remoteproc: Introduce K3 remoteproc driver for R5F subsystem
authorLokesh Vutla <lokeshvutla@ti.com>
Wed, 4 Sep 2019 10:31:34 +0000 (16:01 +0530)
committerTom Rini <trini@konsulko.com>
Fri, 11 Oct 2019 14:07:34 +0000 (10:07 -0400)
SoCs with K3 architecture have an integrated Arm Cortex-R5F subsystem
that is comprised of dual-core Arm Cortex-R5F processor cores. This R5
subsytem can be configured at boot time to be either run in a LockStep
mode or in an Asymmetric Multi Processing (AMP) fashion in Split-mode.
This subsystem has each Tightly-Coupled Memory (TCM) internal memories
for each core split between two banks - TCMA and TCMB.

Add a remoteproc driver to support this subsystem to be able to load
and boot the R5 cores primarily in LockStep mode or split mode.

Signed-off-by: Lokesh Vutla <lokeshvutla@ti.com>
Signed-off-by: Suman Anna <s-anna@ti.com>
drivers/remoteproc/Kconfig
drivers/remoteproc/Makefile
drivers/remoteproc/ti_k3_r5f_rproc.c [new file with mode: 0644]

index f54a245424a91ed64fcdb10d571ffa6f987262b6..c2d59ba0007eee453140ebbb7371da413f0251cc 100644 (file)
@@ -52,6 +52,16 @@ config REMOTEPROC_TI_K3_ARM64
          on various TI K3 family of SoCs through the remote processor
          framework.
 
+config REMOTEPROC_TI_K3_R5F
+       bool "TI K3 R5F remoteproc support"
+       select REMOTEPROC
+       depends on ARCH_K3
+       depends on TI_SCI_PROTOCOL
+       help
+         Say y here to support TI's R5F remote processor subsystems
+         on various TI K3 family of SoCs through the remote processor
+         framework.
+
 config REMOTEPROC_TI_POWER
        bool "Support for TI Power processor"
        select REMOTEPROC
index 271ba55b093dad5dd12beac7fe0c8a19d5e75a66..9d247ba5e7aedec671ff9dbe9927a2ef62653791 100644 (file)
@@ -11,4 +11,5 @@ obj-$(CONFIG_K3_SYSTEM_CONTROLLER) += k3_system_controller.o
 obj-$(CONFIG_REMOTEPROC_SANDBOX) += sandbox_testproc.o
 obj-$(CONFIG_REMOTEPROC_STM32_COPRO) += stm32_copro.o
 obj-$(CONFIG_REMOTEPROC_TI_K3_ARM64) += ti_k3_arm64_rproc.o
+obj-$(CONFIG_REMOTEPROC_TI_K3_R5F) += ti_k3_r5f_rproc.o
 obj-$(CONFIG_REMOTEPROC_TI_POWER) += ti_power_proc.o
diff --git a/drivers/remoteproc/ti_k3_r5f_rproc.c b/drivers/remoteproc/ti_k3_r5f_rproc.c
new file mode 100644 (file)
index 0000000..ae1e4b9
--- /dev/null
@@ -0,0 +1,816 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Texas Instruments' K3 R5 Remoteproc driver
+ *
+ * Copyright (C) 2018-2019 Texas Instruments Incorporated - http://www.ti.com/
+ *     Lokesh Vutla <lokeshvutla@ti.com>
+ */
+
+#include <common.h>
+#include <dm.h>
+#include <remoteproc.h>
+#include <errno.h>
+#include <clk.h>
+#include <reset.h>
+#include <asm/io.h>
+#include <linux/kernel.h>
+#include <linux/soc/ti/ti_sci_protocol.h>
+#include "ti_sci_proc.h"
+
+/*
+ * R5F's view of this address can either be for ATCM or BTCM with the other
+ * at address 0x0 based on loczrama signal.
