[rt-smart] PV_OFFSET as a variable (#6904)

* [rt-smart/mem] remove pv_offset

* [rt-smart] list kernel space command

* [rt-smart] restore ioremap region

* [revert] restore kernel space isolation

* [rt-smart/pv_off] code format

* [rt-smart] add get_pvoff()

* [pvoffset] pvoff as constant for C codes

* [pvoff] pvoff as interfaces

23 files changed, 122 insertions, 64 deletions

diff --git a/bsp/allwinner/d1s/.config b/bsp/allwinner/d1s/.config
index 73a1e6d0e6..9db98662c6 100644
--- a/bsp/allwinner/d1s/.config
+++ b/bsp/allwinner/d1s/.config
@@ -92,7 +92,6 @@ CONFIG_RT_USING_CACHE=y
 # CONFIG_RT_USING_CPU_FFS is not set
 CONFIG_ARCH_MM_MMU=y
 CONFIG_KERNEL_VADDR_START=0x40000000
-CONFIG_PV_OFFSET=0x0
 CONFIG_ARCH_RISCV=y
 CONFIG_ARCH_RISCV64=y
 
diff --git a/bsp/allwinner/d1s/SConstruct b/bsp/allwinner/d1s/SConstruct
index 0b4b6b05fb..4efd6064d4 100644
--- a/bsp/allwinner/d1s/SConstruct
+++ b/bsp/allwinner/d1s/SConstruct
@@ -94,7 +94,7 @@ stack_size = 4096
 
 stack_lds = open('link_stacksize.lds', 'w')
 if GetDepend('__STACKSIZE__'): stack_size = GetDepend('__STACKSIZE__')
-stack_lds.write('__STACKSIZE__ = %d;' % stack_size)
+stack_lds.write('__STACKSIZE__ = %d;\n' % stack_size)
 stack_lds.close()
 
 # make a building
diff --git a/bsp/allwinner/d1s/board/board.c b/bsp/allwinner/d1s/board/board.c
index f121710994..49bed66e90 100644
--- a/bsp/allwinner/d1s/board/board.c
+++ b/bsp/allwinner/d1s/board/board.c
@@ -44,8 +44,8 @@ rt_region_t init_page_region =
 extern volatile rt_size_t MMUTable[__SIZE(VPN2_BIT)] __attribute__((aligned(4 * 1024)));
 
 struct mem_desc platform_mem_desc[] = {
-    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x4000000 - 1, KERNEL_VADDR_START + PV_OFFSET, NORMAL_MEM},
-    {0x1000, 0x3ffff000 - 1, 0x1000 + PV_OFFSET, DEVICE_MEM},
+    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x4000000 - 1, (rt_size_t)ARCH_MAP_FAILED, NORMAL_MEM},
+    {0x1000, 0x3ffff000 - 1, (rt_size_t)ARCH_MAP_FAILED, DEVICE_MEM},
 };
 
 #define NUM_MEM_DESC (sizeof(platform_mem_desc) / sizeof(platform_mem_desc[0]))
diff --git a/bsp/allwinner/d1s/board/board.h b/bsp/allwinner/d1s/board/board.h
index e9e7fce1a3..b5cd40c43e 100644
--- a/bsp/allwinner/d1s/board/board.h
+++ b/bsp/allwinner/d1s/board/board.h
@@ -17,14 +17,14 @@
 extern unsigned int __bss_start;
 extern unsigned int __bss_end;
 
-#define RAM_SIZE    (64 * 1024 * 1024)
-#define RAM_BASE    (0x40000000)
-#define RAM_END     (RAM_BASE + RAM_SIZE - PV_OFFSET)
+#define RAM_SIZE            (64 * 1024 * 1024)
+#define RAM_BASE            (0x40000000)
+#define RAM_END_VADDR       (KERNEL_VADDR_START + RAM_SIZE)
 
