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, 288 insertions, 0 deletions

diff --git a/drivers/macintosh/macio-adb.c b/drivers/macintosh/macio-adb.c
new file mode 100644
index 000000000..19c63959e
--- /dev/null
+++ b/drivers/macintosh/macio-adb.c
@@ -0,0 +1,288 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Driver for the ADB controller in the Mac I/O (Hydra) chip.
+ */
+#include <linux/types.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/delay.h>
+#include <linux/spinlock.h>
+#include <linux/interrupt.h>
+#include <linux/pgtable.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/adb.h>
+
+#include <asm/io.h>
+#include <asm/hydra.h>
+#include <asm/irq.h>
+#include <linux/init.h>
+#include <linux/ioport.h>
+
+struct preg {
+	unsigned char r;
+	char pad[15];
+};
+
+struct adb_regs {
+	struct preg intr;
+	struct preg data[9];
+	struct preg intr_enb;
+	struct preg dcount;
+	struct preg error;
+	struct preg ctrl;
+	struct preg autopoll;
+	struct preg active_hi;
+	struct preg active_lo;
+	struct preg test;
+};
+
+/* Bits in intr and intr_enb registers */
+#define DFB	1		/* data from bus */
+#define TAG	2		/* transfer access grant */
+
+/* Bits in dcount register */
+#define HMB	0x0f		/* how many bytes */
+#define APD	0x10		/* auto-poll data */
+
+/* Bits in error register */
+#define NRE	1		/* no response error */
+#define DLE	2		/* data lost error */
+
+/* Bits in ctrl register */
+#define TAR	1		/* transfer access request */
+#define DTB	2		/* data to bus */
+#define CRE	4		/* command response expected */
+#define ADB_RST	8		/* ADB reset */
+
+/* Bits in autopoll register */
+#define APE	1		/* autopoll enable */
+
+static volatile struct adb_regs __iomem *adb;
+static struct adb_request *current_req, *last_req;
+static DEFINE_SPINLOCK(macio_lock);
+
+static int macio_probe(void);
+static int macio_init(void);
+static irqreturn_t macio_adb_interrupt(int irq, void *arg);
+static int macio_send_request(struct adb_request *req, int sync);
+static int macio_adb_autopoll(int devs);
+static void macio_adb_poll(void);
+static int macio_adb_reset_bus(void);
+
+struct adb_driver macio_adb_driver = {
+	.name         = "MACIO",
+	.probe        = macio_probe,
+	.init         = macio_init,
+	.send_request = macio_send_request,
+	.autopoll     = macio_adb_autopoll,
+	.poll         = macio_adb_poll,
+	.reset_bus    = macio_adb_reset_bus,
+};
+
+int macio_probe(void)
+{
+	struct device_node *np __free(device_node) =
+		of_find_compatible_node(NULL, "adb", "chrp,adb0");
+
+	if (np)
+		return 0;
+
+	return -ENODEV;
+}
+
+int macio_init(void)
+{
+	struct device_node *adbs __free(device_node) =
+		of_find_compatible_node(NULL, "adb", "chrp,adb0");
+	struct resource r;
+	unsigned int irq;
+
+	if (!adbs)
+		return -ENXIO;
+
+	if (of_address_to_resource(adbs, 0, &r))
+		return -ENXIO;
+
+	adb = ioremap(r.start, sizeof(struct adb_regs));
+	if (!adb)
+		return -ENOMEM;
+
+
+	out_8(&adb->ctrl.r, 0);
+	out_8(&adb->intr.r, 0);
+	out_8(&adb->error.r, 0);
+	out_8(&adb->active_hi.r, 0xff); /* for now, set all devices active */
+	out_8(&adb->active_lo.r, 0xff);
+	out_8(&adb->autopoll.r, APE);
+
+	irq = irq_of_parse_and_map(adbs, 0);
+	if (request_irq(irq, macio_adb_interrupt, 0, "ADB", (void *)0)) {
+		iounmap(adb);
+		printk(KERN_ERR "ADB: can't get irq %d\n", irq);
+		return -EAGAIN;
+	}
+	out_8(&adb->intr_enb.r, DFB | TAG);
+
+	printk("adb: mac-io driver 1.0 for unified ADB\n");
+
+	return 0;
+}
+
+static int macio_adb_autopoll(int devs)
+{
+	unsigned long flags;
+	
+	spin_lock_irqsave(&macio_lock, flags);
+	out_8(&adb->active_hi.r, devs >> 8);
+	out_8(&adb->active_lo.r, devs);
+	out_8(&adb->autopoll.r, devs? APE: 0);
+	spin_unlock_irqrestore(&macio_lock, flags);
+	return 0;
+}
+
+static int macio_adb_reset_bus(void)
+{
+	unsigned long flags;
+	int timeout = 1000000;
+
+	/* Hrm... we may want to not lock interrupts for so
+	 * long ... oh well, who uses that chip anyway ? :)
+	 * That function will be seldom used during boot
+	 * on rare machines, so...
