+/*
+ * 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;
-}
+#ifdef CONFIG_SHADOW_PAGING_TELEMETRY
+#include <palacios/vmm_telemetry.h>
+#endif
-int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
-
- if (info->mem_mode == PHYSICAL_MEM) {
- // If paging is not turned on we need to handle the special cases
- return handle_special_page_fault(info, fault_addr, error_code);
- } else if (info->mem_mode == VIRTUAL_MEM) {
-
- switch (info->cpu_mode) {
- case PROTECTED:
- return handle_shadow_pagefault32(info, fault_addr, error_code);
- break;
- case PROTECTED_PAE:
- case LONG:
- // currently not handled
- return -1;
- break;
- default:
- return -1;
- }
- } else {
- PrintDebug("Invalid Memory mode\n");
- return -1;
- }
-}
+#ifdef CONFIG_SYMBIOTIC_SWAP
+#include <palacios/vmm_sym_swap.h>
+#endif
+#ifndef CONFIG_DEBUG_SHADOW_PAGING
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#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)]);
- 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;
- }
+/***
+ *** There be dragons
+ ***/
- guest_pde_entry = (pde32_t *)&(guest_pde[PDE32_INDEX(fault_addr)]);
+struct shadow_page_data {
+ v3_reg_t cr3;
+ addr_t page_pa;
+
+ struct list_head page_list_node;
+};
- // Check the guest page permissions
- guest_pde_access = can_access_pde32(guest_pde, fault_addr, error_code);
- if (guest_pde_access != PT_ACCESS_OK) {
- // inject page fault to the guest (Guest PDE fault)
- info->ctrl_regs.cr2 = fault_addr;
- raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
+static struct shadow_page_data * create_new_shadow_pt(struct guest_info * info);
+static int inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code);
+static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access);
- PrintDebug("Injecting PDE pf to guest\n");
- return 0;
- }
- shadow_pde_access = can_access_pde32(shadow_pde, fault_addr, error_code);
+#include "vmm_shadow_paging_32.h"
+#include "vmm_shadow_paging_32pae.h"
+#include "vmm_shadow_paging_64.h"
- if (shadow_pde_access == PT_ENTRY_NOT_PRESENT) {
- pte32_t * shadow_pte = NULL;
- V3_AllocPages(shadow_pte, 1);
- memset(shadow_pte, 0, PAGE_SIZE);
-
- shadow_pde_entry->pt_base_addr = PD32_BASE_ADDR(shadow_pte);
-
+#ifdef CONFIG_SHADOW_PAGING_TELEMETRY
+static void telemetry_cb(struct guest_info * info, void * private_data, char * hdr) {
+ V3_Print("%s Guest Page faults: %d\n", hdr, info->shdw_pg_state.guest_faults);
+}
+#endif
- 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;
- //
- guest_pde_entry->accessed = 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
- * Also be wary of hooked pages
- */
+int v3_init_shadow_page_state(struct guest_info * info) {
+ struct shadow_page_state * state = &(info->shdw_pg_state);
+
+ state->guest_cr3 = 0;
+ state->guest_cr0 = 0;
+ state->guest_efer.value = 0x0LL;
- PrintDebug("Large PAge!!!\n");
- return -1;
+ INIT_LIST_HEAD(&(state->page_list));
+#ifdef CONFIG_SHADOW_PAGING_TELEMETRY
+ if (info->enable_telemetry) {
+ v3_add_telemetry_cb(info, telemetry_cb, NULL);
}
+#endif
+
+ return 0;
+}
- } 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)));
- // Machine check the guest
-
- raise_exception(info, MC_EXCEPTION);
-
- 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;
+// 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 * info) {
+ switch (v3_get_vm_cpu_mode(info)) {
+
+ case PROTECTED:
+ return activate_shadow_pt_32(info);
+ case PROTECTED_PAE:
+ return activate_shadow_pt_32pae(info);
+ case LONG:
+ case LONG_32_COMPAT:
+ case LONG_16_COMPAT:
+ return activate_shadow_pt_64(info);
+ default:
+ PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(info)));
+ return -1;
}
- } else {
- // Unknown error raise page fault in guest
- info->ctrl_regs.cr2 = fault_addr;
- raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
-
- // For debugging we will return an error here for the time being,
- // this probably shouldn't ever happen
- PrintDebug("Unknown Error occurred\n");
- PrintDebug("Manual Says to inject page fault into guest\n");
- return -1;
- }
-
- //PrintDebugPageTables(shadow_pde);
- PrintDebug("Returning end of PDE function\n");
- return 0;
+ return 0;
}
-/*
- * 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) {
-
- 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)]);
-
+// This must flush any caches
+// and reset the cr3 value to the correct value
+int v3_invalidate_shadow_pts(struct guest_info * info) {
+ return v3_activate_shadow_pt(info);
+}
- // Check the guest page permissions
- guest_pte_access = can_access_pte32(guest_pte, fault_addr, error_code);
+int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
- if (guest_pte_access != PT_ACCESS_OK) {
- // Inject page fault into the guest
-
- info->ctrl_regs.cr2 = fault_addr;
- raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
-
- PrintDebug("Access error injecting pf to guest\n");
- return 0;
- }
-
-
- shadow_pte_access = can_access_pte32(shadow_pte, fault_addr, error_code);
-
- if (shadow_pte_access == PT_ACCESS_OK) {
- // Inconsistent state...
- // Guest Re-Entry will flush page tables and everything should now work
- PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
- return 0;
- } else if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) {
- addr_t shadow_pa;
- addr_t guest_pa = PTE32_T_ADDR((*guest_pte_entry));
-
- // Page Table Entry Not Present
-
- host_region_type_t host_page_type = get_shadow_addr_type(info, guest_pa);
-
- if (host_page_type == HOST_REGION_INVALID) {
- // Inject a machine check in the guest
-
- raise_exception(info, MC_EXCEPTION);
-
- PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_pa);
- return 0;
-
- } else if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
-
- shadow_pa = get_shadow_addr(info, guest_pa);
-
- shadow_pte_entry->page_base_addr = PT32_BASE_ADDR(shadow_pa);
-
- shadow_pte_entry->present = guest_pte_entry->present;
- shadow_pte_entry->user_page = guest_pte_entry->user_page;
-
- //set according to VMM policy
- shadow_pte_entry->write_through = 0;
- shadow_pte_entry->cache_disable = 0;
- shadow_pte_entry->global_page = 0;
- //
-
- guest_pte_entry->accessed = 1;
-
- if (guest_pte_entry->dirty == 1) {
- shadow_pte_entry->writable = guest_pte_entry->writable;
- } else if ((guest_pte_entry->dirty == 0) && (error_code.write == 1)) {
- shadow_pte_entry->writable = guest_pte_entry->writable;
- guest_pte_entry->dirty = 1;
- } else if ((guest_pte_entry->dirty = 0) && (error_code.write == 0)) {
- shadow_pte_entry->writable = 0;
- }
+ if (v3_get_vm_mem_mode(info) == PHYSICAL_MEM) {
+ // If paging is not turned on we need to handle the special cases
+ return v3_handle_passthrough_pagefault(info, fault_addr, error_code);
+ } else if (v3_get_vm_mem_mode(info) == VIRTUAL_MEM) {
+
+ switch (v3_get_vm_cpu_mode(info)) {
+ case PROTECTED:
+ return handle_shadow_pagefault_32(info, fault_addr, error_code);
+ break;
+ case PROTECTED_PAE:
+ return handle_shadow_pagefault_32pae(info, fault_addr, error_code);
+ case LONG:
+ case LONG_32_COMPAT:
+ case LONG_16_COMPAT:
+ return handle_shadow_pagefault_64(info, fault_addr, error_code);
+ break;
+ default:
+ PrintError("Unhandled CPU Mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(info)));
+ return -1;
+ }
} else {
- // Page fault handled by hook functions
- if (handle_special_page_fault(info, fault_addr, error_code) == -1) {
- PrintDebug("Special Page fault handler returned error for address: %x\n", fault_addr);
+ PrintError("Invalid Memory mode\n");
return -1;
- }
}
-
- } else if ((shadow_pte_access == PT_WRITE_ERROR) &&
- (guest_pte_entry->dirty == 0)) {
- guest_pte_entry->dirty = 1;
- shadow_pte_entry->writable = guest_pte_entry->writable;
-
- PrintDebug("Shadow PTE Write Error\n");
-
- return 0;
- } else {
- // Inject page fault into the guest
-
- info->ctrl_regs.cr2 = fault_addr;
- raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
-
- PrintDebug("PTE Page fault fell through... Not sure if this should ever happen\n");
- PrintDebug("Manual Says to inject page fault into guest\n");
- return -1;
- }
-
- PrintDebug("Returning end of function\n");
- return 0;
}
+int v3_handle_shadow_invlpg(struct guest_info * info) {
+ uchar_t instr[15];
+ struct x86_instr dec_instr;
+ int ret = 0;
+ addr_t vaddr = 0;
-addr_t create_new_shadow_pt32(struct guest_info * info) {
- void * host_pde = 0;
-
- V3_AllocPages(host_pde, 1);
- memset(host_pde, 0, PAGE_SIZE);
-
- return (addr_t)host_pde;
-}
-
-
-
-/* Currently Does not work with Segmentation!!! */
-int handle_shadow_invlpg(struct guest_info * info) {
- if (info->mem_mode != VIRTUAL_MEM) {
- // Paging must be turned on...
- // should handle with some sort of fault I think
- PrintDebug("ERROR: INVLPG called in non paged mode\n");
- return -1;
- }
+ if (v3_get_vm_mem_mode(info) != 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(info) == PHYSICAL_MEM) {
+ ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
+ } else {
+ ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
+ }
- if (info->cpu_mode == PROTECTED) {
- char instr[15];
- int ret;
- int index = 0;
+ if (ret == -1) {
+ PrintError("Could not read instruction into buffer\n");
+ return -1;
+ }
- ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
- if (ret != 15) {
- PrintDebug("Could not read instruction 0x%x (ret=%d)\n", info->rip, ret);
- return -1;
+ if (v3_decode(info, (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;
}
-
- /* Can INVLPG work with Segments?? */
- while (is_prefix_byte(instr[index])) {
- index++;
+ vaddr = dec_instr.dst_operand.operand;
+
+ info->rip += dec_instr.instr_length;
+
+ switch (v3_get_vm_cpu_mode(info)) {
+ case PROTECTED:
+ return handle_shadow_invlpg_32(info, vaddr);
+ case PROTECTED_PAE:
+ return handle_shadow_invlpg_32pae(info, vaddr);
+ case LONG:
+ case LONG_32_COMPAT:
+ case LONG_16_COMPAT:
+ return handle_shadow_invlpg_64(info, vaddr);
+ default:
+ PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(info)));
+ return -1;
}
-
-
- if ((instr[index] == (uchar_t)0x0f) &&
- (instr[index + 1] == (uchar_t)0x01)) {
+}
- addr_t first_operand;
- addr_t second_operand;
- operand_type_t addr_type;
- index += 2;
- addr_type = decode_operands32(&(info->vm_regs), instr + index, &index, &first_operand, &second_operand, REG32);
- if (addr_type == MEM_OPERAND) {
- pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3);
- pde32_t * shadow_pde_entry = (pde32_t *)&shadow_pd[PDE32_INDEX(first_operand)];
+static struct shadow_page_data * create_new_shadow_pt(struct guest_info * info) {
+ struct shadow_page_state * state = &(info->shdw_pg_state);
+ v3_reg_t cur_cr3 = info->ctrl_regs.cr3;
+ struct shadow_page_data * page_tail = NULL;
+ addr_t shdw_page = 0;
- //PrintDebug("PDE Index=%d\n", PDE32_INDEX(first_operand));
