Linux Audio

Check our new training course

Linux kernel drivers training

Mar 31-Apr 9, 2025, special US time zones
Register
Loading...
v4.10.11
  1/*
  2 * x86 single-step support code, common to 32-bit and 64-bit.
  3 */
  4#include <linux/sched.h>
  5#include <linux/mm.h>
  6#include <linux/ptrace.h>
  7#include <asm/desc.h>
  8#include <asm/mmu_context.h>
  9
 10unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs)
 11{
 12	unsigned long addr, seg;
 13
 14	addr = regs->ip;
 15	seg = regs->cs & 0xffff;
 16	if (v8086_mode(regs)) {
 17		addr = (addr & 0xffff) + (seg << 4);
 18		return addr;
 19	}
 20
 21#ifdef CONFIG_MODIFY_LDT_SYSCALL
 22	/*
 23	 * We'll assume that the code segments in the GDT
 24	 * are all zero-based. That is largely true: the
 25	 * TLS segments are used for data, and the PNPBIOS
 26	 * and APM bios ones we just ignore here.
 27	 */
 28	if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) {
 29		struct desc_struct *desc;
 30		unsigned long base;
 31
 32		seg >>= 3;
 33
 34		mutex_lock(&child->mm->context.lock);
 35		if (unlikely(!child->mm->context.ldt ||
 36			     seg >= child->mm->context.ldt->size))
 37			addr = -1L; /* bogus selector, access would fault */
 38		else {
 39			desc = &child->mm->context.ldt->entries[seg];
 40			base = get_desc_base(desc);
 41
 42			/* 16-bit code segment? */
 43			if (!desc->d)
 44				addr &= 0xffff;
 45			addr += base;
 46		}
 47		mutex_unlock(&child->mm->context.lock);
 48	}
 49#endif
 50
 51	return addr;
 52}
 53
 54static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
 55{
 56	int i, copied;
 57	unsigned char opcode[15];
 58	unsigned long addr = convert_ip_to_linear(child, regs);
 59
 60	copied = access_process_vm(child, addr, opcode, sizeof(opcode),
 61			FOLL_FORCE);
 62	for (i = 0; i < copied; i++) {
 63		switch (opcode[i]) {
 64		/* popf and iret */
 65		case 0x9d: case 0xcf:
 66			return 1;
 67
 68			/* CHECKME: 64 65 */
 69
 70		/* opcode and address size prefixes */
 71		case 0x66: case 0x67:
 72			continue;
 73		/* irrelevant prefixes (segment overrides and repeats) */
 74		case 0x26: case 0x2e:
 75		case 0x36: case 0x3e:
 76		case 0x64: case 0x65:
 77		case 0xf0: case 0xf2: case 0xf3:
 78			continue;
 79
 80#ifdef CONFIG_X86_64
 81		case 0x40 ... 0x4f:
 82			if (!user_64bit_mode(regs))
 83				/* 32-bit mode: register increment */
 84				return 0;
 85			/* 64-bit mode: REX prefix */
 86			continue;
 87#endif
 88
 89			/* CHECKME: f2, f3 */
 90
 91		/*
 92		 * pushf: NOTE! We should probably not let
 93		 * the user see the TF bit being set. But
 94		 * it's more pain than it's worth to avoid
 95		 * it, and a debugger could emulate this
 96		 * all in user space if it _really_ cares.
 97		 */
 98		case 0x9c:
 99		default:
100			return 0;
101		}
102	}
103	return 0;
104}
105
106/*
107 * Enable single-stepping.  Return nonzero if user mode is not using TF itself.
108 */
109static int enable_single_step(struct task_struct *child)
110{
111	struct pt_regs *regs = task_pt_regs(child);
112	unsigned long oflags;
113
114	/*
115	 * If we stepped into a sysenter/syscall insn, it trapped in
116	 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
117	 * If user-mode had set TF itself, then it's still clear from
118	 * do_debug() and we need to set it again to restore the user
119	 * state so we don't wrongly set TIF_FORCED_TF below.
