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 <asm/apic.h> 10 #include <asm/boot.h> 11 #include <asm/desc.h> 12 #include <asm/i8259.h> 13 #include <asm/mshyperv.h> 14 #include <asm/realmode.h> 15 16 extern struct boot_params boot_params; 17 static struct real_mode_header hv_vtl_real_mode_header; 18 19 static bool __init hv_vtl_msi_ext_dest_id(void) 20 { 21 return true; 22 } 23 24 void __init hv_vtl_init_platform(void) 25 { 26 pr_info("Linux runs in Hyper-V Virtual Trust Level\n"); 27 28 x86_platform.realmode_reserve = x86_init_noop; 29 x86_platform.realmode_init = x86_init_noop; 30 x86_init.irqs.pre_vector_init = x86_init_noop; 31 x86_init.timers.timer_init = x86_init_noop; 32 33 /* Avoid searching for BIOS MP tables */ 34 x86_init.mpparse.find_smp_config = x86_init_noop; 35 x86_init.mpparse.get_smp_config = x86_init_uint_noop; 36 37 x86_platform.get_wallclock = get_rtc_noop; 38 x86_platform.set_wallclock = set_rtc_noop; 39 x86_platform.get_nmi_reason = hv_get_nmi_reason; 40 41 x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT; 42 x86_platform.legacy.rtc = 0; 43 x86_platform.legacy.warm_reset = 0; 44 x86_platform.legacy.reserve_bios_regions = 0; 45 x86_platform.legacy.devices.pnpbios = 0; 46 47 x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id; 48 } 49 50 static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc) 51 { 52 return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) | 53 (desc->base1 << 16) | desc->base0; 54 } 55 56 static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc) 57 { 58 return ((u32)desc->limit1 << 16) | (u32)desc->limit0; 59 } 60 61 typedef void (*secondary_startup_64_fn)(void*, void*); 62 static void hv_vtl_ap_entry(void) 63 { 64 ((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params); 65 } 66 67 static int hv_vtl_bringup_vcpu(u32 target_vp_index, u64 eip_ignored) 68 { 69 u64 status; 70 int ret = 0; 71 struct hv_enable_vp_vtl *input; 72 unsigned long irq_flags; 73 74 struct desc_ptr gdt_ptr; 75 struct desc_ptr idt_ptr; 76 77 struct ldttss_desc *tss; 78 struct ldttss_desc *ldt; 79 struct desc_struct *gdt; 80 81 u64 rsp = current->thread.sp; 82 u64 rip = (u64)&hv_vtl_ap_entry; 83 84 native_store_gdt(&gdt_ptr); 85 store_idt(&idt_ptr); 86 87 gdt = (struct desc_struct *)((void *)(gdt_ptr.address)); 88 tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS); 89 ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT); 90 91 local_irq_save(irq_flags); 92 93 input = *this_cpu_ptr(hyperv_pcpu_input_arg); 94 memset(input, 0, sizeof(*input)); 95 96 input->partition_id = HV_PARTITION_ID_SELF; 97 input->vp_index = target_vp_index; 98 input->target_vtl.target_vtl = HV_VTL_MGMT; 99 100 /* 101 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit 102 * mode transition sequence after waking up an AP with SIPI whose 103 * vector points to the 16-bit AP startup trampoline code. Here in 104 * VTL2, we can't perform that sequence as the AP has to start in 105 * the 64-bit mode. 106 * 107 * To make this happen, we tell the hypervisor to load a valid 64-bit 108 * context (most of which is just magic numbers from the CPU manual) 109 * so that AP jumps right to the 64-bit entry of the kernel, and the 110 * control registers are loaded with values that let the AP fetch the 111 * code and data and carry on with work it gets assigned. 