 #define RT_HW_HEAP_BEGIN    ((void *)&__bss_end)
 #define RT_HW_HEAP_END      ((void *)(((rt_size_t)RT_HW_HEAP_BEGIN) + 16 * 1024 * 1024))
 #define RT_HW_PAGE_START    (RT_HW_HEAP_END)
-#define RT_HW_PAGE_END      ((void *)RAM_END)
+#define RT_HW_PAGE_END      ((void *)RAM_END_VADDR)
 
 void rt_hw_board_init(void);
 void rt_init_user_mem(struct rt_thread *thread, const char *name, unsigned long *entry);
diff --git a/bsp/allwinner/d1s/rtconfig.h b/bsp/allwinner/d1s/rtconfig.h
index 8982bff72f..0f782b583a 100644
--- a/bsp/allwinner/d1s/rtconfig.h
+++ b/bsp/allwinner/d1s/rtconfig.h
@@ -55,7 +55,6 @@
 #define RT_USING_CACHE
 #define ARCH_MM_MMU
 #define KERNEL_VADDR_START 0x40000000
-#define PV_OFFSET 0x0
 #define ARCH_RISCV
 #define ARCH_RISCV64
 
diff --git a/bsp/qemu-vexpress-a9/drivers/board.c b/bsp/qemu-vexpress-a9/drivers/board.c
index 8e5c36cf16..458069b350 100644
--- a/bsp/qemu-vexpress-a9/drivers/board.c
+++ b/bsp/qemu-vexpress-a9/drivers/board.c
@@ -26,7 +26,7 @@
 
 #ifdef RT_USING_SMART
 struct mem_desc platform_mem_desc[] = {
-    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x0fffffff, KERNEL_VADDR_START + PV_OFFSET, NORMAL_MEM}
+    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x0fffffff, (rt_size_t)ARCH_MAP_FAILED, NORMAL_MEM}
 };
 #else
 struct mem_desc platform_mem_desc[] = {
diff --git a/bsp/qemu-virt64-aarch64/drivers/board.c b/bsp/qemu-virt64-aarch64/drivers/board.c
index 823ac8c784..ce8b88151d 100644
--- a/bsp/qemu-virt64-aarch64/drivers/board.c
+++ b/bsp/qemu-virt64-aarch64/drivers/board.c
@@ -11,7 +11,6 @@
  *                           add smp ipi init
  */
 
-#include "mm_aspace.h"
 #include <rthw.h>
 #include <rtthread.h>
 #include <mmu.h>
@@ -21,6 +20,7 @@
 #endif
 
 #include "board.h"
+#include <mm_aspace.h>
 #include <mm_page.h>
 #include <interrupt.h>
 
@@ -33,7 +33,7 @@
 
 #ifdef RT_USING_SMART
 struct mem_desc platform_mem_desc[] = {
-    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x0fffffff, KERNEL_VADDR_START + PV_OFFSET, NORMAL_MEM}
+    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x0fffffff, (rt_size_t)ARCH_MAP_FAILED, NORMAL_MEM}
 };
 #else
 
diff --git a/bsp/qemu-virt64-aarch64/drivers/board.h b/bsp/qemu-virt64-aarch64/drivers/board.h
index 10da6f51e2..cb0aaaea60 100644
--- a/bsp/qemu-virt64-aarch64/drivers/board.h
+++ b/bsp/qemu-virt64-aarch64/drivers/board.h
@@ -26,7 +26,6 @@ extern unsigned char __bss_end;
 #else
 #define HEAP_END    ((void *)HEAP_BEGIN + 64 * 1024 * 1024)
 #define KERNEL_VADDR_START 0x40000000
-#define PV_OFFSET 0
 #endif
 
 void rt_hw_board_init(void);
diff --git a/bsp/qemu-virt64-riscv/SConstruct b/bsp/qemu-virt64-riscv/SConstruct
index 422e592ccd..a1d00bea94 100644
--- a/bsp/qemu-virt64-riscv/SConstruct
+++ b/bsp/qemu-virt64-riscv/SConstruct
@@ -31,7 +31,7 @@ stack_size = 4096
 
 stack_lds = open('link_stacksize.lds', 'w')
 if GetDepend('__STACKSIZE__'): stack_size = GetDepend('__STACKSIZE__')
-stack_lds.write('__STACKSIZE__ = %d;' % stack_size)
+stack_lds.write('__STACKSIZE__ = %d;\n' % stack_size)
 stack_lds.close()
 
