#include <palacios/vmm.h>
#include <palacios/vm_guest_mem.h>
#include <palacios/vmm_decoder.h>
+#include <palacios/vmm_ctrl_regs.h>
-#ifndef DEBUG_SHADOW_PAGING
-#undef PrintDebug
-#define PrintDebug(fmt, args...)
-#endif
-
+#include <palacios/vmm_hashtable.h>
-/***
- *** There be dragons
- ***/
+#include <palacios/vmm_direct_paging.h>
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+#include <palacios/vmm_telemetry.h>
+#endif
-DEFINE_HASHTABLE_INSERT(add_cr3_to_cache, addr_t, struct hashtable *);
-DEFINE_HASHTABLE_SEARCH(find_cr3_in_cache, addr_t, struct hashtable *);
-DEFINE_HASHTABLE_REMOVE(del_cr3_from_cache, addr_t, struct hashtable *, 0);
+#ifdef V3_CONFIG_SYMBIOTIC_SWAP
+#include <palacios/vmm_sym_swap.h>
+#endif
+#ifndef V3_CONFIG_DEBUG_SHADOW_PAGING
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#endif
-DEFINE_HASHTABLE_INSERT(add_pte_map, addr_t, addr_t);
-DEFINE_HASHTABLE_SEARCH(find_pte_map, addr_t, addr_t);
-DEFINE_HASHTABLE_REMOVE(del_pte_map, addr_t, addr_t, 0);
+static const char default_strategy[] = "VTLB";
+static struct hashtable * master_shdw_pg_table = NULL;
-static uint_t pte_hash_fn(addr_t key) {
- return hash_long(key, 32);
-}
-static int pte_equals(addr_t key1, addr_t key2) {
- return (key1 == key2);
-}
+struct event_callback {
+ int (*callback)(struct guest_info *core, struct v3_shdw_pg_event *event, void *priv_data);
+ void *priv_data;
-static uint_t cr3_hash_fn(addr_t key) {
- return hash_long(key, 32);
-}
+ struct list_head node;
+};
-static int cr3_equals(addr_t key1, addr_t key2) {
- return (key1 == key2);
+static uint_t shdw_pg_hash_fn(addr_t key) {
+ char * name = (char *)key;
+ return v3_hash_buffer((uint8_t *)name, strlen(name));
}
+static int shdw_pg_eq_fn(addr_t key1, addr_t key2) {
+ char * name1 = (char *)key1;
+ char * name2 = (char *)key2;
-static 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);
-
-static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code);
-
-int init_shadow_page_state(struct guest_info * info) {
- struct shadow_page_state * state = &(info->shdw_pg_state);
- state->guest_mode = PDE32;
- state->shadow_mode = PDE32;
-
- state->guest_cr3 = 0;
- state->shadow_cr3 = 0;
-
-
- state->cr3_cache = create_hashtable(0, &cr3_hash_fn, &cr3_equals);
-
- state->cached_cr3 = 0;
- state->cached_ptes = NULL;
-
- return 0;
+ return (strcmp(name1, name2) == 0);
}
+static int have_callbacks(struct guest_info *core)
+{
+ // lock acquistion unnecessary
+ // caller will acquire the lock before *iterating* through the list
+ // so any race will be resolved then
+ return !list_empty(&(core->vm_info->shdw_impl.event_callback_list));
+}
+static void dispatch_event(struct guest_info *core, struct v3_shdw_pg_event *event)
+{
+ struct event_callback *cb,*temp;
+ v3_read_lock(&(core->vm_info->shdw_impl.event_callback_lock));
-/*
- For now we'll do something a little more lightweight
-int cache_page_tables32(struct guest_info * info, addr_t pde) {
- struct shadow_page_state * state = &(info->shdw_pg_state);
- addr_t pde_host_addr;
- pde32_t * tmp_pde;
- struct hashtable * pte_cache = NULL;
- int i = 0;
+ list_for_each_entry_safe(cb,
+ temp,
+ &(core->vm_info->shdw_impl.event_callback_list),
+ node) {
+ cb->callback(core,event,cb->priv_data);
+ }
+ v3_read_unlock(&(core->vm_info->shdw_impl.event_callback_lock));
+}
- pte_cache = (struct hashtable *)find_cr3_in_cache(state->cr3_cache, pde);
- if (pte_cache != NULL) {
- PrintError("CR3 already present in cache\n");
- state->current_ptes = pte_cache;
- return 1;
- } else {
- PrintError("Creating new CR3 cache entry\n");
- pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals);
- state->current_ptes = pte_cache;
- add_cr3_to_cache(state->cr3_cache, pde, pte_cache);
- }
- if (guest_pa_to_host_va(info, pde, &pde_host_addr) == -1) {
- PrintError("Could not lookup host address of guest PDE\n");
- return -1;
- }
+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;
- tmp_pde = (pde32_t *)pde_host_addr;
+ master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn);
- add_pte_map(pte_cache, pde, pde_host_addr);
+ while (tmp_impl != __stop__v3_shdw_pg_impls) {
+ V3_Print(VM_NONE, VCORE_NONE, "Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name);
