * 0x10);
+       cnt &= 0xffff;
+       return cnt;
+}
+
+static u32 k3_dma_get_curr_lli(struct k3_dma_phy *phy)
+{
+       return readl_relaxed(phy->base + CX_LLI);
+}
+
+static u32 k3_dma_get_chan_stat(struct k3_dma_dev *d)
+{
+       return readl_relaxed(d->base + CH_STAT);
+}
+
+static void k3_dma_enable_dma(struct k3_dma_dev *d, bool on)
+{
+       if (on) {
+               /* set same priority */
+               writel_relaxed(0x0, d->base + CH_PRI);
+
+               /* unmask irq */
+               writel_relaxed(0xffff, d->base + INT_TC1_MASK);
+               writel_relaxed(0xffff, d->base + INT_ERR1_MASK);
+               writel_relaxed(0xffff, d->base + INT_ERR2_MASK);
+       } else {
+               /* mask irq */
+               writel_relaxed(0x0, d->base + INT_TC1_MASK);
+               writel_relaxed(0x0, d->base + INT_ERR1_MASK);
+               writel_relaxed(0x0, d->base + INT_ERR2_MASK);
+       }
+}
+
+static irqreturn_t k3_dma_int_handler(int irq, void *dev_id)
+{
+       struct k3_dma_dev *d = (struct k3_dma_dev *)dev_id;
+       struct k3_dma_phy *p;
+       struct k3_dma_chan *c;
+       u32 stat = readl_relaxed(d->base + INT_STAT);
+       u32 tc1  = readl_relaxed(d->base + INT_TC1);
+       u32 err1 = readl_relaxed(d->base + INT_ERR1);
+       u32 err2 = readl_relaxed(d->base + INT_ERR2);
+       u32 i, irq_chan = 0;
+
+       while (stat) {
+               i = __ffs(stat);
+               stat &= (stat - 1);
+               if (likely(tc1 & BIT(i))) {
+                       p = &d->phy[i];
+                       c = p->vchan;
+                       if (c) {
+                               unsigned long flags;
+
+                               spin_lock_irqsave(&c->vc.lock, flags);
+                               vchan_cookie_complete(&p->ds_run->vd);
+                               p->ds_done = p->ds_run;
+                               spin_unlock_irqrestore(&c->vc.lock, flags);
+                       }
+                       irq_chan |= BIT(i);
+               }
+               if (unlikely((err1 & BIT(i)) || (err2 & BIT(i))))
+                       dev_warn(d->slave.dev, "DMA ERR\n");
+       }
+
+       writel_relaxed(irq_chan, d->base + INT_TC1_RAW);
+       writel_relaxed(err1, d->base + INT_ERR1_RAW);
+       writel_relaxed(err2, d->base + INT_ERR2_RAW);
+
+       if (irq_chan) {
+               tasklet_schedule(&d->task);
+               return IRQ_HANDLED;
+       } else
+               return IRQ_NONE;
+}
+
+static int k3_dma_start_txd(struct k3_dma_chan *c)
+{
+       struct k3_dma_dev *d = to_k3_dma(c->vc.chan.device);
+       struct virt_dma_desc *vd = vchan_next_desc(&c->vc);
+
+       if (!c->phy)
+               return -EAGAIN;
+
+       if (BIT(c->phy->idx) & k3_dma_get_chan_stat(d))
+               return -EAGAIN;
+
+       if (vd) {
+               struct k3_dma_desc_sw *ds =
+                       container_of(vd, struct k3_dma_desc_sw, vd);
+               /*
+                * fetch and remove request from vc->desc_issued
+                * so vc->desc_issued only contains desc pending
+                */
+               list_del(&ds->vd.node);
+               c->phy->ds_run = ds;
+               c->phy->ds_done = NULL;
+               /* start dma */
+               k3_dma_set_desc(c->phy, &ds->desc_hw[0]);
+               return 0;
+       }
+       c->phy->ds_done = NULL;
+       c->phy->ds_run = NULL;
