dma: ti: add driver to K3 UDMA
authorVignesh R <vigneshr@ti.com>
Tue, 5 Feb 2019 12:01:24 +0000 (17:31 +0530)
committerTom Rini <trini@konsulko.com>
Fri, 12 Apr 2019 00:07:12 +0000 (20:07 -0400)
The UDMA-P is intended to perform similar (but significantly upgraded) functions
as the packet-oriented DMA used on previous SoC devices. The UDMA-P module
supports the transmission and reception of various packet types.
The UDMA-P also supports acting as both a UTC and UDMA-C for its internal
channels. Channels in the UDMA-P can be configured to be either Packet-Based or
Third-Party channels on a channel by channel basis.

The initial driver supports:
- MEM_TO_MEM (TR mode)
- DEV_TO_MEM (Packet mode)
- MEM_TO_DEV (Packet mode)

Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
Signed-off-by: Grygorii Strashko <grygorii.strashko@ti.com>
Signed-off-by: Vignesh R <vigneshr@ti.com>
drivers/dma/Kconfig
drivers/dma/Makefile
drivers/dma/ti/Kconfig [new file with mode: 0644]
drivers/dma/ti/Makefile [new file with mode: 0644]
drivers/dma/ti/k3-udma-hwdef.h [new file with mode: 0644]
drivers/dma/ti/k3-udma.c [new file with mode: 0644]
include/dt-bindings/dma/k3-udma.h [new file with mode: 0644]
include/linux/soc/ti/ti-udma.h [new file with mode: 0644]

index 1820676d7a18d089571c7c4f38fc1b996604e82c..4f37ba7d35eb9776f7a8aff341693e40eff432bd 100644 (file)
@@ -57,4 +57,6 @@ config APBH_DMA_BURST8
 
 endif
 
+source "drivers/dma/ti/Kconfig"
+
 endmenu # menu "DMA Support"
index b5f9147e0a54a0868f4ab639c4495a5c02885ff1..afab324461b906c560e139b8e811bacc8ca70cd3 100644 (file)
@@ -13,3 +13,5 @@ obj-$(CONFIG_SANDBOX_DMA) += sandbox-dma-test.o
 obj-$(CONFIG_TI_KSNAV) += keystone_nav.o keystone_nav_cfg.o
 obj-$(CONFIG_TI_EDMA3) += ti-edma3.o
 obj-$(CONFIG_DMA_LPC32XX) += lpc32xx_dma.o
+
+obj-y += ti/
diff --git a/drivers/dma/ti/Kconfig b/drivers/dma/ti/Kconfig
new file mode 100644 (file)
index 0000000..3d54983
--- /dev/null
@@ -0,0 +1,14 @@
+# SPDX-License-Identifier: GPL-2.0+
+
+if ARCH_K3
+
+config TI_K3_NAVSS_UDMA
+        bool "Texas Instruments UDMA"
+        depends on ARCH_K3
+        select DMA
+        select TI_K3_NAVSS_RINGACC
+        select TI_K3_NAVSS_PSILCFG
+        default n
+        help
+          Support for UDMA used in K3 devices.
+endif
diff --git a/drivers/dma/ti/Makefile b/drivers/dma/ti/Makefile
new file mode 100644 (file)
index 0000000..de2f9ac
--- /dev/null
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0+
+
+obj-$(CONFIG_TI_K3_NAVSS_UDMA) += k3-udma.o
diff --git a/drivers/dma/ti/k3-udma-hwdef.h b/drivers/dma/ti/k3-udma-hwdef.h
new file mode 100644 (file)
index 0000000..c88399a
--- /dev/null
@@ -0,0 +1,184 @@
+/* SPDX-License-Identifier: GPL-2.0+ */
+/*
+ *  Copyright (C) 2018 Texas Instruments Incorporated - http://www.ti.com
+ *
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ */
+
+#ifndef K3_NAVSS_UDMA_HWDEF_H_
+#define K3_NAVSS_UDMA_HWDEF_H_
+
+#define UDMA_PSIL_DST_THREAD_ID_OFFSET 0x8000
+
+/* Global registers */
+#define UDMA_REV_REG                   0x0
+#define UDMA_PERF_CTL_REG              0x4
+#define UDMA_EMU_CTL_REG               0x8
+#define UDMA_PSIL_TO_REG               0x10
+#define UDMA_UTC_CTL_REG               0x1c
+#define UDMA_CAP_REG(i)                        (0x20 + (i * 4))
+#define UDMA_RX_FLOW_ID_FW_OES_REG     0x80
+#define UDMA_RX_FLOW_ID_FW_STATUS_REG  0x88
+
+/* RX Flow regs */
+#define UDMA_RFLOW_RFA_REG             0x0
+#define UDMA_RFLOW_RFB_REG             0x4
+#define UDMA_RFLOW_RFC_REG             0x8
+#define UDMA_RFLOW_RFD_REG             0xc
+#define UDMA_RFLOW_RFE_REG             0x10
+#define UDMA_RFLOW_RFF_REG             0x14
+#define UDMA_RFLOW_RFG_REG             0x18
+#define UDMA_RFLOW_RFH_REG             0x1c
+
+#define UDMA_RFLOW_REG(x) (UDMA_RFLOW_RF##x##_REG)
+
+/* TX chan regs */
+#define UDMA_TCHAN_TCFG_REG            0x0
+#define UDMA_TCHAN_TCREDIT_REG         0x4
+#define UDMA_TCHAN_TCQ_REG             0x14
+#define UDMA_TCHAN_TOES_REG(i)         (0x20 + (i) * 4)
+#define UDMA_TCHAN_TEOES_REG           0x60
+#define UDMA_TCHAN_TPRI_CTRL_REG       0x64
+#define UDMA_TCHAN_THREAD_ID_REG       0x68
+#define UDMA_TCHAN_TFIFO_DEPTH_REG     0x70
+#define UDMA_TCHAN_TST_SCHED_REG       0x80
+
+/* RX chan regs */
+#define UDMA_RCHAN_RCFG_REG            0x0
+#define UDMA_RCHAN_RCQ_REG             0x14
+#define UDMA_RCHAN_ROES_REG(i)         (0x20 + (i) * 4)
+#define UDMA_RCHAN_REOES_REG           0x60
+#define UDMA_RCHAN_RPRI_CTRL_REG       0x64
+#define UDMA_RCHAN_THREAD_ID_REG       0x68
+#define UDMA_RCHAN_RST_SCHED_REG       0x80
+#define UDMA_RCHAN_RFLOW_RNG_REG       0xf0
+
+/* TX chan RT regs */
+#define UDMA_TCHAN_RT_CTL_REG          0x0
+#define UDMA_TCHAN_RT_SWTRIG_REG       0x8
+#define UDMA_TCHAN_RT_STDATA_REG       0x80
+
+#define UDMA_TCHAN_RT_PEERn_REG(i)     (0x200 + (i * 0x4))
+#define UDMA_TCHAN_RT_PEER_STATIC_TR_XY_REG    \
+       UDMA_TCHAN_RT_PEERn_REG(0)      /* PSI-L: 0x400 */
+#define UDMA_TCHAN_RT_PEER_STATIC_TR_Z_REG     \
+       UDMA_TCHAN_RT_PEERn_REG(1)      /* PSI-L: 0x401 */
+#define UDMA_TCHAN_RT_PEER_BCNT_REG            \
+       UDMA_TCHAN_RT_PEERn_REG(4)      /* PSI-L: 0x404 */
+#define UDMA_TCHAN_RT_PEER_RT_EN_REG           \
+       UDMA_TCHAN_RT_PEERn_REG(8)      /* PSI-L: 0x408 */
+
+#define UDMA_TCHAN_RT_PCNT_REG         0x400
+#define UDMA_TCHAN_RT_BCNT_REG         0x408
+#define UDMA_TCHAN_RT_SBCNT_REG                0x410
+
+/* RX chan RT regs */
+#define UDMA_RCHAN_RT_CTL_REG          0x0
+#define UDMA_RCHAN_RT_SWTRIG_REG       0x8
+#define UDMA_RCHAN_RT_STDATA_REG       0x80
+
+#define UDMA_RCHAN_RT_PEERn_REG(i)     (0x200 + (i * 0x4))
+#define UDMA_RCHAN_RT_PEER_STATIC_TR_XY_REG    \
+       UDMA_RCHAN_RT_PEERn_REG(0)      /* PSI-L: 0x400 */
+#define UDMA_RCHAN_RT_PEER_STATIC_TR_Z_REG     \
+       UDMA_RCHAN_RT_PEERn_REG(1)      /* PSI-L: 0x401 */
+#define UDMA_RCHAN_RT_PEER_BCNT_REG            \
+       UDMA_RCHAN_RT_PEERn_REG(4)      /* PSI-L: 0x404 */
+#define UDMA_RCHAN_RT_PEER_RT_EN_REG           \
+       UDMA_RCHAN_RT_PEERn_REG(8)      /* PSI-L: 0x408 */
+
+#define UDMA_RCHAN_RT_PCNT_REG         0x400
+#define UDMA_RCHAN_RT_BCNT_REG         0x408
+#define UDMA_RCHAN_RT_SBCNT_REG                0x410
+
+/* UDMA_TCHAN_TCFG_REG/UDMA_RCHAN_RCFG_REG */
+#define UDMA_CHAN_CFG_PAUSE_ON_ERR             BIT(31)
+#define UDMA_TCHAN_CFG_FILT_EINFO              BIT(30)
+#define UDMA_TCHAN_CFG_FILT_PSWORDS            BIT(29)
+#define UDMA_CHAN_CFG_ATYPE_MASK               GENMASK(25, 24)
+#define UDMA_CHAN_CFG_ATYPE_SHIFT              24
+#define UDMA_CHAN_CFG_CHAN_TYPE_MASK           GENMASK(19, 16)
+#define UDMA_CHAN_CFG_CHAN_TYPE_SHIFT          16
+/*
+ * PBVR - using pass by value rings
+ * PBRR - using pass by reference rings
+ * 3RDP - Third Party DMA
+ * BC - Block Copy
+ * SB - single buffer packet mode enabled
+ */
+#define UDMA_CHAN_CFG_CHAN_TYPE_PACKET_PBRR \
+       (2 << UDMA_CHAN_CFG_CHAN_TYPE_SHIFT)
+#define UDMA_CHAN_CFG_CHAN_TYPE_PACKET_SB_PBRR \
+       (3 << UDMA_CHAN_CFG_CHAN_TYPE_SHIFT)
+#define UDMA_CHAN_CFG_CHAN_TYPE_3RDP_PBRR \
+       (10 << UDMA_CHAN_CFG_CHAN_TYPE_SHIFT)
+#define UDMA_CHAN_CFG_CHAN_TYPE_3RDP_PBVR \
+       (11 << UDMA_CHAN_CFG_CHAN_TYPE_SHIFT)
+#define UDMA_CHAN_CFG_CHAN_TYPE_3RDP_BC_PBRR \
+       (12 << UDMA_CHAN_CFG_CHAN_TYPE_SHIFT)
+#define UDMA_RCHAN_CFG_IGNORE_SHORT            BIT(15)
+#define UDMA_RCHAN_CFG_IGNORE_LONG             BIT(14)
+#define UDMA_TCHAN_CFG_SUPR_TDPKT              BIT(8)
+#define UDMA_CHAN_CFG_FETCH_SIZE_MASK          GENMASK(6, 0)
+#define UDMA_CHAN_CFG_FETCH_SIZE_SHIFT         0
+
+/* UDMA_TCHAN_RT_CTL_REG/UDMA_RCHAN_RT_CTL_REG */
+#define UDMA_CHAN_RT_CTL_EN    BIT(31)
+#define UDMA_CHAN_RT_CTL_TDOWN BIT(30)
+#define UDMA_CHAN_RT_CTL_PAUSE BIT(29)
+#define UDMA_CHAN_RT_CTL_FTDOWN        BIT(28)
+#define UDMA_CHAN_RT_CTL_ERROR BIT(0)
+
+/* UDMA_TCHAN_RT_PEER_RT_EN_REG/UDMA_RCHAN_RT_PEER_RT_EN_REG (PSI-L: 0x408) */
+#define UDMA_PEER_RT_EN_ENABLE         BIT(31)
+#define UDMA_PEER_RT_EN_TEARDOWN       BIT(30)
+#define UDMA_PEER_RT_EN_PAUSE          BIT(29)
+#define UDMA_PEER_RT_EN_FLUSH          BIT(28)
+#define UDMA_PEER_RT_EN_IDLE           BIT(1)
+
+/* RX Flow reg RFA */
+#define UDMA_RFLOW_RFA_EINFO                   BIT(30)
+#define UDMA_RFLOW_RFA_PSINFO                  BIT(29)
+#define UDMA_RFLOW_RFA_ERR_HANDLING            BIT(28)