- for (i = 0; i < MAX_PDE32_ENTRIES; i++) {
- if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) {
- addr_t pte_host_addr;
+ if (v3_htable_search(master_shdw_pg_table, (addr_t)((*tmp_impl)->name))) {
+ PrintError(VM_NONE, VCORE_NONE, "Multiple instances of shadow paging impl (%s)\n", (*tmp_impl)->name);
+ return -1;
+ }
- if (guest_pa_to_host_va(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) {
- PrintError("Could not lookup host address of guest PDE\n");
- return -1;
- }
+ if (v3_htable_insert(master_shdw_pg_table,
+ (addr_t)((*tmp_impl)->name),
+ (addr_t)(*tmp_impl)) == 0) {
+ PrintError(VM_NONE, VCORE_NONE, "Could not register shadow paging impl (%s)\n", (*tmp_impl)->name);
+ return -1;
+ }
- add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr);
+ tmp_impl = &(__start__v3_shdw_pg_impls[++i]);
}
- }
+ return 0;
+}
- return 0;
+int V3_deinit_shdw_paging() {
+ v3_free_htable(master_shdw_pg_table, 0, 0);
+ return 0;
}
-*/
-int cache_page_tables32(struct guest_info * info, addr_t pde) {
- struct shadow_page_state * state = &(info->shdw_pg_state);
- addr_t pde_host_addr;
- pde32_t * tmp_pde;
- struct hashtable * pte_cache = NULL;
- int i = 0;
- if (pde == state->cached_cr3) {
- return 1;
- }
- if (state->cached_ptes != NULL) {
- hashtable_destroy(state->cached_ptes, 0, 0);
- state->cached_ptes = NULL;
- }
+/***
+ *** There be dragons
+ ***/
- state->cached_cr3 = pde;
- pte_cache = create_hashtable(0, &pte_hash_fn, &pte_equals);
- state->cached_ptes = pte_cache;
+#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]);
- if (guest_pa_to_host_pa(info, pde, &pde_host_addr) == -1) {
- PrintError("Could not lookup host address of guest PDE\n");
- return -1;
- }
+ V3_Print(vm, core, "%s Guest Page faults: %d\n", hdr, core->shdw_pg_state.guest_faults);
+ }
+}
+#endif
- tmp_pde = (pde32_t *)pde_host_addr;
- add_pte_map(pte_cache, pde, pde_host_addr);
+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;
+
- for (i = 0; i < MAX_PDE32_ENTRIES; i++) {
- if ((tmp_pde[i].present) && (tmp_pde[i].large_page == 0)) {
- addr_t pte_host_addr;
+ state->guest_cr3 = 0;
+ state->guest_cr0 = 0;
+ state->guest_efer.value = 0x0LL;
- if (guest_pa_to_host_pa(info, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), &pte_host_addr) == -1) {
- PrintError("Could not lookup host address of guest PDE\n");
+ if (impl->local_init(core) == -1) {
+ PrintError(core->vm_info, core, "Error in Shadow paging local initialization (%s)\n", impl->name);
return -1;
- }
-
- add_pte_map(pte_cache, (addr_t)(PDE32_T_ADDR(tmp_pde[i])), pte_host_addr);
}
- }
- return 0;
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+ v3_add_telemetry_cb(core->vm_info, telemetry_cb, NULL);
+#endif
+
+
+ return 0;
}
+int v3_deinit_shdw_pg_state(struct guest_info * core) {
+ struct v3_shdw_pg_impl * impl = NULL;
-int v3_replace_shdw_page32(struct guest_info * info, addr_t location, pte32_t * new_page, pte32_t * old_page) {
- pde32_t * shadow_pd = (pde32_t *)CR3_TO_PDE32(info->shdw_pg_state.shadow_cr3);
- pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(location)]);
+ if (!core || !core->vm_info) {
+ return -1;
+ }
- if (shadow_pde->large_page == 0) {
- pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde));
- pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(location)]);
+ impl = core->vm_info->shdw_impl.current_impl;
- //if (shadow_pte->present == 1) {
- *(uint_t *)old_page = *(uint_t *)shadow_pte;
- //}
+ if (impl && impl->local_deinit(core) == -1) {
+ PrintError(core->vm_info, core, "Error deinitializing shadow paging state\n");
+ return -1;
+ }
- *(uint_t *)shadow_pte = *(uint_t *)new_page;
- } else {
- // currently unhandled
- return -1;
- }
-
- return 0;
+ return 0;
}
+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(vm, VCORE_NONE, "Checking if shadow paging requested.\n");
+ if (pg_mode == NULL) {
+ V3_Print(vm, VCORE_NONE, "No paging mode specified, assuming shadow with defaults\n");
+ pg_mode = "shadow";
+ } else {
+ if (strcasecmp(pg_mode, "nested") == 0) {
+ // this check is repeated here (compare to vmm_config's determine paging mode) since
+ // shadow paging initialization *precedes* per-core pre-config.
