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/msr.h>
15 #include <asm/realmode.h>
16 #include <asm/reboot.h>
17 #include <../kernel/smpboot.h>
18
19 extern struct boot_params boot_params;
20 static struct real_mode_header hv_vtl_real_mode_header;
21
hv_vtl_msi_ext_dest_id(void)22 static bool __init hv_vtl_msi_ext_dest_id(void)
23 {
24 return true;
25 }
26
27 /*
28 * The `native_machine_emergency_restart` function from `reboot.c` writes
29 * to the physical address 0x472 to indicate the type of reboot for the
30 * firmware. We cannot have that in VSM as the memory composition might
31 * be more generic, and such write effectively corrupts the memory thus
32 * making diagnostics harder at the very least.
33 */
hv_vtl_emergency_restart(void)34 static void __noreturn hv_vtl_emergency_restart(void)
35 {
36 /*
37 * Cause a triple fault and the immediate reset. Here the code does not run
38 * on the top of any firmware, whereby cannot reach out to its services.
39 * The inifinite loop is for the improbable case that the triple fault does
40 * not work and have to preserve the state intact for debugging.
41 */
42 for (;;) {
43 idt_invalidate();
44 __asm__ __volatile__("int3");
45 }
46 }
47
48 /*
49 * The only way to restart in the VTL mode is to triple fault as the kernel runs
50 * as firmware.
51 */
hv_vtl_restart(char __maybe_unused * cmd)52 static void __noreturn hv_vtl_restart(char __maybe_unused *cmd)
53 {
54 hv_vtl_emergency_restart();
55 }
56
hv_vtl_init_platform(void)57 void __init hv_vtl_init_platform(void)
58 {
59 /*
60 * This function is a no-op if the VTL mode is not enabled.
61 * If it is, this function runs if and only the kernel boots in
62 * VTL2 which the x86 hv initialization path makes sure of.
63 */
64 pr_info("Linux runs in Hyper-V Virtual Trust Level %d\n", ms_hyperv.vtl);
65
66 x86_platform.realmode_reserve = x86_init_noop;
67 x86_platform.realmode_init = x86_init_noop;
68 x86_init.irqs.pre_vector_init = x86_init_noop;
69 x86_init.timers.timer_init = x86_init_noop;
70 x86_init.resources.probe_roms = x86_init_noop;
71
72 /* Avoid searching for BIOS MP tables */
73 x86_init.mpparse.find_mptable = x86_init_noop;
74 x86_init.mpparse.early_parse_smp_cfg = x86_init_noop;
75
76 x86_platform.get_wallclock = get_rtc_noop;
77 x86_platform.set_wallclock = set_rtc_noop;
78 x86_platform.get_nmi_reason = hv_get_nmi_reason;
79
80 x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT;
81 x86_platform.legacy.rtc = 0;
82 x86_platform.legacy.warm_reset = 0;
83 x86_platform.legacy.reserve_bios_regions = 0;
84 x86_platform.legacy.devices.pnpbios = 0;
85
86 x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id;
87 }
88
hv_vtl_system_desc_base(struct ldttss_desc * desc)89 static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc)
90 {
91 return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) |
92 (desc->base1 << 16) | desc->base0;
93 }
94
hv_vtl_system_desc_limit(struct ldttss_desc * desc)95 static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc)
96 {
97 return ((u32)desc->limit1 << 16) | (u32)desc->limit0;
98 }
99
100 typedef void (*secondary_startup_64_fn)(void*, void*);
hv_vtl_ap_entry(void)101 static void hv_vtl_ap_entry(void)
102 {
103 ((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params);
104 }
105
hv_vtl_bringup_vcpu(u32 target_vp_index,int cpu,u64 eip_ignored)106 static int hv_vtl_bringup_vcpu(u32 target_vp_index, int cpu, u64 eip_ignored)
107 {
108 u64 status;
109 int ret = 0;
110 struct hv_enable_vp_vtl *input;
111 unsigned long irq_flags;
112
113 struct desc_ptr gdt_ptr;
114 struct desc_ptr idt_ptr;
115
116 struct ldttss_desc *tss;
117 struct ldttss_desc *ldt;
118 struct desc_struct *gdt;
119
120 struct task_struct *idle = idle_thread_get(cpu);
121 u64 rsp = (unsigned long)idle->thread.sp;
122
123 u64 rip = (u64)&hv_vtl_ap_entry;
124
125 native_store_gdt(&gdt_ptr);
126 store_idt(&idt_ptr);
127
128 gdt = (struct desc_struct *)((void *)(gdt_ptr.address));
129 tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS);
130 ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT);
131
132 local_irq_save(irq_flags);
133
134 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
135 memset(input, 0, sizeof(*input));
136
137 input->partition_id = HV_PARTITION_ID_SELF;
138 input->vp_index = target_vp_index;
139 input->target_vtl.target_vtl = HV_VTL_MGMT;
140
141 /*
142 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit
143 * mode transition sequence after waking up an AP with SIPI whose
144 * vector points to the 16-bit AP startup trampoline code. Here in
145 * VTL2, we can't perform that sequence as the AP has to start in
146 * the 64-bit mode.
