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