1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (c) 2023, Microsoft Corporation. 4 * 5 * Author: 6 * Saurabh Sengar <ssengar@microsoft.com> 7 */ 8 9 #include <linux/acpi.h> 10 11 #include <asm/acpi.h> 12 #include <asm/apic.h> 13 #include <asm/boot.h> 14 #include <asm/desc.h> 15 #include <asm/fpu/api.h> 16 #include <asm/fpu/types.h> 17 #include <asm/i8259.h> 18 #include <asm/mshyperv.h> 19 #include <asm/msr.h> 20 #include <asm/realmode.h> 21 #include <asm/reboot.h> 22 #include <asm/smap.h> 23 #include <linux/export.h> 24 #include <../kernel/smpboot.h> 25 #include "../../kernel/fpu/legacy.h" 26 27 extern struct boot_params boot_params; 28 static struct real_mode_header hv_vtl_real_mode_header; 29 30 static bool __init hv_vtl_msi_ext_dest_id(void) 31 { 32 return true; 33 } 34 35 /* 36 * The `native_machine_emergency_restart` function from `reboot.c` writes 37 * to the physical address 0x472 to indicate the type of reboot for the 38 * firmware. We cannot have that in VSM as the memory composition might 39 * be more generic, and such write effectively corrupts the memory thus 40 * making diagnostics harder at the very least. 41 */ 42 static void __noreturn hv_vtl_emergency_restart(void) 43 { 44 /* 45 * Cause a triple fault and the immediate reset. Here the code does not run 46 * on the top of any firmware, whereby cannot reach out to its services. 47 * The inifinite loop is for the improbable case that the triple fault does 48 * not work and have to preserve the state intact for debugging. 49 */ 50 for (;;) { 51 idt_invalidate(); 52 __asm__ __volatile__("int3"); 53 } 54 } 55 56 /* 57 * The only way to restart in the VTL mode is to triple fault as the kernel runs 58 * as firmware. 59 */ 60 static void __noreturn hv_vtl_restart(char __maybe_unused *cmd) 61 { 62 hv_vtl_emergency_restart(); 63 } 64 65 static inline bool within_page(u64 addr, u64 start) 66 { 67 return addr >= start && addr < (start + PAGE_SIZE); 68 } 69 70 static bool hv_vtl_is_private_mmio_tdx(u64 addr) 71 { 72 u64 mb_addr = acpi_get_mp_wakeup_mailbox_paddr(); 73 74 return mb_addr && within_page(addr, mb_addr); 75 } 76 77 void __init hv_vtl_init_platform(void) 78 { 79 /* 80 * This function is a no-op if the VTL mode is not enabled. 81 * If it is, this function runs if and only the kernel boots in 82 * VTL2 which the x86 hv initialization path makes sure of. 83 */ 84 pr_info("Linux runs in Hyper-V Virtual Trust Level %d\n", ms_hyperv.vtl); 85 86 /* There is no paravisor present if we are here. */ 87 if (hv_isolation_type_tdx()) { 88 x86_init.resources.realmode_limit = SZ_4G; 89 x86_platform.hyper.is_private_mmio = hv_vtl_is_private_mmio_tdx; 90 91 } else { 92 x86_platform.realmode_reserve = x86_init_noop; 93 x86_platform.realmode_init = x86_init_noop; 94 real_mode_header = &hv_vtl_real_mode_header; 95 } 96 x86_init.irqs.pre_vector_init = x86_init_noop; 97 x86_init.timers.timer_init = x86_init_noop; 98 x86_init.resources.probe_roms = x86_init_noop; 99 100 /* Avoid searching for BIOS MP tables */ 101 x86_init.mpparse.find_mptable = x86_init_noop; 102 x86_init.mpparse.early_parse_smp_cfg = x86_init_noop; 103 104 x86_platform.get_wallclock = get_rtc_noop; 105 x86_platform.set_wallclock = set_rtc_noop; 106 x86_platform.get_nmi_reason = hv_get_nmi_reason; 107 108 x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT; 109 x86_platform.legacy.rtc = 0; 110 x86_platform.legacy.warm_reset = 0; 