+ */
+#define K3_R5_TCM_DEV_ADDR     0x41010000
+
+/* R5 TI-SCI Processor Configuration Flags */
+#define PROC_BOOT_CFG_FLAG_R5_DBG_EN                   0x00000001
+#define PROC_BOOT_CFG_FLAG_R5_DBG_NIDEN                        0x00000002
+#define PROC_BOOT_CFG_FLAG_R5_LOCKSTEP                 0x00000100
+#define PROC_BOOT_CFG_FLAG_R5_TEINIT                   0x00000200
+#define PROC_BOOT_CFG_FLAG_R5_NMFI_EN                  0x00000400
+#define PROC_BOOT_CFG_FLAG_R5_TCM_RSTBASE              0x00000800
+#define PROC_BOOT_CFG_FLAG_R5_BTCM_EN                  0x00001000
+#define PROC_BOOT_CFG_FLAG_R5_ATCM_EN                  0x00002000
+#define PROC_BOOT_CFG_FLAG_GEN_IGN_BOOTVECTOR          0x10000000
+
+/* R5 TI-SCI Processor Control Flags */
+#define PROC_BOOT_CTRL_FLAG_R5_CORE_HALT               0x00000001
+
+/* R5 TI-SCI Processor Status Flags */
+#define PROC_BOOT_STATUS_FLAG_R5_WFE                   0x00000001
+#define PROC_BOOT_STATUS_FLAG_R5_WFI                   0x00000002
+#define PROC_BOOT_STATUS_FLAG_R5_CLK_GATED             0x00000004
+#define PROC_BOOT_STATUS_FLAG_R5_LOCKSTEP_PERMITTED    0x00000100
+
+#define NR_CORES       2
+
+enum cluster_mode {
+       CLUSTER_MODE_SPLIT = 0,
+       CLUSTER_MODE_LOCKSTEP,
+};
+
+/**
+ * struct k3_r5_mem - internal memory structure
+ * @cpu_addr: MPU virtual address of the memory region
+ * @bus_addr: Bus address used to access the memory region
+ * @dev_addr: Device address from remoteproc view
+ * @size: Size of the memory region
+ */
+struct k3_r5f_mem {
+       void __iomem *cpu_addr;
+       phys_addr_t bus_addr;
+       u32 dev_addr;
+       size_t size;
+};
+
+/**
+ * struct k3_r5f_core - K3 R5 core structure
+ * @dev: cached device pointer
+ * @cluster: pointer to the parent cluster.
+ * @reset: reset control handle
+ * @tsp: TI-SCI processor control handle
+ * @mem: Array of available internal memories
+ * @num_mem: Number of available memories
+ * @atcm_enable: flag to control ATCM enablement
+ * @btcm_enable: flag to control BTCM enablement
+ * @loczrama: flag to dictate which TCM is at device address 0x0
+ * @in_use: flag to tell if the core is already in use.
+ */
+struct k3_r5f_core {
+       struct udevice *dev;
+       struct k3_r5f_cluster *cluster;
+       struct reset_ctl reset;
+       struct ti_sci_proc tsp;
+       struct k3_r5f_mem *mem;
+       int num_mems;
+       u32 atcm_enable;
+       u32 btcm_enable;
+       u32 loczrama;
+       bool in_use;
+};
+
+/**
+ * struct k3_r5f_cluster - K3 R5F Cluster structure
+ * @mode: Mode to configure the Cluster - Split or LockStep
+ * @cores: Array of pointers to R5 cores within the cluster
+ */
+struct k3_r5f_cluster {
+       enum cluster_mode mode;
+       struct k3_r5f_core *cores[NR_CORES];
+};
+
+static bool is_primary_core(struct k3_r5f_core *core)
+{
+       return core == core->cluster->cores[0];
+}
+
+static int k3_r5f_proc_request(struct k3_r5f_core *core)
+{
+       struct k3_r5f_cluster *cluster = core->cluster;
+       int i, ret;
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP) {
+               for (i = 0; i < NR_CORES; i++) {
+                       ret = ti_sci_proc_request(&cluster->cores[i]->tsp);
+                       if (ret)
+                               goto proc_release;
+               }
+       } else {
+               ret = ti_sci_proc_request(&core->tsp);
+       }
+
+       return 0;
+
+proc_release:
+       while (i >= 0) {
+               ti_sci_proc_release(&cluster->cores[i]->tsp);
+               i--;
+       }
+       return ret;
+}
+
+static void k3_r5f_proc_release(struct k3_r5f_core *core)
+{
+       struct k3_r5f_cluster *cluster = core->cluster;
+       int i;
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP)
