These changes are the raw update to linux-4.4.6-rt14. Kernel sources
[kvmfornfv.git] / kernel / drivers / dma / edma.c
index bf09db7..16fe773 100644 (file)
 #include <linux/slab.h>
 #include <linux/spinlock.h>
 #include <linux/of.h>
+#include <linux/of_dma.h>
+#include <linux/of_irq.h>
+#include <linux/of_address.h>
+#include <linux/of_device.h>
+#include <linux/pm_runtime.h>
 
 #include <linux/platform_data/edma.h>
 
 #include "dmaengine.h"
 #include "virt-dma.h"
 
-/*
- * This will go away when the private EDMA API is folded
- * into this driver and the platform device(s) are
- * instantiated in the arch code. We can only get away
- * with this simplification because DA8XX may not be built
- * in the same kernel image with other DaVinci parts. This
- * avoids having to sprinkle dmaengine driver platform devices
- * and data throughout all the existing board files.
- */
-#ifdef CONFIG_ARCH_DAVINCI_DA8XX
-#define EDMA_CTLRS     2
-#define EDMA_CHANS     32
-#else
-#define EDMA_CTLRS     1
-#define EDMA_CHANS     64
-#endif /* CONFIG_ARCH_DAVINCI_DA8XX */
+/* Offsets matching "struct edmacc_param" */
+#define PARM_OPT               0x00
+#define PARM_SRC               0x04
+#define PARM_A_B_CNT           0x08
+#define PARM_DST               0x0c
+#define PARM_SRC_DST_BIDX      0x10
+#define PARM_LINK_BCNTRLD      0x14
+#define PARM_SRC_DST_CIDX      0x18
+#define PARM_CCNT              0x1c
+
+#define PARM_SIZE              0x20
+
+/* Offsets for EDMA CC global channel registers and their shadows */
+#define SH_ER                  0x00    /* 64 bits */
+#define SH_ECR                 0x08    /* 64 bits */
+#define SH_ESR                 0x10    /* 64 bits */
+#define SH_CER                 0x18    /* 64 bits */
+#define SH_EER                 0x20    /* 64 bits */
+#define SH_EECR                        0x28    /* 64 bits */
+#define SH_EESR                        0x30    /* 64 bits */
+#define SH_SER                 0x38    /* 64 bits */
+#define SH_SECR                        0x40    /* 64 bits */
+#define SH_IER                 0x50    /* 64 bits */
+#define SH_IECR                        0x58    /* 64 bits */
+#define SH_IESR                        0x60    /* 64 bits */
+#define SH_IPR                 0x68    /* 64 bits */
+#define SH_ICR                 0x70    /* 64 bits */
+#define SH_IEVAL               0x78
+#define SH_QER                 0x80
+#define SH_QEER                        0x84
+#define SH_QEECR               0x88
+#define SH_QEESR               0x8c
+#define SH_QSER                        0x90
+#define SH_QSECR               0x94
+#define SH_SIZE                        0x200
+
+/* Offsets for EDMA CC global registers */
+#define EDMA_REV               0x0000
+#define EDMA_CCCFG             0x0004
+#define EDMA_QCHMAP            0x0200  /* 8 registers */
+#define EDMA_DMAQNUM           0x0240  /* 8 registers (4 on OMAP-L1xx) */
+#define EDMA_QDMAQNUM          0x0260
+#define EDMA_QUETCMAP          0x0280
+#define EDMA_QUEPRI            0x0284
+#define EDMA_EMR               0x0300  /* 64 bits */
+#define EDMA_EMCR              0x0308  /* 64 bits */
+#define EDMA_QEMR              0x0310
+#define EDMA_QEMCR             0x0314
+#define EDMA_CCERR             0x0318
+#define EDMA_CCERRCLR          0x031c
+#define EDMA_EEVAL             0x0320
+#define EDMA_DRAE              0x0340  /* 4 x 64 bits*/
+#define EDMA_QRAE              0x0380  /* 4 registers */
+#define EDMA_QUEEVTENTRY       0x0400  /* 2 x 16 registers */
+#define EDMA_QSTAT             0x0600  /* 2 registers */
+#define EDMA_QWMTHRA           0x0620
+#define EDMA_QWMTHRB           0x0624
+#define EDMA_CCSTAT            0x0640
+
+#define EDMA_M                 0x1000  /* global channel registers */
+#define EDMA_ECR               0x1008
+#define EDMA_ECRH              0x100C
+#define EDMA_SHADOW0           0x2000  /* 4 shadow regions */
+#define EDMA_PARM              0x4000  /* PaRAM entries */
+
+#define PARM_OFFSET(param_no)  (EDMA_PARM + ((param_no) << 5))
+
+#define EDMA_DCHMAP            0x0100  /* 64 registers */
+
+/* CCCFG register */
+#define GET_NUM_DMACH(x)       (x & 0x7) /* bits 0-2 */
+#define GET_NUM_QDMACH(x)      ((x & 0x70) >> 4) /* bits 4-6 */
+#define GET_NUM_PAENTRY(x)     ((x & 0x7000) >> 12) /* bits 12-14 */
+#define GET_NUM_EVQUE(x)       ((x & 0x70000) >> 16) /* bits 16-18 */
+#define GET_NUM_REGN(x)                ((x & 0x300000) >> 20) /* bits 20-21 */
+#define CHMAP_EXIST            BIT(24)
 
 /*
  * Max of 20 segments per channel to conserve PaRAM slots
 #define EDMA_MAX_SLOTS         MAX_NR_SG
 #define EDMA_DESCRIPTORS       16
 
+#define EDMA_CHANNEL_ANY               -1      /* for edma_alloc_channel() */
+#define EDMA_SLOT_ANY                  -1      /* for edma_alloc_slot() */
+#define EDMA_CONT_PARAMS_ANY            1001
+#define EDMA_CONT_PARAMS_FIXED_EXACT    1002
+#define EDMA_CONT_PARAMS_FIXED_NOT_EXACT 1003
+
+/* PaRAM slots are laid out like this */
+struct edmacc_param {
+       u32 opt;
+       u32 src;
+       u32 a_b_cnt;
+       u32 dst;
+       u32 src_dst_bidx;
+       u32 link_bcntrld;
+       u32 src_dst_cidx;
+       u32 ccnt;
+} __packed;
+
+/* fields in edmacc_param.opt */
+#define SAM            BIT(0)
+#define DAM            BIT(1)
+#define SYNCDIM                BIT(2)
+#define STATIC         BIT(3)
+#define EDMA_FWID      (0x07 << 8)
+#define TCCMODE                BIT(11)
+#define EDMA_TCC(t)    ((t) << 12)
+#define TCINTEN                BIT(20)
+#define ITCINTEN       BIT(21)
+#define TCCHEN         BIT(22)
+#define ITCCHEN                BIT(23)
+
 struct edma_pset {
        u32                             len;
        dma_addr_t                      addr;
@@ -105,26 +201,524 @@ struct edma_desc {
 
 struct edma_cc;
 
+struct edma_tc {
+       struct device_node              *node;
+       u16                             id;
+};
+
 struct edma_chan {
        struct virt_dma_chan            vchan;
        struct list_head                node;
        struct edma_desc                *edesc;
        struct edma_cc                  *ecc;
+       struct edma_tc                  *tc;
        int                             ch_num;
        bool                            alloced;
+       bool                            hw_triggered;
        int                             slot[EDMA_MAX_SLOTS];
        int                             missed;
        struct dma_slave_config         cfg;
 };
 
 struct edma_cc {
-       int                             ctlr;
+       struct device                   *dev;
+       struct edma_soc_info            *info;
+       void __iomem                    *base;
+       int                             id;
+       bool                            legacy_mode;
+
+       /* eDMA3 resource information */
+       unsigned                        num_channels;
+       unsigned                        num_qchannels;
+       unsigned                        num_region;
+       unsigned                        num_slots;
+       unsigned                        num_tc;
+       bool                            chmap_exist;
+       enum dma_event_q                default_queue;
+
+       /*
+        * The slot_inuse bit for each PaRAM slot is clear unless the slot is
+        * in use by Linux or if it is allocated to be used by DSP.
+        */
+       unsigned long *slot_inuse;
+
        struct dma_device               dma_slave;
-       struct edma_chan                slave_chans[EDMA_CHANS];
-       int                             num_slave_chans;
+       struct dma_device               *dma_memcpy;
+       struct edma_chan                *slave_chans;
+       struct edma_tc                  *tc_list;
        int                             dummy_slot;
 };
 