+ extern v3_cpu_arch_t v3_mach_type;
+ if ((v3_mach_type == V3_SVM_REV3_CPU) ||
+ (v3_mach_type == V3_VMX_EPT_CPU) ||
+ (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
+ PrintDebug(vm, VCORE_NONE, "Nested paging specified on machine that supports it - not initializing shadow paging\n");
+ return 0;
+ } else {
+ V3_Print(vm, VCORE_NONE, "Nested paging specified but machine does not support it - falling back to shadow paging with defaults\n");
+ pg_mode = "shadow";
+ }
+ } else if (strcasecmp(pg_mode, "shadow") != 0) {
+ V3_Print(vm, VCORE_NONE, "Unknown paging mode '%s' specified - falling back to shadow paging with defaults\n",pg_mode);
+ pg_mode = "shadow";
+ }
+ }
+ if (pg_strat == NULL) {
+ pg_strat = (char *)default_strategy;
+ }
+
+ V3_Print(vm, VCORE_NONE,"Initialization of Shadow Paging implementation\n");
+ impl = (struct v3_shdw_pg_impl *)v3_htable_search(master_shdw_pg_table, (addr_t)pg_strat);
-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, 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:
- default:
- PrintError("Unhandled CPU Mode\n");
- return -1;
+ if (impl == NULL) {
+ PrintError(vm, VCORE_NONE, "Could not find shadow paging impl (%s)\n", pg_strat);
+ return -1;
}
- } else {
- PrintError("Invalid Memory mode\n");
- return -1;
- }
-}
-addr_t create_new_shadow_pt32() {
- void * host_pde = 0;
+ INIT_LIST_HEAD(&(impl_state->event_callback_list));
+ v3_rw_lock_init(&(impl_state->event_callback_lock));
+
+ impl_state->current_impl = impl;
- host_pde = V3_AllocPages(1);
- memset(host_pde, 0, PAGE_SIZE);
+ if (impl->init(vm, pg_cfg) == -1) {
+ PrintError(vm, VCORE_NONE, "Could not initialize Shadow paging implemenation (%s)\n", impl->name);
+ return -1;
+ }
- return (addr_t)host_pde;
-}
-static void inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
- info->ctrl_regs.cr2 = fault_addr;
- v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
+ return 0;
}
+int v3_deinit_shdw_impl(struct v3_vm_info * vm) {
+ struct v3_shdw_pg_impl * impl = vm->shdw_impl.current_impl;
+ struct event_callback *cb,*temp;
+ addr_t flags;
-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
-
- whew...
- */
- if (guest_access != PT_ACCESS_OK) {
- // Guest Access Error
-
- if ((shadow_access != PT_ENTRY_NOT_PRESENT) &&
- (guest_access != PT_ENTRY_NOT_PRESENT)) {
- // aka (guest permission error)
- return 1;
+ if (impl == NULL) {
+ // Shadow paging not implemented
+ return 0;
}
- if ((shadow_access == PT_ENTRY_NOT_PRESENT) &&
- (guest_access == PT_ENTRY_NOT_PRESENT)) {
- // Page tables completely blank, handle guest first
- return 1;
+ if (impl->deinit(vm) == -1) {
+ PrintError(vm, VCORE_NONE,"Error deinitializing shadow paging implementation\n");
+ return -1;
}
- // Otherwise we'll handle the guest fault later...?
- }
-
- return 0;
-}
+ flags=v3_write_lock_irqsave(&(vm->shdw_impl.event_callback_lock));
+ list_for_each_entry_safe(cb,
+ temp,
+ &(vm->shdw_impl.event_callback_list),
+ node) {
+ list_del(&(cb->node));
+ V3_Free(cb);
+ }
+ v3_write_unlock_irqrestore(&(vm->shdw_impl.event_callback_lock),flags);
+ v3_rw_lock_deinit(&(vm->shdw_impl.event_callback_lock));
-/* The guest status checks have already been done,
- * only special case shadow checks remain
- */
-static int handle_large_pagefault32(struct guest_info * info,
- addr_t fault_addr, pf_error_t error_code,
- pte32_t * shadow_pt, pde32_4MB_t * large_guest_pde)
-{
- pt_access_status_t shadow_pte_access = can_access_pte32(shadow_pt, fault_addr, error_code);
- pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]);
-
- if (shadow_pte_access == PT_ACCESS_OK) {
- // Inconsistent state...
- // Guest Re-Entry will flush tables and everything should now workd
- PrintDebug("Inconsistent state... Guest re-entry should flush tlb\n");
return 0;
- }
+}
-
- if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) {
- // Get the guest physical address of the fault
- addr_t guest_fault_pa = PDE32_4MB_T_ADDR(*large_guest_pde) + PD32_4MB_PAGE_OFFSET(fault_addr);
- host_region_type_t host_page_type = get_shadow_addr_type(info, guest_fault_pa);
-
-
- if (host_page_type == HOST_REGION_INVALID) {
- // Inject a machine check in the guest
- PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_fault_pa);
- v3_raise_exception(info, MC_EXCEPTION);
- return 0;
- }
- if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
- struct shadow_page_state * state = &(info->shdw_pg_state);
- addr_t shadow_pa = get_shadow_addr(info, guest_fault_pa);
-
- shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa);
-
- shadow_pte->present = 1;
-
- /* We are assuming that the PDE entry has precedence
- * so the Shadow PDE will mirror the guest PDE settings,
- * and we don't have to worry about them here
- * Allow everything
- */
- shadow_pte->user_page = 1;
-
- if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_fault_pa)) != NULL) {
- // Check if the entry is a page table...