+       return -EAGAIN;
+}
+
+static void k3_dma_tasklet(unsigned long arg)
+{
+       struct k3_dma_dev *d = (struct k3_dma_dev *)arg;
+       struct k3_dma_phy *p;
+       struct k3_dma_chan *c, *cn;
+       unsigned pch, pch_alloc = 0;
+
+       /* check new dma request of running channel in vc->desc_issued */
+       list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
+               spin_lock_irq(&c->vc.lock);
+               p = c->phy;
+               if (p && p->ds_done) {
+                       if (k3_dma_start_txd(c)) {
+                               /* No current txd associated with this channel */
+                               dev_dbg(d->slave.dev, "pchan %u: free\n", p->idx);
+                               /* Mark this channel free */
+                               c->phy = NULL;
+                               p->vchan = NULL;
+                       }
+               }
+               spin_unlock_irq(&c->vc.lock);
+       }
+
+       /* check new channel request in d->chan_pending */
+       spin_lock_irq(&d->lock);
+       for (pch = 0; pch < d->dma_channels; pch++) {
+               p = &d->phy[pch];
+
+               if (p->vchan == NULL && !list_empty(&d->chan_pending)) {
+                       c = list_first_entry(&d->chan_pending,
+                               struct k3_dma_chan, node);
+                       /* remove from d->chan_pending */
+                       list_del_init(&c->node);
+                       pch_alloc |= 1 << pch;
+                       /* Mark this channel allocated */
+                       p->vchan = c;
+                       c->phy = p;
+                       dev_dbg(d->slave.dev, "pchan %u: alloc vchan %p\n", pch, &c->vc);
+               }
+       }
+       spin_unlock_irq(&d->lock);
+
+       for (pch = 0; pch < d->dma_channels; pch++) {
+               if (pch_alloc & (1 << pch)) {
+                       p = &d->phy[pch];
+                       c = p->vchan;
+                       if (c) {
+                               spin_lock_irq(&c->vc.lock);
+                               k3_dma_start_txd(c);
+                               spin_unlock_irq(&c->vc.lock);
+                       }
+               }
+       }
+}
+
+static int k3_dma_alloc_chan_resources(struct dma_chan *chan)
+{
+       return 0;
+}
+
+static void k3_dma_free_chan_resources(struct dma_chan *chan)
+{
+       struct k3_dma_chan *c = to_k3_chan(chan);
+       struct k3_dma_dev *d = to_k3_dma(chan->device);
+       unsigned long flags;
+
+       spin_lock_irqsave(&d->lock, flags);
+       list_del_init(&c->node);
+       spin_unlock_irqrestore(&d->lock, flags);
+
+       vchan_free_chan_resources(&c->vc);
+       c->ccfg = 0;
+}
+
+static enum dma_status k3_dma_tx_status(struct dma_chan *chan,
+       dma_cookie_t cookie, struct dma_tx_state *state)
+{
+       struct k3_dma_chan *c = to_k3_chan(chan);
+       struct k3_dma_dev *d = to_k3_dma(chan->device);
+       struct k3_dma_phy *p;
+       struct virt_dma_desc *vd;
+       unsigned long flags;
+       enum dma_status ret;
+       size_t bytes = 0;
+
+       ret = dma_cookie_status(&c->vc.chan, cookie, state);
+       if (ret == DMA_SUCCESS)
+               return ret;
+
+       spin_lock_irqsave(&c->vc.lock, flags);
+       p = c->phy;
+       ret = c->status;
+
+       /*
+        * If the cookie is on our issue queue, then the residue is
+        * its total size.