- //PrintDebug("FirstOperand = %x\n", first_operand);
+ if (!list_empty(&(state->page_list))) {
+ page_tail = list_tail_entry(&(state->page_list), struct shadow_page_data, page_list_node);
+
+ if (page_tail->cr3 != cur_cr3) {
+ PrintDebug("Reusing old shadow Page: %p (cur_CR3=%p)(page_cr3=%p) \n",
+ (void *)(addr_t)page_tail->page_pa,
+ (void *)(addr_t)cur_cr3,
+ (void *)(addr_t)(page_tail->cr3));
- if (shadow_pde_entry->large_page == 1) {
- shadow_pde_entry->present = 0;
- } else {
- if (shadow_pde_entry->present == 1) {
- pte32_t * shadow_pt = (pte32_t *)PDE32_T_ADDR((*shadow_pde_entry));
- pte32_t * shadow_pte_entry = (pte32_t *)&shadow_pt[PTE32_INDEX(first_operand)];
+ list_move(&(page_tail->page_list_node), &(state->page_list));
- shadow_pte_entry->present = 0;
- }
- }
+ memset(V3_VAddr((void *)(page_tail->page_pa)), 0, PAGE_SIZE_4KB);
- info->rip += index;
- } else {
- PrintDebug("Invalid Operand type\n");
- return -1;
- }
- } else {
- PrintDebug("invalid Instruction Opcode\n");
- PrintTraceMemDump(instr, 15);
- return -1;
+ return page_tail;
+ }
}
- }
- return 0;
-}
+ // else
+ page_tail = (struct shadow_page_data *)V3_Malloc(sizeof(struct shadow_page_data));
+ page_tail->page_pa = (addr_t)V3_AllocPages(1);
+ PrintDebug("Allocating new shadow Page: %p (cur_cr3=%p)\n",
+ (void *)(addr_t)page_tail->page_pa,
+ (void *)(addr_t)cur_cr3);
-/* Deprecated */
-/*
-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;
- }
-
- V3_AllocPages(host_pde, 1);
- memset(host_pde, 0, PAGE_SIZE);
+ page_tail->cr3 = cur_cr3;
+ list_add(&(page_tail->page_list_node), &(state->page_list));
- for (i = 0; i < MAX_PDE32_ENTRIES; i++) {
- if (guest_pde[i].present == 1) {
- addr_t pt_host_addr;
- addr_t host_pte;
+ shdw_page = (addr_t)V3_VAddr((void *)(page_tail->page_pa));
+ memset((void *)shdw_page, 0, PAGE_SIZE_4KB);
- if (guest_pa_to_host_va(info, PDE32_T_ADDR(guest_pde[i]), &pt_host_addr) == -1) {
- return 0;
- }
-
- if ((host_pte = setup_shadow_pte32(info, pt_host_addr)) == 0) {
- return 0;
- }
+ return page_tail;
+}
- host_pde[i].present = 1;
- host_pde[i].pt_base_addr = PD32_BASE_ADDR(host_pte);
- //
- // Set Page DIR flags
- //
- }
- }
+static int inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
+ info->ctrl_regs.cr2 = fault_addr;
- PrintDebugPageTables(host_pde);
+#ifdef CONFIG_SHADOW_PAGING_TELEMETRY
+ info->shdw_pg_state.guest_faults++;
+#endif
- return (addr_t)host_pde;
+ return v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
}
+static int 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
-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);
+ whew...
+ */
+ if (guest_access != PT_ACCESS_OK) {
+ // Guest Access Error
- 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;
+ if ((shadow_access != PT_ACCESS_NOT_PRESENT) &&
+ (guest_access != PT_ACCESS_NOT_PRESENT)) {
+ // aka (guest permission error)
+ return 1;
+ }
- page_type = get_shadow_addr_type(info, guest_pa);
+ /*
+ if ((shadow_access == PT_ACCESS_NOT_PRESENT) &&
+ (guest_access == PT_ACCESS_NOT_PRESENT)) {
+ // Page tables completely blank, handle guest first
+ return 1;
+ }
+ */
- if (page_type == HOST_REGION_PHYSICAL_MEMORY) {
- host_pa = get_shadow_addr(info, guest_pa);
- } else {
+ if (guest_access == PT_ACCESS_NOT_PRESENT) {
+ // Page tables completely blank, handle guest first
+ return 1;
+ }
- //
- // Setup various memory types
- //
- }
-
- host_pte[i].page_base_addr = PT32_BASE_ADDR(host_pa);
- host_pte[i].present = 1;
+ // Otherwise we'll handle the guest fault later...?
}
- }
- return (addr_t)host_pte;
+ return 0;
}
-*/
+
