Skip to contentSkip to conversation

x86/kprobes: Fixes for CONFIG_RETHUNK

PATCH Superseded Masami Hiramatsu (Google) 4 participants 27 messages Sep 7, 2022 – Sep 15, 2022

Subsystems (3)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

cover x86/kprobes: Fixes for CONFIG_RETHUNK

Hi,

Here is a couple of patches to fix kprobes and optprobe to work on the kernel with CONFIG_RETHUNK and CONFIG_SLS.

With these configs, the kernel functions may includes padding INT3 in the function code block (body) in addition to the gaps between functions.

Since kprobes on x86 has to ensure the probe address is a function bondary, it decodes the instructions in the function until the address. If it finds an INT3 which is not embedded by kprobe, it stops decoding because usually the INT3 is used for debugging as a software breakpoint and such INT3 will replace the first byte of an original instruction. Without recovering it, kprobes can not continue to decode it. Thus the kprobes returns -EILSEQ as below.

 # echo "p:probe/vfs_truncate_L19 vfs_truncate+98" >> kprobe_events 
 sh: write error: Invalid or incomplete multibyte or wide character

Actually, those INT3s are just for padding and can be ignored.

To avoid this issue, if kprobe finds an INT3, it gets the address of next non-INT3 byte, and search a branch which jumps to the address. If there is the branch, these INT3 will be for padding, so it can be skipped. [1/2]

Since the optprobe has similar issue, it also skips the padding INT3 in the same way. [2/2]

With thses fixes, kprobe and optprobe can probe the kernel again with CONFIG_RETHUNK=y.

 # echo "p:probe/vfs_truncate_L19 vfs_truncate+98" >> kprobe_events 
 # echo 1 > events/probe/vfs_truncate_L19/enable 
 # cat /sys/kernel/debug/kprobes/list 
 ffffffff81307b52  k  vfs_truncate+0x62    [OPTIMIZED]

Thank you,

---

Masami Hiramatsu (Google) (2):
x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK x86/kprobes: Fix optprobe optimization check with CONFIG_RETHUNK

 arch/x86/kernel/kprobes/common.h |   67 +++++++++++++++++++++++++++
 arch/x86/kernel/kprobes/core.c   |   57 +++++++++++++----------
 arch/x86/kernel/kprobes/opt.c    |   93 ++++++++++++++------------------------
 3 files changed, 133 insertions(+), 84 deletions(-)

--
Masami Hiramatsu (Google) <[email protected]>

#1 lore raw

PATCH 1/2 x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

From: Masami Hiramatsu (Google) <[email protected]>

Since the CONFIG_RETHUNK and CONFIG_SLS will use INT3 for padding after RET instruction, kprobes always failes to check the probed instruction boundary by decoding the function body if the probed address is after such paddings (Note that some conditional code blocks will be placed after RET instruction, if compiler decides it is not on the hot path.) This is because kprobes expects someone (e.g. kgdb) puts the INT3 as a software breakpoint and it will replace the original instruction. But There are INT3 just for padding in the function, it doesn't need to recover the original instruction.

To avoid this issue, if kprobe finds an INT3, it gets the address of next non-INT3 byte, and search a branch which jumps to the address. If there is the branch, these INT3 will be for padding, so it can be skipped.

Fixes 15e67227c49a ("x86: Undo return-thunk damage")
---
 arch/x86/kernel/kprobes/common.h |   67 ++++++++++++++++++++++++++++++++++++++
 arch/x86/kernel/kprobes/core.c   |   57 ++++++++++++++++++--------------
 arch/x86/kernel/kprobes/opt.c    |   23 +------------
 3 files changed, 100 insertions(+), 47 deletions(-)
arch/x86/kernel/kprobes/common.h +67 -0
index c993521d4933..2adb36eaf366 100644
@@ -92,6 +92,73 @@ extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn);
 extern void synthesize_reljump(void *dest, void *from, void *to);
 extern void synthesize_relcall(void *dest, void *from, void *to);
 
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */
+	case 0xe3:	/* jcxz */
+	case 0xe9:	/* near relative jump */
+	case 0xeb:	/* short relative jump */
+		break;
+	case 0x0f:
+		if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
+			break;
+		return 0;
+	default:
+		if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
+			break;
+		return 0;
+	}
+	return (unsigned long)insn->next_byte + insn->immediate.value;
+}
+
+static inline void __decode_insn(struct insn *insn, kprobe_opcode_t *buf,
+				 unsigned long addr)
+{
+	unsigned long recovered_insn;
+
+	/*
+	 * Check if the instruction has been modified by another
+	 * kprobe, in which case we replace the breakpoint by the
+	 * original instruction in our buffer.
+	 * Also, jump optimization will change the breakpoint to
+	 * relative-jump. Since the relative-jump itself is
+	 * normally used, we just go through if there is no kprobe.
+	 */
+	recovered_insn = recover_probed_instruction(buf, addr);
+	if (!recovered_insn ||
+	    insn_decode_kernel(insn, (void *)recovered_insn) < 0) {
+		insn->kaddr = NULL;
+	} else {
+		/* Recover address */
+		insn->kaddr = (void *)addr;
+		insn->next_byte = (void *)(addr + insn->length);
+	}
+}
+
+/* Iterate instructions in [saddr, eaddr), insn->next_byte is loop cursor. */
+#define for_each_insn(insn, saddr, eaddr, buf)				\
+	for (__decode_insn(insn, buf, saddr);				\
+	     (insn)->kaddr && (unsigned long)(insn)->next_byte < eaddr;	\
+	     __decode_insn(insn, buf, (unsigned long)(insn)->next_byte))
+
+/* Return next non-INT3 address, or 0 if failed to access */
+static inline unsigned long skip_padding_int3(unsigned long addr)
+{
+	unsigned char ops;
+
+	while (get_kernel_nofault(ops, (void *)addr) == 0) {
+		if (ops != INT3_INSN_OPCODE)
+			return addr;
+		addr++;
+	}
+
+	return 0;
+}
+
 #ifdef	CONFIG_OPTPROBES
 extern int setup_detour_execution(struct kprobe *p, struct pt_regs *regs, int reenter);
 extern unsigned long __recover_optprobed_insn(kprobe_opcode_t *buf, unsigned long addr);
arch/x86/kernel/kprobes/core.c +31 -26

