A Total Overhaul on the Lunaix's Virtual Memory Model (#26)

[lunaix-os.git] / lunaix-os / kernel / mm / vmm.c

diff --git a/lunaix-os/kernel/mm/vmm.c b/lunaix-os/kernel/mm/vmm.c
index dcd2f2771acbf64af1dd09f0e3ed09adb20e39b2..6b497b8f60b0bb6c4eaea6122d6a18550d6fbe1b 100644 (file)
--- a/lunaix-os/kernel/mm/vmm.c
+++ b/lunaix-os/kernel/mm/vmm.c
@@ -1,9 +1,13 @@
-#include <hal/cpu.h>

 #include <klibc/string.h>
 #include <klibc/string.h>

+#include <lunaix/mm/pmm.h>

 #include <lunaix/mm/vmm.h>
 #include <lunaix/spike.h>
 #include <lunaix/mm/vmm.h>
 #include <lunaix/spike.h>

+#include <lunaix/syslog.h>

 
 

-#include <stdbool.h>
+#include <sys/cpu.h>
+#include <sys/mm/mm_defs.h>
+
+LOG_MODULE("VM")

 
 void
 vmm_init()
 
 void
 vmm_init()


@@ -11,261 +15,138 @@ vmm_init()
     // XXX: something here?
 }
 
     // XXX: something here?
 }
 

-x86_page_table*
-vmm_init_pd()
+pte_t 
+vmm_alloc_page(pte_t* ptep, pte_t pte)

 {
 {

-    x86_page_table* dir = (x86_page_table*)pmm_alloc_page(KERNEL_PID, PP_FGPERSIST);
-    for (size_t i = 0; i < PG_MAX_ENTRIES; i++) {
-        dir->entry[i] = PTE_NULL;
+    ptr_t pa = pmm_alloc_page(PP_FGPERSIST);
+    if (!pa) {
+        return null_pte;

     }
 
     }
 

-    // 递归映射，方便我们在软件层面进行查表地址转换
-    dir->entry[PG_MAX_ENTRIES - 1] = NEW_L1_ENTRY(T_SELF_REF_PERM, dir);
+    pte = pte_setpaddr(pte, pa);
+    pte = pte_mkloaded(pte);
+    set_pte(ptep, pte);
+
+    mount_page(PG_MOUNT_1, pa);
+    memset((void*)PG_MOUNT_1, 0, LFT_SIZE);
+    unmount_page(PG_MOUNT_1);
+
+    cpu_flush_page((ptr_t)ptep);

 
 

-    return dir;
+    return pte;

 }
 
 int
 }
 
 int

-__vmm_map_internal(pid_t pid, 
-                   uint32_t l1_inx,
-                   uint32_t l2_inx,
-                   uintptr_t pa,
-                   pt_attr attr,
-                   int forced)
+vmm_set_mapping(ptr_t mnt, ptr_t va, ptr_t pa, pte_attr_t prot)

 {
 {

-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
-    x86_page_table* l2pt = (x86_page_table*)L2_VADDR(l1_inx);
-
-    // See if attr make sense
-    assert(attr <= 128);
-
-    if (!l1pt->entry[l1_inx]) {
-        x86_page_table* new_l1pt_pa = pmm_alloc_page(pid, PP_FGPERSIST);
-
-        // 物理内存已满！
-        if (!new_l1pt_pa) {
-            return 0;
-        }
-
-        // This must be writable
-        l1pt->entry[l1_inx] = NEW_L1_ENTRY(attr | PG_WRITE, new_l1pt_pa);
-        memset((void*)L2_VADDR(l1_inx), 0, PG_SIZE);
-    }
-
-    x86_pte_t l2pte = l2pt->entry[l2_inx];
-    if (l2pte) {
-        if (!forced) {
-            return 0;
-        }
-    }
+    assert(!va_offset(va));

 
 

-    if ((HAS_FLAGS(attr, PG_PRESENT))) {
-        // add one on reference count, regardless of existence.
-        pmm_ref_page(pid, pa);
-    }
+    pte_t* ptep = mkptep_va(mnt, va);
+    pte_t  pte  = mkpte(pa, prot);

 
 

-    l2pt->entry[l2_inx] = NEW_L2_ENTRY(attr, pa);
+    set_pte(ptep, pte);

 
     return 1;
 }
 
 
     return 1;
 }
 

-void*
-vmm_map_page(pid_t pid, void* va, void* pa, pt_attr tattr)
+ptr_t
+vmm_del_mapping(ptr_t mnt, ptr_t va)

 {
 {

-    // 显然，对空指针进行映射没有意义。
-    if (!pa || !va) {
-        return NULL;
-    }
+    assert(!va_offset(va));

