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 | #include <../kernel/smpboot.h> |
16 | |
17 | extern struct boot_params boot_params; |
18 | static struct real_mode_header ; |
19 | |
20 | static bool __init hv_vtl_msi_ext_dest_id(void) |
21 | { |
22 | return true; |
23 | } |
24 | |
25 | void __init hv_vtl_init_platform(void) |
26 | { |
27 | pr_info("Linux runs in Hyper-V Virtual Trust Level\n" ); |
28 | |
29 | x86_platform.realmode_reserve = x86_init_noop; |
30 | x86_platform.realmode_init = x86_init_noop; |
31 | x86_init.irqs.pre_vector_init = x86_init_noop; |
32 | x86_init.timers.timer_init = x86_init_noop; |
33 | |
34 | /* Avoid searching for BIOS MP tables */ |
35 | x86_init.mpparse.find_mptable = x86_init_noop; |
36 | x86_init.mpparse.early_parse_smp_cfg = x86_init_noop; |
37 | x86_init.mpparse.parse_smp_cfg = x86_init_noop; |
38 | |
39 | x86_platform.get_wallclock = get_rtc_noop; |
40 | x86_platform.set_wallclock = set_rtc_noop; |
41 | x86_platform.get_nmi_reason = hv_get_nmi_reason; |
42 | |
43 | x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT; |
44 | x86_platform.legacy.rtc = 0; |
45 | x86_platform.legacy.warm_reset = 0; |
46 | x86_platform.legacy.reserve_bios_regions = 0; |
47 | x86_platform.legacy.devices.pnpbios = 0; |
48 | |
49 | x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id; |
50 | } |
51 | |
52 | static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc) |
53 | { |
54 | return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) | |
55 | (desc->base1 << 16) | desc->base0; |
56 | } |
57 | |
58 | static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc) |
59 | { |
60 | return ((u32)desc->limit1 << 16) | (u32)desc->limit0; |
61 | } |
62 | |
63 | typedef void (*secondary_startup_64_fn)(void*, void*); |
64 | static void hv_vtl_ap_entry(void) |
65 | { |
66 | ((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params); |
67 | } |
68 | |
69 | static int hv_vtl_bringup_vcpu(u32 target_vp_index, int cpu, u64 eip_ignored) |
70 | { |
71 | u64 status; |
72 | int ret = 0; |
73 | struct hv_enable_vp_vtl *input; |
74 | unsigned long irq_flags; |
75 | |
76 | struct desc_ptr gdt_ptr; |
77 | struct desc_ptr idt_ptr; |
78 | |
79 | struct ldttss_desc *tss; |
80 | struct ldttss_desc *ldt; |
81 | struct desc_struct *gdt; |
82 | |
83 | struct task_struct *idle = idle_thread_get(cpu); |
84 | u64 rsp = (unsigned long)idle->thread.sp; |
85 | |
86 | u64 rip = (u64)&hv_vtl_ap_entry; |
87 | |
88 | native_store_gdt(dtr: &gdt_ptr); |
89 | store_idt(dtr: &idt_ptr); |
90 | |
91 | gdt = (struct desc_struct *)((void *)(gdt_ptr.address)); |
92 | tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS); |
93 | ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT); |
94 | |
95 | local_irq_save(irq_flags); |
96 | |
97 | input = *this_cpu_ptr(hyperv_pcpu_input_arg); |
98 | memset(input, 0, sizeof(*input)); |
99 | |
100 | input->partition_id = HV_PARTITION_ID_SELF; |
101 | input->vp_index = target_vp_index; |
102 | input->target_vtl.target_vtl = HV_VTL_MGMT; |
103 | |
104 | /* |
105 | * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit |
106 | * mode transition sequence after waking up an AP with SIPI whose |
107 | * vector points to the 16-bit AP startup trampoline code. Here in |
108 | * VTL2, we can't perform that sequence as the AP has to start in |
109 | * the 64-bit mode. |
110 | * |
111 | * To make this happen, we tell the hypervisor to load a valid 64-bit |
112 | * context (most of which is just magic numbers from the CPU manual) |
113 | * so that AP jumps right to the 64-bit entry of the kernel, and the |
114 | * control registers are loaded with values that let the AP fetch the |
115 | * code and data and carry on with work it gets assigned. |
116 | */ |
117 | |
118 | input->vp_context.rip = rip; |
119 | input->vp_context.rsp = rsp; |
120 | input->vp_context.rflags = 0x0000000000000002; |
121 | input->vp_context.efer = __rdmsr(MSR_EFER); |
122 | input->vp_context.cr0 = native_read_cr0(); |
123 | input->vp_context.cr3 = __native_read_cr3(); |
124 | input->vp_context.cr4 = native_read_cr4(); |
125 | input->vp_context.msr_cr_pat = __rdmsr(MSR_IA32_CR_PAT); |
126 | input->vp_context.idtr.limit = idt_ptr.size; |
127 | input->vp_context.idtr.base = idt_ptr.address; |
128 | input->vp_context.gdtr.limit = gdt_ptr.size; |
129 | input->vp_context.gdtr.base = gdt_ptr.address; |