- PrintDebug("Marking page as Guest Page Table (large page)\n");
- shadow_pte->vmm_info = PT32_GUEST_PT;
- shadow_pte->writable = 0;
- } else {
- shadow_pte->writable = 1;
- }
-
-
- //set according to VMM policy
- shadow_pte->write_through = 0;
- shadow_pte->cache_disable = 0;
- shadow_pte->global_page = 0;
- //
-
+// 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;
+
+ if (!have_callbacks(core)) {
+ return impl->activate_shdw_pt(core);
} else {
- // Handle hooked pages as well as other special pages
- if (handle_special_page_fault(info, fault_addr, guest_fault_pa, error_code) == -1) {
- PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr);
- return -1;
- }
- }
- } else if ((shadow_pte_access == PT_WRITE_ERROR) &&
- (shadow_pte->vmm_info == PT32_GUEST_PT)) {
-
- struct shadow_page_state * state = &(info->shdw_pg_state);
- PrintDebug("Write operation on Guest PAge Table Page (large page)\n");
- state->cached_cr3 = 0;
- shadow_pte->writable = 1;
+ int rc;
+ struct v3_shdw_pg_event event_pre={SHADOW_ACTIVATE,SHADOW_PREIMPL,0,{0,0,0,0,0,0}};
+ struct v3_shdw_pg_event event_post={SHADOW_ACTIVATE,SHADOW_POSTIMPL,0,{0,0,0,0,0,0}};
+
+ dispatch_event(core,&event_pre);
- } else {
- PrintError("Error in large page fault handler...\n");
- PrintError("This case should have been handled at the top level handler\n");
- return -1;
- }
+ rc =impl->activate_shdw_pt(core);
- PrintDebug("Returning from large page fault handler\n");
- return 0;
+ dispatch_event(core,&event_post);
+
+ return rc;
+ }
}
-static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
- pde32_t * guest_pd = NULL;
- pde32_t * shadow_pd = (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 = NULL;
- pde32_t * shadow_pde = (pde32_t *)&(shadow_pd[PDE32_INDEX(fault_addr)]);
-
- PrintDebug("Shadow page fault handler\n");
- if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) {
- PrintError("Invalid Guest PDE Address: 0x%x\n", guest_cr3);
- return -1;
- }
-
- guest_pde = (pde32_t *)&(guest_pd[PDE32_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 * core) {
+ struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+ struct v3_shdw_pg_impl * impl = state->current_impl;
+ if (!have_callbacks(core)) {
+ return impl->invalidate_shdw_pt(core);
+ } else {
+ int rc;
+ struct v3_shdw_pg_event event_pre={SHADOW_INVALIDATE,SHADOW_PREIMPL,0,{0,0,0,0,0,0}};
+ struct v3_shdw_pg_event event_post={SHADOW_INVALIDATE,SHADOW_POSTIMPL,0,{0,0,0,0,0,0}};
+
+ dispatch_event(core,&event_pre);
- // Check the guest page permissions
- guest_pde_access = can_access_pde32(guest_pd, fault_addr, error_code);
-
- // Check the shadow page permissions
- shadow_pde_access = can_access_pde32(shadow_pd, fault_addr, error_code);
-
- /* Was the page fault caused by the Guest's page tables? */
- if (is_guest_pf(guest_pde_access, shadow_pde_access) == 1) {
- PrintDebug("Injecting PDE pf to guest: (guest access error=%d) (pf error code=%d)\n",
- guest_pde_access, error_code);
- inject_guest_pf(info, fault_addr, error_code);
- return 0;
- }
+ rc = impl->invalidate_shdw_pt(core);
-
- if (shadow_pde_access == PT_ENTRY_NOT_PRESENT)
- {
- pte32_t * shadow_pt = (pte32_t *)create_new_shadow_pt32();
-
- shadow_pde->present = 1;
- shadow_pde->user_page = guest_pde->user_page;
- // shadow_pde->large_page = guest_pde->large_page;
- shadow_pde->large_page = 0;
-
-
- // VMM Specific options
- shadow_pde->write_through = 0;
- shadow_pde->cache_disable = 0;
- shadow_pde->global_page = 0;
- //
-
- guest_pde->accessed = 1;
-
- shadow_pde->pt_base_addr = PD32_BASE_ADDR((addr_t)shadow_pt);
-
- if (guest_pde->large_page == 0) {
- shadow_pde->writable = guest_pde->writable;
- } else {
- ((pde32_4MB_t *)guest_pde)->dirty = 0;
- shadow_pde->writable = 0;
- }
- }
- else if (shadow_pde_access == PT_ACCESS_OK)
- {
- //
- // PTE fault
- //
- pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde));
-
- if (guest_pde->large_page == 0) {
- pte32_t * guest_pt = NULL;
