dmaengine: Enable SpacemiT K1-X mmp_pdma driver 31/316531/1
authorMichal Wilczynski <m.wilczynski@samsung.com>
Thu, 22 Aug 2024 06:52:53 +0000 (08:52 +0200)
committerMichal Wilczynski <m.wilczynski@samsung.com>
Thu, 22 Aug 2024 11:53:26 +0000 (13:53 +0200)
The pdma driver is a dependency for important hardware like i2c8
adapter. Port it from the vendor kernel [1].

[1] - https://github.com/BPI-SINOVOIP/pi-linux.git

Change-Id: Ia16dcbce82fa4b25ed7bf480f500bd2c5bd63e96
Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/mmp_pdma_k1x.c [new file with mode: 0644]

index 7a618f629e86b8b4e8ff493c2d6095c7c63c8845..813cf87c9f03539dca5b8f88c405b0f9f138f9e8 100644 (file)
@@ -416,6 +416,13 @@ config MMP_PDMA
        help
          Support the MMP PDMA engine for PXA and MMP platform.
 
+config MMP_PDMA_SPACEMIT_K1X
+       bool "Spacemit mmp_pdma support"
+       depends on ARCH_MMP || ARCH_PXA || COMPILE_TEST || ARCH_SPACEMIT
+       select DMA_ENGINE
+       help
+         Support the MMP PDMA engine for Spacemit-k1x platform.
+
 config MMP_TDMA
        tristate "MMP Two-Channel DMA support"
        depends on ARCH_MMP || COMPILE_TEST
index 83553a97a010e157b285ad6c47557df31c8bc835..33613aa18b1c8f0d2ef6f4b68f406a4e13bb84f2 100644 (file)
@@ -52,6 +52,7 @@ obj-$(CONFIG_MILBEAUT_HDMAC) += milbeaut-hdmac.o
 obj-$(CONFIG_MILBEAUT_XDMAC) += milbeaut-xdmac.o
 obj-$(CONFIG_MMP_PDMA) += mmp_pdma.o
 obj-$(CONFIG_MMP_TDMA) += mmp_tdma.o
+obj-$(CONFIG_MMP_PDMA_SPACEMIT_K1X) += mmp_pdma_k1x.o
 obj-$(CONFIG_MOXART_DMA) += moxart-dma.o
 obj-$(CONFIG_MPC512X_DMA) += mpc512x_dma.o
 obj-$(CONFIG_MV_XOR) += mv_xor.o
diff --git a/drivers/dma/mmp_pdma_k1x.c b/drivers/dma/mmp_pdma_k1x.c
new file mode 100644 (file)
index 0000000..749640f
--- /dev/null
@@ -0,0 +1,1667 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright 2012 Marvell International Ltd.
+ */
+
+#include <linux/err.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/types.h>
+#include <linux/interrupt.h>
+#include <linux/dma-mapping.h>
+#include <linux/slab.h>
+#include <linux/dmaengine.h>
+#include <linux/platform_device.h>
+#include <linux/device.h>
+#include <linux/platform_data/mmp_dma.h>
+#include <linux/dmapool.h>
+#include <linux/clk.h>
+#include <linux/reset.h>
+#include <linux/of_device.h>
+#include <linux/of_dma.h>
+#include <linux/of.h>
+
+#include <linux/delay.h>
+#include <linux/pm_runtime.h>
+#include <linux/pm_qos.h>
+
+#include "dmaengine.h"
+
+#define DDADRH(n)      (0x0300 + ((n) << 4))
+#define DSADRH(n)      (0x0304 + ((n) << 4))
+#define DTADRH(n)      (0x0308 + ((n) << 4))
+#define DCSR_LPAEEN    BIT(21) /* Long Physical Address Extension enable */
+#define DRCMR_INVALID  100             /* Max DMA request number + 1 */
+#define DCMD_BURST64   (4 << 16)       /* 64 byte burst */
+
+#define DCSR           0x0000
+#define DALGN          0x00a0
+#define DINT           0x00f0
+#define DDADR          0x0200
+#define DSADR(n)       (0x0204 + ((n) << 4))
+#define DTADR(n)       (0x0208 + ((n) << 4))
+#define DCMD           0x020c
+
+#define DCSR_RUN       BIT(31) /* Run Bit (read / write) */
+#define DCSR_NODESC    BIT(30) /* No-Descriptor Fetch (read / write) */
+#define DCSR_STOPIRQEN BIT(29) /* Stop Interrupt Enable (read / write) */
+#define DCSR_REQPEND   BIT(8)  /* Request Pending (read-only) */
+#define DCSR_STOPSTATE BIT(3)  /* Stop State (read-only) */
+#define DCSR_ENDINTR   BIT(2)  /* End Interrupt (read / write) */
+#define DCSR_STARTINTR BIT(1)  /* Start Interrupt (read / write) */
+#define DCSR_BUSERR    BIT(0)  /* Bus Error Interrupt (read / write) */
+
+#define DCSR_EORIRQEN  BIT(28) /* End of Receive Interrupt Enable (R/W) */
+#define DCSR_EORJMPEN  BIT(27) /* Jump to next descriptor on EOR */
+#define DCSR_EORSTOPEN BIT(26) /* STOP on an EOR */
+#define DCSR_SETCMPST  BIT(25) /* Set Descriptor Compare Status */
+#define DCSR_CLRCMPST  BIT(24) /* Clear Descriptor Compare Status */
+#define DCSR_CMPST     BIT(10) /* The Descriptor Compare Status */
+#define DCSR_EORINTR   BIT(9)  /* The end of Receive */
+
+#define DRCMR(n)       ((((n) < 64) ? 0x0100 : 0x1100) + (((n) & 0x3f) << 2))
+#define DRCMR_MAPVLD   BIT(7)  /* Map Valid (read / write) */
+#define DRCMR_CHLNUM   0x1f    /* mask for Channel Number (read / write) */
+
+#define DDADR_DESCADDR 0xfffffff0      /* Address of next descriptor (mask) */
+#define DDADR_STOP     BIT(0)  /* Stop (read / write) */
+
+#define DCMD_INCSRCADDR        BIT(31) /* Source Address Increment Setting. */
+#define DCMD_INCTRGADDR        BIT(30) /* Target Address Increment Setting. */
+#define DCMD_FLOWSRC   BIT(29) /* Flow Control by the source. */
+#define DCMD_FLOWTRG   BIT(28) /* Flow Control by the target. */
+#define DCMD_STARTIRQEN        BIT(22) /* Start Interrupt Enable */
+#define DCMD_ENDIRQEN  BIT(21) /* End Interrupt Enable */
+#define DCMD_ENDIAN    BIT(18) /* Device Endian-ness. */
+#define DCMD_BURST8    (1 << 16)       /* 8 byte burst */
+#define DCMD_BURST16   (2 << 16)       /* 16 byte burst */
+#define DCMD_BURST32   (3 << 16)       /* 32 byte burst */
+#define DCMD_WIDTH1    (1 << 14)       /* 1 byte width */
+#define DCMD_WIDTH2    (2 << 14)       /* 2 byte width (HalfWord) */
+#define DCMD_WIDTH4    (3 << 14)       /* 4 byte width (Word) */
+#define DCMD_LENGTH    0x01fff         /* length mask (max = 8K - 1) */
+
+#define PDMA_MAX_DESC_BYTES    DCMD_LENGTH
+
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+struct mmp_pdma_desc_hw {
+       u32 ddadr;      /* Points to the next descriptor + flags */
+       u32 dsadr;      /* DSADR value for the current transfer */
+       u32 dtadr;      /* DTADR value for the current transfer */
+       u32 dcmd;       /* DCMD value for the current transfer */
+       u32 ddadrh;     /* Points to the next descriptor + flags */
+       u32 dsadrh;     /* DSADR value for the current transfer */
+       u32 dtadrh;     /* DTADR value for the current transfer */
+       u32 rsvd;       /* DCMD value for the current transfer */
+} __aligned(64);
+#else
+struct mmp_pdma_desc_hw {
+       u32 ddadr;      /* Points to the next descriptor + flags */
+       u32 dsadr;      /* DSADR value for the current transfer */
+       u32 dtadr;      /* DTADR value for the current transfer */
+       u32 dcmd;       /* DCMD value for the current transfer */
+} __aligned(32);
+#endif
+
+struct mmp_pdma_desc_sw {
+       struct mmp_pdma_desc_hw desc;
+       struct list_head node;
+       struct list_head tx_list;
+       struct dma_async_tx_descriptor async_tx;
+};
+
+struct mmp_pdma_phy;
+