+               for (i = 0; i < NR_CORES; i++)
+                       ti_sci_proc_release(&cluster->cores[i]->tsp);
+       else
+               ti_sci_proc_release(&core->tsp);
+}
+
+static int k3_r5f_lockstep_release(struct k3_r5f_cluster *cluster)
+{
+       int ret, c;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       for (c = NR_CORES - 1; c >= 0; c--) {
+               ret = ti_sci_proc_power_domain_on(&cluster->cores[c]->tsp);
+               if (ret)
+                       goto unroll_module_reset;
+       }
+
+       /* deassert local reset on all applicable cores */
+       for (c = NR_CORES - 1; c >= 0; c--) {
+               ret = reset_deassert(&cluster->cores[c]->reset);
+               if (ret)
+                       goto unroll_local_reset;
+       }
+
+       return 0;
+
+unroll_local_reset:
+       while (c < NR_CORES) {
+               reset_assert(&cluster->cores[c]->reset);
+               c++;
+       }
+       c = 0;
+unroll_module_reset:
+       while (c < NR_CORES) {
+               ti_sci_proc_power_domain_off(&cluster->cores[c]->tsp);
+               c++;
+       }
+
+       return ret;
+}
+
+static int k3_r5f_split_release(struct k3_r5f_core *core)
+{
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       ret = ti_sci_proc_power_domain_on(&core->tsp);
+       if (ret) {
+               dev_err(core->dev, "module-reset deassert failed, ret = %d\n",
+                       ret);
+               return ret;
+       }
+
+       ret = reset_deassert(&core->reset);
+       if (ret) {
+               dev_err(core->dev, "local-reset deassert failed, ret = %d\n",
+                       ret);
+               if (ti_sci_proc_power_domain_off(&core->tsp))
+                       dev_warn(core->dev, "module-reset assert back failed\n");
+       }
+
+       return ret;
+}
+
+static int k3_r5f_prepare(struct udevice *dev)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       struct k3_r5f_cluster *cluster = core->cluster;
+       int ret = 0;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP)
+               ret = k3_r5f_lockstep_release(cluster);
+       else
+               ret = k3_r5f_split_release(core);
+
+       if (ret)
+               dev_err(dev, "Unable to enable cores for TCM loading %d\n",
+                       ret);
+
+       return ret;
+}
+
+static int k3_r5f_core_sanity_check(struct k3_r5f_core *core)
+{
+       struct k3_r5f_cluster *cluster = core->cluster;
+
+       if (core->in_use) {
+               dev_err(dev, "Invalid op: Trying to load/start on already running core %d\n",
+                       core->tsp.proc_id);
+               return -EINVAL;
+       }
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP && !cluster->cores[1]) {
+               printf("Secondary core is not probed in this cluster\n");
+               return -EAGAIN;
+       }
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP && !is_primary_core(core)) {
+               dev_err(dev, "Invalid op: Trying to start secondary core %d in lockstep mode\n",
+                       core->tsp.proc_id);
+               return -EINVAL;
+       }
+
+       if (cluster->mode == CLUSTER_MODE_SPLIT && !is_primary_core(core)) {
+               if (!core->cluster->cores[0]->in_use) {
+                       dev_err(dev, "Invalid seq: Enable primary core before loading secondary core\n");
+                       return -EINVAL;
+               }
+       }
+
+       return 0;
+}
+
+/**
+ * k3_r5f_load() - Load up the Remote processor image
+ * @dev:       rproc device pointer
+ * @addr:      Address at which image is available
+ * @size:      size of the image
+ *
+ * Return: 0 if all goes good, else appropriate error message.
