xref: /openbmc/linux/arch/x86/hyperv/hv_vtl.c (revision 3d6af30f)
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 hv_vtl_real_mode_header;
19 
hv_vtl_msi_ext_dest_id(void)20 static bool __init hv_vtl_msi_ext_dest_id(void)
21 {
22 	return true;
23 }
24 
hv_vtl_init_platform(void)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_smp_config = x86_init_noop;
36 	x86_init.mpparse.get_smp_config = x86_init_uint_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 
hv_vtl_system_desc_base(struct ldttss_desc * desc)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 
hv_vtl_system_desc_limit(struct ldttss_desc * desc)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*);
hv_vtl_ap_entry(void)63 static void hv_vtl_ap_entry(void)
64 {
65 	((secondary_startup_64_fn)secondary_startup_64)(&boot_params, &boot_params);
66 }
67 
hv_vtl_bringup_vcpu(u32 target_vp_index,int cpu,u64 eip_ignored)68 static int hv_vtl_bringup_vcpu(u32 target_vp_index, int cpu, 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 	struct task_struct *idle = idle_thread_get(cpu);
83 	u64 rsp = (unsigned long)idle->thread.sp;
84 
85 	u64 rip = (u64)&hv_vtl_ap_entry;
86 
87 	native_store_gdt(&gdt_ptr);
88 	store_idt(&idt_ptr);
89 
90 	gdt = (struct desc_struct *)((void *)(gdt_ptr.address));
91 	tss = (struct ldttss_desc *)(gdt + GDT_ENTRY_TSS);
92 	ldt = (struct ldttss_desc *)(gdt + GDT_ENTRY_LDT);
93 
94 	local_irq_save(irq_flags);
95 
96 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
97 	memset(input, 0, sizeof(*input));
98 
99 	input->partition_id = HV_PARTITION_ID_SELF;
100 	input->vp_index = target_vp_index;
101 	input->target_vtl.target_vtl = HV_VTL_MGMT;
102 
103 	/*
104 	 * The x86_64 Linux kernel follows the 16-bit -> 32-bit -> 64-bit
105 	 * mode transition sequence after waking up an AP with SIPI whose
106 	 * vector points to the 16-bit AP startup trampoline code. Here in
107 	 * VTL2, we can't perform that sequence as the AP has to start in
108 	 * the 64-bit mode.
109 	 *
110 	 * To make this happen, we tell the hypervisor to load a valid 64-bit
111 	 * context (most of which is just magic numbers from the CPU manual)
112 	 * so that AP jumps right to the 64-bit entry of the kernel, and the
113 	 * control registers are loaded with values that let the AP fetch the
114 	 * code and data and carry on with work it gets assigned.
115 	 */
116 
117 	input->vp_context.rip = rip;
118 	input->vp_context.rsp = rsp;
119 	input->vp_context.rflags = 0x0000000000000002;
120 	input->vp_context.efer = __rdmsr(MSR_EFER);
121 	input->vp_context.cr0 = native_read_cr0();
122 	input->vp_context.cr3 = __native_read_cr3();
123 	input->vp_context.cr4 = native_read_cr4();
124 	input->vp_context.msr_cr_pat = __rdmsr(MSR_IA32_CR_PAT);
125 	input->vp_context.idtr.limit = idt_ptr.size;
126 	input->vp_context.idtr.base = idt_ptr.address;
127 	input->vp_context.gdtr.limit = gdt_ptr.size;
128 	input->vp_context.gdtr.base = gdt_ptr.address;
129 
130 	/* Non-system desc (64bit), long, code, present */
131 	input->vp_context.cs.selector = __KERNEL_CS;
132 	input->vp_context.cs.base = 0;
133 	input->vp_context.cs.limit = 0xffffffff;
134 	input->vp_context.cs.attributes = 0xa09b;
135 	/* Non-system desc (64bit), data, present, granularity, default */
136 	input->vp_context.ss.selector = __KERNEL_DS;
137 	input->vp_context.ss.base = 0;
138 	input->vp_context.ss.limit = 0xffffffff;
139 	input->vp_context.ss.attributes = 0xc093;
140 
141 	/* System desc (128bit), present, LDT */