75 lines not on this page · raw message

arch/x86/kernel/kprobes/opt.c +2 -21
index e6b8c5362b94..2e41850cab06 100644
@@ -235,28 +235,9 @@ static int __insn_is_indirect_jump(struct insn *insn)
 /* Check whether insn jumps into specified address range */
 static int insn_jump_into_range(struct insn *insn, unsigned long start, int len)
 {
-	unsigned long target = 0;
-
-	switch (insn->opcode.bytes[0]) {
-	case 0xe0:	/* loopne */
-	case 0xe1:	/* loope */
-	case 0xe2:	/* loop */
-	case 0xe3:	/* jcxz */
-	case 0xe9:	/* near relative jump */
-	case 0xeb:	/* short relative jump */
-		break;
-	case 0x0f:
-		if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */
-			break;
-		return 0;
-	default:
-		if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
-			break;
-		return 0;
-	}
-	target = (unsigned long)insn->next_byte + insn->immediate.value;
+	unsigned long target = insn_get_branch_addr(insn);
 
-	return (start <= target && target <= start + len);
+	return target ? (start <= target && target <= start + len) : 0;
 }
 
 static int insn_is_indirect_jump(struct insn *insn)
maintainer
#2 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 101

Quoted 7 lines · Masami Hiramatsu (Google)
arch/x86/kernel/kprobes/common.h +7 -0
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */

Oh cute, objtool doesn't know about those, let me go add them.

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 103

Quoted 2 lines
arch/x86/kernel/kprobes/common.h +2 -0
+	case 0xe3:	/* jcxz */
+	case 0xe9:	/* near relative jump */

/* JMP.d32 */

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 104

Quoted 1 line
arch/x86/kernel/kprobes/common.h +1 -0
+	case 0xeb:	/* short relative jump */

/* JMP.d8 */

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 107

Quoted 3 lines
arch/x86/kernel/kprobes/common.h +3 -0
+		break;
+	case 0x0f:
+		if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */

/* Jcc.d32 */

Are the GNU AS names for these things.

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 113

Quoted 7 lines
arch/x86/kernel/kprobes/common.h +6 -0
+			break;
+		return 0;
 
+	default:
+		if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
+			break;
+		return 0;

You could write that as:

	case 0x70 ... 0x7f: /* Jcc.d8 */
		break;

	default:
		return 0;
Quoted 3 lines
arch/x86/kernel/kprobes/common.h +3 -0
+	}
+	return (unsigned long)insn->next_byte + insn->immediate.value;
+}
maintainer
#3 lore raw

PATCH 1/1 objtool,x86: Teach decode about LOOP* instructions

Quoted 11 lines · Peter Zijlstra

Review

Quoted 7 lines · Masami Hiramatsu (Google)
(quoted diff) +7 -0
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */

Oh cute, objtool doesn't know about those, let me go add them.

---
Subject: objtool,x86: Teach decode about LOOP* instructions

With kprobes also needing to follow control flow; it was found that objtool is missing the branches from the LOOP* instructions.

---
 tools/objtool/arch/x86/decode.c | 6 ++++++
 1 file changed, 6 insertions(+)
tools/objtool/arch/x86/decode.c +6 -0
index c260006106be..1c253b4b7ce0 100644
@@ -635,6 +635,12 @@ int arch_decode_instruction(struct objtool_file *file, const struct section *sec
 		*type = INSN_CONTEXT_SWITCH;
 		break;
 
+	case 0xe0: /* loopne */
+	case 0xe1: /* loope */
+	case 0xe2: /* loop */
+		*type = INSN_JUMP_CONDITIONAL;
+		break;
+
 	case 0xe8:
 		*type = INSN_CALL;
 		/*
#4 lore raw

RE: [PATCH] objtool,x86: Teach decode about LOOP* instructions

From: Peter Zijlstra

Quoted 14 lines

Sent: 07 September 2022 10:01

Quoted 11 lines · Peter Zijlstra

Review

Quoted 7 lines · Masami Hiramatsu (Google)
(quoted diff) +7 -0
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */

Oh cute, objtool doesn't know about those, let me go add them.

Do they ever appear in the kernel?
They are so slow on Intel cpu that finding one ought to deemed a bug!

