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, 378 insertions, 0 deletions

diff --git a/drivers/bus/qcom-ebi2.c b/drivers/bus/qcom-ebi2.c
new file mode 100644
index 000000000..8d2eb955d
--- /dev/null
+++ b/drivers/bus/qcom-ebi2.c
@@ -0,0 +1,378 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Qualcomm External Bus Interface 2 (EBI2) driver
+ * an older version of the Qualcomm Parallel Interface Controller (QPIC)
+ *
+ * Copyright (C) 2016 Linaro Ltd.
+ *
+ * Author: Linus Walleij <linus.walleij@linaro.org>
+ *
+ * See the device tree bindings for this block for more details on the
+ * hardware.
+ */
+
+#include <linux/module.h>
+#include <linux/clk.h>
+#include <linux/err.h>
+#include <linux/io.h>
+#include <linux/of.h>
+#include <linux/of_platform.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/platform_device.h>
+#include <linux/bitops.h>
+
+/*
+ * CS0, CS1, CS4 and CS5 are two bits wide, CS2 and CS3 are one bit.
+ */
+#define EBI2_CS0_ENABLE_MASK BIT(0)|BIT(1)
+#define EBI2_CS1_ENABLE_MASK BIT(2)|BIT(3)
+#define EBI2_CS2_ENABLE_MASK BIT(4)
+#define EBI2_CS3_ENABLE_MASK BIT(5)
+#define EBI2_CS4_ENABLE_MASK BIT(6)|BIT(7)
+#define EBI2_CS5_ENABLE_MASK BIT(8)|BIT(9)
+#define EBI2_CSN_MASK GENMASK(9, 0)
+
+#define EBI2_XMEM_CFG 0x0000 /* Power management etc */
+
+/*
+ * SLOW CSn CFG
+ *
+ * Bits 31-28: RECOVERY recovery cycles (0 = 1, 1 = 2 etc) this is the time the
+ *             memory continues to drive the data bus after OE is de-asserted.
+ *             Inserted when reading one CS and switching to another CS or read
+ *             followed by write on the same CS. Valid values 0 thru 15.
+ * Bits 27-24: WR_HOLD write hold cycles, these are extra cycles inserted after
+ *             every write minimum 1. The data out is driven from the time WE is
+ *             asserted until CS is asserted. With a hold of 1, the CS stays
+ *             active for 1 extra cycle etc. Valid values 0 thru 15.
+ * Bits 23-16: WR_DELTA initial latency for write cycles inserted for the first
+ *             write to a page or burst memory
+ * Bits 15-8:  RD_DELTA initial latency for read cycles inserted for the first
+ *             read to a page or burst memory
+ * Bits 7-4:   WR_WAIT number of wait cycles for every write access, 0=1 cycle
+ *             so 1 thru 16 cycles.
+ * Bits 3-0:   RD_WAIT number of wait cycles for every read access, 0=1 cycle
+ *             so 1 thru 16 cycles.
+ */
+#define EBI2_XMEM_CS0_SLOW_CFG 0x0008
+#define EBI2_XMEM_CS1_SLOW_CFG 0x000C
+#define EBI2_XMEM_CS2_SLOW_CFG 0x0010
+#define EBI2_XMEM_CS3_SLOW_CFG 0x0014
+#define EBI2_XMEM_CS4_SLOW_CFG 0x0018
+#define EBI2_XMEM_CS5_SLOW_CFG 0x001C
+
+#define EBI2_XMEM_RECOVERY_SHIFT	28
+#define EBI2_XMEM_WR_HOLD_SHIFT		24
+#define EBI2_XMEM_WR_DELTA_SHIFT	16
+#define EBI2_XMEM_RD_DELTA_SHIFT	8
+#define EBI2_XMEM_WR_WAIT_SHIFT		4
+#define EBI2_XMEM_RD_WAIT_SHIFT		0
+
+/*
+ * FAST CSn CFG
+ * Bits 31-28: ?
+ * Bits 27-24: RD_HOLD: the length in cycles of the first segment of a read
+ *             transfer. For a single read trandfer this will be the time
+ *             from CS assertion to OE assertion.
+ * Bits 18-24: ?
+ * Bits 17-16: ADV_OE_RECOVERY, the number of cycles elapsed before an OE
+ *             assertion, with respect to the cycle where ADV is asserted.
+ *             2 means 2 cycles between ADV and OE. Values 0, 1, 2 or 3.
+ * Bits 5:     ADDR_HOLD_ENA, The address is held for an extra cycle to meet
+ *             hold time requirements with ADV assertion.
+ *
+ * The manual mentions "write precharge cycles" and "precharge cycles".
+ * We have not been able to figure out which bit fields these correspond to
+ * in the hardware, or what valid values exist. The current hypothesis is that
+ * this is something just used on the FAST chip selects. There is also a "byte
+ * device enable" flag somewhere for 8bit memories.
