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/fpu/api.h>
13 #include <asm/fpu/types.h>
14 #include <asm/i8259.h>
15 #include <asm/mshyperv.h>
16 #include <asm/msr.h>
17 #include <asm/realmode.h>
18 #include <asm/reboot.h>
19 #include <asm/smap.h>
20 #include <linux/export.h>
21 #include <../kernel/smpboot.h>
22 #include "../../kernel/fpu/legacy.h"
23
24 extern struct boot_params boot_params;
25 static struct real_mode_header hv_vtl_real_mode_header;
26
hv_vtl_msi_ext_dest_id(void)27 static bool __init hv_vtl_msi_ext_dest_id(void)
28 {
29 return true;
30 }
31
32 /*
33 * The `native_machine_emergency_restart` function from `reboot.c` writes
34 * to the physical address 0x472 to indicate the type of reboot for the
35 * firmware. We cannot have that in VSM as the memory composition might
36 * be more generic, and such write effectively corrupts the memory thus
37 * making diagnostics harder at the very least.
38 */
hv_vtl_emergency_restart(void)39 static void __noreturn hv_vtl_emergency_restart(void)
40 {
41 /*
42 * Cause a triple fault and the immediate reset. Here the code does not run
43 * on the top of any firmware, whereby cannot reach out to its services.
44 * The inifinite loop is for the improbable case that the triple fault does
45 * not work and have to preserve the state intact for debugging.
46 */
47 for (;;) {
48 idt_invalidate();
49 __asm__ __volatile__("int3");
50 }
51 }
52
53 /*
54 * The only way to restart in the VTL mode is to triple fault as the kernel runs
55 * as firmware.
56 */
hv_vtl_restart(char __maybe_unused * cmd)57 static void __noreturn hv_vtl_restart(char __maybe_unused *cmd)
58 {
59 hv_vtl_emergency_restart();
60 }
61
hv_vtl_init_platform(void)62 void __init hv_vtl_init_platform(void)
63 {
64 /*
65 * This function is a no-op if the VTL mode is not enabled.
66 * If it is, this function runs if and only the kernel boots in
67 * VTL2 which the x86 hv initialization path makes sure of.
68 */
69 pr_info("Linux runs in Hyper-V Virtual Trust Level %d\n", ms_hyperv.vtl);
70
71 x86_platform.realmode_reserve = x86_init_noop;
72 x86_platform.realmode_init = x86_init_noop;
73 x86_init.irqs.pre_vector_init = x86_init_noop;
74 x86_init.timers.timer_init = x86_init_noop;
75 x86_init.resources.probe_roms = x86_init_noop;
76
77 /* Avoid searching for BIOS MP tables */
78 x86_init.mpparse.find_mptable = x86_init_noop;
79 x86_init.mpparse.early_parse_smp_cfg = x86_init_noop;
80
81 x86_platform.get_wallclock = get_rtc_noop;
82 x86_platform.set_wallclock = set_rtc_noop;
83 x86_platform.get_nmi_reason = hv_get_nmi_reason;
84
85 x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT;
86 x86_platform.legacy.rtc = 0;
87 x86_platform.legacy.warm_reset = 0;
88 x86_platform.legacy.reserve_bios_regions = 0;
89 x86_platform.legacy.devices.pnpbios = 0;
90
91 x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id;
92 }
93
hv_vtl_system_desc_base(struct ldttss_desc * desc)94 static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc)
95 {
96 return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) |
97 (desc->base1 << 16) | desc->base0;
98 }
99
hv_vtl_system_desc_limit(struct ldttss_desc * desc)100 static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc)
101 {
102 return ((u32)desc->limit1 << 16) | (u32)desc->limit0;
103 }
104
105 typedef void (*secondary_startup_64_fn)(void*, void*);
hv_vtl_ap_entry(void)106 static void hv_vtl_ap_entry(void)
107 {
108 ((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params);
109 }
110
hv_vtl_bringup_vcpu(u32 target_vp_index,int cpu,u64 eip_ignored)111 static int hv_vtl_bringup_vcpu(u32 target_vp_index, int cpu, u64 eip_ignored)
112 {
113 u64 status, rsp, rip;
114 int ret = 0;
115 struct hv_enable_vp_vtl *input;
116 unsigned long irq_flags;
117
118 struct desc_ptr gdt_ptr;
119 struct desc_ptr idt_ptr;
120
121 struct ldttss_desc *tss;
122 struct ldttss_desc *ldt;
123 struct desc_struct *gdt;
124
125 struct task_struct *idle = idle_thread_get(cpu);
126 if (IS_ERR(idle))
127 return PTR_ERR(idle);
128
129 rsp = (unsigned long)idle->thread.sp;
130 rip = (u64)&hv_vtl_ap_entry;
131
132 native_store_gdt(&gdt_ptr);
133 store_idt(&idt_ptr);
134
135 gdt = (struct desc_struct *)((void *)(gdt_ptr.address));
136 tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS);
137 ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT);
138
139 local_irq_save(irq_flags);
140
141 input = *this_cpu_ptr(hyperv_pcpu_input_arg);
142 memset(input, 0, sizeof(*input));
143
144 input->partition_id = HV_PARTITION_ID_SELF;
145 input->vp_index = target_vp_index;
146 input->target_vtl.target_vtl = HV_VTL_MGMT;
147
148 /*
149 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit
150 * mode transition sequence after waking up an AP with SIPI whose
151 * vector points to the 16-bit AP startup trampoline code. Here in
152 * VTL2, we can't perform that sequence as the AP has to start in
153 * the 64-bit mode.
