From: Jack Lange Date: Sat, 2 Aug 2008 21:42:35 +0000 (+0000) Subject: started the emulation framework X-Git-Tag: vmmhack1-ramdisk-boot-iso-puppy~28 X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?p=palacios.git;a=commitdiff_plain;h=a5c5675571882a9b8a7594ef07fe303b195ef9ae started the emulation framework cleaned up the shadow paging handlers --- diff --git a/misc/decoder_test/vmm_xed.c b/misc/decoder_test/vmm_xed.c index e48a1b8..bb5729b 100644 --- a/misc/decoder_test/vmm_xed.c +++ b/misc/decoder_test/vmm_xed.c @@ -11,6 +11,7 @@ #include #include + #endif static xed_state_t decoder_state; @@ -115,7 +116,8 @@ int init_decoder() { } -int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * size, uint_t * instr_length) { + +int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, struct basic_instr_info * instr_info) { xed_decoded_inst_t xed_instr; xed_error_enum_t xed_error; @@ -137,7 +139,7 @@ int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * siz return -1; } - *instr_length = xed_decoded_inst_get_length(&xed_instr); + instr_info->instr_length = xed_decoded_inst_get_length(&xed_instr); if (xed_decoded_inst_number_of_memory_operands(&xed_instr) == 0) { @@ -145,7 +147,22 @@ int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * siz return -1; } - *size = xed_decoded_inst_get_memory_operand_length(&xed_instr,0); + instr_info->op_size = xed_decoded_inst_get_memory_operand_length(&xed_instr, 0); + + + xed_category_enum_t cat = xed_decoded_inst_get_category(&xed_instr); + if (cat == XED_CATEGORY_STRINGOP) { + instr_info->str_op = 1; + } else { + instr_info->str_op = 0; + } + + xed_operand_values_t * operands = xed_decoded_inst_operands(&xed_instr); + if (xed_operand_values_has_real_rep(operands)) { + instr_info->has_rep = 1; + } else { + instr_info->has_rep = 0; + } return 0; } diff --git a/palacios/build/Makefile b/palacios/build/Makefile index 594c11d..f057439 100644 --- a/palacios/build/Makefile +++ b/palacios/build/Makefile @@ -1,6 +1,6 @@ # Makefile for GeekOS kernel, userspace, and tools # Copyright (c) 2004,2005 David H. Hovemeyer -# $Revision: 1.54 $ +# $Revision: 1.55 $ # This is free software. You are permitted to use, # redistribute, and modify it as specified in the file "COPYING". @@ -189,7 +189,7 @@ VMM_C_SRCS := vm_guest.c \ vmm_shadow_paging.c vm_guest_mem.c \ vm_dev.c vmm_dev_mgr.c vmm_decoder.c \ svm_halt.c svm_pause.c vmm_config.c vmm_hashtable.c \ - vmm_string.c \ + vmm_string.c vmm_emulator.c \ $(DECODER_SRCS) #\ # vmx.c vmcs_gen.c vmcs.c diff --git a/palacios/include/palacios/vm_guest.h b/palacios/include/palacios/vm_guest.h index cc53166..79ba15d 100644 --- a/palacios/include/palacios/vm_guest.h +++ b/palacios/include/palacios/vm_guest.h @@ -11,7 +11,7 @@ #include #include #include - +#include typedef ullong_t v3_reg_t; @@ -71,6 +71,7 @@ struct v3_segments { struct shadow_page_state; struct shadow_map; struct vmm_io_map; +struct emulation_state; struct vm_ctrl_ops { @@ -123,7 +124,7 @@ struct guest_info { struct vm_ctrl_ops vm_ops; - + struct emulation_state emulator; vm_operating_mode_t run_state; void * vmm_data; diff --git a/palacios/include/palacios/vmm_decoder.h b/palacios/include/palacios/vmm_decoder.h index 2617913..e5232b8 100644 --- a/palacios/include/palacios/vmm_decoder.h +++ b/palacios/include/palacios/vmm_decoder.h @@ -47,6 +47,15 @@ struct x86_instr { }; +struct basic_instr_info { + uint_t instr_length; + uint_t op_size; + uint_t str_op : 1; + uint_t has_rep : 1; +}; + + + /************************/ /* EXTERNAL DECODER API */ /************************/ @@ -84,11 +93,12 @@ int v3_encode(struct guest_info * info, struct x86_instr * instr, char * instr_b * Gets the operand size for a memory operation * */ -int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * size, uint_t * instr_len); - +int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, struct basic_instr_info * instr_info); +/* Removes a rep prefix in place */ +void strip_rep_prefix(uchar_t * instr, int length); diff --git a/palacios/include/palacios/vmm_dev_mgr.h b/palacios/include/palacios/vmm_dev_mgr.h index 931bdd3..11b0d28 100644 --- a/palacios/include/palacios/vmm_dev_mgr.h +++ b/palacios/include/palacios/vmm_dev_mgr.h @@ -1,6 +1,8 @@ #ifndef _VMM_DEV_MGR #define _VMM_DEV_MGR +#ifdef __V3VEE__ + #include #include #include @@ -36,7 +38,7 @@ int v3_unattach_device(struct vm_device *dev); -#ifdef __V3VEE__ + struct dev_io_hook { ushort_t port; diff --git a/palacios/include/palacios/vmm_mem.h b/palacios/include/palacios/vmm_mem.h index b5feefd..3e418ec 100644 --- a/palacios/include/palacios/vmm_mem.h +++ b/palacios/include/palacios/vmm_mem.h @@ -141,8 +141,7 @@ int unhook_guest_mem(struct guest_info * info, addr_t guest_addr); -int mem_hook_dispatch(struct guest_info * info, addr_t mem_addr, pf_error_t access_info, struct vmm_mem_hook * hook); -int handle_special_page_fault(struct guest_info * info, addr_t mem_addr, pf_error_t access_info); +int handle_special_page_fault(struct guest_info * info, addr_t fault_addr, addr_t gp_addr, pf_error_t access_info); #endif // ! __V3VEE__ diff --git a/palacios/include/palacios/vmm_shadow_paging.h b/palacios/include/palacios/vmm_shadow_paging.h index ac20544..779f323 100644 --- a/palacios/include/palacios/vmm_shadow_paging.h +++ b/palacios/include/palacios/vmm_shadow_paging.h @@ -35,7 +35,7 @@ struct guest_info; int init_shadow_page_state(struct guest_info * info); -addr_t create_new_shadow_pt32(struct guest_info * info); +addr_t create_new_shadow_pt32(); int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_error_t error_code); int handle_shadow_invlpg(struct guest_info * info); diff --git a/palacios/src/palacios/vmm_config.c b/palacios/src/palacios/vmm_config.c index bdb471f..8e80bfd 100644 --- a/palacios/src/palacios/vmm_config.c +++ b/palacios/src/palacios/vmm_config.c @@ -2,6 +2,7 @@ #include #include + #include #include #include @@ -37,6 +38,8 @@ int config_guest(struct guest_info * info, void * config_ptr) { init_interrupt_state(info); dev_mgr_init(info); + + init_emulator(info); // SerialPrint("Guest Mem Dump at 0x%x\n", 0x100000); diff --git a/palacios/src/palacios/vmm_ctrl_regs.c b/palacios/src/palacios/vmm_ctrl_regs.c index 6473e4a..2e566ef 100644 --- a/palacios/src/palacios/vmm_ctrl_regs.c +++ b/palacios/src/palacios/vmm_ctrl_regs.c @@ -92,7 +92,7 @@ int handle_cr0_write(struct guest_info * info) { struct cr0_32 *real_cr0 = (struct cr0_32*)&(info->ctrl_regs.cr0); struct cr0_32 *new_cr0= (struct cr0_32 *)(dec_instr.src_operand.operand); - PrintDebug("OperandVal = %x, length=%d\n", *new_cr0, dec_instr.dst_operand.size); + PrintDebug("OperandVal = %x, length=%d\n", *new_cr0, dec_instr_src_operand.size); PrintDebug("Old CR0=%x\n", *real_cr0); @@ -253,7 +253,7 @@ int handle_cr3_write(struct guest_info * info) { delete_page_tables_pde32((pde32_t *)CR3_TO_PDE32(*(uint_t*)shadow_cr3)); - shadow_pt = create_new_shadow_pt32(info); + shadow_pt = create_new_shadow_pt32(); shadow_cr3->pdt_base_addr = PD32_BASE_ADDR(shadow_pt); } diff --git a/palacios/src/palacios/vmm_decoder.c b/palacios/src/palacios/vmm_decoder.c index 6d1299a..10fb2dc 100644 --- a/palacios/src/palacios/vmm_decoder.c +++ b/palacios/src/palacios/vmm_decoder.c @@ -8,3 +8,23 @@ int opcode_cmp(const uchar_t * op1, const uchar_t * op2) { return memcmp(op1 + 1, op2 + 1, op1[0]); } } + + +void strip_rep_prefix(uchar_t * instr, int length) { + int read_ctr = 0; + int write_ctr = 0; + int found = 0; + + while (read_ctr < length) { + if ((!found) && + ( (instr[read_ctr] == 0xF2) || + (instr[read_ctr] == 0xF3))) { + read_ctr++; + found = 1; + } else { + instr[write_ctr] = instr[read_ctr]; + write_ctr++; + read_ctr++; + } + } +} diff --git a/palacios/src/palacios/vmm_mem.c b/palacios/src/palacios/vmm_mem.c index 36d866a..7480da7 100644 --- a/palacios/src/palacios/vmm_mem.c +++ b/palacios/src/palacios/vmm_mem.c @@ -70,7 +70,8 @@ struct vmm_mem_hook * get_mem_hook(struct guest_info * info, addr_t guest_addr) } -int mem_hook_dispatch(struct guest_info * info, addr_t mem_addr, pf_error_t access_info, struct vmm_mem_hook * hook) { +/* mem_addr is the guest physical memory address */ +static int mem_hook_dispatch(struct guest_info * info, addr_t fault_addr, addr_t guest_phys_page, pf_error_t access_info, struct vmm_mem_hook * hook) { // emulate and then dispatch // or dispatch and emulate @@ -79,10 +80,11 @@ int mem_hook_dispatch(struct guest_info * info, addr_t mem_addr, pf_error_t acce if (access_info.write == 1) { void * src = NULL; uint_t length = 0; + PrintDebug("Memory hook write\n"); return -1; - if (hook->write(mem_addr, src, length, hook->priv_data) != length) { + if (hook->write(fault_addr, src, length, hook->priv_data) != length) { return -1; } } else { @@ -94,12 +96,12 @@ int mem_hook_dispatch(struct guest_info * info, addr_t mem_addr, pf_error_t acce } -int handle_special_page_fault(struct guest_info * info, addr_t mem_addr, pf_error_t access_info) { - struct shadow_region * reg = get_shadow_region_by_addr(&(info->mem_map), mem_addr); +int handle_special_page_fault(struct guest_info * info, addr_t fault_addr, addr_t guest_phys_page, pf_error_t access_info) { + struct shadow_region * reg = get_shadow_region_by_addr(&(info->mem_map), guest_phys_page); switch (reg->host_type) { case HOST_REGION_HOOK: - return mem_hook_dispatch(info, mem_addr, access_info, (struct vmm_mem_hook *)(reg->host_addr)); + return mem_hook_dispatch(info, fault_addr, guest_phys_page, access_info, (struct vmm_mem_hook *)(reg->host_addr)); default: return -1; } @@ -199,7 +201,7 @@ int delete_shadow_region(struct shadow_map * map, struct shadow_region *get_shadow_region_by_index(struct shadow_map * map, - uint_t index) { + uint_t index) { struct shadow_region * reg = map->head; uint_t i = 0; @@ -215,7 +217,7 @@ struct shadow_region *get_shadow_region_by_index(struct shadow_map * map, struct shadow_region * get_shadow_region_by_addr(struct shadow_map * map, - addr_t addr) { + addr_t addr) { struct shadow_region * reg = map->head; while (reg) { diff --git a/palacios/src/palacios/vmm_shadow_paging.c b/palacios/src/palacios/vmm_shadow_paging.c index 0097853..9e70bb2 100644 --- a/palacios/src/palacios/vmm_shadow_paging.c +++ b/palacios/src/palacios/vmm_shadow_paging.c @@ -11,6 +11,9 @@ #endif +/*** + *** There be dragons + ***/ @@ -38,7 +41,7 @@ int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_erro 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); + return handle_special_page_fault(info, fault_addr, fault_addr, error_code); } else if (info->mem_mode == VIRTUAL_MEM) { switch (info->cpu_mode) { @@ -57,7 +60,7 @@ int handle_shadow_pagefault(struct guest_info * info, addr_t fault_addr, pf_erro } } -addr_t create_new_shadow_pt32(struct guest_info * info) { +addr_t