+ */
+#define EBI2_XMEM_CS0_FAST_CFG 0x0028
+#define EBI2_XMEM_CS1_FAST_CFG 0x002C
+#define EBI2_XMEM_CS2_FAST_CFG 0x0030
+#define EBI2_XMEM_CS3_FAST_CFG 0x0034
+#define EBI2_XMEM_CS4_FAST_CFG 0x0038
+#define EBI2_XMEM_CS5_FAST_CFG 0x003C
+
+#define EBI2_XMEM_RD_HOLD_SHIFT		24
+#define EBI2_XMEM_ADV_OE_RECOVERY_SHIFT	16
+#define EBI2_XMEM_ADDR_HOLD_ENA_SHIFT	5
+
+/**
+ * struct cs_data - struct with info on a chipselect setting
+ * @enable_mask: mask to enable the chipselect in the EBI2 config
+ * @slow_cfg: offset to XMEMC slow CS config
+ * @fast_cfg: offset to XMEMC fast CS config
+ */
+struct cs_data {
+	u32 enable_mask;
+	u16 slow_cfg;
+	u16 fast_cfg;
+};
+
+static const struct cs_data cs_info[] = {
+	{
+		/* CS0 */
+		.enable_mask = EBI2_CS0_ENABLE_MASK,
+		.slow_cfg = EBI2_XMEM_CS0_SLOW_CFG,
+		.fast_cfg = EBI2_XMEM_CS0_FAST_CFG,
+	},
+	{
+		/* CS1 */
+		.enable_mask = EBI2_CS1_ENABLE_MASK,
+		.slow_cfg = EBI2_XMEM_CS1_SLOW_CFG,
+		.fast_cfg = EBI2_XMEM_CS1_FAST_CFG,
+	},
+	{
+		/* CS2 */
+		.enable_mask = EBI2_CS2_ENABLE_MASK,
+		.slow_cfg = EBI2_XMEM_CS2_SLOW_CFG,
+		.fast_cfg = EBI2_XMEM_CS2_FAST_CFG,
+	},
+	{
+		/* CS3 */
+		.enable_mask = EBI2_CS3_ENABLE_MASK,
+		.slow_cfg = EBI2_XMEM_CS3_SLOW_CFG,
+		.fast_cfg = EBI2_XMEM_CS3_FAST_CFG,
+	},
+	{
+		/* CS4 */
+		.enable_mask = EBI2_CS4_ENABLE_MASK,
+		.slow_cfg = EBI2_XMEM_CS4_SLOW_CFG,
+		.fast_cfg = EBI2_XMEM_CS4_FAST_CFG,
+	},
+	{
+		/* CS5 */
+		.enable_mask = EBI2_CS5_ENABLE_MASK,
+		.slow_cfg = EBI2_XMEM_CS5_SLOW_CFG,
+		.fast_cfg = EBI2_XMEM_CS5_FAST_CFG,
+	},
+};
+
+/**
+ * struct ebi2_xmem_prop - describes an XMEM config property
+ * @prop: the device tree binding name
+ * @max: maximum value for the property
+ * @slowreg: true if this property is in the SLOW CS config register
+ * else it is assumed to be in the FAST config register
+ * @shift: the bit field start in the SLOW or FAST register for this
+ * property
+ */
+struct ebi2_xmem_prop {
+	const char *prop;
+	u32 max;
+	bool slowreg;
+	u16 shift;
+};
+
+static const struct ebi2_xmem_prop xmem_props[] = {
+	{
+		.prop = "qcom,xmem-recovery-cycles",
+		.max = 15,
+		.slowreg = true,
+		.shift = EBI2_XMEM_RECOVERY_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-write-hold-cycles",
+		.max = 15,
+		.slowreg = true,
+		.shift = EBI2_XMEM_WR_HOLD_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-write-delta-cycles",
+		.max = 255,
+		.slowreg = true,
+		.shift = EBI2_XMEM_WR_DELTA_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-read-delta-cycles",
+		.max = 255,
+		.slowreg = true,
+		.shift = EBI2_XMEM_RD_DELTA_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-write-wait-cycles",
+		.max = 15,
+		.slowreg = true,
+		.shift = EBI2_XMEM_WR_WAIT_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-read-wait-cycles",
+		.max = 15,
+		.slowreg = true,
+		.shift = EBI2_XMEM_RD_WAIT_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-address-hold-enable",
+		.max = 1, /* boolean prop */
+		.slowreg = false,
+		.shift = EBI2_XMEM_ADDR_HOLD_ENA_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-adv-to-oe-recovery-cycles",
+		.max = 3,
+		.slowreg = false,
+		.shift = EBI2_XMEM_ADV_OE_RECOVERY_SHIFT,
+	},
+	{
+		.prop = "qcom,xmem-read-hold-cycles",
+		.max = 15,
+		.slowreg = false,
+		.shift = EBI2_XMEM_RD_HOLD_SHIFT,
+	},
+};
+
+static void qcom_ebi2_setup_chipselect(struct device_node *np,
+				       struct device *dev,
+				       void __iomem *ebi2_base,
+				       void __iomem *ebi2_xmem,
+				       u32 csindex)
+{