+struct mmp_pdma_chan {
+       struct device *dev;
+       struct dma_chan chan;
+       struct dma_async_tx_descriptor desc;
+       struct mmp_pdma_phy *phy;
+       enum dma_transfer_direction dir;
+       struct dma_slave_config slave_config;
+
+       struct mmp_pdma_desc_sw *cyclic_first;  /* first desc_sw if channel
+                                                * is in cyclic mode */
+
+       /* channel's basic info */
+       struct tasklet_struct tasklet;
+       u32 dcmd;
+       u32 drcmr;
+       u32 dev_addr;
+
+       /* list for desc */
+       spinlock_t desc_lock;           /* Descriptor list lock */
+       struct list_head chain_pending; /* Link descriptors queue for pending */
+       struct list_head chain_running; /* Link descriptors queue for running */
+       bool idle;                      /* channel statue machine */
+       bool byte_align;
+
+       int user_do_qos;
+       int qos_count; /* Per-channel qos count */
+       enum dma_status status; /* channel state machine */
+       u32 bytes_residue;
+
+       struct dma_pool *desc_pool;     /* Descriptors pool */
+};
+
+struct mmp_pdma_phy {
+       int idx;
+       void __iomem *base;
+       struct mmp_pdma_chan *vchan;
+};
+
+struct reserved_chan{
+       int     chan_id;
+       int     drcmr;
+};
+
+struct mmp_pdma_device {
+       int                             dma_channels;
+       int                             nr_reserved_channels;
+       struct reserved_chan            *reserved_channels;
+       s32                             lpm_qos;
+       struct clk                      *clk;
+       struct reset_control            *resets;
+       int                             max_burst_size;
+       void __iomem                    *base;
+       struct device                   *dev;
+       struct dma_device               device;
+       struct mmp_pdma_phy             *phy;
+       spinlock_t phy_lock; /* protect alloc/free phy channels */
+};
+
+#define tx_to_mmp_pdma_desc(tx)                                        \
+       container_of(tx, struct mmp_pdma_desc_sw, async_tx)
+#define to_mmp_pdma_desc(lh)                                   \
+       container_of(lh, struct mmp_pdma_desc_sw, node)
+#define to_mmp_pdma_chan(dchan)                                        \
+       container_of(dchan, struct mmp_pdma_chan, chan)
+#define to_mmp_pdma_dev(dmadev)                                        \
+       container_of(dmadev, struct mmp_pdma_device, device)
+
+static void mmp_pdma_qos_get(struct mmp_pdma_chan *chan);
+static void mmp_pdma_qos_put(struct mmp_pdma_chan *chan);
+
+#define QSPI_PHY_CHAN  15
+
+static int mmp_pdma_config_write(struct dma_chan *dchan,
+                          struct dma_slave_config *cfg,
+                          enum dma_transfer_direction direction);
+
+static void set_desc(struct mmp_pdma_phy *phy, dma_addr_t addr)
+{
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+       u32 ddadrh;
+#endif
+       u32 reg = (phy->idx << 4) + DDADR;
+
+       writel(addr & 0xffffffff, phy->base + reg);
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+       /* config higher bits for desc address */
+       ddadrh = (addr >> 32);
+       writel(ddadrh, phy->base + DDADRH(phy->idx));
+#endif
+}
+
+static void enable_chan(struct mmp_pdma_phy *phy)
+{
+       u32 reg, dalgn;
+       u32 dcsr;
+       unsigned long flags;
+       struct mmp_pdma_device *pdev;
+
+       if (phy == NULL)
+               return;
+
+       if (!phy->vchan)
+               return;
+
+       pdev = to_mmp_pdma_dev(phy->vchan->chan.device);
+
+       spin_lock_irqsave(&pdev->phy_lock, flags);
+
+       reg = DRCMR(phy->vchan->drcmr);
+       writel(DRCMR_MAPVLD | phy->idx, phy->base + reg);
+
+       dalgn = readl(phy->base + DALGN);
+       if (phy->vchan->byte_align)
+               dalgn |= 1 << phy->idx;
+       else
+               dalgn &= ~(1 << phy->idx);
+       writel(dalgn, phy->base + DALGN);
+
+       reg = (phy->idx << 2) + DCSR;
+
+       dcsr = readl(phy->base + reg);
+       dcsr |= (DCSR_RUN | DCSR_EORIRQEN | DCSR_EORSTOPEN);
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+       /* use long descriptor mode: set DCSR_LPAEEN bit */
+       dcsr |= DCSR_LPAEEN;
+#endif
+       writel(dcsr, phy->base + reg);
+
+       spin_unlock_irqrestore(&pdev->phy_lock, flags);
+}
+
+static void disable_chan(struct mmp_pdma_phy *phy)
+{
+       u32 reg;
+       u32 dcsr, cnt = 1000;
+
+       if (!phy)
+               return;
+
+       reg = (phy->idx << 2) + DCSR;
+
+       dcsr = readl(phy->base + reg);
+       dcsr &= ~(DCSR_RUN | DCSR_EORIRQEN | DCSR_EORSTOPEN);
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+       /* use long descriptor mode: set DCSR_LPAEEN bit */
+       dcsr &= ~DCSR_LPAEEN;
+#endif
+       writel(dcsr, phy->base + reg);
+
+       /* ensure dma is stopped. */
+       dcsr = readl(phy->base + reg);
+       while (!(dcsr & (0x1 << 3)) && --cnt) {
+               udelay(10);
+               dcsr = readl(phy->base + reg);
+       }
+
+       WARN_ON(!cnt);
+}
+
+static int clear_chan_irq(struct mmp_pdma_phy *phy)
+{
+       u32 dcsr;
+       u32 dint = readl(phy->base + DINT);
+       u32 reg = (phy->idx << 2) + DCSR;
+
+       if (!(dint & BIT(phy->idx)))
+               return -EAGAIN;
+
+       /* clear irq */
+       dcsr = readl(phy->base + reg);
+       writel(dcsr, phy->base + reg);
+       if ((dcsr & DCSR_BUSERR) && (phy->vchan))
+               dev_warn(phy->vchan->dev, "DCSR_BUSERR\n");
+
+       return 0;
+}
+
+static irqreturn_t mmp_pdma_chan_handler(int irq, void *dev_id)
+{
+       struct mmp_pdma_phy *phy = dev_id;
+       struct mmp_pdma_chan *pchan = phy->vchan;
+
+       if (clear_chan_irq(phy) != 0)
+               return IRQ_NONE;
+
+       if (pchan)
+               tasklet_schedule(&pchan->tasklet);
+       tasklet_schedule(&phy->vchan->tasklet);
+
+       return IRQ_HANDLED;
+}
+
+static bool is_channel_reserved(struct mmp_pdma_device *pdev, int chan_id)
+{
+       int i;
+
+       for (i = 0; i < pdev->nr_reserved_channels; i++) {
+               if (chan_id == pdev->reserved_channels[i].chan_id)
+                       return true;
+       }
+
+       return false;
+}
+
+static struct mmp_pdma_phy * lookup_phy_for_drcmr(struct mmp_pdma_device *pdev, int drcmr)
+{
+       int i;
+       int chan_id;
+       struct mmp_pdma_phy *phy;
+
+       for (i = 0; i < pdev->nr_reserved_channels; i++) {
+               if (drcmr == pdev->reserved_channels[i].drcmr) {
+                       chan_id = pdev->reserved_channels[i].chan_id;