111 x86_platform.legacy.reserve_bios_regions = 0; 112 x86_platform.legacy.devices.pnpbios = 0; 113 114 x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id; 115 } 116 117 static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc) 118 { 119 return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) | 120 (desc->base1 << 16) | desc->base0; 121 } 122 123 static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc) 124 { 125 return ((u32)desc->limit1 << 16) | (u32)desc->limit0; 126 } 127 128 typedef void (*secondary_startup_64_fn)(void*, void*); 129 static void hv_vtl_ap_entry(void) 130 { 131 ((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params); 132 } 133 134 static int hv_vtl_bringup_vcpu(u32 target_vp_index, int cpu, u64 eip_ignored) 135 { 136 u64 status, rsp, rip; 137 int ret = 0; 138 struct hv_enable_vp_vtl *input; 139 unsigned long irq_flags; 140 141 struct desc_ptr gdt_ptr; 142 struct desc_ptr idt_ptr; 143 144 struct ldttss_desc *tss; 145 struct ldttss_desc *ldt; 146 struct desc_struct *gdt; 147 148 struct task_struct *idle = idle_thread_get(cpu); 149 if (IS_ERR(idle)) 150 return PTR_ERR(idle); 151 152 rsp = (unsigned long)idle->thread.sp; 153 rip = (u64)&hv_vtl_ap_entry; 154 155 native_store_gdt(&gdt_ptr); 156 store_idt(&idt_ptr); 157 158 gdt = (struct desc_struct *)((void *)(gdt_ptr.address)); 159 tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS); 160 ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT); 161 162 local_irq_save(irq_flags); 163 164 input = *this_cpu_ptr(hyperv_pcpu_input_arg); 165 memset(input, 0, sizeof(*input)); 166 167 input->partition_id = HV_PARTITION_ID_SELF; 168 input->vp_index = target_vp_index; 169 input->target_vtl.target_vtl = HV_VTL_MGMT; 170 171 /* 172 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit 173 * mode transition sequence after waking up an AP with SIPI whose 174 * vector points to the 16-bit AP startup trampoline code. Here in 175 * VTL2, we can't perform that sequence as the AP has to start in 176 * the 64-bit mode. 177 * 178 * To make this happen, we tell the hypervisor to load a valid 64-bit 179 * context (most of which is just magic numbers from the CPU manual) 180 * so that AP jumps right to the 64-bit entry of the kernel, and the 181 * control registers are loaded with values that let the AP fetch the 182 * code and data and carry on with work it gets assigned. 183 */ 184 185 input->vp_context.rip = rip; 186 input->vp_context.rsp = rsp; 187 input->vp_context.rflags = 0x0000000000000002; 188 input->vp_context.efer = native_rdmsrq(MSR_EFER); 189 input->vp_context.cr0 = native_read_cr0(); 190 input->vp_context.cr3 = __native_read_cr3(); 191 input->vp_context.cr4 = native_read_cr4(); 192 input->vp_context.msr_cr_pat = native_rdmsrq(MSR_IA32_CR_PAT); 193 input->vp_context.idtr.limit = idt_ptr.size; 194 input->vp_context.idtr.base = idt_ptr.address; 195 input->vp_context.gdtr.limit = gdt_ptr.size; 196 input->vp_context.gdtr.base = gdt_ptr.address; 197 198 /* Non-system desc (64bit), long, code, present */ 199 input->vp_context.cs.selector = __KERNEL_CS; 200 input->vp_context.cs.base = 0; 201 input->vp_context.cs.limit = 0xffffffff; 202 input->vp_context.cs.attributes = 0xa09b; 203 /* Non-system desc (64bit), data, present, granularity, default */ 204 input->vp_context.ss.selector = __KERNEL_DS; 205 input->vp_context.ss.base = 0; 206 input->vp_context.ss.limit = 0xffffffff; 