+/*
+ * This file is part of the Palacios Virtual Machine Monitor developed
+ * by the V3VEE Project with funding from the United States National
+ * Science Foundation and the Department of Energy.
+ *
+ * The V3VEE Project is a joint project between Northwestern University
+ * and the University of New Mexico. You can find out more at
+ * http://www.v3vee.org
+ *
+ * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
+ * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
+ * All rights reserved.
+ *
+ * Author: Jack Lange <jarusl@cs.northwestern.edu>
+ *
+ * This is free software. You are permitted to use,
+ * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
+ */
+
+
#include <palacios/vmm_shadow_paging.h>
#include <palacios/vmm.h>
#include <palacios/vm_guest_mem.h>
+#include <palacios/vmm_decoder.h>
+#include <palacios/vmm_ctrl_regs.h>
+#include <palacios/vmm_hashtable.h>
+#include <palacios/vmm_direct_paging.h>
-int init_shadow_page_state(struct shadow_page_state * state) {
- state->guest_mode = PDE32;
- state->shadow_mode = PDE32;
-
- state->guest_cr3 = 0;
- state->shadow_cr3 = 0;
- return 0;
-}
-int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
- if (info->cpu_mode == PROTECTED_PG) {
- return handle_shadow_pagefault32(info, fault_addr, error_code);
- } else {
- return -1;
- }
-}
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+#include <palacios/vmm_telemetry.h>
+#endif
+#ifdef V3_CONFIG_SYMBIOTIC_SWAP
+#include <palacios/vmm_sym_swap.h>
+#endif
-int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
- pde32_t * guest_pde = NULL;
- pde32_t * shadow_pde = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3);
- addr_t guest_cr3 = CR3_TO_PDE32(info->shdw_pg_state.guest_cr3);
- pt_access_status_t guest_pde_access;
- pt_access_status_t shadow_pde_access;
- pde32_t * guest_pde_entry = NULL;
- pde32_t * shadow_pde_entry = (pde32_t *)&(shadow_pde[PDE32_INDEX(fault_addr)]);
+#ifndef V3_CONFIG_DEBUG_SHADOW_PAGING
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#endif
- if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pde) == -1) {
- PrintDebug("Invalid Guest PDE Address: 0x%x\n", guest_cr3);
- return -1;
- }
+static const char default_strategy[] = "VTLB";
- guest_pde_entry = (pde32_t *)&(guest_pde[PDE32_INDEX(fault_addr)]);
- // Check the guest page permissions
- guest_pde_access = can_access_pde32(guest_pde, fault_addr, error_code);
+static struct hashtable * master_shdw_pg_table = NULL;
- if (guest_pde_access != PT_ACCESS_OK) {
+static uint_t shdw_pg_hash_fn(addr_t key) {
+ char * name = (char *)key;
+ return v3_hash_buffer((uint8_t *)name, strlen(name));
+}
- //
- // inject page fault to the guest (Guest PDE fault)
- //
+static int shdw_pg_eq_fn(addr_t key1, addr_t key2) {
+ char * name1 = (char *)key1;
+ char * name2 = (char *)key2;
- PrintDebug("Guest Page fault (currently not handled)\n");
- return -1;
- }
+ return (strcmp(name1, name2) == 0);
+}
- shadow_pde_access = can_access_pde32(shadow_pde, fault_addr, error_code);
+int V3_init_shdw_paging() {
+ extern struct v3_shdw_pg_impl * __start__v3_shdw_pg_impls[];
+ extern struct v3_shdw_pg_impl * __stop__v3_shdw_pg_impls[];
+ struct v3_shdw_pg_impl ** tmp_impl = __start__v3_shdw_pg_impls;
+ int i = 0;
- if (shadow_pde_access == PT_ENTRY_NOT_PRESENT) {
- pte32_t * shadow_pte = NULL;
+ master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn);
- V3_AllocPages(shadow_pte, 1);
- memset(shadow_pte, 0, PAGE_SIZE);
- shadow_pde_entry->pt_base_addr = PD32_BASE_ADDR(shadow_pte);
-
+ while (tmp_impl != __stop__v3_shdw_pg_impls) {
+ V3_Print("Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name);
- shadow_pde_entry->present = 1;
- shadow_pde_entry->user_page = guest_pde_entry->user_page;
-
- // VMM Specific options
- shadow_pde_entry->write_through = 0;
- shadow_pde_entry->cache_disable = 0;
- shadow_pde_entry->global_page = 0;
- //
+ if (v3_htable_search(master_shdw_pg_table, (addr_t)((*tmp_impl)->name))) {
