fix: corner case for x87 context restore on execve

[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 a076a976cbbe2fa9f495bc9cafcc6af932f7b328..3340f114c895ea05bc153aed0f52a9ac88589d2a 100644 (file)
--- a/lunaix-os/kernel/mm/vmm.c
+++ b/lunaix-os/kernel/mm/vmm.c
@@ -1,154 +1,198 @@
-#include <libc/string.h>
-#include <lunaix/mm/page.h>
+#include <hal/cpu.h>
+#include <klibc/string.h>

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

+#include <lunaix/spike.h>
+#include <lunaix/syslog.h>

 
 

-// TODO: Move these nasty inline asm stuff into hal
-//      These should be arch dependent
-ptd_t*
-get_pd()
-{
-    ptd_t* pd;
-#ifdef __ARCH_IA32
-    __asm__("movl %%cr3, %0\n"
-            "andl $0xfffff000, %0"
-            : "=r"(pd));
-#endif
-    return (ptd_t*)P2V(pd);
-}
-
-void
-set_pd(ptd_t* pd)
-{
-#ifdef __ARCH_IA32
-    __asm__("movl %0, %%eax\n"
-            "andl $0xfffff000, %%eax\n"
-            "movl %%eax, %%cr3\n"
-            :
-            : "r"(pd));
-#endif
-}
+LOG_MODULE("VMM")

 
 void
 vmm_init()
 {
 
 void
 vmm_init()
 {

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

 }
 
 }
 

-ptd_t*
+x86_page_table*

 vmm_init_pd()
 {
 vmm_init_pd()
 {

-    ptd_t* dir = pmm_alloc_page();
-    for (size_t i = 0; i < 1024; i++) {
-        dir[i] = 0;
+    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;

     }
 
     }
 

-    // 自己映射自己，方便我们在软件层面进行查表地址转换
-    dir[1023] = PDE(T_SELF_REF_PERM, dir);
+    // 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);

 
     return dir;
 }
 
 
     return dir;
 }
 

-void*
-vmm_map_page(void* va, void* pa, pt_attr dattr, pt_attr tattr)
+int
+vmm_set_mapping(ptr_t mnt, ptr_t va, ptr_t pa, pt_attr attr, int options)

 {
 {

-    // 显然，对空指针进行映射没有意义。
-    if (!pa || !va) {
-        return NULL;
-    }
+    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);
+
+    if (!l1pt->entry[l1_inx]) {
+        x86_page_table* new_l1pt_pa =
+          (x86_page_table*)pmm_alloc_page(KERNEL_PID, PP_FGPERSIST);

 
 

-    uintptr_t pd_offset = PD_INDEX(va);
-    uintptr_t pt_offset = PT_INDEX(va);
-    ptd_t* ptd = (ptd_t*)PTD_BASE_VADDR;
-
-    // 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）
-    ptd_t pde = ptd[pd_offset];
-    pt_t* pt = (pt_t*)PT_VADDR(pd_offset);
-    while (pde && pd_offset < 1024) {
-        if (pt_offset == 1024) {
-            pd_offset++;
-            pt_offset = 0;
-            pde = ptd[pd_offset];
-            pt = (pt_t*)PT_VADDR(pd_offset);
+        // 物理内存已满！
+        if (!new_l1pt_pa) {
+            return 0;

         }
         }

-        // 页表有空位，只需要开辟一个新的 PTE
-        if (pt && !pt[pt_offset]) {
-            pt[pt_offset] = PTE(tattr, pa);
-            return (void*)V_ADDR(pd_offset, pt_offset, PG_OFFSET(va));
+
+        // This must be writable
+        l1pt->entry[l1_inx] =
+          NEW_L1_ENTRY(attr | PG_WRITE | PG_PRESENT, new_l1pt_pa);
+
+        // make sure our new l2 table is visible to CPU
+        cpu_invplg((ptr_t)l2pt);
+
+        memset((void*)l2pt, 0, PG_SIZE);
+    } else {
+        x86_pte_t pte = l2pt->entry[l2_inx];
+        if (pte && (options & VMAP_IGNORE)) {
+            return 1;

         }
         }

-        pt_offset++;

     }
 
     }
 

-    // 页目录与所有页表已满！
-    if (pd_offset > 1024) {
-        return NULL;
+    if (mnt == VMS_SELF) {
+        cpu_invplg(va);
+    }
+
+    if ((options & VMAP_NOMAP)) {
+        return 1;

