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, 335 insertions, 0 deletions

diff --git a/drivers/ptp/ptp_dte.c b/drivers/ptp/ptp_dte.c
new file mode 100644
index 000000000..834ecacc7
--- /dev/null
+++ b/drivers/ptp/ptp_dte.c
@@ -0,0 +1,335 @@
+// SPDX-License-Identifier: GPL-2.0-only
+// Copyright 2017 Broadcom
+
+#include <linux/err.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/platform_device.h>
+#include <linux/ptp_clock_kernel.h>
+#include <linux/types.h>
+
+#define DTE_NCO_LOW_TIME_REG	0x00
+#define DTE_NCO_TIME_REG	0x04
+#define DTE_NCO_OVERFLOW_REG	0x08
+#define DTE_NCO_INC_REG		0x0c
+
+#define DTE_NCO_SUM2_MASK	0xffffffff
+#define DTE_NCO_SUM2_SHIFT	4ULL
+
+#define DTE_NCO_SUM3_MASK	0xff
+#define DTE_NCO_SUM3_SHIFT	36ULL
+#define DTE_NCO_SUM3_WR_SHIFT	8
+
+#define DTE_NCO_TS_WRAP_MASK	0xfff
+#define DTE_NCO_TS_WRAP_LSHIFT	32
+
+#define DTE_NCO_INC_DEFAULT	0x80000000
+#define DTE_NUM_REGS_TO_RESTORE	4
+
+/* Full wrap around is 44bits in ns (~4.887 hrs) */
+#define DTE_WRAP_AROUND_NSEC_SHIFT 44
+
+/* 44 bits NCO */
+#define DTE_NCO_MAX_NS	0xFFFFFFFFFFFLL
+
+/* 125MHz with 3.29 reg cfg */
+#define DTE_PPB_ADJ(ppb) (u32)(div64_u64((((u64)abs(ppb) * BIT(28)) +\
+				      62500000ULL), 125000000ULL))
+
+/* ptp dte priv structure */
+struct ptp_dte {
+	void __iomem *regs;
+	struct ptp_clock *ptp_clk;
+	struct ptp_clock_info caps;
+	struct device *dev;
+	u32 ts_ovf_last;
+	u32 ts_wrap_cnt;
+	spinlock_t lock;
+	u32 reg_val[DTE_NUM_REGS_TO_RESTORE];
+};
+
+static void dte_write_nco(void __iomem *regs, s64 ns)
+{
+	u32 sum2, sum3;
+
+	sum2 = (u32)((ns >> DTE_NCO_SUM2_SHIFT) & DTE_NCO_SUM2_MASK);
+	/* compensate for ignoring sum1 */
+	if (sum2 != DTE_NCO_SUM2_MASK)
+		sum2++;
+
+	/* to write sum3, bits [15:8] needs to be written */
+	sum3 = (u32)(((ns >> DTE_NCO_SUM3_SHIFT) & DTE_NCO_SUM3_MASK) <<
+		     DTE_NCO_SUM3_WR_SHIFT);
+
+	writel(0, (regs + DTE_NCO_LOW_TIME_REG));
+	writel(sum2, (regs + DTE_NCO_TIME_REG));
+	writel(sum3, (regs + DTE_NCO_OVERFLOW_REG));
+}
+
+static s64 dte_read_nco(void __iomem *regs)
+{
+	u32 sum2, sum3;
+	s64 ns;
+
+	/*
+	 * ignoring sum1 (4 bits) gives a 16ns resolution, which
+	 * works due to the async register read.
