1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * VMX-preemption timer test
4  *
5  * Copyright (C) 2020, Google, LLC.
6  *
7  * Test to ensure the VM-Enter after migration doesn't
8  * incorrectly restarts the timer with the full timer
9  * value instead of partially decayed timer value
10  *
11  */
12 #define _GNU_SOURCE /* for program_invocation_short_name */
13 #include <fcntl.h>
14 #include <stdio.h>
15 #include <stdlib.h>
16 #include <string.h>
17 #include <sys/ioctl.h>
18 
19 #include "test_util.h"
20 
21 #include "kvm_util.h"
22 #include "processor.h"
23 #include "vmx.h"
24 
25 #define VCPU_ID		5
26 #define PREEMPTION_TIMER_VALUE			100000000ull
27 #define PREEMPTION_TIMER_VALUE_THRESHOLD1	 80000000ull
28 
29 u32 vmx_pt_rate;
30 bool l2_save_restore_done;
31 static u64 l2_vmx_pt_start;
32 volatile u64 l2_vmx_pt_finish;
33 
34 union vmx_basic basic;
35 union vmx_ctrl_msr ctrl_pin_rev;
36 union vmx_ctrl_msr ctrl_exit_rev;
37 
38 void l2_guest_code(void)
39 {
40 	u64 vmx_pt_delta;
41 
42 	vmcall();
43 	l2_vmx_pt_start = (rdtsc() >> vmx_pt_rate) << vmx_pt_rate;
44 
45 	/*
46 	 * Wait until the 1st threshold has passed
47 	 */
48 	do {
49 		l2_vmx_pt_finish = rdtsc();
50 		vmx_pt_delta = (l2_vmx_pt_finish - l2_vmx_pt_start) >>
51 				vmx_pt_rate;
52 	} while (vmx_pt_delta < PREEMPTION_TIMER_VALUE_THRESHOLD1);
53 
54 	/*
55 	 * Force L2 through Save and Restore cycle
56 	 */
57 	GUEST_SYNC(1);
58 
59 	l2_save_restore_done = 1;
60 
61 	/*
62 	 * Now wait for the preemption timer to fire and
63 	 * exit to L1
64 	 */
65 	while ((l2_vmx_pt_finish = rdtsc()))
66 		;
67 }
68 
69 void l1_guest_code(struct vmx_pages *vmx_pages)
70 {
71 #define L2_GUEST_STACK_SIZE 64
72 	unsigned long l2_guest_stack[L2_GUEST_STACK_SIZE];
73 	u64 l1_vmx_pt_start;
74 	u64 l1_vmx_pt_finish;
75 	u64 l1_tsc_deadline, l2_tsc_deadline;
76 
77 	GUEST_ASSERT(vmx_pages->vmcs_gpa);
78 	GUEST_ASSERT(prepare_for_vmx_operation(vmx_pages));
79 	GUEST_ASSERT(load_vmcs(vmx_pages));
80 	GUEST_ASSERT(vmptrstz() == vmx_pages->vmcs_gpa);
81 
82 	prepare_vmcs(vmx_pages, l2_guest_code,
83 		     &l2_guest_stack[L2_GUEST_STACK_SIZE]);
84 
85 	/*
86 	 * Check for Preemption timer support
87 	 */
88 	basic.val = rdmsr(MSR_IA32_VMX_BASIC);
89 	ctrl_pin_rev.val = rdmsr(basic.ctrl ? MSR_IA32_VMX_TRUE_PINBASED_CTLS
90 			: MSR_IA32_VMX_PINBASED_CTLS);
91 	ctrl_exit_rev.val = rdmsr(basic.ctrl ? MSR_IA32_VMX_TRUE_EXIT_CTLS
92 			: MSR_IA32_VMX_EXIT_CTLS);
93 
94 	if (!(ctrl_pin_rev.clr & PIN_BASED_VMX_PREEMPTION_TIMER) ||
95 	    !(ctrl_exit_rev.clr & VM_EXIT_SAVE_VMX_PREEMPTION_TIMER))
96 		return;
97 
98 	GUEST_ASSERT(!vmlaunch());
99 	GUEST_ASSERT(vmreadz(VM_EXIT_REASON) == EXIT_REASON_VMCALL);
100 	vmwrite(GUEST_RIP, vmreadz(GUEST_RIP) + vmreadz(VM_EXIT_INSTRUCTION_LEN));
101 
102 	/*
103 	 * Turn on PIN control and resume the guest
104 	 */
105 	GUEST_ASSERT(!vmwrite(PIN_BASED_VM_EXEC_CONTROL,
106 			      vmreadz(PIN_BASED_VM_EXEC_CONTROL) |
107 			      PIN_BASED_VMX_PREEMPTION_TIMER));
108 
109 	GUEST_ASSERT(!vmwrite(VMX_PREEMPTION_TIMER_VALUE,
110 			      PREEMPTION_TIMER_VALUE));
111 
112 	vmx_pt_rate = rdmsr(MSR_IA32_VMX_MISC) & 0x1F;
113 
114 	l2_save_restore_done = 0;
115 
116 	l1_vmx_pt_start = (rdtsc() >> vmx_pt_rate) << vmx_pt_rate;
117 
118 	GUEST_ASSERT(!vmresume());
119 
120 	l1_vmx_pt_finish = rdtsc();
121 
122 	/*
123 	 * Ensure exit from L2 happens after L2 goes through
124 	 * save and restore
125 	 */
126 	GUEST_ASSERT(l2_save_restore_done);
127 
128 	/*
129 	 * Ensure the exit from L2 is due to preemption timer expiry
130 	 */
131 	GUEST_ASSERT(vmreadz(VM_EXIT_REASON) == EXIT_REASON_PREEMPTION_TIMER);
132 
133 	l1_tsc_deadline = l1_vmx_pt_start +
134 		(PREEMPTION_TIMER_VALUE << vmx_pt_rate);
135 
136 	l2_tsc_deadline = l2_vmx_pt_start +
137 		(PREEMPTION_TIMER_VALUE << vmx_pt_rate);
138 
139 	/*
140 	 * Sync with the host and pass the l1|l2 pt_expiry_finish times and
141 	 * tsc deadlines so that host can verify they are as expected
142 	 */
143 	GUEST_SYNC_ARGS(2, l1_vmx_pt_finish, l1_tsc_deadline,
144 		l2_vmx_pt_finish, l2_tsc_deadline);
145 }
146 
147 void guest_code(struct vmx_pages *vmx_pages)
148 {
149 	if (vmx_pages)
150 		l1_guest_code(vmx_pages);
151 
152 	GUEST_DONE();
153 }
154 
155 int main(int argc, char *argv[])
156 {
157 	vm_vaddr_t vmx_pages_gva = 0;
158 
159 	struct kvm_regs regs1, regs2;
160 	struct kvm_vm *vm;
161 	struct kvm_run *run;
162 	struct kvm_x86_state *state;
163 	struct ucall uc;
164 	int stage;
165 
166 	/*
167 	 * AMD currently does not implement any VMX features, so for now we
168 	 * just early out.