+ PrintError("Multiple instances of shadow paging impl (%s)\n", (*tmp_impl)->name);
+ return -1;
+ }
- guest_pde_entry->accessed = 1;
+ if (v3_htable_insert(master_shdw_pg_table,
+ (addr_t)((*tmp_impl)->name),
+ (addr_t)(*tmp_impl)) == 0) {
+ PrintError("Could not register shadow paging impl (%s)\n", (*tmp_impl)->name);
+ return -1;
+ }
- if (guest_pde_entry->large_page == 0) {
- shadow_pde_entry->writable = guest_pde_entry->writable;
- } else {
- /*
- * Check the Intel manual because we are ignoring Large Page issues here
- */
+ tmp_impl = &(__start__v3_shdw_pg_impls[++i]);
}
- } else if (shadow_pde_access == PT_WRITE_ERROR) {
-
- //
- // Page Directory Entry marked read-only
- //
-
- PrintDebug("Shadow Paging Write Error\n");
- return -1;
- } else if (shadow_pde_access == PT_USER_ERROR) {
-
- //
- // Page Directory Entry marked non-user
- //
-
- PrintDebug("Shadow Paging User access error\n");
- return -1;
- } else if (shadow_pde_access == PT_ACCESS_OK) {
- pte32_t * shadow_pte = (pte32_t *)PDE32_T_ADDR((*shadow_pde_entry));
- pte32_t * guest_pte = NULL;
-
- // Page Table entry fault
-
- if (guest_pa_to_host_va(info, PDE32_T_ADDR((*guest_pde_entry)), (addr_t*)&guest_pte) == -1) {
- PrintDebug("Invalid Guest PTE Address: 0x%x\n", PDE32_T_ADDR((*guest_pde_entry)));
- return -1;
- }
+ return 0;
+}
+int V3_deinit_shdw_paging() {
+ v3_free_htable(master_shdw_pg_table, 0, 0);
+ return 0;
+}
- if (handle_shadow_pte32_fault(info, fault_addr, error_code, shadow_pte, guest_pte) == -1) {
- PrintDebug("Error handling Page fault caused by PTE\n");
- return -1;
- }
- }
- PrintDebugPageTables(shadow_pde);
+/***
+ *** There be dragons
+ ***/
- return 0;
+
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+static void telemetry_cb(struct v3_vm_info * vm, void * private_data, char * hdr) {
+ int i = 0;
+ for (i = 0; i < vm->num_cores; i++) {
+ struct guest_info * core = &(vm->cores[i]);
+
+ V3_Print("%s Guest Page faults: %d\n", hdr, core->shdw_pg_state.guest_faults);
+ }
}
+#endif
-/*
- * We assume the the guest pte pointer has already been translated to a host virtual address
- */
-int handle_shadow_pte32_fault(struct guest_info* info,
- addr_t fault_addr,
- pf_error_t error_code,
- pte32_t * shadow_pte,
- pte32_t * guest_pte) {
+int v3_init_shdw_pg_state(struct guest_info * core) {
+ struct v3_shdw_pg_state * state = &(core->shdw_pg_state);
+ struct v3_shdw_pg_impl * impl = core->vm_info->shdw_impl.current_impl;
+
- pt_access_status_t guest_pte_access;
- pt_access_status_t shadow_pte_access;
- // pte32_t * guest_pte_entry = (pte32_t *)&(guest_pte[PTE32_INDEX(fault_addr)]);;
- // pte32_t * shadow_pte_entry = (pte32_t *)&(shadow_pte[PTE32_INDEX(fault_addr)]);
+ state->guest_cr3 = 0;
+ state->guest_cr0 = 0;
+ state->guest_efer.value = 0x0LL;
- // Check the guest page permissions
- guest_pte_access = can_access_pte32(guest_pte, fault_addr, error_code);
+ if (impl->local_init(core) == -1) {
+ PrintError("Error in Shadow paging local initialization (%s)\n", impl->name);
+ return -1;
+ }
+
- if (guest_pte_access != PT_ACCESS_OK) {
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+ v3_add_telemetry_cb(core->vm_info, telemetry_cb, NULL);
+#endif
+
+ return 0;
+}
- //
- // Inject page fault into the guest
- //
- PrintDebug("Guest Page fault (currently not handled)\n");
- return -1;
- }
+int v3_deinit_shdw_pg_state(struct guest_info * core) {
+ struct v3_shdw_pg_impl * impl = core->vm_info->shdw_impl.current_impl;
+ if (impl->local_deinit(core) == -1) {
+ PrintError("Error deinitializing shadow paging state\n");
+ return -1;
+ }
- shadow_pte_access = can_access_pte32(shadow_pte, fault_addr, error_code);
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+ v3_remove_telemetry_cb(core->vm_info, telemetry_cb, NULL);
+#endif
- if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) {
+ return 0;
+}
- //
- // Page Table Entry Not Present
- //
- } else if (shadow_pte_access == PT_WRITE_ERROR) {
- //
- // Page Table Entry marked read-only
- //