+	 */
+	sum3 = readl(regs + DTE_NCO_OVERFLOW_REG) & DTE_NCO_SUM3_MASK;
+	sum2 = readl(regs + DTE_NCO_TIME_REG);
+	ns = ((s64)sum3 << DTE_NCO_SUM3_SHIFT) |
+		 ((s64)sum2 << DTE_NCO_SUM2_SHIFT);
+
+	return ns;
+}
+
+static void dte_write_nco_delta(struct ptp_dte *ptp_dte, s64 delta)
+{
+	s64 ns;
+
+	ns = dte_read_nco(ptp_dte->regs);
+
+	/* handle wraparound conditions */
+	if ((delta < 0) && (abs(delta) > ns)) {
+		if (ptp_dte->ts_wrap_cnt) {
+			ns += DTE_NCO_MAX_NS + delta;
+			ptp_dte->ts_wrap_cnt--;
+		} else {
+			ns = 0;
+		}
+	} else {
+		ns += delta;
+		if (ns > DTE_NCO_MAX_NS) {
+			ptp_dte->ts_wrap_cnt++;
+			ns -= DTE_NCO_MAX_NS;
+		}
+	}
+
+	dte_write_nco(ptp_dte->regs, ns);
+
+	ptp_dte->ts_ovf_last = (ns >> DTE_NCO_TS_WRAP_LSHIFT) &
+			DTE_NCO_TS_WRAP_MASK;
+}
+
+static s64 dte_read_nco_with_ovf(struct ptp_dte *ptp_dte)
+{
+	u32 ts_ovf;
+	s64 ns = 0;
+
+	ns = dte_read_nco(ptp_dte->regs);
+
+	/*Timestamp overflow: 8 LSB bits of sum3, 4 MSB bits of sum2 */
+	ts_ovf = (ns >> DTE_NCO_TS_WRAP_LSHIFT) & DTE_NCO_TS_WRAP_MASK;
+
+	/* Check for wrap around */
+	if (ts_ovf < ptp_dte->ts_ovf_last)
+		ptp_dte->ts_wrap_cnt++;
+
+	ptp_dte->ts_ovf_last = ts_ovf;
+
+	/* adjust for wraparounds */
+	ns += (s64)(BIT_ULL(DTE_WRAP_AROUND_NSEC_SHIFT) * ptp_dte->ts_wrap_cnt);
+
+	return ns;
+}
+
+static int ptp_dte_adjfine(struct ptp_clock_info *ptp, long scaled_ppm)
+{
+	s32 ppb = scaled_ppm_to_ppb(scaled_ppm);
+	u32 nco_incr;
+	unsigned long flags;
+	struct ptp_dte *ptp_dte = container_of(ptp, struct ptp_dte, caps);
+
+	if (abs(ppb) > ptp_dte->caps.max_adj) {
+		dev_err(ptp_dte->dev, "ppb adj too big\n");
+		return -EINVAL;
+	}
+
+	if (ppb < 0)
+		nco_incr = DTE_NCO_INC_DEFAULT - DTE_PPB_ADJ(ppb);
+	else
+		nco_incr = DTE_NCO_INC_DEFAULT + DTE_PPB_ADJ(ppb);
+
+	spin_lock_irqsave(&ptp_dte->lock, flags);
+	writel(nco_incr, ptp_dte->regs + DTE_NCO_INC_REG);
+	spin_unlock_irqrestore(&ptp_dte->lock, flags);
+
+	return 0;
+}
+
+static int ptp_dte_adjtime(struct ptp_clock_info *ptp, s64 delta)
+{
+	unsigned long flags;
+	struct ptp_dte *ptp_dte = container_of(ptp, struct ptp_dte, caps);
+
+	spin_lock_irqsave(&ptp_dte->lock, flags);
+	dte_write_nco_delta(ptp_dte, delta);
+	spin_unlock_irqrestore(&ptp_dte->lock, flags);
+
+	return 0;
+}
+
+static int ptp_dte_gettime(struct ptp_clock_info *ptp, struct timespec64 *ts)
+{
+	unsigned long flags;
+	struct ptp_dte *ptp_dte = container_of(ptp, struct ptp_dte, caps);
+
+	spin_lock_irqsave(&ptp_dte->lock, flags);
+	*ts = ns_to_timespec64(dte_read_nco_with_ovf(ptp_dte));
+	spin_unlock_irqrestore(&ptp_dte->lock, flags);
+
+	return 0;
+}
+
+static int ptp_dte_settime(struct ptp_clock_info *ptp,
+			     const struct timespec64 *ts)
+{
+	unsigned long flags;
+	struct ptp_dte *ptp_dte = container_of(ptp, struct ptp_dte, caps);
+
+	spin_lock_irqsave(&ptp_dte->lock, flags);
+
+	/* Disable nco increment */
+	writel(0, ptp_dte->regs + DTE_NCO_INC_REG);
+
+	dte_write_nco(ptp_dte->regs, timespec64_to_ns(ts));
+
+	/* reset overflow and wrap counter */
+	ptp_dte->ts_ovf_last = 0;
+	ptp_dte->ts_wrap_cnt = 0;
+
+	/* Enable nco increment */
