xref: /linux/arch/x86/hyperv/hv_vtl.c (revision eb01fe7abbe2d0b38824d2a93fdb4cc3eaf2ccc1)
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 hv_vtl_real_mode_header;
18 
19 static bool __init hv_vtl_msi_ext_dest_id(void)
20 {
21 	return true;
22 }
23 
24 void __init hv_vtl_init_platform(void)
25 {
26 	pr_info("Linux runs in Hyper-V Virtual Trust Level\n");
27 
28 	x86_platform.realmode_reserve = x86_init_noop;
29 	x86_platform.realmode_init = x86_init_noop;
30 	x86_init.irqs.pre_vector_init = x86_init_noop;
31 	x86_init.timers.timer_init = x86_init_noop;
32 
33 	/* Avoid searching for BIOS MP tables */
34 	x86_init.mpparse.find_mptable = x86_init_noop;
35 	x86_init.mpparse.early_parse_smp_cfg = x86_init_noop;
36 	x86_init.mpparse.parse_smp_cfg = x86_init_noop;
37 
38 	x86_platform.get_wallclock = get_rtc_noop;
39 	x86_platform.set_wallclock = set_rtc_noop;
40 	x86_platform.get_nmi_reason = hv_get_nmi_reason;
41 
42 	x86_platform.legacy.i8042 = X86_LEGACY_I8042_PLATFORM_ABSENT;
43 	x86_platform.legacy.rtc = 0;
44 	x86_platform.legacy.warm_reset = 0;
45 	x86_platform.legacy.reserve_bios_regions = 0;
46 	x86_platform.legacy.devices.pnpbios = 0;
47 
48 	x86_init.hyper.msi_ext_dest_id = hv_vtl_msi_ext_dest_id;
49 }
50 
51 static inline u64 hv_vtl_system_desc_base(struct ldttss_desc *desc)
52 {
53 	return ((u64)desc->base3 << 32) | ((u64)desc->base2 << 24) |
54 		(desc->base1 << 16) | desc->base0;
55 }
56 
57 static inline u32 hv_vtl_system_desc_limit(struct ldttss_desc *desc)
58 {
59 	return ((u32)desc->limit1 << 16) | (u32)desc->limit0;
60 }
61 
62 typedef void (*secondary_startup_64_fn)(void*, void*);
63 static void hv_vtl_ap_entry(void)
64 {
65 	((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params);
66 }
67 
68 static int hv_vtl_bringup_vcpu(u32 target_vp_index, u64 eip_ignored)
69 {
70 	u64 status;
71 	int ret = 0;
72 	struct hv_enable_vp_vtl *input;
73 	unsigned long irq_flags;
74 
75 	struct desc_ptr gdt_ptr;
76 	struct desc_ptr idt_ptr;
77 
78 	struct ldttss_desc *tss;
79 	struct ldttss_desc *ldt;
80 	struct desc_struct *gdt;
81 
82 	u64 rsp = current->thread.sp;
83 	u64 rip = (u64)&hv_vtl_ap_entry;
84 
85 	native_store_gdt(&gdt_ptr);
86 	store_idt(&idt_ptr);
87 
88 	gdt = (struct desc_struct *)((void *)(gdt_ptr.address));
89 	tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS);
90 	ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT);
91 
92 	local_irq_save(irq_flags);
93 
94 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
95 	memset(input, 0, sizeof(*input));
96 
97 	input->partition_id = HV_PARTITION_ID_SELF;
98 	input->vp_index = target_vp_index;
99 	input->target_vtl.target_vtl = HV_VTL_MGMT;
100 
101 	/*
102 	 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit
103 	 * mode transition sequence after waking up an AP with SIPI whose
104 	 * vector points to the 16-bit AP startup trampoline code. Here in
105 	 * VTL2, we can't perform that sequence as the AP has to start in
106 	 * the 64-bit mode.
107 	 *
108 	 * To make this happen, we tell the hypervisor to load a valid 64-bit
109 	 * context (most of which is just magic numbers from the CPU manual)
110 	 * so that AP jumps right to the 64-bit entry of the kernel, and the
111 	 * control registers are loaded with values that let the AP fetch the
112 	 * code and data and carry on with work it gets assigned.