+/* dummy param set used to (re)initialize parameter RAM slots */
+static const struct edmacc_param dummy_paramset = {
+       .link_bcntrld = 0xffff,
+       .ccnt = 1,
+};
+
+#define EDMA_BINDING_LEGACY    0
+#define EDMA_BINDING_TPCC      1
+static const struct of_device_id edma_of_ids[] = {
+       {
+               .compatible = "ti,edma3",
+               .data = (void *)EDMA_BINDING_LEGACY,
+       },
+       {
+               .compatible = "ti,edma3-tpcc",
+               .data = (void *)EDMA_BINDING_TPCC,
+       },
+       {}
+};
+
+static const struct of_device_id edma_tptc_of_ids[] = {
+       { .compatible = "ti,edma3-tptc", },
+       {}
+};
+
+static inline unsigned int edma_read(struct edma_cc *ecc, int offset)
+{
+       return (unsigned int)__raw_readl(ecc->base + offset);
+}
+
+static inline void edma_write(struct edma_cc *ecc, int offset, int val)
+{
+       __raw_writel(val, ecc->base + offset);
+}
+
+static inline void edma_modify(struct edma_cc *ecc, int offset, unsigned and,
+                              unsigned or)
+{
+       unsigned val = edma_read(ecc, offset);
+
+       val &= and;
+       val |= or;
+       edma_write(ecc, offset, val);
+}
+
+static inline void edma_and(struct edma_cc *ecc, int offset, unsigned and)
+{
+       unsigned val = edma_read(ecc, offset);
+
+       val &= and;
+       edma_write(ecc, offset, val);
+}
+
+static inline void edma_or(struct edma_cc *ecc, int offset, unsigned or)
+{
+       unsigned val = edma_read(ecc, offset);
+
+       val |= or;
+       edma_write(ecc, offset, val);
+}
+
+static inline unsigned int edma_read_array(struct edma_cc *ecc, int offset,
+                                          int i)
+{
+       return edma_read(ecc, offset + (i << 2));
+}
+
+static inline void edma_write_array(struct edma_cc *ecc, int offset, int i,
+                                   unsigned val)
+{
+       edma_write(ecc, offset + (i << 2), val);
+}
+
+static inline void edma_modify_array(struct edma_cc *ecc, int offset, int i,
+                                    unsigned and, unsigned or)
+{
+       edma_modify(ecc, offset + (i << 2), and, or);
+}
+
+static inline void edma_or_array(struct edma_cc *ecc, int offset, int i,
+                                unsigned or)
+{
+       edma_or(ecc, offset + (i << 2), or);
+}
+
+static inline void edma_or_array2(struct edma_cc *ecc, int offset, int i, int j,
+                                 unsigned or)
+{
+       edma_or(ecc, offset + ((i * 2 + j) << 2), or);
+}
+
+static inline void edma_write_array2(struct edma_cc *ecc, int offset, int i,
+                                    int j, unsigned val)
+{
+       edma_write(ecc, offset + ((i * 2 + j) << 2), val);
+}
+
+static inline unsigned int edma_shadow0_read(struct edma_cc *ecc, int offset)
+{
+       return edma_read(ecc, EDMA_SHADOW0 + offset);
+}
+
+static inline unsigned int edma_shadow0_read_array(struct edma_cc *ecc,
+                                                  int offset, int i)
+{
+       return edma_read(ecc, EDMA_SHADOW0 + offset + (i << 2));
+}
+
+static inline void edma_shadow0_write(struct edma_cc *ecc, int offset,
+                                     unsigned val)
+{
+       edma_write(ecc, EDMA_SHADOW0 + offset, val);
+}
+
+static inline void edma_shadow0_write_array(struct edma_cc *ecc, int offset,
+                                           int i, unsigned val)
+{
+       edma_write(ecc, EDMA_SHADOW0 + offset + (i << 2), val);
+}
+
+static inline unsigned int edma_param_read(struct edma_cc *ecc, int offset,
+                                          int param_no)
+{
+       return edma_read(ecc, EDMA_PARM + offset + (param_no << 5));
+}
+
+static inline void edma_param_write(struct edma_cc *ecc, int offset,
+                                   int param_no, unsigned val)
+{
+       edma_write(ecc, EDMA_PARM + offset + (param_no << 5), val);
+}
+
+static inline void edma_param_modify(struct edma_cc *ecc, int offset,
+                                    int param_no, unsigned and, unsigned or)
+{
+       edma_modify(ecc, EDMA_PARM + offset + (param_no << 5), and, or);
+}
+
+static inline void edma_param_and(struct edma_cc *ecc, int offset, int param_no,
+                                 unsigned and)
+{
+       edma_and(ecc, EDMA_PARM + offset + (param_no << 5), and);
+}
+
+static inline void edma_param_or(struct edma_cc *ecc, int offset, int param_no,
+                                unsigned or)
+{
+       edma_or(ecc, EDMA_PARM + offset + (param_no << 5), or);
+}
+
+static inline void set_bits(int offset, int len, unsigned long *p)
+{
+       for (; len > 0; len--)
+               set_bit(offset + (len - 1), p);
+}
+
+static inline void clear_bits(int offset, int len, unsigned long *p)
+{
+       for (; len > 0; len--)
+               clear_bit(offset + (len - 1), p);
+}
+
+static void edma_assign_priority_to_queue(struct edma_cc *ecc, int queue_no,
+                                         int priority)
+{
+       int bit = queue_no * 4;
+
+       edma_modify(ecc, EDMA_QUEPRI, ~(0x7 << bit), ((priority & 0x7) << bit));
+}
+
+static void edma_set_chmap(struct edma_chan *echan, int slot)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+
+       if (ecc->chmap_exist) {
+               slot = EDMA_CHAN_SLOT(slot);
+               edma_write_array(ecc, EDMA_DCHMAP, channel, (slot << 5));
+       }
+}
+
+static void edma_setup_interrupt(struct edma_chan *echan, bool enable)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+
+       if (enable) {
+               edma_shadow0_write_array(ecc, SH_ICR, channel >> 5,
+                                        BIT(channel & 0x1f));
+               edma_shadow0_write_array(ecc, SH_IESR, channel >> 5,
+                                        BIT(channel & 0x1f));
+       } else {
+               edma_shadow0_write_array(ecc, SH_IECR, channel >> 5,
+                                        BIT(channel & 0x1f));
+       }
+}
+
+/*
+ * paRAM slot management functions
+ */
+static void edma_write_slot(struct edma_cc *ecc, unsigned slot,
+                           const struct edmacc_param *param)
+{
+       slot = EDMA_CHAN_SLOT(slot);
+       if (slot >= ecc->num_slots)
+               return;
+       memcpy_toio(ecc->base + PARM_OFFSET(slot), param, PARM_SIZE);
+}
+
+static void edma_read_slot(struct edma_cc *ecc, unsigned slot,
+                          struct edmacc_param *param)
+{
+       slot = EDMA_CHAN_SLOT(slot);
+       if (slot >= ecc->num_slots)
+               return;
+       memcpy_fromio(param, ecc->base + PARM_OFFSET(slot), PARM_SIZE);
+}
+
+/**
+ * edma_alloc_slot - allocate DMA parameter RAM
+ * @ecc: pointer to edma_cc struct
+ * @slot: specific slot to allocate; negative for "any unused slot"
+ *
+ * This allocates a parameter RAM slot, initializing it to hold a
+ * dummy transfer.  Slots allocated using this routine have not been
+ * mapped to a hardware DMA channel, and will normally be used by
+ * linking to them from a slot associated with a DMA channel.
+ *
+ * Normal use is to pass EDMA_SLOT_ANY as the @slot, but specific
+ * slots may be allocated on behalf of DSP firmware.
+ *
+ * Returns the number of the slot, else negative errno.
+ */
+static int edma_alloc_slot(struct edma_cc *ecc, int slot)
+{
+       if (slot > 0) {
+               slot = EDMA_CHAN_SLOT(slot);
+               /* Requesting entry paRAM slot for a HW triggered channel. */
+               if (ecc->chmap_exist && slot < ecc->num_channels)
+                       slot = EDMA_SLOT_ANY;
+       }
+
+       if (slot < 0) {
+               if (ecc->chmap_exist)
+                       slot = 0;
+               else
+                       slot = ecc->num_channels;
+               for (;;) {
+                       slot = find_next_zero_bit(ecc->slot_inuse,
+                                                 ecc->num_slots,
+                                                 slot);
+                       if (slot == ecc->num_slots)
+                               return -ENOMEM;
+                       if (!test_and_set_bit(slot, ecc->slot_inuse))
+                               break;
+               }
+       } else if (slot >= ecc->num_slots) {
+               return -EINVAL;
+       } else if (test_and_set_bit(slot, ecc->slot_inuse)) {
+               return -EBUSY;
+       }
+
+       edma_write_slot(ecc, slot, &dummy_paramset);
+
+       return EDMA_CTLR_CHAN(ecc->id, slot);
+}
+
+static void edma_free_slot(struct edma_cc *ecc, unsigned slot)
+{
+       slot = EDMA_CHAN_SLOT(slot);
+       if (slot >= ecc->num_slots)
+               return;
+
+       edma_write_slot(ecc, slot, &dummy_paramset);
+       clear_bit(slot, ecc->slot_inuse);
+}
+
+/**
+ * edma_link - link one parameter RAM slot to another
+ * @ecc: pointer to edma_cc struct
+ * @from: parameter RAM slot originating the link
+ * @to: parameter RAM slot which is the link target
+ *
+ * The originating slot should not be part of any active DMA transfer.
+ */
+static void edma_link(struct edma_cc *ecc, unsigned from, unsigned to)
+{
+       if (unlikely(EDMA_CTLR(from) != EDMA_CTLR(to)))
+               dev_warn(ecc->dev, "Ignoring eDMA instance for linking\n");
+
+       from = EDMA_CHAN_SLOT(from);
+       to = EDMA_CHAN_SLOT(to);
+       if (from >= ecc->num_slots || to >= ecc->num_slots)
+               return;
+
+       edma_param_modify(ecc, PARM_LINK_BCNTRLD, from, 0xffff0000,
+                         PARM_OFFSET(to));
+}
+
+/**
+ * edma_get_position - returns the current transfer point
+ * @ecc: pointer to edma_cc struct
+ * @slot: parameter RAM slot being examined
+ * @dst:  true selects the dest position, false the source
+ *
+ * Returns the position of the current active slot
+ */
+static dma_addr_t edma_get_position(struct edma_cc *ecc, unsigned slot,
+                                   bool dst)
+{
+       u32 offs;
+
+       slot = EDMA_CHAN_SLOT(slot);
+       offs = PARM_OFFSET(slot);
+       offs += dst ? PARM_DST : PARM_SRC;
+
+       return edma_read(ecc, offs);
+}
+
+/*
+ * Channels with event associations will be triggered by their hardware
+ * events, and channels without such associations will be triggered by
+ * software.  (At this writing there is no interface for using software
+ * triggers except with channels that don't support hardware triggers.)
+ */
+static void edma_start(struct edma_chan *echan)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       int j = (channel >> 5);
+       unsigned int mask = BIT(channel & 0x1f);
+
+       if (!echan->hw_triggered) {
+               /* EDMA channels without event association */
+               dev_dbg(ecc->dev, "ESR%d %08x\n", j,
+                       edma_shadow0_read_array(ecc, SH_ESR, j));
+               edma_shadow0_write_array(ecc, SH_ESR, j, mask);
+       } else {
+               /* EDMA channel with event association */
+               dev_dbg(ecc->dev, "ER%d %08x\n", j,
+                       edma_shadow0_read_array(ecc, SH_ER, j));
+               /* Clear any pending event or error */
+               edma_write_array(ecc, EDMA_ECR, j, mask);
+               edma_write_array(ecc, EDMA_EMCR, j, mask);
+               /* Clear any SER */
+               edma_shadow0_write_array(ecc, SH_SECR, j, mask);
+               edma_shadow0_write_array(ecc, SH_EESR, j, mask);
+               dev_dbg(ecc->dev, "EER%d %08x\n", j,
+                       edma_shadow0_read_array(ecc, SH_EER, j));
+       }
+}
+
+static void edma_stop(struct edma_chan *echan)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       int j = (channel >> 5);
+       unsigned int mask = BIT(channel & 0x1f);
+
+       edma_shadow0_write_array(ecc, SH_EECR, j, mask);
+       edma_shadow0_write_array(ecc, SH_ECR, j, mask);
+       edma_shadow0_write_array(ecc, SH_SECR, j, mask);
+       edma_write_array(ecc, EDMA_EMCR, j, mask);
+
+       /* clear possibly pending completion interrupt */
+       edma_shadow0_write_array(ecc, SH_ICR, j, mask);
+
+       dev_dbg(ecc->dev, "EER%d %08x\n", j,
+               edma_shadow0_read_array(ecc, SH_EER, j));
+
+       /* REVISIT:  consider guarding against inappropriate event
+        * chaining by overwriting with dummy_paramset.
+        */
+}
+
+/*
+ * Temporarily disable EDMA hardware events on the specified channel,
+ * preventing them from triggering new transfers
+ */
+static void edma_pause(struct edma_chan *echan)
+{
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       unsigned int mask = BIT(channel & 0x1f);
+
+       edma_shadow0_write_array(echan->ecc, SH_EECR, channel >> 5, mask);
+}
+
+/* Re-enable EDMA hardware events on the specified channel.  */
+static void edma_resume(struct edma_chan *echan)
+{
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       unsigned int mask = BIT(channel & 0x1f);
+
+       edma_shadow0_write_array(echan->ecc, SH_EESR, channel >> 5, mask);
+}
+
+static void edma_trigger_channel(struct edma_chan *echan)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       unsigned int mask = BIT(channel & 0x1f);
+
+       edma_shadow0_write_array(ecc, SH_ESR, (channel >> 5), mask);
+
+       dev_dbg(ecc->dev, "ESR%d %08x\n", (channel >> 5),
+               edma_shadow0_read_array(ecc, SH_ESR, (channel >> 5)));
+}
+
+static void edma_clean_channel(struct edma_chan *echan)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       int j = (channel >> 5);
+       unsigned int mask = BIT(channel & 0x1f);
+
+       dev_dbg(ecc->dev, "EMR%d %08x\n", j, edma_read_array(ecc, EDMA_EMR, j));
+       edma_shadow0_write_array(ecc, SH_ECR, j, mask);
+       /* Clear the corresponding EMR bits */
+       edma_write_array(ecc, EDMA_EMCR, j, mask);
+       /* Clear any SER */
+       edma_shadow0_write_array(ecc, SH_SECR, j, mask);
+       edma_write(ecc, EDMA_CCERRCLR, BIT(16) | BIT(1) | BIT(0));
+}
+
+/* Move channel to a specific event queue */
+static void edma_assign_channel_eventq(struct edma_chan *echan,
+                                      enum dma_event_q eventq_no)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+       int bit = (channel & 0x7) * 4;
+
+       /* default to low priority queue */
+       if (eventq_no == EVENTQ_DEFAULT)
+               eventq_no = ecc->default_queue;
+       if (eventq_no >= ecc->num_tc)
+               return;
+
+       eventq_no &= 7;
+       edma_modify_array(ecc, EDMA_DMAQNUM, (channel >> 3), ~(0x7 << bit),
+                         eventq_no << bit);
+}
+
+static int edma_alloc_channel(struct edma_chan *echan,
+                             enum dma_event_q eventq_no)
+{
+       struct edma_cc *ecc = echan->ecc;
+       int channel = EDMA_CHAN_SLOT(echan->ch_num);
+
+       /* ensure access through shadow region 0 */
+       edma_or_array2(ecc, EDMA_DRAE, 0, channel >> 5, BIT(channel & 0x1f));
+
+       /* ensure no events are pending */
+       edma_stop(echan);
+
+       edma_setup_interrupt(echan, true);
+
+       edma_assign_channel_eventq(echan, eventq_no);
+
+       return 0;
+}
+
+static void edma_free_channel(struct edma_chan *echan)
+{
+       /* ensure no events are pending */
+       edma_stop(echan);
+       /* REVISIT should probably take out of shadow region 0 */
+       edma_setup_interrupt(echan, false);
+}
+
 static inline struct edma_cc *to_edma_cc(struct dma_device *d)
 {
        return container_of(d, struct edma_cc, dma_slave);
@@ -135,8 +729,7 @@ static inline struct edma_chan *to_edma_chan(struct dma_chan *c)
        return container_of(c, struct edma_chan, vchan.chan);
 }
 
