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 <asm/reboot.h>
16 #include <../kernel/smpboot.h>
17
18 extern struct boot_params boot_params;
19 static struct real_mode_header hv_vtl_real_mode_header;
20
hv_vtl_msi_ext_dest_id(void)21 static bool __init hv_vtl_msi_ext_dest_id(void)
22 {
23 return true;
24 }
25
26 /*
27 * The `native_machine_emergency_restart` function from `reboot.c` writes
28 * to the physical address 0x472 to indicate the type of reboot for the
29 * firmware. We cannot have that in VSM as the memory composition might
30 * be more generic, and such write effectively corrupts the memory thus
31 * making diagnostics harder at the very least.
32 */
hv_vtl_emergency_restart(void)33 static void __noreturn hv_vtl_emergency_restart(void)
34 {
35 /*
36 * Cause a triple fault and the immediate reset. Here the code does not run
37 * on the top of any firmware, whereby cannot reach out to its services.
38 * The inifinite loop is for the improbable case that the triple fault does
39 * not work and have to preserve the state intact for debugging.
40 */
41 for (;;) {
42 idt_invalidate();
43 __asm__ __volatile__("int3");
44 }
45 }
46
47 /*
48 * The only way to restart in the VTL mode is to triple fault as the kernel runs
49 * as firmware.
50 */
hv_vtl_restart(char __maybe_unused * cmd)51 static void __noreturn hv_vtl_restart(char __maybe_unused *cmd)
52 {
53 hv_vtl_emergency_restart();
54 }
55
hv_vtl_init_platform(void)56 void __init hv_vtl_init_platform(void)
57 {
58 pr_info("Linux runs in Hyper-V Virtual Trust Level\n");
59
60 x86_platform.realmode_reserve = x86_init_noop;
61 x86_platform.realmode_init = x86_init_noop;
62 x86_init.irqs.pre_vector_init = x86_init_noop;
63 x86_init.timers.timer_init = x86_init_noop;
64 x86_init.resources.probe_roms = x86_init_noop;
65
66 /* Avoid searching for BIOS MP tables */
67 x86_init.mpparse.find_mptable = x86_init_noop;
68 x86_init.mpparse.early_parse_smp_cfg = x86_init_noop;
69
70 x86_platform.get_wallclock = get_rtc_noop;
71 x86_platform.set_wallclock = set_rtc_noop;
72 x86_platform.get_nmi_reason = hv_get_nmi_reason;
73
74 x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT;
75 x86_platform.legacy.rtc = 0;
76 x86_platform.legacy.warm_reset = 0;
77 x86_platform.legacy.reserve_bios_regions = 0;
78 x86_platform.legacy.devices.pnpbios = 0;
79
80 x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id;
81 }
82
hv_vtl_system_desc_base(struct ldttss_desc * desc)83 static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc)
84 {
85 return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) |
86 (desc->base1 << 16) | desc->base0;
87 }
88
hv_vtl_system_desc_limit(struct ldttss_desc * desc)89 static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc)
90 {
91 return ((u32)desc->limit1 << 16) | (u32)desc->limit0;
92 }
93
94 typedef void (*secondary_startup_64_fn)(void*, void*);
hv_vtl_ap_entry(void)95 static void hv_vtl_ap_entry(void)
96 {
97 ((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params);
98 }
99
hv_vtl_bringup_vcpu(u32 target_vp_index,int cpu,u64 eip_ignored)100 static int hv_vtl_bringup_vcpu(u32 target_vp_index, int cpu, u64 eip_ignored)
101 {
102 u64 status;
103 int ret = 0;
104 struct hv_enable_vp_vtl *input;
105 unsigned long irq_flags;
106
107 struct desc_ptr gdt_ptr;
108 struct desc_ptr idt_ptr;
109
110 struct ldttss_desc *tss;
111 struct ldttss_desc *ldt;
112 struct desc_struct *gdt;
113
114 struct task_struct *idle = idle_thread_get(cpu);
115 u64 rsp = (unsigned long)idle->thread.sp;
116
117 u64 rip = (u64)&hv_vtl_ap_entry;
118
119 native_store_gdt(&gdt_ptr);
120 store_idt(&idt_ptr);
121
122 gdt = (struct desc_struct *)((void *)(gdt_ptr.address));
123 tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS);
124 ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT);
125
126 local_irq_save(irq_flags);
127
128 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
129 memset(input, 0, sizeof(*input));
130
131 input->partition_id = HV_PARTITION_ID_SELF;
132 input->vp_index = target_vp_index;
133 input->target_vtl.target_vtl = HV_VTL_MGMT;
134
135 /*
136 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit
137 * mode transition sequence after waking up an AP with SIPI whose
138 * vector points to the 16-bit AP startup trampoline code. Here in
139 * VTL2, we can't perform that sequence as the AP has to start in
140 * the 64-bit mode.
141 *
142 * To make this happen, we tell the hypervisor to load a valid 64-bit
143 * context (most of which is just magic numbers from the CPU manual)
144 * so that AP jumps right to the 64-bit entry of the kernel, and the
145 * control registers are loaded with values that let the AP fetch the
146 * code and data and carry on with work it gets assigned.