- if (guest_pa_to_host_va(info, PDE32_T_ADDR((*guest_pde)), (addr_t*)&guest_pt) == -1) {
- // Machine check the guest
- PrintDebug("Invalid Guest PTE Address: 0x%x\n", PDE32_T_ADDR((*guest_pde)));
- v3_raise_exception(info, MC_EXCEPTION);
- return 0;
- }
+ dispatch_event(core,&event_post);
- if (handle_shadow_pte32_fault(info, fault_addr, error_code, shadow_pt, guest_pt) == -1) {
- PrintError("Error handling Page fault caused by PTE\n");
- return -1;
- }
- } else if (guest_pde->large_page == 1) {
- if (handle_large_pagefault32(info, fault_addr, error_code, shadow_pt, (pde32_4MB_t *)guest_pde) == -1) {
- PrintError("Error handling large pagefault\n");
- return -1;
- }
- }
+ return rc;
}
- else if ((shadow_pde_access == PT_WRITE_ERROR) &&
- (guest_pde->large_page == 1) &&
- (((pde32_4MB_t *)guest_pde)->dirty == 0))
- {
- //
- // Page Directory Entry marked read-only
- // Its a large page and we need to update the dirty bit in the guest
- //
-
- PrintDebug("Large page write error... Setting dirty bit and returning\n");
- ((pde32_4MB_t *)guest_pde)->dirty = 1;
- shadow_pde->writable = guest_pde->writable;
- return 0;
-
- }
- else if (shadow_pde_access == PT_USER_ERROR)
- {
- //
- // Page Directory Entry marked non-user
- //
- PrintDebug("Shadow Paging User access error (shadow_pde_access=0x%x, guest_pde_access=0x%x)\n",
- shadow_pde_access, guest_pde_access);
- inject_guest_pf(info, fault_addr, error_code);
- return 0;
- }
- else
- {
- // inject page fault in guest
- inject_guest_pf(info, fault_addr, error_code);
- PrintDebug("Unknown Error occurred (shadow_pde_access=%d)\n", shadow_pde_access);
- PrintDebug("Manual Says to inject page fault into guest\n");
-#ifdef DEBUG_SHADOW_PAGING
- PrintDebug("Guest PDE: (access=%d)\n\t", guest_pde_access);
- PrintPDE32(fault_addr, guest_pde);
- PrintDebug("Shadow PDE: (access=%d)\n\t", shadow_pde_access);
- PrintPDE32(fault_addr, shadow_pde);
-#endif
-
- return 0;
- }
-
- PrintDebug("Returning end of PDE function (rip=%x)\n", info->rip);
- return 0;
}
+int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code)
+{
+ int rc;
+
-/*
- * We assume the the guest pte pointer has already been translated to a host virtual address
- */
-static int handle_shadow_pte32_fault(struct guest_info * info,
- addr_t fault_addr,
- pf_error_t error_code,
- pte32_t * shadow_pt,
- pte32_t * guest_pt) {
-
- pt_access_status_t guest_pte_access;
- pt_access_status_t shadow_pte_access;
- pte32_t * guest_pte = (pte32_t *)&(guest_pt[PTE32_INDEX(fault_addr)]);;
- pte32_t * shadow_pte = (pte32_t *)&(shadow_pt[PTE32_INDEX(fault_addr)]);
-
-
- // Check the guest page permissions
- guest_pte_access = can_access_pte32(guest_pt, fault_addr, error_code);
-
- // Check the shadow page permissions
- shadow_pte_access = can_access_pte32(shadow_pt, fault_addr, error_code);
-
-#ifdef DEBUG_SHADOW_PAGING
- PrintDebug("Guest PTE: (access=%d)\n\t", guest_pte_access);
- PrintPTE32(fault_addr, guest_pte);
- PrintDebug("Shadow PTE: (access=%d)\n\t", shadow_pte_access);
- PrintPTE32(fault_addr, shadow_pte);
-#endif
-
- /* Was the page fault caused by the Guest's page tables? */
- if (is_guest_pf(guest_pte_access, shadow_pte_access) == 1) {
- PrintDebug("Access error injecting pf to guest (guest access error=%d) (pf error code=%d)\n",
- guest_pte_access, *(uint_t*)&error_code);
- inject_guest_pf(info, fault_addr, error_code);
- return 0;
- }
-
-
- 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;
- }
-
-
- if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) {
-
- addr_t guest_pa = PTE32_T_ADDR((*guest_pte)) + PT32_PAGE_OFFSET(fault_addr);
-
- // 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
- PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_pa);
- v3_raise_exception(info, MC_EXCEPTION);
- return 0;
+ if (have_callbacks(core)) {
+ struct v3_shdw_pg_event event={SHADOW_PAGEFAULT,SHADOW_PREIMPL,fault_addr,error_code};
+ dispatch_event(core,&event);
}
-
- // else...