+                       phy = &pdev->phy[chan_id];
+                       return phy;
+               }
+       }
+
+       return NULL;
+}
+
+static irqreturn_t mmp_pdma_int_handler(int irq, void *dev_id)
+{
+       struct mmp_pdma_device *pdev = dev_id;
+       struct mmp_pdma_phy *phy;
+       u32 dint = readl(pdev->base + DINT);
+       int i, ret;
+       int irq_num = 0;
+       unsigned long flags;
+
+       while (dint) {
+               i = __ffs(dint);
+               /* only handle interrupts belonging to pdma driver*/
+               if (i >= pdev->dma_channels)
+                       break;
+
+               dint &= (dint - 1);
+               phy = &pdev->phy[i];
+               spin_lock_irqsave(&pdev->phy_lock, flags);
+
+               ret = mmp_pdma_chan_handler(irq, phy);
+
+               spin_unlock_irqrestore(&pdev->phy_lock, flags);
+               if (ret == IRQ_HANDLED)
+                       irq_num++;
+       }
+
+       if (irq_num)
+               return IRQ_HANDLED;
+
+       return IRQ_NONE;
+}
+
+/* lookup free phy channel as descending priority */
+static struct mmp_pdma_phy *lookup_phy(struct mmp_pdma_chan *pchan)
+{
+       int prio, i;
+       struct mmp_pdma_device *pdev = to_mmp_pdma_dev(pchan->chan.device);
+       struct mmp_pdma_phy *phy, *found = NULL;
+       unsigned long flags;
+
+       /*
+        * dma channel priorities
+        * ch 0 - 3,  16 - 19  <--> (0)
+        * ch 4 - 7,  20 - 23  <--> (1)
+        * ch 8 - 11, 24 - 27  <--> (2)
+        * ch 12 - 15, 28 - 31  <--> (3)
+        */
+
+       spin_lock_irqsave(&pdev->phy_lock, flags);
+
+       phy = lookup_phy_for_drcmr(pdev, pchan->drcmr);
+
+       if (phy != NULL) {
+               if (!phy->vchan) {
+                       phy->vchan = pchan;
+                       found = phy;
+               }
+
+               goto out_unlock;
+       }
+
+       for (prio = 0; prio <= ((pdev->dma_channels - 1) & 0xf) >> 2; prio++) {
+               for (i = 0; i < pdev->dma_channels; i++) {
+                       if (prio != (i & 0xf) >> 2)
+                               continue;
+
+                       if (is_channel_reserved(pdev, i))
+                               continue;
+                       phy = &pdev->phy[i];
+                       if (!phy->vchan) {
+                               phy->vchan = pchan;
+                               found = phy;
+                               goto out_unlock;
+                       }
+               }
+       }
+
+out_unlock:
+       spin_unlock_irqrestore(&pdev->phy_lock, flags);
+       return found;
+}
+
+static void mmp_pdma_free_phy(struct mmp_pdma_chan *pchan)
+{
+       struct mmp_pdma_device *pdev = to_mmp_pdma_dev(pchan->chan.device);
+       unsigned long flags;
+       u32 reg;
+
+       if (!pchan->phy)
+               return;
+
+       /* clear the channel mapping in DRCMR */
+       reg = DRCMR(pchan->drcmr);
+       writel(0, pchan->phy->base + reg);
+
+       spin_lock_irqsave(&pdev->phy_lock, flags);
+       pchan->phy->vchan = NULL;
+       pchan->phy = NULL;
+
+       spin_unlock_irqrestore(&pdev->phy_lock, flags);
+}
+
+/*
+ * start_pending_queue - transfer any pending transactions
+ * pending list ==> running list
+ */
+static int start_pending_queue(struct mmp_pdma_chan *chan)
+{
+       struct mmp_pdma_desc_sw *desc;
+       struct mmp_pdma_desc_sw *_desc;
+
+       /* still in running, irq will start the pending list */
+       if (chan->status == DMA_IN_PROGRESS) {
+               dev_dbg(chan->dev, "DMA controller still busy\n");
+               return -1;
+       }
+
+       if (list_empty(&chan->chain_pending)) {
+               /* chance to re-fetch phy channel with higher prio */
+               mmp_pdma_free_phy(chan);
+               dev_dbg(chan->dev, "no pending list\n");
+
+               return -1;
+       }
+
+       if (!chan->phy) {
+               chan->phy = lookup_phy(chan);
+               if (!chan->phy) {
+                       dev_dbg(chan->dev, "no free dma channel\n");
+
+                       return -1;
+               }
+       }
+
+       /*
+        * pending -> running
+        * reintilize pending list
+        */
+       list_for_each_entry_safe(desc, _desc, &chan->chain_pending, node) {
+               list_del(&desc->node);
+               list_add_tail(&desc->node, &chan->chain_running);
+               if (desc->desc.ddadr & DDADR_STOP)
+                       break;
+       }
+
+       desc = list_first_entry(&chan->chain_running,
+                               struct mmp_pdma_desc_sw, node);
+
+       /*
+        * Program the descriptor's address into the DMA controller,
+        * then start the DMA transaction
+        */
+       set_desc(chan->phy, desc->async_tx.phys);
+       enable_chan(chan->phy);
+       chan->idle = false;
+       chan->status = DMA_IN_PROGRESS;
+       chan->bytes_residue = 0;
+       return 0;
+}
+
+
+/* desc->tx_list ==> pending list */
+static dma_cookie_t mmp_pdma_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(tx->chan);
+       struct mmp_pdma_desc_sw *desc = tx_to_mmp_pdma_desc(tx);
+       struct mmp_pdma_desc_sw *child;
+       unsigned long flags;
+       dma_cookie_t cookie = -EBUSY;
+
+       spin_lock_irqsave(&chan->desc_lock, flags);
+
+       list_for_each_entry(child, &desc->tx_list, node) {
+               cookie = dma_cookie_assign(&child->async_tx);
+       }
+
+       /* softly link to pending list - desc->tx_list ==> pending list */
+       list_splice_tail_init(&desc->tx_list, &chan->chain_pending);
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+       return cookie;
+}
+
+static struct mmp_pdma_desc_sw *
+mmp_pdma_alloc_descriptor(struct mmp_pdma_chan *chan)
+{
+       struct mmp_pdma_desc_sw *desc;
+       dma_addr_t pdesc;
+
+       desc = dma_pool_zalloc(chan->desc_pool, GFP_ATOMIC, &pdesc);
+       if (!desc) {
+               dev_err(chan->dev, "out of memory for link descriptor\n");
+               return NULL;
+       }
+
+       INIT_LIST_HEAD(&desc->tx_list);
+       dma_async_tx_descriptor_init(&desc->async_tx, &chan->chan);
+       /* each desc has submit */
+       desc->async_tx.tx_submit = mmp_pdma_tx_submit;
+       desc->async_tx.phys = pdesc;
+
+       return desc;
+}
+
+/*
+ * mmp_pdma_alloc_chan_resources - Allocate resources for DMA channel.
+ *
+ * This function will create a dma pool for descriptor allocation.
+ * Request irq only when channel is requested
+ * Return - The number of allocated descriptors.