 # make a building
diff --git a/bsp/qemu-virt64-riscv/driver/board.c b/bsp/qemu-virt64-riscv/driver/board.c
index e1aa2ab977..cbbe80ee03 100644
--- a/bsp/qemu-virt64-riscv/driver/board.c
+++ b/bsp/qemu-virt64-riscv/driver/board.c
@@ -35,7 +35,7 @@ rt_region_t init_page_region = {(rt_size_t)RT_HW_PAGE_START, (rt_size_t)RT_HW_PA
 extern size_t MMUTable[];
 
 struct mem_desc platform_mem_desc[] = {
-    {KERNEL_VADDR_START, KERNEL_VADDR_START + 0x10000000 - 1, KERNEL_VADDR_START + PV_OFFSET, NORMAL_MEM},
+    {KERNEL_VADDR_START, (rt_size_t)RT_HW_PAGE_END - 1, (rt_size_t)ARCH_MAP_FAILED, NORMAL_MEM},
 };
 
 #define NUM_MEM_DESC (sizeof(platform_mem_desc) / sizeof(platform_mem_desc[0]))
@@ -54,11 +54,17 @@ void primary_cpu_entry(void)
 
 #define IOREMAP_SIZE (1ul << 30)
 
+#ifndef ARCH_KERNEL_IN_HIGH_VA
+#define IOREMAP_VEND USER_VADDR_START
+#else
+#define IOREMAP_VEND 0ul
+#endif
+
 void rt_hw_board_init(void)
 {
 #ifdef RT_USING_SMART
     /* init data structure */
-    rt_hw_mmu_map_init(&rt_kernel_space, (void *)(USER_VADDR_START - IOREMAP_SIZE), IOREMAP_SIZE, (rt_size_t *)MMUTable, 0);
+    rt_hw_mmu_map_init(&rt_kernel_space, (void *)(IOREMAP_VEND - IOREMAP_SIZE), IOREMAP_SIZE, (rt_size_t *)MMUTable, PV_OFFSET);
 
     /* init page allocator */
     rt_page_init(init_page_region);
diff --git a/components/mm/mm_aspace.c b/components/mm/mm_aspace.c
index df296b1782..6a969074dc 100644
--- a/components/mm/mm_aspace.c
+++ b/components/mm/mm_aspace.c
@@ -31,10 +31,6 @@
 #include <mmu.h>
 #include <tlb.h>
 
-#ifndef RT_USING_SMART
-#define PV_OFFSET 0
-#endif
-
 static void _aspace_unmap(rt_aspace_t aspace, void *addr, rt_size_t length);
 static void *_find_free(rt_aspace_t aspace, void *prefer, rt_size_t req_size,
                         void *limit_start, rt_size_t limit_size,
@@ -488,6 +484,7 @@ int rt_aspace_map_phy_static(rt_aspace_t aspace, rt_varea_t varea,
         varea->start = hint->prefer;
         varea->size = hint->map_size;
         hint->flags |= MMF_MAP_FIXED;
+        LOG_D("%s: start %p size %p phy at %p", __func__, varea->start, varea->size, pa_off << MM_PAGE_SHIFT);
         err = _mm_aspace_map_phy(aspace, varea, hint, attr, pa_off, ret_va);
     }
     else
@@ -743,8 +740,16 @@ int rt_aspace_traversal(rt_aspace_t aspace,
 
 static int _dump(rt_varea_t varea, void *arg)
 {
-    rt_kprintf("%s[%p - %p]\n", varea->mem_obj->get_name(varea), varea->start,
-               varea->start + varea->size);
+    if (varea->mem_obj && varea->mem_obj->get_name)
+    {
+        rt_kprintf("[%p - %p] %s\n", varea->start, varea->start + varea->size,
+                   varea->mem_obj->get_name(varea));
+    }
+    else
+    {
+        rt_kprintf("[%p - %p] phy-map\n", varea->start, varea->start + varea->size);
+        rt_kprintf("\t\\_ paddr = %p\n",  varea->offset << MM_PAGE_SHIFT);
+    }
     return 0;
 }
 
diff --git a/components/mm/mm_aspace.h b/components/mm/mm_aspace.h
index a8cf6e8339..8687f9a093 100644
--- a/components/mm/mm_aspace.h
+++ b/components/mm/mm_aspace.h
@@ -20,6 +20,7 @@
 
