1/* SPDX-License-Identifier: GPL-2.0-only */
2/*
3 * KVM nVHE hypervisor stack tracing support.
4 *
5 * The unwinder implementation depends on the nVHE mode:
6 *
7 * 1) Non-protected nVHE mode - the host can directly access the
8 * HYP stack pages and unwind the HYP stack in EL1. This saves having
9 * to allocate shared buffers for the host to read the unwinded
10 * stacktrace.
11 *
12 * 2) pKVM (protected nVHE) mode - the host cannot directly access
13 * the HYP memory. The stack is unwinded in EL2 and dumped to a shared
14 * buffer where the host can read and print the stacktrace.
15 *
16 * Copyright (C) 2022 Google LLC
17 */
18
19#include <linux/kvm.h>
20#include <linux/kvm_host.h>
21
22#include <asm/kvm_mmu.h>
23#include <asm/stacktrace/nvhe.h>
24
25static struct stack_info stackinfo_get_overflow(void)
26{
27 struct kvm_nvhe_stacktrace_info *stacktrace_info
28 = this_cpu_ptr_nvhe_sym(kvm_stacktrace_info);
29 unsigned long low = (unsigned long)stacktrace_info->overflow_stack_base;
30 unsigned long high = low + OVERFLOW_STACK_SIZE;
31
32 return (struct stack_info) {
33 .low = low,
34 .high = high,
35 };
36}
37
38static struct stack_info stackinfo_get_overflow_kern_va(void)
39{
40 unsigned long low = (unsigned long)this_cpu_ptr_nvhe_sym(overflow_stack);
41 unsigned long high = low + OVERFLOW_STACK_SIZE;
42
43 return (struct stack_info) {
44 .low = low,
45 .high = high,
46 };
47}
48
49static struct stack_info stackinfo_get_hyp(void)
50{
51 struct kvm_nvhe_stacktrace_info *stacktrace_info
52 = this_cpu_ptr_nvhe_sym(kvm_stacktrace_info);
53 unsigned long low = (unsigned long)stacktrace_info->stack_base;
54 unsigned long high = low + NVHE_STACK_SIZE;
55
56 return (struct stack_info) {
57 .low = low,
58 .high = high,
59 };
60}
61
62static struct stack_info stackinfo_get_hyp_kern_va(void)
63{
64 unsigned long low = (unsigned long)*this_cpu_ptr(&kvm_arm_hyp_stack_base);
65 unsigned long high = low + NVHE_STACK_SIZE;
66
67 return (struct stack_info) {
68 .low = low,
69 .high = high,
70 };
71}
72
73/*
74 * kvm_nvhe_stack_kern_va - Convert KVM nVHE HYP stack addresses to a kernel VAs
75 *
76 * The nVHE hypervisor stack is mapped in the flexible 'private' VA range, to
77 * allow for guard pages below the stack. Consequently, the fixed offset address
78 * translation macros won't work here.
79 *
80 * The kernel VA is calculated as an offset from the kernel VA of the hypervisor
81 * stack base.
82 *
83 * Returns true on success and updates @addr to its corresponding kernel VA;
84 * otherwise returns false.
85 */
86static bool kvm_nvhe_stack_kern_va(unsigned long *addr, unsigned long size)
87{
88 struct stack_info stack_hyp, stack_kern;
89
90 stack_hyp = stackinfo_get_hyp();
91 stack_kern = stackinfo_get_hyp_kern_va();
92 if (stackinfo_on_stack(&stack_hyp, *addr, size))
93 goto found;
94
95 stack_hyp = stackinfo_get_overflow();
96 stack_kern = stackinfo_get_overflow_kern_va();
97 if (stackinfo_on_stack(&stack_hyp, *addr, size))
98 goto found;
99
100 return false;
101
102found:
103 *addr = *addr - stack_hyp.low + stack_kern.low;
104 return true;
105}
106
107/*
108 * Convert a KVN nVHE HYP frame record address to a kernel VA
109 */
110static bool kvm_nvhe_stack_kern_record_va(unsigned long *addr)
111{
112 return kvm_nvhe_stack_kern_va(addr, size: 16);
113}
114
115static int unwind_next(struct unwind_state *state)
116{
117 /*
118 * The FP is in the hypervisor VA space. Convert it to the kernel VA
119 * space so it can be unwound by the regular unwind functions.
