Linux 6.12-rc1
[linux-2.6-block.git] / arch / x86 / kernel / step.c
CommitLineData
b2441318 1// SPDX-License-Identifier: GPL-2.0
fa1e03ea
RM
2/*
3 * x86 single-step support code, common to 32-bit and 64-bit.
4 */
5#include <linux/sched.h>
68db0cf1 6#include <linux/sched/task_stack.h>
fa1e03ea
RM
7#include <linux/mm.h>
8#include <linux/ptrace.h>
154fcf3a 9
254e0a6b 10#include <asm/desc.h>
154fcf3a 11#include <asm/debugreg.h>
37868fe1 12#include <asm/mmu_context.h>
fa1e03ea 13
37cd9cf3 14unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs)
fa1e03ea
RM
15{
16 unsigned long addr, seg;
17
65ea5b03 18 addr = regs->ip;
99504819 19 seg = regs->cs;
65ea5b03 20 if (v8086_mode(regs)) {
7122ec81
RM
21 addr = (addr & 0xffff) + (seg << 4);
22 return addr;
23 }
fa1e03ea 24
a5b9e5a2 25#ifdef CONFIG_MODIFY_LDT_SYSCALL
fa1e03ea
RM
26 /*
27 * We'll assume that the code segments in the GDT
28 * are all zero-based. That is largely true: the
29 * TLS segments are used for data, and the PNPBIOS
30 * and APM bios ones we just ignore here.
31 */
3f80c1ad 32 if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) {
254e0a6b 33 struct desc_struct *desc;
fa1e03ea
RM
34 unsigned long base;
35
136d9d83 36 seg >>= 3;
fa1e03ea
RM
37
38 mutex_lock(&child->mm->context.lock);
37868fe1 39 if (unlikely(!child->mm->context.ldt ||
bbf79d21 40 seg >= child->mm->context.ldt->nr_entries))
fa1e03ea
RM
41 addr = -1L; /* bogus selector, access would fault */
42 else {
37868fe1 43 desc = &child->mm->context.ldt->entries[seg];
254e0a6b 44 base = get_desc_base(desc);
fa1e03ea
RM
45
46 /* 16-bit code segment? */
254e0a6b 47 if (!desc->d)
fa1e03ea
RM
48 addr &= 0xffff;
49 addr += base;
50 }
51 mutex_unlock(&child->mm->context.lock);
52 }
a5b9e5a2 53#endif
fa1e03ea
RM
54
55 return addr;
56}
57
58static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
59{
60 int i, copied;
61 unsigned char opcode[15];
37cd9cf3 62 unsigned long addr = convert_ip_to_linear(child, regs);
fa1e03ea 63
f307ab6d
LS
64 copied = access_process_vm(child, addr, opcode, sizeof(opcode),
65 FOLL_FORCE);
fa1e03ea
RM
66 for (i = 0; i < copied; i++) {
67 switch (opcode[i]) {
68 /* popf and iret */
69 case 0x9d: case 0xcf:
70 return 1;
71
72 /* CHECKME: 64 65 */
73
74 /* opcode and address size prefixes */
75 case 0x66: case 0x67:
76 continue;
77 /* irrelevant prefixes (segment overrides and repeats) */
78 case 0x26: case 0x2e:
79 case 0x36: case 0x3e:
80 case 0x64: case 0x65:
5f76cb1f 81 case 0xf0: case 0xf2: case 0xf3:
fa1e03ea
RM
82 continue;
83
7122ec81 84#ifdef CONFIG_X86_64
fa1e03ea 85 case 0x40 ... 0x4f:
318f5a2a 86 if (!user_64bit_mode(regs))
fa1e03ea
RM
87 /* 32-bit mode: register increment */
88 return 0;
89 /* 64-bit mode: REX prefix */
90 continue;
7122ec81 91#endif
fa1e03ea
RM
92
93 /* CHECKME: f2, f3 */
94
95 /*
96 * pushf: NOTE! We should probably not let
97 * the user see the TF bit being set. But
98 * it's more pain than it's worth to avoid
99 * it, and a debugger could emulate this
100 * all in user space if it _really_ cares.
101 */
102 case 0x9c:
103 default:
104 return 0;
105 }
106 }
107 return 0;
108}
109
10faa81e
RM
110/*
111 * Enable single-stepping. Return nonzero if user mode is not using TF itself.
112 */
113static int enable_single_step(struct task_struct *child)
fa1e03ea
RM
114{
115 struct pt_regs *regs = task_pt_regs(child);
6718d0d6 116 unsigned long oflags;
fa1e03ea 117
380fdd75
RM
118 /*
119 * If we stepped into a sysenter/syscall insn, it trapped in
120 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
121 * If user-mode had set TF itself, then it's still clear from
122 * do_debug() and we need to set it again to restore the user
123 * state so we don't wrongly set TIF_FORCED_TF below.
