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, 296 insertions, 0 deletions

diff --git a/drivers/ata/pata_pxa.c b/drivers/ata/pata_pxa.c
new file mode 100644
index 000000000..a694f6a17
--- /dev/null
+++ b/drivers/ata/pata_pxa.c
@@ -0,0 +1,296 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Generic PXA PATA driver
+ *
+ * Copyright (C) 2010 Marek Vasut <marek.vasut@gmail.com>
+ */
+
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/blkdev.h>
+#include <linux/ata.h>
+#include <linux/libata.h>
+#include <linux/platform_device.h>
+#include <linux/dmaengine.h>
+#include <linux/slab.h>
+#include <linux/completion.h>
+
+#include <scsi/scsi_host.h>
+
+#include <linux/platform_data/ata-pxa.h>
+
+#define DRV_NAME	"pata_pxa"
+#define DRV_VERSION	"0.1"
+
+struct pata_pxa_data {
+	struct dma_chan		*dma_chan;
+	dma_cookie_t		dma_cookie;
+	struct completion	dma_done;
+};
+
+/*
+ * DMA interrupt handler.
+ */
+static void pxa_ata_dma_irq(void *d)
+{
+	struct pata_pxa_data *pd = d;
+	enum dma_status status;
+
+	status = dmaengine_tx_status(pd->dma_chan, pd->dma_cookie, NULL);
+	if (status == DMA_ERROR || status == DMA_COMPLETE)
+		complete(&pd->dma_done);
+}
+
+/*
+ * Prepare taskfile for submission.
+ */
+static enum ata_completion_errors pxa_qc_prep(struct ata_queued_cmd *qc)
+{
+	struct pata_pxa_data *pd = qc->ap->private_data;
+	struct dma_async_tx_descriptor *tx;
+	enum dma_transfer_direction dir;
+
+	if (!(qc->flags & ATA_QCFLAG_DMAMAP))
+		return AC_ERR_OK;
+
+	dir = (qc->dma_dir == DMA_TO_DEVICE ? DMA_MEM_TO_DEV : DMA_DEV_TO_MEM);
+	tx = dmaengine_prep_slave_sg(pd->dma_chan, qc->sg, qc->n_elem, dir,
+				     DMA_PREP_INTERRUPT);
+	if (!tx) {
+		ata_dev_err(qc->dev, "prep_slave_sg() failed\n");
+		return AC_ERR_OK;
+	}
+	tx->callback = pxa_ata_dma_irq;
+	tx->callback_param = pd;
+	pd->dma_cookie = dmaengine_submit(tx);
+
+	return AC_ERR_OK;
+}
+
+/*
+ * Configure the DMA controller, load the DMA descriptors, but don't start the
+ * DMA controller yet. Only issue the ATA command.
+ */
+static void pxa_bmdma_setup(struct ata_queued_cmd *qc)
+{
+	qc->ap->ops->sff_exec_command(qc->ap, &qc->tf);
+}
+
+/*
+ * Execute the DMA transfer.
+ */
+static void pxa_bmdma_start(struct ata_queued_cmd *qc)
+{
+	struct pata_pxa_data *pd = qc->ap->private_data;
+	init_completion(&pd->dma_done);
+	dma_async_issue_pending(pd->dma_chan);
+}
+
+/*
+ * Wait until the DMA transfer completes, then stop the DMA controller.
+ */
+static void pxa_bmdma_stop(struct ata_queued_cmd *qc)
+{
+	struct pata_pxa_data *pd = qc->ap->private_data;
+	enum dma_status status;
+
+	status = dmaengine_tx_status(pd->dma_chan, pd->dma_cookie, NULL);
+	if (status != DMA_ERROR && status != DMA_COMPLETE &&
+	    wait_for_completion_timeout(&pd->dma_done, HZ))
+		ata_dev_err(qc->dev, "Timeout waiting for DMA completion!");
+
+	dmaengine_terminate_all(pd->dma_chan);
+}
+
+/*
+ * Read DMA status. The bmdma_stop() will take care of properly finishing the
+ * DMA transfer so we always have DMA-complete interrupt here.
+ */
+static unsigned char pxa_bmdma_status(struct ata_port *ap)
+{
+	struct pata_pxa_data *pd = ap->private_data;
+	unsigned char ret = ATA_DMA_INTR;
+	struct dma_tx_state state;
+	enum dma_status status;
+
+	status = dmaengine_tx_status(pd->dma_chan, pd->dma_cookie, &state);
+	if (status != DMA_COMPLETE)
+		ret |= ATA_DMA_ERR;
+
+	return ret;
+}
+
+/*
+ * No IRQ register present so we do nothing.
+ */
+static void pxa_irq_clear(struct ata_port *ap)
+{
+}
+
+/*
+ * Check for ATAPI DMA. ATAPI DMA is unsupported by this driver. It's still
+ * unclear why ATAPI has DMA issues.