Have you got jcxz (0xe3) in there?
They are fast on both Intel and AMD cpus - so are usable.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

maintainer
#5 lore raw

Re: [PATCH] objtool,x86: Teach decode about LOOP* instructions

Quoted 16 lines · David Laight

From: Peter Zijlstra

Quoted 14 lines

Sent: 07 September 2022 10:01

Quoted 11 lines · Peter Zijlstra

Review

Quoted 7 lines · Masami Hiramatsu (Google)
(quoted diff) +7 -0
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */

Oh cute, objtool doesn't know about those, let me go add them.

Quoted 1 line

Do they ever appear in the kernel?

No; that is, not on any of the random vmlinux.o images I checked this morning.

Still, best to properly decode them anyway.

#6 lore raw

RE: [PATCH] objtool,x86: Teach decode about LOOP* instructions

From: Peter Zijlstra

Quoted 19 lines

Sent: 07 September 2022 10:40

Quoted 16 lines · David Laight

From: Peter Zijlstra

Quoted 14 lines

Sent: 07 September 2022 10:01

Quoted 11 lines · Peter Zijlstra

Review

Quoted 7 lines · Masami Hiramatsu (Google)
(quoted diff) +7 -0
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */

Oh cute, objtool doesn't know about those, let me go add them.

Quoted 6 lines
Quoted 1 line

Do they ever appear in the kernel?

No; that is, not on any of the random vmlinux.o images I checked this morning.

Still, best to properly decode them anyway.

It is annoying that cpu with adox/adcx have slow loop. You really want to be able to do:

	1:	adox ...
		adcx ...
		loop	1b

That would never run with one iteration/clock.
But unrolling once would probably be enough.

What you can do (and gives the fastest IPcsum loop) is:

	1:	jcxz	2f
		....
		lea	%rcx,...
		jmp	1b
	2:

The extra instructions mean that needs unrolling 4 times. I've got over 12 bytes/clock that way.

David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK Registration No: 1397386 (Wales)

merge bot Applied
#7 lore raw

[tip: objtool/core] objtool,x86: Teach decode about LOOP* instructions

In reply to Peter Zijlstra
The following commit has been merged into the objtool/core branch of tip:

The following commit has been merged into the objtool/core branch of tip:

Commit-ID:     7a7621dfa417aa3715d2a3bd1bdd6cf5018274d0
Gitweb:        https://git.kernel.org/tip/7a7621dfa417aa3715d2a3bd1bdd6cf5018274d0
Author:        Peter Zijlstra <[email protected]>
AuthorDate:    Wed, 07 Sep 2022 11:01:20 +02:00
Committer:     Peter Zijlstra <[email protected]>
CommitterDate: Thu, 15 Sep 2022 16:13:55 +02:00

objtool,x86: Teach decode about LOOP* instructions

When 'discussing' control flow Masami mentioned the LOOP* instructions and I realized objtool doesn't decode them properly.

As it turns out, these instructions are somewhat inefficient and as such unlikely to be emitted by the compiler (a few vmlinux.o checks can't find a single one) so this isn't critical, but still, best to decode them properly.

---
 tools/objtool/arch/x86/decode.c | 6 ++++++
 1 file changed, 6 insertions(+)
tools/objtool/arch/x86/decode.c +6 -0
index c260006..1c253b4 100644
@@ -635,6 +635,12 @@ int arch_decode_instruction(struct objtool_file *file, const struct section *sec
 		*type = INSN_CONTEXT_SWITCH;
 		break;
 
+	case 0xe0: /* loopne */
+	case 0xe1: /* loope */
+	case 0xe2: /* loop */
+		*type = INSN_JUMP_CONDITIONAL;
+		break;
+
 	case 0xe8:
 		*type = INSN_CALL;
 		/*
#8 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

In reply to Peter Zijlstra
Quoted 21 lines · On Wed, 7 Sep 2022 09:06:44 +0200 Peter Zijlstra <[email protected]> wrote:

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 101

Quoted 7 lines · Masami Hiramatsu (Google)
arch/x86/kernel/kprobes/common.h +7 -0
+/* Return the jump target address or 0 */
+static inline unsigned long insn_get_branch_addr(struct insn *insn)
+{
+	switch (insn->opcode.bytes[0]) {
+	case 0xe0:	/* loopne */
+	case 0xe1:	/* loope */
+	case 0xe2:	/* loop */

Oh cute, objtool doesn't know about those, let me go add them.

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 103

Quoted 2 lines
arch/x86/kernel/kprobes/common.h +2 -0
+	case 0xe3:	/* jcxz */
+	case 0xe9:	/* near relative jump */

/* JMP.d32 */

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 104

Quoted 1 line
arch/x86/kernel/kprobes/common.h +1 -0
+	case 0xeb:	/* short relative jump */

/* JMP.d8 */

Quoted 7 lines

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 107

Quoted 3 lines
arch/x86/kernel/kprobes/common.h +3 -0
+		break;
+	case 0x0f:
+		if ((insn->opcode.bytes[1] & 0xf0) == 0x80) /* jcc near */

/* Jcc.d32 */

Are the GNU AS names for these things.

OK, it should be updated. Where can I refer the names (especially '.dX' suffixes)?

Quoted 11 lines

arch/x86/kernel/kprobes/common.h extern int __copy_instruction(u8 *dest, u8 *src, u8 *real, struct insn *insn); · line 113

Quoted 8 lines
arch/x86/kernel/kprobes/common.h +6 -0
+			break;
+		return 0;
 
+	default:
+		if ((insn->opcode.bytes[0] & 0xf0) == 0x70) /* jcc short */
+			break;
+		return 0;

You could write that as:

Quoted 5 lines
	case 0x70 ... 0x7f: /* Jcc.d8 */
		break;

	default:
		return 0;

Thanks! I'll update it.

