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, 167 insertions, 0 deletions

diff --git a/arch/mips/sni/time.c b/arch/mips/sni/time.c
new file mode 100644
index 000000000..ff3ba7e77
--- /dev/null
+++ b/arch/mips/sni/time.c
@@ -0,0 +1,167 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/types.h>
+#include <linux/i8253.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/smp.h>
+#include <linux/time.h>
+#include <linux/clockchips.h>
+
+#include <asm/sni.h>
+#include <asm/time.h>
+
+#define SNI_CLOCK_TICK_RATE	3686400
+#define SNI_COUNTER2_DIV	64
+#define SNI_COUNTER0_DIV	((SNI_CLOCK_TICK_RATE / SNI_COUNTER2_DIV) / HZ)
+
+static int a20r_set_periodic(struct clock_event_device *evt)
+{
+	*(volatile u8 *)(A20R_PT_CLOCK_BASE + 12) = 0x34;
+	wmb();
+	*(volatile u8 *)(A20R_PT_CLOCK_BASE + 0) = SNI_COUNTER0_DIV & 0xff;
+	wmb();
+	*(volatile u8 *)(A20R_PT_CLOCK_BASE + 0) = SNI_COUNTER0_DIV >> 8;
+	wmb();
+
+	*(volatile u8 *)(A20R_PT_CLOCK_BASE + 12) = 0xb4;
+	wmb();
+	*(volatile u8 *)(A20R_PT_CLOCK_BASE + 8) = SNI_COUNTER2_DIV & 0xff;
+	wmb();
+	*(volatile u8 *)(A20R_PT_CLOCK_BASE + 8) = SNI_COUNTER2_DIV >> 8;
+	wmb();
+	return 0;
+}
+
+static struct clock_event_device a20r_clockevent_device = {
+	.name			= "a20r-timer",
+	.features		= CLOCK_EVT_FEAT_PERIODIC,
+
+	/* .mult, .shift, .max_delta_ns and .min_delta_ns left uninitialized */
+
+	.rating			= 300,
+	.irq			= SNI_A20R_IRQ_TIMER,
+	.set_state_periodic	= a20r_set_periodic,
+};
+
+static irqreturn_t a20r_interrupt(int irq, void *dev_id)
+{
+	struct clock_event_device *cd = dev_id;
+
+	*(volatile u8 *)A20R_PT_TIM0_ACK = 0;
+	wmb();
+
+	cd->event_handler(cd);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * a20r platform uses 2 counters to divide the input frequency.
+ * Counter 2 output is connected to Counter 0 & 1 input.
+ */
+static void __init sni_a20r_timer_setup(void)
+{
+	struct clock_event_device *cd = &a20r_clockevent_device;
+	unsigned int cpu = smp_processor_id();
+
+	cd->cpumask		= cpumask_of(cpu);
+	clockevents_register_device(cd);
+	if (request_irq(SNI_A20R_IRQ_TIMER, a20r_interrupt,
+			IRQF_PERCPU | IRQF_TIMER, "a20r-timer", cd))
+		pr_err("Failed to register a20r-timer interrupt\n");
+}
+
+#define SNI_8254_TICK_RATE	  1193182UL
+
+#define SNI_8254_TCSAMP_COUNTER	  ((SNI_8254_TICK_RATE / HZ) + 255)
+
+static __init unsigned long dosample(void)
+{
+	u32 ct0, ct1;
+	volatile u8 msb;
+
+	/* Start the counter. */
+	outb_p(0x34, 0x43);
+	outb_p(SNI_8254_TCSAMP_COUNTER & 0xff, 0x40);
+	outb(SNI_8254_TCSAMP_COUNTER >> 8, 0x40);
+
+	/* Get initial counter invariant */
+	ct0 = read_c0_count();
+
+	/* Latch and spin until top byte of counter0 is zero */
+	do {
+		outb(0x00, 0x43);
+		(void) inb(0x40);
+		msb = inb(0x40);
+		ct1 = read_c0_count();
+	} while (msb);
+
+	/* Stop the counter. */
+	outb(0x38, 0x43);
+	/*
+	 * Return the difference, this is how far the r4k counter increments
+	 * for every 1/HZ seconds. We round off the nearest 1 MHz of master
+	 * clock (= 1000000 / HZ / 2).
+	 */
+	/*return (ct1 - ct0 + (500000/HZ/2)) / (500000/HZ) * (500000/HZ);*/
+	return (ct1 - ct0) / (500000/HZ) * (500000/HZ);
+}
+
+/*
+ * Here we need to calibrate the cycle counter to at least be close.
+ */
+void __init plat_time_init(void)
+{
+	unsigned long r4k_ticks[3];
+	unsigned long r4k_tick;
+
+	/*
+	 * Figure out the r4k offset, the algorithm is very simple and works in
+	 * _all_ cases as long as the 8254 counter register itself works ok (as
+	 * an interrupt driving timer it does not because of bug, this is why
+	 * we are using the onchip r4k counter/compare register to serve this
+	 * purpose, but for r4k_offset calculation it will work ok for us).
+	 * There are other very complicated ways of performing this calculation
+	 * but this one works just fine so I am not going to futz around. ;-)
+	 */
+	printk(KERN_INFO "Calibrating system timer... ");
+	dosample();	/* Prime cache. */
+	dosample();	/* Prime cache. */
+	/* Zero is NOT an option. */
+	do {
+		r4k_ticks[0] = dosample();
+	} while (!r4k_ticks[0]);
+	do {
+		r4k_ticks[1] = dosample();
+	} while (!r4k_ticks[1]);
+
+	if (r4k_ticks[0] != r4k_ticks[1]) {
+		printk("warning: timer counts differ, retrying... ");
+		r4k_ticks[2] = dosample();
+		if (r4k_ticks[2] == r4k_ticks[0]
+		    || r4k_ticks[2] == r4k_ticks[1])
+			r4k_tick = r4k_ticks[2];
+		else {
+			printk("disagreement, using average... ");
+			r4k_tick = (r4k_ticks[0] + r4k_ticks[1]
+				   + r4k_ticks[2]) / 3;
+		}
+	} else
+		r4k_tick = r4k_ticks[0];
+
+	printk("%d [%d.%04d MHz CPU]\n", (int) r4k_tick,
+		(int) (r4k_tick / (500000 / HZ)),
+		(int) (r4k_tick % (500000 / HZ)));
+
+	mips_hpt_frequency = r4k_tick * HZ;
+
+	switch (sni_brd_type) {
+	case SNI_BRD_10:
+	case SNI_BRD_10NEW:
+	case SNI_BRD_TOWER_OASIC:
+	case SNI_BRD_MINITOWER:
+		sni_a20r_timer_setup();
+		break;
+	}
+	setup_pit_timer();
+}