X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?p=palacios.git;a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_sym_iface.c;h=83959555e33fee6587f3152666e889e777c1dc9f;hp=0c6414d1c6fb28f3892b719fc46121466f58964b;hb=cfcc5717f659b3ed2954f41cf363d3bceae8dc84;hpb=e63be432894673d56526c4f1c2cb4fa64daf01d9 diff --git a/palacios/src/palacios/vmm_sym_iface.c b/palacios/src/palacios/vmm_sym_iface.c index 0c6414d..8395955 100644 --- a/palacios/src/palacios/vmm_sym_iface.c +++ b/palacios/src/palacios/vmm_sym_iface.c @@ -21,17 +21,60 @@ #include #include #include +#include +#include -#define SYM_MSR_NUM 0x535 +#define SYM_PAGE_MSR 0x535 +#define SYM_CPUID_NUM 0x90000000 +// A succesfull symcall returns via the RET_HCALL, with the return values in registers +// A symcall error returns via the ERR_HCALL with the error code in rbx +#define SYM_CALL_RET_HCALL 0x535 +#define SYM_CALL_ERR_HCALL 0x536 + + +/* Notes: We use a combination of SYSCALL and SYSENTER Semantics + * SYSCALL just sets an EIP, CS/SS seg, and GS seg via swapgs + * the RSP is loaded via the structure pointed to by GS + * This is safe because it assumes that system calls are guaranteed to be made with an empty kernel stack. + * We cannot make that assumption with a symcall, so we have to have our own stack area somewhere. + * SYSTENTER does not really use the GS base MSRs, but we do to map to 64 bit kernels + */ + +#define SYMCALL_RIP_MSR 0x536 +#define SYMCALL_RSP_MSR 0x537 +#define SYMCALL_CS_MSR 0x538 +#define SYMCALL_GS_MSR 0x539 +#define SYMCALL_FS_MSR 0x540 static int msr_read(uint_t msr, struct v3_msr * dst, void * priv_data) { struct guest_info * info = (struct guest_info *)priv_data; struct v3_sym_state * state = &(info->sym_state); - dst->value = state->guest_pg_addr; + switch (msr) { + case SYM_PAGE_MSR: + dst->value = state->guest_pg_addr; + break; + case SYMCALL_RIP_MSR: + dst->value = state->sym_call_rip; + break; + case SYMCALL_RSP_MSR: + dst->value = state->sym_call_rsp; + break; + case SYMCALL_CS_MSR: + dst->value = state->sym_call_cs; + break; + case SYMCALL_GS_MSR: + dst->value = state->sym_call_gs; + break; + case SYMCALL_FS_MSR: + dst->value = state->sym_call_fs; + break; + default: + return -1; + } return 0; } @@ -40,70 +83,275 @@ static int msr_write(uint_t msr, struct v3_msr src, void * priv_data) { struct guest_info * info = (struct guest_info *)priv_data; struct v3_sym_state * state = &(info->sym_state); - if (state->active == 1) { - // unmap page - struct v3_shadow_region * old_reg = v3_get_shadow_region(info, (addr_t)state->guest_pg_addr); + if (msr == SYM_PAGE_MSR) { + PrintDebug("Symbiotic MSR write for page %p\n", (void *)src.value); - if (old_reg == NULL) { - PrintError("Could not find previously active symbiotic page (%p)\n", (void *)state->guest_pg_addr); - return -1; + if (state->active == 1) { + // unmap page + struct v3_shadow_region * old_reg = v3_get_shadow_region(info, (addr_t)state->guest_pg_addr); + + if (old_reg == NULL) { + PrintError("Could not find previously active symbiotic page (%p)\n", (void *)state->guest_pg_addr); + return -1; + } + + v3_delete_shadow_region(info, old_reg); } - v3_delete_shadow_region(info, old_reg); + state->guest_pg_addr = src.value; + state->guest_pg_addr &= ~0xfffLL; + + state->active = 1; + + // map page + v3_add_shadow_mem(info, (addr_t)state->guest_pg_addr, + (addr_t)(state->guest_pg_addr + PAGE_SIZE_4KB - 1), + state->sym_page_pa); + + + } else if (msr == SYMCALL_RIP_MSR) { + state->sym_call_rip = src.value; + } else if (msr == SYMCALL_RSP_MSR) { + state->sym_call_rsp = src.value; + } else if (msr == SYMCALL_CS_MSR) { + state->sym_call_cs = src.value; + } else if (msr == SYMCALL_GS_MSR) { + state->sym_call_gs = src.value; + } else if (msr == SYMCALL_FS_MSR) { + state->sym_call_fs = src.value; + } else { + PrintError("Invalid Symbiotic MSR write (0x%x)\n", msr); + return -1; } - state->guest_pg_addr = src.value; - state->guest_pg_addr &= ~0xfffLL; + return 0; +} + +static int cpuid_fn(struct guest_info * info, uint32_t cpuid, + uint32_t * eax, uint32_t * ebx, + uint32_t * ecx, uint32_t * edx, + void * private_data) { + extern v3_cpu_arch_t v3_cpu_types[]; - state->active = 1; + *eax = *(uint32_t *)"V3V"; + + if ((v3_cpu_types[info->cpu_id] == V3_SVM_CPU) || + (v3_cpu_types[info->cpu_id] == V3_SVM_REV3_CPU)) { + *ebx = *(uint32_t *)"SVM"; + } else if ((v3_cpu_types[info->cpu_id] == V3_VMX_CPU) || + (v3_cpu_types[info->cpu_id] == V3_VMX_EPT_CPU)) { + *ebx = *(uint32_t *)"VMX"; + } - // map page - v3_add_shadow_mem(info, (addr_t)state->guest_pg_addr, - (addr_t)(state->guest_pg_addr + PAGE_SIZE_4KB - 1), - state->sym_page_pa); return 0; } + + +static int sym_call_ret(struct guest_info * info, uint_t hcall_id, void * private_data); +static int sym_call_err(struct guest_info * info, uint_t hcall_id, void * private_data); int v3_init_sym_iface(struct guest_info * info) { struct v3_sym_state * state = &(info->sym_state); - memset(state, 0, sizeof(struct v3_sym_state)); - PrintDebug("Allocating symbiotic page\n"); state->sym_page_pa = (addr_t)V3_AllocPages(1); state->sym_page = (struct v3_sym_interface *)V3_VAddr((void *)state->sym_page_pa); - - PrintDebug("Clearing symbiotic page\n"); memset(state->sym_page, 0, PAGE_SIZE_4KB); - PrintDebug("hooking MSR\n"); - v3_hook_msr(info, SYM_MSR_NUM, msr_read, msr_write, info); + + memcpy(&(state->sym_page->magic), "V3V", 3); + + v3_hook_msr(info, SYM_PAGE_MSR, msr_read, msr_write, info); + + v3_hook_cpuid(info, SYM_CPUID_NUM, cpuid_fn, info); + + v3_hook_msr(info, SYMCALL_RIP_MSR, msr_read, msr_write, info); + v3_hook_msr(info, SYMCALL_RSP_MSR, msr_read, msr_write, info); + v3_hook_msr(info, SYMCALL_CS_MSR, msr_read, msr_write, info); + v3_hook_msr(info, SYMCALL_GS_MSR, msr_read, msr_write, info); + v3_hook_msr(info, SYMCALL_FS_MSR, msr_read, msr_write, info); + + v3_register_hypercall(info, SYM_CALL_RET_HCALL, sym_call_ret, NULL); + v3_register_hypercall(info, SYM_CALL_ERR_HCALL, sym_call_err, NULL); - PrintDebug("Done\n"); return 0; } int v3_sym_map_pci_passthrough(struct guest_info * info, uint_t bus, uint_t dev, uint_t fn) { struct v3_sym_state * state = &(info->sym_state); - uint_t dev_index = (bus << 16) + (dev << 8) + fn; + uint_t dev_index = (bus << 8) + (dev << 3) + fn; uint_t major = dev_index / 8; uint_t minor = dev_index % 8; + if (bus > 3) { + PrintError("Invalid PCI bus %d\n", bus); + return -1; + } + + PrintDebug("Setting passthrough pci map for index=%d\n", dev_index); + state->sym_page->pci_pt_map[major] |= 0x1 << minor; + PrintDebug("pt_map entry=%x\n", state->sym_page->pci_pt_map[major]); + + PrintDebug("pt map vmm addr=%p\n", state->sym_page->pci_pt_map); + return 0; } int v3_sym_unmap_pci_passthrough(struct guest_info * info, uint_t bus, uint_t dev, uint_t fn) { struct v3_sym_state * state = &(info->sym_state); - uint_t dev_index = (bus << 16) + (dev << 8) + fn; + uint_t dev_index = (bus << 8) + (dev << 3) + fn; uint_t major = dev_index / 8; uint_t minor = dev_index % 8; + if (bus > 3) { + PrintError("Invalid PCI bus %d\n", bus); + return -1; + } + state->sym_page->pci_pt_map[major] &= ~(0x1 << minor); return 0; } + + +static int sym_call_err(struct guest_info * info, uint_t hcall_id, void * private_data) { + struct v3_sym_state * state = (struct v3_sym_state *)&(info->sym_state); + + PrintError("sym call error\n"); + + state->sym_call_errno = (int)info->vm_regs.rbx; + v3_print_guest_state(info); + v3_print_mem_map(info); + + // clear sym flags + state->sym_call_error = 1; + state->sym_call_returned = 1; + + return -1; +} + +static int sym_call_ret(struct guest_info * info, uint_t hcall_id, void * private_data) { + struct v3_sym_state * state = (struct v3_sym_state *)&(info->sym_state); + + // PrintError("Return from sym call (ID=%x)\n", hcall_id); + // v3_print_guest_state(info); + + state->sym_call_returned = 1; + + return 0; +} + +static int execute_symcall(struct guest_info * info) { + + while (info->sym_state.sym_call_returned == 0) { + if (v3_vm_enter(info) == -1) { + PrintError("Error in Sym call\n"); + return -1; + } + } + + return 0; +} + + +int v3_sym_call(struct guest_info * info, + uint64_t call_num, sym_arg_t * arg0, + sym_arg_t * arg1, sym_arg_t * arg2, + sym_arg_t * arg3, sym_arg_t * arg4) { + struct v3_sym_state * state = (struct v3_sym_state *)&(info->sym_state); + struct v3_sym_context * old_ctx = (struct v3_sym_context *)&(state->old_ctx); + struct v3_segment sym_cs; + struct v3_segment sym_ss; + uint64_t trash_args[5] = { [0 ... 4] = 0 }; + + // PrintDebug("Making Sym call\n"); + // v3_print_guest_state(info); + + if ((state->sym_page->sym_call_enabled == 0) || + (state->sym_call_active == 1)) { + return -1; + } + + if (!arg0) arg0 = &trash_args[0]; + if (!arg1) arg1 = &trash_args[1]; + if (!arg2) arg2 = &trash_args[2]; + if (!arg3) arg3 = &trash_args[3]; + if (!arg4) arg4 = &trash_args[4]; + + // Save the old context + memcpy(&(old_ctx->vm_regs), &(info->vm_regs), sizeof(struct v3_gprs)); + memcpy(&(old_ctx->cs), &(info->segments.cs), sizeof(struct v3_segment)); + memcpy(&(old_ctx->ss), &(info->segments.ss), sizeof(struct v3_segment)); + old_ctx->gs_base = info->segments.gs.base; + old_ctx->fs_base = info->segments.fs.base; + old_ctx->rip = info->rip; + old_ctx->cpl = info->cpl; + old_ctx->flags = info->ctrl_regs.rflags; + + // Setup the sym call context + info->rip = state->sym_call_rip; + info->vm_regs.rsp = state->sym_call_rsp; // old contest rsp is saved in vm_regs + + v3_translate_segment(info, state->sym_call_cs, &sym_cs); + memcpy(&(info->segments.cs), &sym_cs, sizeof(struct v3_segment)); + + v3_translate_segment(info, state->sym_call_cs + 8, &sym_ss); + memcpy(&(info->segments.ss), &sym_ss, sizeof(struct v3_segment)); + + info->segments.gs.base = state->sym_call_gs; + info->segments.fs.base = state->sym_call_fs; + info->cpl = 0; + + info->vm_regs.rax = call_num; + info->vm_regs.rbx = *arg0; + info->vm_regs.rcx = *arg1; + info->vm_regs.rdx = *arg2; + info->vm_regs.rsi = *arg3; + info->vm_regs.rdi = *arg4; + + // Mark sym call as active + state->sym_call_active = 1; + state->sym_call_returned = 0; + + // PrintDebug("Sym state\n"); + // v3_print_guest_state(info); + + // Do the sym call entry + if (execute_symcall(info) == -1) { + PrintError("SYMCALL error\n"); + return -1; + } + + // clear sym flags + state->sym_call_active = 0; + + *arg0 = info->vm_regs.rbx; + *arg1 = info->vm_regs.rcx; + *arg2 = info->vm_regs.rdx; + *arg3 = info->vm_regs.rsi; + *arg4 = info->vm_regs.rdi; + + // restore guest state + memcpy(&(info->vm_regs), &(old_ctx->vm_regs), sizeof(struct v3_gprs)); + memcpy(&(info->segments.cs), &(old_ctx->cs), sizeof(struct v3_segment)); + memcpy(&(info->segments.ss), &(old_ctx->ss), sizeof(struct v3_segment)); + info->segments.gs.base = old_ctx->gs_base; + info->segments.fs.base = old_ctx->fs_base; + info->rip = old_ctx->rip; + info->cpl = old_ctx->cpl; + info->ctrl_regs.rflags = old_ctx->flags; + + + + // PrintError("restoring guest state\n"); + // v3_print_guest_state(info); + + return 0; +} + +