Quoted 6 lines
Quoted 4 lines
arch/x86/kernel/kprobes/common.h +3 -0
+	}
+	return (unsigned long)insn->next_byte + insn->immediate.value;
+}
Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
#10 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 6 lines · On Wed, 7 Sep 2022 11:38:36 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 1 line · Masami Hiramatsu

OK, it should be updated. Where can I refer the names (especially '.dX' suffixes)?

https://sourceware.org/binutils/docs-2.23.1/as/i386_002dMnemonics.html

.d8' or .d32' suffix prefers 8bit or 32bit displacement in encoding.

This is good to know. OK, I'll use this.

Thanks!

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
maintainer
#11 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

In reply to Masami Hiramatsu (Google)
Quoted 8 lines · Masami Hiramatsu (Google)

static int can_probe(unsigned long paddr)
{

arch/x86/kernel/kprobes/core.c +2 -0
 	kprobe_opcode_t buf[MAX_INSN_SIZE];
+	unsigned long addr, offset = 0;
+	struct insn insn;
 
 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
 		return 0;

arch/x86/kernel/kprobes/core.c unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long add · line 273

Quoted 10 lines
arch/x86/kernel/kprobes/core.c +7 -0
 
+	/* The first address must be instruction boundary. */
+	if (!offset)
+		return 1;
 
+	/* Decode instructions */
+	for_each_insn(&insn, paddr - offset, paddr, buf) {
 		/*
+		 * CONFIG_RETHUNK or CONFIG_SLS or another debug feature
+		 * may install INT3.

Note: they are not debug features.

Quoted 18 lines
arch/x86/kernel/kprobes/core.c +16 -0
 		 */
+		if (insn.opcode.bytes[0] == INT3_INSN_OPCODE) {
+			/* Find the next non-INT3 instruction address */
+			addr = skip_padding_int3((unsigned long)insn.kaddr);
+			if (!addr)
+				return 0;
+			/*
+			 * This can be a padding INT3 for CONFIG_RETHUNK or
+			 * CONFIG_SLS. If a branch jumps to the address next
+			 * to the INT3 sequence, this is just for padding,
+			 * then we can continue decoding.
+			 */
+			for_each_insn(&insn, paddr - offset, addr, buf) {
+				if (insn_get_branch_addr(&insn) == addr)
+					goto found;
+			}
 
+			/* This INT3 can not be decoded safely. */

arch/x86/kernel/kprobes/core.c unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long add · line 301

Quoted 10 lines
arch/x86/kernel/kprobes/core.c +6 -0
 			return 0;
+found:
+			/* Set loop cursor */
+			insn.next_byte = (void *)addr;
+			continue;
+		}
 	}
 
+	return ((unsigned long)insn.next_byte == paddr);
 }

If I understand correctly, it'll fail on something like this:

foo: insn
insn
insn
jmp 2f
int3

1: insn
insn
2: insn
jcc 1b

ret
int3

Which isn't weird code by any means. And soon to be generated by compilers.

Maybe something like:

struct queue {
	int head, tail;
	unsigned long val[16]; /* insufficient; probably should allocate something */
};

void push(struct queue *q, unsigned long val)

{
	/* break loops, been here already */
	for (int i = 0; i < q->head; i++) {
		if (q->val[i] == val)
			return;
	}

	q->val[q->head++] = val;

	WARN_ON(q->head > ARRAY_SIZE(q->val)
}

unsigned long pop(struct queue *q)

{
	if (q->tail == q->head)
		return 0;

	return q->val[q->tail++];
}

bool dead_end_insn(struct instruction *insn)

{
	switch (insn->opcode.bytes[0]) {
	case INT3_INSN_OPCODE:
	case JMP8_INSN_OPCODE:
	case JMP32_INSN_OPCODE:
		return true; /* no arch execution after these */

	case 0xff:
		/* jmp *%reg; jmpf */
		if (modrm_reg == 4 || modrm_reg == 5)
			return true;
		break;

	default:
		break;
	}

	return false;
}
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;

			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}

It's a bit of a pain, but I think it should cover things.

maintainer
#12 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 28 lines · Peter Zijlstra

Maybe something like:

struct queue {
	int head, tail;
	unsigned long val[16]; /* insufficient; probably should allocate something */
};

void push(struct queue *q, unsigned long val)

{
	/* break loops, been here already */
	for (int i = 0; i < q->head; i++) {
		if (q->val[i] == val)
			return;
	}

	q->val[q->head++] = val;

	WARN_ON(q->head > ARRAY_SIZE(q->val)
}

unsigned long pop(struct queue *q)

{
	if (q->tail == q->head)
		return 0;

	return q->val[q->tail++];
}
Quoted 6 lines

bool dead_end_insn(struct instruction *insn)

{
	switch (insn->opcode.bytes[0]) {
	case INT3_INSN_OPCODE:
	case JMP8_INSN_OPCODE:
	case JMP32_INSN_OPCODE:

case RET_INSN_OPCODE:

Quoted 40 lines
		return true; /* no arch execution after these */

	case 0xff:
		/* jmp *%reg; jmpf */
		if (modrm_reg == 4 || modrm_reg == 5)
			return true;
		break;

	default:
		break;
	}

	return false;
}
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;

			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}

It's a bit of a pain, but I think it should cover things.

