remoteproc: Extend device_to_virt with a is_iomem parameter

Some areas needs to be initialized by using memcpy_toio and memset_io.
Following Linux Kernel commit: 40df0a91b2a5 ("remoteproc: add is_iomem to
da_to_va"), add this to U-Boot.

Reviewed-by: Ye Li <ye.li@nxp.com>
Signed-off-by: Peng Fan <peng.fan@nxp.com>

drivers/remoteproc/renesas_apmu.c

@@ -170,11 +170,12 @@ static int renesas_apmu_rproc_init(struct udevice *dev)
  * @dev:	corresponding remote processor device
  * @da:		device address
  * @size:	Size of the memory region @da is pointing to
+ * @is_iomem:	optional pointer filled in to indicate if @da is iomapped memory
  *
  * Return: converted virtual address
  */
 static void *renesas_apmu_rproc_device_to_virt(struct udevice *dev, ulong da,
-					       ulong size)
+					       ulong size, bool *is_iomem)
 {
 	/*
 	 * The Cortex R52 and A76 share the same address space,

drivers/remoteproc/rproc-elf-loader.c

@@ -1,6 +1,7 @@
 // SPDX-License-Identifier: GPL-2.0+ OR BSD-3-Clause
 /*
  * Copyright (C) 2019, STMicroelectronics - All Rights Reserved
+ * Copyright 2025 NXP
  */
 #include <cpu_func.h>
 #include <dm.h>
@@ -9,6 +10,7 @@
 #include <mapmem.h>
 #include <remoteproc.h>
 #include <asm/cache.h>
+#include <asm/io.h>
 #include <dm/device_compat.h>
 #include <linux/compat.h>
 #include <linux/printk.h>
@@ -181,27 +183,38 @@ int rproc_elf32_load_image(struct udevice *dev, unsigned long addr, ulong size)
 	for (i = 0; i < ehdr->e_phnum; i++, phdr++) {
 		void *dst = (void *)(uintptr_t)phdr->p_paddr;
 		void *src = (void *)addr + phdr->p_offset;
+		bool is_iomem = false;
 		ulong dst_addr;
 
-		if (phdr->p_type != PT_LOAD)
+		if (phdr->p_type != PT_LOAD || !phdr->p_memsz)
 			continue;
 
 		if (ops->device_to_virt)
 			dst = ops->device_to_virt(dev, (ulong)dst,
-						  phdr->p_memsz);
+						  phdr->p_memsz, &is_iomem);
 
 		dev_dbg(dev, "Loading phdr %i to 0x%p (%i bytes)\n",
 			i, dst, phdr->p_filesz);
-		if (phdr->p_filesz)
+		if (phdr->p_filesz) {
-			memcpy(dst, src, phdr->p_filesz);
+			if (is_iomem)
-		if (phdr->p_filesz != phdr->p_memsz)
+				memcpy_toio(dst, src, phdr->p_filesz);
-			memset(dst + phdr->p_filesz, 0x00,
+			else
-			       phdr->p_memsz - phdr->p_filesz);
+				memcpy(dst, src, phdr->p_filesz);
+		}
+		if (phdr->p_filesz != phdr->p_memsz) {
+			if (is_iomem)
+				memset_io(dst + phdr->p_filesz, 0x00,
+					  phdr->p_memsz - phdr->p_filesz);
+			else
+				memset(dst + phdr->p_filesz, 0x00,
+				       phdr->p_memsz - phdr->p_filesz);
+		}
 		dst_addr = map_to_sysmem(dst);
-		flush_cache(rounddown(dst_addr, ARCH_DMA_MINALIGN),
+		if (!is_iomem) {
-			    roundup(dst_addr + phdr->p_filesz,
+			flush_cache(rounddown(dst_addr, ARCH_DMA_MINALIGN),
-				    ARCH_DMA_MINALIGN) -
+				    roundup(dst_addr + phdr->p_filesz, ARCH_DMA_MINALIGN) -
-			    rounddown(dst_addr, ARCH_DMA_MINALIGN));
+				    rounddown(dst_addr, ARCH_DMA_MINALIGN));
+		}
 	}
 
 	return 0;
@@ -230,6 +243,7 @@ int rproc_elf64_load_image(struct udevice *dev, ulong addr, ulong size)
 		memsz = phdr->p_memsz;
 		filesz = phdr->p_filesz;
 		offset = phdr->p_offset;
+		bool is_iomem = false;
 
 		if (phdr->p_type != PT_LOAD)
 			continue;
@@ -239,7 +253,7 @@ int rproc_elf64_load_image(struct udevice *dev, ulong addr, ulong size)
 
 		ptr = (void *)(uintptr_t)da;
 		if (ops->device_to_virt) {
-			ptr = ops->device_to_virt(dev, da, phdr->p_memsz);
+			ptr = ops->device_to_virt(dev, da, phdr->p_memsz, &is_iomem);
 			if (!ptr) {
 				dev_err(dev, "bad da 0x%llx mem 0x%llx\n", da,
 					memsz);
@@ -248,14 +262,24 @@ int rproc_elf64_load_image(struct udevice *dev, ulong addr, ulong size)
 			}
 		}
 