154 *
155 * To make this happen, we tell the hypervisor to load a valid 64-bit
156 * context (most of which is just magic numbers from the CPU manual)
157 * so that AP jumps right to the 64-bit entry of the kernel, and the
158 * control registers are loaded with values that let the AP fetch the
159 * code and data and carry on with work it gets assigned.
160 */
161
162 input->vp_context.rip = rip;
163 input->vp_context.rsp = rsp;
164 input->vp_context.rflags = 0x0000000000000002;
165 input->vp_context.efer = native_rdmsrq(MSR_EFER);
166 input->vp_context.cr0 = native_read_cr0();
167 input->vp_context.cr3 = __native_read_cr3();
168 input->vp_context.cr4 = native_read_cr4();
169 input->vp_context.msr_cr_pat = native_rdmsrq(MSR_IA32_CR_PAT);
170 input->vp_context.idtr.limit = idt_ptr.size;
171 input->vp_context.idtr.base = idt_ptr.address;
172 input->vp_context.gdtr.limit = gdt_ptr.size;
173 input->vp_context.gdtr.base = gdt_ptr.address;
174
175 /* Non-system desc (64bit), long, code, present */
176 input->vp_context.cs.selector = __KERNEL_CS;
177 input->vp_context.cs.base = 0;
178 input->vp_context.cs.limit = 0xffffffff;
179 input->vp_context.cs.attributes = 0xa09b;
180 /* Non-system desc (64bit), data, present, granularity, default */
181 input->vp_context.ss.selector = __KERNEL_DS;
182 input->vp_context.ss.base = 0;
183 input->vp_context.ss.limit = 0xffffffff;
184 input->vp_context.ss.attributes = 0xc093;
185
186 /* System desc (128bit), present, LDT */
187 input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8;
188 input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt);
189 input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt);
190 input->vp_context.ldtr.attributes = 0x82;
191
192 /* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */
193 input->vp_context.tr.selector = GDT_ENTRY_TSS * 8;
194 input->vp_context.tr.base = hv_vtl_system_desc_base(tss);
195 input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss);
196 input->vp_context.tr.attributes = 0x8b;
197
198 status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL);
199
200 if (!hv_result_success(status) &&
201 hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) {
202 pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]",
203 target_vp_index, status);
204 ret = -EINVAL;
205 goto free_lock;
206 }
207
208 status = hv_do_hypercall(HVCALL_START_VP, input, NULL);
209
210 if (!hv_result_success(status)) {
211 pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n",
212 target_vp_index, status);
213 ret = -EINVAL;
214 }
215
216 free_lock:
217 local_irq_restore(irq_flags);
218
219 return ret;
220 }
221
hv_vtl_wakeup_secondary_cpu(u32 apicid,unsigned long start_eip,unsigned int cpu)222 static int hv_vtl_wakeup_secondary_cpu(u32 apicid, unsigned long start_eip, unsigned int cpu)
223 {
224 int vp_index;
225
226 pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid);
227 vp_index = hv_apicid_to_vp_index(apicid);
228
229 if (vp_index < 0) {
230 pr_err("Couldn't find CPU with APIC ID %d\n", apicid);
231 return -EINVAL;
232 }
233 if (vp_index > ms_hyperv.max_vp_index) {
234 pr_err("Invalid CPU id %d for APIC ID %d\n", vp_index, apicid);
235 return -EINVAL;
236 }
237
238 return hv_vtl_bringup_vcpu(vp_index, cpu, start_eip);
239 }
240
hv_vtl_early_init(void)241 int __init hv_vtl_early_init(void)
242 {
243 machine_ops.emergency_restart = hv_vtl_emergency_restart;
244 machine_ops.restart = hv_vtl_restart;
245
246 /*
247 * `boot_cpu_has` returns the runtime feature support,
248 * and here is the earliest it can be used.
249 */
250 if (cpu_feature_enabled(X86_FEATURE_XSAVE))
251 panic("XSAVE has to be disabled as it is not supported by this module.\n"
252 "Please add 'noxsave' to the kernel command line.\n");
253
254 real_mode_header = &hv_vtl_real_mode_header;
255 apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu);
256
257 return 0;
258 }
259
260 DEFINE_STATIC_CALL_NULL(__mshv_vtl_return_hypercall, void (*)(void));
261
mshv_vtl_return_call_init(u64 vtl_return_offset)262 void mshv_vtl_return_call_init(u64 vtl_return_offset)
263 {
264 static_call_update(__mshv_vtl_return_hypercall,
265 (void *)((u8 *)hv_hypercall_pg + vtl_return_offset));
266 }
267 EXPORT_SYMBOL(mshv_vtl_return_call_init);
268
mshv_vtl_return_call(struct mshv_vtl_cpu_context * vtl0)269 void mshv_vtl_return_call(struct mshv_vtl_cpu_context *vtl0)
270 {
271 struct hv_vp_assist_page *hvp;
272
273 hvp = hv_vp_assist_page[smp_processor_id()];
274 hvp->vtl_ret_x64rax = vtl0->rax;
275 hvp->vtl_ret_x64rcx = vtl0->rcx;
276
277 kernel_fpu_begin_mask(0);
278 fxrstor(&vtl0->fx_state);
279 __mshv_vtl_return_call(vtl0);
280 fxsave(&vtl0->fx_state);
281 kernel_fpu_end();
282 }
283 EXPORT_SYMBOL(mshv_vtl_return_call);
284