#13 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

In reply to Peter Zijlstra
Quoted 15 lines · On Wed, 7 Sep 2022 10:02:40 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 13 lines · Masami Hiramatsu (Google)

static int can_probe(unsigned long paddr)
{

arch/x86/kernel/kprobes/core.c +5 -0
 	kprobe_opcode_t buf[MAX_INSN_SIZE];
+	unsigned long addr, offset = 0;
+	struct insn insn;
 
 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
 		return 0;
 
+	/* The first address must be instruction boundary. */
+	if (!offset)
+		return 1;
Quoted 7 lines

arch/x86/kernel/kprobes/core.c unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long add · line 273

Quoted 5 lines
arch/x86/kernel/kprobes/core.c +4 -0
+	/* Decode instructions */
+	for_each_insn(&insn, paddr - offset, paddr, buf) {
 		/*
+		 * CONFIG_RETHUNK or CONFIG_SLS or another debug feature
+		 * may install INT3.

Note: they are not debug features.

Yes, sorry for confusion. CONFIG_RETHUNK/CONFIG_SLS are security feature, and something like kgdb is debug feature, what I meant here.

Quoted 43 lines
Quoted 19 lines
arch/x86/kernel/kprobes/core.c +16 -0
 		 */
+		if (insn.opcode.bytes[0] == INT3_INSN_OPCODE) {
+			/* Find the next non-INT3 instruction address */
+			addr = skip_padding_int3((unsigned long)insn.kaddr);
+			if (!addr)
+				return 0;
+			/*
+			 * This can be a padding INT3 for CONFIG_RETHUNK or
+			 * CONFIG_SLS. If a branch jumps to the address next
+			 * to the INT3 sequence, this is just for padding,
+			 * then we can continue decoding.
+			 */
+			for_each_insn(&insn, paddr - offset, addr, buf) {
+				if (insn_get_branch_addr(&insn) == addr)
+					goto found;
+			}
 
+			/* This INT3 can not be decoded safely. */

arch/x86/kernel/kprobes/core.c unsigned long recover_probed_instruction(kprobe_opcode_t *buf, unsigned long add · line 301

Quoted 10 lines
arch/x86/kernel/kprobes/core.c +6 -0
 			return 0;
+found:
+			/* Set loop cursor */
+			insn.next_byte = (void *)addr;
+			continue;
+		}
 	}
 
+	return ((unsigned long)insn.next_byte == paddr);
 }

If I understand correctly, it'll fail on something like this:

foo: insn
insn
insn
jmp 2f
int3

1: insn
insn
2: insn
jcc 1b

Quoted 5 lines

ret
int3

Which isn't weird code by any means. And soon to be generated by compilers.

Hmm, yeah, I thought that was rare case.

Quoted 75 lines

Maybe something like:

struct queue {
	int head, tail;
	unsigned long val[16]; /* insufficient; probably should allocate something */
};

void push(struct queue *q, unsigned long val)

{
	/* break loops, been here already */
	for (int i = 0; i < q->head; i++) {
		if (q->val[i] == val)
			return;
	}

	q->val[q->head++] = val;

	WARN_ON(q->head > ARRAY_SIZE(q->val)
}

unsigned long pop(struct queue *q)

{
	if (q->tail == q->head)
		return 0;

	return q->val[q->tail++];
}

bool dead_end_insn(struct instruction *insn)

{
	switch (insn->opcode.bytes[0]) {
	case INT3_INSN_OPCODE:
	case JMP8_INSN_OPCODE:
	case JMP32_INSN_OPCODE:
		return true; /* no arch execution after these */

	case 0xff:
		/* jmp *%reg; jmpf */
		if (modrm_reg == 4 || modrm_reg == 5)
			return true;
		break;

	default:
		break;
	}

	return false;
}
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;

			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}
Quoted 1 line

It's a bit of a pain, but I think it should cover things.

Yeah, this looks good to me. What I just need is to add expanding queue buffer. (can we use xarray for this purpose?)

Thank you!

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
maintainer
#14 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 2 lines · Masami Hiramatsu

Yeah, this looks good to me. What I just need is to add expanding queue buffer. (can we use xarray for this purpose?)

Yeah, xarray might just work.

I need to go fetch the kids from school, but if I remember I'll modify objtool to tell us the max number required here (for any one particular build obviously).

maintainer
#15 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 9 lines · Peter Zijlstra
Quoted 2 lines · Masami Hiramatsu

Yeah, this looks good to me. What I just need is to add expanding queue buffer. (can we use xarray for this purpose?)

Yeah, xarray might just work.

I need to go fetch the kids from school, but if I remember I'll modify objtool to tell us the max number required here (for any one particular build obviously).

quick hack below suggests we need 405 for a x86_64 defconfig+kvm_guest.config vmlinux.o.

  $ OBJTOOL_ARGS="--stats" make O=defconfig-build/ -j12 vmlinux.o
  ...
  max_targets: 405 (hidinput_connect)
  ...

And if you look at the output of:

  $ ./scripts/objdump-func defconfig-build/vmlinux.o hidinput_connect

I'm inclined to believe this. That function is crazy. So yeah, we definitely need something dynamic.