+ */
+
+static int mmp_pdma_alloc_chan_resources(struct dma_chan *dchan)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+
+       if (chan->desc_pool)
+               return 1;
+
+       chan->desc_pool = dma_pool_create(dev_name(&dchan->dev->device),
+                                         chan->dev,
+                                         sizeof(struct mmp_pdma_desc_sw),
+                                         __alignof__(struct mmp_pdma_desc_sw),
+                                         0);
+       if (!chan->desc_pool) {
+               dev_err(chan->dev, "unable to allocate descriptor pool\n");
+               return -ENOMEM;
+       }
+
+       chan->status = DMA_COMPLETE;
+       chan->dir = 0;
+       chan->dcmd = 0;
+
+       mmp_pdma_free_phy(chan);
+
+       chan->idle = true;
+       chan->dev_addr = 0;
+       return 1;
+}
+
+static void mmp_pdma_free_desc_list(struct mmp_pdma_chan *chan,
+                                   struct list_head *list)
+{
+       struct mmp_pdma_desc_sw *desc, *_desc;
+
+       list_for_each_entry_safe(desc, _desc, list, node) {
+               list_del(&desc->node);
+               dma_pool_free(chan->desc_pool, desc, desc->async_tx.phys);
+       }
+}
+
+static void mmp_pdma_free_chan_resources(struct dma_chan *dchan)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+       unsigned long flags;
+
+       /* wait until task ends if necessary */
+       tasklet_kill(&chan->tasklet);
+
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       mmp_pdma_free_desc_list(chan, &chan->chain_pending);
+       mmp_pdma_free_desc_list(chan, &chan->chain_running);
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+       dma_pool_destroy(chan->desc_pool);
+       chan->desc_pool = NULL;
+       chan->idle = true;
+       chan->dev_addr = 0;
+
+       chan->status = DMA_COMPLETE;
+       chan->dir = 0;
+       chan->dcmd = 0;
+
+       mmp_pdma_free_phy(chan);
+       return;
+}
+
+#define INVALID_BURST_SETTING  -1
+#define DEFAULT_MAX_BURST_SIZE 32
+
+static int get_max_burst_setting(unsigned int max_burst_size)
+{
+       switch (max_burst_size) {
+       case 8:
+               return DCMD_BURST8;
+       case 16:
+               return DCMD_BURST16;
+       case 32:
+               return DCMD_BURST32;
+       case 64:
+               return DCMD_BURST64;
+       default:
+               return INVALID_BURST_SETTING;
+       }
+}
+
+static struct dma_async_tx_descriptor *
+mmp_pdma_prep_memcpy(struct dma_chan *dchan,
+                    dma_addr_t dma_dst, dma_addr_t dma_src,
+                    size_t len, unsigned long flags)
+{
+       struct mmp_pdma_chan *chan;
+       struct mmp_pdma_desc_sw *first = NULL, *prev = NULL, *new;
+       size_t copy = 0;
+       struct mmp_pdma_device *dev;
+       int value;
+
+       if (!dchan)
+               return NULL;
+
+       if (!len)
+               return NULL;
+
+       chan = to_mmp_pdma_chan(dchan);
+       chan->byte_align = false;
+
+       if (!chan->dir) {
+               chan->dir = DMA_MEM_TO_MEM;
+               chan->dcmd = DCMD_INCTRGADDR | DCMD_INCSRCADDR;
+               dev = to_mmp_pdma_dev(dchan->device);
+               value = get_max_burst_setting(dev->max_burst_size);
+
+               BUG_ON(value == INVALID_BURST_SETTING);
+
+               chan->dcmd |= value;
+       }
+
+       do {
+               /* Allocate the link descriptor from DMA pool */
+               new = mmp_pdma_alloc_descriptor(chan);
+               if (!new) {
+                       dev_err(chan->dev, "no memory for desc\n");
+                       goto fail;
+               }
+
+               copy = min_t(size_t, len, PDMA_MAX_DESC_BYTES);
+               if (dma_src & 0x7 || dma_dst & 0x7)
+                       chan->byte_align = true;
+
+               new->desc.dcmd = chan->dcmd | (DCMD_LENGTH & copy);
+
+               /*
+                * Check whether descriptor/source-addr/target-addr is in
+                * region higher than 4G. If so, set related higher bits to 1.
+                */
+               if (chan->dir == DMA_MEM_TO_DEV) {
+                       new->desc.dsadr = dma_src & 0xffffffff;
+                       new->desc.dtadr = dma_dst;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                       new->desc.dsadrh = (dma_src >> 32);
+                       new->desc.dtadrh = 0;
+#endif
+               } else if (chan->dir == DMA_DEV_TO_MEM) {
+                       new->desc.dsadr = dma_src;
+                       new->desc.dtadr = dma_dst & 0xffffffff;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                       new->desc.dsadrh = 0;
+                       new->desc.dtadrh = (dma_dst >> 32);
+#endif
+               } else if (chan->dir == DMA_MEM_TO_MEM) {
+                       new->desc.dsadr = dma_src & 0xffffffff;
+                       new->desc.dtadr = dma_dst & 0xffffffff;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                       new->desc.dsadrh = (dma_src >> 32);
+                       new->desc.dtadrh = (dma_dst >> 32);
+#endif
+               } else {
+                       dev_err(chan->dev, "wrong direction: 0x%x\n", chan->dir);
+                       goto fail;
+               }
+
+               if (!first)
+                       first = new;
+               else {
+                       prev->desc.ddadr = new->async_tx.phys;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                       prev->desc.ddadrh = (new->async_tx.phys >> 32);
+#endif
+               }
+
+               new->async_tx.cookie = 0;
+               async_tx_ack(&new->async_tx);
+
+               prev = new;
+               len -= copy;
+
+               if (chan->dir == DMA_MEM_TO_DEV) {
+                       dma_src += copy;
+               } else if (chan->dir == DMA_DEV_TO_MEM) {
+                       dma_dst += copy;
+               } else if (chan->dir == DMA_MEM_TO_MEM) {
+                       dma_src += copy;
+                       dma_dst += copy;
+               }
+
+               /* Insert the link descriptor to the LD ring */
+               list_add_tail(&new->node, &first->tx_list);
+       } while (len);
+
+       first->async_tx.flags = flags; /* client is in control of this ack */
+       first->async_tx.cookie = -EBUSY;
+
+       /* last desc and fire IRQ */
+       new->desc.ddadr = DDADR_STOP;
+       new->desc.dcmd |= DCMD_ENDIRQEN;
+
+       chan->cyclic_first = NULL;
+
+       return &first->async_tx;
+
+fail:
+       if (first)
+               mmp_pdma_free_desc_list(chan, &first->tx_list);
+       return NULL;
+}
+
+static struct dma_async_tx_descriptor *
+mmp_pdma_prep_slave_sg(struct dma_chan *dchan, struct scatterlist *sgl,
+                      unsigned int sg_len, enum dma_transfer_direction dir,
+                      unsigned long flags, void *context)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+       struct mmp_pdma_desc_sw *first = NULL, *prev = NULL, *new = NULL;
+       size_t len, avail;
+       struct scatterlist *sg;
+       dma_addr_t addr;
+       int i;
+
+       if ((sgl == NULL) || (sg_len == 0))
+               return NULL;
+
+       chan->byte_align = true;
+
+       mmp_pdma_config_write(dchan, &chan->slave_config, dir);
+
+       for_each_sg(sgl, sg, sg_len, i) {
+               addr = sg_dma_address(sg);
+               avail = sg_dma_len(sgl);
+
+               do {
+                       len = min_t(size_t, avail, PDMA_MAX_DESC_BYTES);
+                       if (addr & 0x7)
+                               chan->byte_align = true;
+
+                       /* allocate and populate the descriptor */
+                       new = mmp_pdma_alloc_descriptor(chan);
+                       if (!new) {
+                               dev_err(chan->dev, "no memory for desc\n");
+                               goto fail;
+                       }
+
+                       new->desc.dcmd = chan->dcmd | (DCMD_LENGTH & len);
+
+                       /*
+                        * Check whether descriptor/source-addr/target-addr is in
+                        * region higher than 4G. If so, set related higher bits to 1.