120	 * If enable_single_step() was used last and that is what
121	 * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are
122	 * already set and our bookkeeping is fine.
123	 */
124	if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP)))
125		regs->flags |= X86_EFLAGS_TF;
126
127	/*
128	 * Always set TIF_SINGLESTEP - this guarantees that
129	 * we single-step system calls etc..  This will also
130	 * cause us to set TF when returning to user mode.
131	 */
132	set_tsk_thread_flag(child, TIF_SINGLESTEP);
133
134	oflags = regs->flags;
135
136	/* Set TF on the kernel stack.. */
137	regs->flags |= X86_EFLAGS_TF;
138
139	/*
140	 * ..but if TF is changed by the instruction we will trace,
141	 * don't mark it as being "us" that set it, so that we
142	 * won't clear it by hand later.
143	 *
144	 * Note that if we don't actually execute the popf because
145	 * of a signal arriving right now or suchlike, we will lose
146	 * track of the fact that it really was "us" that set it.
147	 */
148	if (is_setting_trap_flag(child, regs)) {
149		clear_tsk_thread_flag(child, TIF_FORCED_TF);
150		return 0;
151	}
152
153	/*
154	 * If TF was already set, check whether it was us who set it.
155	 * If not, we should never attempt a block step.
156	 */
157	if (oflags & X86_EFLAGS_TF)
158		return test_tsk_thread_flag(child, TIF_FORCED_TF);
159
160	set_tsk_thread_flag(child, TIF_FORCED_TF);
161
162	return 1;
163}
164
165void set_task_blockstep(struct task_struct *task, bool on)
166{
167	unsigned long debugctl;
168
169	/*
170	 * Ensure irq/preemption can't change debugctl in between.
171	 * Note also that both TIF_BLOCKSTEP and debugctl should
172	 * be changed atomically wrt preemption.
173	 *
174	 * NOTE: this means that set/clear TIF_BLOCKSTEP is only safe if
175	 * task is current or it can't be running, otherwise we can race
176	 * with __switch_to_xtra(). We rely on ptrace_freeze_traced() but
177	 * PTRACE_KILL is not safe.
178	 */
179	local_irq_disable();
180	debugctl = get_debugctlmsr();
181	if (on) {
182		debugctl |= DEBUGCTLMSR_BTF;
183		set_tsk_thread_flag(task, TIF_BLOCKSTEP);
184	} else {
185		debugctl &= ~DEBUGCTLMSR_BTF;
186		clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
187	}
188	if (task == current)
189		update_debugctlmsr(debugctl);
190	local_irq_enable();
191}
192
193/*
194 * Enable single or block step.
195 */
196static void enable_step(struct task_struct *child, bool block)
197{
198	/*
199	 * Make sure block stepping (BTF) is not enabled unless it should be.
200	 * Note that we don't try to worry about any is_setting_trap_flag()
201	 * instructions after the first when using block stepping.
202	 * So no one should try to use debugger block stepping in a program
203	 * that uses user-mode single stepping itself.
204	 */
205	if (enable_single_step(child) && block)
206		set_task_blockstep(child, true);
207	else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
208		set_task_blockstep(child, false);
209}
210
211void user_enable_single_step(struct task_struct *child)
212{
213	enable_step(child, 0);
214}
215
216void user_enable_block_step(struct task_struct *child)
217{
218	enable_step(child, 1);
219}
220
221void user_disable_single_step(struct task_struct *child)
222{
223	/*
224	 * Make sure block stepping (BTF) is disabled.
225	 */
226	if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
227		set_task_blockstep(child, false);
228
229	/* Always clear TIF_SINGLESTEP... */
230	clear_tsk_thread_flag(child, TIF_SINGLESTEP);
231
232	/* But touch TF only if it was set by us.. */
233	if (test_and_clear_tsk_thread_flag(child, TIF_FORCED_TF))
234		task_pt_regs(child)->flags &= ~X86_EFLAGS_TF;
235}
v4.6
  1/*
  2 * x86 single-step support code, common to 32-bit and 64-bit.
