x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 133 insertions, 0 deletions

diff --git a/lib/math/int_log.c b/lib/math/int_log.c
new file mode 100644
index 000000000..8f9da3a2a
--- /dev/null
+++ b/lib/math/int_log.c
@@ -0,0 +1,133 @@
+// SPDX-License-Identifier: LGPL-2.1-or-later
+/*
+ * Provides fixed-point logarithm operations.
+ *
+ * Copyright (C) 2006 Christoph Pfister (christophpfister@gmail.com)
+ */
+
+#include <linux/bitops.h>
+#include <linux/export.h>
+#include <linux/int_log.h>
+#include <linux/kernel.h>
+#include <linux/types.h>
+
+#include <asm/bug.h>
+
+static const unsigned short logtable[256] = {
+	0x0000, 0x0171, 0x02e0, 0x044e, 0x05ba, 0x0725, 0x088e, 0x09f7,
+	0x0b5d, 0x0cc3, 0x0e27, 0x0f8a, 0x10eb, 0x124b, 0x13aa, 0x1508,
+	0x1664, 0x17bf, 0x1919, 0x1a71, 0x1bc8, 0x1d1e, 0x1e73, 0x1fc6,
+	0x2119, 0x226a, 0x23ba, 0x2508, 0x2656, 0x27a2, 0x28ed, 0x2a37,
+	0x2b80, 0x2cc8, 0x2e0f, 0x2f54, 0x3098, 0x31dc, 0x331e, 0x345f,
+	0x359f, 0x36de, 0x381b, 0x3958, 0x3a94, 0x3bce, 0x3d08, 0x3e41,
+	0x3f78, 0x40af, 0x41e4, 0x4319, 0x444c, 0x457f, 0x46b0, 0x47e1,
+	0x4910, 0x4a3f, 0x4b6c, 0x4c99, 0x4dc5, 0x4eef, 0x5019, 0x5142,
+	0x526a, 0x5391, 0x54b7, 0x55dc, 0x5700, 0x5824, 0x5946, 0x5a68,
+	0x5b89, 0x5ca8, 0x5dc7, 0x5ee5, 0x6003, 0x611f, 0x623a, 0x6355,
+	0x646f, 0x6588, 0x66a0, 0x67b7, 0x68ce, 0x69e4, 0x6af8, 0x6c0c,
+	0x6d20, 0x6e32, 0x6f44, 0x7055, 0x7165, 0x7274, 0x7383, 0x7490,
+	0x759d, 0x76aa, 0x77b5, 0x78c0, 0x79ca, 0x7ad3, 0x7bdb, 0x7ce3,
+	0x7dea, 0x7ef0, 0x7ff6, 0x80fb, 0x81ff, 0x8302, 0x8405, 0x8507,
+	0x8608, 0x8709, 0x8809, 0x8908, 0x8a06, 0x8b04, 0x8c01, 0x8cfe,
+	0x8dfa, 0x8ef5, 0x8fef, 0x90e9, 0x91e2, 0x92db, 0x93d2, 0x94ca,
+	0x95c0, 0x96b6, 0x97ab, 0x98a0, 0x9994, 0x9a87, 0x9b7a, 0x9c6c,
+	0x9d5e, 0x9e4f, 0x9f3f, 0xa02e, 0xa11e, 0xa20c, 0xa2fa, 0xa3e7,
+	0xa4d4, 0xa5c0, 0xa6ab, 0xa796, 0xa881, 0xa96a, 0xaa53, 0xab3c,
+	0xac24, 0xad0c, 0xadf2, 0xaed9, 0xafbe, 0xb0a4, 0xb188, 0xb26c,
+	0xb350, 0xb433, 0xb515, 0xb5f7, 0xb6d9, 0xb7ba, 0xb89a, 0xb97a,
+	0xba59, 0xbb38, 0xbc16, 0xbcf4, 0xbdd1, 0xbead, 0xbf8a, 0xc065,
+	0xc140, 0xc21b, 0xc2f5, 0xc3cf, 0xc4a8, 0xc580, 0xc658, 0xc730,
+	0xc807, 0xc8de, 0xc9b4, 0xca8a, 0xcb5f, 0xcc34, 0xcd08, 0xcddc,
+	0xceaf, 0xcf82, 0xd054, 0xd126, 0xd1f7, 0xd2c8, 0xd399, 0xd469,
+	0xd538, 0xd607, 0xd6d6, 0xd7a4, 0xd872, 0xd93f, 0xda0c, 0xdad9,
+	0xdba5, 0xdc70, 0xdd3b, 0xde06, 0xded0, 0xdf9a, 0xe063, 0xe12c,
+	0xe1f5, 0xe2bd, 0xe385, 0xe44c, 0xe513, 0xe5d9, 0xe69f, 0xe765,
+	0xe82a, 0xe8ef, 0xe9b3, 0xea77, 0xeb3b, 0xebfe, 0xecc1, 0xed83,