+ */
+static int k3_r5f_load(struct udevice *dev, ulong addr, ulong size)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       u32 boot_vector;
+       int ret;
+
+       dev_dbg(dev, "%s addr = 0x%lx, size = 0x%lx\n", __func__, addr, size);
+
+       ret = k3_r5f_core_sanity_check(core);
+       if (ret)
+               return ret;
+
+       ret = k3_r5f_proc_request(core);
+       if (ret)
+               return ret;
+
+       ret = k3_r5f_prepare(dev);
+       if (ret) {
+               dev_err(dev, "R5f prepare failed for core %d\n",
+                       core->tsp.proc_id);
+               goto proc_release;
+       }
+
+       /* Zero out TCMs so that ECC can be effective on all TCM addresses */
+       if (core->atcm_enable)
+               memset(core->mem[0].cpu_addr, 0x00, core->mem[0].size);
+       if (core->btcm_enable)
+               memset(core->mem[1].cpu_addr, 0x00, core->mem[1].size);
+
+       ret = rproc_elf_load_image(dev, addr, size);
+       if (ret < 0) {
+               dev_err(dev, "Loading elf failedi %d\n", ret);
+               goto proc_release;
+       }
+
+       boot_vector = rproc_elf_get_boot_addr(dev, addr);
+
+       dev_dbg(dev, "%s: Boot vector = 0x%x\n", __func__, boot_vector);
+
+       ret = ti_sci_proc_set_config(&core->tsp, boot_vector, 0, 0);
+
+proc_release:
+       k3_r5f_proc_release(core);
+
+       return ret;
+}
+
+static int k3_r5f_core_halt(struct k3_r5f_core *core)
+{
+       int ret;
+
+       ret = ti_sci_proc_set_control(&core->tsp,
+                                     PROC_BOOT_CTRL_FLAG_R5_CORE_HALT, 0);
+       if (ret)
+               dev_err(core->dev, "Core %d failed to stop\n",
+                       core->tsp.proc_id);
+
+       return ret;
+}
+
+static int k3_r5f_core_run(struct k3_r5f_core *core)
+{
+       int ret;
+
+       ret = ti_sci_proc_set_control(&core->tsp,
+                                     0, PROC_BOOT_CTRL_FLAG_R5_CORE_HALT);
+       if (ret) {
+               dev_err(core->dev, "Core %d failed to start\n",
+                       core->tsp.proc_id);
+               return ret;
+       }
+
+       return 0;
+}
+
+/**
+ * k3_r5f_start() - Start the remote processor
+ * @dev:       rproc device pointer
+ *
+ * Return: 0 if all went ok, else return appropriate error
+ */
+static int k3_r5f_start(struct udevice *dev)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       struct k3_r5f_cluster *cluster = core->cluster;
+       int ret, c;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       ret = k3_r5f_core_sanity_check(core);
+       if (ret)
+               return ret;
+
+       ret = k3_r5f_proc_request(core);
+       if (ret)
+               return ret;
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP) {
+               if (is_primary_core(core)) {
+                       for (c = NR_CORES - 1; c >= 0; c--) {
+                               ret = k3_r5f_core_run(cluster->cores[c]);
+                               if (ret)
+                                       goto unroll_core_run;
+                       }
+               } else {
+                       dev_err(dev, "Invalid op: Trying to start secondary core %d in lockstep mode\n",
+                               core->tsp.proc_id);
+                       ret = -EINVAL;
+                       goto proc_release;
+               }
+       } else {
+               ret = k3_r5f_core_run(core);
+               if (ret)
+                       goto proc_release;
+       }
+
+       core->in_use = true;
+
+       k3_r5f_proc_release(core);
+       return 0;
+
+unroll_core_run:
+       while (c < NR_CORES) {
+               k3_r5f_core_halt(cluster->cores[c]);
+               c++;
+       }
+proc_release:
+       k3_r5f_proc_release(core);
+
+       return ret;
+}
+
+static int k3_r5f_split_reset(struct k3_r5f_core *core)
+{
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       if (reset_assert(&core->reset))
+               ret = -EINVAL;
+
+       if (ti_sci_proc_power_domain_off(&core->tsp))
+               ret = -EINVAL;
+
+       return ret;
+}
+
+static int k3_r5f_lockstep_reset(struct k3_r5f_cluster *cluster)
+{
+       int ret = 0, c;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       for (c = 0; c < NR_CORES; c++)