120 */
121 if (!kvm_nvhe_stack_kern_record_va(addr: &state->fp))
122 return -EINVAL;
123
124 return unwind_next_frame_record(state);
125}
126
127static void unwind(struct unwind_state *state,
128 stack_trace_consume_fn consume_entry, void *cookie)
129{
130 while (1) {
131 int ret;
132
133 if (!consume_entry(cookie, state->pc))
134 break;
135 ret = unwind_next(state);
136 if (ret < 0)
137 break;
138 }
139}
140
141/*
142 * kvm_nvhe_dump_backtrace_entry - Symbolize and print an nVHE backtrace entry
143 *
144 * @arg : the hypervisor offset, used for address translation
145 * @where : the program counter corresponding to the stack frame
146 */
147static bool kvm_nvhe_dump_backtrace_entry(void *arg, unsigned long where)
148{
149 unsigned long va_mask = GENMASK_ULL(__hyp_va_bits - 1, 0);
150 unsigned long hyp_offset = (unsigned long)arg;
151
152 /* Mask tags and convert to kern addr */
153 where = (where & va_mask) + hyp_offset;
154 kvm_err(" [<%016lx>] %pB\n", where, (void *)(where + kaslr_offset()));
155
156 return true;
157}
158
159static void kvm_nvhe_dump_backtrace_start(void)
160{
161 kvm_err("nVHE call trace:\n");
162}
163
164static void kvm_nvhe_dump_backtrace_end(void)
165{
166 kvm_err("---[ end nVHE call trace ]---\n");
167}
168
169/*
170 * hyp_dump_backtrace - Dump the non-protected nVHE backtrace.
171 *
172 * @hyp_offset: hypervisor offset, used for address translation.
173 *
174 * The host can directly access HYP stack pages in non-protected
175 * mode, so the unwinding is done directly from EL1. This removes
176 * the need for shared buffers between host and hypervisor for
177 * the stacktrace.
178 */
179static void hyp_dump_backtrace(unsigned long hyp_offset)
180{
181 struct kvm_nvhe_stacktrace_info *stacktrace_info;
182 struct stack_info stacks[] = {
183 stackinfo_get_overflow_kern_va(),
184 stackinfo_get_hyp_kern_va(),
185 };
186 struct unwind_state state = {
187 .stacks = stacks,
188 .nr_stacks = ARRAY_SIZE(stacks),
189 };
190
191 stacktrace_info = this_cpu_ptr_nvhe_sym(kvm_stacktrace_info);
192
193 kvm_nvhe_unwind_init(&state, stacktrace_info->fp, stacktrace_info->pc);
194
195 kvm_nvhe_dump_backtrace_start();
196 unwind(state: &state, consume_entry: kvm_nvhe_dump_backtrace_entry, cookie: (void *)hyp_offset);
197 kvm_nvhe_dump_backtrace_end();
198}
199
200#ifdef CONFIG_PROTECTED_NVHE_STACKTRACE
201DECLARE_KVM_NVHE_PER_CPU(unsigned long [NVHE_STACKTRACE_SIZE/sizeof(long)],
202 pkvm_stacktrace);
203
204/*
205 * pkvm_dump_backtrace - Dump the protected nVHE HYP backtrace.
206 *
207 * @hyp_offset: hypervisor offset, used for address translation.
208 *
209 * Dumping of the pKVM HYP backtrace is done by reading the
210 * stack addresses from the shared stacktrace buffer, since the
211 * host cannot directly access hypervisor memory in protected
212 * mode.
213 */
214static void pkvm_dump_backtrace(unsigned long hyp_offset)
215{
216 unsigned long *stacktrace
217 = (unsigned long *) this_cpu_ptr_nvhe_sym(pkvm_stacktrace);
218 int i;
219
220 kvm_nvhe_dump_backtrace_start();
221 /* The saved stacktrace is terminated by a null entry */
222 for (i = 0;
223 i < ARRAY_SIZE(kvm_nvhe_sym(pkvm_stacktrace)) && stacktrace[i];
224 i++)
225 kvm_nvhe_dump_backtrace_entry((void *)hyp_offset, stacktrace[i]);
226 kvm_nvhe_dump_backtrace_end();
227}
228#else /* !CONFIG_PROTECTED_NVHE_STACKTRACE */
229static void pkvm_dump_backtrace(unsigned long hyp_offset)
230{
231 kvm_err("Cannot dump pKVM nVHE stacktrace: !CONFIG_PROTECTED_NVHE_STACKTRACE\n");
232}
233#endif /* CONFIG_PROTECTED_NVHE_STACKTRACE */
234
235/*
236 * kvm_nvhe_dump_backtrace - Dump KVM nVHE hypervisor backtrace.
237 *
238 * @hyp_offset: hypervisor offset, used for address translation.
239 */
240void kvm_nvhe_dump_backtrace(unsigned long hyp_offset)
241{
242 if (is_protected_kvm_enabled())
243 pkvm_dump_backtrace(hyp_offset);
244 else
245 hyp_dump_backtrace(hyp_offset);
246}
247

source code of linux/arch/arm64/kvm/stacktrace.c