+ */
+static int pxa_check_atapi_dma(struct ata_queued_cmd *qc)
+{
+	return -EOPNOTSUPP;
+}
+
+static const struct scsi_host_template pxa_ata_sht = {
+	ATA_BMDMA_SHT(DRV_NAME),
+};
+
+static struct ata_port_operations pxa_ata_port_ops = {
+	.inherits		= &ata_bmdma_port_ops,
+	.cable_detect		= ata_cable_40wire,
+
+	.bmdma_setup		= pxa_bmdma_setup,
+	.bmdma_start		= pxa_bmdma_start,
+	.bmdma_stop		= pxa_bmdma_stop,
+	.bmdma_status		= pxa_bmdma_status,
+
+	.check_atapi_dma	= pxa_check_atapi_dma,
+
+	.sff_irq_clear		= pxa_irq_clear,
+
+	.qc_prep		= pxa_qc_prep,
+};
+
+static int pxa_ata_probe(struct platform_device *pdev)
+{
+	struct ata_host *host;
+	struct ata_port *ap;
+	struct pata_pxa_data *data;
+	struct resource *dma_res;
+	struct pata_pxa_pdata *pdata = dev_get_platdata(&pdev->dev);
+	struct dma_slave_config	config;
+	int ret = 0;
+	int irq;
+
+	/*
+	 * Resource validation, three resources are needed:
+	 *  - CMD port base address
+	 *  - CTL port base address
+	 *  - DMA port base address
+	 *  - IRQ pin
+	 */
+	if (pdev->num_resources != 4) {
+		dev_err(&pdev->dev, "invalid number of resources\n");
+		return -EINVAL;
+	}
+
+	/*
+	 * DMA port base address
+	 */
+	dma_res = platform_get_resource(pdev, IORESOURCE_DMA, 0);
+	if (unlikely(dma_res == NULL))
+		return -EINVAL;
+
+	/*
+	 * IRQ pin
+	 */
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0)
+		return irq;
+
+	/*
+	 * Allocate the host
+	 */
+	host = ata_host_alloc(&pdev->dev, 1);
+	if (!host)
+		return -ENOMEM;
+
+	ap		= host->ports[0];
+	ap->ops		= &pxa_ata_port_ops;
+	ap->pio_mask	= ATA_PIO4;
+	ap->mwdma_mask	= ATA_MWDMA2;
+
+	ap->ioaddr.cmd_addr	= devm_platform_ioremap_resource(pdev, 0);
+	if (IS_ERR(ap->ioaddr.cmd_addr))
+		return PTR_ERR(ap->ioaddr.cmd_addr);
+	ap->ioaddr.ctl_addr	= devm_platform_ioremap_resource(pdev, 1);
+	if (IS_ERR(ap->ioaddr.ctl_addr))
+		return PTR_ERR(ap->ioaddr.ctl_addr);
+	ap->ioaddr.bmdma_addr	= devm_ioremap(&pdev->dev, dma_res->start,
+						resource_size(dma_res));
+	if (!ap->ioaddr.bmdma_addr)
+		return -ENOMEM;
+
+	/*
+	 * Adjust register offsets
+	 */
+	ap->ioaddr.altstatus_addr = ap->ioaddr.ctl_addr;
+	ap->ioaddr.data_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_DATA << pdata->reg_shift);
+	ap->ioaddr.error_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_ERR << pdata->reg_shift);
+	ap->ioaddr.feature_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_FEATURE << pdata->reg_shift);
+	ap->ioaddr.nsect_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_NSECT << pdata->reg_shift);
+	ap->ioaddr.lbal_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_LBAL << pdata->reg_shift);
+	ap->ioaddr.lbam_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_LBAM << pdata->reg_shift);
+	ap->ioaddr.lbah_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_LBAH << pdata->reg_shift);
+	ap->ioaddr.device_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_DEVICE << pdata->reg_shift);
+	ap->ioaddr.status_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_STATUS << pdata->reg_shift);
+	ap->ioaddr.command_addr	= ap->ioaddr.cmd_addr +
+					(ATA_REG_CMD << pdata->reg_shift);
+
+	/*
+	 * Allocate and load driver's internal data structure
+	 */
+	data = devm_kzalloc(&pdev->dev, sizeof(struct pata_pxa_data),
+								GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	ap->private_data = data;
+
+	memset(&config, 0, sizeof(config));
+	config.src_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
+	config.dst_addr_width = DMA_SLAVE_BUSWIDTH_2_BYTES;
+	config.src_addr = dma_res->start;
+	config.dst_addr = dma_res->start;
+	config.src_maxburst = 32;
+	config.dst_maxburst = 32;
+
+	/*
+	 * Request the DMA channel
+	 */
+	data->dma_chan = devm_dma_request_chan(&pdev->dev, "data");
+	if (IS_ERR(data->dma_chan))
+		return PTR_ERR(data->dma_chan);
+
+	ret = dmaengine_slave_config(data->dma_chan, &config);
+	if (ret < 0) {
+		dev_err(&pdev->dev, "dma configuration failed: %d\n", ret);
+		return ret;
+	}
+
+	/*
+	 * Activate the ATA host
+	 */
+	return ata_host_activate(host, irq, ata_sff_interrupt,
+				pdata->irq_flags, &pxa_ata_sht);
+}
+
+static struct platform_driver pxa_ata_driver = {
+	.probe		= pxa_ata_probe,
+	.remove		= ata_platform_remove_one,
+	.driver		= {
+		.name		= DRV_NAME,
+	},
+};
+
+module_platform_driver(pxa_ata_driver);
+
+MODULE_AUTHOR("Marek Vasut <marek.vasut@gmail.com>");
+MODULE_DESCRIPTION("DMA-capable driver for PATA on PXA CPU");
+MODULE_LICENSE("GPL");
+MODULE_VERSION(DRV_VERSION);
+MODULE_ALIAS("platform:" DRV_NAME);