---
 tools/objtool/check.c               | 12 ++++++++++++
 tools/objtool/include/objtool/elf.h |  1 +
 2 files changed, 13 insertions(+)
tools/objtool/check.c +12 -0
index e55fdf952a3a..897d3b83ab70 100644
@@ -3295,6 +3295,9 @@ static struct instruction *next_insn_to_validate(struct objtool_file *file,
 	return next_insn_same_sec(file, insn);
 }
 
+static int max_targets = 0;
+static char *max_name = NULL;
+
 /*
  * Follow the branch starting at the given instruction, and recursively follow
  * any other branches (jumps).  Meanwhile, track the frame pointer state at
@@ -3312,6 +3315,14 @@ static int validate_branch(struct objtool_file *file, struct symbol *func,
 
 	sec = insn->sec;
 
+	if (!insn->visited && func) {
+		func->targets++;
+		if (func->targets > max_targets) {
+			max_targets = func->targets;
+			max_name = func->name;
+		}
+	}
+
 	while (1) {
 		next_insn = next_insn_to_validate(file, insn);
 
@@ -4305,6 +4316,7 @@ int check(struct objtool_file *file)
 		printf("nr_cfi: %ld\n", nr_cfi);
 		printf("nr_cfi_reused: %ld\n", nr_cfi_reused);
 		printf("nr_cfi_cache: %ld\n", nr_cfi_cache);
+		printf("max_targets: %d (%s)\n", max_targets, max_name);
 	}
 
 out:
tools/objtool/include/objtool/elf.h +1 -0
index 16f4067b82ae..a707becdef50 100644
@@ -61,6 +61,7 @@ struct symbol {
 	u8 fentry            : 1;
 	u8 profiling_func    : 1;
 	struct list_head pv_target;
+	int targets;
 };
 
 struct reloc {
maintainer
#16 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

In reply to Peter Zijlstra
Quoted 11 lines · Peter Zijlstra
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;
Quoted 8 lines
			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}

There is the very rare case of intra-function-calls; but I *think* they're all in noinstr/nokprobe code anyway.

For instance we have RSB stuffing code like:

	.rept 16
	call 1f;
	int3
	1:
	.endr
	add $(BITS_PER_LONG/8) * 16, %_ASM_SP

And the proposed will be horribly confused by that. But like said; it should also never try and untangle it.

#17 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 34 lines · On Wed, 7 Sep 2022 15:05:13 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 11 lines · Peter Zijlstra
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;
Quoted 8 lines
			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}

There is the very rare case of intra-function-calls; but I *think* they're all in noinstr/nokprobe code anyway.

For instance we have RSB stuffing code like:

	.rept 16
	call 1f;
	int3
	1:
	.endr
	add $(BITS_PER_LONG/8) * 16, %_ASM_SP
Quoted 2 lines

And the proposed will be horribly confused by that. But like said; it should also never try and untangle it.

Yeah, but I guess if we break the decoding (internal) loop when we hit an INT3, it maybe possible to be handled?

Thank you,

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
maintainer
#18 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 37 lines · Masami Hiramatsu
Quoted 34 lines · On Wed, 7 Sep 2022 15:05:13 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 11 lines · Peter Zijlstra
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;
Quoted 8 lines
			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}

There is the very rare case of intra-function-calls; but I *think* they're all in noinstr/nokprobe code anyway.

For instance we have RSB stuffing code like:

	.rept 16
	call 1f;
	int3
	1:
	.endr
	add $(BITS_PER_LONG/8) * 16, %_ASM_SP
Quoted 5 lines
Quoted 2 lines

And the proposed will be horribly confused by that. But like said; it should also never try and untangle it.

Yeah, but I guess if we break the decoding (internal) loop when we hit an INT3, it maybe possible to be handled?

If you make insn_get_branch_addr() return the target of CALL instructions when this target is between function start and end, it should work I think.

But like said; this construct is rare and all instances I can remember should not be kprobes to begin with. These are all 'fun' things like retpoline stubs and the the above RSB stuff loop.

#19 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 41 lines · On Wed, 7 Sep 2022 16:27:59 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 40 lines · Masami Hiramatsu
Quoted 36 lines · On Wed, 7 Sep 2022 15:05:13 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 11 lines · Peter Zijlstra
	struct queue q;

	start = paddr - offset;
	end = start + size;
	push(&q, paddr - offset);

	while (start = pop(&q)) {
		for_each_insn(&insn, start, end, buf) {
			if (insn.kaddr == paddr)
				return 1;
Quoted 8 lines
			target = insn_get_branch_addr(&insn);
			if (target)
				push(&q, target);

			if (dead_end_insn(&insn))
				break;
		}
	}

There is the very rare case of intra-function-calls; but I *think* they're all in noinstr/nokprobe code anyway.

For instance we have RSB stuffing code like:

	.rept 16
	call 1f;
	int3
	1:
	.endr
	add $(BITS_PER_LONG/8) * 16, %_ASM_SP

And the proposed will be horribly confused by that. But like said; it should also never try and untangle it.

Quoted 6 lines
Quoted 2 lines

Yeah, but I guess if we break the decoding (internal) loop when we hit an INT3, it maybe possible to be handled?

If you make insn_get_branch_addr() return the target of CALL instructions when this target is between function start and end, it should work I think.

Ah Indeed. Anyway, I would like to use INT3 as a stop instruction, instread of checking dead_end_instruction. Is there any problem?