-static inline struct edma_desc
-*to_edma_desc(struct dma_async_tx_descriptor *tx)
+static inline struct edma_desc *to_edma_desc(struct dma_async_tx_descriptor *tx)
 {
        return container_of(tx, struct edma_desc, vdesc.tx);
 }
@@ -149,20 +742,17 @@ static void edma_desc_free(struct virt_dma_desc *vdesc)
 /* Dispatch a queued descriptor to the controller (caller holds lock) */
 static void edma_execute(struct edma_chan *echan)
 {
+       struct edma_cc *ecc = echan->ecc;
        struct virt_dma_desc *vdesc;
        struct edma_desc *edesc;
        struct device *dev = echan->vchan.chan.device->dev;
        int i, j, left, nslots;
 
-       /* If either we processed all psets or we're still not started */
-       if (!echan->edesc ||
-           echan->edesc->pset_nr == echan->edesc->processed) {
-               /* Get next vdesc */
+       if (!echan->edesc) {
+               /* Setup is needed for the first transfer */
                vdesc = vchan_next_desc(&echan->vchan);
-               if (!vdesc) {
-                       echan->edesc = NULL;
+               if (!vdesc)
                        return;
-               }
                list_del(&vdesc->node);
                echan->edesc = to_edma_desc(&vdesc->tx);
        }
@@ -177,32 +767,32 @@ static void edma_execute(struct edma_chan *echan)
        /* Write descriptor PaRAM set(s) */
        for (i = 0; i < nslots; i++) {
                j = i + edesc->processed;
-               edma_write_slot(echan->slot[i], &edesc->pset[j].param);
+               edma_write_slot(ecc, echan->slot[i], &edesc->pset[j].param);
                edesc->sg_len += edesc->pset[j].len;
-               dev_vdbg(echan->vchan.chan.device->dev,
-                       "\n pset[%d]:\n"
-                       "  chnum\t%d\n"
-                       "  slot\t%d\n"
-                       "  opt\t%08x\n"
-                       "  src\t%08x\n"
-                       "  dst\t%08x\n"
-                       "  abcnt\t%08x\n"
-                       "  ccnt\t%08x\n"
-                       "  bidx\t%08x\n"
-                       "  cidx\t%08x\n"
-                       "  lkrld\t%08x\n",
-                       j, echan->ch_num, echan->slot[i],
-                       edesc->pset[j].param.opt,
-                       edesc->pset[j].param.src,
-                       edesc->pset[j].param.dst,
-                       edesc->pset[j].param.a_b_cnt,
-                       edesc->pset[j].param.ccnt,
-                       edesc->pset[j].param.src_dst_bidx,
-                       edesc->pset[j].param.src_dst_cidx,
-                       edesc->pset[j].param.link_bcntrld);
+               dev_vdbg(dev,
+                        "\n pset[%d]:\n"
+                        "  chnum\t%d\n"
+                        "  slot\t%d\n"
+                        "  opt\t%08x\n"
+                        "  src\t%08x\n"
+                        "  dst\t%08x\n"
+                        "  abcnt\t%08x\n"
+                        "  ccnt\t%08x\n"
+                        "  bidx\t%08x\n"
+                        "  cidx\t%08x\n"
+                        "  lkrld\t%08x\n",
+                        j, echan->ch_num, echan->slot[i],
+                        edesc->pset[j].param.opt,
+                        edesc->pset[j].param.src,
+                        edesc->pset[j].param.dst,
+                        edesc->pset[j].param.a_b_cnt,
+                        edesc->pset[j].param.ccnt,
+                        edesc->pset[j].param.src_dst_bidx,
+                        edesc->pset[j].param.src_dst_cidx,
+                        edesc->pset[j].param.link_bcntrld);
                /* Link to the previous slot if not the last set */
                if (i != (nslots - 1))
-                       edma_link(echan->slot[i], echan->slot[i+1]);
+                       edma_link(ecc, echan->slot[i], echan->slot[i + 1]);
        }
 
        edesc->processed += nslots;
@@ -214,34 +804,32 @@ static void edma_execute(struct edma_chan *echan)
         */
        if (edesc->processed == edesc->pset_nr) {
                if (edesc->cyclic)
-                       edma_link(echan->slot[nslots-1], echan->slot[1]);
+                       edma_link(ecc, echan->slot[nslots - 1], echan->slot[1]);
                else
-                       edma_link(echan->slot[nslots-1],
+                       edma_link(ecc, echan->slot[nslots - 1],
                                  echan->ecc->dummy_slot);
        }
 
-       if (edesc->processed <= MAX_NR_SG) {
+       if (echan->missed) {
+               /*
+                * This happens due to setup times between intermediate
+                * transfers in long SG lists which have to be broken up into
+                * transfers of MAX_NR_SG
+                */
+               dev_dbg(dev, "missed event on channel %d\n", echan->ch_num);
+               edma_clean_channel(echan);
+               edma_stop(echan);
+               edma_start(echan);
+               edma_trigger_channel(echan);
+               echan->missed = 0;
+       } else if (edesc->processed <= MAX_NR_SG) {
                dev_dbg(dev, "first transfer starting on channel %d\n",
                        echan->ch_num);
-               edma_start(echan->ch_num);
+               edma_start(echan);
        } else {
                dev_dbg(dev, "chan: %d: completed %d elements, resuming\n",
                        echan->ch_num, edesc->processed);
-               edma_resume(echan->ch_num);
-       }
-
-       /*
-        * This happens due to setup times between intermediate transfers
-        * in long SG lists which have to be broken up into transfers of
-        * MAX_NR_SG
-        */
-       if (echan->missed) {
-               dev_dbg(dev, "missed event on channel %d\n", echan->ch_num);
-               edma_clean_channel(echan->ch_num);
-               edma_stop(echan->ch_num);
-               edma_start(echan->ch_num);
-               edma_trigger_channel(echan->ch_num);
-               echan->missed = 0;
+               edma_resume(echan);
        }
 }
 
@@ -259,20 +847,16 @@ static int edma_terminate_all(struct dma_chan *chan)
         * echan->edesc is NULL and exit.)
         */
        if (echan->edesc) {
-               int cyclic = echan->edesc->cyclic;
-
+               edma_stop(echan);
+               /* Move the cyclic channel back to default queue */
+               if (!echan->tc && echan->edesc->cyclic)
+                       edma_assign_channel_eventq(echan, EVENTQ_DEFAULT);
                /*
                 * free the running request descriptor
                 * since it is not in any of the vdesc lists
                 */
                edma_desc_free(&echan->edesc->vdesc);
-
                echan->edesc = NULL;
-               edma_stop(echan->ch_num);
-               /* Move the cyclic channel back to default queue */
-               if (cyclic)
-                       edma_assign_channel_eventq(echan->ch_num,
-                                                  EVENTQ_DEFAULT);
        }
 
        vchan_get_all_descriptors(&echan->vchan, &head);
@@ -300,11 +884,10 @@ static int edma_dma_pause(struct dma_chan *chan)
 {
        struct edma_chan *echan = to_edma_chan(chan);
 