+	writel(DTE_NCO_INC_DEFAULT, ptp_dte->regs + DTE_NCO_INC_REG);
+
+	spin_unlock_irqrestore(&ptp_dte->lock, flags);
+
+	return 0;
+}
+
+static int ptp_dte_enable(struct ptp_clock_info *ptp,
+			    struct ptp_clock_request *rq, int on)
+{
+	return -EOPNOTSUPP;
+}
+
+static const struct ptp_clock_info ptp_dte_caps = {
+	.owner		= THIS_MODULE,
+	.name		= "DTE PTP timer",
+	.max_adj	= 50000000,
+	.n_ext_ts	= 0,
+	.n_pins		= 0,
+	.pps		= 0,
+	.adjfine	= ptp_dte_adjfine,
+	.adjtime	= ptp_dte_adjtime,
+	.gettime64	= ptp_dte_gettime,
+	.settime64	= ptp_dte_settime,
+	.enable		= ptp_dte_enable,
+};
+
+static int ptp_dte_probe(struct platform_device *pdev)
+{
+	struct ptp_dte *ptp_dte;
+	struct device *dev = &pdev->dev;
+
+	ptp_dte = devm_kzalloc(dev, sizeof(struct ptp_dte), GFP_KERNEL);
+	if (!ptp_dte)
+		return -ENOMEM;
+
+	ptp_dte->regs = devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(ptp_dte->regs))
+		return PTR_ERR(ptp_dte->regs);
+
+	spin_lock_init(&ptp_dte->lock);
+
+	ptp_dte->dev = dev;
+	ptp_dte->caps = ptp_dte_caps;
+	ptp_dte->ptp_clk = ptp_clock_register(&ptp_dte->caps, &pdev->dev);
+	if (IS_ERR(ptp_dte->ptp_clk)) {
+		dev_err(dev,
+			"%s: Failed to register ptp clock\n", __func__);
+		return PTR_ERR(ptp_dte->ptp_clk);
+	}
+
+	platform_set_drvdata(pdev, ptp_dte);
+
+	dev_info(dev, "ptp clk probe done\n");
+
+	return 0;
+}
+
+static void ptp_dte_remove(struct platform_device *pdev)
+{
+	struct ptp_dte *ptp_dte = platform_get_drvdata(pdev);
+	u8 i;
+
+	ptp_clock_unregister(ptp_dte->ptp_clk);
+
+	for (i = 0; i < DTE_NUM_REGS_TO_RESTORE; i++)
+		writel(0, ptp_dte->regs + (i * sizeof(u32)));
+}
+
+#ifdef CONFIG_PM_SLEEP
+static int ptp_dte_suspend(struct device *dev)
+{
+	struct ptp_dte *ptp_dte = dev_get_drvdata(dev);
+	u8 i;
+
+	for (i = 0; i < DTE_NUM_REGS_TO_RESTORE; i++) {
+		ptp_dte->reg_val[i] =
+			readl(ptp_dte->regs + (i * sizeof(u32)));
+	}
+
+	/* disable the nco */
+	writel(0, ptp_dte->regs + DTE_NCO_INC_REG);
+
+	return 0;
+}
+
+static int ptp_dte_resume(struct device *dev)
+{
+	struct ptp_dte *ptp_dte = dev_get_drvdata(dev);
+	u8 i;
+
+	for (i = 0; i < DTE_NUM_REGS_TO_RESTORE; i++) {
+		if ((i * sizeof(u32)) != DTE_NCO_OVERFLOW_REG)
+			writel(ptp_dte->reg_val[i],
+				(ptp_dte->regs + (i * sizeof(u32))));
+		else
+			writel(((ptp_dte->reg_val[i] &
+				DTE_NCO_SUM3_MASK) << DTE_NCO_SUM3_WR_SHIFT),
+				(ptp_dte->regs + (i * sizeof(u32))));
+	}
+
+	return 0;
+}
+
+static const struct dev_pm_ops ptp_dte_pm_ops = {
+	.suspend = ptp_dte_suspend,
+	.resume = ptp_dte_resume
+};
+
+#define PTP_DTE_PM_OPS	(&ptp_dte_pm_ops)
+#else
+#define PTP_DTE_PM_OPS	NULL
+#endif
+
+static const struct of_device_id ptp_dte_of_match[] = {
+	{ .compatible = "brcm,ptp-dte", },
+	{},
+};
+MODULE_DEVICE_TABLE(of, ptp_dte_of_match);
+
+static struct platform_driver ptp_dte_driver = {
+	.driver = {
+		.name = "ptp-dte",
+		.pm = PTP_DTE_PM_OPS,
+		.of_match_table = ptp_dte_of_match,
+	},
+	.probe    = ptp_dte_probe,
+	.remove   = ptp_dte_remove,
+};
+module_platform_driver(ptp_dte_driver);
+
+MODULE_AUTHOR("Broadcom");
+MODULE_DESCRIPTION("Broadcom DTE PTP Clock driver");
+MODULE_LICENSE("GPL v2");