Quoted 4 lines

But like said; this construct is rare and all instances I can remember should not be kprobes to begin with. These are all 'fun' things like retpoline stubs and the the above RSB stuff loop.

Agree. That should not appear on normal code.

Thank you,

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
maintainer
#20 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

In reply to Masami Hiramatsu (Google)

Review

Quoted 3 lines · Masami Hiramatsu (Google)
(quoted diff) +0 -0
 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
 		return 0;

One more thing:

https://lkml.kernel.org/r/[email protected]

can result in negative offsets. The expression:

'paddr - offset'

will still get you to +0, but I might not have fully considered things when I wrote that patch.

#21 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 8 lines · On Wed, 7 Sep 2022 14:56:52 +0200 Peter Zijlstra <[email protected]> wrote:

Review

Quoted 3 lines · Masami Hiramatsu (Google)
(quoted diff) +0 -0
 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
 		return 0;

One more thing:

Quoted 8 lines

https://lkml.kernel.org/r/[email protected]

can result in negative offsets. The expression:

'paddr - offset'

will still get you to +0, but I might not have fully considered things when I wrote that patch.

Hmm, isn't 'offset' unsigned? If 'paddr - offset' is still available to find the function entry address, it is OK to me.

Thank you,

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
maintainer
#22 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 15 lines · Masami Hiramatsu
Quoted 12 lines · On Wed, 7 Sep 2022 14:56:52 +0200 Peter Zijlstra <[email protected]> wrote:

Review

Quoted 3 lines · Masami Hiramatsu (Google)
(quoted diff) +0 -0
 	if (!kallsyms_lookup_size_offset(paddr, NULL, &offset))
 		return 0;

One more thing:

https://lkml.kernel.org/r/[email protected]

can result in negative offsets. The expression:

Quoted 7 lines
Quoted 4 lines

'paddr - offset'

will still get you to +0, but I might not have fully considered things when I wrote that patch.

Hmm, isn't 'offset' unsigned? If 'paddr - offset' is still available to find the function entry address, it is OK to me.

Yeah, but the magic of 2s complement means it doesn't matter ;-)

maintainer
#23 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

In reply to Masami Hiramatsu (Google)
Quoted 12 lines · Masami Hiramatsu (Google)

From: Masami Hiramatsu (Google) <[email protected]>

Since the CONFIG_RETHUNK and CONFIG_SLS will use INT3 for padding after RET instruction, kprobes always failes to check the probed instruction boundary by decoding the function body if the probed address is after such paddings (Note that some conditional code blocks will be placed after RET instruction, if compiler decides it is not on the hot path.) This is because kprobes expects someone (e.g. kgdb) puts the INT3 as a software breakpoint and it will replace the original instruction. But There are INT3 just for padding in the function, it doesn't need to recover the original instruction.

Quoted 8 lines

To avoid this issue, if kprobe finds an INT3, it gets the address of next non-INT3 byte, and search a branch which jumps to the address. If there is the branch, these INT3 will be for padding, so it can be skipped.

Fixes 15e67227c49a ("x86: Undo return-thunk damage")

I take objection with this Fixes tag.. if anything it should be the SLS commit that predates this.

e463a09af2f0 ("x86: Add straight-line-speculation mitigation")

#24 lore raw

Re: [PATCH 1/2] x86/kprobes: Fix kprobes instruction boudary check with CONFIG_RETHUNK

Quoted 18 lines · On Wed, 7 Sep 2022 14:59:41 +0200 Peter Zijlstra <[email protected]> wrote:
Quoted 17 lines · Masami Hiramatsu (Google)

From: Masami Hiramatsu (Google) <[email protected]>

Since the CONFIG_RETHUNK and CONFIG_SLS will use INT3 for padding after RET instruction, kprobes always failes to check the probed instruction boundary by decoding the function body if the probed address is after such paddings (Note that some conditional code blocks will be placed after RET instruction, if compiler decides it is not on the hot path.) This is because kprobes expects someone (e.g. kgdb) puts the INT3 as a software breakpoint and it will replace the original instruction. But There are INT3 just for padding in the function, it doesn't need to recover the original instruction.

To avoid this issue, if kprobe finds an INT3, it gets the address of next non-INT3 byte, and search a branch which jumps to the address. If there is the branch, these INT3 will be for padding, so it can be skipped.

Quoted 8 lines
Quoted 3 lines
Fixes 15e67227c49a ("x86: Undo return-thunk damage")

I take objection with this Fixes tag.. if anything it should be the SLS commit that predates this.

e463a09af2f0 ("x86: Add straight-line-speculation mitigation")

Thanks, I'll change to it.

Signature · 1 line
Masami Hiramatsu (Google) <[email protected]>
#25 lore raw

PATCH 2/2 x86/kprobes: Fix optprobe optimization check with CONFIG_RETHUNK

In reply to Masami Hiramatsu (Google)

From: Masami Hiramatsu (Google) <[email protected]>

Since the CONFIG_RETHUNK and CONFIG_SLS will use INT3 for padding after RET instruction, kprobe jump optimization always fails on the functions with INT3 padding inside the function body. (It already checks the INT3 padding between functions, but not inside the function)

To avoid this issue, when it finds an INT3, read following bytes and find the next non-INT3 instruction, and decode the function again to search a branch which jumps to that address. If it can not find such branch instruction, it thinks that INT3 does not come from RETHUNK or SLS.