  3 */
  4#include <linux/sched.h>
  5#include <linux/mm.h>
  6#include <linux/ptrace.h>
  7#include <asm/desc.h>
  8#include <asm/mmu_context.h>
  9
 10unsigned long convert_ip_to_linear(struct task_struct *child, struct pt_regs *regs)
 11{
 12	unsigned long addr, seg;
 13
 14	addr = regs->ip;
 15	seg = regs->cs & 0xffff;
 16	if (v8086_mode(regs)) {
 17		addr = (addr & 0xffff) + (seg << 4);
 18		return addr;
 19	}
 20
 21#ifdef CONFIG_MODIFY_LDT_SYSCALL
 22	/*
 23	 * We'll assume that the code segments in the GDT
 24	 * are all zero-based. That is largely true: the
 25	 * TLS segments are used for data, and the PNPBIOS
 26	 * and APM bios ones we just ignore here.
 27	 */
 28	if ((seg & SEGMENT_TI_MASK) == SEGMENT_LDT) {
 29		struct desc_struct *desc;
 30		unsigned long base;
 31
 32		seg >>= 3;
 33
 34		mutex_lock(&child->mm->context.lock);
 35		if (unlikely(!child->mm->context.ldt ||
 36			     seg >= child->mm->context.ldt->size))
 37			addr = -1L; /* bogus selector, access would fault */
 38		else {
 39			desc = &child->mm->context.ldt->entries[seg];
 40			base = get_desc_base(desc);
 41
 42			/* 16-bit code segment? */
 43			if (!desc->d)
 44				addr &= 0xffff;
 45			addr += base;
 46		}
 47		mutex_unlock(&child->mm->context.lock);
 48	}
 49#endif
 50
 51	return addr;
 52}
 53
 54static int is_setting_trap_flag(struct task_struct *child, struct pt_regs *regs)
 55{
 56	int i, copied;
 57	unsigned char opcode[15];
 58	unsigned long addr = convert_ip_to_linear(child, regs);
 59
 60	copied = access_process_vm(child, addr, opcode, sizeof(opcode), 0);
 
 61	for (i = 0; i < copied; i++) {
 62		switch (opcode[i]) {
 63		/* popf and iret */
 64		case 0x9d: case 0xcf:
 65			return 1;
 66
 67			/* CHECKME: 64 65 */
 68
 69		/* opcode and address size prefixes */
 70		case 0x66: case 0x67:
 71			continue;
 72		/* irrelevant prefixes (segment overrides and repeats) */
 73		case 0x26: case 0x2e:
 74		case 0x36: case 0x3e:
 75		case 0x64: case 0x65:
 76		case 0xf0: case 0xf2: case 0xf3:
 77			continue;
 78
 79#ifdef CONFIG_X86_64
 80		case 0x40 ... 0x4f:
 81			if (!user_64bit_mode(regs))
 82				/* 32-bit mode: register increment */
 83				return 0;
 84			/* 64-bit mode: REX prefix */
 85			continue;
 86#endif
 87
 88			/* CHECKME: f2, f3 */
 89
 90		/*
 91		 * pushf: NOTE! We should probably not let
 92		 * the user see the TF bit being set. But
 93		 * it's more pain than it's worth to avoid
 94		 * it, and a debugger could emulate this
 95		 * all in user space if it _really_ cares.
 96		 */
 97		case 0x9c:
 98		default:
 99			return 0;
100		}
101	}
102	return 0;
103}
104
105/*
106 * Enable single-stepping.  Return nonzero if user mode is not using TF itself.
107 */
108static int enable_single_step(struct task_struct *child)
109{
110	struct pt_regs *regs = task_pt_regs(child);
111	unsigned long oflags;
112
113	/*
114	 * If we stepped into a sysenter/syscall insn, it trapped in
115	 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
116	 * If user-mode had set TF itself, then it's still clear from
117	 * do_debug() and we need to set it again to restore the user
118	 * state so we don't wrongly set TIF_FORCED_TF below.
119	 * If enable_single_step() was used last and that is what
120	 * set TIF_SINGLESTEP, then both TF and TIF_FORCED_TF are
121	 * already set and our bookkeeping is fine.