-       /* Pause/Resume only allowed with cyclic mode */
-       if (!echan->edesc || !echan->edesc->cyclic)
+       if (!echan->edesc)
                return -EINVAL;
 
-       edma_pause(echan->ch_num);
+       edma_pause(echan);
        return 0;
 }
 
@@ -312,11 +895,7 @@ static int edma_dma_resume(struct dma_chan *chan)
 {
        struct edma_chan *echan = to_edma_chan(chan);
 
-       /* Pause/Resume only allowed with cyclic mode */
-       if (!echan->edesc->cyclic)
-               return -EINVAL;
-
-       edma_resume(echan->ch_num);
+       edma_resume(echan);
        return 0;
 }
 
@@ -332,19 +911,17 @@ static int edma_dma_resume(struct dma_chan *chan)
  * @direction: Direction of the transfer
  */
 static int edma_config_pset(struct dma_chan *chan, struct edma_pset *epset,
-       dma_addr_t src_addr, dma_addr_t dst_addr, u32 burst,
-       enum dma_slave_buswidth dev_width, unsigned int dma_length,
-       enum dma_transfer_direction direction)
+                           dma_addr_t src_addr, dma_addr_t dst_addr, u32 burst,
+                           unsigned int acnt, unsigned int dma_length,
+                           enum dma_transfer_direction direction)
 {
        struct edma_chan *echan = to_edma_chan(chan);
        struct device *dev = chan->device->dev;
        struct edmacc_param *param = &epset->param;
-       int acnt, bcnt, ccnt, cidx;
+       int bcnt, ccnt, cidx;
        int src_bidx, dst_bidx, src_cidx, dst_cidx;
        int absync;
 
-       acnt = dev_width;
-
        /* src/dst_maxburst == 0 is the same case as src/dst_maxburst == 1 */
        if (!burst)
                burst = 1;
@@ -480,8 +1057,8 @@ static struct dma_async_tx_descriptor *edma_prep_slave_sg(
                return NULL;
        }
 
-       edesc = kzalloc(sizeof(*edesc) + sg_len *
-               sizeof(edesc->pset[0]), GFP_ATOMIC);
+       edesc = kzalloc(sizeof(*edesc) + sg_len * sizeof(edesc->pset[0]),
+                       GFP_ATOMIC);
        if (!edesc) {
                dev_err(dev, "%s: Failed to allocate a descriptor\n", __func__);
                return NULL;
@@ -498,8 +1075,7 @@ static struct dma_async_tx_descriptor *edma_prep_slave_sg(
        for (i = 0; i < nslots; i++) {
                if (echan->slot[i] < 0) {
                        echan->slot[i] =
-                               edma_alloc_slot(EDMA_CTLR(echan->ch_num),
-                                               EDMA_SLOT_ANY);
+                               edma_alloc_slot(echan->ecc, EDMA_SLOT_ANY);
                        if (echan->slot[i] < 0) {
                                kfree(edesc);
                                dev_err(dev, "%s: Failed to allocate slot\n",
@@ -546,36 +1122,98 @@ static struct dma_async_tx_descriptor *edma_prep_dma_memcpy(
        struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
        size_t len, unsigned long tx_flags)
 {
-       int ret;
+       int ret, nslots;
        struct edma_desc *edesc;
        struct device *dev = chan->device->dev;
        struct edma_chan *echan = to_edma_chan(chan);
+       unsigned int width, pset_len;
 
        if (unlikely(!echan || !len))
                return NULL;
 
-       edesc = kzalloc(sizeof(*edesc) + sizeof(edesc->pset[0]), GFP_ATOMIC);
+       if (len < SZ_64K) {
+               /*
+                * Transfer size less than 64K can be handled with one paRAM
+                * slot and with one burst.
+                * ACNT = length
+                */
+               width = len;
+               pset_len = len;
+               nslots = 1;
+       } else {
+               /*
+                * Transfer size bigger than 64K will be handled with maximum of
+                * two paRAM slots.
+                * slot1: (full_length / 32767) times 32767 bytes bursts.
+                *        ACNT = 32767, length1: (full_length / 32767) * 32767
+                * slot2: the remaining amount of data after slot1.
+                *        ACNT = full_length - length1, length2 = ACNT
+                *
+                * When the full_length is multibple of 32767 one slot can be
+                * used to complete the transfer.
+                */
+               width = SZ_32K - 1;
+               pset_len = rounddown(len, width);
+               /* One slot is enough for lengths multiple of (SZ_32K -1) */
+               if (unlikely(pset_len == len))
+                       nslots = 1;
+               else
+                       nslots = 2;
+       }
+
+       edesc = kzalloc(sizeof(*edesc) + nslots * sizeof(edesc->pset[0]),
+                       GFP_ATOMIC);
        if (!edesc) {
                dev_dbg(dev, "Failed to allocate a descriptor\n");
                return NULL;
        }
 
-       edesc->pset_nr = 1;
+       edesc->pset_nr = nslots;
+       edesc->residue = edesc->residue_stat = len;
+       edesc->direction = DMA_MEM_TO_MEM;
+       edesc->echan = echan;
 
        ret = edma_config_pset(chan, &edesc->pset[0], src, dest, 1,
-                              DMA_SLAVE_BUSWIDTH_4_BYTES, len, DMA_MEM_TO_MEM);
-       if (ret < 0)
+                              width, pset_len, DMA_MEM_TO_MEM);
+       if (ret < 0) {
+               kfree(edesc);
                return NULL;
+       }
 
        edesc->absync = ret;
 
-       /*
-        * Enable intermediate transfer chaining to re-trigger channel
-        * on completion of every TR, and enable transfer-completion
-        * interrupt on completion of the whole transfer.
-        */
        edesc->pset[0].param.opt |= ITCCHEN;
-       edesc->pset[0].param.opt |= TCINTEN;
+       if (nslots == 1) {
+               /* Enable transfer complete interrupt */
+               edesc->pset[0].param.opt |= TCINTEN;
+       } else {
+               /* Enable transfer complete chaining for the first slot */
+               edesc->pset[0].param.opt |= TCCHEN;
+
+               if (echan->slot[1] < 0) {
+                       echan->slot[1] = edma_alloc_slot(echan->ecc,
+                                                        EDMA_SLOT_ANY);
+                       if (echan->slot[1] < 0) {
+                               kfree(edesc);
+                               dev_err(dev, "%s: Failed to allocate slot\n",
+                                       __func__);
+                               return NULL;
+                       }
+               }
+               dest += pset_len;
+               src += pset_len;
+               pset_len = width = len % (SZ_32K - 1);
+
+               ret = edma_config_pset(chan, &edesc->pset[1], src, dest, 1,
+                                      width, pset_len, DMA_MEM_TO_MEM);
+               if (ret < 0) {
+                       kfree(edesc);
+                       return NULL;
+               }
+
+               edesc->pset[1].param.opt |= ITCCHEN;
+               edesc->pset[1].param.opt |= TCINTEN;
+       }
 
        return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
 }
@@ -634,8 +1272,8 @@ static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
        if (nslots > MAX_NR_SG)
                return NULL;
 
-       edesc = kzalloc(sizeof(*edesc) + nslots *
-               sizeof(edesc->pset[0]), GFP_ATOMIC);
+       edesc = kzalloc(sizeof(*edesc) + nslots * sizeof(edesc->pset[0]),
+                       GFP_ATOMIC);
        if (!edesc) {
                dev_err(dev, "%s: Failed to allocate a descriptor\n", __func__);
                return NULL;
@@ -654,8 +1292,7 @@ static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
                /* Allocate a PaRAM slot, if needed */
                if (echan->slot[i] < 0) {
                        echan->slot[i] =
-                               edma_alloc_slot(EDMA_CTLR(echan->ch_num),
-                                               EDMA_SLOT_ANY);
+                               edma_alloc_slot(echan->ecc, EDMA_SLOT_ANY);
                        if (echan->slot[i] < 0) {
                                kfree(edesc);
                                dev_err(dev, "%s: Failed to allocate slot\n",
@@ -716,128 +1353,281 @@ static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
        }
 
        /* Place the cyclic channel to highest priority queue */
-       edma_assign_channel_eventq(echan->ch_num, EVENTQ_0);
+       if (!echan->tc)
+               edma_assign_channel_eventq(echan, EVENTQ_0);
 
        return vchan_tx_prep(&echan->vchan, &edesc->vdesc, tx_flags);
 }
 
-static void edma_callback(unsigned ch_num, u16 ch_status, void *data)
+static void edma_completion_handler(struct edma_chan *echan)
 {
-       struct edma_chan *echan = data;
        struct device *dev = echan->vchan.chan.device->dev;
-       struct edma_desc *edesc;
-       struct edmacc_param p;
+       struct edma_desc *edesc = echan->edesc;
 
-       edesc = echan->edesc;
+       if (!edesc)
+               return;
 
-       /* Pause the channel for non-cyclic */
-       if (!edesc || (edesc && !edesc->cyclic))
-               edma_pause(echan->ch_num);
-
-       switch (ch_status) {
-       case EDMA_DMA_COMPLETE:
-               spin_lock(&echan->vchan.lock);
-
-               if (edesc) {
-                       if (edesc->cyclic) {
-                               vchan_cyclic_callback(&edesc->vdesc);
-                       } else if (edesc->processed == edesc->pset_nr) {
-                               dev_dbg(dev, "Transfer complete, stopping channel %d\n", ch_num);
-                               edesc->residue = 0;
-                               edma_stop(echan->ch_num);
-                               vchan_cookie_complete(&edesc->vdesc);
-                               edma_execute(echan);
-                       } else {
-                               dev_dbg(dev, "Intermediate transfer complete on channel %d\n", ch_num);
-
-                               /* Update statistics for tx_status */
-                               edesc->residue -= edesc->sg_len;
-                               edesc->residue_stat = edesc->residue;
-                               edesc->processed_stat = edesc->processed;
-
-                               edma_execute(echan);
-                       }
+       spin_lock(&echan->vchan.lock);
+       if (edesc->cyclic) {
+               vchan_cyclic_callback(&edesc->vdesc);
+               spin_unlock(&echan->vchan.lock);
+               return;
+       } else if (edesc->processed == edesc->pset_nr) {
+               edesc->residue = 0;
+               edma_stop(echan);
+               vchan_cookie_complete(&edesc->vdesc);
+               echan->edesc = NULL;
+
+               dev_dbg(dev, "Transfer completed on channel %d\n",
+                       echan->ch_num);
+       } else {
+               dev_dbg(dev, "Sub transfer completed on channel %d\n",
+                       echan->ch_num);
+
+               edma_pause(echan);
+
+               /* Update statistics for tx_status */
+               edesc->residue -= edesc->sg_len;
+               edesc->residue_stat = edesc->residue;
+               edesc->processed_stat = edesc->processed;
+       }
+       edma_execute(echan);
+
+       spin_unlock(&echan->vchan.lock);
+}
+
+/* eDMA interrupt handler */
+static irqreturn_t dma_irq_handler(int irq, void *data)
+{
+       struct edma_cc *ecc = data;
+       int ctlr;
+       u32 sh_ier;
+       u32 sh_ipr;
+       u32 bank;
+
+       ctlr = ecc->id;
+       if (ctlr < 0)
+               return IRQ_NONE;
+
+       dev_vdbg(ecc->dev, "dma_irq_handler\n");
+
+       sh_ipr = edma_shadow0_read_array(ecc, SH_IPR, 0);
+       if (!sh_ipr) {
+               sh_ipr = edma_shadow0_read_array(ecc, SH_IPR, 1);
+               if (!sh_ipr)
+                       return IRQ_NONE;
+               sh_ier = edma_shadow0_read_array(ecc, SH_IER, 1);
+               bank = 1;
+       } else {
+               sh_ier = edma_shadow0_read_array(ecc, SH_IER, 0);
+               bank = 0;
+       }
+
+       do {
+               u32 slot;
+               u32 channel;
+
+               slot = __ffs(sh_ipr);
+               sh_ipr &= ~(BIT(slot));
+
+               if (sh_ier & BIT(slot)) {
+                       channel = (bank << 5) | slot;
+                       /* Clear the corresponding IPR bits */
+                       edma_shadow0_write_array(ecc, SH_ICR, bank, BIT(slot));
+                       edma_completion_handler(&ecc->slave_chans[channel]);
                }
+       } while (sh_ipr);
 