+        */
+       vd = vchan_find_desc(&c->vc, cookie);
+       if (vd) {
+               bytes = container_of(vd, struct k3_dma_desc_sw, vd)->size;
+       } else if ((!p) || (!p->ds_run)) {
+               bytes = 0;
+       } else {
+               struct k3_dma_desc_sw *ds = p->ds_run;
+               u32 clli = 0, index = 0;
+
+               bytes = k3_dma_get_curr_cnt(d, p);
+               clli = k3_dma_get_curr_lli(p);
+               index = (clli - ds->desc_hw_lli) / sizeof(struct k3_desc_hw);
+               for (; index < ds->desc_num; index++) {
+                       bytes += ds->desc_hw[index].count;
+                       /* end of lli */
+                       if (!ds->desc_hw[index].lli)
+                               break;
+               }
+       }
+       spin_unlock_irqrestore(&c->vc.lock, flags);
+       dma_set_residue(state, bytes);
+       return ret;
+}
+
+static void k3_dma_issue_pending(struct dma_chan *chan)
+{
+       struct k3_dma_chan *c = to_k3_chan(chan);
+       struct k3_dma_dev *d = to_k3_dma(chan->device);
+       unsigned long flags;
+
+       spin_lock_irqsave(&c->vc.lock, flags);
+       /* add request to vc->desc_issued */
+       if (vchan_issue_pending(&c->vc)) {
+               spin_lock(&d->lock);
+               if (!c->phy) {
+                       if (list_empty(&c->node)) {
+                               /* if new channel, add chan_pending */
+                               list_add_tail(&c->node, &d->chan_pending);
+                               /* check in tasklet */
+                               tasklet_schedule(&d->task);
+                               dev_dbg(d->slave.dev, "vchan %p: issued\n", &c->vc);
+                       }
+               }
+               spin_unlock(&d->lock);
+       } else
+               dev_dbg(d->slave.dev, "vchan %p: nothing to issue\n", &c->vc);
+       spin_unlock_irqrestore(&c->vc.lock, flags);
+}
+
+static void k3_dma_fill_desc(struct k3_dma_desc_sw *ds, dma_addr_t dst,
+                       dma_addr_t src, size_t len, u32 num, u32 ccfg)
+{
+       if ((num + 1) < ds->desc_num)
+               ds->desc_hw[num].lli = ds->desc_hw_lli + (num + 1) *
+                       sizeof(struct k3_desc_hw);
+       ds->desc_hw[num].lli |= CX_LLI_CHAIN_EN;
+       ds->desc_hw[num].count = len;
+       ds->desc_hw[num].saddr = src;
+       ds->desc_hw[num].daddr = dst;
+       ds->desc_hw[num].config = ccfg;
+}
+
+static struct dma_async_tx_descriptor *k3_dma_prep_memcpy(
+       struct dma_chan *chan,  dma_addr_t dst, dma_addr_t src,
+       size_t len, unsigned long flags)
+{
+       struct k3_dma_chan *c = to_k3_chan(chan);
+       struct k3_dma_desc_sw *ds;
+       size_t copy = 0;
+       int num = 0;
+
+       if (!len)
+               return NULL;
+
+       num = DIV_ROUND_UP(len, DMA_MAX_SIZE);
+       ds = kzalloc(sizeof(*ds) + num * sizeof(ds->desc_hw[0]), GFP_ATOMIC);
+       if (!ds) {
+               dev_dbg(chan->device->dev, "vchan %p: kzalloc fail\n", &c->vc);
+               return NULL;
+       }
+       ds->desc_hw_lli = __virt_to_phys((unsigned long)&ds->desc_hw[0]);
+       ds->size = len;
+       ds->desc_num = num;
+       num = 0;
+
+       if (!c->ccfg) {
+               /* default is memtomem, without calling device_control */
+               c->ccfg = CX_CFG_SRCINCR | CX_CFG_DSTINCR | CX_CFG_EN;
+               c->ccfg |= (0xf << 20) | (0xf << 24);   /* burst = 16 */
+               c->ccfg |= (0x3 << 12) | (0x3 << 16);   /* width = 64 bit */
+       }
+
+       do {
+               copy = min_t(size_t, len, DMA_MAX_SIZE);
+               k3_dma_fill_desc(ds, dst, src, copy, num++, c->ccfg);
+
+               if (c->dir == DMA_MEM_TO_DEV) {
+                       src += copy;
+               } else if (c->dir == DMA_DEV_TO_MEM) {
+                       dst += copy;
+               } else {
+                       src += copy;
+                       dst += copy;
+               }
+               len -= copy;
+       } while (len);
+
+       ds->desc_hw[num-1].lli = 0;     /* end of link */
+       return vchan_tx_prep(&c->vc, &ds->vd, flags);
+}
+
+static struct dma_async_tx_descriptor *k3_dma_prep_slave_sg(