142 	input->vp_context.ldtr.selector = GDT_ENTRY_LDT * 8;
143 	input->vp_context.ldtr.base = hv_vtl_system_desc_base(ldt);
144 	input->vp_context.ldtr.limit = hv_vtl_system_desc_limit(ldt);
145 	input->vp_context.ldtr.attributes = 0x82;
146 
147 	/* System desc (128bit), present, TSS, 0x8b - busy, 0x89 -- default */
148 	input->vp_context.tr.selector = GDT_ENTRY_TSS * 8;
149 	input->vp_context.tr.base = hv_vtl_system_desc_base(tss);
150 	input->vp_context.tr.limit = hv_vtl_system_desc_limit(tss);
151 	input->vp_context.tr.attributes = 0x8b;
152 
153 	status = hv_do_hypercall(HVCALL_ENABLE_VP_VTL, input, NULL);
154 
155 	if (!hv_result_success(status) &&
156 	    hv_result(status) != HV_STATUS_VTL_ALREADY_ENABLED) {
157 		pr_err("HVCALL_ENABLE_VP_VTL failed for VP : %d ! [Err: %#llx\n]",
158 		       target_vp_index, status);
159 		ret = -EINVAL;
160 		goto free_lock;
161 	}
162 
163 	status = hv_do_hypercall(HVCALL_START_VP, input, NULL);
164 
165 	if (!hv_result_success(status)) {
166 		pr_err("HVCALL_START_VP failed for VP : %d ! [Err: %#llx]\n",
167 		       target_vp_index, status);
168 		ret = -EINVAL;
169 	}
170 
171 free_lock:
172 	local_irq_restore(irq_flags);
173 
174 	return ret;
175 }
176 
hv_vtl_apicid_to_vp_id(u32 apic_id)177 static int hv_vtl_apicid_to_vp_id(u32 apic_id)
178 {
179 	u64 control;
180 	u64 status;
181 	unsigned long irq_flags;
182 	struct hv_get_vp_from_apic_id_in *input;
183 	u32 *output, ret;
184 
185 	local_irq_save(irq_flags);
186 
187 	input = *this_cpu_ptr(hyperv_pcpu_input_arg);
188 	memset(input, 0, sizeof(*input));
189 	input->partition_id = HV_PARTITION_ID_SELF;
190 	input->apic_ids[0] = apic_id;
191 
192 	output = (u32 *)input;
193 
194 	control = HV_HYPERCALL_REP_COMP_1 | HVCALL_GET_VP_ID_FROM_APIC_ID;
195 	status = hv_do_hypercall(control, input, output);
196 	ret = output[0];
197 
198 	local_irq_restore(irq_flags);
199 
200 	if (!hv_result_success(status)) {
201 		pr_err("failed to get vp id from apic id %d, status %#llx\n",
202 		       apic_id, status);
203 		return -EINVAL;
204 	}
205 
206 	return ret;
207 }
208 
hv_vtl_wakeup_secondary_cpu(int apicid,unsigned long start_eip)209 static int hv_vtl_wakeup_secondary_cpu(int apicid, unsigned long start_eip)
210 {
211 	int vp_id, cpu;
212 
213 	/* Find the logical CPU for the APIC ID */
214 	for_each_present_cpu(cpu) {
215 		if (arch_match_cpu_phys_id(cpu, apicid))
216 			break;
217 	}
218 	if (cpu >= nr_cpu_ids)
219 		return -EINVAL;
220 
221 	pr_debug("Bringing up CPU with APIC ID %d in VTL2...\n", apicid);
222 	vp_id = hv_vtl_apicid_to_vp_id(apicid);
223 
224 	if (vp_id < 0) {
225 		pr_err("Couldn't find CPU with APIC ID %d\n", apicid);
226 		return -EINVAL;
227 	}
228 	if (vp_id > ms_hyperv.max_vp_index) {
229 		pr_err("Invalid CPU id %d for APIC ID %d\n", vp_id, apicid);
230 		return -EINVAL;
231 	}
232 
233 	return hv_vtl_bringup_vcpu(vp_id, cpu, start_eip);
234 }
235 
hv_vtl_early_init(void)236 int __init hv_vtl_early_init(void)
237 {
238 	/*
239 	 * `boot_cpu_has` returns the runtime feature support,
240 	 * and here is the earliest it can be used.
241 	 */
242 	if (cpu_feature_enabled(X86_FEATURE_XSAVE))
243 		panic("XSAVE has to be disabled as it is not supported by this module.\n"
244 			  "Please add 'noxsave' to the kernel command line.\n");
245 
246 	real_mode_header = &hv_vtl_real_mode_header;
247 	apic_update_callback(wakeup_secondary_cpu_64, hv_vtl_wakeup_secondary_cpu);
248 
249 	return 0;
250 }
251