-               spin_unlock(&echan->vchan.lock);
+       edma_shadow0_write(ecc, SH_IEVAL, 1);
+       return IRQ_HANDLED;
+}
+
+static void edma_error_handler(struct edma_chan *echan)
+{
+       struct edma_cc *ecc = echan->ecc;
+       struct device *dev = echan->vchan.chan.device->dev;
+       struct edmacc_param p;
 
-               break;
-       case EDMA_DMA_CC_ERROR:
-               spin_lock(&echan->vchan.lock);
+       if (!echan->edesc)
+               return;
 
-               edma_read_slot(EDMA_CHAN_SLOT(echan->slot[0]), &p);
+       spin_lock(&echan->vchan.lock);
 
+       edma_read_slot(ecc, echan->slot[0], &p);
+       /*
+        * Issue later based on missed flag which will be sure
+        * to happen as:
+        * (1) we finished transmitting an intermediate slot and
+        *     edma_execute is coming up.
+        * (2) or we finished current transfer and issue will
+        *     call edma_execute.
+        *
+        * Important note: issuing can be dangerous here and
+        * lead to some nasty recursion when we are in a NULL
+        * slot. So we avoid doing so and set the missed flag.
+        */
+       if (p.a_b_cnt == 0 && p.ccnt == 0) {
+               dev_dbg(dev, "Error on null slot, setting miss\n");
+               echan->missed = 1;
+       } else {
                /*
-                * Issue later based on missed flag which will be sure
-                * to happen as:
-                * (1) we finished transmitting an intermediate slot and
-                *     edma_execute is coming up.
-                * (2) or we finished current transfer and issue will
-                *     call edma_execute.
-                *
-                * Important note: issuing can be dangerous here and
-                * lead to some nasty recursion when we are in a NULL
-                * slot. So we avoid doing so and set the missed flag.
+                * The slot is already programmed but the event got
+                * missed, so its safe to issue it here.
                 */
-               if (p.a_b_cnt == 0 && p.ccnt == 0) {
-                       dev_dbg(dev, "Error occurred, looks like slot is null, just setting miss\n");
-                       echan->missed = 1;
-               } else {
-                       /*
-                        * The slot is already programmed but the event got
-                        * missed, so its safe to issue it here.
-                        */
-                       dev_dbg(dev, "Error occurred but slot is non-null, TRIGGERING\n");
-                       edma_clean_channel(echan->ch_num);
-                       edma_stop(echan->ch_num);
-                       edma_start(echan->ch_num);
-                       edma_trigger_channel(echan->ch_num);
+               dev_dbg(dev, "Missed event, TRIGGERING\n");
+               edma_clean_channel(echan);
+               edma_stop(echan);
+               edma_start(echan);
+               edma_trigger_channel(echan);
+       }
+       spin_unlock(&echan->vchan.lock);
+}
+
+static inline bool edma_error_pending(struct edma_cc *ecc)
+{
+       if (edma_read_array(ecc, EDMA_EMR, 0) ||
+           edma_read_array(ecc, EDMA_EMR, 1) ||
+           edma_read(ecc, EDMA_QEMR) || edma_read(ecc, EDMA_CCERR))
+               return true;
+
+       return false;
+}
+
+/* eDMA error interrupt handler */
+static irqreturn_t dma_ccerr_handler(int irq, void *data)
+{
+       struct edma_cc *ecc = data;
+       int i, j;
+       int ctlr;
+       unsigned int cnt = 0;
+       unsigned int val;
+
+       ctlr = ecc->id;
+       if (ctlr < 0)
+               return IRQ_NONE;
+
+       dev_vdbg(ecc->dev, "dma_ccerr_handler\n");
+
+       if (!edma_error_pending(ecc))
+               return IRQ_NONE;
+
+       while (1) {
+               /* Event missed register(s) */
+               for (j = 0; j < 2; j++) {
+                       unsigned long emr;
+
+                       val = edma_read_array(ecc, EDMA_EMR, j);
+                       if (!val)
+                               continue;
+
+                       dev_dbg(ecc->dev, "EMR%d 0x%08x\n", j, val);
+                       emr = val;
+                       for (i = find_next_bit(&emr, 32, 0); i < 32;
+                            i = find_next_bit(&emr, 32, i + 1)) {
+                               int k = (j << 5) + i;
+
+                               /* Clear the corresponding EMR bits */
+                               edma_write_array(ecc, EDMA_EMCR, j, BIT(i));
+                               /* Clear any SER */
+                               edma_shadow0_write_array(ecc, SH_SECR, j,
+                                                        BIT(i));
+                               edma_error_handler(&ecc->slave_chans[k]);
+                       }
                }
 
-               spin_unlock(&echan->vchan.lock);
+               val = edma_read(ecc, EDMA_QEMR);
+               if (val) {
+                       dev_dbg(ecc->dev, "QEMR 0x%02x\n", val);
+                       /* Not reported, just clear the interrupt reason. */
+                       edma_write(ecc, EDMA_QEMCR, val);
+                       edma_shadow0_write(ecc, SH_QSECR, val);
+               }
+
+               val = edma_read(ecc, EDMA_CCERR);
+               if (val) {
+                       dev_warn(ecc->dev, "CCERR 0x%08x\n", val);
+                       /* Not reported, just clear the interrupt reason. */
+                       edma_write(ecc, EDMA_CCERRCLR, val);
+               }
 
-               break;
-       default:
-               break;
+               if (!edma_error_pending(ecc))
+                       break;
+               cnt++;
+               if (cnt > 10)
+                       break;
        }
+       edma_write(ecc, EDMA_EEVAL, 1);
+       return IRQ_HANDLED;
+}
+
+static void edma_tc_set_pm_state(struct edma_tc *tc, bool enable)
+{
+       struct platform_device *tc_pdev;
+       int ret;
+
+       if (!IS_ENABLED(CONFIG_OF) || !tc)
+               return;
+
+       tc_pdev = of_find_device_by_node(tc->node);
+       if (!tc_pdev) {
+               pr_err("%s: TPTC device is not found\n", __func__);
+               return;
+       }
+       if (!pm_runtime_enabled(&tc_pdev->dev))
+               pm_runtime_enable(&tc_pdev->dev);
+
+       if (enable)
+               ret = pm_runtime_get_sync(&tc_pdev->dev);
+       else
+               ret = pm_runtime_put_sync(&tc_pdev->dev);
+
+       if (ret < 0)
+               pr_err("%s: pm_runtime_%s_sync() failed for %s\n", __func__,
+                      enable ? "get" : "put", dev_name(&tc_pdev->dev));
 }
 
 /* Alloc channel resources */
 static int edma_alloc_chan_resources(struct dma_chan *chan)
 {
        struct edma_chan *echan = to_edma_chan(chan);
-       struct device *dev = chan->device->dev;
+       struct edma_cc *ecc = echan->ecc;
+       struct device *dev = ecc->dev;
+       enum dma_event_q eventq_no = EVENTQ_DEFAULT;
        int ret;
-       int a_ch_num;
-       LIST_HEAD(descs);
-
-       a_ch_num = edma_alloc_channel(echan->ch_num, edma_callback,
-                                       echan, EVENTQ_DEFAULT);
 
-       if (a_ch_num < 0) {
-               ret = -ENODEV;
-               goto err_no_chan;
+       if (echan->tc) {
+               eventq_no = echan->tc->id;
+       } else if (ecc->tc_list) {
+               /* memcpy channel */
+               echan->tc = &ecc->tc_list[ecc->info->default_queue];
+               eventq_no = echan->tc->id;
        }
 
-       if (a_ch_num != echan->ch_num) {
-               dev_err(dev, "failed to allocate requested channel %u:%u\n",
-                       EDMA_CTLR(echan->ch_num),
+       ret = edma_alloc_channel(echan, eventq_no);
+       if (ret)
+               return ret;
+
+       echan->slot[0] = edma_alloc_slot(ecc, echan->ch_num);
+       if (echan->slot[0] < 0) {
+               dev_err(dev, "Entry slot allocation failed for channel %u\n",
                        EDMA_CHAN_SLOT(echan->ch_num));
-               ret = -ENODEV;
-               goto err_wrong_chan;
+               goto err_slot;
        }
 
+       /* Set up channel -> slot mapping for the entry slot */
+       edma_set_chmap(echan, echan->slot[0]);
        echan->alloced = true;
-       echan->slot[0] = echan->ch_num;
 
-       dev_dbg(dev, "allocated channel %d for %u:%u\n", echan->ch_num,
-               EDMA_CTLR(echan->ch_num), EDMA_CHAN_SLOT(echan->ch_num));
+       dev_dbg(dev, "Got eDMA channel %d for virt channel %d (%s trigger)\n",
+               EDMA_CHAN_SLOT(echan->ch_num), chan->chan_id,
+               echan->hw_triggered ? "HW" : "SW");
+
+       edma_tc_set_pm_state(echan->tc, true);
 
        return 0;
 
-err_wrong_chan:
-       edma_free_channel(a_ch_num);
-err_no_chan:
+err_slot:
+       edma_free_channel(echan);
        return ret;
 }
 