     }
 
     }
 

-    // 页目录有空位，需要开辟一个新的 PDE
-    uint8_t* new_pt_pa = pmm_alloc_page();
+    l2pt->entry[l2_inx] = NEW_L2_ENTRY(attr, pa);
+    return 1;
+}
+
+ptr_t
+vmm_del_mapping(ptr_t mnt, ptr_t va)
+{
+    assert(((ptr_t)va & 0xFFFU) == 0);

 
 

-    // 物理内存已满！
-    if (!new_pt_pa) {
-        return NULL;
+    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;

     }
 
     }
 

-    ptd[pd_offset] = PDE(dattr, new_pt_pa);
+    x86_page_table* l1pt = (x86_page_table*)(mnt | (1023 << 12));
+
+    x86_pte_t l1pte = l1pt->entry[l1_index];

 
 

-    memset((void*)PT_VADDR(pd_offset), 0, PM_PAGE_SIZE);
-    pt[pt_offset] = PTE(tattr, pa);
+    if (l1pte) {
+        x86_page_table* l2pt = (x86_page_table*)(mnt | (l1_index << 12));
+        x86_pte_t l2pte = l2pt->entry[l2_index];

 
 

-    return (void*)V_ADDR(pd_offset, pt_offset, PG_OFFSET(va));
+        cpu_invplg(va);
+        l2pt->entry[l2_index] = PTE_NULL;
+
+        return PG_ENTRY_ADDR(l2pte);
+    }
+
+    return 0;
+}
+
+int
+vmm_lookup(ptr_t va, v_mapping* mapping)
+{
+    return vmm_lookupat(VMS_SELF, va, mapping);

 }
 
 }
 

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

 {
 {

-    void* pp = pmm_alloc_page();
-    void* result = vmm_map_page(vpn, pp, dattr, tattr);
-    if (!result) {
-        pmm_free_page(pp);
+    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 result;
+
+    return 0;

 }
 
 }
 

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

 {
 {

-    uintptr_t pd_offset = PD_INDEX(vpn);
-    uintptr_t pt_offset = PT_INDEX(vpn);
-    ptd_t* self_pde = (ptd_t*)PTD_BASE_VADDR;
-
-    ptd_t pde = self_pde[pd_offset];
-
-    if (pde) {
-        pt_t* pt = (pt_t*)PT_VADDR(pd_offset);
-        uint32_t pte = pt[pt_offset];
-        if (IS_CACHED(pte) && pmm_free_page((void*)pte)) {
-// 刷新TLB
-#ifdef __ARCH_IA32
-            __asm__("invlpg (%0)" ::"r"((uintptr_t)vpn) : "memory");
-#endif
+    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_pte_t l1pte = l1pt->entry[l1_index];
+
+    if (l1pte) {
+        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);

         }
         }

-        pt[pt_offset] = 0;

     }
     }

+    return 0;

 }
 
 }
 

-void*
-vmm_v2p(void* va)
+ptr_t
+vmm_v2pat(ptr_t mnt, ptr_t va)

 {
 {

-    uintptr_t pd_offset = PD_INDEX(va);
-    uintptr_t pt_offset = PT_INDEX(va);
-    uintptr_t po = PG_OFFSET(va);
-    ptd_t* self_pde = (ptd_t*)PTD_BASE_VADDR;
-
-    ptd_t pde = self_pde[pd_offset];
-    if (pde) {
-        pt_t pte = ((pt_t*)PT_VADDR(pd_offset))[pt_offset];
-        if (pte) {
-            uintptr_t ppn = pte >> 12;
-            return (void*)P_ADDR(ppn, po);
+    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) {
+            return PG_ENTRY_ADDR(*l2pte) | ((ptr_t)va & 0xfff);

         }
     }
         }
     }

+    return 0;
+}

 
 

-    return NULL;
+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_invplg(mnt);
+    return mnt;
+}
+
+ptr_t
+vmm_unmount_pd(ptr_t mnt)
+{
+    x86_page_table* l1pt = (x86_page_table*)L1_BASE_VADDR;
+    l1pt->entry[(mnt >> 22)] = 0;
+    cpu_invplg(mnt);
+    return mnt;

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