169 	 */
170 	nested_vmx_check_supported();
171 
172 	if (!kvm_check_cap(KVM_CAP_NESTED_STATE)) {
173 		print_skip("KVM_CAP_NESTED_STATE not supported");
174 		exit(KSFT_SKIP);
175 	}
176 
177 	/* Create VM */
178 	vm = vm_create_default(VCPU_ID, 0, guest_code);
179 	run = vcpu_state(vm, VCPU_ID);
180 
181 	vcpu_regs_get(vm, VCPU_ID, &regs1);
182 
183 	vcpu_alloc_vmx(vm, &vmx_pages_gva);
184 	vcpu_args_set(vm, VCPU_ID, 1, vmx_pages_gva);
185 
186 	for (stage = 1;; stage++) {
187 		_vcpu_run(vm, VCPU_ID);
188 		TEST_ASSERT(run->exit_reason == KVM_EXIT_IO,
189 			    "Stage %d: unexpected exit reason: %u (%s),\n",
190 			    stage, run->exit_reason,
191 			    exit_reason_str(run->exit_reason));
192 
193 		switch (get_ucall(vm, VCPU_ID, &uc)) {
194 		case UCALL_ABORT:
195 			TEST_FAIL("%s at %s:%ld", (const char *)uc.args[0],
196 				  __FILE__, uc.args[1]);
197 			/* NOT REACHED */
198 		case UCALL_SYNC:
199 			break;
200 		case UCALL_DONE:
201 			goto done;
202 		default:
203 			TEST_FAIL("Unknown ucall %lu", uc.cmd);
204 		}
205 
206 		/* UCALL_SYNC is handled here.  */
207 		TEST_ASSERT(!strcmp((const char *)uc.args[0], "hello") &&
208 			    uc.args[1] == stage, "Stage %d: Unexpected register values vmexit, got %lx",
209 			    stage, (ulong)uc.args[1]);
210 		/*
211 		 * If this stage 2 then we should verify the vmx pt expiry
212 		 * is as expected.
213 		 * From L1's perspective verify Preemption timer hasn't
214 		 * expired too early.
215 		 * From L2's perspective verify Preemption timer hasn't
216 		 * expired too late.
217 		 */
218 		if (stage == 2) {
219 
220 			pr_info("Stage %d: L1 PT expiry TSC (%lu) , L1 TSC deadline (%lu)\n",
221 				stage, uc.args[2], uc.args[3]);
222 
223 			pr_info("Stage %d: L2 PT expiry TSC (%lu) , L2 TSC deadline (%lu)\n",
224 				stage, uc.args[4], uc.args[5]);
225 
226 			TEST_ASSERT(uc.args[2] >= uc.args[3],
227 				"Stage %d: L1 PT expiry TSC (%lu) < L1 TSC deadline (%lu)",
228 				stage, uc.args[2], uc.args[3]);
229 
230 			TEST_ASSERT(uc.args[4] < uc.args[5],
231 				"Stage %d: L2 PT expiry TSC (%lu) > L2 TSC deadline (%lu)",
232 				stage, uc.args[4], uc.args[5]);
233 		}
234 
235 		state = vcpu_save_state(vm, VCPU_ID);
236 		memset(&regs1, 0, sizeof(regs1));
237 		vcpu_regs_get(vm, VCPU_ID, &regs1);
238 
239 		kvm_vm_release(vm);
240 
241 		/* Restore state in a new VM.  */
242 		kvm_vm_restart(vm, O_RDWR);
243 		vm_vcpu_add(vm, VCPU_ID);
244 		vcpu_set_cpuid(vm, VCPU_ID, kvm_get_supported_cpuid());
245 		vcpu_load_state(vm, VCPU_ID, state);
246 		run = vcpu_state(vm, VCPU_ID);
247 		kvm_x86_state_cleanup(state);
248 
249 		memset(&regs2, 0, sizeof(regs2));
250 		vcpu_regs_get(vm, VCPU_ID, &regs2);
251 		TEST_ASSERT(!memcmp(&regs1, &regs2, sizeof(regs2)),
252 			    "Unexpected register values after vcpu_load_state; rdi: %lx rsi: %lx",
253 			    (ulong) regs2.rdi, (ulong) regs2.rsi);
254 	}
255 
256 done:
257 	kvm_vm_free(vm);
258 }
259