113 	 */
114 
115 	input->vp_context.rip = rip;
116 	input->vp_context.rsp = rsp;
117 	input->vp_context.rflags = 0x0000000000000002;
118 	input->vp_context.efer = __rdmsr(MSR_EFER);
119 	input->vp_context.cr0 = native_read_cr0();
120 	input->vp_context.cr3 = __native_read_cr3();
121 	input->vp_context.cr4 = native_read_cr4();
122 	input->vp_context.msr_cr_pat = __rdmsr(MSR_IA32_CR_PAT);
123 	input->vp_context.idtr.limit = idt_ptr.size;
124 	input->vp_context.idtr.base = idt_ptr.address;
125 	input->vp_context.gdtr.limit = gdt_ptr.size;
126 	input->vp_context.gdtr.base = gdt_ptr.address;
127 
128 	/* Non-system desc (64bit), long, code, present */
129 	input->vp_context.cs.selector = __KERNEL_CS;
130 	input->vp_context.cs.base = 0;
131 	input->vp_context.cs.limit = 0xffffffff;
132 	input->vp_context.cs.attributes = 0xa09b;
133 	/* Non-system desc (64bit), data, present, granularity, default */
134 	input->vp_context.ss.selector = __KERNEL_DS;
135 	input->vp_context.ss.base = 0;
136 	input->vp_context.ss.limit = 0xffffffff;
137 	input->vp_context.ss.attributes = 0xc093;
138 
139 	/* System desc (128bit), present, LDT */
140 	input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8;
141 	input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt);
142 	input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt);
143 	input->vp_context.ldtr.attributes = 0x82;
144 
145 	/* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */
146 	input->vp_context.tr.selector = GDT_ENTRY_TSS * 8;
147 	input->vp_context.tr.base = hv_vtl_system_desc_base(tss);
148 	input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss);
149 	input->vp_context.tr.attributes = 0x8b;
150 
151 	status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL);
152 
153 	if (!hv_result_success(status) &&
154 	    hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) {
155 		pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]",
156 		       target_vp_index, status);
157 		ret = -EINVAL;
158 		goto free_lock;
159 	}
160 
161 	status = hv_do_hypercall(HVCALL_START_VP, input, NULL);
162 
163 	if (!hv_result_success(status)) {
164 		pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n",
165 		       target_vp_index, status);
166 		ret = -EINVAL;
167 	}
168 
169 free_lock:
170 	local_irq_restore(irq_flags);
171 
172 	return ret;
173 }
174 
175 static int hv_vtl_apicid_to_vp_id(u32 apic_id)
176 {
177 	u64 control;
178 	u64 status;
179 	unsigned long irq_flags;
180 	struct hv_get_vp_from_apic_id_in *input;
181 	u32 *output, ret;
182 
183 	local_irq_save(irq_flags);
184 
185 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
186 	memset(input, 0, sizeof(*input));
187 	input->partition_id = HV_PARTITION_ID_SELF;
188 	input->apic_ids[0] = apic_id;
189 
190 	output = (u32 *)input;
191 
192 	control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_ID_FROM_APIC_ID;
193 	status = hv_do_hypercall(control, input, output);
194 	ret = output[0];
195 
196 	local_irq_restore(irq_flags);
197 
198 	if (!hv_result_success(status)) {
199 		pr_err("failed to get vp id from apic id %d, status %#llx\n",
200 		       apic_id, status);
201 		return -EINVAL;
202 	}
203 
204 	return ret;
205 }
206 
207 static int hv_vtl_wakeup_secondary_cpu(u32 apicid, unsigned long start_eip)
208 {
209 	int vp_id;
210 
211 	pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid);
212 	vp_id = hv_vtl_apicid_to_vp_id(apicid);
213 
214 	if (vp_id < 0) {
215 		pr_err("Couldn't find CPU with APIC ID %d\n", apicid);
216 		return -EINVAL;
217 	}
218 	if (vp_id > ms_hyperv.max_vp_index) {
219 		pr_err("Invalid CPU id %d for APIC ID %d\n", vp_id, apicid);
220 		return -EINVAL;
221 	}
222 
223 	return hv_vtl_bringup_vcpu(vp_id, start_eip);
224 }
225 
226 int __init hv_vtl_early_init(void)
227 {
228 	/*
229 	 * `boot_cpu_has` returns the runtime feature support,
230 	 * and here is the earliest it can be used.
231 	 */
232 	if (cpu_feature_enabled(X86_FEATURE_XSAVE))
233 		panic("XSAVE has to be disabled as it is not supported by this module.\n"
234 			  "Please add 'noxsave' to the kernel command line.\n");
235 
236 	real_mode_header = &hv_vtl_real_mode_header;
237 	apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu);
238 
239 	return 0;
240 }
241