+#define UDMA_RFLOW_RFA_DESC_TYPE_MASK          GENMASK(27, 26)
+#define UDMA_RFLOW_RFA_DESC_TYPE_SHIFT         26
+#define UDMA_RFLOW_RFA_PS_LOC                  BIT(25)
+#define UDMA_RFLOW_RFA_SOP_OFF_MASK            GENMASK(24, 16)
+#define UDMA_RFLOW_RFA_SOP_OFF_SHIFT           16
+#define UDMA_RFLOW_RFA_DEST_QNUM_MASK          GENMASK(15, 0)
+#define UDMA_RFLOW_RFA_DEST_QNUM_SHIFT         0
+
+/* RX Flow reg RFC */
+#define UDMA_RFLOW_RFC_SRC_TAG_HI_SEL_SHIFT    28
+#define UDMA_RFLOW_RFC_SRC_TAG_LO_SEL_SHIFT    24
+#define UDMA_RFLOW_RFC_DST_TAG_HI_SEL_SHIFT    20
+#define UDMA_RFLOW_RFC_DST_TAG_LO_SE_SHIFT     16
+
+/*
+ * UDMA_TCHAN_RT_PEER_STATIC_TR_XY_REG /
+ * UDMA_RCHAN_RT_PEER_STATIC_TR_XY_REG
+ */
+#define PDMA_STATIC_TR_X_MASK          GENMASK(26, 24)
+#define PDMA_STATIC_TR_X_SHIFT         (24)
+#define PDMA_STATIC_TR_Y_MASK          GENMASK(11, 0)
+#define PDMA_STATIC_TR_Y_SHIFT         (0)
+
+#define PDMA_STATIC_TR_Y(x)    \
+       (((x) << PDMA_STATIC_TR_Y_SHIFT) & PDMA_STATIC_TR_Y_MASK)
+#define PDMA_STATIC_TR_X(x)    \
+       (((x) << PDMA_STATIC_TR_X_SHIFT) & PDMA_STATIC_TR_X_MASK)
+
+/*
+ * UDMA_TCHAN_RT_PEER_STATIC_TR_Z_REG /
+ * UDMA_RCHAN_RT_PEER_STATIC_TR_Z_REG
+ */
+#define PDMA_STATIC_TR_Z_MASK          GENMASK(11, 0)
+#define PDMA_STATIC_TR_Z_SHIFT         (0)
+#define PDMA_STATIC_TR_Z(x)    \
+       (((x) << PDMA_STATIC_TR_Z_SHIFT) & PDMA_STATIC_TR_Z_MASK)
+
+#endif /* K3_NAVSS_UDMA_HWDEF_H_ */
diff --git a/drivers/dma/ti/k3-udma.c b/drivers/dma/ti/k3-udma.c
new file mode 100644 (file)
index 0000000..f78a01a
--- /dev/null
@@ -0,0 +1,1730 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ *  Copyright (C) 2018 Texas Instruments Incorporated - http://www.ti.com
+ *  Author: Peter Ujfalusi <peter.ujfalusi@ti.com>
+ */
+#define pr_fmt(fmt) "udma: " fmt
+
+#include <common.h>
+#include <asm/io.h>
+#include <asm/bitops.h>
+#include <malloc.h>
+#include <asm/dma-mapping.h>
+#include <dm.h>
+#include <dm/read.h>
+#include <dm/of_access.h>
+#include <dma.h>
+#include <dma-uclass.h>
+#include <linux/delay.h>
+#include <dt-bindings/dma/k3-udma.h>
+#include <linux/soc/ti/k3-navss-ringacc.h>
+#include <linux/soc/ti/cppi5.h>
+#include <linux/soc/ti/ti-udma.h>
+#include <linux/soc/ti/ti_sci_protocol.h>
+
+#include "k3-udma-hwdef.h"
+
+#if BITS_PER_LONG == 64
+#define RINGACC_RING_USE_PROXY (0)
+#else
+#define RINGACC_RING_USE_PROXY (1)
+#endif
+
+struct udma_chan;
+
+enum udma_mmr {
+       MMR_GCFG = 0,
+       MMR_RCHANRT,
+       MMR_TCHANRT,
+       MMR_LAST,
+};
+
+static const char * const mmr_names[] = {
+       "gcfg", "rchanrt", "tchanrt"
+};
+
+struct udma_tchan {
+       void __iomem *reg_rt;
+
+       int id;
+       struct k3_nav_ring *t_ring; /* Transmit ring */
+       struct k3_nav_ring *tc_ring; /* Transmit Completion ring */
+};
+
+struct udma_rchan {
+       void __iomem *reg_rt;
+
+       int id;
+       struct k3_nav_ring *fd_ring; /* Free Descriptor ring */
+       struct k3_nav_ring *r_ring; /* Receive ring*/
+};
+
+struct udma_rflow {
+       int id;
+};
+
+struct udma_dev {
+       struct device *dev;
+       void __iomem *mmrs[MMR_LAST];
+
+       struct k3_nav_ringacc *ringacc;
+
+       u32 features;
+
+       int tchan_cnt;
+       int echan_cnt;
+       int rchan_cnt;
+       int rflow_cnt;
+       unsigned long *tchan_map;
+       unsigned long *rchan_map;
+       unsigned long *rflow_map;
+
+       struct udma_tchan *tchans;
+       struct udma_rchan *rchans;
+       struct udma_rflow *rflows;
+
+       struct udma_chan *channels;
+       u32 psil_base;
+
+       u32 ch_count;
+       const struct ti_sci_handle *tisci;
+       const struct ti_sci_rm_udmap_ops *tisci_udmap_ops;
+       const struct ti_sci_rm_psil_ops *tisci_psil_ops;
+       u32  tisci_dev_id;
+       u32  tisci_navss_dev_id;
+       bool is_coherent;
+};
+
+struct udma_chan {
+       struct udma_dev *ud;
+       char name[20];
+
+       struct udma_tchan *tchan;
+       struct udma_rchan *rchan;
+       struct udma_rflow *rflow;
+
+       u32 bcnt; /* number of bytes completed since the start of the channel */
+
+       bool pkt_mode; /* TR or packet */
+       bool needs_epib; /* EPIB is needed for the communication or not */
+       u32 psd_size; /* size of Protocol Specific Data */
+       u32 metadata_size; /* (needs_epib ? 16:0) + psd_size */
+       int slave_thread_id;
+       u32 src_thread;
+       u32 dst_thread;
+       u32 static_tr_type;
+
+       u32 id;
+       enum dma_direction dir;
+
+       struct cppi5_host_desc_t *desc_tx;
+       u32 hdesc_size;
+       bool in_use;
+       void    *desc_rx;
+       u32     num_rx_bufs;
+       u32     desc_rx_cur;
+
+};
+
+#define UDMA_CH_1000(ch)               (ch * 0x1000)
+#define UDMA_CH_100(ch)                        (ch * 0x100)
+#define UDMA_CH_40(ch)                 (ch * 0x40)
+
+#ifdef PKTBUFSRX
+#define UDMA_RX_DESC_NUM PKTBUFSRX
+#else
+#define UDMA_RX_DESC_NUM 4
+#endif
+
+/* Generic register access functions */
+static inline u32 udma_read(void __iomem *base, int reg)
+{
+       u32 v;
+
+       v = __raw_readl(base + reg);
+       pr_debug("READL(32): v(%08X)<--reg(%p)\n", v, base + reg);
+       return v;
+}
+
+static inline void udma_write(void __iomem *base, int reg, u32 val)
+{
+       pr_debug("WRITEL(32): v(%08X)-->reg(%p)\n", val, base + reg);
+       __raw_writel(val, base + reg);
+}
+
+static inline void udma_update_bits(void __iomem *base, int reg,
+                                   u32 mask, u32 val)
+{
+       u32 tmp, orig;
+
+       orig = udma_read(base, reg);
+       tmp = orig & ~mask;
+       tmp |= (val & mask);
+
+       if (tmp != orig)
+               udma_write(base, reg, tmp);
+}
+
+/* TCHANRT */
+static inline u32 udma_tchanrt_read(struct udma_tchan *tchan, int reg)
+{
+       if (!tchan)
+               return 0;
+       return udma_read(tchan->reg_rt, reg);
+}
+
+static inline void udma_tchanrt_write(struct udma_tchan *tchan,
+                                     int reg, u32 val)
+{
+       if (!tchan)
+               return;
+       udma_write(tchan->reg_rt, reg, val);
+}
+
+/* RCHANRT */
+static inline u32 udma_rchanrt_read(struct udma_rchan *rchan, int reg)
+{
+       if (!rchan)
+               return 0;
+       return udma_read(rchan->reg_rt, reg);
+}
+
+static inline void udma_rchanrt_write(struct udma_rchan *rchan,
+                                     int reg, u32 val)
+{
+       if (!rchan)
+               return;
+       udma_write(rchan->reg_rt, reg, val);
+}
+
+static inline int udma_navss_psil_pair(struct udma_dev *ud, u32 src_thread,
+                                      u32 dst_thread)
+{
+       dst_thread |= UDMA_PSIL_DST_THREAD_ID_OFFSET;
+       return ud->tisci_psil_ops->pair(ud->tisci,
+                                       ud->tisci_navss_dev_id,
+                                       src_thread, dst_thread);
+}
+
+static inline int udma_navss_psil_unpair(struct udma_dev *ud, u32 src_thread,
+                                        u32 dst_thread)
+{
+       dst_thread |= UDMA_PSIL_DST_THREAD_ID_OFFSET;
+       return ud->tisci_psil_ops->unpair(ud->tisci,
+                                         ud->tisci_navss_dev_id,
+                                         src_thread, dst_thread);
+}
+
+static inline char *udma_get_dir_text(enum dma_direction dir)
+{
+       switch (dir) {
+       case DMA_DEV_TO_MEM:
+               return "DEV_TO_MEM";
+       case DMA_MEM_TO_DEV:
+               return "MEM_TO_DEV";
+       case DMA_MEM_TO_MEM:
+               return "MEM_TO_MEM";
+       case DMA_DEV_TO_DEV:
+               return "DEV_TO_DEV";
+       default:
+               break;
+       }
+
+       return "invalid";
+}
+
+static inline bool udma_is_chan_running(struct udma_chan *uc)
+{
+       u32 trt_ctl = 0;
+       u32 rrt_ctl = 0;
+
+       switch (uc->dir) {
+       case DMA_DEV_TO_MEM:
+               rrt_ctl = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_CTL_REG);
+               pr_debug("%s: rrt_ctl: 0x%08x (peer: 0x%08x)\n",
+                        __func__, rrt_ctl,
+                        udma_rchanrt_read(uc->rchan,
+                                          UDMA_RCHAN_RT_PEER_RT_EN_REG));
+               break;
+       case DMA_MEM_TO_DEV:
+               trt_ctl = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