+               if (reset_assert(&cluster->cores[c]->reset))
+                       ret = -EINVAL;
+
+       /* disable PSC modules on all applicable cores */
+       for (c = 0; c < NR_CORES; c++)
+               if (ti_sci_proc_power_domain_off(&cluster->cores[c]->tsp))
+                       ret = -EINVAL;
+
+       return ret;
+}
+
+static int k3_r5f_unprepare(struct udevice *dev)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       struct k3_r5f_cluster *cluster = core->cluster;
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP) {
+               if (is_primary_core(core))
+                       ret = k3_r5f_lockstep_reset(cluster);
+       } else {
+               ret = k3_r5f_split_reset(core);
+       }
+
+       if (ret)
+               dev_warn(dev, "Unable to enable cores for TCM loading %d\n",
+                        ret);
+
+       return 0;
+}
+
+static int k3_r5f_stop(struct udevice *dev)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       struct k3_r5f_cluster *cluster = core->cluster;
+       int c, ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       ret = k3_r5f_proc_request(core);
+       if (ret)
+               return ret;
+
+       core->in_use = false;
+
+       if (cluster->mode == CLUSTER_MODE_LOCKSTEP) {
+               if (is_primary_core(core)) {
+                       for (c = 0; c < NR_CORES; c++)
+                               k3_r5f_core_halt(cluster->cores[c]);
+               } else {
+                       dev_err(dev, "Invalid op: Trying to stop secondary core in lockstep mode\n");
+                       ret = -EINVAL;
+                       goto proc_release;
+               }
+       } else {
+               k3_r5f_core_halt(core);
+       }
+
+       ret = k3_r5f_unprepare(dev);
+proc_release:
+       k3_r5f_proc_release(core);
+       return ret;
+}
+
+static void *k3_r5f_da_to_va(struct udevice *dev, ulong da, ulong size)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       void __iomem *va = NULL;
+       phys_addr_t bus_addr;
+       u32 dev_addr, offset;
+       ulong mem_size;
+       int i;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       if (size <= 0)
+               return NULL;
+
+       for (i = 0; i < core->num_mems; i++) {
+               bus_addr = core->mem[i].bus_addr;
+               dev_addr = core->mem[i].dev_addr;
+               mem_size = core->mem[i].size;
+
+               if (da >= bus_addr && (da + size) <= (bus_addr + mem_size)) {
+                       offset = da - bus_addr;
+                       va = core->mem[i].cpu_addr + offset;
+                       return (__force void *)va;
+               }
+
+               if (da >= dev_addr && (da + size) <= (dev_addr + mem_size)) {
+                       offset = da - dev_addr;
+                       va = core->mem[i].cpu_addr + offset;
+                       return (__force void *)va;
+               }
+       }
+
+       /* Assume it is DDR region and return da */
+       return map_physmem(da, size, MAP_NOCACHE);
+}
+
+static int k3_r5f_init(struct udevice *dev)
+{
+       return 0;
+}
+
+static int k3_r5f_reset(struct udevice *dev)
+{
+       return 0;
+}
+
+static const struct dm_rproc_ops k3_r5f_rproc_ops = {
+       .init = k3_r5f_init,
+       .reset = k3_r5f_reset,
+       .start = k3_r5f_start,
+       .stop = k3_r5f_stop,
+       .load = k3_r5f_load,
+       .device_to_virt = k3_r5f_da_to_va,
+};
+
+static int k3_r5f_rproc_configure(struct k3_r5f_core *core)
+{
+       struct k3_r5f_cluster *cluster = core->cluster;
+       u32 set_cfg = 0, clr_cfg = 0, cfg, ctrl, sts;
+       u64 boot_vec = 0;
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       ret = ti_sci_proc_request(&core->tsp);
+       if (ret < 0)
+               return ret;
+
+       /* Do not touch boot vector now. Load will take care of it. */
+       clr_cfg |= PROC_BOOT_CFG_FLAG_GEN_IGN_BOOTVECTOR;
+
+       ret = ti_sci_proc_get_status(&core->tsp, &boot_vec, &cfg, &ctrl, &sts);
+       if (ret)
+               goto out;
+
+       /* Sanity check for Lockstep mode */
+       if (cluster->mode && is_primary_core(core) &&
+           !(sts & PROC_BOOT_STATUS_FLAG_R5_LOCKSTEP_PERMITTED)) {
+               dev_err(core->dev, "LockStep mode not permitted on this device\n");
