feat: support ARCH_REMAP_KERNEL on libcpu/c906 (#9123) (#9171)

feat: [libcpu/c906] support ARCH_REMAP_KERNEL

This change was necessary to enable the remapping of the kernel image to
a high virtual address region on the c906 platform.

Changes:
- Introduced new configuration options `ARCH_REMAP_KERNEL`, and
  `ARCH_USING_ASID` under the `ARCH_RISCV64` section.
- Updated MMU initialization and switching functions to incorporate
  remapping handling.
- Modified page table setup for proper memory attribute settings.
- Added support for early memory setup, kernel remapping
- Added conditional compilation for ASID support in the `rt_aspace` struct,
  since this is not enable currently for most architecture.

Signed-off-by: Shell <smokewood@qq.com>
Co-authored-by: Shell <smokewood@qq.com>

components/mm/mm_aspace.h

@@ -49,7 +49,11 @@ typedef struct rt_aspace
     struct rt_mutex bst_lock;
 
     struct rt_mem_obj *private_object;
+
+#ifdef ARCH_USING_ASID
     rt_uint64_t asid;
+#endif /* ARCH_USING_ASID */
+
 } *rt_aspace_t;
 
 typedef struct rt_varea

libcpu/Kconfig

@@ -262,13 +262,17 @@ config ARCH_RISCV64
     select ARCH_CPU_64BIT
     bool
 
-if ARCH_RISCV64
+config ARCH_REMAP_KERNEL
-    config ARCH_REMAP_KERNEL
+    bool
-        bool
+    depends on RT_USING_SMART
-        depends on RT_USING_SMART
+    help
-        help
+        Remapping kernel image to high virtual address region
-            Remapping kernel image to high virtual address region
+
-endif
+config ARCH_USING_ASID
+    bool
+    depends on RT_USING_SMART
+    help
+        Using ASID support from architecture
 
 config ARCH_IA32
     bool

libcpu/risc-v/t-head/c906/mmu.c

@@ -43,6 +43,7 @@ static void *current_mmu_table = RT_NULL;
 volatile __attribute__((aligned(4 * 1024)))
 rt_ubase_t MMUTable[__SIZE(VPN2_BIT)];
 
+#ifdef ARCH_USING_ASID
 static rt_uint8_t ASID_BITS = 0;
 static rt_uint32_t next_asid;
 static rt_uint64_t global_asid_generation;
@@ -109,6 +110,24 @@ void rt_hw_aspace_switch(rt_aspace_t aspace)
     asm volatile("sfence.vma x0,%0"::"r"(asid):"memory");
 }
 
+#define ASID_INIT() _asid_init()
+
+#else /* ARCH_USING_ASID */
+
+#define ASID_INIT()
+
+void rt_hw_aspace_switch(rt_aspace_t aspace)
+{
+    uintptr_t page_table = (uintptr_t)rt_kmem_v2p(aspace->page_table);
+    current_mmu_table = aspace->page_table;
+
+    write_csr(satp, (((size_t)SATP_MODE) << SATP_MODE_OFFSET) |
+                        ((rt_ubase_t)page_table >> PAGE_OFFSET_BIT));
+    rt_hw_tlb_invalidate_all_local();
+}
+
+#endif /* ARCH_USING_ASID */
+
 void *rt_hw_mmu_tbl_get()
 {
     return current_mmu_table;
@@ -552,7 +571,7 @@ void rt_hw_mmu_setup(rt_aspace_t aspace, struct mem_desc *mdesc, int desc_nr)
         mdesc++;
     }
 
-    _asid_init();
+    ASID_INIT();
 
     rt_hw_aspace_switch(&rt_kernel_space);
     rt_page_cleanup();
@@ -601,13 +620,15 @@ void rt_hw_mem_setup_early(void)
             LOG_E("%s: not aligned virtual address. pv_offset %p", __func__, pv_off);
             RT_ASSERT(0);
         }
+
         /**
          * identical mapping,
          * PC are still at lower region before relocating to high memory
          */
         for (size_t i = 0; i < __SIZE(PPN0_BIT); i++)
         {
-            early_pgtbl[i] = COMBINEPTE(ps, PAGE_ATTR_RWX | PTE_G | PTE_V);
+            early_pgtbl[i] = COMBINEPTE(ps, PAGE_ATTR_RWX | PTE_G | PTE_V | PTE_CACHE |
+                                        PTE_SHARE | PTE_BUF | PTE_A | PTE_D);
             ps += L1_PAGE_SIZE;
         }
 
@@ -621,7 +642,8 @@ void rt_hw_mem_setup_early(void)
         rt_size_t ve_idx = GET_L1(vs + 0x80000000);
         for (size_t i = vs_idx; i < ve_idx; i++)
         {
-            early_pgtbl[i] = COMBINEPTE(ps, PAGE_ATTR_RWX | PTE_G | PTE_V);
+            early_pgtbl[i] = COMBINEPTE(ps, PAGE_ATTR_RWX | PTE_G | PTE_V | PTE_CACHE |
+                                        PTE_SHARE | PTE_BUF | PTE_A | PTE_D);
             ps += L1_PAGE_SIZE;
         }
 

libcpu/risc-v/t-head/c906/startup_gcc.S

@@ -8,13 +8,16 @@
  * 2018/10/01     Bernard      The first version
  * 2018/12/27     Jesven       Add SMP support
  * 2020/6/12      Xim          Port to QEMU and remove SMP support
+ * 2024-06-30     Shell        Support of kernel remapping
  */
 
 #include <encoding.h>
 #include <cpuport.h>
 
-boot_hartid: .int
+    .data
-    .global         boot_hartid
+    .global boot_hartid    /* global varible rt_boot_hartid in .data section */
+boot_hartid:
+    .word 0xdeadbeef
 
     .global         _start
     .section ".start", "ax"
@@ -72,8 +75,6 @@ _start:
     li x31,0
 
     /* set to disable FPU */
-    li t0, SSTATUS_FS
-    csrc sstatus, t0
     li t0, SSTATUS_SUM
     csrs sstatus, t0
 
@@ -86,8 +87,45 @@ _start:
     la   sp, __stack_start__
     li   t0, __STACKSIZE__
     add  sp, sp, t0
-    csrw sscratch, sp
 
+    /**
+     * sscratch is always zero on kernel mode
+     */
+    csrw sscratch, zero
     call init_bss
-    call sbi_init
+#ifdef ARCH_MM_MMU
-    j primary_cpu_entry
+    call    rt_hw_mem_setup_early
+    call    rt_kmem_pvoff
+    /* a0 := pvoff  */
+    beq     a0, zero, 1f
+
+    /* relocate pc */
+    la      x1, _after_pc_relocation
+    sub     x1, x1, a0
+    ret
+_after_pc_relocation:
+    /* relocate gp */
+    sub     gp, gp, a0
+
+    /* relocate context: sp */
+    la      sp, __stack_start__
+    li      t0, __STACKSIZE__
+    add     sp, sp, t0
+
+    /* reset s0-fp */
+    mv      s0, zero
+
+    /* relocate stvec */
+    la      t0, trap_entry
+    csrw    stvec, t0
+1:
+#endif
+    call    sbi_init
+    call    primary_cpu_entry
+
+_never_return_here:
+    j       .
+
+.global _start_link_addr
+_start_link_addr:
+    .dword __text_start