+               pr_debug("%s: trt_ctl: 0x%08x (peer: 0x%08x)\n",
+                        __func__, trt_ctl,
+                        udma_tchanrt_read(uc->tchan,
+                                          UDMA_TCHAN_RT_PEER_RT_EN_REG));
+               break;
+       case DMA_MEM_TO_MEM:
+               trt_ctl = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
+               rrt_ctl = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_CTL_REG);
+               break;
+       default:
+               break;
+       }
+
+       if (trt_ctl & UDMA_CHAN_RT_CTL_EN || rrt_ctl & UDMA_CHAN_RT_CTL_EN)
+               return true;
+
+       return false;
+}
+
+static int udma_is_coherent(struct udma_chan *uc)
+{
+       return uc->ud->is_coherent;
+}
+
+static int udma_pop_from_ring(struct udma_chan *uc, dma_addr_t *addr)
+{
+       struct k3_nav_ring *ring = NULL;
+       int ret = -ENOENT;
+
+       switch (uc->dir) {
+       case DMA_DEV_TO_MEM:
+               ring = uc->rchan->r_ring;
+               break;
+       case DMA_MEM_TO_DEV:
+               ring = uc->tchan->tc_ring;
+               break;
+       case DMA_MEM_TO_MEM:
+               ring = uc->tchan->tc_ring;
+               break;
+       default:
+               break;
+       }
+
+       if (ring && k3_nav_ringacc_ring_get_occ(ring))
+               ret = k3_nav_ringacc_ring_pop(ring, addr);
+
+       return ret;
+}
+
+static void udma_reset_rings(struct udma_chan *uc)
+{
+       struct k3_nav_ring *ring1 = NULL;
+       struct k3_nav_ring *ring2 = NULL;
+
+       switch (uc->dir) {
+       case DMA_DEV_TO_MEM:
+               ring1 = uc->rchan->fd_ring;
+               ring2 = uc->rchan->r_ring;
+               break;
+       case DMA_MEM_TO_DEV:
+               ring1 = uc->tchan->t_ring;
+               ring2 = uc->tchan->tc_ring;
+               break;
+       case DMA_MEM_TO_MEM:
+               ring1 = uc->tchan->t_ring;
+               ring2 = uc->tchan->tc_ring;
+               break;
+       default:
+               break;
+       }
+
+       if (ring1)
+               k3_nav_ringacc_ring_reset_dma(ring1, 0);
+       if (ring2)
+               k3_nav_ringacc_ring_reset(ring2);
+}
+
+static void udma_reset_counters(struct udma_chan *uc)
+{
+       u32 val;
+
+       if (uc->tchan) {
+               val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_BCNT_REG);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_BCNT_REG, val);
+
+               val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_SBCNT_REG);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_SBCNT_REG, val);
+
+               val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PCNT_REG);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PCNT_REG, val);
+
+               val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_BCNT_REG, val);
+       }
+
+       if (uc->rchan) {
+               val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_BCNT_REG);
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_BCNT_REG, val);
+
+               val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_SBCNT_REG);
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_SBCNT_REG, val);
+
+               val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PCNT_REG);
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PCNT_REG, val);
+
+               val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PEER_BCNT_REG);
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_BCNT_REG, val);
+       }
+
+       uc->bcnt = 0;
+}
+
+static inline int udma_stop_hard(struct udma_chan *uc)
+{
+       pr_debug("%s: ENTER (chan%d)\n", __func__, uc->id);
+
+       switch (uc->dir) {
+       case DMA_DEV_TO_MEM:
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG, 0);
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
+               break;
+       case DMA_MEM_TO_DEV:
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG, 0);
+               break;
+       case DMA_MEM_TO_MEM:
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static int udma_start(struct udma_chan *uc)
+{
+       /* Channel is already running, no need to proceed further */
+       if (udma_is_chan_running(uc))
+               goto out;
+
+       pr_debug("%s: chan:%d dir:%s (static_tr_type: %d)\n",
+                __func__, uc->id, udma_get_dir_text(uc->dir),
+                uc->static_tr_type);
+
+       /* Make sure that we clear the teardown bit, if it is set */
+       udma_stop_hard(uc);
+
+       /* Reset all counters */
+       udma_reset_counters(uc);
+
+       switch (uc->dir) {
+       case DMA_DEV_TO_MEM:
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
+                                  UDMA_CHAN_RT_CTL_EN);
+
+               /* Enable remote */
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG,
+                                  UDMA_PEER_RT_EN_ENABLE);
+
+               pr_debug("%s(rx): RT_CTL:0x%08x PEER RT_ENABLE:0x%08x\n",
+                        __func__,
+                        udma_rchanrt_read(uc->rchan,
+                                          UDMA_RCHAN_RT_CTL_REG),
+                        udma_rchanrt_read(uc->rchan,
+                                          UDMA_RCHAN_RT_PEER_RT_EN_REG));
+               break;
+       case DMA_MEM_TO_DEV:
+               /* Enable remote */
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG,
+                                  UDMA_PEER_RT_EN_ENABLE);
+
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
+                                  UDMA_CHAN_RT_CTL_EN);
+
+               pr_debug("%s(tx): RT_CTL:0x%08x PEER RT_ENABLE:0x%08x\n",
+                        __func__,
+                        udma_rchanrt_read(uc->rchan,
+                                          UDMA_TCHAN_RT_CTL_REG),
+                        udma_rchanrt_read(uc->rchan,
+                                          UDMA_TCHAN_RT_PEER_RT_EN_REG));
+               break;
+       case DMA_MEM_TO_MEM:
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG,
+                                  UDMA_CHAN_RT_CTL_EN);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
+                                  UDMA_CHAN_RT_CTL_EN);
+
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       pr_debug("%s: DONE chan:%d\n", __func__, uc->id);
+out:
+       return 0;
+}
+
+static inline void udma_stop_mem2dev(struct udma_chan *uc, bool sync)
+{
+       int i = 0;
+       u32 val;
+
+       udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG,
+                          UDMA_CHAN_RT_CTL_EN |
+                          UDMA_CHAN_RT_CTL_TDOWN);
+
+       val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
+
+       while (sync && (val & UDMA_CHAN_RT_CTL_EN)) {
+               val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_CTL_REG);
+               udelay(1);
+               if (i > 1000) {
+                       printf(" %s TIMEOUT !\n", __func__);
+                       break;
+               }
+               i++;
+       }
+
+       val = udma_tchanrt_read(uc->tchan, UDMA_TCHAN_RT_PEER_RT_EN_REG);
+       if (val & UDMA_PEER_RT_EN_ENABLE)
+               printf("%s: peer not stopped TIMEOUT !\n", __func__);
+}
+
+static inline void udma_stop_dev2mem(struct udma_chan *uc, bool sync)
+{
+       int i = 0;
+       u32 val;
+
+       udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG,
+                          UDMA_PEER_RT_EN_ENABLE |
+                          UDMA_PEER_RT_EN_TEARDOWN);
+
+       val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_CTL_REG);
+
+       while (sync && (val & UDMA_CHAN_RT_CTL_EN)) {
+               val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_CTL_REG);
+               udelay(1);
+               if (i > 1000) {
+                       printf("%s TIMEOUT !\n", __func__);
+                       break;
+               }
+               i++;
+       }
+
+       val = udma_rchanrt_read(uc->rchan, UDMA_RCHAN_RT_PEER_RT_EN_REG);
+       if (val & UDMA_PEER_RT_EN_ENABLE)
+               printf("%s: peer not stopped TIMEOUT !\n", __func__);
+}
+
+static inline int udma_stop(struct udma_chan *uc)
+{
+       pr_debug("%s: chan:%d dir:%s\n",
+                __func__, uc->id, udma_get_dir_text(uc->dir));
+
+       udma_reset_counters(uc);
+       switch (uc->dir) {
+       case DMA_DEV_TO_MEM:
+               udma_stop_dev2mem(uc, true);
+               break;
+       case DMA_MEM_TO_DEV:
+               udma_stop_mem2dev(uc, true);
+               break;
+       case DMA_MEM_TO_MEM:
+               udma_rchanrt_write(uc->rchan, UDMA_RCHAN_RT_CTL_REG, 0);
+               udma_tchanrt_write(uc->tchan, UDMA_TCHAN_RT_CTL_REG, 0);
+               break;
+       default:
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
+static void udma_poll_completion(struct udma_chan *uc, dma_addr_t *paddr)
+{
+       int i = 1;
+
+       while (udma_pop_from_ring(uc, paddr)) {
+               udelay(1);
+               if (!(i % 1000000))
+                       printf(".");
+               i++;
+       }
+}
+
+#define UDMA_RESERVE_RESOURCE(res)                                     \
+static struct udma_##res *__udma_reserve_##res(struct udma_dev *ud,    \
+                                              int id)                  \
+{                                                                      \
+       if (id >= 0) {                                                  \
+               if (test_bit(id, ud->res##_map)) {                      \
+                       dev_err(ud->dev, "res##%d is in use\n", id);    \
+                       return ERR_PTR(-ENOENT);                        \
+               }                                                       \
+       } else {                                                        \
+               id = find_first_zero_bit(ud->res##_map, ud->res##_cnt); \
+               if (id == ud->res##_cnt) {                              \
+                       return ERR_PTR(-ENOENT);                        \
+               }                                                       \
+       }                                                               \
+                                                                       \
+       __set_bit(id, ud->res##_map);                                   \
+       return &ud->res##s[id];                                         \
+}
+
+UDMA_RESERVE_RESOURCE(tchan);
+UDMA_RESERVE_RESOURCE(rchan);
+UDMA_RESERVE_RESOURCE(rflow);
+
+static int udma_get_tchan(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+
+       if (uc->tchan) {
+               dev_dbg(ud->dev, "chan%d: already have tchan%d allocated\n",
+                       uc->id, uc->tchan->id);
+               return 0;
+       }
+
+       uc->tchan = __udma_reserve_tchan(ud, -1);
+       if (IS_ERR(uc->tchan))
+               return PTR_ERR(uc->tchan);
+
+       pr_debug("chan%d: got tchan%d\n", uc->id, uc->tchan->id);
+
+       if (udma_is_chan_running(uc)) {
+               dev_warn(ud->dev, "chan%d: tchan%d is running!\n", uc->id,
+                        uc->tchan->id);
+               udma_stop(uc);
+               if (udma_is_chan_running(uc))
+                       dev_err(ud->dev, "chan%d: won't stop!\n", uc->id);
+       }
+
+       return 0;
+}
+
+static int udma_get_rchan(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+
+       if (uc->rchan) {
+               dev_dbg(ud->dev, "chan%d: already have rchan%d allocated\n",
+                       uc->id, uc->rchan->id);
+               return 0;
+       }
+
+       uc->rchan = __udma_reserve_rchan(ud, -1);
+       if (IS_ERR(uc->rchan))
+               return PTR_ERR(uc->rchan);
+
+       pr_debug("chan%d: got rchan%d\n", uc->id, uc->rchan->id);
+
+       if (udma_is_chan_running(uc)) {
+               dev_warn(ud->dev, "chan%d: rchan%d is running!\n", uc->id,
+                        uc->rchan->id);
+               udma_stop(uc);
+               if (udma_is_chan_running(uc))
+                       dev_err(ud->dev, "chan%d: won't stop!\n", uc->id);
+       }
+
+       return 0;
+}
+
+static int udma_get_chan_pair(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+       int chan_id, end;
+
+       if ((uc->tchan && uc->rchan) && uc->tchan->id == uc->rchan->id) {
+               dev_info(ud->dev, "chan%d: already have %d pair allocated\n",
+                        uc->id, uc->tchan->id);
+               return 0;
+       }
+
+       if (uc->tchan) {
+               dev_err(ud->dev, "chan%d: already have tchan%d allocated\n",
+                       uc->id, uc->tchan->id);
+               return -EBUSY;
+       } else if (uc->rchan) {
+               dev_err(ud->dev, "chan%d: already have rchan%d allocated\n",
+                       uc->id, uc->rchan->id);
+               return -EBUSY;
+       }
+
+       /* Can be optimized, but let's have it like this for now */
+       end = min(ud->tchan_cnt, ud->rchan_cnt);
+       for (chan_id = 0; chan_id < end; chan_id++) {
+               if (!test_bit(chan_id, ud->tchan_map) &&
+                   !test_bit(chan_id, ud->rchan_map))
+                       break;
+       }
+
+       if (chan_id == end)
+               return -ENOENT;
+
+       __set_bit(chan_id, ud->tchan_map);
+       __set_bit(chan_id, ud->rchan_map);
+       uc->tchan = &ud->tchans[chan_id];
+       uc->rchan = &ud->rchans[chan_id];
+
+       pr_debug("chan%d: got t/rchan%d pair\n", uc->id, chan_id);
+
+       if (udma_is_chan_running(uc)) {
+               dev_warn(ud->dev, "chan%d: t/rchan%d pair is running!\n",
+                        uc->id, chan_id);
+               udma_stop(uc);
+               if (udma_is_chan_running(uc))
+                       dev_err(ud->dev, "chan%d: won't stop!\n", uc->id);
+       }
+
+       return 0;
+}
+
+static int udma_get_rflow(struct udma_chan *uc, int flow_id)
+{
+       struct udma_dev *ud = uc->ud;
+
+       if (uc->rflow) {
+               dev_dbg(ud->dev, "chan%d: already have rflow%d allocated\n",
+                       uc->id, uc->rflow->id);
+               return 0;
+       }
+
+       if (!uc->rchan)
+               dev_warn(ud->dev, "chan%d: does not have rchan??\n", uc->id);
+
+       uc->rflow = __udma_reserve_rflow(ud, flow_id);
+       if (IS_ERR(uc->rflow))
+               return PTR_ERR(uc->rflow);
+
+       pr_debug("chan%d: got rflow%d\n", uc->id, uc->rflow->id);
+       return 0;
+}
+
+static void udma_put_rchan(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+
+       if (uc->rchan) {
+               dev_dbg(ud->dev, "chan%d: put rchan%d\n", uc->id,
+                       uc->rchan->id);
+               __clear_bit(uc->rchan->id, ud->rchan_map);
+               uc->rchan = NULL;
+       }
+}
+
+static void udma_put_tchan(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+
+       if (uc->tchan) {
+               dev_dbg(ud->dev, "chan%d: put tchan%d\n", uc->id,
+                       uc->tchan->id);
+               __clear_bit(uc->tchan->id, ud->tchan_map);
+               uc->tchan = NULL;
+       }
+}
+
+static void udma_put_rflow(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+
+       if (uc->rflow) {
+               dev_dbg(ud->dev, "chan%d: put rflow%d\n", uc->id,
+                       uc->rflow->id);
+               __clear_bit(uc->rflow->id, ud->rflow_map);
+               uc->rflow = NULL;
+       }
+}
+
+static void udma_free_tx_resources(struct udma_chan *uc)
+{
+       if (!uc->tchan)
+               return;
+
+       k3_nav_ringacc_ring_free(uc->tchan->t_ring);
+       k3_nav_ringacc_ring_free(uc->tchan->tc_ring);
+       uc->tchan->t_ring = NULL;
+       uc->tchan->tc_ring = NULL;
+
+       udma_put_tchan(uc);
+}
+
+static int udma_alloc_tx_resources(struct udma_chan *uc)
+{
+       struct k3_nav_ring_cfg ring_cfg;
+       struct udma_dev *ud = uc->ud;
+       int ret;
+
+       ret = udma_get_tchan(uc);
+       if (ret)
+               return ret;
+
+       uc->tchan->t_ring = k3_nav_ringacc_request_ring(
+                               ud->ringacc, uc->tchan->id,
+                               RINGACC_RING_USE_PROXY);
+       if (!uc->tchan->t_ring) {
+               ret = -EBUSY;
+               goto err_tx_ring;
+       }
+
+       uc->tchan->tc_ring = k3_nav_ringacc_request_ring(
+                               ud->ringacc, -1, RINGACC_RING_USE_PROXY);
+       if (!uc->tchan->tc_ring) {
+               ret = -EBUSY;
+               goto err_txc_ring;
+       }
+
+       memset(&ring_cfg, 0, sizeof(ring_cfg));
+       ring_cfg.size = 16;
+       ring_cfg.elm_size = K3_NAV_RINGACC_RING_ELSIZE_8;
+       ring_cfg.mode = K3_NAV_RINGACC_RING_MODE_MESSAGE;
+
+       ret = k3_nav_ringacc_ring_cfg(uc->tchan->t_ring, &ring_cfg);