 #define MM_PAGE_SHIFT    12
 #define MM_PA_TO_OFF(pa) ((uintptr_t)(pa) >> MM_PAGE_SHIFT)
+#define PV_OFFSET        (rt_kmem_pvoff())
 
 #ifndef RT_USING_SMP
 typedef rt_spinlock_t mm_spinlock;
@@ -201,4 +202,8 @@ void rt_varea_free_pages(rt_varea_t varea);
 
 void rt_varea_offload_page(rt_varea_t varea, void *vaddr, rt_size_t size);
 
+rt_ubase_t rt_kmem_pvoff(void);
+
+void rt_kmem_pvoff_set(rt_ubase_t pvoff);
+
 #endif /* __MM_ASPACE_H__ */
diff --git a/components/mm/mm_kmem.c b/components/mm/mm_kmem.c
new file mode 100644
index 0000000000..4d7854f04a
--- /dev/null
+++ b/components/mm/mm_kmem.c
@@ -0,0 +1,32 @@
+/*
+ * Copyright (c) 2006-2022, RT-Thread Development Team
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ *
+ * Change Logs:
+ * Date           Author       Notes
+ * 2022-11-14     WangXiaoyao  the first version
+ */
+
+#include "mm_aspace.h"
+#include "mm_private.h"
+
+static void list_kernel_space(void)
+{
+    rt_aspace_print_all(&rt_kernel_space);
+}
+MSH_CMD_EXPORT(list_kernel_space, List varea in kernel space);
+
+void rt_kmem_list_varea(void) __attribute__((alias("list_kernel_space")));
+
+static rt_ubase_t rt_pv_offset;
+
+rt_ubase_t rt_kmem_pvoff(void)
+{
+    return rt_pv_offset;
+}
+
+void rt_kmem_pvoff_set(rt_ubase_t pvoff)
+{
+    rt_pv_offset = pvoff;
+}
diff --git a/components/mm/mm_page.c b/components/mm/mm_page.c
index 2da22457a8..bacade4f5e 100644
--- a/components/mm/mm_page.c
+++ b/components/mm/mm_page.c
@@ -40,8 +40,6 @@ CT_ASSERT(order_huge_pg, RT_PAGE_MAX_ORDER > ARCH_PAGE_SHIFT - 2);
 CT_ASSERT(size_width, sizeof(rt_size_t) == sizeof(rt_size_t));
 #endif /* ARCH_CPU_64BIT */
 
-#else
-#define PV_OFFSET 0
 #endif /* RT_USING_SMART */
 
 static rt_size_t init_mpr_align_start;
diff --git a/libcpu/Kconfig b/libcpu/Kconfig
index 616600b179..17a57199b9 100644
--- a/libcpu/Kconfig
+++ b/libcpu/Kconfig
@@ -87,13 +87,6 @@ config KERNEL_VADDR_START
     default 0x80000000 if ARCH_RISCV
     depends on ARCH_MM_MMU
 
-config PV_OFFSET
-    hex "The offset of kernel physical address and virtural address"
-    default 0x1000040000000 if ARCH_ARMV8
-    default 0xa0000000      if ARCH_ARM
-    default 0x0             if ARCH_RISCV
-    depends on ARCH_MM_MMU
-
 config RT_IOREMAP_LATE
     bool "Support to create IO mapping in the kernel address space after system initlalization."
     default n
diff --git a/libcpu/aarch64/common/mmu.c b/libcpu/aarch64/common/mmu.c
index d567d111d9..b175450dc7 100644
--- a/libcpu/aarch64/common/mmu.c
+++ b/libcpu/aarch64/common/mmu.c
@@ -309,6 +309,9 @@ void rt_hw_mmu_setup(rt_aspace_t aspace, struct mem_desc *mdesc, int desc_nr)
                                                 mdesc->vaddr_start + 1,
                                     .prefer = (void *)mdesc->vaddr_start};
 