-		if (filesz)
+		if (filesz) {
-			memcpy(ptr, (void *)addr + offset, filesz);
+			if (is_iomem)
-		if (filesz != memsz)
+				memcpy_toio(ptr, (void *)addr + offset, filesz);
-			memset(ptr + filesz, 0x00, memsz - filesz);
+			else
+				memcpy(ptr, (void *)addr + offset, filesz);
+		}
+		if (filesz != memsz) {
+			if (is_iomem)
+				memset_io(ptr + filesz, 0x00, memsz - filesz);
+			else
+				memset(ptr + filesz, 0x00, memsz - filesz);
+		}
 
-		flush_cache(rounddown((ulong)ptr, ARCH_DMA_MINALIGN),
+		if (!is_iomem) {
-			    roundup((ulong)ptr + filesz, ARCH_DMA_MINALIGN) -
+			flush_cache(rounddown((ulong)ptr, ARCH_DMA_MINALIGN),
-			    rounddown((ulong)ptr, ARCH_DMA_MINALIGN));
+				    roundup((ulong)ptr + filesz, ARCH_DMA_MINALIGN) -
+				    rounddown((ulong)ptr, ARCH_DMA_MINALIGN));
+		}
 	}
 
 	return ret;
@@ -381,6 +405,7 @@ int rproc_elf32_load_rsc_table(struct udevice *dev, ulong fw_addr,
 	Elf32_Shdr *shdr;
 	void *src, *dst;
 	ulong dst_addr;
+	bool is_iomem = false;
 
 	shdr = rproc_elf32_find_rsc_table(dev, fw_addr, fw_size);
 	if (!shdr)
@@ -394,18 +419,22 @@ int rproc_elf32_load_rsc_table(struct udevice *dev, ulong fw_addr,
 
 	src = (void *)fw_addr + shdr->sh_offset;
 	if (ops->device_to_virt)
-		dst = (void *)ops->device_to_virt(dev, *rsc_addr, *rsc_size);
+		dst = (void *)ops->device_to_virt(dev, *rsc_addr, *rsc_size, &is_iomem);
 	else
 		dst = (void *)rsc_addr;
 
 	dev_dbg(dev, "Loading resource table to 0x%8lx (%ld bytes)\n",
 		(ulong)dst, *rsc_size);
 
-	memcpy(dst, src, *rsc_size);
+	if (is_iomem) {
-	dst_addr = map_to_sysmem(dst);
+		memcpy_toio(dst, src, *rsc_size);
-	flush_cache(rounddown(dst_addr, ARCH_DMA_MINALIGN),
+	} else {
-		    roundup(dst_addr + *rsc_size, ARCH_DMA_MINALIGN) -
+		memcpy(dst, src, *rsc_size);
-		    rounddown(dst_addr, ARCH_DMA_MINALIGN));
+		dst_addr = map_to_sysmem(dst);
+		flush_cache(rounddown(dst_addr, ARCH_DMA_MINALIGN),
+			    roundup(dst_addr + *rsc_size, ARCH_DMA_MINALIGN) -
+			    rounddown(dst_addr, ARCH_DMA_MINALIGN));
+	}
 
 	return 0;
 }
@@ -490,6 +519,7 @@ int rproc_elf64_load_rsc_table(struct udevice *dev, ulong fw_addr,
 	const struct dm_rproc_ops *ops;
 	Elf64_Shdr *shdr;
 	void *src, *dst;
+	bool is_iomem = false;
 
 	shdr = rproc_elf64_find_rsc_table(dev, fw_addr, fw_size);
 	if (!shdr)
@@ -503,18 +533,21 @@ int rproc_elf64_load_rsc_table(struct udevice *dev, ulong fw_addr,
 
 	src = (void *)fw_addr + shdr->sh_offset;
 	if (ops->device_to_virt)
-		dst = (void *)ops->device_to_virt(dev, *rsc_addr, *rsc_size);
+		dst = (void *)ops->device_to_virt(dev, *rsc_addr, *rsc_size, &is_iomem);
 	else
 		dst = (void *)rsc_addr;
 
 	dev_dbg(dev, "Loading resource table to 0x%8lx (%ld bytes)\n",
 		(ulong)dst, *rsc_size);
 