+       ret |= k3_nav_ringacc_ring_cfg(uc->tchan->tc_ring, &ring_cfg);
+
+       if (ret)
+               goto err_ringcfg;
+
+       return 0;
+
+err_ringcfg:
+       k3_nav_ringacc_ring_free(uc->tchan->tc_ring);
+       uc->tchan->tc_ring = NULL;
+err_txc_ring:
+       k3_nav_ringacc_ring_free(uc->tchan->t_ring);
+       uc->tchan->t_ring = NULL;
+err_tx_ring:
+       udma_put_tchan(uc);
+
+       return ret;
+}
+
+static void udma_free_rx_resources(struct udma_chan *uc)
+{
+       if (!uc->rchan)
+               return;
+
+       k3_nav_ringacc_ring_free(uc->rchan->fd_ring);
+       k3_nav_ringacc_ring_free(uc->rchan->r_ring);
+       uc->rchan->fd_ring = NULL;
+       uc->rchan->r_ring = NULL;
+
+       udma_put_rflow(uc);
+       udma_put_rchan(uc);
+}
+
+static int udma_alloc_rx_resources(struct udma_chan *uc)
+{
+       struct k3_nav_ring_cfg ring_cfg;
+       struct udma_dev *ud = uc->ud;
+       int fd_ring_id;
+       int ret;
+
+       ret = udma_get_rchan(uc);
+       if (ret)
+               return ret;
+
+       /* For MEM_TO_MEM we don't need rflow or rings */
+       if (uc->dir == DMA_MEM_TO_MEM)
+               return 0;
+
+       ret = udma_get_rflow(uc, uc->rchan->id);
+       if (ret) {
+               ret = -EBUSY;
+               goto err_rflow;
+       }
+
+       fd_ring_id = ud->tchan_cnt + ud->echan_cnt + uc->rchan->id;
+
+       uc->rchan->fd_ring = k3_nav_ringacc_request_ring(
+                               ud->ringacc, fd_ring_id,
+                               RINGACC_RING_USE_PROXY);
+       if (!uc->rchan->fd_ring) {
+               ret = -EBUSY;
+               goto err_rx_ring;
+       }
+
+       uc->rchan->r_ring = k3_nav_ringacc_request_ring(
+                               ud->ringacc, -1, RINGACC_RING_USE_PROXY);
+       if (!uc->rchan->r_ring) {
+               ret = -EBUSY;
+               goto err_rxc_ring;
+       }
+
+       memset(&ring_cfg, 0, sizeof(ring_cfg));
+       ring_cfg.size = 16;
+       ring_cfg.elm_size = K3_NAV_RINGACC_RING_ELSIZE_8;
+       ring_cfg.mode = K3_NAV_RINGACC_RING_MODE_MESSAGE;
+
+       ret = k3_nav_ringacc_ring_cfg(uc->rchan->fd_ring, &ring_cfg);
+       ret |= k3_nav_ringacc_ring_cfg(uc->rchan->r_ring, &ring_cfg);
+
+       if (ret)
+               goto err_ringcfg;
+
+       return 0;
+
+err_ringcfg:
+       k3_nav_ringacc_ring_free(uc->rchan->r_ring);
+       uc->rchan->r_ring = NULL;
+err_rxc_ring:
+       k3_nav_ringacc_ring_free(uc->rchan->fd_ring);
+       uc->rchan->fd_ring = NULL;
+err_rx_ring:
+       udma_put_rflow(uc);
+err_rflow:
+       udma_put_rchan(uc);
+
+       return ret;
+}
+
+static int udma_alloc_tchan_sci_req(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+       int tc_ring = k3_nav_ringacc_get_ring_id(uc->tchan->tc_ring);
+       struct ti_sci_msg_rm_udmap_tx_ch_cfg req;
+       u32 mode;
+       int ret;
+
+       if (uc->pkt_mode)
+               mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
+       else
+               mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
+
+       req.valid_params = TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID;
+       req.nav_id = ud->tisci_dev_id;
+       req.index = uc->tchan->id;
+       req.tx_chan_type = mode;
+       if (uc->dir == DMA_MEM_TO_MEM)
+               req.tx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
+       else
+               req.tx_fetch_size = cppi5_hdesc_calc_size(uc->needs_epib,
+                                                         uc->psd_size,
+                                                         0) >> 2;
+       req.txcq_qnum = tc_ring;
+
+       ret = ud->tisci_udmap_ops->tx_ch_cfg(ud->tisci, &req);
+       if (ret)
+               dev_err(ud->dev, "tisci tx alloc failed %d\n", ret);
+
+       return ret;
+}
+
+static int udma_alloc_rchan_sci_req(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+       int fd_ring = k3_nav_ringacc_get_ring_id(uc->rchan->fd_ring);
+       int rx_ring = k3_nav_ringacc_get_ring_id(uc->rchan->r_ring);
+       int tc_ring = k3_nav_ringacc_get_ring_id(uc->tchan->tc_ring);
+       struct ti_sci_msg_rm_udmap_rx_ch_cfg req = { 0 };
+       struct ti_sci_msg_rm_udmap_flow_cfg flow_req = { 0 };
+       u32 mode;
+       int ret;
+
+       if (uc->pkt_mode)
+               mode = TI_SCI_RM_UDMAP_CHAN_TYPE_PKT_PBRR;
+       else
+               mode = TI_SCI_RM_UDMAP_CHAN_TYPE_3RDP_BCOPY_PBRR;
+
+       req.valid_params = TI_SCI_MSG_VALUE_RM_UDMAP_CH_FETCH_SIZE_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_CH_CQ_QNUM_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_CH_CHAN_TYPE_VALID;
+       req.nav_id = ud->tisci_dev_id;
+       req.index = uc->rchan->id;
+       req.rx_chan_type = mode;
+       if (uc->dir == DMA_MEM_TO_MEM) {
+               req.rx_fetch_size = sizeof(struct cppi5_desc_hdr_t) >> 2;
+               req.rxcq_qnum = tc_ring;
+       } else {
+               req.rx_fetch_size = cppi5_hdesc_calc_size(uc->needs_epib,
+                                                         uc->psd_size,
+                                                         0) >> 2;
+               req.rxcq_qnum = rx_ring;
+       }
+       if (uc->rflow->id != uc->rchan->id && uc->dir != DMA_MEM_TO_MEM) {
+               req.flowid_start = uc->rflow->id;
+               req.flowid_cnt = 1;
+               req.valid_params |=
+                       TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_START_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_CH_RX_FLOWID_CNT_VALID;
+       }
+
+       ret = ud->tisci_udmap_ops->rx_ch_cfg(ud->tisci, &req);
+       if (ret) {
+               dev_err(ud->dev, "tisci rx %u cfg failed %d\n",
+                       uc->rchan->id, ret);
+               return ret;
+       }
+       if (uc->dir == DMA_MEM_TO_MEM)
+               return ret;
+
+       flow_req.valid_params =
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_EINFO_PRESENT_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_PSINFO_PRESENT_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_ERROR_HANDLING_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DESC_TYPE_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_QNUM_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_HI_SEL_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_SRC_TAG_LO_SEL_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_HI_SEL_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_DEST_TAG_LO_SEL_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ0_SZ0_QNUM_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ1_QNUM_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ2_QNUM_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_FDQ3_QNUM_VALID |
+                       TI_SCI_MSG_VALUE_RM_UDMAP_FLOW_PS_LOCATION_VALID;
+
+       flow_req.nav_id = ud->tisci_dev_id;
+       flow_req.flow_index = uc->rflow->id;
+
+       if (uc->needs_epib)
+               flow_req.rx_einfo_present = 1;
+       else
+               flow_req.rx_einfo_present = 0;
+
+       if (uc->psd_size)
+               flow_req.rx_psinfo_present = 1;
+       else
+               flow_req.rx_psinfo_present = 0;
+
+       flow_req.rx_error_handling = 0;
+       flow_req.rx_desc_type = 0;
+       flow_req.rx_dest_qnum = rx_ring;
+       flow_req.rx_src_tag_hi_sel = 2;
+       flow_req.rx_src_tag_lo_sel = 4;
+       flow_req.rx_dest_tag_hi_sel = 5;
+       flow_req.rx_dest_tag_lo_sel = 4;
+       flow_req.rx_fdq0_sz0_qnum = fd_ring;
+       flow_req.rx_fdq1_qnum = fd_ring;
+       flow_req.rx_fdq2_qnum = fd_ring;
+       flow_req.rx_fdq3_qnum = fd_ring;
+       flow_req.rx_ps_location = 0;
+
+       ret = ud->tisci_udmap_ops->rx_flow_cfg(ud->tisci, &flow_req);
+       if (ret)
+               dev_err(ud->dev, "tisci rx %u flow %u cfg failed %d\n",
+                       uc->rchan->id, uc->rflow->id, ret);
+
+       return ret;
+}
+
+static int udma_alloc_chan_resources(struct udma_chan *uc)
+{
+       struct udma_dev *ud = uc->ud;
+       int ret;
+
+       pr_debug("%s: chan:%d as %s\n",
+                __func__, uc->id, udma_get_dir_text(uc->dir));
+
+       switch (uc->dir) {
+       case DMA_MEM_TO_MEM:
+               /* Non synchronized - mem to mem type of transfer */
+               ret = udma_get_chan_pair(uc);
+               if (ret)
+                       return ret;
+