+int v3_init_shdw_impl(struct v3_vm_info * vm) {
+ struct v3_shdw_impl_state * impl_state = &(vm->shdw_impl);
+ v3_cfg_tree_t * pg_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "paging");
+ char * pg_mode = v3_cfg_val(pg_cfg, "mode");
+ char * pg_strat = v3_cfg_val(pg_cfg, "strategy");
+ struct v3_shdw_pg_impl * impl = NULL;
+
+ PrintDebug("Checking if shadow paging requested.\n");
+ if ((pg_mode != NULL) && (strcasecmp(pg_mode, "nested") == 0)) {
+ PrintDebug("Nested paging specified - not initializing shadow paging.\n");
+ return 0;
+ }
- PrintDebug("Shadow Paging Write Error\n");
- return -1;
- } else if (shadow_pte_access == PT_USER_ERROR) {
+ if (pg_strat == NULL) {
+ pg_strat = (char *)default_strategy;
+ }
+
+ V3_Print("Initialization of Shadow Paging implementation\n");
- //
- // Page Table Entry marked non-user
- //
-
- PrintDebug("Shadow Paging User access error\n");
- return -1;
- } else if (shadow_pte_access == PT_ACCESS_OK) {
+ impl = (struct v3_shdw_pg_impl *)v3_htable_search(master_shdw_pg_table, (addr_t)pg_strat);
- PrintDebug("Page Fault occurred for No Reason\n");
- return -1;
- }
+ if (impl == NULL) {
+ PrintError("Could not find shadow paging impl (%s)\n", pg_strat);
+ return -1;
+ }
+
+ impl_state->current_impl = impl;
+ if (impl->init(vm, pg_cfg) == -1) {
+ PrintError("Could not initialize Shadow paging implemenation (%s)\n", impl->name);
+ return -1;
+ }
- return 0;
+ return 0;
}
+int v3_deinit_shdw_impl(struct v3_vm_info * vm) {
+ struct v3_shdw_pg_impl * impl = vm->shdw_impl.current_impl;
+ if (impl->deinit(vm) == -1) {
+ PrintError("Error deinitializing shadow paging implementation\n");
+ return -1;
+ }
-addr_t create_new_shadow_pt32(struct guest_info * info) {
- void * host_pde = 0;
+ return 0;
+}
- V3_AllocPages(host_pde, 1);
- memset(host_pde, 0, PAGE_SIZE);
- return (addr_t)host_pde;
+// Reads the guest CR3 register
+// creates new shadow page tables
+// updates the shadow CR3 register to point to the new pts
+int v3_activate_shadow_pt(struct guest_info * core) {
+ struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+ struct v3_shdw_pg_impl * impl = state->current_impl;
+ return impl->activate_shdw_pt(core);
}
+// This must flush any caches
+// and reset the cr3 value to the correct value
+int v3_invalidate_shadow_pts(struct guest_info * core) {
+ struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+ struct v3_shdw_pg_impl * impl = state->current_impl;
+ return impl->invalidate_shdw_pt(core);
+}
-addr_t setup_shadow_pt32(struct guest_info * info, addr_t virt_cr3) {
- addr_t cr3_guest_addr = CR3_TO_PDE32(virt_cr3);
- pde32_t * guest_pde;
- pde32_t * host_pde = NULL;
- int i;
-
- // Setup up guest_pde to point to the PageDir in host addr
- if (guest_pa_to_host_va(info, cr3_guest_addr, (addr_t*)&guest_pde) == -1) {
- return 0;
- }
+
+int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
- V3_AllocPages(host_pde, 1);
- memset(host_pde, 0, PAGE_SIZE);
+ if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
+ // If paging is not turned on we need to handle the special cases
+ return v3_handle_passthrough_pagefault(core, fault_addr, error_code);
+ } else if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
+ struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+ struct v3_shdw_pg_impl * impl = state->current_impl;
+
+ return impl->handle_pagefault(core, fault_addr, error_code);
+ } else {
+ PrintError("Invalid Memory mode\n");
+ return -1;
+ }
+}
- for (i = 0; i < MAX_PDE32_ENTRIES; i++) {
- if (guest_pde[i].present == 1) {
- addr_t pt_host_addr;
- addr_t host_pte;
- if (guest_pa_to_host_va(info, PDE32_T_ADDR(guest_pde[i]), &pt_host_addr) == -1) {
- return 0;
- }
+int v3_handle_shadow_invlpg(struct guest_info * core) {
+ uchar_t instr[15];
+ struct x86_instr dec_instr;
+ int ret = 0;
+ addr_t vaddr = 0;
- if ((host_pte = setup_shadow_pte32(info, pt_host_addr)) == 0) {
- return 0;
- }
+ if (v3_get_vm_mem_mode(core) != VIRTUAL_MEM) {
+ // Paging must be turned on...