+        if (mdesc->paddr_start == (rt_size_t)ARCH_MAP_FAILED)
+            mdesc->paddr_start = mdesc->vaddr_start + PV_OFFSET;
+
         rt_aspace_map_phy_static(aspace, &mdesc->varea, &hint, attr,
                                  mdesc->paddr_start >> MM_PAGE_SHIFT, &err);
         mdesc++;
@@ -656,10 +659,14 @@ int rt_hw_mmu_control(struct rt_aspace *aspace, void *vaddr, size_t size,
     return err;
 }
 
-void rt_hw_mmu_setup_early(unsigned long *tbl0, unsigned long *tbl1,
+void rt_hw_mem_setup_early(unsigned long *tbl0, unsigned long *tbl1,
                            unsigned long size, unsigned long pv_off)
 {
     int ret;
+
+    /* setup pv off */
+    rt_kmem_pvoff_set(pv_off);
+
     unsigned long va = KERNEL_VADDR_START;
     unsigned long count = (size + ARCH_SECTION_MASK) >> ARCH_SECTION_SHIFT;
     unsigned long normal_attr = MMU_MAP_CUSTOM(MMU_AP_KAUN, NORMAL_MEM);
diff --git a/libcpu/aarch64/common/mmu.h b/libcpu/aarch64/common/mmu.h
index 6dc8cdca31..912ee6fdb9 100644
--- a/libcpu/aarch64/common/mmu.h
+++ b/libcpu/aarch64/common/mmu.h
@@ -74,7 +74,7 @@ struct mem_desc
 struct rt_aspace;
 
 void rt_hw_mmu_ktbl_set(unsigned long tbl);
-void rt_hw_mmu_setup_early(unsigned long *tbl0, unsigned long *tbl1,
+void rt_hw_mem_setup_early(unsigned long *tbl0, unsigned long *tbl1,
                            unsigned long size, unsigned long pv_off);
 void rt_hw_mmu_setup(struct rt_aspace *aspace, struct mem_desc *mdesc,
                      int desc_nr);
diff --git a/libcpu/aarch64/common/trap.c b/libcpu/aarch64/common/trap.c
index 0a2dd2e9ed..7527440f9f 100644
--- a/libcpu/aarch64/common/trap.c
+++ b/libcpu/aarch64/common/trap.c
@@ -45,6 +45,7 @@ void check_user_fault(struct rt_hw_exp_stack *regs, uint32_t pc_adj, char *info)
 #ifdef LWP_USING_CORE_DUMP
         lwp_core_dump(regs, pc_adj);
 #endif
+        backtrace((unsigned long)regs->pc, (unsigned long)regs->x30, (unsigned long)regs->x29);
         sys_exit(-1);
     }
 }
diff --git a/libcpu/aarch64/cortex-a/entry_point.S b/libcpu/aarch64/cortex-a/entry_point.S
index 89c383a1e5..a73b420781 100644
--- a/libcpu/aarch64/cortex-a/entry_point.S
+++ b/libcpu/aarch64/cortex-a/entry_point.S
@@ -78,12 +78,12 @@ __start:
 
     eret                            /* exception return. from EL2. continue from .L__in_el1 */
 
+.macro GET_PHY reg, symbol
+    adrp    \reg, \symbol
+    add     \reg, \reg, #:lo12:\symbol
+.endm
+
 .L__in_el1:
-#ifdef RT_USING_LWP
-    ldr     x9, =PV_OFFSET
-#else
-    mov     x9, #0
-#endif
     mov     sp, x1                  /* in EL1. Set sp to _start */
 
     /* Set CPACR_EL1 (Architecture Feature Access Control Register) to avoid trap from SIMD or float point instruction */
@@ -91,15 +91,13 @@ __start:
     msr     cpacr_el1, x1
 
     /* clear bss */
-    adrp    x1, __bss_start     /* get bss start address */
-    add	    x1, x1, #:lo12:__bss_start
-    adrp    x2, __bss_end
-    add	    x1, x1, #:lo12:__bss_end
-    sub     x2, x2, x1            /* get bss size          */
+    GET_PHY x1, __bss_start
+    GET_PHY x2, __bss_end
+    sub     x2, x2, x1              /* get bss size */
 
-    and     x3, x2, #7           /* x3 is < 7 */
+    and     x3, x2, #7              /* x3 is < 7 */
     ldr     x4, =~0x7
-    and     x2, x2, x4            /* mask ~7 */
+    and     x2, x2, x4              /* mask ~7 */
 