+                        */
+                       if (dir == DMA_MEM_TO_DEV) {
+                               new->desc.dsadr = addr & 0xffffffff;
+                               new->desc.dtadr = chan->dev_addr;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                               new->desc.dsadrh = (addr >> 32);
+                               new->desc.dtadrh = 0;
+#endif
+                       } else if (dir == DMA_DEV_TO_MEM) {
+                               new->desc.dsadr = chan->dev_addr;
+                               new->desc.dtadr = addr & 0xffffffff;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                               new->desc.dsadrh = 0;
+                               new->desc.dtadrh = (addr >> 32);
+#endif
+                       } else {
+                               dev_err(chan->dev, "wrong direction: 0x%x\n", chan->dir);
+                               goto fail;
+                       }
+
+                       if (!first)
+                               first = new;
+                       else {
+                               prev->desc.ddadr = new->async_tx.phys;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                               prev->desc.ddadrh = (new->async_tx.phys >> 32);
+#endif
+                       }
+
+                       new->async_tx.cookie = 0;
+                       async_tx_ack(&new->async_tx);
+                       prev = new;
+
+                       /* Insert the link descriptor to the LD ring */
+                       list_add_tail(&new->node, &first->tx_list);
+
+                       /* update metadata */
+                       addr += len;
+                       avail -= len;
+               } while (avail);
+       }
+
+       first->async_tx.cookie = -EBUSY;
+       first->async_tx.flags = flags;
+
+       /* last desc and fire IRQ */
+       new->desc.ddadr = DDADR_STOP;
+       new->desc.dcmd |= DCMD_ENDIRQEN;
+
+       chan->dir = dir;
+       chan->cyclic_first = NULL;
+
+       return &first->async_tx;
+
+fail:
+       if (first)
+               mmp_pdma_free_desc_list(chan, &first->tx_list);
+       return NULL;
+}
+
+static struct dma_async_tx_descriptor *
+mmp_pdma_prep_dma_cyclic(struct dma_chan *dchan,
+                        dma_addr_t buf_addr, size_t len, size_t period_len,
+                        enum dma_transfer_direction direction,
+                        unsigned long flags)
+{
+       struct mmp_pdma_chan *chan;
+       struct mmp_pdma_desc_sw *first = NULL, *prev = NULL, *new;
+       dma_addr_t dma_src, dma_dst;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+       dma_addr_t dma_srch, dma_dsth;
+#endif
+
+       if (!dchan || !len || !period_len)
+               return NULL;
+
+       /* the buffer length must be a multiple of period_len */
+       if (len % period_len != 0)
+               return NULL;
+
+       if (period_len > PDMA_MAX_DESC_BYTES)
+               return NULL;
+
+       chan = to_mmp_pdma_chan(dchan);
+       mmp_pdma_config_write(dchan, &chan->slave_config, direction);
+
+       switch (direction) {
+       case DMA_MEM_TO_DEV:
+               dma_src = buf_addr & 0xffffffff;
+               dma_dst = chan->dev_addr;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+               dma_srch = (buf_addr >> 32);
+               dma_dsth = 0;
+#endif
+               break;
+       case DMA_DEV_TO_MEM:
+               dma_dst = buf_addr & 0xffffffff;
+               dma_src = chan->dev_addr;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+               dma_dsth = (buf_addr >> 32);
+               dma_srch = 0;
+#endif
+               break;
+       default:
+               dev_err(chan->dev, "Unsupported direction for cyclic DMA\n");
+               return NULL;
+       }
+
+       chan->dir = direction;
+
+       do {
+               /* Allocate the link descriptor from DMA pool */
+               new = mmp_pdma_alloc_descriptor(chan);
+               if (!new) {
+                       dev_err(chan->dev, "no memory for desc\n");
+                       goto fail;
+               }
+
+               new->desc.dcmd = (chan->dcmd | DCMD_ENDIRQEN |
+                                 (DCMD_LENGTH & period_len));
+               new->desc.dsadr = dma_src;
+               new->desc.dtadr = dma_dst;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+               new->desc.dsadrh = dma_dsth;
+               new->desc.dtadrh = dma_srch;
+#endif
+
+               if (!first)
+                       first = new;
+               else {
+                       prev->desc.ddadr = new->async_tx.phys;
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+                       prev->desc.ddadrh = (new->async_tx.phys >> 32);
+#endif
+               }
+
+               new->async_tx.cookie = 0;
+               async_tx_ack(&new->async_tx);
+
+               prev = new;
+               len -= period_len;
+
+               if (chan->dir == DMA_MEM_TO_DEV)
+                       dma_src += period_len;
+               else
+                       dma_dst += period_len;
+
+               /* Insert the link descriptor to the LD ring */
+               list_add_tail(&new->node, &first->tx_list);
+       } while (len);
+
+       first->async_tx.flags = flags; /* client is in control of this ack */
+       first->async_tx.cookie = -EBUSY;
+
+       /* make the cyclic link */
+       new->desc.ddadr = first->async_tx.phys;
+       chan->cyclic_first = first;
+
+       return &first->async_tx;
+
+fail:
+       if (first)
+               mmp_pdma_free_desc_list(chan, &first->tx_list);
+       return NULL;
+}
+
+static int mmp_pdma_config_write(struct dma_chan *dchan,
+                          struct dma_slave_config *cfg,
+                          enum dma_transfer_direction direction)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+       u32 maxburst = 0, addr = 0;
+       enum dma_slave_buswidth width = DMA_SLAVE_BUSWIDTH_UNDEFINED;
+
+       if (!dchan)
+               return -EINVAL;
+
+       if (direction == DMA_DEV_TO_MEM) {
+               chan->dcmd = DCMD_INCTRGADDR | DCMD_FLOWSRC;
+               maxburst = cfg->src_maxburst;
+               width = cfg->src_addr_width;
+               addr = cfg->src_addr;
+       } else if (direction == DMA_MEM_TO_DEV) {
+               chan->dcmd = DCMD_INCSRCADDR | DCMD_FLOWTRG;
+               maxburst = cfg->dst_maxburst;
+               width = cfg->dst_addr_width;
+               addr = cfg->dst_addr;
+       }
+
+       if (width == DMA_SLAVE_BUSWIDTH_1_BYTE)
+               chan->dcmd |= DCMD_WIDTH1;
+       else if (width == DMA_SLAVE_BUSWIDTH_2_BYTES)
+               chan->dcmd |= DCMD_WIDTH2;
+       else if (width == DMA_SLAVE_BUSWIDTH_4_BYTES)
+               chan->dcmd |= DCMD_WIDTH4;
+
+       if (maxburst == 8)
+               chan->dcmd |= DCMD_BURST8;
+       else if (maxburst == 16)
+               chan->dcmd |= DCMD_BURST16;
+       else if (maxburst == 32)
+               chan->dcmd |= DCMD_BURST32;
+
+       chan->dir = direction;
+       chan->dev_addr = addr;
+
+       return 0;
+}
+
+static int mmp_pdma_pause_chan(struct dma_chan *dchan)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+
+       if (!chan->phy)
+               return -1;
+
+       disable_chan(chan->phy);
+       chan->status = DMA_PAUSED;
+
+       return 0;
+}
+
+static int mmp_pdma_config(struct dma_chan *dchan,
+                          struct dma_slave_config *cfg)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+
+       memcpy(&chan->slave_config, cfg, sizeof(*cfg));
+       return 0;
+}
+
+static int mmp_pdma_terminate_all(struct dma_chan *dchan)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+       unsigned long flags;
+
+       if (!dchan)
+               return -EINVAL;
+
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       disable_chan(chan->phy);
+       chan->status = DMA_COMPLETE;
+       mmp_pdma_free_phy(chan);
+
+       mmp_pdma_free_desc_list(chan, &chan->chain_pending);
+       mmp_pdma_free_desc_list(chan, &chan->chain_running);
+       chan->bytes_residue = 0;
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+       chan->idle = true;
+
+       mmp_pdma_qos_put(chan);
+
+       return 0;
+}
+
+static unsigned int mmp_pdma_residue(struct mmp_pdma_chan *chan,
+                                    dma_cookie_t cookie)
+{
+       struct mmp_pdma_desc_sw *sw;
+       u32 curr, residue = 0;
+       bool passed = false;
+       bool cyclic = chan->cyclic_first != NULL;
+
+       /*
+        * If the channel does not have a phy pointer anymore, it has already
+        * been completed. Therefore, its residue is 0.
