Palacios Public Git Repository

To checkout Palacios execute

  git clone http://v3vee.org/palacios/palacios.web/palacios.git
This will give you the master branch. You probably want the devel branch or one of the release branches. To switch to the devel branch, simply execute
  cd palacios
  git checkout --track -b devel origin/devel
The other branches are similar.


build fixes to merge the Palacios configuration parameters with Linux parameters.
[palacios.git] / palacios / src / palacios / vmm_shadow_paging.c
index 91004f9..9743073 100644 (file)
 #include <palacios/vmm_direct_paging.h>
 
 
-// set this to 1 if you want us to attempt to
-// fetch multiple entries on a page fault
-#define SPECULATIVE_PAGING 1
 
-#define REGULAR_PAGE_FAULT 0
-#define SPECULATIVE_PAGE_FAULT 1
 
-#ifndef DEBUG_SHADOW_PAGING
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+#include <palacios/vmm_telemetry.h>
+#endif
+
+#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
 
 
-/*** 
- ***  There be dragons
- ***/
+static const char default_strategy[] = "VTLB";
 
 
-struct shadow_page_data {
-    v3_reg_t cr3;
-    addr_t page_pa;
-  
-    struct list_head page_list_node;
-};
+static struct hashtable * master_shdw_pg_table = NULL;
+
+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;
+
+    return (strcmp(name1, name2) == 0);
+}
+
+
+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;
 
+    master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn);
 
-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);
 
+    while (tmp_impl != __stop__v3_shdw_pg_impls) {
+       V3_Print("Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name);
+
+       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;
+       }
+
+       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;
+       }
 
+       tmp_impl = &(__start__v3_shdw_pg_impls[++i]);
+    }
 
-static uint_t pte_hash_fn(addr_t key) {
-    return hash_long(key, 32);
+    return 0;
 }
 
-static int pte_equals(addr_t key1, addr_t key2) {
-    return (key1 == key2);
+int V3_deinit_shdw_paging() {
+    v3_free_htable(master_shdw_pg_table, 0, 0);
+    return 0;
 }
 
-static struct shadow_page_data * create_new_shadow_pt(struct guest_info * info);
-static void 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);
 
 
-#include "vmm_shadow_paging_32.h"
-#include "vmm_shadow_paging_32pae.h"
-#include "vmm_shadow_paging_64.h"
+/*** 
+ ***  There be dragons
+ ***/
+
+
+#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
 
 
 
-int v3_init_shadow_page_state(struct guest_info * info) {
-    struct shadow_page_state * state = &(info->shdw_pg_state);
+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;
   
+
     state->guest_cr3 = 0;
     state->guest_cr0 = 0;
+    state->guest_efer.value = 0x0LL;
 
-    INIT_LIST_HEAD(&(state->page_list));
 
-    state->cached_ptes = NULL;
-    state->cached_cr3 = 0;
+    if (impl->local_init(core) == -1) {
+       PrintError("Error in Shadow paging local initialization (%s)\n", impl->name);
+       return -1;
+    }
+
+
+#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 = core->vm_info->shdw_impl.current_impl;
 
-// 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_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_cpu_mode(info)));
-           return -1;
+    if (impl->local_deinit(core) == -1) {
+       PrintError("Error deinitializing shadow paging state\n");
+       return -1;
     }
 
+#ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
+    v3_remove_telemetry_cb(core->vm_info, telemetry_cb, NULL);
+#endif
+
     return 0;
 }
 
 
-int v3_activate_passthrough_pt(struct guest_info * info) {
-    // For now... But we need to change this....
-    // As soon as shadow paging becomes active the passthrough tables are hosed
-    // So this will cause chaos if it is called at that time
 
-    info->ctrl_regs.cr3 = *(addr_t*)&(info->direct_map_pt);
-    //PrintError("Activate Passthrough Page tables not implemented\n");
+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;
+    }
+
+    if (pg_strat == NULL) {
+       pg_strat = (char *)default_strategy;
+    }
+       
+    V3_Print("Initialization of Shadow Paging implementation\n");
+
+    impl = (struct v3_shdw_pg_impl *)v3_htable_search(master_shdw_pg_table, (addr_t)pg_strat);
+
+    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;
 }
 
+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;
+    }
+
+    return 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;
+    return impl->activate_shdw_pt(core);
+}
+
 
 
-int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
-  int rc;
+// 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);
+}
+
+
+int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
   