 
 

-    assert(((uintptr_t)va & 0xFFFU) == 0) assert(((uintptr_t)pa & 0xFFFU) == 0);
+    pte_t* ptep = mkptep_va(mnt, va);

 
 

-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
+    pte_t old = *ptep;

 
 

-    // 在页表与页目录中找到一个可用的空位进行映射（位于va或其附近）
-    x86_pte_t l1pte = l1pt->entry[l1_index];
-    x86_page_table* l2pt = (x86_page_table*)L2_VADDR(l1_index);
-    while (l1pte && l1_index < PG_MAX_ENTRIES) {
-        if (l2_index == PG_MAX_ENTRIES) {
-            l1_index++;
-            l2_index = 0;
-            l1pte = l1pt->entry[l1_index];
-            l2pt = (x86_page_table*)L2_VADDR(l1_index);
-        }
-        // 页表有空位，只需要开辟一个新的 PTE (Level 2)
-        if (__vmm_map_internal(pid, l1_index, l2_index, pa, tattr, false)) {
-            return (void*)V_ADDR(l1_index, l2_index, PG_OFFSET(va));
-        }
-        l2_index++;
-    }
+    set_pte(ptep, null_pte);

 
 

-    // 页目录与所有页表已满！
-    if (l1_index > PG_MAX_ENTRIES) {
-        return NULL;
-    }
-
-    if (!__vmm_map_internal(pid, l1_index, l2_index, (uintptr_t)pa, tattr, false)) {
-        return NULL;
-    }
-
-    return (void*)V_ADDR(l1_index, l2_index, PG_OFFSET(va));
+    return pte_paddr(old);

 }
 
 }
 

-void*
-vmm_fmap_page(pid_t pid, void* va, void* pa, pt_attr tattr)
+pte_t
+vmm_tryptep(pte_t* ptep, size_t lvl_size)

 {
 {

-    if (!pa || !va) {
-        return NULL;
-    }
-
-    assert(((uintptr_t)va & 0xFFFU) == 0) assert(((uintptr_t)pa & 0xFFFU) == 0);
-
-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
-
-    if (!__vmm_map_internal(pid, l1_index, l2_index, (uintptr_t)pa, tattr, true)) {
-        return NULL;
-    }
-
-    cpu_invplg(va);
-
-    return va;
+    ptr_t va = ptep_va(ptep, lvl_size);
+    pte_t* _ptep = mkl0tep(ptep);
+    pte_t pte;
+
+    if (pte_isnull(pte = *_ptep) || _ptep == ptep) 
+        return pte;
+
+#if LnT_ENABLED(1)
+    _ptep = getl1tep(_ptep, va);
+    if (_ptep == ptep || pte_isnull(pte = *_ptep)) 
+        return pte;
+#endif
+#if LnT_ENABLED(2)
+    _ptep = getl2tep(_ptep, va);
+    if (_ptep == ptep || pte_isnull(pte = *_ptep)) 
+        return pte;
+#endif
+#if LnT_ENABLED(3)
+    _ptep = getl3tep(_ptep, va);
+    if (_ptep == ptep || pte_isnull(pte = *_ptep)) 
+        return pte;
+#endif
+    _ptep = getlftep(_ptep, va);
+    return *_ptep;

 }
 
 }
 

-void*
-vmm_alloc_page(pid_t pid, void* vpn, void** pa, pt_attr tattr, pp_attr_t pattr)
+ptr_t
+vmm_v2pat(ptr_t mnt, ptr_t va)

 {
 {

-    void* pp = pmm_alloc_page(pid, pattr);
-    void* result = vmm_map_page(pid, vpn, pp, tattr);
-    if (!result) {
-        pmm_free_page(pp, pid);
-    }
-    pa ? (*pa = pp) : 0;
-    return result;
+    ptr_t  va_off = va_offset(va);
+    pte_t* ptep   = mkptep_va(mnt, va);
+
+    return pte_paddr(pte_at(ptep)) + va_off;

 }
 
 }
 

-int
-vmm_alloc_pages(pid_t pid, void* va, size_t sz, pt_attr tattr, pp_attr_t pattr)
+ptr_t
+vms_mount(ptr_t mnt, ptr_t vms_root)

 {
 {

-    assert((uintptr_t)va % PG_SIZE == 0) assert(sz % PG_SIZE == 0);
+    assert(vms_root);

 
 