112 */ 113 114 input->vp_context.rip = rip; 115 input->vp_context.rsp = rsp; 116 input->vp_context.rflags = 0x0000000000000002; 117 input->vp_context.efer = __rdmsr(MSR_EFER); 118 input->vp_context.cr0 = native_read_cr0(); 119 input->vp_context.cr3 = __native_read_cr3(); 120 input->vp_context.cr4 = native_read_cr4(); 121 input->vp_context.msr_cr_pat = __rdmsr(MSR_IA32_CR_PAT); 122 input->vp_context.idtr.limit = idt_ptr.size; 123 input->vp_context.idtr.base = idt_ptr.address; 124 input->vp_context.gdtr.limit = gdt_ptr.size; 125 input->vp_context.gdtr.base = gdt_ptr.address; 126 127 /* Non-system desc (64bit), long, code, present */ 128 input->vp_context.cs.selector = __KERNEL_CS; 129 input->vp_context.cs.base = 0; 130 input->vp_context.cs.limit = 0xffffffff; 131 input->vp_context.cs.attributes = 0xa09b; 132 /* Non-system desc (64bit), data, present, granularity, default */ 133 input->vp_context.ss.selector = __KERNEL_DS; 134 input->vp_context.ss.base = 0; 135 input->vp_context.ss.limit = 0xffffffff; 136 input->vp_context.ss.attributes = 0xc093; 137 138 /* System desc (128bit), present, LDT */ 139 input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8; 140 input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt); 141 input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt); 142 input->vp_context.ldtr.attributes = 0x82; 143 144 /* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */ 145 input->vp_context.tr.selector = GDT_ENTRY_TSS * 8; 146 input->vp_context.tr.base = hv_vtl_system_desc_base(tss); 147 input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss); 148 input->vp_context.tr.attributes = 0x8b; 149 150 status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL); 151 152 if (!hv_result_success(status) && 153 hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) { 154 pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]", 155 target_vp_index, status); 156 ret = -EINVAL; 157 goto free_lock; 158 } 159 160 status = hv_do_hypercall(HVCALL_START_VP, input, NULL); 161 162 if (!hv_result_success(status)) { 163 pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n", 164 target_vp_index, status); 165 ret = -EINVAL; 166 } 167 168 free_lock: 169 local_irq_restore(irq_flags); 170 171 return ret; 172 } 173 174 static int hv_vtl_apicid_to_vp_id(u32 apic_id) 175 { 176 u64 control; 177 u64 status; 178 unsigned long irq_flags; 179 struct hv_get_vp_from_apic_id_in *input; 180 u32 *output, ret; 181 182 local_irq_save(irq_flags); 183 184 input = *this_cpu_ptr(hyperv_pcpu_input_arg); 185 memset(input, 0, sizeof(*input)); 186 input->partition_id = HV_PARTITION_ID_SELF; 187 input->apic_ids[0] = apic_id; 188 189 output = (u32 *)input; 190 191 control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_ID_FROM_APIC_ID; 192 status = hv_do_hypercall(control, input, output); 193 ret = output[0]; 194 195 local_irq_restore(irq_flags); 196 197 if (!hv_result_success(status)) { 198 pr_err("failed to get vp id from apic id %d, status %#llx\n", 199 apic_id, status); 200 return -EINVAL; 201 } 202 203 return ret; 204 } 205 206 static int hv_vtl_wakeup_secondary_cpu(int apicid, unsigned long start_eip) 207 { 208 int vp_id; 209 210 pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid); 211 vp_id = hv_vtl_apicid_to_vp_id(apicid); 212 213 if (vp_id < 0) { 214 pr_err("Couldn't find CPU with APIC ID %d\n", apicid); 215 return -EINVAL; 216 } 217 if (vp_id > ms_hyperv.max_vp_index) { 218 pr_err("Invalid CPU id %d for APIC ID %d\n", vp_id, apicid); 219 return -EINVAL; 220 } 221 222 return hv_vtl_bringup_vcpu(vp_id, start_eip); 223 } 224 225 int __init hv_vtl_early_init(void) 226 { 227 /* 228 * `boot_cpu_has` returns the runtime feature support, 229 * and here is the earliest it can be used. 230 */ 231 if (cpu_feature_enabled(X86_FEATURE_XSAVE)) 232 panic("XSAVE has to be disabled as it is not supported by this module.\n" 233 "Please add 'noxsave' to the kernel command line.\n"); 234 235 real_mode_header = &hv_vtl_real_mode_header; 236 apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu); 237 238 return 0; 239 } 240