+               ret = -EINVAL;
+               goto out;
+       }
+
+       /* Primary core only configuration */
+       if (is_primary_core(core)) {
+               /* always enable ARM mode */
+               clr_cfg |= PROC_BOOT_CFG_FLAG_R5_TEINIT;
+               if (cluster->mode == CLUSTER_MODE_LOCKSTEP)
+                       set_cfg |= PROC_BOOT_CFG_FLAG_R5_LOCKSTEP;
+               else
+                       clr_cfg |= PROC_BOOT_CFG_FLAG_R5_LOCKSTEP;
+       }
+
+       if (core->atcm_enable)
+               set_cfg |= PROC_BOOT_CFG_FLAG_R5_ATCM_EN;
+       else
+               clr_cfg |= PROC_BOOT_CFG_FLAG_R5_ATCM_EN;
+
+       if (core->btcm_enable)
+               set_cfg |= PROC_BOOT_CFG_FLAG_R5_BTCM_EN;
+       else
+               clr_cfg |= PROC_BOOT_CFG_FLAG_R5_BTCM_EN;
+
+       if (core->loczrama)
+               set_cfg |= PROC_BOOT_CFG_FLAG_R5_TCM_RSTBASE;
+       else
+               clr_cfg |= PROC_BOOT_CFG_FLAG_R5_TCM_RSTBASE;
+
+       ret = k3_r5f_core_halt(core);
+       if (ret)
+               goto out;
+
+       ret = ti_sci_proc_set_config(&core->tsp, boot_vec, set_cfg, clr_cfg);
+out:
+       ti_sci_proc_release(&core->tsp);
+       return ret;
+}
+
+static int ti_sci_proc_of_to_priv(struct udevice *dev, struct ti_sci_proc *tsp)
+{
+       u32 ids[2];
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       tsp->sci = ti_sci_get_by_phandle(dev, "ti,sci");
+       if (IS_ERR(tsp->sci)) {
+               dev_err(dev, "ti_sci get failed: %ld\n", PTR_ERR(tsp->sci));
+               return PTR_ERR(tsp->sci);
+       }
+
+       ret = dev_read_u32_array(dev, "ti,sci-proc-ids", ids, 2);
+       if (ret) {
+               dev_err(dev, "Proc IDs not populated %d\n", ret);
+               return ret;
+       }
+
+       tsp->ops = &tsp->sci->ops.proc_ops;
+       tsp->proc_id = ids[0];
+       tsp->host_id = ids[1];
+       tsp->dev_id = dev_read_u32_default(dev, "ti,sci-dev-id",
+                                          TI_SCI_RESOURCE_NULL);
+       if (tsp->dev_id == TI_SCI_RESOURCE_NULL) {
+               dev_err(dev, "Device ID not populated %d\n", ret);
+               return -ENODEV;
+       }
+
+       return 0;
+}
+
+static int k3_r5f_of_to_priv(struct k3_r5f_core *core)
+{
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       core->atcm_enable = dev_read_u32_default(core->dev, "atcm-enable", 0);
+       core->btcm_enable = dev_read_u32_default(core->dev, "btcm-enable", 1);
+       core->loczrama = dev_read_u32_default(core->dev, "loczrama", 1);
+
+       ret = ti_sci_proc_of_to_priv(core->dev, &core->tsp);
+       if (ret)
+               return ret;
+
+       ret = reset_get_by_index(core->dev, 0, &core->reset);
+       if (ret) {
+               dev_err(core->dev, "Reset lines not available: %d\n", ret);
+               return ret;
+       }
+
+       return 0;
+}
+
+static int k3_r5f_core_of_get_memories(struct k3_r5f_core *core)
+{
+       static const char * const mem_names[] = {"atcm", "btcm"};
+       struct udevice *dev = core->dev;
+       int i;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       core->num_mems = ARRAY_SIZE(mem_names);
+       core->mem = calloc(core->num_mems, sizeof(*core->mem));
+       if (!core->mem)
+               return -ENOMEM;
+
+       for (i = 0; i < core->num_mems; i++) {
+               core->mem[i].bus_addr = dev_read_addr_size_name(dev,
+                                                               mem_names[i],
+                                       (fdt_addr_t *)&core->mem[i].size);
+               if (core->mem[i].bus_addr == FDT_ADDR_T_NONE) {
+                       dev_err(dev, "%s bus address not found\n",
+                               mem_names[i]);
+                       return -EINVAL;
+               }
+               core->mem[i].cpu_addr = map_physmem(core->mem[i].bus_addr,
+                                                   core->mem[i].size,
+                                                   MAP_NOCACHE);
+               if (!strcmp(mem_names[i], "atcm")) {
+                       core->mem[i].dev_addr = core->loczrama ?
+                                                       0 : K3_R5_TCM_DEV_ADDR;
+               } else {
+                       core->mem[i].dev_addr = core->loczrama ?