-
- if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
- struct shadow_page_state * state = &(info->shdw_pg_state);
- addr_t shadow_pa = get_shadow_addr(info, guest_pa);
-
- shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa);
-
- shadow_pte->present = guest_pte->present;
- shadow_pte->user_page = guest_pte->user_page;
-
- //set according to VMM policy
- shadow_pte->write_through = 0;
- shadow_pte->cache_disable = 0;
- shadow_pte->global_page = 0;
- //
-
- guest_pte->accessed = 1;
-
- if (find_pte_map(state->cached_ptes, PT32_PAGE_ADDR(guest_pa)) != NULL) {
- // Check if the entry is a page table...
- PrintDebug("Marking page as Guest Page Table\n", shadow_pte->writable);
- shadow_pte->vmm_info = PT32_GUEST_PT;
- }
-
- if (guest_pte->dirty == 1) {
- shadow_pte->writable = guest_pte->writable;
- } else if ((guest_pte->dirty == 0) && (error_code.write == 1)) {
- shadow_pte->writable = guest_pte->writable;
- guest_pte->dirty = 1;
+
+ if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
+ // If paging is not turned on we need to handle the special cases
+ rc = v3_handle_passthrough_pagefault(core, fault_addr, error_code,NULL,NULL);
+ } 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;
- if (shadow_pte->vmm_info == PT32_GUEST_PT) {
- // Well that was quick...
- struct shadow_page_state * state = &(info->shdw_pg_state);
- PrintDebug("Immediate Write operation on Guest PAge Table Page\n");
- state->cached_cr3 = 0;
- }
-
- } else if ((guest_pte->dirty = 0) && (error_code.write == 0)) {
- shadow_pte->writable = 0;
- }
-
-
-
+ rc = impl->handle_pagefault(core, fault_addr, error_code);
} else {
- // Page fault handled by hook functions
- if (handle_special_page_fault(info, fault_addr, guest_pa, error_code) == -1) {
- PrintError("Special Page fault handler returned error for address: %x\n", fault_addr);
- return -1;
- }
+ PrintError(core->vm_info, core, "Invalid Memory mode\n");
+ rc = -1;
}
-
- } else if ((shadow_pte_access == PT_WRITE_ERROR) &&
- (guest_pte->dirty == 0)) {
-
- PrintDebug("Shadow PTE Write Error\n");
- guest_pte->dirty = 1;
- shadow_pte->writable = guest_pte->writable;
-
- if (shadow_pte->vmm_info == PT32_GUEST_PT) {
- struct shadow_page_state * state = &(info->shdw_pg_state);
- PrintDebug("Write operation on Guest PAge Table Page\n");
- state->cached_cr3 = 0;
+
+ if (have_callbacks(core)) {
+ struct v3_shdw_pg_event event={SHADOW_PAGEFAULT,SHADOW_POSTIMPL,fault_addr,error_code};
+ dispatch_event(core,&event);
}
- return 0;
-
- } else {
- // Inject page fault into the guest
- inject_guest_pf(info, fault_addr, error_code);
- PrintError("PTE Page fault fell through... Not sure if this should ever happen\n");
- PrintError("Manual Says to inject page fault into guest\n");
- return -1;
- }
-
- PrintDebug("Returning end of function\n");
- return 0;
+ return rc;
}
+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 (v3_get_vm_mem_mode(core) != VIRTUAL_MEM) {
+ // Paging must be turned on...
