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