-    if (v3_get_mem_mode(info) == PHYSICAL_MEM) {
+    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(info, fault_addr, error_code);
-    } else if (v3_get_mem_mode(info) == VIRTUAL_MEM) {
-
-       switch (v3_get_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: {
-             addr_t curr_addr;
-             addr_t fault_addr_base;
-             // first, we will handle the actual fault, non-speculatively
-             rc=handle_shadow_pagefault_64(info, fault_addr, error_code, REGULAR_PAGE_FAULT);
-             if (rc) {
-               return -1;
-             } 
-             if (!SPECULATIVE_PAGING) { 
-               return 0;
-             }
-             fault_addr_base=PAGE_ADDR_4KB(fault_addr);
-             PrintDebug("Attempting speculative paging around %p\n",(void*)fault_addr_base);
-             for (curr_addr = (fault_addr_base & (~0x1fffffLL)) ; 
-                  curr_addr < (fault_addr_base | (0x1fffffLL)) ;
-                  curr_addr+=PAGE_SIZE) {
-               if (curr_addr!=fault_addr_base) { 
-                 rc=handle_shadow_pagefault_64(info, curr_addr, error_code, SPECULATIVE_PAGE_FAULT);
-                 if (rc) {
-                   PrintDebug("Speculative page fault handler failed at %p\n",(void*)curr_addr);
-                   return -1;
-                 } 
-               }
-             }
-             return 0;
-           }
-               break;
-           default:
-               PrintError("Unhandled CPU Mode: %s\n", v3_cpu_mode_to_str(v3_get_cpu_mode(info)));
-               return -1;
-       }
+       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;
@@ -187,23 +246,23 @@ int v3_handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_e
 }
 
 
-int v3_handle_shadow_invlpg(struct guest_info * info) {
+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_mem_mode(info) != VIRTUAL_MEM) {
+    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_mem_mode(info) == PHYSICAL_MEM) { 
-       ret = read_guest_pa_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
+    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 = read_guest_va_memory(info, get_addr_linear(info, info->rip, &(info->segments.cs)), 15, instr);
+       ret = v3_read_gva_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
     }
 
     if (ret == -1) {
@@ -211,7 +270,7 @@ int v3_handle_shadow_invlpg(struct guest_info * info) {
        return -1;
     }
 
-    if (v3_decode(info, (addr_t)instr, &dec_instr) == -1) {
+    if (v3_decode(core, (addr_t)instr, &dec_instr) == -1) {
        PrintError("Decoding Error\n");
        return -1;
     }
@@ -225,80 +284,33 @@ int v3_handle_shadow_invlpg(struct guest_info * info) {
 
     vaddr = dec_instr.dst_operand.operand;
 
-    info->rip += dec_instr.instr_length;
-
-    switch (v3_get_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_cpu_mode(info)));
-           return -1;
-    }
-}
-
-
-
-
-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;
+    core->rip += dec_instr.instr_length;
 
-    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));
+    {
+       struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
+       struct v3_shdw_pg_impl * impl = state->current_impl;
 
-           list_move(&(page_tail->page_list_node), &(state->page_list));
-
-           memset(V3_VAddr((void *)(page_tail->page_pa)), 0, PAGE_SIZE_4KB);
-
-
-           return page_tail;
-       }
+       return impl->handle_invlpg(core, vaddr);
     }
+}
 
-    // 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);
 
-    page_tail->cr3 = cur_cr3;
-    list_add(&(page_tail->page_list_node), &(state->page_list));
 
-    shdw_page = (addr_t)V3_VAddr((void *)(page_tail->page_pa));
-    memset((void *)shdw_page, 0, PAGE_SIZE_4KB);
 
-    return page_tail;
-}
+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
 
-static void inject_guest_pf(struct guest_info * info, addr_t fault_addr, pf_error_t error_code) {
-    if (info->enable_profiler) {
-       info->profiler.guest_pf_cnt++;
-    }
-
-    info->ctrl_regs.cr2 = fault_addr;
-    v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code);
+    return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code);
 }
 
 
-static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) {
+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,
@@ -317,12 +329,19 @@ static int is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shado
            return 1;
        }
 
-       if ((shadow_access == PT_ACCESS_NOT_PRESENT) &&
-           (guest_access == PT_ACCESS_NOT_PRESENT)) {
+       /*
+         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...?
     }