+               ret = udma_alloc_tx_resources(uc);
+               if (ret)
+                       goto err_free_res;
+
+               ret = udma_alloc_rx_resources(uc);
+               if (ret)
+                       goto err_free_res;
+
+               uc->src_thread = ud->psil_base + uc->tchan->id;
+               uc->dst_thread = (ud->psil_base + uc->rchan->id) | 0x8000;
+               break;
+       case DMA_MEM_TO_DEV:
+               /* Slave transfer synchronized - mem to dev (TX) trasnfer */
+               ret = udma_alloc_tx_resources(uc);
+               if (ret)
+                       goto err_free_res;
+
+               uc->src_thread = ud->psil_base + uc->tchan->id;
+               uc->dst_thread = uc->slave_thread_id;
+               if (!(uc->dst_thread & 0x8000))
+                       uc->dst_thread |= 0x8000;
+
+               break;
+       case DMA_DEV_TO_MEM:
+               /* Slave transfer synchronized - dev to mem (RX) trasnfer */
+               ret = udma_alloc_rx_resources(uc);
+               if (ret)
+                       goto err_free_res;
+
+               uc->src_thread = uc->slave_thread_id;
+               uc->dst_thread = (ud->psil_base + uc->rchan->id) | 0x8000;
+
+               break;
+       default:
+               /* Can not happen */
+               pr_debug("%s: chan:%d invalid direction (%u)\n",
+                        __func__, uc->id, uc->dir);
+               return -EINVAL;
+       }
+
+       /* We have channel indexes and rings */
+       if (uc->dir == DMA_MEM_TO_MEM) {
+               ret = udma_alloc_tchan_sci_req(uc);
+               if (ret)
+                       goto err_free_res;
+
+               ret = udma_alloc_rchan_sci_req(uc);
+               if (ret)
+                       goto err_free_res;
+       } else {
+               /* Slave transfer */
+               if (uc->dir == DMA_MEM_TO_DEV) {
+                       ret = udma_alloc_tchan_sci_req(uc);
+                       if (ret)
+                               goto err_free_res;
+               } else {
+                       ret = udma_alloc_rchan_sci_req(uc);
+                       if (ret)
+                               goto err_free_res;
+               }
+       }
+
+       /* PSI-L pairing */
+       ret = udma_navss_psil_pair(ud, uc->src_thread, uc->dst_thread);
+       if (ret) {
+               dev_err(ud->dev, "k3_nav_psil_request_link fail\n");
+               goto err_free_res;
+       }
+
+       return 0;
+
+err_free_res:
+       udma_free_tx_resources(uc);
+       udma_free_rx_resources(uc);
+       uc->slave_thread_id = -1;
+       return ret;
+}
+
+static void udma_free_chan_resources(struct udma_chan *uc)
+{
+       /* Some configuration to UDMA-P channel: disable, reset, whatever */
+
+       /* Release PSI-L pairing */
+       udma_navss_psil_unpair(uc->ud, uc->src_thread, uc->dst_thread);
+
+       /* Reset the rings for a new start */
+       udma_reset_rings(uc);
+       udma_free_tx_resources(uc);
+       udma_free_rx_resources(uc);
+
+       uc->slave_thread_id = -1;
+       uc->dir = DMA_MEM_TO_MEM;
+}
+
+static int udma_get_mmrs(struct udevice *dev)
+{
+       struct udma_dev *ud = dev_get_priv(dev);
+       int i;
+
+       for (i = 0; i < MMR_LAST; i++) {
+               ud->mmrs[i] = (uint32_t *)devfdt_get_addr_name(dev,
+                               mmr_names[i]);
+               if (!ud->mmrs[i])
+                       return -EINVAL;
+       }
+
+       return 0;
+}
+
+#define UDMA_MAX_CHANNELS      192
+
+static int udma_probe(struct udevice *dev)
+{
+       struct dma_dev_priv *uc_priv = dev_get_uclass_priv(dev);
+       struct udma_dev *ud = dev_get_priv(dev);
+       int i, ret;
+       u32 cap2, cap3;
+       struct udevice *tmp;
+       struct udevice *tisci_dev = NULL;
+
+       ret = udma_get_mmrs(dev);
+       if (ret)
+               return ret;
+
+       ret = uclass_get_device_by_phandle(UCLASS_MISC, dev,
+                                          "ti,ringacc", &tmp);
+       ud->ringacc = dev_get_priv(tmp);
+       if (IS_ERR(ud->ringacc))
+               return PTR_ERR(ud->ringacc);
+
+       ud->psil_base = dev_read_u32_default(dev, "ti,psil-base", 0);
+       if (!ud->psil_base) {
+               dev_info(dev,
+                        "Missing ti,psil-base property, using %d.\n", ret);
+               return -EINVAL;
+       }
+
+       ret = uclass_get_device_by_name(UCLASS_FIRMWARE, "dmsc", &tisci_dev);
+       if (ret) {
+               debug("TISCI RA RM get failed (%d)\n", ret);
+               ud->tisci = NULL;
+               return 0;
+       }
+       ud->tisci = (struct ti_sci_handle *)
+                        (ti_sci_get_handle_from_sysfw(tisci_dev));
+
+       ret = dev_read_u32_default(dev, "ti,sci", 0);
+       if (!ret) {
+               dev_err(dev, "TISCI RA RM disabled\n");
+               ud->tisci = NULL;
+       }
+
+       if (ud->tisci) {
+               ofnode navss_ofnode = ofnode_get_parent(dev_ofnode(dev));
+
+               ud->tisci_dev_id = -1;
+               ret = dev_read_u32(dev, "ti,sci-dev-id", &ud->tisci_dev_id);
+               if (ret) {
+                       dev_err(dev, "ti,sci-dev-id read failure %d\n", ret);
+                       return ret;
+               }
+
+               ud->tisci_navss_dev_id = -1;
+               ret = ofnode_read_u32(navss_ofnode, "ti,sci-dev-id",
+                                     &ud->tisci_navss_dev_id);
+               if (ret) {
+                       dev_err(dev, "navss sci-dev-id read failure %d\n", ret);
+                       return ret;
+               }
+
+               ud->tisci_udmap_ops = &ud->tisci->ops.rm_udmap_ops;
+               ud->tisci_psil_ops = &ud->tisci->ops.rm_psil_ops;
+       }
+
+       ud->is_coherent = dev_read_bool(dev, "dma-coherent");
+
+       cap2 = udma_read(ud->mmrs[MMR_GCFG], 0x28);
+       cap3 = udma_read(ud->mmrs[MMR_GCFG], 0x2c);
+
+       ud->rflow_cnt = cap3 & 0x3fff;
+       ud->tchan_cnt = cap2 & 0x1ff;
+       ud->echan_cnt = (cap2 >> 9) & 0x1ff;
+       ud->rchan_cnt = (cap2 >> 18) & 0x1ff;
+       ud->ch_count  = ud->tchan_cnt + ud->rchan_cnt;
+
+       dev_info(dev,
+                "Number of channels: %u (tchan: %u, echan: %u, rchan: %u dev-id %u)\n",
+                ud->ch_count, ud->tchan_cnt, ud->echan_cnt, ud->rchan_cnt,
+                ud->tisci_dev_id);
+       dev_info(dev, "Number of rflows: %u\n", ud->rflow_cnt);
+
+       ud->channels = devm_kcalloc(dev, ud->ch_count, sizeof(*ud->channels),
+                                   GFP_KERNEL);
+       ud->tchan_map = devm_kcalloc(dev, BITS_TO_LONGS(ud->tchan_cnt),
+                                    sizeof(unsigned long), GFP_KERNEL);
+       ud->tchans = devm_kcalloc(dev, ud->tchan_cnt,
+                                 sizeof(*ud->tchans), GFP_KERNEL);
+       ud->rchan_map = devm_kcalloc(dev, BITS_TO_LONGS(ud->rchan_cnt),
+                                    sizeof(unsigned long), GFP_KERNEL);
+       ud->rchans = devm_kcalloc(dev, ud->rchan_cnt,
+                                 sizeof(*ud->rchans), GFP_KERNEL);
+       ud->rflow_map = devm_kcalloc(dev, BITS_TO_LONGS(ud->rflow_cnt),
+                                    sizeof(unsigned long), GFP_KERNEL);
+       ud->rflows = devm_kcalloc(dev, ud->rflow_cnt,
+                                 sizeof(*ud->rflows), GFP_KERNEL);
+
+       if (!ud->channels || !ud->tchan_map || !ud->rchan_map ||
+           !ud->rflow_map || !ud->tchans || !ud->rchans || !ud->rflows)
+               return -ENOMEM;
+
+       for (i = 0; i < ud->tchan_cnt; i++) {
+               struct udma_tchan *tchan = &ud->tchans[i];
+
+               tchan->id = i;
+               tchan->reg_rt = ud->mmrs[MMR_TCHANRT] + UDMA_CH_1000(i);
+       }
+
+       for (i = 0; i < ud->rchan_cnt; i++) {
+               struct udma_rchan *rchan = &ud->rchans[i];
+
+               rchan->id = i;
+               rchan->reg_rt = ud->mmrs[MMR_RCHANRT] + UDMA_CH_1000(i);
+       }
+
+       for (i = 0; i < ud->rflow_cnt; i++) {
+               struct udma_rflow *rflow = &ud->rflows[i];
+
+               rflow->id = i;
+       }
+
+       for (i = 0; i < ud->ch_count; i++) {
+               struct udma_chan *uc = &ud->channels[i];
+
+               uc->ud = ud;
+               uc->id = i;
+               uc->slave_thread_id = -1;
