# seL4 map\_kernel\_window on riscv platform

**URL:** <https://sel4.discourse.group/t/sel4-map-kernel-window-on-riscv-platform/674>\
**Category:** New to seL4\
**Created:** [March 8, 2023, 1:48am UTC](https://sel4.discourse.group/t/sel4-map-kernel-window-on-riscv-platform/674 "2023-03-08T01:48:24Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![CtrlZ](https://avatars.discourse-cdn.com/v4/letter/c/bcef8e/32.png) [@CtrlZ](https://sel4.discourse.group/u/CtrlZ)\
**Post date:** [March 8, 2023, 1:48am UTC](https://sel4.discourse.group/t/sel4-map-kernel-window-on-riscv-platform/674/1 "2023-03-08T01:48:24Z")

</div>

I’m reading the sel4test source code, and the there is a function called map\_kernel\_window when system is booting. After called map\_kernel\_window function, the system will enable the pagetable mode.

When I read the map\_kernel\_window function and found that just two-level page tables are populated, but using the SV39, so I think the page table is not fully populated, and when enable the pagetable mode, I thought the kernel execution would go wrong, but it didn’t. The map\_kernel\_window function was followed:

```c
BOOT_CODE VISIBLE void map_kernel_window(void)
{
    /* mapping of KERNEL_ELF_BASE (virtual address) to kernel's
     * KERNEL_ELF_PHYS_BASE */
    assert(CONFIG_PT_LEVELS > 1 && CONFIG_PT_LEVELS <= 4);

    /* kernel window starts at PPTR_BASE */
    word_t pptr = PPTR_BASE;

    /* first we map in memory from PADDR_BASE */
    word_t paddr = PADDR_BASE;

    while (pptr < PPTR_TOP) {
        assert(IS_ALIGNED(pptr, RISCV_GET_LVL_PGSIZE_BITS(0)));
        assert(IS_ALIGNED(paddr, RISCV_GET_LVL_PGSIZE_BITS(0)));
        // printf("RISCV_GET_PT_INDEX(pptr, 0): %llx\n", (long long unsigned int)(RISCV_GET_PT_INDEX(pptr, 0)));
        kernel_root_pageTable[RISCV_GET_PT_INDEX(pptr, 0)] = pte_next(paddr, true);

        pptr += RISCV_GET_LVL_PGSIZE(0);
        paddr += RISCV_GET_LVL_PGSIZE(0);
    }
    /* now we should be mapping the 1GiB kernel base */
    assert(pptr == PPTR_TOP);

    
    pptr = ROUND_DOWN(KERNEL_ELF_BASE, RISCV_GET_LVL_PGSIZE_BITS(0));
    paddr = ROUND_DOWN(KERNEL_ELF_PADDR_BASE, RISCV_GET_LVL_PGSIZE_BITS(0));

#if __riscv_xlen == 32
    kernel_root_pageTable[RISCV_GET_PT_INDEX(pptr, 0)] = pte_next(paddr, true);
    pptr += RISCV_GET_LVL_PGSIZE(0);
    paddr += RISCV_GET_LVL_PGSIZE(0);
#ifdef CONFIG_KERNEL_LOG_BUFFER
    kernel_root_pageTable[RISCV_GET_PT_INDEX(KS_LOG_PPTR, 0)] =
        pte_next(kpptr_to_paddr(kernel_image_level2_log_buffer_pt), false);
#endif
#else
    word_t index = 0;
    /* The kernel image is mapped twice, locating the two indexes in the
     * root page table, pointing them to the same second level page table.
     */
    kernel_root_pageTable[RISCV_GET_PT_INDEX(KERNEL_ELF_PADDR_BASE + PPTR_BASE_OFFSET, 0)] =
        pte_next(kpptr_to_paddr(kernel_image_level2_pt), false);
    kernel_root_pageTable[RISCV_GET_PT_INDEX(pptr, 0)] =
        pte_next(kpptr_to_paddr(kernel_image_level2_pt), false);
    while (pptr < PPTR_TOP + RISCV_GET_LVL_PGSIZE(0)) {
        kernel_image_level2_pt[index] = pte_next(paddr, true);
        index++;
        pptr += RISCV_GET_LVL_PGSIZE(1);
        paddr += RISCV_GET_LVL_PGSIZE(1);
    }

    /* Map kernel device page table */
    kernel_root_pageTable[RISCV_GET_PT_INDEX(KDEV_BASE, 0)] =
        pte_next(kpptr_to_paddr(kernel_image_level2_dev_pt), false);
#endif

    /* There should be 1GiB free where we put device mapping */
    assert(pptr == UINTPTR_MAX - RISCV_GET_LVL_PGSIZE(0) + 1);
    map_kernel_devices();
}

```

Can someone tell me what’s going on?