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