+ // should handle with some sort of fault I think
+ PrintError(core->vm_info, core, "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);
+ }
+ if (ret == -1) {
+ PrintError(core->vm_info, core, "Could not read instruction into buffer\n");
+ return -1;
+ }
+ if (v3_decode(core, (addr_t)instr, &dec_instr) == -1) {
+ PrintError(core->vm_info, core, "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(core->vm_info, core, "Decoder Error: Not a valid INVLPG instruction...\n");
+ return -1;
+ }
-/* 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
- PrintError("ERROR: INVLPG called in non paged mode\n");
- return -1;
- }
+ vaddr = dec_instr.dst_operand.operand;
+ core->rip += dec_instr.instr_length;
- if (info->cpu_mode == PROTECTED) {
- uchar_t instr[15];
- int ret;
- int index = 0;
+ {
+ struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+ struct v3_shdw_pg_impl * impl = state->current_impl;
+ int rc;
- ret = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
- if (ret != 15) {
- PrintError("Could not read instruction 0x%x (ret=%d)\n", info->rip, ret);
- return -1;
- }
+ if (have_callbacks(core)) {
+ struct v3_shdw_pg_event event={SHADOW_INVLPG,SHADOW_PREIMPL,vaddr,{0,0,0,0,0,0}};
+ dispatch_event(core,&event);
+ }
-
- /* Can INVLPG work with Segments?? */
- while (is_prefix_byte(instr[index])) {
- index++;
- }
-
-
- if ((instr[index] == (uchar_t)0x0f) &&
- (instr[index + 1] == (uchar_t)0x01)) {
+ rc=impl->handle_invlpg(core, vaddr);
- addr_t first_operand;
- addr_t second_operand;
- operand_type_t addr_type;
- addr_t guest_cr3 = CR3_TO_PDE32(info->shdw_pg_state.guest_cr3);
+ if (have_callbacks(core)) {
+ struct v3_shdw_pg_event event={SHADOW_INVLPG,SHADOW_POSTIMPL,vaddr,{0,0,0,0,0,0}};
+ dispatch_event(core,&event);
+ }
- pde32_t * guest_pd = NULL;
+ return rc;
+ }
+}
- if (guest_pa_to_host_va(info, guest_cr3, (addr_t*)&guest_pd) == -1) {
- PrintError("Invalid Guest PDE Address: 0x%x\n", guest_cr3);
- return -1;
- }
-
- 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 = (pde32_t *)&shadow_pd[PDE32_INDEX(first_operand)];
- pde32_t * guest_pde;
+int v3_inject_guest_pf(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
+ core->ctrl_regs.cr2 = fault_addr;
- //PrintDebug("PDE Index=%d\n", PDE32_INDEX(first_operand));
- //PrintDebug("FirstOperand = %x\n", first_operand);
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+ core->shdw_pg_state.guest_faults++;
+#endif
- PrintDebug("Invalidating page for %x\n", first_operand);
+ return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code);
+}
- guest_pde = (pde32_t *)&(guest_pd[PDE32_INDEX(first_operand)]);
- if (guest_pde->large_page == 1) {
- shadow_pde->present = 0;
- PrintDebug("Invalidating Large Page\n");
- } else {
-
- if (shadow_pde->present == 1) {
- pte32_t * shadow_pt = (pte32_t *)(addr_t)PDE32_T_ADDR((*shadow_pde));
- pte32_t * shadow_pte = (pte32_t *)&shadow_pt[PTE32_INDEX(first_operand)];
+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
-#ifdef DEBUG_SHADOW_PAGING
- PrintDebug("Setting not present\n");
- PrintPTE32(first_operand, shadow_pte);
-#endif
+ whew...
+ */
+ if (guest_access != PT_ACCESS_OK) {
+ // Guest Access Error
- shadow_pte->present = 0;
- }
+ if ((shadow_access != PT_ACCESS_NOT_PRESENT) &&
+ (guest_access != PT_ACCESS_NOT_PRESENT)) {
+ // aka (guest permission error)
+ return 1;
}
- info->rip += index;
+ /*
+ if ((shadow_access == PT_ACCESS_NOT_PRESENT) &&
+ (guest_access == PT_ACCESS_NOT_PRESENT)) {
+ // Page tables completely blank, handle guest first
+ return 1;
+ }
+ */
- } else {
- PrintError("Invalid Operand type\n");
- return -1;
- }
- } else {
- PrintError("invalid Instruction Opcode\n");
- PrintTraceMemDump(instr, 15);
- 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 0;
+ return 0;
}
-/*
-
-
-static int create_pd32_nonaligned_4MB_page(struct guest_info * info, pte32_t * pt, addr_t guest_addr, pde32_4MB_t * large_shadow_pde) {
- uint_t i = 0;
- pte32_t * pte_cursor;
- addr_t guest_pa = 0;
+int v3_register_shadow_paging_event_callback(struct v3_vm_info *vm,
+ int (*callback)(struct guest_info *core,
+ struct v3_shdw_pg_event *event,
+ void *priv_data),
+ void *priv_data)
+{
+ struct event_callback *ec = V3_Malloc(sizeof(struct event_callback));
+ addr_t flags;
- for (i = 0; i < 1024; i++) {
- guest_pa = guest_addr + (PAGE_SIZE * i);
- host_region_type_t host_page_type = get_shadow_addr_type(info, guest_pa);
+ if (!ec) {
+ PrintError(vm, VCORE_NONE, "Unable to allocate for a shadow paging event callback\n");
+ return -1;
+ }
- pte_cursor = &(pt[i]);
-
- if (host_page_type == HOST_REGION_INVALID) {
- // Currently we don't support this, but in theory we could
- PrintError("Invalid Host Memory Type\n");
- return -1;
- } else if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
- addr_t shadow_pa = get_shadow_addr(info, guest_pa);
+ ec->callback = callback;
+ ec->priv_data = priv_data;
+ flags=v3_write_lock_irqsave(&(vm->shdw_impl.event_callback_lock));
+ list_add(&(ec->node),&(vm->shdw_impl.event_callback_list));
+ v3_write_unlock_irqrestore(&(vm->shdw_impl.event_callback_lock),flags);
- pte_cursor->page_base_addr = PT32_BASE_ADDR(shadow_pa);
- pte_cursor->present = 1;
- pte_cursor->writable = large_shadow_pde->writable;
- pte_cursor->user_page = large_shadow_pde->user_page;
- pte_cursor->write_through = 0;
- pte_cursor->cache_disable = 0;
- pte_cursor->global_page = 0;
+ return 0;
- } else {
- PrintError("Unsupported Host Memory Type\n");
- return -1;
- }
- }
- return 0;
}
-
-static int handle_large_pagefault32(struct guest_info * info,
- pde32_t * guest_pde, pde32_t * shadow_pde,
- addr_t fault_addr, pf_error_t error_code ) {
- struct shadow_region * mem_reg;
- pde32_4MB_t * large_guest_pde = (pde32_4MB_t *)guest_pde;
- pde32_4MB_t * large_shadow_pde = (pde32_4MB_t *)shadow_pde;
- host_region_type_t host_page_type;
- addr_t guest_start_addr = PDE32_4MB_T_ADDR(*large_guest_pde);
- // addr_t guest_end_addr = guest_start_addr + PAGE_SIZE_4MB; // start address + 4MB
-
-
- // Check that the Guest PDE entry points to valid memory
- // else Machine Check the guest
- PrintDebug("Large Page: Page Base Addr=%x\n", guest_start_addr);
-
- host_page_type = get_shadow_addr_type(info, guest_start_addr);
-
- if (host_page_type == HOST_REGION_INVALID) {
- PrintError("Invalid guest address in large page (0x%x)\n", guest_start_addr);
- v3_raise_exception(info, MC_EXCEPTION);
- return -1;
- }
-
- // else...