122	 */
123	if (unlikely(test_tsk_thread_flag(child, TIF_SINGLESTEP)))
124		regs->flags |= X86_EFLAGS_TF;
125
126	/*
127	 * Always set TIF_SINGLESTEP - this guarantees that
128	 * we single-step system calls etc..  This will also
129	 * cause us to set TF when returning to user mode.
130	 */
131	set_tsk_thread_flag(child, TIF_SINGLESTEP);
132
133	oflags = regs->flags;
134
135	/* Set TF on the kernel stack.. */
136	regs->flags |= X86_EFLAGS_TF;
137
138	/*
139	 * ..but if TF is changed by the instruction we will trace,
140	 * don't mark it as being "us" that set it, so that we
141	 * won't clear it by hand later.
142	 *
143	 * Note that if we don't actually execute the popf because
144	 * of a signal arriving right now or suchlike, we will lose
145	 * track of the fact that it really was "us" that set it.
146	 */
147	if (is_setting_trap_flag(child, regs)) {
148		clear_tsk_thread_flag(child, TIF_FORCED_TF);
149		return 0;
150	}
151
152	/*
153	 * If TF was already set, check whether it was us who set it.
154	 * If not, we should never attempt a block step.
155	 */
156	if (oflags & X86_EFLAGS_TF)
157		return test_tsk_thread_flag(child, TIF_FORCED_TF);
158
159	set_tsk_thread_flag(child, TIF_FORCED_TF);
160
161	return 1;
162}
163
164void set_task_blockstep(struct task_struct *task, bool on)
165{
166	unsigned long debugctl;
167
168	/*
169	 * Ensure irq/preemption can't change debugctl in between.
170	 * Note also that both TIF_BLOCKSTEP and debugctl should
171	 * be changed atomically wrt preemption.
172	 *
173	 * NOTE: this means that set/clear TIF_BLOCKSTEP is only safe if
174	 * task is current or it can't be running, otherwise we can race
175	 * with __switch_to_xtra(). We rely on ptrace_freeze_traced() but
176	 * PTRACE_KILL is not safe.
177	 */
178	local_irq_disable();
179	debugctl = get_debugctlmsr();
180	if (on) {
181		debugctl |= DEBUGCTLMSR_BTF;
182		set_tsk_thread_flag(task, TIF_BLOCKSTEP);
183	} else {
184		debugctl &= ~DEBUGCTLMSR_BTF;
185		clear_tsk_thread_flag(task, TIF_BLOCKSTEP);
186	}
187	if (task == current)
188		update_debugctlmsr(debugctl);
189	local_irq_enable();
190}
191
192/*
193 * Enable single or block step.
194 */
195static void enable_step(struct task_struct *child, bool block)
196{
197	/*
198	 * Make sure block stepping (BTF) is not enabled unless it should be.
199	 * Note that we don't try to worry about any is_setting_trap_flag()
200	 * instructions after the first when using block stepping.
201	 * So no one should try to use debugger block stepping in a program
202	 * that uses user-mode single stepping itself.
203	 */
204	if (enable_single_step(child) && block)
205		set_task_blockstep(child, true);
206	else if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
207		set_task_blockstep(child, false);
208}
209
210void user_enable_single_step(struct task_struct *child)
211{
212	enable_step(child, 0);
213}
214
215void user_enable_block_step(struct task_struct *child)
216{
217	enable_step(child, 1);
218}
219
220void user_disable_single_step(struct task_struct *child)
221{
222	/*
223	 * Make sure block stepping (BTF) is disabled.
224	 */
225	if (test_tsk_thread_flag(child, TIF_BLOCKSTEP))
226		set_task_blockstep(child, false);
227
228	/* Always clear TIF_SINGLESTEP... */
229	clear_tsk_thread_flag(child, TIF_SINGLESTEP);
230
231	/* But touch TF only if it was set by us.. */
232	if (test_and_clear_tsk_thread_flag(child, TIF_FORCED_TF))
233		task_pt_regs(child)->flags &= ~X86_EFLAGS_TF;
234}