refactor: use a more decent physical memory map

[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 54da656e695519ac61734665226df0be721be80e..4cfb1a68bc40d1a912f951b3ef2256dcff4dd4e9 100644 (file)
--- a/lunaix-os/kernel/mm/vmm.c
+++ b/lunaix-os/kernel/mm/vmm.c
@@ -1,22 +1,23 @@
-#include <hal/cpu.h>
-#include <libc/string.h>
-#include <lunaix/mm/page.h>
+#include <klibc/string.h>

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

+#include <lunaix/syslog.h>
+#include <sys/cpu.h>

 
 

-#include <stdbool.h>
+LOG_MODULE("VMM")

 
 void
 vmm_init()
 {
 
 void
 vmm_init()
 {

-    // TODO: something here?
+    // XXX: something here?

 }
 
 x86_page_table*
 vmm_init_pd()
 {
 }
 
 x86_page_table*
 vmm_init_pd()
 {

-    x86_page_table* dir = (x86_page_table*)pmm_alloc_page();
+    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;
     }
     for (size_t i = 0; i < PG_MAX_ENTRIES; i++) {
         dir->entry[i] = PTE_NULL;
     }


@@ -28,189 +29,174 @@ vmm_init_pd()
 }
 
 int
 }
 
 int

-__vmm_map_internal(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, pt_attr attr, int options)

 {
 {

-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
-    x86_page_table* l2pt = (x86_page_table*)L2_VADDR(l1_inx);
+    assert((ptr_t)va % PG_SIZE == 0);
+
+    ptr_t l1_inx = L1_INDEX(va);
+    ptr_t l2_inx = L2_INDEX(va);
+    x86_page_table* l1pt = (x86_page_table*)(mnt | (1023 << 12));
+    x86_page_table* l2pt = (x86_page_table*)(mnt | (l1_inx << 12));

 
     // See if attr make sense
     assert(attr <= 128);
 
 
     // See if attr make sense
     assert(attr <= 128);
 

-    if (!l1pt->entry[l1_inx]) {
-        x86_page_table* new_l1pt_pa = pmm_alloc_page();
+    x86_pte_t* l1pte = &l1pt->entry[l1_inx];
+    if (!*l1pte) {
+        x86_page_table* new_l1pt_pa =
+          (x86_page_table*)pmm_alloc_page(KERNEL_PID, PP_FGPERSIST);

 
         // 物理内存已满！
         if (!new_l1pt_pa) {
             return 0;
         }
 
 
         // 物理内存已满！
         if (!new_l1pt_pa) {
             return 0;
         }
 

-        l1pt->entry[l1_inx] = NEW_L1_ENTRY(attr, new_l1pt_pa);
-        memset((void*)L2_VADDR(l1_inx), 0, PG_SIZE);
-    }
-
-    if (!forced && l2pt->entry[l2_inx]) {
-        return 0;
-    }
-
-    l2pt->entry[l2_inx] = NEW_L2_ENTRY(attr, pa);
-
-    return 1;
-}
-
-void*
-vmm_map_page(void* va, void* pa, pt_attr tattr)
-{
-    // 显然，对空指针进行映射没有意义。
-    if (!pa || !va) {
-        return NULL;
-    }
-
-    assert(((uintptr_t)va & 0xFFFU) == 0) assert(((uintptr_t)pa & 0xFFFU) == 0);
+        // This must be writable
+        *l1pte = NEW_L1_ENTRY(attr | PG_WRITE | PG_PRESENT, new_l1pt_pa);

 
 

-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
+        // make sure our new l2 table is visible to CPU
+        cpu_flush_page((ptr_t)l2pt);

 
 

-    // 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);
+        memset((void*)l2pt, 0, PG_SIZE);
+    } else {
+        if ((attr & PG_ALLOW_USER) && !(*l1pte & PG_ALLOW_USER)) {
+            *l1pte |= PG_ALLOW_USER;