 .L__clean_bss_loop:
     cbz     x2, .L__clean_bss_loop_1
@@ -116,7 +114,7 @@ __start:
 .L__jump_to_entry:          /* jump to C code, should not return */        
     bl mmu_tcr_init
 
-    adr x1, __start
+    adr x1, __start         /* install early page table */
     ldr x0, =~0x1fffff
     and x0, x1, x0
     add x1, x0, #0x1000
@@ -125,11 +123,13 @@ __start:
     msr ttbr1_el1, x1
     dsb sy
 
-    ldr x2, =0x40000000      /* map 1G memory for kernel space */
-#ifdef RT_USING_LWP
-    ldr x3, =PV_OFFSET
+#ifdef RT_USING_SMART
+    ldr     x2, =__start
+    GET_PHY x3, __start
+    sub     x3, x3, x2
 #endif
-    bl rt_hw_mmu_setup_early
+    ldr x2, =0x40000000     /* map 1G memory for kernel space */
+    bl rt_hw_mem_setup_early
 
     ldr x30, =after_mmu_enable  /* set LR to after_mmu_enable function, it's a v_addr */
 
@@ -152,7 +152,7 @@ __start:
     ret
 
 after_mmu_enable:
-#ifdef RT_USING_LWP
+#ifdef RT_USING_SMART
     mrs x0, tcr_el1          /* disable ttbr0, only using kernel space */
     orr x0, x0, #(1 << 7)
     msr tcr_el1, x0
@@ -172,7 +172,7 @@ after_mmu_enable:
  *  secondary cpu
  */
 
-.globl _secondary_cpu_entry
+.global _secondary_cpu_entry
 _secondary_cpu_entry:
     bl      rt_hw_cpu_id_set
     adr     x1, __start
@@ -225,10 +225,6 @@ _secondary_cpu_entry:
     eret                            /* exception return. from EL2. continue from .L__in_el1 */
 
 .L__in_el1_cpux:
-    adr     x19, .L__in_el1_cpux
-    ldr     x8, =.L__in_el1_cpux
-    sub     x19, x19, x8            /* get PV_OFFSET            */
-
     mrs     x0, tpidr_el1
     /* each cpu init stack is 8k */
     sub     x1, x1, x0, lsl #13
@@ -244,7 +240,7 @@ _secondary_cpu_entry:
 
     bl mmu_tcr_init
 
-    adr x1, __start
+    adr x1, __start             /* GET & setup early page table */
     ldr x0, =~0x1fffff
     and x0, x1, x0
     add x1, x0, #0x1000
diff --git a/libcpu/arm/cortex-a/mmu.c b/libcpu/arm/cortex-a/mmu.c
index 150dd98b27..12d441d8e9 100644
--- a/libcpu/arm/cortex-a/mmu.c
+++ b/libcpu/arm/cortex-a/mmu.c
@@ -24,7 +24,6 @@
 #include "ioremap.h"
 #else
 #define KERNEL_VADDR_START 0
-#define PV_OFFSET 0
 #endif
 
 /* level1 page table, each entry for 1MB memory. */
@@ -59,6 +58,8 @@ void rt_hw_init_mmu_table(struct mem_desc *mdesc, rt_uint32_t size)
     /* set page table */
     for(; size > 0; size--)
     {
+        if (mdesc->paddr_start == (rt_uint32_t)ARCH_MAP_FAILED)
+            mdesc->paddr_start = mdesc->vaddr_start + PV_OFFSET;
         rt_hw_mmu_setmtt(mdesc->vaddr_start, mdesc->vaddr_end,
                 mdesc->paddr_start, mdesc->attr);
         mdesc++;
diff --git a/libcpu/arm/cortex-a/start_gcc.S b/libcpu/arm/cortex-a/start_gcc.S
index 94f96468fa..b1cddebe0b 100644
--- a/libcpu/arm/cortex-a/start_gcc.S
+++ b/libcpu/arm/cortex-a/start_gcc.S
@@ -102,7 +102,10 @@ continue:
     isb
 