-	memcpy(dst, src, *rsc_size);
+	if (is_iomem) {
-	flush_cache(rounddown((unsigned long)dst, ARCH_DMA_MINALIGN),
+		memcpy_toio(dst, src, *rsc_size);
-		    roundup((unsigned long)dst + *rsc_size,
+	} else {
-			    ARCH_DMA_MINALIGN) -
+		memcpy(dst, src, *rsc_size);
-		    rounddown((unsigned long)dst, ARCH_DMA_MINALIGN));
+		flush_cache(rounddown((unsigned long)dst, ARCH_DMA_MINALIGN),
+			    roundup((unsigned long)dst + *rsc_size, ARCH_DMA_MINALIGN) -
+			    rounddown((unsigned long)dst, ARCH_DMA_MINALIGN));
+	}
 
 	return 0;
 }

drivers/remoteproc/sandbox_testproc.c

@@ -308,10 +308,11 @@ static int sandbox_testproc_ping(struct udevice *dev)
  * @dev:	device to operate upon
  * @da:		device address
  * @size:	Size of the memory region @da is pointing to
+ * @is_iomem:	optional pointer filled in to indicate if @da is iomapped memory
  * Return: converted virtual address
  */
 static void *sandbox_testproc_device_to_virt(struct udevice *dev, ulong da,
-					     ulong size)
+					     ulong size, bool *is_iomem)
 {
 	u64 paddr;
 

drivers/remoteproc/stm32_copro.c

@@ -61,10 +61,11 @@ static int stm32_copro_probe(struct udevice *dev)
  * @dev:	corresponding STM32 remote processor device
  * @da:		device address
  * @size:	Size of the memory region @da is pointing to
+ * @is_iomem:	optional pointer filled in to indicate if @da is iomapped memory
  * Return: converted virtual address
  */
 static void *stm32_copro_device_to_virt(struct udevice *dev, ulong da,
-					ulong size)
+					ulong size, bool *is_iomem)
 {
 	fdt32_t in_addr = cpu_to_be32(da), end_addr;
 	u64 paddr;

drivers/remoteproc/ti_k3_dsp_rproc.c

@@ -261,7 +261,7 @@ static int k3_dsp_reset(struct udevice *dev)
 	return 0;
 }
 
-static void *k3_dsp_da_to_va(struct udevice *dev, ulong da, ulong len)
+static void *k3_dsp_da_to_va(struct udevice *dev, ulong da, ulong len, bool *is_iomem)
 {
 	struct k3_dsp_privdata *dsp = dev_get_priv(dev);
 	phys_addr_t bus_addr, dev_addr;

drivers/remoteproc/ti_k3_m4_rproc.c

@@ -181,7 +181,7 @@ static int k3_m4_stop(struct udevice *dev)
 	return 0;
 }
 
-static void *k3_m4_da_to_va(struct udevice *dev, ulong da, ulong len)
+static void *k3_m4_da_to_va(struct udevice *dev, ulong da, ulong len, bool *is_iomem)
 {
 	struct k3_m4_privdata *m4 = dev_get_priv(dev);
 	phys_addr_t bus_addr, dev_addr;

drivers/remoteproc/ti_k3_r5f_rproc.c

@@ -534,7 +534,7 @@ proc_release:
 	return ret;
 }
 
-static void *k3_r5f_da_to_va(struct udevice *dev, ulong da, ulong size)
+static void *k3_r5f_da_to_va(struct udevice *dev, ulong da, ulong size, bool *is_iomem)
 {
 	struct k3_r5f_core *core = dev_get_priv(dev);
 	void __iomem *va = NULL;

include/remoteproc.h

@@ -495,9 +495,10 @@ struct dm_rproc_ops {
 	 * @dev:	Remote proc device
 	 * @da:		Device address
 	 * @size:	Size of the memory region @da is pointing to
+	 * @is_iomem:	optional pointer filled in to indicate if @da is iomapped memory
 	 * @return virtual address.
 	 */
-	void * (*device_to_virt)(struct udevice *dev, ulong da, ulong size);
+	void * (*device_to_virt)(struct udevice *dev, ulong da, ulong size, bool *is_iomem);
 	int (*add_res)(struct udevice *dev,
 		       struct rproc_mem_entry *mapping);
 	void * (*alloc_mem)(struct udevice *dev, unsigned long len,