+       struct dma_chan *chan, struct scatterlist *sgl, unsigned int sglen,
+       enum dma_transfer_direction dir, unsigned long flags, void *context)
+{
+       struct k3_dma_chan *c = to_k3_chan(chan);
+       struct k3_dma_desc_sw *ds;
+       size_t len, avail, total = 0;
+       struct scatterlist *sg;
+       dma_addr_t addr, src = 0, dst = 0;
+       int num = sglen, i;
+
+       if (sgl == 0)
+               return NULL;
+
+       for_each_sg(sgl, sg, sglen, i) {
+               avail = sg_dma_len(sg);
+               if (avail > DMA_MAX_SIZE)
+                       num += DIV_ROUND_UP(avail, DMA_MAX_SIZE) - 1;
+       }
+
+       ds = kzalloc(sizeof(*ds) + num * sizeof(ds->desc_hw[0]), GFP_ATOMIC);
+       if (!ds) {
+               dev_dbg(chan->device->dev, "vchan %p: kzalloc fail\n", &c->vc);
+               return NULL;
+       }
+       ds->desc_hw_lli = __virt_to_phys((unsigned long)&ds->desc_hw[0]);
+       ds->desc_num = num;
+       num = 0;
+
+       for_each_sg(sgl, sg, sglen, i) {
+               addr = sg_dma_address(sg);
+               avail = sg_dma_len(sg);
+               total += avail;
+
+               do {
+                       len = min_t(size_t, avail, DMA_MAX_SIZE);
+
+                       if (dir == DMA_MEM_TO_DEV) {
+                               src = addr;
+                               dst = c->dev_addr;
+                       } else if (dir == DMA_DEV_TO_MEM) {
+                               src = c->dev_addr;
+                               dst = addr;
+                       }
+
+                       k3_dma_fill_desc(ds, dst, src, len, num++, c->ccfg);
+
+                       addr += len;
+                       avail -= len;
+               } while (avail);
+       }
+
+       ds->desc_hw[num-1].lli = 0;     /* end of link */
+       ds->size = total;
+       return vchan_tx_prep(&c->vc, &ds->vd, flags);
+}
+
+static int k3_dma_control(struct dma_chan *chan, enum dma_ctrl_cmd cmd,
+       unsigned long arg)
+{
+       struct k3_dma_chan *c = to_k3_chan(chan);
+       struct k3_dma_dev *d = to_k3_dma(chan->device);
+       struct dma_slave_config *cfg = (void *)arg;
+       struct k3_dma_phy *p = c->phy;
+       unsigned long flags;
+       u32 maxburst = 0, val = 0;
+       enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
+       LIST_HEAD(head);
+
+       switch (cmd) {
+       case DMA_SLAVE_CONFIG:
+               if (cfg == NULL)
+                       return -EINVAL;
+               c->dir = cfg->direction;
+               if (c->dir == DMA_DEV_TO_MEM) {
+                       c->ccfg = CX_CFG_DSTINCR;
+                       c->dev_addr = cfg->src_addr;
+                       maxburst = cfg->src_maxburst;
+                       width = cfg->src_addr_width;
+               } else if (c->dir == DMA_MEM_TO_DEV) {
+                       c->ccfg = CX_CFG_SRCINCR;
+                       c->dev_addr = cfg->dst_addr;
+                       maxburst = cfg->dst_maxburst;
+                       width = cfg->dst_addr_width;
+               }
+               switch (width) {
+               case DMA_SLAVE_BUSWIDTH_1_BYTE:
+               case DMA_SLAVE_BUSWIDTH_2_BYTES:
+               case DMA_SLAVE_BUSWIDTH_4_BYTES:
+               case DMA_SLAVE_BUSWIDTH_8_BYTES:
+                       val =  __ffs(width);
+                       break;
+               default:
+                       val = 3;
+                       break;
+               }
+               c->ccfg |= (val << 12) | (val << 16);
+
+               if ((maxburst == 0) || (maxburst > 16))
+                       val = 16;
+               else
+                       val = maxburst - 1;
+               c->ccfg |= (val << 20) | (val << 24);
+               c->ccfg |= CX_CFG_MEM2PER | CX_CFG_EN;
+
+               /* specific request line */
+               c->ccfg |= c->vc.chan.chan_id << 4;
+               break;
+
+       case DMA_TERMINATE_ALL:
+               dev_dbg(d->slave.dev, "vchan %p: terminate all\n", &c->vc);
+
+               /* Prevent this channel being scheduled */
+               spin_lock(&d->lock);
+               list_del_init(&c->node);
+               spin_unlock(&d->lock);
