From: Peter Dinda <pdinda@northwestern.edu>
Date: Thu, 12 Mar 2015 02:00:40 +0000 (-0500)
Subject: Add proper cleanup for a nested VMX guest This also fixes a bug that was preventing... 
X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?p=palacios.git;a=commitdiff_plain;h=d389e7022c2afa3fb72bc87ab7b324fd9183e5b1

Add proper cleanup for a nested VMX guest This also fixes a bug that was preventing such a guest from being freed
---

diff --git a/palacios/src/palacios/vmm_direct_paging.c b/palacios/src/palacios/vmm_direct_paging.c
index 2884923..0458155 100644
--- a/palacios/src/palacios/vmm_direct_paging.c
+++ b/palacios/src/palacios/vmm_direct_paging.c
@@ -182,6 +182,10 @@ static addr_t create_generic_pt_page(struct guest_info *core) {
 
 
 int v3_init_passthrough_pts(struct guest_info * info) {
+    if (info->shdw_pg_mode == NESTED_PAGING && is_vmx_nested()) { 
+        // skip - ept_init will do this allocation
+        return 0;
+    }
     info->direct_map_pt = (addr_t)V3_PAddr((void *)create_generic_pt_page(info));
     return 0;
 }
@@ -190,6 +194,18 @@ int v3_init_passthrough_pts(struct guest_info * info) {
 int v3_free_passthrough_pts(struct guest_info * core) {
     v3_cpu_mode_t mode = v3_get_vm_cpu_mode(core);
 
+    if (core->shdw_pg_mode == NESTED_PAGING && is_vmx_nested()) { 
+        // there are no passthrough page tables, but
+        // the EPT implementation is using direct_map_pt to store
+        // the EPT root table pointer...  and the EPT tables
+        // are not compatible with regular x86 tables, so we
+        // must not attempt to free them here...
+        return 0;
+    }
+  
+    // we are either in shadow or in SVM nested
+    // in either case, we can nuke the PTs
+
     // Delete the old direct map page tables
     switch(mode) {
 	case REAL:
@@ -609,6 +625,8 @@ int v3_init_nested_paging_core(struct guest_info *core, void *hwinfo)
     return init_ept(core, (struct vmx_hw_info *) hwinfo);
   } else {
     // no initialization for SVM
+    // the direct map page tables are used since the 
+    // nested pt format is identical to the main pt format
     return 0;
   }
 }
@@ -637,9 +655,17 @@ int v3_deinit_nested_paging(struct v3_vm_info *vm)
 
 int v3_deinit_nested_paging_core(struct guest_info *core)
 {
-  // nothing to do..  probably dealloc?  FIXME PAD
-
-  return 0;
+  if (core->shdw_pg_mode == NESTED_PAGING) {
+    if (is_vmx_nested()) {
+     return deinit_ept(core);
+    } else {
+      // SVM nested deinit is handled by the passthrough paging teardown
+      return 0;
+    }
+  } else {
+    // not relevant
+    return 0;
+  }
 }
 
 
diff --git a/palacios/src/palacios/vmx_npt.h b/palacios/src/palacios/vmx_npt.h
index aa1bc14..041d67b 100644
--- a/palacios/src/palacios/vmx_npt.h
+++ b/palacios/src/palacios/vmx_npt.h
@@ -96,6 +96,66 @@ static int init_ept(struct guest_info * core, struct vmx_hw_info * hw_info) {
 
     ept_ptr->pml_base_addr = PAGE_BASE_ADDR(ept_pa);
 
+    PrintDebug(core->vm_info,core,"init_ept direct_map_pt=%p\n",(void*)(core->direct_map_pt));
+
+
+    return 0;
+}
+
+// 
+// You would think we could just the regular 64 bit PT free
+// routine, but no, because the EPT format is slightly different, in that
+// it has no present bit....   We signify present via the read 
+static void delete_page_tables_ept64(ept_pml4_t * pml4) {
+    int i, j, k;
+
+    if (pml4 == NULL) {
+	return;
+    }
+
+    PrintDebug(VM_NONE, VCORE_NONE,"Deleting EPT Page Tables -- PML4 (%p)\n", pml4);
+
+    for (i = 0; i < MAX_PML4E64_ENTRIES; i++) {
+	if (!pml4[i].read && !pml4[i].write && !pml4[i].exec) {
+	    continue;
+	}
+
+	ept_pdp_t * pdpe = (ept_pdp_t *)V3_VAddr((void *)(addr_t)BASE_TO_PAGE_ADDR_4KB(pml4[i].pdp_base_addr));
+
+	for (j = 0; j < MAX_PDPE64_ENTRIES; j++) {
+	    if ((!pdpe[j].read && !pdpe[j].write && !pdpe[j].exec) || (pdpe[j].large_page == 1)) {
+		continue;
+	    }
+
+	    ept_pde_t * pde = (ept_pde_t *)V3_VAddr((void *)(addr_t)BASE_TO_PAGE_ADDR_4KB(pdpe[j].pd_base_addr));
+
+	    for (k = 0; k < MAX_PDE64_ENTRIES; k++) {
+		if ((!pde[k].read && !pde[k].write && !pde[k].exec) || (pde[k].large_page == 1)) {
+		    continue;
+		}
+
+		V3_FreePages((void *)(addr_t)BASE_TO_PAGE_ADDR_4KB(pde[k].pt_base_addr), 1);
+	    }
+	    
+	    V3_FreePages(V3_PAddr(pde), 1);
+	}
+
+	V3_FreePages(V3_PAddr(pdpe), 1);
+    }
+
+    V3_FreePages(V3_PAddr(pml4), 1);
+}
+
+
+
+static int deinit_ept(struct guest_info * core) {
+    ept_pml4_t    *pml;
+
+    pml = (ept_pml4_t *)CR3_TO_PML4E64_VA(core->direct_map_pt);
+
+    delete_page_tables_ept64(pml);
+ 
+    core->direct_map_pt = 0;
 
     return 0;
 }