 #ifdef RT_USING_SMART
-    ldr r5, =PV_OFFSET
+    /* load r5 with PV_OFFSET */
+    ldr r7, =_reset
+    adr r5, _reset
+    sub r5, r5, r7
 
     mov r7, #0x100000
     sub r7, #1
@@ -166,6 +169,9 @@ bss_loop:
     strlo r0,[r1],#4            /* clear 4 bytes                    */
     blo bss_loop                /* loop until done                  */
 
+    mov r0, r5
+    bl rt_kmem_pvoff_set
+
 #ifdef RT_USING_SMP
     mrc p15, 0, r1, c1, c0, 1
     mov r0, #(1<<6)
@@ -173,7 +179,10 @@ bss_loop:
     mcr p15, 0, r1, c1, c0, 1   /* enable smp */
 #endif
 
-    /* initialize the mmu table and enable mmu */
+    /**
+     * void rt_hw_init_mmu_table(struct mem_desc *mdesc, rt_uint32_t size)
+     * initialize the mmu table and enable mmu
+     */
     ldr r0, =platform_mem_desc
     ldr r1, =platform_mem_desc_size
     ldr r1, [r1]
@@ -595,16 +604,19 @@ vector_resv:
     bl      rt_hw_trap_resv
     b       .
 
-#ifdef RT_USING_SMP
 .global rt_hw_clz
 rt_hw_clz:
     clz r0, r0
     bx lr
 
+#ifdef RT_USING_SMP
+
 .global rt_secondary_cpu_entry
 rt_secondary_cpu_entry:
 #ifdef RT_USING_SMART
-    ldr     r5, =PV_OFFSET
+    ldr     r0, =_reset
+    adr     r5, _reset
+    sub     r5, r5, r0
 
     ldr     lr, =after_enable_mmu_n
     ldr     r0, =init_mtbl
diff --git a/libcpu/risc-v/t-head/c906/mmu.c b/libcpu/risc-v/t-head/c906/mmu.c
index dc3ba0cfdd..cc696ec6cb 100644
--- a/libcpu/risc-v/t-head/c906/mmu.c
+++ b/libcpu/risc-v/t-head/c906/mmu.c
@@ -30,7 +30,6 @@
 #endif
 
 #ifndef RT_USING_SMART
-#define PV_OFFSET 0
 #define USER_VADDR_START 0
 #endif
 
@@ -533,6 +532,9 @@ void rt_hw_mmu_setup(rt_aspace_t aspace, struct mem_desc *mdesc, int desc_nr)
                                                 mdesc->vaddr_start + 1,
                                     .prefer = (void *)mdesc->vaddr_start};
 
+        if (mdesc->paddr_start == (rt_size_t)ARCH_MAP_FAILED)
+            mdesc->paddr_start = mdesc->vaddr_start + PV_OFFSET;
+
         rt_aspace_map_phy_static(aspace, &mdesc->varea, &hint, attr,
                                  mdesc->paddr_start >> MM_PAGE_SHIFT, &err);
         mdesc++;
diff --git a/libcpu/risc-v/virt64/mmu.c b/libcpu/risc-v/virt64/mmu.c
index 6752ff306c..6d4cc6463e 100644
--- a/libcpu/risc-v/virt64/mmu.c
+++ b/libcpu/risc-v/virt64/mmu.c
@@ -25,12 +25,12 @@
 #include <tlb.h>
 
 #ifdef RT_USING_SMART
+#include <board.h>
 #include <ioremap.h>
 #include <lwp_user_mm.h>
 #endif
 
 #ifndef RT_USING_SMART
-#define PV_OFFSET 0
 #define USER_VADDR_START 0
 #endif
 
@@ -476,6 +476,9 @@ void rt_hw_mmu_setup(rt_aspace_t aspace, struct mem_desc *mdesc, int desc_nr)
                                                 mdesc->vaddr_start + 1,
                                     .prefer = (void *)mdesc->vaddr_start};
 
+        if (mdesc->paddr_start == (rt_size_t)ARCH_MAP_FAILED)
+            mdesc->paddr_start = mdesc->vaddr_start + PV_OFFSET;
+
         rt_aspace_map_phy_static(aspace, &mdesc->varea, &hint, attr,
                                  mdesc->paddr_start >> MM_PAGE_SHIFT, &err);
         mdesc++;