+ // should handle with some sort of fault I think
+ PrintError("ERROR: INVLPG called in non paged mode\n");
+ return -1;
+ }
+
+ if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
+ ret = v3_read_gpa_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
+ } else {
+ ret = v3_read_gva_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
+ }
- host_pde[i].present = 1;
- host_pde[i].pt_base_addr = PD32_BASE_ADDR(host_pte);
+ if (ret == -1) {
+ PrintError("Could not read instruction into buffer\n");
+ return -1;
+ }
- //
- // Set Page DIR flags
- //
+ if (v3_decode(core, (addr_t)instr, &dec_instr) == -1) {
+ PrintError("Decoding Error\n");
+ return -1;
+ }
+
+ if ((dec_instr.op_type != V3_OP_INVLPG) ||
+ (dec_instr.num_operands != 1) ||
+ (dec_instr.dst_operand.type != MEM_OPERAND)) {
+ PrintError("Decoder Error: Not a valid INVLPG instruction...\n");
+ return -1;
}
- }
- PrintDebugPageTables(host_pde);
+ vaddr = dec_instr.dst_operand.operand;
+
+ core->rip += dec_instr.instr_length;
- return (addr_t)host_pde;
+ {
+ struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+ struct v3_shdw_pg_impl * impl = state->current_impl;
+
+ return impl->handle_invlpg(core, vaddr);
+ }
}
-addr_t setup_shadow_pte32(struct guest_info * info, addr_t pt_host_addr) {
- pte32_t * guest_pte = (pte32_t *)pt_host_addr;
- pte32_t * host_pte = NULL;
- int i;
- V3_AllocPages(host_pte, 1);
- memset(host_pte, 0, PAGE_SIZE);
- for (i = 0; i < MAX_PTE32_ENTRIES; i++) {
- if (guest_pte[i].present == 1) {
- addr_t guest_pa = PTE32_T_ADDR(guest_pte[i]);
- shadow_mem_type_t page_type;
- addr_t host_pa = 0;
- page_type = get_shadow_addr_type(info, guest_pa);
+int v3_inject_guest_pf(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
+ core->ctrl_regs.cr2 = fault_addr;
+
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+ core->shdw_pg_state.guest_faults++;
+#endif
+
+ return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code);
+}
- if (page_type == HOST_REGION_PHYSICAL_MEMORY) {
- host_pa = get_shadow_addr(info, guest_pa);
- } else {
-
- //
- // Setup various memory types
- //
- }
- host_pte[i].page_base_addr = PT32_BASE_ADDR(host_pa);
- host_pte[i].present = 1;
+int v3_is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) {
+ /* basically the reasoning is that there can be multiple reasons for a page fault:
+ If there is a permissions failure for a page present in the guest _BUT_
+ the reason for the fault was that the page is not present in the shadow,
+ _THEN_ we have to map the shadow page in and reexecute, this will generate
+ a permissions fault which is _THEN_ valid to send to the guest
+ _UNLESS_ both the guest and shadow have marked the page as not present
+
+ whew...
+ */
+ if (guest_access != PT_ACCESS_OK) {
+ // Guest Access Error
+
+ if ((shadow_access != PT_ACCESS_NOT_PRESENT) &&
+ (guest_access != PT_ACCESS_NOT_PRESENT)) {
+ // aka (guest permission error)
+ return 1;
+ }
+
+ /*
+ if ((shadow_access == PT_ACCESS_NOT_PRESENT) &&
+ (guest_access == PT_ACCESS_NOT_PRESENT)) {
+ // Page tables completely blank, handle guest first
+ return 1;
+ }
+ */
+
+ if (guest_access == PT_ACCESS_NOT_PRESENT) {
+ // Page tables completely blank, handle guest first
+ return 1;
+ }
+
+ // Otherwise we'll handle the guest fault later...?
}
- }
- return (addr_t)host_pte;
+ return 0;
}