124 * If enable_single_step() was used last and that is what
125 * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are
126 * already set and our bookkeeping is fine.
127 */
128 if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP)))
129 regs->flags |= X86_EFLAGS_TF;
130
fa1e03ea 131 /*
6342adca 132 * Always set TIF_SINGLESTEP. This will also
fa1e03ea
RM
133 * cause us to set TF when returning to user mode.
134 */
135 set_tsk_thread_flag(child, TIF_SINGLESTEP);
136
6342adca
GKB
137 /*
138 * Ensure that a trap is triggered once stepping out of a system
139 * call prior to executing any user instruction.
140 */
141 set_task_syscall_work(child, SYSCALL_EXIT_TRAP);
142
6718d0d6 143 oflags = regs->flags;
fa1e03ea
RM
144
145 /* Set TF on the kernel stack.. */
65ea5b03 146 regs->flags |= X86_EFLAGS_TF;
fa1e03ea
RM
147
148 /*
149 * ..but if TF is changed by the instruction we will trace,
150 * don't mark it as being "us" that set it, so that we
151 * won't clear it by hand later.
6718d0d6
RM
152 *
153 * Note that if we don't actually execute the popf because
154 * of a signal arriving right now or suchlike, we will lose
155 * track of the fact that it really was "us" that set it.
fa1e03ea 156 */
6718d0d6
RM
157 if (is_setting_trap_flag(child, regs)) {
158 clear_tsk_thread_flag(child, TIF_FORCED_TF);
10faa81e 159 return 0;
6718d0d6
RM
160 }
161
162 /*
163 * If TF was already set, check whether it was us who set it.
164 * If not, we should never attempt a block step.
165 */
166 if (oflags & X86_EFLAGS_TF)
167 return test_tsk_thread_flag(child, TIF_FORCED_TF);
fa1e03ea 168
e1f28773 169 set_tsk_thread_flag(child, TIF_FORCED_TF);
10faa81e
RM
170
171 return 1;
172}
173
9bd1190a 174void set_task_blockstep(struct task_struct *task, bool on)
848e8f5f
ON
175{
176 unsigned long debugctl;
177
95cf00fa
ON
178 /*
179 * Ensure irq/preemption can't change debugctl in between.
180 * Note also that both TIF_BLOCKSTEP and debugctl should
181 * be changed atomically wrt preemption.
9899d11f
ON
182 *
183 * NOTE: this means that set/clear TIF_BLOCKSTEP is only safe if
184 * task is current or it can't be running, otherwise we can race
6a2d90ba 185 * with __switch_to_xtra(). We rely on ptrace_freeze_traced().
95cf00fa
ON
186 */
187 local_irq_disable();
848e8f5f
ON
188 debugctl = get_debugctlmsr();
189 if (on) {
190 debugctl |= DEBUGCTLMSR_BTF;
191 set_tsk_thread_flag(task, TIF_BLOCKSTEP);
192 } else {
193 debugctl &= ~DEBUGCTLMSR_BTF;
194 clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
195 }
95cf00fa
ON
196 if (task == current)
197 update_debugctlmsr(debugctl);
198 local_irq_enable();
848e8f5f
ON
199}
200
10faa81e
RM
201/*
202 * Enable single or block step.
203 */
204static void enable_step(struct task_struct *child, bool block)
205{
206 /*
207 * Make sure block stepping (BTF) is not enabled unless it should be.
208 * Note that we don't try to worry about any is_setting_trap_flag()
209 * instructions after the first when using block stepping.
0d2eb44f 210 * So no one should try to use debugger block stepping in a program
10faa81e
RM
211 * that uses user-mode single stepping itself.
212 */
848e8f5f
ON
213 if (enable_single_step(child) && block)
214 set_task_blockstep(child, true);
215 else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
216 set_task_blockstep(child, false);
10faa81e
RM
217}
218
219void user_enable_single_step(struct task_struct *child)
220{
221 enable_step(child, 0);
222}
223
224void user_enable_block_step(struct task_struct *child)
225{
226 enable_step(child, 1);
fa1e03ea
RM
227}
228
229void user_disable_single_step(struct task_struct *child)
230{
10faa81e
RM
231 /*
232 * Make sure block stepping (BTF) is disabled.
233 */
848e8f5f
ON
234 if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
235 set_task_blockstep(child, false);
10faa81e 236
fa1e03ea
RM
237 /* Always clear TIF_SINGLESTEP... */
238 clear_tsk_thread_flag(child, TIF_SINGLESTEP);
6342adca 239 clear_task_syscall_work(child, SYSCALL_EXIT_TRAP);
fa1e03ea
RM
240
241 /* But touch TF only if it was set by us.. */
e1f28773 242 if (test_and_clear_tsk_thread_flag(child, TIF_FORCED_TF))
65ea5b03 243 task_pt_regs(child)->flags &= ~X86_EFLAGS_TF;
fa1e03ea 244}