+
+               /* Clear the tx descriptor lists */
+               spin_lock_irqsave(&c->vc.lock, flags);
+               vchan_get_all_descriptors(&c->vc, &head);
+               if (p) {
+                       /* vchan is assigned to a pchan - stop the channel */
+                       k3_dma_terminate_chan(p, d);
+                       c->phy = NULL;
+                       p->vchan = NULL;
+                       p->ds_run = p->ds_done = NULL;
+               }
+               spin_unlock_irqrestore(&c->vc.lock, flags);
+               vchan_dma_desc_free_list(&c->vc, &head);
+               break;
+
+       case DMA_PAUSE:
+               dev_dbg(d->slave.dev, "vchan %p: pause\n", &c->vc);
+               if (c->status == DMA_IN_PROGRESS) {
+                       c->status = DMA_PAUSED;
+                       if (p) {
+                               k3_dma_pause_dma(p, false);
+                       } else {
+                               spin_lock(&d->lock);
+                               list_del_init(&c->node);
+                               spin_unlock(&d->lock);
+                       }
+               }
+               break;
+
+       case DMA_RESUME:
+               dev_dbg(d->slave.dev, "vchan %p: resume\n", &c->vc);
+               spin_lock_irqsave(&c->vc.lock, flags);
+               if (c->status == DMA_PAUSED) {
+                       c->status = DMA_IN_PROGRESS;
+                       if (p) {
+                               k3_dma_pause_dma(p, true);
+                       } else if (!list_empty(&c->vc.desc_issued)) {
+                               spin_lock(&d->lock);
+                               list_add_tail(&c->node, &d->chan_pending);
+                               spin_unlock(&d->lock);
+                       }
+               }
+               spin_unlock_irqrestore(&c->vc.lock, flags);
+               break;
+       default:
+               return -ENXIO;
+       }
+       return 0;
+}
+
+static void k3_dma_free_desc(struct virt_dma_desc *vd)
+{
+       struct k3_dma_desc_sw *ds =
+               container_of(vd, struct k3_dma_desc_sw, vd);
+
+       kfree(ds);
+}
+
+static struct of_device_id k3_pdma_dt_ids[] = {
+       { .compatible = "hisilicon,k3-dma-1.0", },
+       {}
+};
+MODULE_DEVICE_TABLE(of, k3_pdma_dt_ids);
+
+static struct dma_chan *k3_of_dma_simple_xlate(struct of_phandle_args *dma_spec,
+                                               struct of_dma *ofdma)
+{
+       struct k3_dma_dev *d = ofdma->of_dma_data;
+       unsigned int request = dma_spec->args[0];
+
+       if (request > d->dma_requests)
+               return NULL;
+
+       return dma_get_slave_channel(&(d->chans[request].vc.chan));
+}
+
+static int k3_dma_probe(struct platform_device *op)
+{
+       struct k3_dma_dev *d;
+       const struct of_device_id *of_id;
+       struct resource *iores;
+       int i, ret, irq = 0;
+
+       iores = platform_get_resource(op, IORESOURCE_MEM, 0);
+       if (!iores)
+               return -EINVAL;
+
+       d = devm_kzalloc(&op->dev, sizeof(*d), GFP_KERNEL);
+       if (!d)
+               return -ENOMEM;
+
+       d->base = devm_request_and_ioremap(&op->dev, iores);
+       if (!d->base)
+               return -EADDRNOTAVAIL;
+
+       of_id = of_match_device(k3_pdma_dt_ids, &op->dev);
+       if (of_id) {
+               of_property_read_u32((&op->dev)->of_node,
+                               "dma-channels", &d->dma_channels);
+               of_property_read_u32((&op->dev)->of_node,
+                               "dma-requests", &d->dma_requests);
+       }
+
+       d->clk = devm_clk_get(&op->dev, NULL);
+       if (IS_ERR(d->clk)) {
+               dev_err(&op->dev, "no dma clk\n");
+               return PTR_ERR(d->clk);
+       }
+
+       irq = platform_get_irq(op, 0);
+       ret = devm_request_irq(&op->dev, irq,
+                       k3_dma_int_handler, IRQF_DISABLED, DRIVER_NAME, d);
+       if (ret)
+               return ret;
+
+       /* init phy channel */
+       d->phy = devm_kzalloc(&op->dev,
+               d->dma_channels * sizeof(struct k3_dma_phy), GFP_KERNEL);
+       if (d->phy == NULL)
+               return -ENOMEM;
+