@@ -845,29 +1635,37 @@ err_no_chan:
 static void edma_free_chan_resources(struct dma_chan *chan)
 {
        struct edma_chan *echan = to_edma_chan(chan);
-       struct device *dev = chan->device->dev;
+       struct device *dev = echan->ecc->dev;
        int i;
 
        /* Terminate transfers */
-       edma_stop(echan->ch_num);
+       edma_stop(echan);
 
        vchan_free_chan_resources(&echan->vchan);
 
        /* Free EDMA PaRAM slots */
-       for (i = 1; i < EDMA_MAX_SLOTS; i++) {
+       for (i = 0; i < EDMA_MAX_SLOTS; i++) {
                if (echan->slot[i] >= 0) {
-                       edma_free_slot(echan->slot[i]);
+                       edma_free_slot(echan->ecc, echan->slot[i]);
                        echan->slot[i] = -1;
                }
        }
 
+       /* Set entry slot to the dummy slot */
+       edma_set_chmap(echan, echan->ecc->dummy_slot);
+
        /* Free EDMA channel */
        if (echan->alloced) {
-               edma_free_channel(echan->ch_num);
+               edma_free_channel(echan);
                echan->alloced = false;
        }
 
-       dev_dbg(dev, "freeing channel for %u\n", echan->ch_num);
+       edma_tc_set_pm_state(echan->tc, false);
+       echan->tc = NULL;
+       echan->hw_triggered = false;
+
+       dev_dbg(dev, "Free eDMA channel %d for virt channel %d\n",
+               EDMA_CHAN_SLOT(echan->ch_num), chan->chan_id);
 }
 
 /* Send pending descriptor to hardware */
@@ -893,7 +1691,7 @@ static u32 edma_residue(struct edma_desc *edesc)
         * We always read the dst/src position from the first RamPar
         * pset. That's the one which is active now.
         */
-       pos = edma_get_position(edesc->echan->slot[0], dst);
+       pos = edma_get_position(edesc->echan->ecc, edesc->echan->slot[0], dst);
 
        /*
         * Cyclic is simple. Just subtract pset[0].addr from pos.
@@ -954,19 +1752,99 @@ static enum dma_status edma_tx_status(struct dma_chan *chan,
        return ret;
 }
 
-static void __init edma_chan_init(struct edma_cc *ecc,
-                                 struct dma_device *dma,
-                                 struct edma_chan *echans)
+static bool edma_is_memcpy_channel(int ch_num, s32 *memcpy_channels)
+{
+       if (!memcpy_channels)
+               return false;
+       while (*memcpy_channels != -1) {
+               if (*memcpy_channels == ch_num)
+                       return true;
+               memcpy_channels++;
+       }
+       return false;
+}
+
+#define EDMA_DMA_BUSWIDTHS     (BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
+                                BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
+                                BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
+                                BIT(DMA_SLAVE_BUSWIDTH_4_BYTES))
+
+static void edma_dma_init(struct edma_cc *ecc, bool legacy_mode)
 {
+       struct dma_device *s_ddev = &ecc->dma_slave;
+       struct dma_device *m_ddev = NULL;
+       s32 *memcpy_channels = ecc->info->memcpy_channels;
        int i, j;
 
-       for (i = 0; i < EDMA_CHANS; i++) {
-               struct edma_chan *echan = &echans[i];
-               echan->ch_num = EDMA_CTLR_CHAN(ecc->ctlr, i);
+       dma_cap_zero(s_ddev->cap_mask);
+       dma_cap_set(DMA_SLAVE, s_ddev->cap_mask);
+       dma_cap_set(DMA_CYCLIC, s_ddev->cap_mask);
+       if (ecc->legacy_mode && !memcpy_channels) {
+               dev_warn(ecc->dev,
+                        "Legacy memcpy is enabled, things might not work\n");
+
+               dma_cap_set(DMA_MEMCPY, s_ddev->cap_mask);
+               s_ddev->device_prep_dma_memcpy = edma_prep_dma_memcpy;
+               s_ddev->directions = BIT(DMA_MEM_TO_MEM);
+       }
+
+       s_ddev->device_prep_slave_sg = edma_prep_slave_sg;
+       s_ddev->device_prep_dma_cyclic = edma_prep_dma_cyclic;
+       s_ddev->device_alloc_chan_resources = edma_alloc_chan_resources;
+       s_ddev->device_free_chan_resources = edma_free_chan_resources;
+       s_ddev->device_issue_pending = edma_issue_pending;
+       s_ddev->device_tx_status = edma_tx_status;
+       s_ddev->device_config = edma_slave_config;
+       s_ddev->device_pause = edma_dma_pause;
+       s_ddev->device_resume = edma_dma_resume;
+       s_ddev->device_terminate_all = edma_terminate_all;
+
+       s_ddev->src_addr_widths = EDMA_DMA_BUSWIDTHS;
+       s_ddev->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
+       s_ddev->directions |= (BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV));
+       s_ddev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
+
+       s_ddev->dev = ecc->dev;
+       INIT_LIST_HEAD(&s_ddev->channels);
+
+       if (memcpy_channels) {
+               m_ddev = devm_kzalloc(ecc->dev, sizeof(*m_ddev), GFP_KERNEL);
+               ecc->dma_memcpy = m_ddev;
+
+               dma_cap_zero(m_ddev->cap_mask);
+               dma_cap_set(DMA_MEMCPY, m_ddev->cap_mask);
+
+               m_ddev->device_prep_dma_memcpy = edma_prep_dma_memcpy;
+               m_ddev->device_alloc_chan_resources = edma_alloc_chan_resources;
+               m_ddev->device_free_chan_resources = edma_free_chan_resources;
+               m_ddev->device_issue_pending = edma_issue_pending;
+               m_ddev->device_tx_status = edma_tx_status;
+               m_ddev->device_config = edma_slave_config;
+               m_ddev->device_pause = edma_dma_pause;
+               m_ddev->device_resume = edma_dma_resume;
+               m_ddev->device_terminate_all = edma_terminate_all;
+
+               m_ddev->src_addr_widths = EDMA_DMA_BUSWIDTHS;
+               m_ddev->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
+               m_ddev->directions = BIT(DMA_MEM_TO_MEM);
+               m_ddev->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
+
+               m_ddev->dev = ecc->dev;
+               INIT_LIST_HEAD(&m_ddev->channels);
+       } else if (!ecc->legacy_mode) {
+               dev_info(ecc->dev, "memcpy is disabled\n");
+       }
+
+       for (i = 0; i < ecc->num_channels; i++) {
+               struct edma_chan *echan = &ecc->slave_chans[i];
+               echan->ch_num = EDMA_CTLR_CHAN(ecc->id, i);
                echan->ecc = ecc;
                echan->vchan.desc_free = edma_desc_free;
 
-               vchan_init(&echan->vchan, dma);
+               if (m_ddev && edma_is_memcpy_channel(i, memcpy_channels))
+                       vchan_init(&echan->vchan, m_ddev);
+               else
+                       vchan_init(&echan->vchan, s_ddev);
 
                INIT_LIST_HEAD(&echan->node);
                for (j = 0; j < EDMA_MAX_SLOTS; j++)
@@ -974,85 +1852,493 @@ static void __init edma_chan_init(struct edma_cc *ecc,
        }
 }
 
-#define EDMA_DMA_BUSWIDTHS     (BIT(DMA_SLAVE_BUSWIDTH_1_BYTE) | \
-                                BIT(DMA_SLAVE_BUSWIDTH_2_BYTES) | \
-                                BIT(DMA_SLAVE_BUSWIDTH_3_BYTES) | \
-                                BIT(DMA_SLAVE_BUSWIDTH_4_BYTES))
-
-static void edma_dma_init(struct edma_cc *ecc, struct dma_device *dma,
-                         struct device *dev)
+static int edma_setup_from_hw(struct device *dev, struct edma_soc_info *pdata,
+                             struct edma_cc *ecc)
 {
-       dma->device_prep_slave_sg = edma_prep_slave_sg;
-       dma->device_prep_dma_cyclic = edma_prep_dma_cyclic;
-       dma->device_prep_dma_memcpy = edma_prep_dma_memcpy;
-       dma->device_alloc_chan_resources = edma_alloc_chan_resources;
-       dma->device_free_chan_resources = edma_free_chan_resources;
-       dma->device_issue_pending = edma_issue_pending;
-       dma->device_tx_status = edma_tx_status;
-       dma->device_config = edma_slave_config;
-       dma->device_pause = edma_dma_pause;
-       dma->device_resume = edma_dma_resume;
-       dma->device_terminate_all = edma_terminate_all;
+       int i;
+       u32 value, cccfg;
+       s8 (*queue_priority_map)[2];
+
+       /* Decode the eDMA3 configuration from CCCFG register */
+       cccfg = edma_read(ecc, EDMA_CCCFG);
 
-       dma->src_addr_widths = EDMA_DMA_BUSWIDTHS;
-       dma->dst_addr_widths = EDMA_DMA_BUSWIDTHS;
-       dma->directions = BIT(DMA_DEV_TO_MEM) | BIT(DMA_MEM_TO_DEV);
-       dma->residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
+       value = GET_NUM_REGN(cccfg);
+       ecc->num_region = BIT(value);
 
-       dma->dev = dev;
+       value = GET_NUM_DMACH(cccfg);
+       ecc->num_channels = BIT(value + 1);
+
+       value = GET_NUM_QDMACH(cccfg);
+       ecc->num_qchannels = value * 2;
+
+       value = GET_NUM_PAENTRY(cccfg);
+       ecc->num_slots = BIT(value + 4);
+
+       value = GET_NUM_EVQUE(cccfg);
+       ecc->num_tc = value + 1;
+
+       ecc->chmap_exist = (cccfg & CHMAP_EXIST) ? true : false;
+
+       dev_dbg(dev, "eDMA3 CC HW configuration (cccfg: 0x%08x):\n", cccfg);
+       dev_dbg(dev, "num_region: %u\n", ecc->num_region);
+       dev_dbg(dev, "num_channels: %u\n", ecc->num_channels);
+       dev_dbg(dev, "num_qchannels: %u\n", ecc->num_qchannels);
+       dev_dbg(dev, "num_slots: %u\n", ecc->num_slots);
+       dev_dbg(dev, "num_tc: %u\n", ecc->num_tc);
+       dev_dbg(dev, "chmap_exist: %s\n", ecc->chmap_exist ? "yes" : "no");
+
+       /* Nothing need to be done if queue priority is provided */
+       if (pdata->queue_priority_mapping)
+               return 0;
 
