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, 272 insertions, 0 deletions

diff --git a/drivers/ata/pata_cmd640.c b/drivers/ata/pata_cmd640.c
new file mode 100644
index 000000000..84eb36d02
--- /dev/null
+++ b/drivers/ata/pata_cmd640.c
@@ -0,0 +1,272 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * pata_cmd640.c 	- CMD640 PCI PATA for new ATA layer
+ *			  (C) 2007 Red Hat Inc
+ *
+ * Based upon
+ *  linux/drivers/ide/pci/cmd640.c		Version 1.02  Sep 01, 1996
+ *
+ *  Copyright (C) 1995-1996  Linus Torvalds & authors (see driver)
+ *
+ *	This drives only the PCI version of the controller. If you have a
+ *	VLB one then we have enough docs to support it but you can write
+ *	your own code.
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/blkdev.h>
+#include <linux/delay.h>
+#include <linux/gfp.h>
+#include <scsi/scsi_host.h>
+#include <linux/libata.h>
+
+#define DRV_NAME "pata_cmd640"
+#define DRV_VERSION "0.0.5"
+
+struct cmd640_reg {
+	int last;
+	u8 reg58[ATA_MAX_DEVICES];
+};
+
+enum {
+	CFR = 0x50,
+	CNTRL = 0x51,
+	CMDTIM = 0x52,
+	ARTIM0 = 0x53,
+	DRWTIM0 = 0x54,
+	ARTIM23 = 0x57,
+	DRWTIM23 = 0x58,
+	BRST = 0x59
+};
+
+/**
+ *	cmd640_set_piomode	-	set initial PIO mode data
+ *	@ap: ATA port
+ *	@adev: ATA device
+ *
+ *	Called to do the PIO mode setup.
+ */
+
+static void cmd640_set_piomode(struct ata_port *ap, struct ata_device *adev)
+{
+	struct cmd640_reg *timing = ap->private_data;
+	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
+	struct ata_timing t;
+	const unsigned long T = 1000000 / 33;
+	const u8 setup_data[] = { 0x40, 0x40, 0x40, 0x80, 0x00 };
+	u8 reg;
+	int arttim = ARTIM0 + 2 * adev->devno;
+	struct ata_device *pair = ata_dev_pair(adev);
+
+	if (ata_timing_compute(adev, adev->pio_mode, &t, T, 0) < 0) {
+		ata_dev_err(adev, DRV_NAME ": mode computation failed.\n");
+		return;
+	}
+
+	/* The second channel has shared timings and the setup timing is
+	   messy to switch to merge it for worst case */
+	if (ap->port_no && pair) {
+		struct ata_timing p;
+		ata_timing_compute(pair, pair->pio_mode, &p, T, 1);
+		ata_timing_merge(&p, &t, &t, ATA_TIMING_SETUP);
+	}
+
+	/* Make the timings fit */
+	if (t.recover > 16) {
+		t.active += t.recover - 16;
+		t.recover = 16;
+	}
+	if (t.active > 16)
+		t.active = 16;
+
+	/* Now convert the clocks into values we can actually stuff into
+	   the chip */
+
+	if (t.recover > 1)
+		t.recover--;	/* 640B only */
+	else
+		t.recover = 15;
+
+	if (t.setup > 4)
+		t.setup = 0xC0;
+	else
+		t.setup = setup_data[t.setup];
+
+	if (ap->port_no == 0) {
+		t.active &= 0x0F;	/* 0 = 16 */
+
+		/* Load setup timing */
+		pci_read_config_byte(pdev, arttim, &reg);
+		reg &= 0x3F;
+		reg |= t.setup;
+		pci_write_config_byte(pdev, arttim, reg);
+
+		/* Load active/recovery */
+		pci_write_config_byte(pdev, arttim + 1, (t.active << 4) | t.recover);
+	} else {
+		/* Save the shared timings for channel, they will be loaded
+		   by qc_issue. Reloading the setup time is expensive so we
+		   keep a merged one loaded */
+		pci_read_config_byte(pdev, ARTIM23, &reg);
+		reg &= 0x3F;
+		reg |= t.setup;
+		pci_write_config_byte(pdev, ARTIM23, reg);
+		timing->reg58[adev->devno] = (t.active << 4) | t.recover;
+	}
+}
+
+
+/**
+ *	cmd640_qc_issue	-	command preparation hook
+ *	@qc: Command to be issued
+ *
+ *	Channel 1 has shared timings. We must reprogram the
+ *	clock each drive 2/3 switch we do.
+ */
+
+static unsigned int cmd640_qc_issue(struct ata_queued_cmd *qc)
+{
+	struct ata_port *ap = qc->ap;
+	struct ata_device *adev = qc->dev;
+	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
+	struct cmd640_reg *timing = ap->private_data;
+
+	if (ap->port_no != 0 && adev->devno != timing->last) {
+		pci_write_config_byte(pdev, DRWTIM23, timing->reg58[adev->devno]);
+		timing->last = adev->devno;
+	}
+	return ata_sff_qc_issue(qc);
+}
+
+/**
+ *	cmd640_port_start	-	port setup
+ *	@ap: ATA port being set up
+ *
+ *	The CMD640 needs to maintain private data structures so we
+ *	allocate space here.
