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, 219 insertions, 0 deletions

diff --git a/arch/powerpc/boot/oflib.c b/arch/powerpc/boot/oflib.c
new file mode 100644
index 000000000..8759c985e
--- /dev/null
+++ b/arch/powerpc/boot/oflib.c
@@ -0,0 +1,219 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) Paul Mackerras 1997.
+ */
+#include <stddef.h>
+#include "types.h"
+#include "elf.h"
+#include "string.h"
+#include "stdio.h"
+#include "page.h"
+#include "ops.h"
+
+#include "of.h"
+
+typedef u32 prom_arg_t;
+
+/* The following structure is used to communicate with open firmware.
+ * All arguments in and out are in big endian format. */
+struct prom_args {
+	__be32 service;	/* Address of service name string. */
+	__be32 nargs;	/* Number of input arguments. */
+	__be32 nret;	/* Number of output arguments. */
+	__be32 args[10];	/* Input/output arguments. */
+};
+
+#ifdef __powerpc64__
+extern int prom(void *);
+#else
+static int (*prom) (void *);
+#endif
+
+void of_init(void *promptr)
+{
+#ifndef __powerpc64__
+	prom = (int (*)(void *))promptr;
+#endif
+}
+
+#define ADDR(x)		(u32)(unsigned long)(x)
+
+int of_call_prom(const char *service, int nargs, int nret, ...)
+{
+	int i;
+	struct prom_args args;
+	va_list list;
+
+	args.service = cpu_to_be32(ADDR(service));
+	args.nargs = cpu_to_be32(nargs);
+	args.nret = cpu_to_be32(nret);
+
+	va_start(list, nret);
+	for (i = 0; i < nargs; i++)
+		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
+	va_end(list);
+
+	for (i = 0; i < nret; i++)
+		args.args[nargs+i] = 0;
+
+	if (prom(&args) < 0)
+		return PROM_ERROR;
+
+	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
+}
+
+static int of_call_prom_ret(const char *service, int nargs, int nret,
+			    prom_arg_t *rets, ...)
+{
+	int i;
+	struct prom_args args;
+	va_list list;
+
+	args.service = cpu_to_be32(ADDR(service));
+	args.nargs = cpu_to_be32(nargs);
+	args.nret = cpu_to_be32(nret);
+
+	va_start(list, rets);
+	for (i = 0; i < nargs; i++)
+		args.args[i] = cpu_to_be32(va_arg(list, prom_arg_t));
+	va_end(list);
+
+	for (i = 0; i < nret; i++)
+		args.args[nargs+i] = 0;
+
+	if (prom(&args) < 0)
+		return PROM_ERROR;
+
+	if (rets != NULL)
+		for (i = 1; i < nret; ++i)
+			rets[i-1] = be32_to_cpu(args.args[nargs+i]);
+
+	return (nret > 0) ? be32_to_cpu(args.args[nargs]) : 0;
+}
+
+/* returns true if s2 is a prefix of s1 */
+static int string_match(const char *s1, const char *s2)
+{
+	for (; *s2; ++s2)
+		if (*s1++ != *s2)
+			return 0;
+	return 1;
+}
+
+/*
+ * Older OF's require that when claiming a specific range of addresses,
+ * we claim the physical space in the /memory node and the virtual
+ * space in the chosen mmu node, and then do a map operation to
+ * map virtual to physical.
+ */
+static int need_map = -1;
+static ihandle chosen_mmu;
+static ihandle memory;
+
+static int check_of_version(void)
+{
+	phandle oprom, chosen;
+	char version[64];
+
+	oprom = of_finddevice("/openprom");
+	if (oprom == (phandle) -1)
+		return 0;
+	if (of_getprop(oprom, "model", version, sizeof(version)) <= 0)
+		return 0;
+	version[sizeof(version)-1] = 0;
+	printf("OF version = '%s'\r\n", version);
+	if (!string_match(version, "Open Firmware, 1.")
+	    && !string_match(version, "FirmWorks,3."))
+		return 0;
+	chosen = of_finddevice("/chosen");
+	if (chosen == (phandle) -1) {
+		chosen = of_finddevice("/chosen@0");
+		if (chosen == (phandle) -1) {
+			printf("no chosen\n");
+			return 0;
+		}
+	}
+	if (of_getprop(chosen, "mmu", &chosen_mmu, sizeof(chosen_mmu)) <= 0) {
+		printf("no mmu\n");
+		return 0;
+	}
+	memory = of_call_prom("open", 1, 1, "/memory");
+	if (memory == PROM_ERROR) {
+		memory = of_call_prom("open", 1, 1, "/memory@0");
+		if (memory == PROM_ERROR) {
+			printf("no memory node\n");
+			return 0;
+		}
+	}
+	printf("old OF detected\r\n");
+	return 1;
+}
+
+unsigned int of_claim(unsigned long virt, unsigned long size,
+		      unsigned long align)
+{
+	int ret;
+	prom_arg_t result;
+
+	if (need_map < 0)
+		need_map = check_of_version();
+	if (align || !need_map)
+		return of_call_prom("claim", 3, 1, virt, size, align);
+
+	ret = of_call_prom_ret("call-method", 5, 2, &result, "claim", memory,
+			       align, size, virt);
+	if (ret != 0 || result == -1)
+		return  -1;
+	ret = of_call_prom_ret("call-method", 5, 2, &result, "claim", chosen_mmu,
+			       align, size, virt);
+	/* 0x12 == coherent + read/write */
+	ret = of_call_prom("call-method", 6, 1, "map", chosen_mmu,
+			   0x12, size, virt, virt);
+	return virt;
+}
+
+void *of_vmlinux_alloc(unsigned long size)
+{
+	unsigned long start = (unsigned long)_start, end = (unsigned long)_end;
+	unsigned long addr;
+	void *p;
+
+	/* With some older POWER4 firmware we need to claim the area the kernel
+	 * will reside in.  Newer firmwares don't need this so we just ignore
+	 * the return value.
+	 */
+	addr = (unsigned long) of_claim(start, end - start, 0);
+	printf("Trying to claim from 0x%lx to 0x%lx (0x%lx) got %lx\r\n",
+	       start, end, end - start, addr);
+
+	p = malloc(size);
+	if (!p)
+		fatal("Can't allocate memory for kernel image!\n\r");
+
+	return p;
+}
+
+void of_exit(void)
+{
+	of_call_prom("exit", 0, 0);
+}
+
+/*
+ * OF device tree routines
+ */
+void *of_finddevice(const char *name)
+{
+	return (void *) (unsigned long) of_call_prom("finddevice", 1, 1, name);
+}
+
+int of_getprop(const void *phandle, const char *name, void *buf,
+	       const int buflen)
+{
+	return of_call_prom("getprop", 4, 1, phandle, name, buf, buflen);
+}
+
+int of_setprop(const void *phandle, const char *name, const void *buf,
+	       const int buflen)
+{
+	return of_call_prom("setprop", 4, 1, phandle, name, buf, buflen);
+}