create_new_shadow_pt32() { void * host_pde = 0; host_pde = V3_AllocPages(1); @@ -67,41 +70,116 @@ addr_t create_new_shadow_pt32(struct guest_info * info) { } -static int handle_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; +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); +} - 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); + +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 - pte_cursor = &(pt[i]); + if ((shadow_access != PT_ENTRY_NOT_PRESENT) && + (guest_access != PT_ENTRY_NOT_PRESENT)) { + // aka (guest permission error) + return 1; + } + + if ((shadow_access == PT_ENTRY_NOT_PRESENT) && + (guest_access == PT_ENTRY_NOT_PRESENT)) { + // Page tables completely blank, handle guest first + return 1; + } + + // Otherwise we'll handle the guest fault later...? + } + + return 0; +} + + + + +/* 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) { - // 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); + // 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) { + addr_t shadow_pa = get_shadow_addr(info, guest_fault_pa); - 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; + 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; + shadow_pte->writable = 1; + //set according to VMM policy + shadow_pte->write_through = 0; + shadow_pte->cache_disable = 0; + shadow_pte->global_page = 0; + // + } else { - PrintError("Unsupported Host Memory Type\n"); - return -1; + // Handle hooked pages as well as other special pages + if (handle_special_page_fault(info, fault_addr, PT32_PAGE_ADDR(guest_fault_pa), error_code) == -1) { + PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr); + return -1; + } } + } else { + PrintError("Error in large page fault handler...\n"); + PrintError("This case should have been handled at the top level handler\n"); + return -1; } + + PrintDebug("Returning from large page fault handler\n"); return 0; } + 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); @@ -111,233 +189,125 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr 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)]); + // 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); - /* This should be redone, - but 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_pde_access != PT_ACCESS_OK) && - ( - ( (shadow_pde_access != PT_ENTRY_NOT_PRESENT) && - (guest_pde_access != PT_ENTRY_NOT_PRESENT)) // aka (guest permission error) - || - ( (shadow_pde_access == PT_ENTRY_NOT_PRESENT) && - (guest_pde_access == PT_ENTRY_NOT_PRESENT)))) { - // inject page fault to the guest (Guest PDE fault) - - info->ctrl_regs.cr2 = fault_addr; - v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code); - - - PrintDebug("Injecting PDE pf to guest: (guest access error=%d) (pf error code=%d)\n", guest_pde_access, 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; - - -#ifdef DEBUG_SHADOW_PAGING - PrintDebug("Guest CR3=%x\n", guest_cr3); - PrintDebug("Guest PD\n"); - PrintPD32(guest_pd); - PrintDebug("Shadow PD\n"); - PrintPD32(shadow_pd); -#endif - - return -1; } + + 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; + - //shadow_pde_access = can_access_pde32(shadow_pd, fault_addr, error_code); - - - if (shadow_pde_access == PT_ENTRY_NOT_PRESENT) { - - shadow_pde->present = 1; - shadow_pde->user_page = guest_pde->user_page; - shadow_pde->large_page = guest_pde->large_page; - - - // VMM Specific options - shadow_pde->write_through = 0; - shadow_pde->cache_disable = 0; - shadow_pde->global_page = 0; - // - - guest_pde->accessed = 1; - - if (guest_pde->large_page == 0) { - pte32_t * shadow_pt = NULL; + // VMM