147 *
148 * To make this happen, we tell the hypervisor to load a valid 64-bit
149 * context (most of which is just magic numbers from the CPU manual)
150 * so that AP jumps right to the 64-bit entry of the kernel, and the
151 * control registers are loaded with values that let the AP fetch the
152 * code and data and carry on with work it gets assigned.
153 */
154
155 input->vp_context.rip = rip;
156 input->vp_context.rsp = rsp;
157 input->vp_context.rflags = 0x0000000000000002;
158 input->vp_context.efer = native_rdmsrq(MSR_EFER);
159 input->vp_context.cr0 = native_read_cr0();
160 input->vp_context.cr3 = __native_read_cr3();
161 input->vp_context.cr4 = native_read_cr4();
162 input->vp_context.msr_cr_pat = native_rdmsrq(MSR_IA32_CR_PAT);
163 input->vp_context.idtr.limit = idt_ptr.size;
164 input->vp_context.idtr.base = idt_ptr.address;
165 input->vp_context.gdtr.limit = gdt_ptr.size;
166 input->vp_context.gdtr.base = gdt_ptr.address;
167
168 /* Non-system desc (64bit), long, code, present */
169 input->vp_context.cs.selector = __KERNEL_CS;
170 input->vp_context.cs.base = 0;
171 input->vp_context.cs.limit = 0xffffffff;
172 input->vp_context.cs.attributes = 0xa09b;
173 /* Non-system desc (64bit), data, present, granularity, default */
174 input->vp_context.ss.selector = __KERNEL_DS;
175 input->vp_context.ss.base = 0;
176 input->vp_context.ss.limit = 0xffffffff;
177 input->vp_context.ss.attributes = 0xc093;
178
179 /* System desc (128bit), present, LDT */
180 input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8;
181 input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt);
182 input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt);
183 input->vp_context.ldtr.attributes = 0x82;
184
185 /* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */
186 input->vp_context.tr.selector = GDT_ENTRY_TSS * 8;
187 input->vp_context.tr.base = hv_vtl_system_desc_base(tss);
188 input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss);
189 input->vp_context.tr.attributes = 0x8b;
190
191 status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL);
192
193 if (!hv_result_success(status) &&
194 hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) {
195 pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]",
196 target_vp_index, status);
197 ret = -EINVAL;
198 goto free_lock;
199 }
200
201 status = hv_do_hypercall(HVCALL_START_VP, input, NULL);
202
203 if (!hv_result_success(status)) {
204 pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n",
205 target_vp_index, status);
206 ret = -EINVAL;
207 }
208
209 free_lock:
210 local_irq_restore(irq_flags);
211
212 return ret;
213 }
214
hv_vtl_wakeup_secondary_cpu(u32 apicid,unsigned long start_eip,unsigned int cpu)215 static int hv_vtl_wakeup_secondary_cpu(u32 apicid, unsigned long start_eip, unsigned int cpu)
216 {
217 int vp_index;
218
219 pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid);
220 vp_index = hv_apicid_to_vp_index(apicid);
221
222 if (vp_index < 0) {
223 pr_err("Couldn't find CPU with APIC ID %d\n", apicid);
224 return -EINVAL;
225 }
226 if (vp_index > ms_hyperv.max_vp_index) {
227 pr_err("Invalid CPU id %d for APIC ID %d\n", vp_index, apicid);
228 return -EINVAL;
229 }
230
231 return hv_vtl_bringup_vcpu(vp_index, cpu, start_eip);
232 }
233
hv_vtl_early_init(void)234 int __init hv_vtl_early_init(void)
235 {
236 machine_ops.emergency_restart = hv_vtl_emergency_restart;
237 machine_ops.restart = hv_vtl_restart;
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("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