        /*
-        * code using dma memcpy must make sure alignment of
-        * length is at dma->copy_align boundary.
+        * Configure TC/queue priority as follows:
+        * Q0 - priority 0
+        * Q1 - priority 1
+        * Q2 - priority 2
+        * ...
+        * The meaning of priority numbers: 0 highest priority, 7 lowest
+        * priority. So Q0 is the highest priority queue and the last queue has
+        * the lowest priority.
         */
-       dma->copy_align = DMA_SLAVE_BUSWIDTH_4_BYTES;
+       queue_priority_map = devm_kcalloc(dev, ecc->num_tc + 1, sizeof(s8),
+                                         GFP_KERNEL);
+       if (!queue_priority_map)
+               return -ENOMEM;
+
+       for (i = 0; i < ecc->num_tc; i++) {
+               queue_priority_map[i][0] = i;
+               queue_priority_map[i][1] = i;
+       }
+       queue_priority_map[i][0] = -1;
+       queue_priority_map[i][1] = -1;
+
+       pdata->queue_priority_mapping = queue_priority_map;
+       /* Default queue has the lowest priority */
+       pdata->default_queue = i - 1;
 
-       INIT_LIST_HEAD(&dma->channels);
+       return 0;
 }
 
+#if IS_ENABLED(CONFIG_OF)
+static int edma_xbar_event_map(struct device *dev, struct edma_soc_info *pdata,
+                              size_t sz)
+{
+       const char pname[] = "ti,edma-xbar-event-map";
+       struct resource res;
+       void __iomem *xbar;
+       s16 (*xbar_chans)[2];
+       size_t nelm = sz / sizeof(s16);
+       u32 shift, offset, mux;
+       int ret, i;
+
+       xbar_chans = devm_kcalloc(dev, nelm + 2, sizeof(s16), GFP_KERNEL);
+       if (!xbar_chans)
+               return -ENOMEM;
+
+       ret = of_address_to_resource(dev->of_node, 1, &res);
+       if (ret)
+               return -ENOMEM;
+
+       xbar = devm_ioremap(dev, res.start, resource_size(&res));
+       if (!xbar)
+               return -ENOMEM;
+
+       ret = of_property_read_u16_array(dev->of_node, pname, (u16 *)xbar_chans,
+                                        nelm);
+       if (ret)
+               return -EIO;
+
+       /* Invalidate last entry for the other user of this mess */
+       nelm >>= 1;
+       xbar_chans[nelm][0] = -1;
+       xbar_chans[nelm][1] = -1;
+
+       for (i = 0; i < nelm; i++) {
+               shift = (xbar_chans[i][1] & 0x03) << 3;
+               offset = xbar_chans[i][1] & 0xfffffffc;
+               mux = readl(xbar + offset);
+               mux &= ~(0xff << shift);
+               mux |= xbar_chans[i][0] << shift;
+               writel(mux, (xbar + offset));
+       }
+
+       pdata->xbar_chans = (const s16 (*)[2]) xbar_chans;
+       return 0;
+}
+
+static struct edma_soc_info *edma_setup_info_from_dt(struct device *dev,
+                                                    bool legacy_mode)
+{
+       struct edma_soc_info *info;
+       struct property *prop;
+       size_t sz;
+       int ret;
+
+       info = devm_kzalloc(dev, sizeof(struct edma_soc_info), GFP_KERNEL);
+       if (!info)
+               return ERR_PTR(-ENOMEM);
+
+       if (legacy_mode) {
+               prop = of_find_property(dev->of_node, "ti,edma-xbar-event-map",
+                                       &sz);
+               if (prop) {
+                       ret = edma_xbar_event_map(dev, info, sz);
+                       if (ret)
+                               return ERR_PTR(ret);
+               }
+               return info;
+       }
+
+       /* Get the list of channels allocated to be used for memcpy */
+       prop = of_find_property(dev->of_node, "ti,edma-memcpy-channels", &sz);
+       if (prop) {
+               const char pname[] = "ti,edma-memcpy-channels";
+               size_t nelm = sz / sizeof(s32);
+               s32 *memcpy_ch;
+
+               memcpy_ch = devm_kcalloc(dev, nelm + 1, sizeof(s32),
+                                        GFP_KERNEL);
+               if (!memcpy_ch)
+                       return ERR_PTR(-ENOMEM);
+
+               ret = of_property_read_u32_array(dev->of_node, pname,
+                                                (u32 *)memcpy_ch, nelm);
+               if (ret)
+                       return ERR_PTR(ret);
+
+               memcpy_ch[nelm] = -1;
+               info->memcpy_channels = memcpy_ch;
+       }
+
+       prop = of_find_property(dev->of_node, "ti,edma-reserved-slot-ranges",
+                               &sz);
+       if (prop) {
+               const char pname[] = "ti,edma-reserved-slot-ranges";
+               u32 (*tmp)[2];
+               s16 (*rsv_slots)[2];
+               size_t nelm = sz / sizeof(*tmp);
+               struct edma_rsv_info *rsv_info;
+               int i;
+
+               if (!nelm)
+                       return info;
+
+               tmp = kcalloc(nelm, sizeof(*tmp), GFP_KERNEL);
+               if (!tmp)
+                       return ERR_PTR(-ENOMEM);
+
+               rsv_info = devm_kzalloc(dev, sizeof(*rsv_info), GFP_KERNEL);
+               if (!rsv_info) {
+                       kfree(tmp);
+                       return ERR_PTR(-ENOMEM);
+               }
+
+               rsv_slots = devm_kcalloc(dev, nelm + 1, sizeof(*rsv_slots),
+                                        GFP_KERNEL);
+               if (!rsv_slots) {
+                       kfree(tmp);
+                       return ERR_PTR(-ENOMEM);
+               }
+
+               ret = of_property_read_u32_array(dev->of_node, pname,
+                                                (u32 *)tmp, nelm * 2);
+               if (ret) {
+                       kfree(tmp);
+                       return ERR_PTR(ret);
+               }
+
+               for (i = 0; i < nelm; i++) {
+                       rsv_slots[i][0] = tmp[i][0];
+                       rsv_slots[i][1] = tmp[i][1];
+               }
+               rsv_slots[nelm][0] = -1;
+               rsv_slots[nelm][1] = -1;
+
+               info->rsv = rsv_info;
+               info->rsv->rsv_slots = (const s16 (*)[2])rsv_slots;
+
+               kfree(tmp);
+       }
+
+       return info;
+}
+
+static struct dma_chan *of_edma_xlate(struct of_phandle_args *dma_spec,
+                                     struct of_dma *ofdma)
+{
+       struct edma_cc *ecc = ofdma->of_dma_data;
+       struct dma_chan *chan = NULL;
+       struct edma_chan *echan;
+       int i;
+
+       if (!ecc || dma_spec->args_count < 1)
+               return NULL;
+
+       for (i = 0; i < ecc->num_channels; i++) {
+               echan = &ecc->slave_chans[i];
+               if (echan->ch_num == dma_spec->args[0]) {
+                       chan = &echan->vchan.chan;
+                       break;
+               }
+       }
+
+       if (!chan)
+               return NULL;
+
+       if (echan->ecc->legacy_mode && dma_spec->args_count == 1)
+               goto out;
+
+       if (!echan->ecc->legacy_mode && dma_spec->args_count == 2 &&
+           dma_spec->args[1] < echan->ecc->num_tc) {
+               echan->tc = &echan->ecc->tc_list[dma_spec->args[1]];
+               goto out;
+       }
+
+       return NULL;
+out:
+       /* The channel is going to be used as HW synchronized */
+       echan->hw_triggered = true;
+       return dma_get_slave_channel(chan);
+}
+#else
+static struct edma_soc_info *edma_setup_info_from_dt(struct device *dev,
+                                                    bool legacy_mode)
+{
+       return ERR_PTR(-EINVAL);
+}
+
+static struct dma_chan *of_edma_xlate(struct of_phandle_args *dma_spec,
+                                     struct of_dma *ofdma)
+{
+       return NULL;
+}
+#endif
+
 static int edma_probe(struct platform_device *pdev)
 {
-       struct edma_cc *ecc;
+       struct edma_soc_info    *info = pdev->dev.platform_data;
+       s8                      (*queue_priority_mapping)[2];
+       int                     i, off, ln;
+       const s16               (*rsv_slots)[2];
+       const s16               (*xbar_chans)[2];
+       int                     irq;
+       char                    *irq_name;
+       struct resource         *mem;
+       struct device_node      *node = pdev->dev.of_node;
+       struct device           *dev = &pdev->dev;
+       struct edma_cc          *ecc;
+       bool                    legacy_mode = true;
        int ret;
 
-       ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
+       if (node) {
+               const struct of_device_id *match;
+
+               match = of_match_node(edma_of_ids, node);
+               if (match && (u32)match->data == EDMA_BINDING_TPCC)
+                       legacy_mode = false;
+
+               info = edma_setup_info_from_dt(dev, legacy_mode);
+               if (IS_ERR(info)) {
+                       dev_err(dev, "failed to get DT data\n");
+                       return PTR_ERR(info);
+               }
+       }
+
+       if (!info)
+               return -ENODEV;
+
+       pm_runtime_enable(dev);
+       ret = pm_runtime_get_sync(dev);
+       if (ret < 0) {
+               dev_err(dev, "pm_runtime_get_sync() failed\n");
+               return ret;
+       }
+
+       ret = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(32));
        if (ret)
                return ret;
 
-       ecc = devm_kzalloc(&pdev->dev, sizeof(*ecc), GFP_KERNEL);
+       ecc = devm_kzalloc(dev, sizeof(*ecc), GFP_KERNEL);
        if (!ecc) {
-               dev_err(&pdev->dev, "Can't allocate controller\n");
+               dev_err(dev, "Can't allocate controller\n");
                return -ENOMEM;
        }
 