+        */
+       if (!chan->phy)
+               return chan->bytes_residue; /* special case for EORIRQEN */
+
+       if (chan->dir == DMA_DEV_TO_MEM)
+               curr = readl(chan->phy->base + DTADR(chan->phy->idx));
+       else
+               curr = readl(chan->phy->base + DSADR(chan->phy->idx));
+
+       list_for_each_entry(sw, &chan->chain_running, node) {
+               u32 start, end, len;
+
+               if (chan->dir == DMA_DEV_TO_MEM)
+                       start = sw->desc.dtadr;
+               else
+                       start = sw->desc.dsadr;
+
+               len = sw->desc.dcmd & DCMD_LENGTH;
+               end = start + len;
+
+               /*
+                * 'passed' will be latched once we found the descriptor which
+                * lies inside the boundaries of the curr pointer. All
+                * descriptors that occur in the list _after_ we found that
+                * partially handled descriptor are still to be processed and
+                * are hence added to the residual bytes counter.
+                */
+
+               if (passed) {
+                       residue += len;
+               } else if (curr >= start && curr <= end) {
+                       residue += end - curr;
+                       passed = true;
+               }
+
+               /*
+                * Descriptors that have the ENDIRQEN bit set mark the end of a
+                * transaction chain, and the cookie assigned with it has been
+                * returned previously from mmp_pdma_tx_submit().
+                *
+                * In case we have multiple transactions in the running chain,
+                * and the cookie does not match the one the user asked us
+                * about, reset the state variables and start over.
+                *
+                * This logic does not apply to cyclic transactions, where all
+                * descriptors have the ENDIRQEN bit set, and for which we
+                * can't have multiple transactions on one channel anyway.
+                */
+               if (cyclic || !(sw->desc.dcmd & DCMD_ENDIRQEN))
+                       continue;
+
+               if (sw->async_tx.cookie == cookie) {
+                       return residue;
+               } else {
+                       residue = 0;
+                       passed = false;
+               }
+       }
+
+       /* We should only get here in case of cyclic transactions */
+       return residue;
+}
+
+static enum dma_status mmp_pdma_tx_status(struct dma_chan *dchan,
+                                         dma_cookie_t cookie,
+                                         struct dma_tx_state *txstate)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+       enum dma_status ret;
+       unsigned long flags;
+
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       ret = dma_cookie_status(dchan, cookie, txstate);
+       if (likely(ret != DMA_ERROR))
+               dma_set_residue(txstate, mmp_pdma_residue(chan, cookie));
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+       if (ret == DMA_COMPLETE)
+               return ret;
+       else
+               return chan->status;
+}
+
+/*
+ * mmp_pdma_issue_pending - Issue the DMA start command
+ * pending list ==> running list
+ */
+static void mmp_pdma_issue_pending(struct dma_chan *dchan)
+{
+       struct mmp_pdma_chan *chan = to_mmp_pdma_chan(dchan);
+       unsigned long flags;
+       int ret = 0;
+
+       mmp_pdma_qos_get(chan);
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       ret = start_pending_queue(chan);
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+       if (ret)
+               mmp_pdma_qos_put(chan);
+}
+
+/*
+ * dma_do_tasklet
+ * Do call back
+ * Start pending list
+ */
+static void dma_do_tasklet(struct tasklet_struct *t)
+{
+       struct mmp_pdma_chan *chan = from_tasklet(chan, t, tasklet);
+       struct mmp_pdma_desc_sw *desc, *_desc;
+       LIST_HEAD(chain_cleanup);
+       unsigned long flags;
+       struct dmaengine_desc_callback cb;
+
+       int ret = 0;
+
+       /* return if this channel has been stopped */
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       if (chan->status == DMA_COMPLETE) {
+               spin_unlock_irqrestore(&chan->desc_lock, flags);
+               return;
+       }
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+       if (chan->cyclic_first) {
+               spin_lock_irqsave(&chan->desc_lock, flags);
+               desc = chan->cyclic_first;
+               dmaengine_desc_get_callback(&desc->async_tx, &cb);
+               spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+               dmaengine_desc_callback_invoke(&cb, NULL);
+
+               return;
+       }
+
+       /* submit pending list; callback for each desc; free desc */
+       spin_lock_irqsave(&chan->desc_lock, flags);
+
+       /* special for the EORIRQEN case, residue is not 0 */
+       list_for_each_entry(desc, &chan->chain_running, node) {
+               if (desc->desc.dcmd & DCMD_ENDIRQEN) {
+                       chan->bytes_residue =
+                               mmp_pdma_residue(chan, desc->async_tx.cookie);
+                       break;
+               }
+       }
+
+       list_for_each_entry_safe(desc, _desc, &chan->chain_running, node) {
+               /*
+                * move the descriptors to a temporary list so we can drop
+                * the lock during the entire cleanup operation
+                */
+               list_move(&desc->node, &chain_cleanup);
+
+               /*
+                * Look for the first list entry which has the ENDIRQEN flag
+                * set. That is the descriptor we got an interrupt for, so
+                * complete that transaction and its cookie.
+                */
+               if (desc->desc.dcmd & DCMD_ENDIRQEN) {
+                       dma_cookie_t cookie = desc->async_tx.cookie;
+                       dma_cookie_complete(&desc->async_tx);
+                       dev_dbg(chan->dev, "completed_cookie=%d\n", cookie);
+                       break;
+               }
+       }
+
+       /*
+        * The hardware is idle and ready for more when the
+        * chain_running list is empty.
+        */
+       chan->status = list_empty(&chan->chain_running) ?