+                                                       K3_R5_TCM_DEV_ADDR : 0;
+               }
+
+               dev_dbg(dev, "memory %8s: bus addr %pa size 0x%zx va %p da 0x%x\n",
+                       mem_names[i], &core->mem[i].bus_addr,
+                       core->mem[i].size, core->mem[i].cpu_addr,
+                       core->mem[i].dev_addr);
+       }
+
+       return 0;
+}
+
+/**
+ * k3_r5f_probe() - Basic probe
+ * @dev:       corresponding k3 remote processor device
+ *
+ * Return: 0 if all goes good, else appropriate error message.
+ */
+static int k3_r5f_probe(struct udevice *dev)
+{
+       struct k3_r5f_cluster *cluster = dev_get_priv(dev->parent);
+       struct k3_r5f_core *core = dev_get_priv(dev);
+       bool r_state;
+       int ret;
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       core->dev = dev;
+       ret = k3_r5f_of_to_priv(core);
+       if (ret)
+               return ret;
+
+       core->cluster = cluster;
+       /* Assume Primary core gets probed first */
+       if (!cluster->cores[0])
+               cluster->cores[0] = core;
+       else
+               cluster->cores[1] = core;
+
+       ret = k3_r5f_core_of_get_memories(core);
+       if (ret) {
+               dev_err(dev, "Rproc getting internal memories failed\n");
+               return ret;
+       }
+
+       ret = core->tsp.sci->ops.dev_ops.is_on(core->tsp.sci, core->tsp.dev_id,
+                                              &r_state, &core->in_use);
+       if (ret)
+               return ret;
+
+       if (core->in_use) {
+               dev_info(dev, "Core %d is already in use. No rproc commands work\n",
+                        core->tsp.proc_id);
+               return 0;
+       }
+
+       /* Make sure Local reset is asserted. Redundant? */
+       reset_assert(&core->reset);
+
+       ret = k3_r5f_rproc_configure(core);
+       if (ret) {
+               dev_err(dev, "rproc configure failed %d\n", ret);
+               return ret;
+       }
+
+       dev_dbg(dev, "Remoteproc successfully probed\n");
+
+       return 0;
+}
+
+static int k3_r5f_remove(struct udevice *dev)
+{
+       struct k3_r5f_core *core = dev_get_priv(dev);
+
+       free(core->mem);
+
+       ti_sci_proc_release(&core->tsp);
+
+       return 0;
+}
+
+static const struct udevice_id k3_r5f_rproc_ids[] = {
+       { .compatible = "ti,am654-r5f"},
+       { .compatible = "ti,j721e-r5f"},
+       {}
+};
+
+U_BOOT_DRIVER(k3_r5f_rproc) = {
+       .name = "k3_r5f_rproc",
+       .of_match = k3_r5f_rproc_ids,
+       .id = UCLASS_REMOTEPROC,
+       .ops = &k3_r5f_rproc_ops,
+       .probe = k3_r5f_probe,
+       .remove = k3_r5f_remove,
+       .priv_auto_alloc_size = sizeof(struct k3_r5f_core),
+};
+
+static int k3_r5f_cluster_probe(struct udevice *dev)
+{
+       struct k3_r5f_cluster *cluster = dev_get_priv(dev);
+
+       dev_dbg(dev, "%s\n", __func__);
+
+       cluster->mode = dev_read_u32_default(dev, "lockstep-mode",
+                                            CLUSTER_MODE_LOCKSTEP);
+
+       if (device_get_child_count(dev) != 2) {
+               dev_err(dev, "Invalid number of R5 cores");
+               return -EINVAL;
+       }
+
+       dev_dbg(dev, "%s: Cluster successfully probed in %s mode\n",
+               __func__, cluster->mode ? "lockstep" : "split");
+
+       return 0;
+}
+
+static const struct udevice_id k3_r5fss_ids[] = {
+       { .compatible = "ti,am654-r5fss"},
+       { .compatible = "ti,j721e-r5fss"},
+       {}
+};
+
+U_BOOT_DRIVER(k3_r5fss) = {
+       .name = "k3_r5fss",
+       .of_match = k3_r5fss_ids,
+       .id = UCLASS_MISC,
+       .probe = k3_r5f_cluster_probe,
+       .priv_auto_alloc_size = sizeof(struct k3_r5f_cluster),
+};