147 */
148
149 input->vp_context.rip = rip;
150 input->vp_context.rsp = rsp;
151 input->vp_context.rflags = 0x0000000000000002;
152 input->vp_context.efer = __rdmsr(MSR_EFER);
153 input->vp_context.cr0 = native_read_cr0();
154 input->vp_context.cr3 = __native_read_cr3();
155 input->vp_context.cr4 = native_read_cr4();
156 input->vp_context.msr_cr_pat = __rdmsr(MSR_IA32_CR_PAT);
157 input->vp_context.idtr.limit = idt_ptr.size;
158 input->vp_context.idtr.base = idt_ptr.address;
159 input->vp_context.gdtr.limit = gdt_ptr.size;
160 input->vp_context.gdtr.base = gdt_ptr.address;
161
162 /* Non-system desc (64bit), long, code, present */
163 input->vp_context.cs.selector = __KERNEL_CS;
164 input->vp_context.cs.base = 0;
165 input->vp_context.cs.limit = 0xffffffff;
166 input->vp_context.cs.attributes = 0xa09b;
167 /* Non-system desc (64bit), data, present, granularity, default */
168 input->vp_context.ss.selector = __KERNEL_DS;
169 input->vp_context.ss.base = 0;
170 input->vp_context.ss.limit = 0xffffffff;
171 input->vp_context.ss.attributes = 0xc093;
172
173 /* System desc (128bit), present, LDT */
174 input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8;
175 input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt);
176 input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt);
177 input->vp_context.ldtr.attributes = 0x82;
178
179 /* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */
180 input->vp_context.tr.selector = GDT_ENTRY_TSS * 8;
181 input->vp_context.tr.base = hv_vtl_system_desc_base(tss);
182 input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss);
183 input->vp_context.tr.attributes = 0x8b;
184
185 status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL);
186
187 if (!hv_result_success(status) &&
188 hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) {
189 pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]",
190 target_vp_index, status);
191 ret = -EINVAL;
192 goto free_lock;
193 }
194
195 status = hv_do_hypercall(HVCALL_START_VP, input, NULL);
196
197 if (!hv_result_success(status)) {
198 pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n",
199 target_vp_index, status);
200 ret = -EINVAL;
201 }
202
203 free_lock:
204 local_irq_restore(irq_flags);
205
206 return ret;
207 }
208
hv_vtl_apicid_to_vp_id(u32 apic_id)209 static int hv_vtl_apicid_to_vp_id(u32 apic_id)
210 {
211 u64 control;
212 u64 status;
213 unsigned long irq_flags;
214 struct hv_get_vp_from_apic_id_in *input;
215 u32 *output, ret;
216
217 local_irq_save(irq_flags);
218
219 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
220 memset(input, 0, sizeof(*input));
221 input->partition_id = HV_PARTITION_ID_SELF;
222 input->apic_ids[0] = apic_id;
223
224 output = *this_cpu_ptr(hyperv_pcpu_output_arg);
225
226 control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_ID_FROM_APIC_ID;
227 status = hv_do_hypercall(control, input, output);
228 ret = output[0];
229
230 local_irq_restore(irq_flags);
231
232 if (!hv_result_success(status)) {
233 pr_err("failed to get vp id from apic id %d, status %#llx\n",
234 apic_id, status);
235 return -EINVAL;
236 }
237
238 return ret;
239 }
240
hv_vtl_wakeup_secondary_cpu(u32 apicid,unsigned long start_eip)241 static int hv_vtl_wakeup_secondary_cpu(u32 apicid, unsigned long start_eip)
242 {
243 int vp_id, cpu;
244
245 /* Find the logical CPU for the APIC ID */
246 for_each_present_cpu(cpu) {
247 if (arch_match_cpu_phys_id(cpu, apicid))
248 break;
249 }
250 if (cpu >= nr_cpu_ids)
251 return -EINVAL;
252
253 pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid);
254 vp_id = hv_vtl_apicid_to_vp_id(apicid);
255
256 if (vp_id < 0) {
257 pr_err("Couldn't find CPU with APIC ID %d\n", apicid);
258 return -EINVAL;
259 }
260 if (vp_id > ms_hyperv.max_vp_index) {
261 pr_err("Invalid CPU id %d for APIC ID %d\n", vp_id, apicid);
262 return -EINVAL;
263 }
264
265 return hv_vtl_bringup_vcpu(vp_id, cpu, start_eip);
266 }
267
hv_vtl_early_init(void)268 int __init hv_vtl_early_init(void)
269 {
270 machine_ops.emergency_restart = hv_vtl_emergency_restart;
271 machine_ops.restart = hv_vtl_restart;
272
273 /*
274 * `boot_cpu_has` returns the runtime feature support,
275 * and here is the earliest it can be used.
276 */
277 if (cpu_feature_enabled(X86_FEATURE_XSAVE))
278 panic("XSAVE has to be disabled as it is not supported by this module.\n"
279 "Please add 'noxsave' to the kernel command line.\n");
280
281 real_mode_header = &hv_vtl_real_mode_header;
282 apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu);
283
284 return 0;
285 }
286