+	const struct cs_data *csd;
+	u32 slowcfg, fastcfg;
+	u32 val;
+	int ret;
+	int i;
+
+	csd = &cs_info[csindex];
+	val = readl(ebi2_base);
+	val |= csd->enable_mask;
+	writel(val, ebi2_base);
+	dev_dbg(dev, "enabled CS%u\n", csindex);
+
+	/* Next set up the XMEMC */
+	slowcfg = 0;
+	fastcfg = 0;
+
+	for (i = 0; i < ARRAY_SIZE(xmem_props); i++) {
+		const struct ebi2_xmem_prop *xp = &xmem_props[i];
+
+		/* All are regular u32 values */
+		ret = of_property_read_u32(np, xp->prop, &val);
+		if (ret) {
+			dev_dbg(dev, "could not read %s for CS%d\n",
+				xp->prop, csindex);
+			continue;
+		}
+
+		/* First check boolean props */
+		if (xp->max == 1 && val) {
+			if (xp->slowreg)
+				slowcfg |= BIT(xp->shift);
+			else
+				fastcfg |= BIT(xp->shift);
+			dev_dbg(dev, "set %s flag\n", xp->prop);
+			continue;
+		}
+
+		/* We're dealing with an u32 */
+		if (val > xp->max) {
+			dev_err(dev,
+				"too high value for %s: %u, capped at %u\n",
+				xp->prop, val, xp->max);
+			val = xp->max;
+		}
+		if (xp->slowreg)
+			slowcfg |= (val << xp->shift);
+		else
+			fastcfg |= (val << xp->shift);
+		dev_dbg(dev, "set %s to %u\n", xp->prop, val);
+	}
+
+	dev_info(dev, "CS%u: SLOW CFG 0x%08x, FAST CFG 0x%08x\n",
+		 csindex, slowcfg, fastcfg);
+
+	if (slowcfg)
+		writel(slowcfg, ebi2_xmem + csd->slow_cfg);
+	if (fastcfg)
+		writel(fastcfg, ebi2_xmem + csd->fast_cfg);
+}
+
+static int qcom_ebi2_probe(struct platform_device *pdev)
+{
+	struct device_node *np = pdev->dev.of_node;
+	struct device *dev = &pdev->dev;
+	struct resource *res;
+	void __iomem *ebi2_base;
+	void __iomem *ebi2_xmem;
+	struct clk *ebi2xclk;
+	struct clk *ebi2clk;
+	bool have_children = false;
+	u32 val;
+	int ret;
+
+	ebi2xclk = devm_clk_get_enabled(dev, "ebi2x");
+	if (IS_ERR(ebi2xclk))
+		return PTR_ERR(ebi2xclk);
+
+	ebi2clk = devm_clk_get_enabled(dev, "ebi2");
+	if (IS_ERR(ebi2clk))
+		return PTR_ERR(ebi2clk);
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	ebi2_base = devm_ioremap_resource(dev, res);
+	if (IS_ERR(ebi2_base))
+		return PTR_ERR(ebi2_base);
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
+	ebi2_xmem = devm_ioremap_resource(dev, res);
+	if (IS_ERR(ebi2_xmem))
+		return PTR_ERR(ebi2_xmem);
+
+	/* Allegedly this turns the power save mode off */
+	writel(0UL, ebi2_xmem + EBI2_XMEM_CFG);
+
+	/* Disable all chipselects */
+	val = readl(ebi2_base);
+	val &= ~EBI2_CSN_MASK;
+	writel(val, ebi2_base);
+
+	/* Walk over the child nodes and see what chipselects we use */
+	for_each_available_child_of_node_scoped(np, child) {
+		u32 csindex;
+
+		/* Figure out the chipselect */
+		ret = of_property_read_u32(child, "reg", &csindex);
+		if (ret)
+			return ret;
+
+		if (csindex > 5) {
+			dev_err(dev,
+				"invalid chipselect %u, we only support 0-5\n",
+				csindex);
+			continue;
+		}
+
+		qcom_ebi2_setup_chipselect(child,
+					   dev,
+					   ebi2_base,
+					   ebi2_xmem,
+					   csindex);
+
+		/* We have at least one child */
+		have_children = true;
+	}
+
+	if (have_children)
+		return devm_of_platform_populate(dev);
+
+	return 0;
+}
+
+static const struct of_device_id qcom_ebi2_of_match[] = {
+	{ .compatible = "qcom,msm8660-ebi2", },
+	{ .compatible = "qcom,apq8060-ebi2", },
+	{ }
+};
+
+static struct platform_driver qcom_ebi2_driver = {
+	.probe = qcom_ebi2_probe,
+	.driver = {
+		.name = "qcom-ebi2",
+		.of_match_table = qcom_ebi2_of_match,
+	},
+};
+module_platform_driver(qcom_ebi2_driver);
+MODULE_AUTHOR("Linus Walleij <linus.walleij@linaro.org>");
+MODULE_DESCRIPTION("Qualcomm EBI2 driver");