+               DMA_COMPLETE : DMA_IN_PROGRESS;
+
+       /* Start any pending transactions automatically */
+       ret = start_pending_queue(chan);
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+
+       /* restart pending transactions failed, do not need qos anymore */
+       if (ret)
+               mmp_pdma_qos_put(chan);
+
+       /* Run the callback for each descriptor, in order */
+       list_for_each_entry_safe(desc, _desc, &chain_cleanup, node) {
+               struct dma_async_tx_descriptor *txd = &desc->async_tx;
+
+               /* Remove from the list of transactions */
+               list_del(&desc->node);
+               /* Run the link descriptor callback function */
+               dmaengine_desc_get_callback(txd, &cb);
+               dmaengine_desc_callback_invoke(&cb, NULL);
+
+               dma_pool_free(chan->desc_pool, desc, txd->phys);
+       }
+}
+
+static int mmp_pdma_remove(struct platform_device *op)
+{
+       struct mmp_pdma_device *pdev = platform_get_drvdata(op);
+       struct mmp_pdma_phy *phy;
+       int i, irq = 0, irq_num = 0;
+
+       if (op->dev.of_node)
+               of_dma_controller_free(op->dev.of_node);
+
+       for (i = 0; i < pdev->dma_channels; i++) {
+               if (platform_get_irq(op, i) > 0)
+                       irq_num++;
+       }
+
+       if (irq_num != pdev->dma_channels) {
+               irq = platform_get_irq(op, 0);
+               devm_free_irq(&op->dev, irq, pdev);
+       } else {
+               for (i = 0; i < pdev->dma_channels; i++) {
+                       phy = &pdev->phy[i];
+                       irq = platform_get_irq(op, i);
+                       devm_free_irq(&op->dev, irq, phy);
+               }
+       }
+
+       dma_async_device_unregister(&pdev->device);
+
+       reset_control_assert(pdev->resets);
+       clk_disable_unprepare(pdev->clk);
+
+       kfree(pdev->reserved_channels);
+       platform_set_drvdata(op, NULL);
+
+       return 0;
+}
+
+static int mmp_pdma_chan_init(struct mmp_pdma_device *pdev, int idx, int irq)
+{
+       struct mmp_pdma_phy *phy  = &pdev->phy[idx];
+       struct mmp_pdma_chan *chan;
+       int ret;
+
+       chan = devm_kzalloc(pdev->dev, sizeof(*chan), GFP_KERNEL);
+       if (chan == NULL)
+               return -ENOMEM;
+
+       phy->idx = idx;
+       phy->base = pdev->base;
+
+       if (irq) {
+               ret = devm_request_irq(pdev->dev, irq, mmp_pdma_chan_handler,
+                                      IRQF_SHARED, "pdma", phy);
+               if (ret) {
+                       dev_err(pdev->dev, "channel request irq fail!\n");
+                       return ret;
+               }
+       }
+
+       spin_lock_init(&chan->desc_lock);
+       chan->dev = pdev->dev;
+       chan->chan.device = &pdev->device;
+       tasklet_setup(&chan->tasklet, dma_do_tasklet);
+       INIT_LIST_HEAD(&chan->chain_pending);
+       INIT_LIST_HEAD(&chan->chain_running);
+
+       chan->status = DMA_COMPLETE;
+       chan->bytes_residue = 0;
+       chan->qos_count = 0;
+       chan->user_do_qos = 1;
+
+       /* register virt channel to dma engine */
+       list_add_tail(&chan->chan.device_node, &pdev->device.channels);
+
+       return 0;
+}
+
+static const struct of_device_id mmp_pdma_dt_ids[] = {
+       { .compatible = "spacemit,pdma-1.0", },
+       {}
+};
+MODULE_DEVICE_TABLE(of, mmp_pdma_dt_ids);
+
+static struct dma_chan *mmp_pdma_dma_xlate(struct of_phandle_args *dma_spec,
+                                          struct of_dma *ofdma)
+{
+       struct mmp_pdma_device *d = ofdma->of_dma_data;
+       struct dma_chan *chan;
+#ifdef CONFIG_PM
+       struct mmp_pdma_chan *c;
+#endif
+
+       chan = dma_get_any_slave_channel(&d->device);
+       if (!chan)
+               return NULL;
+
+       to_mmp_pdma_chan(chan)->drcmr = dma_spec->args[0];
+#ifdef CONFIG_PM
+       if (unlikely(dma_spec->args_count != 2))
+               dev_err(d->dev, "#dma-cells should be 2!\n");
+
+       c = to_mmp_pdma_chan(chan);
+       c->user_do_qos = dma_spec->args[1] ? 1 : 0;
+
+       if (c->user_do_qos)
+               dev_dbg(d->dev, "channel %d: user does qos itself\n",
+                        c->chan.chan_id);
+       else
+               dev_dbg(d->dev, "channel %d: pdma does qos\n",
+                        c->chan.chan_id);
+#endif
+
+       return chan;
+}
+
+static int mmp_pdma_probe(struct platform_device *op)
+{
+       struct mmp_pdma_device *pdev;
+       const struct of_device_id *of_id;
+       struct mmp_dma_platdata *pdata = dev_get_platdata(&op->dev);
+       struct resource *iores;
+       int i, ret, irq = 0;
+       int dma_channels = 0, irq_num = 0;
+       const enum dma_slave_buswidth widths =
+               DMA_SLAVE_BUSWIDTH_1_BYTE   | DMA_SLAVE_BUSWIDTH_2_BYTES |
+               DMA_SLAVE_BUSWIDTH_4_BYTES;
+
+       int nr_reserved_channels;
+       const int *list;
+       unsigned int max_burst_size = DEFAULT_MAX_BURST_SIZE;
+
+       pdev = devm_kzalloc(&op->dev, sizeof(*pdev), GFP_KERNEL);
+       if (!pdev)
+               return -ENOMEM;
+
+       pdev->dev = &op->dev;
+
+       spin_lock_init(&pdev->phy_lock);
+
+       iores = platform_get_resource(op, IORESOURCE_MEM, 0);
+       pdev->base = devm_ioremap_resource(pdev->dev, iores);
+       if (IS_ERR(pdev->base))
+               return PTR_ERR(pdev->base);
+
+       pdev->clk = devm_clk_get(pdev->dev,NULL);
+       if(IS_ERR(pdev->clk))
+               return PTR_ERR(pdev->clk);
+
+       ret = clk_prepare_enable(pdev->clk);
+       if (ret)
+               return dev_err_probe(pdev->dev, ret, "could not enable dma bus clock\n");
+
+       pdev->resets = devm_reset_control_get_optional(pdev->dev,NULL);
+       if(IS_ERR(pdev->resets)) {
+               ret = PTR_ERR(pdev->resets);
+               goto err_rst;
+       }
+       ret = reset_control_deassert(pdev->resets);
+       if(ret)
+               goto err_rst;
+
+       of_id = of_match_device(mmp_pdma_dt_ids, pdev->dev);
+
+       if (of_id) {
+               int n;
+               of_property_read_u32(pdev->dev->of_node, "#dma-channels",
+                               &dma_channels);
+
+               list = of_get_property(pdev->dev->of_node, "reserved-channels",
+                       &n);
+
+               if (of_property_read_u32(pdev->dev->of_node, "max-burst-size",
+                   &max_burst_size)) {
+                       dev_err(pdev->dev, "Cannot find the max-burst-size node "
+                                      "in the device tree, set it to %d\n",
+                                      DEFAULT_MAX_BURST_SIZE);
+                       max_burst_size = DEFAULT_MAX_BURST_SIZE;
+               }
+
+               if (get_max_burst_setting(max_burst_size) == INVALID_BURST_SETTING) {
+                       dev_err(pdev->dev, "Unsupported max-burst-size value %d "
+                                      "in the device tree, set it to %d\n",
+                                       max_burst_size, DEFAULT_MAX_BURST_SIZE);
+                       max_burst_size = DEFAULT_MAX_BURST_SIZE;
+               }
+
+               if (list) {
+                       int num_args = 2;
+
+                       nr_reserved_channels = n / (sizeof(u32) * num_args);
+
+                       pdev->nr_reserved_channels = nr_reserved_channels;
+
+                       pdev->reserved_channels = kzalloc(nr_reserved_channels * sizeof(struct reserved_chan),
+                                                       GFP_KERNEL);
+
+                       if (pdev->reserved_channels == NULL)
+                               return -ENOMEM;
+
+                       for (i = 0; i < nr_reserved_channels; i++) {
+                               int value;
+
+                               of_property_read_u32_index(pdev->dev->of_node, "reserved-channels", i * num_args, &value);
+                               pdev->reserved_channels[i].chan_id = value;
+                               of_property_read_u32_index(pdev->dev->of_node, "reserved-channels", i * num_args + 1, &value);
+                               pdev->reserved_channels[i].drcmr   = value;
+                       }
+               }
+       } else if (pdata && pdata->dma_channels) {
+               dma_channels = pdata->dma_channels;
+       } else {
+               dma_channels = 32;      /* default 32 channel */
+       }
+       pdev->dma_channels = dma_channels;
+
+       pdev->max_burst_size = max_burst_size;
+       dev_dbg(pdev->dev, "set max burst size to %d\n", max_burst_size);
+
+#ifdef CONFIG_PM
+       pm_runtime_enable(&op->dev);
+       /*
+        * We can't ensure the pm operations are always in non-atomic context.