+	 */
+	spin_lock_irqsave(&macio_lock, flags);
+	out_8(&adb->ctrl.r, in_8(&adb->ctrl.r) | ADB_RST);
+	while ((in_8(&adb->ctrl.r) & ADB_RST) != 0) {
+		if (--timeout == 0) {
+			out_8(&adb->ctrl.r, in_8(&adb->ctrl.r) & ~ADB_RST);
+			spin_unlock_irqrestore(&macio_lock, flags);
+			return -1;
+		}
+	}
+	spin_unlock_irqrestore(&macio_lock, flags);
+	return 0;
+}
+
+/* Send an ADB command */
+static int macio_send_request(struct adb_request *req, int sync)
+{
+	unsigned long flags;
+	int i;
+	
+	if (req->data[0] != ADB_PACKET)
+		return -EINVAL;
+	
+	for (i = 0; i < req->nbytes - 1; ++i)
+		req->data[i] = req->data[i+1];
+	--req->nbytes;
+	
+	req->next = NULL;
+	req->sent = 0;
+	req->complete = 0;
+	req->reply_len = 0;
+
+	spin_lock_irqsave(&macio_lock, flags);
+	if (current_req) {
+		last_req->next = req;
+		last_req = req;
+	} else {
+		current_req = last_req = req;
+		out_8(&adb->ctrl.r, in_8(&adb->ctrl.r) | TAR);
+	}
+	spin_unlock_irqrestore(&macio_lock, flags);
+	
+	if (sync) {
+		while (!req->complete)
+			macio_adb_poll();
+	}
+
+	return 0;
+}
+
+static irqreturn_t macio_adb_interrupt(int irq, void *arg)
+{
+	int i, n, err;
+	struct adb_request *req = NULL;
+	unsigned char ibuf[16];
+	int ibuf_len = 0;
+	int complete = 0;
+	int autopoll = 0;
+	int handled = 0;
+
+	spin_lock(&macio_lock);
+	if (in_8(&adb->intr.r) & TAG) {
+		handled = 1;
+		req = current_req;
+		if (req) {
+			/* put the current request in */
+			for (i = 0; i < req->nbytes; ++i)
+				out_8(&adb->data[i].r, req->data[i]);
+			out_8(&adb->dcount.r, req->nbytes & HMB);
+			req->sent = 1;
+			if (req->reply_expected) {
+				out_8(&adb->ctrl.r, DTB + CRE);
+			} else {
+				out_8(&adb->ctrl.r, DTB);
+				current_req = req->next;
+				complete = 1;
+				if (current_req)
+					out_8(&adb->ctrl.r, in_8(&adb->ctrl.r) | TAR);
+			}
+		}
+		out_8(&adb->intr.r, 0);
+	}
+
+	if (in_8(&adb->intr.r) & DFB) {
+		handled = 1;
+		err = in_8(&adb->error.r);
+		if (current_req && current_req->sent) {
+			/* this is the response to a command */
+			req = current_req;
+			if (err == 0) {
+				req->reply_len = in_8(&adb->dcount.r) & HMB;
+				for (i = 0; i < req->reply_len; ++i)
+					req->reply[i] = in_8(&adb->data[i].r);
+			}
+			current_req = req->next;
+			complete = 1;
+			if (current_req)
+				out_8(&adb->ctrl.r, in_8(&adb->ctrl.r) | TAR);
+		} else if (err == 0) {
+			/* autopoll data */
+			n = in_8(&adb->dcount.r) & HMB;
+			for (i = 0; i < n; ++i)
+				ibuf[i] = in_8(&adb->data[i].r);
+			ibuf_len = n;
+			autopoll = (in_8(&adb->dcount.r) & APD) != 0;
+		}
+		out_8(&adb->error.r, 0);
+		out_8(&adb->intr.r, 0);
+	}
+	spin_unlock(&macio_lock);
+	if (complete && req) {
+	    void (*done)(struct adb_request *) = req->done;
+	    mb();
+	    req->complete = 1;
+	    /* Here, we assume that if the request has a done member, the
+    	     * struct request will survive to setting req->complete to 1
+	     */
+	    if (done)
+		(*done)(req);
+	}
+	if (ibuf_len)
+		adb_input(ibuf, ibuf_len, autopoll);
+
+	return IRQ_RETVAL(handled);
+}
+
+static void macio_adb_poll(void)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+	if (in_8(&adb->intr.r) != 0)
+		macio_adb_interrupt(0, NULL);
+	local_irq_restore(flags);
+}