+	0xee45, 0xef06, 0xefc8, 0xf088, 0xf149, 0xf209, 0xf2c8, 0xf387,
+	0xf446, 0xf505, 0xf5c3, 0xf680, 0xf73e, 0xf7fb, 0xf8b7, 0xf973,
+	0xfa2f, 0xfaea, 0xfba5, 0xfc60, 0xfd1a, 0xfdd4, 0xfe8e, 0xff47,
+};
+
+unsigned int intlog2(u32 value)
+{
+	/**
+	 *	returns: log2(value) * 2^24
+	 *	wrong result if value = 0 (log2(0) is undefined)
+	 */
+	unsigned int msb;
+	unsigned int logentry;
+	unsigned int significand;
+	unsigned int interpolation;
+
+	if (unlikely(value == 0)) {
+		WARN_ON(1);
+		return 0;
+	}
+
+	/* first detect the msb (count begins at 0) */
+	msb = fls(value) - 1;
+
+	/**
+	 *	now we use a logtable after the following method:
+	 *
+	 *	log2(2^x * y) * 2^24 = x * 2^24 + log2(y) * 2^24
+	 *	where x = msb and therefore 1 <= y < 2
+	 *	first y is determined by shifting the value left
+	 *	so that msb is bit 31
+	 *		0x00231f56 -> 0x8C7D5800
+	 *	the result is y * 2^31 -> "significand"
+	 *	then the highest 9 bits are used for a table lookup
+	 *	the highest bit is discarded because it's always set
+	 *	the highest nine bits in our example are 100011000
+	 *	so we would use the entry 0x18
+	 */
+	significand = value << (31 - msb);
+	logentry = (significand >> 23) % ARRAY_SIZE(logtable);
+
+	/**
+	 *	last step we do is interpolation because of the
+	 *	limitations of the log table the error is that part of
+	 *	the significand which isn't used for lookup then we
+	 *	compute the ratio between the error and the next table entry
+	 *	and interpolate it between the log table entry used and the
+	 *	next one the biggest error possible is 0x7fffff
+	 *	(in our example it's 0x7D5800)
+	 *	needed value for next table entry is 0x800000
+	 *	so the interpolation is
+	 *	(error / 0x800000) * (logtable_next - logtable_current)
+	 *	in the implementation the division is moved to the end for
+	 *	better accuracy there is also an overflow correction if
+	 *	logtable_next is 256
+	 */
+	interpolation = ((significand & 0x7fffff) *
+			((logtable[(logentry + 1) % ARRAY_SIZE(logtable)] -
+			  logtable[logentry]) & 0xffff)) >> 15;
+
+	/* now we return the result */
+	return ((msb << 24) + (logtable[logentry] << 8) + interpolation);
+}
+EXPORT_SYMBOL(intlog2);
+
+unsigned int intlog10(u32 value)
+{
+	/**
+	 *	returns: log10(value) * 2^24
+	 *	wrong result if value = 0 (log10(0) is undefined)
+	 */
+	u64 log;
+
+	if (unlikely(value == 0)) {
+		WARN_ON(1);
+		return 0;
+	}
+
+	log = intlog2(value);
+
+	/**
+	 *	we use the following method:
+	 *	log10(x) = log2(x) * log10(2)
+	 */
+
+	return (log * 646456993) >> 31;
+}
+EXPORT_SYMBOL(intlog10);