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