+       for (i = 0; i < d->dma_channels; i++) {
+               struct k3_dma_phy *p = &d->phy[i];
+
+               p->idx = i;
+               p->base = d->base + i * 0x40;
+       }
+
+       INIT_LIST_HEAD(&d->slave.channels);
+       dma_cap_set(DMA_SLAVE, d->slave.cap_mask);
+       dma_cap_set(DMA_MEMCPY, d->slave.cap_mask);
+       d->slave.dev = &op->dev;
+       d->slave.device_alloc_chan_resources = k3_dma_alloc_chan_resources;
+       d->slave.device_free_chan_resources = k3_dma_free_chan_resources;
+       d->slave.device_tx_status = k3_dma_tx_status;
+       d->slave.device_prep_dma_memcpy = k3_dma_prep_memcpy;
+       d->slave.device_prep_slave_sg = k3_dma_prep_slave_sg;
+       d->slave.device_issue_pending = k3_dma_issue_pending;
+       d->slave.device_control = k3_dma_control;
+       d->slave.copy_align = DMA_ALIGN;
+       d->slave.chancnt = d->dma_requests;
+
+       /* init virtual channel */
+       d->chans = devm_kzalloc(&op->dev,
+               d->dma_requests * sizeof(struct k3_dma_chan), GFP_KERNEL);
+       if (d->chans == NULL)
+               return -ENOMEM;
+
+       for (i = 0; i < d->dma_requests; i++) {
+               struct k3_dma_chan *c = &d->chans[i];
+
+               c->status = DMA_IN_PROGRESS;
+               INIT_LIST_HEAD(&c->node);
+               c->vc.desc_free = k3_dma_free_desc;
+               vchan_init(&c->vc, &d->slave);
+       }
+
+       /* Enable clock before accessing registers */
+       ret = clk_prepare_enable(d->clk);
+       if (ret < 0) {
+               dev_err(&op->dev, "clk_prepare_enable failed: %d\n", ret);
+               return ret;
+       }
+
+       k3_dma_enable_dma(d, true);
+
+       ret = dma_async_device_register(&d->slave);
+       if (ret)
+               return ret;
+
+       ret = of_dma_controller_register((&op->dev)->of_node,
+                                       k3_of_dma_simple_xlate, d);
+       if (ret)
+               goto of_dma_register_fail;
+
+       spin_lock_init(&d->lock);
+       INIT_LIST_HEAD(&d->chan_pending);
+       tasklet_init(&d->task, k3_dma_tasklet, (unsigned long)d);
+       platform_set_drvdata(op, d);
+       dev_info(&op->dev, "initialized\n");
+
+       return 0;
+
+of_dma_register_fail:
+       dma_async_device_unregister(&d->slave);
+       return ret;
+}
+
+static int k3_dma_remove(struct platform_device *op)
+{
+       struct k3_dma_chan *c, *cn;
+       struct k3_dma_dev *d = platform_get_drvdata(op);
+
+       dma_async_device_unregister(&d->slave);
+       of_dma_controller_free((&op->dev)->of_node);
+
+       list_for_each_entry_safe(c, cn, &d->slave.channels, vc.chan.device_node) {
+               list_del(&c->vc.chan.device_node);
+               tasklet_kill(&c->vc.task);
+       }
+       tasklet_kill(&d->task);
+       clk_disable_unprepare(d->clk);
+       return 0;
+}
+
+static int k3_dma_suspend(struct device *dev)
+{
+       struct k3_dma_dev *d = dev_get_drvdata(dev);
+       u32 stat = 0;
+
+       stat = k3_dma_get_chan_stat(d);
+       if (stat) {
+               dev_warn(d->slave.dev,
+                       "chan %d is running fail to suspend\n", stat);
+               return -1;
+       }
+       k3_dma_enable_dma(d, false);
+       clk_disable_unprepare(d->clk);
+       return 0;
+}
+
+static int k3_dma_resume(struct device *dev)
+{
+       struct k3_dma_dev *d = dev_get_drvdata(dev);
+       int ret = 0;
+
+       ret = clk_prepare_enable(d->clk);
+       if (ret < 0) {
+               dev_err(d->slave.dev, "clk_prepare_enable failed: %d\n", ret);
+               return ret;
+       }
+       k3_dma_enable_dma(d, true);
+       return 0;
+}
+
+SIMPLE_DEV_PM_OPS(k3_dma_pmops, k3_dma_suspend, k3_dma_resume);
+
+static struct platform_driver k3_pdma_driver = {
+       .driver         = {
+               .name   = DRIVER_NAME,
+               .owner  = THIS_MODULE,
+               .pm     = &k3_dma_pmops,
+               .of_match_table = k3_pdma_dt_ids,
+       },
+       .probe          = k3_dma_probe,
+       .remove         = k3_dma_remove,
+};
+
+module_platform_driver(k3_pdma_driver);
+
+MODULE_DESCRIPTION("Hisilicon k3 DMA Driver");
+MODULE_ALIAS("platform:k3dma");
+MODULE_LICENSE("GPL v2");