130 | |
131 | /* Non-system desc (64bit), long, code, present */ |
132 | input->vp_context.cs.selector = __KERNEL_CS; |
133 | input->vp_context.cs.base = 0; |
134 | input->vp_context.cs.limit = 0xffffffff; |
135 | input->vp_context.cs.attributes = 0xa09b; |
136 | /* Non-system desc (64bit), data, present, granularity, default */ |
137 | input->vp_context.ss.selector = __KERNEL_DS; |
138 | input->vp_context.ss.base = 0; |
139 | input->vp_context.ss.limit = 0xffffffff; |
140 | input->vp_context.ss.attributes = 0xc093; |
141 | |
142 | /* System desc (128bit), present, LDT */ |
143 | input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8; |
144 | input->vp_context.ldtr.base = hv_vtl_system_desc_base(desc: ldt); |
145 | input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(desc: ldt); |
146 | input->vp_context.ldtr.attributes = 0x82; |
147 | |
148 | /* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */ |
149 | input->vp_context.tr.selector = GDT_ENTRY_TSS * 8; |
150 | input->vp_context.tr.base = hv_vtl_system_desc_base(desc: tss); |
151 | input->vp_context.tr.limit = hv_vtl_system_desc_limit(desc: tss); |
152 | input->vp_context.tr.attributes = 0x8b; |
153 | |
154 | status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, inputaddr: input, NULL); |
155 | |
156 | if (!hv_result_success(status) && |
157 | hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) { |
158 | pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]" , |
159 | target_vp_index, status); |
160 | ret = -EINVAL; |
161 | goto free_lock; |
162 | } |
163 | |
164 | status = hv_do_hypercall(HVCALL_START_VP, inputaddr: input, NULL); |
165 | |
166 | if (!hv_result_success(status)) { |
167 | pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n" , |
168 | target_vp_index, status); |
169 | ret = -EINVAL; |
170 | } |
171 | |
172 | free_lock: |
173 | local_irq_restore(irq_flags); |
174 | |
175 | return ret; |
176 | } |
177 | |
178 | static int hv_vtl_apicid_to_vp_id(u32 apic_id) |
179 | { |
180 | u64 control; |
181 | u64 status; |
182 | unsigned long irq_flags; |
183 | struct hv_get_vp_from_apic_id_in *input; |
184 | u32 *output, ret; |
185 | |
186 | local_irq_save(irq_flags); |
187 | |
188 | input = *this_cpu_ptr(hyperv_pcpu_input_arg); |
189 | memset(input, 0, sizeof(*input)); |
190 | input->partition_id = HV_PARTITION_ID_SELF; |
191 | input->apic_ids[0] = apic_id; |
192 | |
193 | output = (u32 *)input; |
194 | |
195 | control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_ID_FROM_APIC_ID; |
196 | status = hv_do_hypercall(control, inputaddr: input, outputaddr: output); |
197 | ret = output[0]; |
198 | |
199 | local_irq_restore(irq_flags); |
200 | |
201 | if (!hv_result_success(status)) { |
202 | pr_err("failed to get vp id from apic id %d, status %#llx\n" , |
203 | apic_id, status); |
204 | return -EINVAL; |
205 | } |
206 | |
207 | return ret; |
208 | } |
209 | |
210 | static int hv_vtl_wakeup_secondary_cpu(u32 apicid, unsigned long start_eip) |
211 | { |
212 | int vp_id, cpu; |
213 | |
214 | /* Find the logical CPU for the APIC ID */ |
215 | for_each_present_cpu(cpu) { |
216 | if (arch_match_cpu_phys_id(cpu, phys_id: apicid)) |
217 | break; |
218 | } |
219 | if (cpu >= nr_cpu_ids) |
220 | return -EINVAL; |
221 | |
222 | pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n" , apicid); |
223 | vp_id = hv_vtl_apicid_to_vp_id(apic_id: apicid); |
224 | |
225 | if (vp_id < 0) { |
226 | pr_err("Couldn't find CPU with APIC ID %d\n" , apicid); |
227 | return -EINVAL; |
228 | } |
229 | if (vp_id > ms_hyperv.max_vp_index) { |
230 | pr_err("Invalid CPU id %d for APIC ID %d\n" , vp_id, apicid); |
231 | return -EINVAL; |
232 | } |
233 | |
234 | return hv_vtl_bringup_vcpu(target_vp_index: vp_id, cpu, eip_ignored: start_eip); |
235 | } |
236 | |
237 | int __init hv_vtl_early_init(void) |
238 | { |
239 | /* |
240 | * `boot_cpu_has` returns the runtime feature support, |
241 | * and here is the earliest it can be used. |
242 | */ |
243 | if (cpu_feature_enabled(X86_FEATURE_XSAVE)) |
244 | panic(fmt: "XSAVE has to be disabled as it is not supported by this module.\n" |
245 | "Please add 'noxsave' to the kernel command line.\n" ); |
246 | |
247 | real_mode_header = &hv_vtl_real_mode_header; |
248 | apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu); |
249 | |
250 | return 0; |
251 | } |
252 | |