Fixes 15e67227c49a ("x86: Undo return-thunk damage")
---
 arch/x86/kernel/kprobes/opt.c |   70 +++++++++++++++++++----------------------
 1 file changed, 33 insertions(+), 37 deletions(-)
arch/x86/kernel/kprobes/opt.c +33 -37
index 2e41850cab06..ed77eeeef4ed 100644
@@ -260,25 +260,12 @@ static int insn_is_indirect_jump(struct insn *insn)
 	return ret;
 }
 
-static bool is_padding_int3(unsigned long addr, unsigned long eaddr)
-{
-	unsigned char ops;
-
-	for (; addr < eaddr; addr++) {
-		if (get_kernel_nofault(ops, (void *)addr) < 0 ||
-		    ops != INT3_INSN_OPCODE)
-			return false;
-	}
-
-	return true;
-}
-
 /* Decode whole function to ensure any instructions don't jump into target */
 static int can_optimize(unsigned long paddr)
 {
-	unsigned long addr, size = 0, offset = 0;
-	struct insn insn;
 	kprobe_opcode_t buf[MAX_INSN_SIZE];
+	unsigned long size = 0, offset = 0;
+	struct insn insn;
 
 	/* Lookup symbol including addr */
 	if (!kallsyms_lookup_size_offset(paddr, &size, &offset))
@@ -296,11 +283,9 @@ static int can_optimize(unsigned long paddr)
 	if (size - offset < JMP32_INSN_SIZE)
 		return 0;
 
-	/* Decode instructions */
-	addr = paddr - offset;
-	while (addr < paddr - offset + size) { /* Decode until function end */
-		unsigned long recovered_insn;
-		int ret;
+	/* Decode all instructions in the function */
+	for_each_insn(&insn, paddr - offset, paddr - offset + size, buf) {
+		unsigned long addr = (unsigned long)insn.kaddr;
 
 		if (search_exception_tables(addr))
 			/*
@@ -308,31 +293,42 @@ static int can_optimize(unsigned long paddr)
 			 * we can't optimize kprobe in this function.
 			 */
 			return 0;
-		recovered_insn = recover_probed_instruction(buf, addr);
-		if (!recovered_insn)
-			return 0;
 
-		ret = insn_decode_kernel(&insn, (void *)recovered_insn);
-		if (ret < 0)
+		if (insn.opcode.bytes[0] == INT3_INSN_OPCODE) {
+			addr = skip_padding_int3(addr);
+			if (!addr)
+				return 0;
+			/*
+			 * If addr becomes the next function entry, this is
+			 * the INT3 padding between functions.
+			 */
+			if (addr - 1 == paddr - offset + size)
+				return 1;
+
+			/*
+			 * This can be padding INT3 for CONFIG_RETHUNK or
+			 * CONFIG_SLS. If a branch jumps to the address next
+			 * to the INT3 sequence, this is just for padding,
+			 * then we can continue decoding.
+			 */
+			for_each_insn(&insn, paddr - offset, addr, buf) {
+				if (insn_get_branch_addr(&insn) == addr)
+					goto found;
+			}
+
+			/* This INT3 can not be decoded safely. */
 			return 0;
+found:
+			/* Set loop cursor */
+			insn.next_byte = (void *)addr;
+			continue;
+		}
 
-		/*
-		 * In the case of detecting unknown breakpoint, this could be
-		 * a padding INT3 between functions. Let's check that all the
-		 * rest of the bytes are also INT3.
-		 */
-		if (insn.opcode.bytes[0] == INT3_INSN_OPCODE)
-			return is_padding_int3(addr, paddr - offset + size) ? 1 : 0;
-
-		/* Recover address */
-		insn.kaddr = (void *)addr;
-		insn.next_byte = (void *)(addr + insn.length);
 		/* Check any instructions don't jump into target */
 		if (insn_is_indirect_jump(&insn) ||
 		    insn_jump_into_range(&insn, paddr + INT3_INSN_SIZE,
 					 DISP32_SIZE))
 			return 0;
-		addr += insn.length;
 	}
 
 	return 1;
maintainer
#26 lore raw
In reply to Masami Hiramatsu (Google)
Quoted 9 lines · Masami Hiramatsu (Google)

Hi,

Here is a couple of patches to fix kprobes and optprobe to work on the kernel with CONFIG_RETHUNK and CONFIG_SLS.

With these configs, the kernel functions may includes padding INT3 in the function code block (body) in addition to the gaps between functions.

Since kprobes on x86 has to ensure the probe address is a function

s/function/instruction/

Quoted 6 lines

bondary, it decodes the instructions in the function until the address. If it finds an INT3 which is not embedded by kprobe, it stops decoding because usually the INT3 is used for debugging as a software breakpoint and such INT3 will replace the first byte of an original instruction. Without recovering it, kprobes can not continue to decode it. Thus the kprobes returns -EILSEQ as below.

In the absence of kgdb nobody else except kprobes itself will do this.

Quoted 5 lines
 # echo "p:probe/vfs_truncate_L19 vfs_truncate+98" >> kprobe_events 
 sh: write error: Invalid or incomplete multibyte or wide character

Actually, those INT3s are just for padding and can be ignored.

They are speculations stops, not mere padding.

Anyway, let me get on with reading the actual patches :-)