Specific options + shadow_pde->write_through = 0; + shadow_pde->cache_disable = 0; + shadow_pde->global_page = 0; + // - shadow_pt = V3_AllocPages(1); - memset(shadow_pt, 0, PAGE_SIZE); + guest_pde->accessed = 1; shadow_pde->pt_base_addr = PD32_BASE_ADDR(shadow_pt); - - shadow_pde->writable = guest_pde->writable; - } else { - 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) { - - v3_raise_exception(info, MC_EXCEPTION); - PrintError("Invalid guest address in large page (0x%x)\n", guest_start_addr); - 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 = V3_AllocPages(1); - memset(shadow_pt, 0, PAGE_SIZE); - - if (handle_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); + 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 *)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; } - - } else { - // Handle hooked pages as well as other special pages - if (handle_special_page_fault(info, fault_addr, error_code) == -1) { - PrintError("Special Page Fault handler returned error for address: %x\n", fault_addr); + 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; } } } - - } 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 - // - - ((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 - injecting into guest\n", shadow_pde_access, guest_pde_access); - info->ctrl_regs.cr2 = fault_addr; - v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code); - return 0; - - } else if (shadow_pde_access == PT_ACCESS_OK) { - pte32_t * shadow_pt = (pte32_t *)PDE32_T_ADDR((*shadow_pde)); - pte32_t * guest_pt = NULL; - - // Page Table Entry fault - - if (guest_pa_to_host_va(info, PDE32_T_ADDR((*guest_pde)), (addr_t*)&guest_pt) == -1) { - PrintDebug("Invalid Guest PTE Address: 0x%x\n", PDE32_T_ADDR((*guest_pde))); - // Machine check the guest - - v3_raise_exception(info, MC_EXCEPTION); + 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 - - 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; + return 0; } - } else { - // Unknown error raise page fault in guest - info->ctrl_regs.cr2 = fault_addr; - v3_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 0; - - } - - //PrintDebugPageTables(shadow_pd); PrintDebug("Returning end of PDE function (rip=%x)\n", info->rip); return 0; } @@ -348,10 +318,10 @@ static int handle_shadow_pagefault32(struct guest_info * info, addr_t fault_addr * 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) { + 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; @@ -372,42 +342,25 @@ static int handle_shadow_pte32_fault(struct guest_info * info, PrintPTE32(fault_addr, shadow_pte); #endif - /* This should be redone, - but 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_pte_access != PT_ACCESS_OK) && - ( - ((shadow_pte_access != PT_ENTRY_NOT_PRESENT) && - (guest_pte_access != PT_ENTRY_NOT_PRESENT)) // aka (guest permission error) - || - ((shadow_pte_access == PT_ENTRY_NOT_PRESENT) && - (guest_pte_access == PT_ENTRY_NOT_PRESENT)))) { - // Inject page fault into the guest - - info->ctrl_regs.cr2 = fault_addr; - v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code); - - PrintDebug("Access error injecting pf to guest (guest access error=%d) (pf error code=%d)\n", guest_pte_access, *(uint_t*)&error_code); + /* 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; - } else if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) { - addr_t shadow_pa; + } + + + if (shadow_pte_access == PT_ENTRY_NOT_PRESENT) { + addr_t guest_pa = PTE32_T_ADDR((*guest_pte)); // Page Table Entry Not Present @@ -416,20 +369,15 @@ static int handle_shadow_pte32_fault(struct guest_info * info, if (host_page_type == HOST_REGION_INVALID) { // Inject a machine check in the guest - - v3_raise_exception(info, MC_EXCEPTION); -#ifdef DEBUG_SHADOW_PAGING PrintDebug("Invalid Guest Address in page table (0x%x)\n", guest_pa); - PrintDebug("fault_addr=0x%x next are guest and shadow ptes \n",fault_addr); - PrintPTE32(fault_addr,guest_pte); - PrintPTE32(fault_addr,shadow_pte); - PrintDebug("Done.