207 input->vp_context.ss.attributes = 0xc093; 208 209 /* System desc (128bit), present, LDT */ 210 input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8; 211 input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt); 212 input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt); 213 input->vp_context.ldtr.attributes = 0x82; 214 215 /* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */ 216 input->vp_context.tr.selector = GDT_ENTRY_TSS * 8; 217 input->vp_context.tr.base = hv_vtl_system_desc_base(tss); 218 input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss); 219 input->vp_context.tr.attributes = 0x8b; 220 221 status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL); 222 223 if (!hv_result_success(status) && 224 hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) { 225 pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]", 226 target_vp_index, status); 227 ret = -EINVAL; 228 goto free_lock; 229 } 230 231 status = hv_do_hypercall(HVCALL_START_VP, input, NULL); 232 233 if (!hv_result_success(status)) { 234 pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n", 235 target_vp_index, status); 236 ret = -EINVAL; 237 } 238 239 free_lock: 240 local_irq_restore(irq_flags); 241 242 return ret; 243 } 244 245 static int hv_vtl_wakeup_secondary_cpu(u32 apicid, unsigned long start_eip, unsigned int cpu) 246 { 247 int vp_index; 248 249 pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid); 250 vp_index = hv_apicid_to_vp_index(apicid); 251 252 if (vp_index < 0) { 253 pr_err("Couldn't find CPU with APIC ID %d\n", apicid); 254 return -EINVAL; 255 } 256 if (vp_index > ms_hyperv.max_vp_index) { 257 pr_err("Invalid CPU id %d for APIC ID %d\n", vp_index, apicid); 258 return -EINVAL; 259 } 260 261 return hv_vtl_bringup_vcpu(vp_index, cpu, start_eip); 262 } 263 264 int __init hv_vtl_early_init(void) 265 { 266 machine_ops.emergency_restart = hv_vtl_emergency_restart; 267 machine_ops.restart = hv_vtl_restart; 268 269 /* 270 * `boot_cpu_has` returns the runtime feature support, 271 * and here is the earliest it can be used. 272 */ 273 if (cpu_feature_enabled(X86_FEATURE_XSAVE)) 274 panic("XSAVE has to be disabled as it is not supported by this module.\n" 275 "Please add 'noxsave' to the kernel command line.\n"); 276 277 /* 278 * TDX confidential VMs do not trust the hypervisor and cannot use it to 279 * boot secondary CPUs. Instead, they will be booted using the wakeup 280 * mailbox if detected during boot. See setup_arch(). 281 * 282 * There is no paravisor present if we are here. 283 */ 284 if (!hv_isolation_type_tdx()) 285 apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu); 286 287 return 0; 288 } 289 290 DEFINE_STATIC_CALL_NULL(__mshv_vtl_return_hypercall, void (*)(void)); 291 292 void mshv_vtl_return_call_init(u64 vtl_return_offset) 293 { 294 static_call_update(__mshv_vtl_return_hypercall, 295 (void *)((u8 *)hv_hypercall_pg + vtl_return_offset)); 296 } 297 EXPORT_SYMBOL(mshv_vtl_return_call_init); 298 299 void mshv_vtl_return_call(struct mshv_vtl_cpu_context *vtl0) 300 { 301 struct hv_vp_assist_page *hvp; 302 303 hvp = hv_vp_assist_page[smp_processor_id()]; 304 hvp->vtl_ret_x64rax = vtl0->rax; 305 hvp->vtl_ret_x64rcx = vtl0->rcx; 306 307 kernel_fpu_begin_mask(0); 308 fxrstor(&vtl0->fx_state); 309 __mshv_vtl_return_call(vtl0); 310 fxsave(&vtl0->fx_state); 311 kernel_fpu_end(); 312 } 313 EXPORT_SYMBOL(mshv_vtl_return_call); 314