+               uc->tchan = NULL;
+               uc->rchan = NULL;
+               uc->dir = DMA_MEM_TO_MEM;
+               sprintf(uc->name, "UDMA chan%d\n", i);
+               if (!i)
+                       uc->in_use = true;
+       }
+
+       pr_debug("UDMA(rev: 0x%08x) CAP0-3: 0x%08x, 0x%08x, 0x%08x, 0x%08x\n",
+                udma_read(ud->mmrs[MMR_GCFG], 0),
+                udma_read(ud->mmrs[MMR_GCFG], 0x20),
+                udma_read(ud->mmrs[MMR_GCFG], 0x24),
+                udma_read(ud->mmrs[MMR_GCFG], 0x28),
+                udma_read(ud->mmrs[MMR_GCFG], 0x2c));
+
+       uc_priv->supported = DMA_SUPPORTS_MEM_TO_MEM | DMA_SUPPORTS_MEM_TO_DEV;
+
+       return ret;
+}
+
+static int *udma_prep_dma_memcpy(struct udma_chan *uc, dma_addr_t dest,
+                                dma_addr_t src, size_t len)
+{
+       u32 tc_ring_id = k3_nav_ringacc_get_ring_id(uc->tchan->tc_ring);
+       struct cppi5_tr_type15_t *tr_req;
+       int num_tr;
+       size_t tr_size = sizeof(struct cppi5_tr_type15_t);
+       u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
+       unsigned long dummy;
+       void *tr_desc;
+       size_t desc_size;
+
+       if (len < SZ_64K) {
+               num_tr = 1;
+               tr0_cnt0 = len;
+               tr0_cnt1 = 1;
+       } else {
+               unsigned long align_to = __ffs(src | dest);
+
+               if (align_to > 3)
+                       align_to = 3;
+               /*
+                * Keep simple: tr0: SZ_64K-alignment blocks,
+                *              tr1: the remaining
+                */
+               num_tr = 2;
+               tr0_cnt0 = (SZ_64K - BIT(align_to));
+               if (len / tr0_cnt0 >= SZ_64K) {
+                       dev_err(uc->ud->dev, "size %zu is not supported\n",
+                               len);
+                       return NULL;
+               }
+
+               tr0_cnt1 = len / tr0_cnt0;
+               tr1_cnt0 = len % tr0_cnt0;
+       }
+
+       desc_size = cppi5_trdesc_calc_size(num_tr, tr_size);
+       tr_desc = dma_alloc_coherent(desc_size, &dummy);
+       if (!tr_desc)
+               return NULL;
+       memset(tr_desc, 0, desc_size);
+
+       cppi5_trdesc_init(tr_desc, num_tr, tr_size, 0, 0);
+       cppi5_desc_set_pktids(tr_desc, uc->id, 0x3fff);
+       cppi5_desc_set_retpolicy(tr_desc, 0, tc_ring_id);
+
+       tr_req = tr_desc + tr_size;
+
+       cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
+                     CPPI5_TR_EVENT_SIZE_COMPLETION, 1);
+       cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
+
+       tr_req[0].addr = src;
+       tr_req[0].icnt0 = tr0_cnt0;
+       tr_req[0].icnt1 = tr0_cnt1;
+       tr_req[0].icnt2 = 1;
+       tr_req[0].icnt3 = 1;
+       tr_req[0].dim1 = tr0_cnt0;
+
+       tr_req[0].daddr = dest;
+       tr_req[0].dicnt0 = tr0_cnt0;
+       tr_req[0].dicnt1 = tr0_cnt1;
+       tr_req[0].dicnt2 = 1;
+       tr_req[0].dicnt3 = 1;
+       tr_req[0].ddim1 = tr0_cnt0;
+
+       if (num_tr == 2) {
+               cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
+                             CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
+               cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
+
+               tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
+               tr_req[1].icnt0 = tr1_cnt0;
+               tr_req[1].icnt1 = 1;
+               tr_req[1].icnt2 = 1;
+               tr_req[1].icnt3 = 1;
+
+               tr_req[1].daddr = dest + tr0_cnt1 * tr0_cnt0;
+               tr_req[1].dicnt0 = tr1_cnt0;
+               tr_req[1].dicnt1 = 1;
+               tr_req[1].dicnt2 = 1;
+               tr_req[1].dicnt3 = 1;
+       }
+
+       cppi5_tr_csf_set(&tr_req[num_tr - 1].flags, CPPI5_TR_CSF_EOP);
+
+       if (!udma_is_coherent(uc)) {
+               flush_dcache_range((u64)tr_desc,
+                                  ALIGN((u64)tr_desc + desc_size,
+                                        ARCH_DMA_MINALIGN));
+       }
+
+       k3_nav_ringacc_ring_push(uc->tchan->t_ring, &tr_desc);
+
+       return 0;
+}
+
+static int udma_transfer(struct udevice *dev, int direction,
+                        void *dst, void *src, size_t len)
+{
+       struct udma_dev *ud = dev_get_priv(dev);
+       /* Channel0 is reserved for memcpy */
+       struct udma_chan *uc = &ud->channels[0];
+       dma_addr_t paddr = 0;
+       int ret;
+
+       ret = udma_alloc_chan_resources(uc);
+       if (ret)
+               return ret;
+
+       udma_prep_dma_memcpy(uc, (dma_addr_t)dst, (dma_addr_t)src, len);
+       udma_start(uc);
+       udma_poll_completion(uc, &paddr);
+       udma_stop(uc);
+
+       udma_free_chan_resources(uc);
+       return 0;
+}
+
+static int udma_request(struct dma *dma)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct udma_chan *uc;
+       unsigned long dummy;
+       int ret;
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+
+       uc = &ud->channels[dma->id];
+       ret = udma_alloc_chan_resources(uc);
+       if (ret) {
+               dev_err(dma->dev, "alloc dma res failed %d\n", ret);
+               return -EINVAL;
+       }
+
+       uc->hdesc_size = cppi5_hdesc_calc_size(uc->needs_epib,
+                                              uc->psd_size, 0);
+       uc->hdesc_size = ALIGN(uc->hdesc_size, ARCH_DMA_MINALIGN);
+
+       if (uc->dir == DMA_MEM_TO_DEV) {
+               uc->desc_tx = dma_alloc_coherent(uc->hdesc_size, &dummy);
+               memset(uc->desc_tx, 0, uc->hdesc_size);
+       } else {
+               uc->desc_rx = dma_alloc_coherent(
+                               uc->hdesc_size * UDMA_RX_DESC_NUM, &dummy);
+               memset(uc->desc_rx, 0, uc->hdesc_size * UDMA_RX_DESC_NUM);
+       }
+
+       uc->in_use = true;
+       uc->desc_rx_cur = 0;
+       uc->num_rx_bufs = 0;
+
+       return 0;
+}
+
+static int udma_free(struct dma *dma)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct udma_chan *uc;
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+       uc = &ud->channels[dma->id];
+
+       if (udma_is_chan_running(uc))
+               udma_stop(uc);
+       udma_free_chan_resources(uc);
+
+       uc->in_use = false;
+
+       return 0;
+}
+
+static int udma_enable(struct dma *dma)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct udma_chan *uc;
+       int ret;
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+       uc = &ud->channels[dma->id];
+
+       ret = udma_start(uc);
+
+       return ret;
+}
+
+static int udma_disable(struct dma *dma)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct udma_chan *uc;
+       int ret = 0;
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+       uc = &ud->channels[dma->id];
+
+       if (udma_is_chan_running(uc))
+               ret = udma_stop(uc);
+       else
+               dev_err(dma->dev, "%s not running\n", __func__);
+
+       return ret;
+}
+
+static int udma_send(struct dma *dma, void *src, size_t len, void *metadata)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct cppi5_host_desc_t *desc_tx;
+       dma_addr_t dma_src = (dma_addr_t)src;
+       struct ti_udma_drv_packet_data packet_data = { 0 };
+       dma_addr_t paddr;
+       struct udma_chan *uc;
+       u32 tc_ring_id;
+       int ret;
+
+       if (!metadata)
+               packet_data = *((struct ti_udma_drv_packet_data *)metadata);
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+       uc = &ud->channels[dma->id];
+
+       if (uc->dir != DMA_MEM_TO_DEV)
+               return -EINVAL;
+
+       tc_ring_id = k3_nav_ringacc_get_ring_id(uc->tchan->tc_ring);
+
+       desc_tx = uc->desc_tx;
+
+       cppi5_hdesc_reset_hbdesc(desc_tx);
+
+       cppi5_hdesc_init(desc_tx,
+                        uc->needs_epib ? CPPI5_INFO0_HDESC_EPIB_PRESENT : 0,
+                        uc->psd_size);
+       cppi5_hdesc_set_pktlen(desc_tx, len);
+       cppi5_hdesc_attach_buf(desc_tx, dma_src, len, dma_src, len);
+       cppi5_desc_set_pktids(&desc_tx->hdr, uc->id, 0x3fff);
+       cppi5_desc_set_retpolicy(&desc_tx->hdr, 0, tc_ring_id);
+       /* pass below information from caller */
+       cppi5_hdesc_set_pkttype(desc_tx, packet_data.pkt_type);
+       cppi5_desc_set_tags_ids(&desc_tx->hdr, 0, packet_data.dest_tag);
+
+       if (!udma_is_coherent(uc)) {
+               flush_dcache_range((u64)dma_src,
