X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_shadow_paging.c;h=65d19ee6fd49af042d66014bed43d2b66645b748;hb=862392a8cadf9e2a2fa0e395bb33835356ad0f36;hp=c9546e4d30bfece0df576771d146f085bba7e568;hpb=a109eb919a162bd7de58d62020801bc2e633be50;p=palacios.git diff --git a/palacios/src/palacios/vmm_shadow_paging.c b/palacios/src/palacios/vmm_shadow_paging.c index c9546e4..65d19ee 100644 --- a/palacios/src/palacios/vmm_shadow_paging.c +++ b/palacios/src/palacios/vmm_shadow_paging.c @@ -1,209 +1,322 @@ +/* + * 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 + * Copyright (c) 2008, The V3VEE Project + * All rights reserved. + * + * Author: Jack Lange + * + * This is free software. You are permitted to use, + * redistribute, and modify it as specified in the file "V3VEE_LICENSE". + */ + + #include + #include #include +#include +#include -extern struct vmm_os_hooks * os_hooks; +#include +#include + + + + +#ifdef CONFIG_SHADOW_PAGING_TELEMETRY +#include +#endif + +#ifdef CONFIG_SYMBIOTIC_SWAP +#include +#endif + +#ifndef CONFIG_DEBUG_SHADOW_PAGING +#undef PrintDebug +#define PrintDebug(fmt, args...) +#endif + + +static const char default_strategy[] = "VTLB"; -int init_shadow_page_state(struct shadow_page_state * state) { - state->guest_mode = PDE32; - state->shadow_mode = PDE32; - - state->guest_cr3.r_reg = 0; - state->shadow_cr3.r_reg = 0; - return 0; +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)); } - -int wholesale_update_shadow_page_state(struct guest_info * guest_info) { - unsigned i, j; - pde32_t * guest_pde; - pde32_t * shadow_pde; +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; - struct shadow_page_state * state = &(guest_info->shdw_pg_state); + master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn); - // For now, we'll only work with PDE32 - if (state->guest_mode != PDE32) { - return -1; - } + while (tmp_impl != __stop__v3_shdw_pg_impls) { + V3_Print("Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name); - shadow_pde = (pde32_t *)(CR3_TO_PDE32(state->shadow_cr3.e_reg.low)); + 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; + } - if (host_pa_to_host_va(CR3_TO_PDE32(state->guest_cr3.e_reg.low), (addr_t*)&guest_pde) != 0) { - return -1; - } + tmp_impl = &(__start__v3_shdw_pg_impls[++i]); + } + + return 0; +} - // Delete the current page table - delete_page_tables_pde32(shadow_pde); - shadow_pde = os_hooks->allocate_pages(1); - state->shadow_cr3.e_reg.low = (addr_t)shadow_pde; +/*** + *** There be dragons + ***/ - state->shadow_mode = PDE32; - for (i = 0; i < MAX_PDE32_ENTRIES; i++) { - shadow_pde[i] = guest_pde[i]; +#ifdef 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]); - // The shadow can be identical to the guest if it's not present - if (!shadow_pde[i].present) { - continue; + V3_Print("%s Guest Page faults: %d\n", hdr, core->shdw_pg_state.guest_faults); } +} +#endif + + + +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; - if (shadow_pde[i].large_pages) { - // large page - just map it through shadow map to generate its physical location - addr_t guest_addr = PAGE_ADDR(shadow_pde[i].pt_base_addr); - addr_t host_addr; - shadow_region_t * ent; - - ent = get_shadow_region_by_addr(&(guest_info->mem_map), guest_addr); - - if (!ent) { - // FIXME Panic here - guest is trying to map to physical memory - // it does not own in any way! + + if (impl->local_init(core) == -1) { + PrintError("Error in Shadow paging local initialization (%s)\n", impl->name); return -1; - } - - // FIXME Bounds check here to see if it's trying to trick us - - switch (ent->host_type) { - case HOST_REGION_PHYSICAL_MEMORY: - // points into currently allocated physical memory, so we just - // set up the shadow to point to the mapped location - if (guest_pa_to_host_pa(guest_info, guest_addr, &host_addr)) { - // Panic here - return -1; - } + } + + +#ifdef CONFIG_SHADOW_PAGING_TELEMETRY + v3_add_telemetry_cb(core->vm_info, telemetry_cb, NULL); +#endif + + 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("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); - shadow_pde[i].pt_base_addr = PAGE_ALIGNED_ADDR(host_addr); - // FIXME set vmm_info bits here - break; - case HOST_REGION_UNALLOCATED: - // points to physical memory that is *allowed* but that we - // have not yet allocated. We mark as not present and set a - // bit to remind us to allocate it later - shadow_pde[i].present = 0; - // FIXME Set vminfo bits here so that we know that we will be - // allocating it later - break; - case HOST_REGION_NOTHING: - // points to physical memory that is NOT ALLOWED. - // We will