         }
         }

-        // 页表有空位，只需要开辟一个新的 PTE (Level 2)
-        if (l2pt && !l2pt->entry[l2_index]) {
-            l2pt->entry[l2_index] = NEW_L2_ENTRY(tattr, pa);
-            return (void*)V_ADDR(l1_index, l2_index, PG_OFFSET(va));
+
+        x86_pte_t pte = l2pt->entry[l2_inx];
+        if (pte && (options & VMAP_IGNORE)) {
+            return 1;

         }
         }

-        l2_index++;

     }
 
     }
 

-    // 页目录与所有页表已满！
-    if (l1_index > PG_MAX_ENTRIES) {
-        return NULL;
+    if (mnt == VMS_SELF) {
+        cpu_flush_page(va);

     }
 
     }
 

-    if (!__vmm_map_internal(l1_index, l2_index, (uintptr_t)pa, tattr, false)) {
-        return NULL;
+    if ((options & VMAP_NOMAP)) {
+        return 1;

     }
 
     }
 

-    return (void*)V_ADDR(l1_index, l2_index, PG_OFFSET(va));
+    l2pt->entry[l2_inx] = NEW_L2_ENTRY(attr, pa);
+    return 1;

 }
 
 }
 

-void*
-vmm_fmap_page(void* va, void* pa, pt_attr tattr)
+ptr_t
+vmm_del_mapping(ptr_t mnt, ptr_t va)

 {
 {

-    if (!pa || !va) {
-        return NULL;
+    assert(((ptr_t)va & 0xFFFU) == 0);
+
+    u32_t l1_index = L1_INDEX(va);
+    u32_t l2_index = L2_INDEX(va);
+
+    // prevent unmap of recursive mapping region
+    if (l1_index == 1023) {
+        return 0;

     }
 
     }
 

-    assert(((uintptr_t)va & 0xFFFU) == 0) assert(((uintptr_t)pa & 0xFFFU) == 0);
+    x86_page_table* l1pt = (x86_page_table*)(mnt | (1023 << 12));

 
 

-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
+    x86_pte_t l1pte = l1pt->entry[l1_index];

 
 

-    if (!__vmm_map_internal(l1_index, l2_index, (uintptr_t)pa, tattr, true)) {
-        return NULL;
-    }
+    if (l1pte) {
+        x86_page_table* l2pt = (x86_page_table*)(mnt | (l1_index << 12));
+        x86_pte_t l2pte = l2pt->entry[l2_index];
+
+        cpu_flush_page(va);
+        l2pt->entry[l2_index] = PTE_NULL;

 
 

-    cpu_invplg(va);
+        return PG_ENTRY_ADDR(l2pte);
+    }

 
 

-    return (void*)V_ADDR(l1_index, l2_index, PG_OFFSET(va));
+    return 0;

 }
 
 }
 

-void*
-vmm_alloc_page(void* vpn, pt_attr tattr)
+int
+vmm_lookup(ptr_t va, v_mapping* mapping)

 {
 {

-    void* pp = pmm_alloc_page();
-    void* result = vmm_map_page(vpn, pp, tattr);
-    if (!result) {
-        pmm_free_page(pp);
-    }
-    return result;
+    return vmm_lookupat(VMS_SELF, va, mapping);

 }
 
 int
 }
 
 int

-vmm_alloc_pages(void* va, size_t sz, pt_attr tattr)
+vmm_lookupat(ptr_t mnt, ptr_t va, v_mapping* mapping)