-    void* va_ = va;
-    for (size_t i = 0; i < (sz >> PG_SIZE_BITS); i++, va_ += PG_SIZE) {
-        void* pp = pmm_alloc_page(pid, pattr);
-        uint32_t l1_index = L1_INDEX(va_);
-        uint32_t l2_index = L2_INDEX(va_);
-        if (!pp || !__vmm_map_internal(
-                     pid,
-                     l1_index, l2_index, (uintptr_t)pp, tattr, false)) {
-            // if one failed, release previous allocated pages.
-            va_ = va;
-            for (size_t j = 0; j < i; j++, va_ += PG_SIZE) {
-                vmm_unmap_page(pid, va_);
-            }
-
-            return false;
-        }
-    }
-
-    return true;
+    pte_t* ptep = mkl0tep_va(VMS_SELF, mnt);
+    set_pte(ptep, mkpte(vms_root, KERNEL_DATA));
+    cpu_flush_page(mnt);
+    return mnt;

 }
 
 }
 

-int
-vmm_set_mapping(pid_t pid, void* va, void* pa, pt_attr attr) {
-    assert(((uintptr_t)va & 0xFFFU) == 0);
-
-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
-
-    // prevent map of recursive mapping region
-    if (l1_index == 1023) {
-        return 0;
-    }
-    
-    __vmm_map_internal(pid, l1_index, l2_index, (uintptr_t)pa, attr, false);
-    return 1;
+ptr_t
+vms_unmount(ptr_t mnt)
+{
+    pte_t* ptep = mkl0tep_va(VMS_SELF, mnt);
+    set_pte(ptep, null_pte);
+    cpu_flush_page(mnt);
+    return mnt;

 }
 
 }
 

-void
-__vmm_unmap_internal(pid_t pid, void* va, int free_ppage) {
-    assert(((uintptr_t)va & 0xFFFU) == 0);

 
 

-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
-
-    // prevent unmap of recursive mapping region
-    if (l1_index == 1023) {
+void
+ptep_alloc_hierarchy(pte_t* ptep, ptr_t va, pte_attr_t prot)
+{
+    pte_t* _ptep;
+    
+    _ptep = mkl0tep(ptep);
+    if (_ptep == ptep) {

         return;
     }
 
         return;
     }
 

-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
-
-    x86_pte_t l1pte = l1pt->entry[l1_index];
-
-    if (l1pte) {
-        x86_page_table* l2pt = (x86_page_table*)L2_VADDR(l1_index);
-        x86_pte_t l2pte = l2pt->entry[l2_index];
-        if (IS_CACHED(l2pte) && free_ppage) {
-            pmm_free_page(pid, (void*)l2pte);
-        }
-        cpu_invplg(va);
-        l2pt->entry[l2_index] = PTE_NULL;
+    _ptep = mkl1t(_ptep, va, prot);
+    if (_ptep == ptep) {
+        return;

     }
     }

-}
-
-void
-vmm_unset_mapping(void* va) {
-    __vmm_unmap_internal(0, va, false);
-}
-
-void
-vmm_unmap_page(pid_t pid, void* va)
-{
-    __vmm_unmap_internal(pid, va, true);
-}
-
-v_mapping
-vmm_lookup(void* va)
-{
-    assert(((uintptr_t)va & 0xFFFU) == 0);

 
 

-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
-
-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
-    x86_pte_t l1pte = l1pt->entry[l1_index];
-
-    v_mapping mapping = { .flags = 0, .pa = 0, .pn = 0 };
-    if (l1pte) {
-        x86_pte_t* l2pte =
-          &((x86_page_table*)L2_VADDR(l1_index))->entry[l2_index];
-        if (l2pte) {
-            mapping.flags = PG_ENTRY_FLAGS(*l2pte);
-            mapping.pa = PG_ENTRY_ADDR(*l2pte);
-            mapping.pn = mapping.pa >> PG_SIZE_BITS;
-            mapping.pte = l2pte;
-        }
+    _ptep = mkl2t(_ptep, va, prot);
+    if (_ptep == ptep) {
+        return;

     }
 
     }
 

-    return mapping;
-}
-
-void*
-vmm_v2p(void* va)
-{
-    return (void*)vmm_lookup(va).pa;
-}
-
-void*
-vmm_mount_pd(uintptr_t mnt, void* pde) {
-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
-    l1pt->entry[(mnt >> 22)] = NEW_L1_ENTRY(PG_PREM_RW, pde);
-    return mnt;
-}
+    _ptep = mkl3t(_ptep, va, prot);
+    if (_ptep == ptep) {
+        return;
+    }

 
 

-void*
-vmm_unmount_pd(uintptr_t mnt) {
-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
-    l1pt->entry[(mnt >> 22)] = 0;
+    _ptep = mklft(_ptep, va, prot);
+    assert(_ptep == ptep);

 }
\ No newline at end of file
 }
\ No newline at end of file