+                                  ALIGN((u64)dma_src + len,
+                                        ARCH_DMA_MINALIGN));
+               flush_dcache_range((u64)desc_tx,
+                                  ALIGN((u64)desc_tx + uc->hdesc_size,
+                                        ARCH_DMA_MINALIGN));
+       }
+
+       ret = k3_nav_ringacc_ring_push(uc->tchan->t_ring, &uc->desc_tx);
+       if (ret) {
+               dev_err(dma->dev, "TX dma push fail ch_id %lu %d\n",
+                       dma->id, ret);
+               return ret;
+       }
+
+       udma_poll_completion(uc, &paddr);
+
+       return 0;
+}
+
+static int udma_receive(struct dma *dma, void **dst, void *metadata)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct cppi5_host_desc_t *desc_rx;
+       dma_addr_t buf_dma;
+       struct udma_chan *uc;
+       u32 buf_dma_len, pkt_len;
+       u32 port_id = 0;
+       int ret;
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+       uc = &ud->channels[dma->id];
+
+       if (uc->dir != DMA_DEV_TO_MEM)
+               return -EINVAL;
+       if (!uc->num_rx_bufs)
+               return -EINVAL;
+
+       ret = k3_nav_ringacc_ring_pop(uc->rchan->r_ring, &desc_rx);
+       if (ret && ret != -ENODATA) {
+               dev_err(dma->dev, "rx dma fail ch_id:%lu %d\n", dma->id, ret);
+               return ret;
+       } else if (ret == -ENODATA) {
+               return 0;
+       }
+
+       /* invalidate cache data */
+       if (!udma_is_coherent(uc)) {
+               invalidate_dcache_range((ulong)desc_rx,
+                                       (ulong)(desc_rx + uc->hdesc_size));
+       }
+
+       cppi5_hdesc_get_obuf(desc_rx, &buf_dma, &buf_dma_len);
+       pkt_len = cppi5_hdesc_get_pktlen(desc_rx);
+
+       /* invalidate cache data */
+       if (!udma_is_coherent(uc)) {
+               invalidate_dcache_range((ulong)buf_dma,
+                                       (ulong)(buf_dma + buf_dma_len));
+       }
+
+       cppi5_desc_get_tags_ids(&desc_rx->hdr, &port_id, NULL);
+
+       *dst = (void *)buf_dma;
+       uc->num_rx_bufs--;
+
+       return pkt_len;
+}
+
+static int udma_of_xlate(struct dma *dma, struct ofnode_phandle_args *args)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct udma_chan *uc = &ud->channels[0];
+       ofnode chconf_node, slave_node;
+       char prop[50];
+       u32 val;
+
+       for (val = 0; val < ud->ch_count; val++) {
+               uc = &ud->channels[val];
+               if (!uc->in_use)
+                       break;
+       }
+
+       if (val == ud->ch_count)
+               return -EBUSY;
+
+       uc->dir = DMA_DEV_TO_MEM;
+       if (args->args[2] == UDMA_DIR_TX)
+               uc->dir = DMA_MEM_TO_DEV;
+
+       slave_node = ofnode_get_by_phandle(args->args[0]);
+       if (!ofnode_valid(slave_node)) {
+               dev_err(ud->dev, "slave node is missing\n");
+               return -EINVAL;
+       }
+
+       snprintf(prop, sizeof(prop), "ti,psil-config%u", args->args[1]);
+       chconf_node = ofnode_find_subnode(slave_node, prop);
+       if (!ofnode_valid(chconf_node)) {
+               dev_err(ud->dev, "Channel configuration node is missing\n");
+               return -EINVAL;
+       }
+
+       if (!ofnode_read_u32(chconf_node, "linux,udma-mode", &val)) {
+               if (val == UDMA_PKT_MODE)
+                       uc->pkt_mode = true;
+       }
+
+       if (!ofnode_read_u32(chconf_node, "statictr-type", &val))
+               uc->static_tr_type = val;
+
+       uc->needs_epib = ofnode_read_bool(chconf_node, "ti,needs-epib");
+       if (!ofnode_read_u32(chconf_node, "ti,psd-size", &val))
+               uc->psd_size = val;
+       uc->metadata_size = (uc->needs_epib ? 16 : 0) + uc->psd_size;
+
+       if (ofnode_read_u32(slave_node, "ti,psil-base", &val)) {
+               dev_err(ud->dev, "ti,psil-base is missing\n");
+               return -EINVAL;
+       }
+
+       uc->slave_thread_id = val + args->args[1];
+
+       dma->id = uc->id;
+       pr_debug("Allocated dma chn:%lu epib:%d psdata:%u meta:%u thread_id:%x\n",
+                dma->id, uc->needs_epib,
+                uc->psd_size, uc->metadata_size,
+                uc->slave_thread_id);
+
+       return 0;
+}
+
+int udma_prepare_rcv_buf(struct dma *dma, void *dst, size_t size)
+{
+       struct udma_dev *ud = dev_get_priv(dma->dev);
+       struct cppi5_host_desc_t *desc_rx;
+       dma_addr_t dma_dst;
+       struct udma_chan *uc;
+       u32 desc_num;
+
+       if (dma->id >= (ud->rchan_cnt + ud->tchan_cnt)) {
+               dev_err(dma->dev, "invalid dma ch_id %lu\n", dma->id);
+               return -EINVAL;
+       }
+       uc = &ud->channels[dma->id];
+
+       if (uc->dir != DMA_DEV_TO_MEM)
+               return -EINVAL;
+
+       if (uc->num_rx_bufs >= UDMA_RX_DESC_NUM)
+               return -EINVAL;
+
+       desc_num = uc->desc_rx_cur % UDMA_RX_DESC_NUM;
+       desc_rx = uc->desc_rx + (desc_num * uc->hdesc_size);
+       dma_dst = (dma_addr_t)dst;
+
+       cppi5_hdesc_reset_hbdesc(desc_rx);
+
+       cppi5_hdesc_init(desc_rx,
+                        uc->needs_epib ? CPPI5_INFO0_HDESC_EPIB_PRESENT : 0,
+                        uc->psd_size);
+       cppi5_hdesc_set_pktlen(desc_rx, size);
+       cppi5_hdesc_attach_buf(desc_rx, dma_dst, size, dma_dst, size);
+
+       if (!udma_is_coherent(uc)) {
+               flush_dcache_range((u64)desc_rx,
+                                  ALIGN((u64)desc_rx + uc->hdesc_size,
+                                        ARCH_DMA_MINALIGN));
+       }
+
+       k3_nav_ringacc_ring_push(uc->rchan->fd_ring, &desc_rx);
+
+       uc->num_rx_bufs++;
+       uc->desc_rx_cur++;
+
+       return 0;
+}
+
+static const struct dma_ops udma_ops = {
+       .transfer       = udma_transfer,
+       .of_xlate       = udma_of_xlate,
+       .request        = udma_request,
+       .free           = udma_free,
+       .enable         = udma_enable,
+       .disable        = udma_disable,
+       .send           = udma_send,
+       .receive        = udma_receive,
+       .prepare_rcv_buf = udma_prepare_rcv_buf,
+};
+
+static const struct udevice_id udma_ids[] = {
+       { .compatible = "ti,k3-navss-udmap" },
+       { }
+};
+
+U_BOOT_DRIVER(ti_edma3) = {
+       .name   = "ti-udma",
+       .id     = UCLASS_DMA,
+       .of_match = udma_ids,
+       .ops    = &udma_ops,
+       .probe  = udma_probe,
+       .priv_auto_alloc_size = sizeof(struct udma_dev),
+};
diff --git a/include/dt-bindings/dma/k3-udma.h b/include/dt-bindings/dma/k3-udma.h
new file mode 100644 (file)
index 0000000..670e123
--- /dev/null
@@ -0,0 +1,31 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ *  Copyright (C) 2018 Texas Instruments Incorporated - http://www.ti.com
+ */
+
+#ifndef __DT_TI_UDMA_H
+#define __DT_TI_UDMA_H
+
+#define UDMA_TR_MODE           0
+#define UDMA_PKT_MODE          1
+
+#define UDMA_DIR_TX            0
+#define UDMA_DIR_RX            1
+
+#define PSIL_STATIC_TR_NONE    0
+#define PSIL_STATIC_TR_XY      1
+#define PSIL_STATIC_TR_MCAN    2
+
+#define UDMA_PDMA_TR_XY(id)                            \
+       ti,psil-config##id {                            \
+               linux,udma-mode = <UDMA_TR_MODE>;       \
+               statictr-type = <PSIL_STATIC_TR_XY>;    \
+       }
+
+#define UDMA_PDMA_PKT_XY(id)                           \
+       ti,psil-config##id {                            \
+               linux,udma-mode = <UDMA_PKT_MODE>;      \
+               statictr-type = <PSIL_STATIC_TR_XY>;    \
+       }
+
+#endif /* __DT_TI_UDMA_H */
diff --git a/include/linux/soc/ti/ti-udma.h b/include/linux/soc/ti/ti-udma.h
new file mode 100644 (file)
index 0000000..e9d4226
--- /dev/null
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ *  Copyright (C) 2018 Texas Instruments Incorporated - http://www.ti.com
+ *  Author: Peter Ujfalusi <peter.ujfalusi@ti.com>
+ */
+
+#ifndef __TI_UDMA_H
+#define __TI_UDMA_H
+
+/**
+ * struct ti_udma_drv_packet_data - TI UDMA transfer specific data
+ *
+ * @pkt_type: Packet Type - specific for each DMA client HW
+ * @dest_tag: Destination tag The source pointer.
+ *
+ * TI UDMA transfer specific data passed as part of DMA transfer to
+ * the DMA client HW in UDMA descriptors.
+ */
+struct ti_udma_drv_packet_data {
+       u32     pkt_type;
+       u32     dest_tag;
+};
+
+#endif /* __TI_UDMA_H */