+        * Actually it depends on the drivers' behavior. So mark it as irq safe.
+        */
+       pm_runtime_irq_safe(&op->dev);
+#endif
+       for (i = 0; i < dma_channels; i++) {
+               if (platform_get_irq_optional(op, i) > 0)
+                       irq_num++;
+       }
+
+       pdev->phy = devm_kcalloc(pdev->dev, dma_channels, sizeof(*pdev->phy),
+                                GFP_KERNEL);
+       if (pdev->phy == NULL)
+               return -ENOMEM;
+
+       INIT_LIST_HEAD(&pdev->device.channels);
+
+       if (irq_num != dma_channels) {
+               /* all chan share one irq, demux inside */
+               irq = platform_get_irq(op, 0);
+               ret = devm_request_irq(pdev->dev, irq, mmp_pdma_int_handler,
+                                      IRQF_SHARED, "pdma", pdev);
+               if (ret)
+                       return ret;
+       }
+
+       for (i = 0; i < dma_channels; i++) {
+               irq = (irq_num != dma_channels) ? 0 : platform_get_irq(op, i);
+               ret = mmp_pdma_chan_init(pdev, i, irq);
+               if (ret)
+                       return ret;
+       }
+
+       dma_cap_set(DMA_SLAVE, pdev->device.cap_mask);
+       dma_cap_set(DMA_MEMCPY, pdev->device.cap_mask);
+       dma_cap_set(DMA_CYCLIC, pdev->device.cap_mask);
+       dma_cap_set(DMA_PRIVATE, pdev->device.cap_mask);
+       pdev->device.dev = &op->dev;
+       pdev->device.device_alloc_chan_resources = mmp_pdma_alloc_chan_resources;
+       pdev->device.device_free_chan_resources = mmp_pdma_free_chan_resources;
+       pdev->device.device_tx_status = mmp_pdma_tx_status;
+       pdev->device.device_prep_dma_memcpy = mmp_pdma_prep_memcpy;
+       pdev->device.device_prep_slave_sg = mmp_pdma_prep_slave_sg;
+       pdev->device.device_prep_dma_cyclic = mmp_pdma_prep_dma_cyclic;
+       pdev->device.device_issue_pending = mmp_pdma_issue_pending;
+       pdev->device.device_config = mmp_pdma_config;
+       pdev->device.device_pause = mmp_pdma_pause_chan;
+       pdev->device.device_terminate_all = mmp_pdma_terminate_all;
+       pdev->device.copy_align = DMAENGINE_ALIGN_8_BYTES;
+       pdev->device.src_addr_widths = widths;
+       pdev->device.dst_addr_widths = widths;
+       pdev->device.directions = BIT(DMA_MEM_TO_DEV) | BIT(DMA_DEV_TO_MEM);
+       pdev->device.residue_granularity = DMA_RESIDUE_GRANULARITY_DESCRIPTOR;
+
+#ifdef CONFIG_SPACEMIT_PDMA_SUPPORT_64BIT
+       dma_set_mask(pdev->dev, DMA_BIT_MASK(64));
+#else
+       dma_set_mask(pdev->dev, pdev->dev->coherent_dma_mask);
+#endif
+
+       ret = dma_async_device_register(&pdev->device);
+       if (ret) {
+               dev_err(pdev->device.dev, "unable to register\n");
+               return ret;
+       }
+
+       if (op->dev.of_node) {
+               /* Device-tree DMA controller registration */
+               ret = of_dma_controller_register(op->dev.of_node,
+                                                mmp_pdma_dma_xlate, pdev);
+               if (ret < 0) {
+                       dev_err(&op->dev, "of_dma_controller_register failed\n");
+                       dma_async_device_unregister(&pdev->device);
+                       return ret;
+               }
+       }
+
+       platform_set_drvdata(op, pdev);
+       dev_dbg(pdev->device.dev, "initialized %d channels\n", dma_channels);
+
+       return 0;
+
+err_rst:
+       clk_disable_unprepare(pdev->clk);
+       return ret;
+}
+
+/*
+ * Per-channel qos get/put function. This function ensures that pm_
+ * runtime_get/put are not called multi times for one channel.
+ * This guarantees pm_runtime_get/put always match for the entire device.
+ */
+static void mmp_pdma_qos_get(struct mmp_pdma_chan *chan)
+{
+       unsigned long flags;
+
+       if (chan->user_do_qos)
+               return;
+
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       if (chan->qos_count == 0) {
+               chan->qos_count = 1;
+               /*
+                * Safe in spin_lock because it's marked as irq safe.
+                * Similar case for mmp_pdma_qos_put().
+                */
+               pm_runtime_get_sync(chan->dev);
+       }
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+}
+
+static void mmp_pdma_qos_put(struct mmp_pdma_chan *chan)
+{
+       unsigned long flags;
+
+       if (chan->user_do_qos)
+               return;
+
+       spin_lock_irqsave(&chan->desc_lock, flags);
+       if (chan->qos_count == 1) {
+               chan->qos_count = 0;
+               pm_runtime_put_autosuspend(chan->dev);
+       }
+
+       spin_unlock_irqrestore(&chan->desc_lock, flags);
+}
+
+static const struct platform_device_id mmp_pdma_id_table[] = {
+       { "mmp-pdma", },
+       { },
+};
+
+#ifdef CONFIG_PM_SLEEP
+static int mmp_pdma_suspend_noirq(struct device *dev)
+{
+       struct mmp_pdma_device *pdev = dev_get_drvdata(dev);
+
+       clk_disable_unprepare(pdev->clk);
+
+       return 0;
+}
+
+static int mmp_pdma_resume_noirq(struct device *dev)
+{
+       struct mmp_pdma_device *pdev = dev_get_drvdata(dev);
+
+       clk_prepare_enable(pdev->clk);
+
+       return 0;
+}
+
+static const struct dev_pm_ops k1x_mmp_pdma_pm_qos = {
+       .suspend_noirq = mmp_pdma_suspend_noirq,
+       .resume_noirq = mmp_pdma_resume_noirq,
+};
+#endif
+
+static struct platform_driver mmp_pdma_driver = {
+       .driver         = {
+               .name   = "mmp-pdma",
+#ifdef CONFIG_PM_SLEEP
+               .pm     = &k1x_mmp_pdma_pm_qos,
+#endif
+               .of_match_table = mmp_pdma_dt_ids,
+       },
+       .id_table       = mmp_pdma_id_table,
+       .probe          = mmp_pdma_probe,
+       .remove         = mmp_pdma_remove,
+};
+
+static int __init mmp_pdma_init(void)
+{
+       return platform_driver_register(&mmp_pdma_driver);
+}
+
+static void __exit mmp_pdma_exit(void)
+{
+       platform_driver_unregister(&mmp_pdma_driver);
+}
+
+subsys_initcall(mmp_pdma_init);
+module_exit(mmp_pdma_exit);
+
+MODULE_DESCRIPTION("MARVELL MMP Peripheral DMA Driver");
+MODULE_AUTHOR("Marvell International Ltd.");
+MODULE_LICENSE("GPL v2");