-
- if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) {
-
- addr_t host_start_addr = 0;
- addr_t region_end_addr = 0;
-
- // Check for a large enough region in host memory
- mem_reg = get_shadow_region_by_addr(&(info->mem_map), guest_start_addr);
- PrintDebug("Host region: host_addr=%x (guest_start=%x, end=%x)\n",
- mem_reg->host_addr, mem_reg->guest_start, mem_reg->guest_end);
- host_start_addr = mem_reg->host_addr + (guest_start_addr - mem_reg->guest_start);
- region_end_addr = mem_reg->host_addr + (mem_reg->guest_end - mem_reg->guest_start);
-
- PrintDebug("Host Start Addr=%x; Region End Addr=%x\n", host_start_addr, region_end_addr);
-
-
- //4f
- if (large_guest_pde->dirty == 1) { // dirty
- large_shadow_pde->writable = guest_pde->writable;
- } else if (error_code.write == 1) { // not dirty, access is write
- large_shadow_pde->writable = guest_pde->writable;
- large_guest_pde->dirty = 1;
- } else { // not dirty, access is read
- large_shadow_pde->writable = 0;
- }
-
-
- // Check if the region is at least an additional 4MB
-
-
- //4b.
- if ((PD32_4MB_PAGE_OFFSET(host_start_addr) == 0) &&
- (region_end_addr >= host_start_addr + PAGE_SIZE_4MB)) { // if 4MB boundary
- large_shadow_pde->page_base_addr = PD32_4MB_BASE_ADDR(host_start_addr);
- } else { // else generate 4k pages
- pte32_t * shadow_pt = NULL;
- PrintDebug("Handling non aligned large page\n");
-
- shadow_pde->large_page = 0;
-
- shadow_pt = create_new_shadow_pt32();
-
- if (create_pd32_nonaligned_4MB_page(info, shadow_pt, guest_start_addr, large_shadow_pde) == -1) {
- PrintError("Non Aligned Large Page Error\n");
- V3_Free(shadow_pt);
- return -1;
- }
-
-
-#ifdef DEBUG_SHADOW_PAGING
- PrintDebug("non-aligned Shadow PT\n");
- PrintPT32(PT32_PAGE_ADDR(fault_addr), shadow_pt);
-#endif
- shadow_pde->pt_base_addr = PD32_BASE_ADDR(shadow_pt);
+int v3_unregister_shadow_paging_event_callback(struct v3_vm_info *vm,
+ int (*callback)(struct guest_info *core,
+ struct v3_shdw_pg_event *event,
+ void *priv_data),
+ void *priv_data)
+{
+ struct event_callback *cb,*temp;
+ addr_t flags;
+
+ flags=v3_write_lock_irqsave(&(vm->shdw_impl.event_callback_lock));
+
+ list_for_each_entry_safe(cb,
+ temp,
+ &(vm->shdw_impl.event_callback_list),
+ node) {
+ if ((callback == cb->callback) && (priv_data == cb->priv_data)) {
+ list_del(&(cb->node));
+ v3_write_unlock_irqrestore(&(vm->shdw_impl.event_callback_lock),flags);
+ V3_Free(cb);
+ return 0;
+ }
}
- } else {
- // Handle hooked pages as well as other special pages
- if (handle_special_page_fault(info, fault_addr, guest_start_addr, error_code) == -1) {
- PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr);
- return -1;
- }
- }
+ v3_write_unlock_irqrestore(&(vm->shdw_impl.event_callback_lock),flags);
- return 0;
+ PrintError(vm, VCORE_NONE, "No callback found!\n");
+
+ return -1;
}
-*/
+
+