+ */
+
+static int cmd640_port_start(struct ata_port *ap)
+{
+	struct pci_dev *pdev = to_pci_dev(ap->host->dev);
+	struct cmd640_reg *timing;
+
+	timing = devm_kzalloc(&pdev->dev, sizeof(struct cmd640_reg), GFP_KERNEL);
+	if (timing == NULL)
+		return -ENOMEM;
+	timing->last = -1;	/* Force a load */
+	ap->private_data = timing;
+	return 0;
+}
+
+static bool cmd640_sff_irq_check(struct ata_port *ap)
+{
+	struct pci_dev *pdev	= to_pci_dev(ap->host->dev);
+	int irq_reg		= ap->port_no ? ARTIM23 : CFR;
+	u8  irq_stat, irq_mask	= ap->port_no ? 0x10 : 0x04;
+
+	pci_read_config_byte(pdev, irq_reg, &irq_stat);
+
+	return irq_stat & irq_mask;
+}
+
+static const struct scsi_host_template cmd640_sht = {
+	ATA_PIO_SHT(DRV_NAME),
+};
+
+static struct ata_port_operations cmd640_port_ops = {
+	.inherits	= &ata_sff_port_ops,
+	/* In theory xfer_noirq is not needed once we kill the prefetcher */
+	.sff_data_xfer	= ata_sff_data_xfer32,
+	.sff_irq_check	= cmd640_sff_irq_check,
+	.qc_issue	= cmd640_qc_issue,
+	.cable_detect	= ata_cable_40wire,
+	.set_piomode	= cmd640_set_piomode,
+	.port_start	= cmd640_port_start,
+};
+
+static void cmd640_hardware_init(struct pci_dev *pdev)
+{
+	u8 ctrl;
+
+	/* CMD640 detected, commiserations */
+	pci_write_config_byte(pdev, 0x5B, 0x00);
+	/* PIO0 command cycles */
+	pci_write_config_byte(pdev, CMDTIM, 0);
+	/* 512 byte bursts (sector) */
+	pci_write_config_byte(pdev, BRST, 0x40);
+	/*
+	 * A reporter a long time ago
+	 * Had problems with the data fifo
+	 * So don't run the risk
+	 * Of putting crap on the disk
+	 * For its better just to go slow
+	 */
+	/* Do channel 0 */
+	pci_read_config_byte(pdev, CNTRL, &ctrl);
+	pci_write_config_byte(pdev, CNTRL, ctrl | 0xC0);
+	/* Ditto for channel 1 */
+	pci_read_config_byte(pdev, ARTIM23, &ctrl);
+	ctrl |= 0x0C;
+	pci_write_config_byte(pdev, ARTIM23, ctrl);
+}
+
+static int cmd640_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+	static const struct ata_port_info info = {
+		.flags = ATA_FLAG_SLAVE_POSS,
+		.pio_mask = ATA_PIO4,
+		.port_ops = &cmd640_port_ops
+	};
+	const struct ata_port_info *ppi[] = { &info, NULL };
+	int rc;
+
+	rc = pcim_enable_device(pdev);
+	if (rc)
+		return rc;
+
+	cmd640_hardware_init(pdev);
+
+	return ata_pci_sff_init_one(pdev, ppi, &cmd640_sht, NULL, 0);
+}
+
+#ifdef CONFIG_PM_SLEEP
+static int cmd640_reinit_one(struct pci_dev *pdev)
+{
+	struct ata_host *host = pci_get_drvdata(pdev);
+	int rc;
+
+	rc = ata_pci_device_do_resume(pdev);
+	if (rc)
+		return rc;
+	cmd640_hardware_init(pdev);
+	ata_host_resume(host);
+	return 0;
+}
+#endif
+
+static const struct pci_device_id cmd640[] = {
+	{ PCI_VDEVICE(CMD, 0x0640) },
+	{ }
+};
+
+static struct pci_driver cmd640_pci_driver = {
+	.name 		= DRV_NAME,
+	.id_table	= cmd640,
+	.probe 		= cmd640_init_one,
+	.remove		= ata_pci_remove_one,
+#ifdef CONFIG_PM_SLEEP
+	.suspend	= ata_pci_device_suspend,
+	.resume		= cmd640_reinit_one,
+#endif
+};
+
+module_pci_driver(cmd640_pci_driver);
+
+MODULE_AUTHOR("Alan Cox");
+MODULE_DESCRIPTION("low-level driver for CMD640 PATA controllers");
+MODULE_LICENSE("GPL");
+MODULE_DEVICE_TABLE(pci, cmd640);
+MODULE_VERSION(DRV_VERSION);