 {
 {

-    assert((uintptr_t)va % PG_SIZE == 0) assert(sz % PG_SIZE == 0);
-
-    void* va_ = va;
-    for (size_t i = 0; i < (sz >> PG_SIZE_BITS); i++, va_ += PG_SIZE) {
-        void* pp = pmm_alloc_page();
-        uint32_t l1_index = L1_INDEX(va_);
-        uint32_t l2_index = L2_INDEX(va_);
-        if (!pp || !__vmm_map_internal(
-                     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(va_);
-            }
-
-            return false;
+    u32_t l1_index = L1_INDEX(va);
+    u32_t l2_index = L2_INDEX(va);
+
+    x86_page_table* l1pt = (x86_page_table*)(mnt | 1023 << 12);
+    x86_pte_t l1pte = l1pt->entry[l1_index];
+
+    if (l1pte) {
+        x86_pte_t* l2pte =
+          &((x86_page_table*)(mnt | (l1_index << 12)))->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;
+            mapping->va = va;
+            return 1;

         }
     }
 
         }
     }
 

-    return true;
+    return 0;

 }
 
 }
 

-void
-vmm_unmap_page(void* va)
+ptr_t
+vmm_v2p(ptr_t va)

 {
 {

-    assert(((uintptr_t)va & 0xFFFU) == 0);
+    u32_t l1_index = L1_INDEX(va);
+    u32_t l2_index = L2_INDEX(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;
     x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;

-

     x86_pte_t l1pte = l1pt->entry[l1_index];
 
     if (l1pte) {
     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)) {
-            pmm_free_page((void*)l2pte);
+        x86_pte_t* l2pte =
+          &((x86_page_table*)L2_VADDR(l1_index))->entry[l2_index];
+
+        if (l2pte) {
+            return PG_ENTRY_ADDR(*l2pte) | ((ptr_t)va & 0xfff);

         }
         }

-        cpu_invplg(va);
-        l2pt->entry[l2_index] = PTE_NULL;

     }
     }

+    return 0;

 }
 
 }
 

-v_mapping
-vmm_lookup(void* va)
+ptr_t
+vmm_v2pat(ptr_t mnt, ptr_t va)

 {
 {

-    assert(((uintptr_t)va & 0xFFFU) == 0);
-
-    uint32_t l1_index = L1_INDEX(va);
-    uint32_t l2_index = L2_INDEX(va);
+    u32_t l1_index = L1_INDEX(va);
+    u32_t l2_index = L2_INDEX(va);

 
 

-    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
+    x86_page_table* l1pt = (x86_page_table*)(mnt | 1023 << 12);

     x86_pte_t l1pte = l1pt->entry[l1_index];
 
     x86_pte_t l1pte = l1pt->entry[l1_index];
 

-    v_mapping mapping = { .flags = 0, .pa = 0, .pn = 0 };

     if (l1pte) {
     if (l1pte) {

-        x86_pte_t l2pte =
-          ((x86_page_table*)L2_VADDR(l1_index))->entry[l2_index];
+        x86_pte_t* l2pte =
+          &((x86_page_table*)(mnt | (l1_index << 12)))->entry[l2_index];
+

         if (l2pte) {
         if (l2pte) {

-            mapping.flags = PG_ENTRY_FLAGS(l2pte);
-            mapping.pa = PG_ENTRY_ADDR(l2pte);
-            mapping.pn = mapping.pa >> PG_SIZE_BITS;
+            return PG_ENTRY_ADDR(*l2pte) | ((ptr_t)va & 0xfff);

         }
     }
         }
     }

+    return 0;
+}

 
 

-    return mapping;
+ptr_t
+vmm_mount_pd(ptr_t mnt, ptr_t pde)
+{
+    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
+    l1pt->entry[(mnt >> 22)] = NEW_L1_ENTRY(T_SELF_REF_PERM, pde);
+    cpu_flush_page(mnt);
+    return mnt;

 }
 
 }
 

-void*
-vmm_v2p(void* va)
+ptr_t
+vmm_unmount_pd(ptr_t mnt)

 {
 {

-    return (void*)vmm_lookup(va).pa;
+    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
+    l1pt->entry[(mnt >> 22)] = 0;
+    cpu_flush_page(mnt);
+    return mnt;

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