-       ecc->ctlr = pdev->id;
-       ecc->dummy_slot = edma_alloc_slot(ecc->ctlr, EDMA_SLOT_ANY);
+       ecc->dev = dev;
+       ecc->id = pdev->id;
+       ecc->legacy_mode = legacy_mode;
+       /* When booting with DT the pdev->id is -1 */
+       if (ecc->id < 0)
+               ecc->id = 0;
+
+       mem = platform_get_resource_byname(pdev, IORESOURCE_MEM, "edma3_cc");
+       if (!mem) {
+               dev_dbg(dev, "mem resource not found, using index 0\n");
+               mem = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+               if (!mem) {
+                       dev_err(dev, "no mem resource?\n");
+                       return -ENODEV;
+               }
+       }
+       ecc->base = devm_ioremap_resource(dev, mem);
+       if (IS_ERR(ecc->base))
+               return PTR_ERR(ecc->base);
+
+       platform_set_drvdata(pdev, ecc);
+
+       /* Get eDMA3 configuration from IP */
+       ret = edma_setup_from_hw(dev, info, ecc);
+       if (ret)
+               return ret;
+
+       /* Allocate memory based on the information we got from the IP */
+       ecc->slave_chans = devm_kcalloc(dev, ecc->num_channels,
+                                       sizeof(*ecc->slave_chans), GFP_KERNEL);
+       if (!ecc->slave_chans)
+               return -ENOMEM;
+
+       ecc->slot_inuse = devm_kcalloc(dev, BITS_TO_LONGS(ecc->num_slots),
+                                      sizeof(unsigned long), GFP_KERNEL);
+       if (!ecc->slot_inuse)
+               return -ENOMEM;
+
+       ecc->default_queue = info->default_queue;
+
+       for (i = 0; i < ecc->num_slots; i++)
+               edma_write_slot(ecc, i, &dummy_paramset);
+
+       if (info->rsv) {
+               /* Set the reserved slots in inuse list */
+               rsv_slots = info->rsv->rsv_slots;
+               if (rsv_slots) {
+                       for (i = 0; rsv_slots[i][0] != -1; i++) {
+                               off = rsv_slots[i][0];
+                               ln = rsv_slots[i][1];
+                               set_bits(off, ln, ecc->slot_inuse);
+                       }
+               }
+       }
+
+       /* Clear the xbar mapped channels in unused list */
+       xbar_chans = info->xbar_chans;
+       if (xbar_chans) {
+               for (i = 0; xbar_chans[i][1] != -1; i++) {
+                       off = xbar_chans[i][1];
+               }
+       }
+
+       irq = platform_get_irq_byname(pdev, "edma3_ccint");
+       if (irq < 0 && node)
+               irq = irq_of_parse_and_map(node, 0);
+
+       if (irq >= 0) {
+               irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_ccint",
+                                         dev_name(dev));
+               ret = devm_request_irq(dev, irq, dma_irq_handler, 0, irq_name,
+                                      ecc);
+               if (ret) {
+                       dev_err(dev, "CCINT (%d) failed --> %d\n", irq, ret);
+                       return ret;
+               }
+       }
+
+       irq = platform_get_irq_byname(pdev, "edma3_ccerrint");
+       if (irq < 0 && node)
+               irq = irq_of_parse_and_map(node, 2);
+
+       if (irq >= 0) {
+               irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_ccerrint",
+                                         dev_name(dev));
+               ret = devm_request_irq(dev, irq, dma_ccerr_handler, 0, irq_name,
+                                      ecc);
+               if (ret) {
+                       dev_err(dev, "CCERRINT (%d) failed --> %d\n", irq, ret);
+                       return ret;
+               }
+       }
+
+       ecc->dummy_slot = edma_alloc_slot(ecc, EDMA_SLOT_ANY);
        if (ecc->dummy_slot < 0) {
-               dev_err(&pdev->dev, "Can't allocate PaRAM dummy slot\n");
+               dev_err(dev, "Can't allocate PaRAM dummy slot\n");
                return ecc->dummy_slot;
        }
 
-       dma_cap_zero(ecc->dma_slave.cap_mask);
-       dma_cap_set(DMA_SLAVE, ecc->dma_slave.cap_mask);
-       dma_cap_set(DMA_CYCLIC, ecc->dma_slave.cap_mask);
-       dma_cap_set(DMA_MEMCPY, ecc->dma_slave.cap_mask);
+       queue_priority_mapping = info->queue_priority_mapping;
+
+       if (!ecc->legacy_mode) {
+               int lowest_priority = 0;
+               struct of_phandle_args tc_args;
+
+               ecc->tc_list = devm_kcalloc(dev, ecc->num_tc,
+                                           sizeof(*ecc->tc_list), GFP_KERNEL);
+               if (!ecc->tc_list)
+                       return -ENOMEM;
+
+               for (i = 0;; i++) {
+                       ret = of_parse_phandle_with_fixed_args(node, "ti,tptcs",
+                                                              1, i, &tc_args);
+                       if (ret || i == ecc->num_tc)
+                               break;
+
+                       ecc->tc_list[i].node = tc_args.np;
+                       ecc->tc_list[i].id = i;
+                       queue_priority_mapping[i][1] = tc_args.args[0];
+                       if (queue_priority_mapping[i][1] > lowest_priority) {
+                               lowest_priority = queue_priority_mapping[i][1];
+                               info->default_queue = i;
+                       }
+               }
+       }
+
+       /* Event queue priority mapping */
+       for (i = 0; queue_priority_mapping[i][0] != -1; i++)
+               edma_assign_priority_to_queue(ecc, queue_priority_mapping[i][0],
+                                             queue_priority_mapping[i][1]);
 
-       edma_dma_init(ecc, &ecc->dma_slave, &pdev->dev);
+       for (i = 0; i < ecc->num_region; i++) {
+               edma_write_array2(ecc, EDMA_DRAE, i, 0, 0x0);
+               edma_write_array2(ecc, EDMA_DRAE, i, 1, 0x0);
+               edma_write_array(ecc, EDMA_QRAE, i, 0x0);
+       }
+       ecc->info = info;
 
-       edma_chan_init(ecc, &ecc->dma_slave, ecc->slave_chans);
+       /* Init the dma device and channels */
+       edma_dma_init(ecc, legacy_mode);
+
+       for (i = 0; i < ecc->num_channels; i++) {
+               /* Assign all channels to the default queue */
+               edma_assign_channel_eventq(&ecc->slave_chans[i],
+                                          info->default_queue);
+               /* Set entry slot to the dummy slot */
+               edma_set_chmap(&ecc->slave_chans[i], ecc->dummy_slot);
+       }
 
        ret = dma_async_device_register(&ecc->dma_slave);
-       if (ret)
+       if (ret) {
+               dev_err(dev, "slave ddev registration failed (%d)\n", ret);
                goto err_reg1;
+       }
 
-       platform_set_drvdata(pdev, ecc);
+       if (ecc->dma_memcpy) {
+               ret = dma_async_device_register(ecc->dma_memcpy);
+               if (ret) {
+                       dev_err(dev, "memcpy ddev registration failed (%d)\n",
+                               ret);
+                       dma_async_device_unregister(&ecc->dma_slave);
+                       goto err_reg1;
+               }
+       }
 
-       dev_info(&pdev->dev, "TI EDMA DMA engine driver\n");
+       if (node)
+               of_dma_controller_register(node, of_edma_xlate, ecc);
+
+       dev_info(dev, "TI EDMA DMA engine driver\n");
 
        return 0;
 
 err_reg1:
-       edma_free_slot(ecc->dummy_slot);
+       edma_free_slot(ecc, ecc->dummy_slot);
        return ret;
 }
 
@@ -1061,33 +2347,112 @@ static int edma_remove(struct platform_device *pdev)
        struct device *dev = &pdev->dev;
        struct edma_cc *ecc = dev_get_drvdata(dev);
 
+       if (dev->of_node)
+               of_dma_controller_free(dev->of_node);
        dma_async_device_unregister(&ecc->dma_slave);
-       edma_free_slot(ecc->dummy_slot);
+       if (ecc->dma_memcpy)
+               dma_async_device_unregister(ecc->dma_memcpy);
+       edma_free_slot(ecc, ecc->dummy_slot);
+
+       return 0;
+}
+
+#ifdef CONFIG_PM_SLEEP
+static int edma_pm_suspend(struct device *dev)
+{
+       struct edma_cc *ecc = dev_get_drvdata(dev);
+       struct edma_chan *echan = ecc->slave_chans;
+       int i;
+
+       for (i = 0; i < ecc->num_channels; i++) {
+               if (echan[i].alloced) {
+                       edma_setup_interrupt(&echan[i], false);
+                       edma_tc_set_pm_state(echan[i].tc, false);
+               }
+       }
 
        return 0;
 }
 
+static int edma_pm_resume(struct device *dev)
+{
+       struct edma_cc *ecc = dev_get_drvdata(dev);
+       struct edma_chan *echan = ecc->slave_chans;
+       int i;
+       s8 (*queue_priority_mapping)[2];
+
+       queue_priority_mapping = ecc->info->queue_priority_mapping;
+
+       /* Event queue priority mapping */
+       for (i = 0; queue_priority_mapping[i][0] != -1; i++)
+               edma_assign_priority_to_queue(ecc, queue_priority_mapping[i][0],
+                                             queue_priority_mapping[i][1]);
+
+       for (i = 0; i < ecc->num_channels; i++) {
+               if (echan[i].alloced) {
+                       /* ensure access through shadow region 0 */
+                       edma_or_array2(ecc, EDMA_DRAE, 0, i >> 5,
+                                      BIT(i & 0x1f));
+
+                       edma_setup_interrupt(&echan[i], true);
+
+                       /* Set up channel -> slot mapping for the entry slot */
+                       edma_set_chmap(&echan[i], echan[i].slot[0]);
+
+                       edma_tc_set_pm_state(echan[i].tc, true);
+               }
+       }
+
+       return 0;
+}
+#endif
+
+static const struct dev_pm_ops edma_pm_ops = {
+       SET_LATE_SYSTEM_SLEEP_PM_OPS(edma_pm_suspend, edma_pm_resume)
+};
+
 static struct platform_driver edma_driver = {
        .probe          = edma_probe,
        .remove         = edma_remove,
        .driver = {
-               .name = "edma-dma-engine",
+               .name   = "edma",
+               .pm     = &edma_pm_ops,
+               .of_match_table = edma_of_ids,
+       },
+};
+
+static struct platform_driver edma_tptc_driver = {
+       .driver = {
+               .name   = "edma3-tptc",
+               .of_match_table = edma_tptc_of_ids,
        },
 };
 
 bool edma_filter_fn(struct dma_chan *chan, void *param)
 {
+       bool match = false;
+
        if (chan->device->dev->driver == &edma_driver.driver) {
                struct edma_chan *echan = to_edma_chan(chan);
                unsigned ch_req = *(unsigned *)param;
-               return ch_req == echan->ch_num;
+               if (ch_req == echan->ch_num) {
+                       /* The channel is going to be used as HW synchronized */
+                       echan->hw_triggered = true;
+                       match = true;
+               }
        }
-       return false;
+       return match;
 }
 EXPORT_SYMBOL(edma_filter_fn);
 
 static int edma_init(void)
 {
+       int ret;
+
+       ret = platform_driver_register(&edma_tptc_driver);
+       if (ret)
+               return ret;
+
        return platform_driver_register(&edma_driver);
 }
 subsys_initcall(edma_init);
@@ -1095,6 +2460,7 @@ subsys_initcall(edma_init);
 static void __exit edma_exit(void)
 {
        platform_driver_unregister(&edma_driver);
+       platform_driver_unregister(&edma_tptc_driver);
 }
 module_exit(edma_exit);