\n"); -#endif + v3_raise_exception(info, MC_EXCEPTION); return 0; + } - } else if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { - - shadow_pa = get_shadow_addr(info, guest_pa); + // else... + + if (host_page_type == HOST_REGION_PHYSICAL_MEMORY) { + addr_t shadow_pa = get_shadow_addr(info, guest_pa); shadow_pte->page_base_addr = PT32_BASE_ADDR(shadow_pa); @@ -454,7 +402,7 @@ static int handle_shadow_pte32_fault(struct guest_info * info, } } else { // Page fault handled by hook functions - if (handle_special_page_fault(info, fault_addr, error_code) == -1) { + 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; } @@ -462,18 +410,15 @@ static int handle_shadow_pte32_fault(struct guest_info * info, } else if ((shadow_pte_access == PT_WRITE_ERROR) && (guest_pte->dirty == 0)) { - guest_pte->dirty = 1; - shadow_pte->writable = guest_pte->writable; PrintDebug("Shadow PTE Write Error\n"); - + guest_pte->dirty = 1; + shadow_pte->writable = guest_pte->writable; return 0; + } else { // Inject page fault into the guest - - info->ctrl_regs.cr2 = fault_addr; - v3_raise_exception_with_error(info, PF_EXCEPTION, *(uint_t *)&error_code); - + 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; @@ -585,3 +530,133 @@ int handle_shadow_invlpg(struct guest_info * info) { } +/* + + +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; + + 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); + + 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); + + + 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; + + } 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); + } + + } 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; + } + } + + return 0; +} +*/ diff --git a/palacios/src/palacios/vmm_xed.c b/palacios/src/palacios/vmm_xed.c index e48a1b8..bb5729b 100644 --- a/palacios/src/palacios/vmm_xed.c +++ b/palacios/src/palacios/vmm_xed.c @@ -11,6 +11,7 @@ #include #include + #endif static xed_state_t decoder_state; @@ -115,7 +116,8 @@ int init_decoder() { } -int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * size, uint_t * instr_length) { + +int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, struct basic_instr_info * instr_info) { xed_decoded_inst_t xed_instr; xed_error_enum_t xed_error; @@ -137,7 +139,7 @@ int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * siz return -1; } - *instr_length = xed_decoded_inst_get_length(&xed_instr); + instr_info->instr_length = xed_decoded_inst_get_length(&xed_instr); if (xed_decoded_inst_number_of_memory_operands(&xed_instr) == 0) { @@ -145,7 +147,22 @@ int v3_basic_mem_decode(struct guest_info * info, addr_t instr_ptr, uint_t * siz return -1; } - *size = xed_decoded_inst_get_memory_operand_length(&xed_instr,0); + instr_info->op_size = xed_decoded_inst_get_memory_operand_length(&xed_instr, 0); + + + xed_category_enum_t cat = xed_decoded_inst_get_category(&xed_instr); + if (cat == XED_CATEGORY_STRINGOP) { + instr_info->str_op = 1; + } else { + instr_info->str_op = 0; + } + + xed_operand_values_t * operands = xed_decoded_inst_operands(&xed_instr); + if (xed_operand_values_has_real_rep(operands)) { + instr_info->has_rep = 1; + } else { + instr_info->has_rep = 0; + } return 0; }