mark it as not present and set a bit to remind - // us that it's bad later and insert a GPF then - shadow_pde[i].present = 0; - break; - case HOST_REGION_MEMORY_MAPPED_DEVICE: - case HOST_REGION_REMOTE: - case HOST_REGION_SWAPPED: - default: - // Panic. Currently unhandled + if (impl == NULL) { + PrintError("Could not find shadow paging impl (%s)\n", pg_strat); return -1; - break; - } - } else { - pte32_t * guest_pte; - pte32_t * shadow_pte; - addr_t guest_addr; - addr_t guest_pte_host_addr; - shadow_region_t * ent; + } + + 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; +} - // small page - set PDE and follow down to the child table - shadow_pde[i] = guest_pde[i]; - guest_addr = PAGE_ADDR(guest_pde[i].pt_base_addr); +// 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); +} + - // Allocate a new second level page table for the shadow - shadow_pte = os_hooks->allocate_pages(1); - // make our first level page table in the shadow point to it - shadow_pde[i].pt_base_addr = PAGE_ALIGNED_ADDR(shadow_pte); - - ent = get_shadow_region_by_addr(&(guest_info->mem_map), guest_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; + return impl->invalidate_shdw_pt(core); +} - /* JRL: This is bad.... */ - // For now the guest Page Table must always be mapped to host physical memory - /* If we swap out a page table or if it isn't present for some reason, this turns real ugly */ - if ((!ent) || (ent->host_type != HOST_REGION_PHYSICAL_MEMORY)) { - // FIXME Panic here - guest is trying to map to physical memory - // it does not own in any way! +int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) { + + 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; - } + } +} + + +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; - // Address of the relevant second level page table in the guest - if (guest_pa_to_host_pa(guest_info, guest_addr, &guest_pte_host_addr)) { - // Panic here + 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_addr now contains the host physical address for the guest's 2nd level page table - // Now we transform it to relevant virtual address - guest_pte = os_hooks->paddr_to_vaddr((void *)guest_pte_host_addr); + if (ret == -1) { + PrintError("Could not read instruction into buffer\n"); + return -1; + } - // Now we walk through the second level guest page table - // and clone it into the shadow - for (j = 0; j < MAX_PTE32_ENTRIES; j++) { - shadow_pte[j] = guest_pte[j]; + 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; + } - addr_t guest_addr = PAGE_ADDR(shadow_pte[j].page_base_addr); - - shadow_region_t * ent; - - ent = get_shadow_region_by_addr(&(guest_info->mem_map), guest_addr); - - if (!ent) { - // FIXME Panic here - guest is trying to map to physical memory - // it does not own in any way! - return -1; + vaddr = dec_instr.dst_operand.operand; + + core->rip += dec_instr.instr_length; + + { + 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); + } +} + + + + + + +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 CONFIG_SHADOW_PAGING_TELEMETRY + core->shdw_pg_state.guest_faults++; +#endif + + return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code); +} + + +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; } - switch (ent->host_type) { - case HOST_REGION_PHYSICAL_MEMORY: - { - addr_t host_addr; - - // points into currently allocated physical memory, so we just - // set up the shadow to point to the mapped location - if (guest_pa_to_host_pa(guest_info, guest_addr, &host_addr)) { - // Panic here - return -1; - } - - shadow_pte[j].page_base_addr = PAGE_ALIGNED_ADDR(host_addr); - // FIXME set vmm_info bits here - break; + /* + if ((shadow_access == PT_ACCESS_NOT_PRESENT) && + (guest_access == PT_ACCESS_NOT_PRESENT)) { + // Page tables completely blank, handle guest first + return 1; } - case HOST_REGION_UNALLOCATED: - // points to physical memory that is *allowed* but that we - // have not yet allocated. We mark as not present and set a - // bit to remind us to allocate it later - shadow_pte[j].present = 0; - // FIXME Set vminfo bits here so that we know that we will be - // allocating it later - break; - case HOST_REGION_NOTHING: - // points to physical memory that is NOT ALLOWED. - // We will mark it as not present and set a bit to remind - // us that it's bad later and insert a GPF then - shadow_pte[j].present = 0; - break; - case HOST_REGION_MEMORY_MAPPED_DEVICE: - case HOST_REGION_REMOTE: - case HOST_REGION_SWAPPED: - default: - // Panic. Currently unhandled - return -1; - break; + */ + + 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; } - +