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, 464 insertions, 0 deletions

diff --git a/drivers/misc/sram.c b/drivers/misc/sram.c
new file mode 100644
index 000000000..c8990e83a
--- /dev/null
+++ b/drivers/misc/sram.c
@@ -0,0 +1,464 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Generic on-chip SRAM allocation driver
+ *
+ * Copyright (C) 2012 Philipp Zabel, Pengutronix
+ */
+
+#include <linux/clk.h>
+#include <linux/delay.h>
+#include <linux/genalloc.h>
+#include <linux/io.h>
+#include <linux/list_sort.h>
+#include <linux/of.h>
+#include <linux/of_address.h>
+#include <linux/platform_device.h>
+#include <linux/regmap.h>
+#include <linux/slab.h>
+#include <linux/mfd/syscon.h>
+#include <soc/at91/atmel-secumod.h>
+
+#include "sram.h"
+
+#define SRAM_GRANULARITY	32
+
+static ssize_t sram_read(struct file *filp, struct kobject *kobj,
+			 const struct bin_attribute *attr,
+			 char *buf, loff_t pos, size_t count)
+{
+	struct sram_partition *part;
+
+	/* Cast away the const as the attribute is part of a larger structure */
+	part = (struct sram_partition *)container_of(attr, struct sram_partition, battr);
+
+	mutex_lock(&part->lock);
+	memcpy_fromio(buf, part->base + pos, count);
+	mutex_unlock(&part->lock);
+
+	return count;
+}
+
+static ssize_t sram_write(struct file *filp, struct kobject *kobj,
+			  const struct bin_attribute *attr,
+			  char *buf, loff_t pos, size_t count)
+{
+	struct sram_partition *part;
+
+	/* Cast away the const as the attribute is part of a larger structure */
+	part = (struct sram_partition *)container_of(attr, struct sram_partition, battr);
+
+	mutex_lock(&part->lock);
+	memcpy_toio(part->base + pos, buf, count);
+	mutex_unlock(&part->lock);
+
+	return count;
+}
+
+static int sram_add_pool(struct sram_dev *sram, struct sram_reserve *block,
+			 phys_addr_t start, struct sram_partition *part)
+{
+	int ret;
+
+	part->pool = devm_gen_pool_create(sram->dev, ilog2(SRAM_GRANULARITY),
+					  NUMA_NO_NODE, block->label);
+	if (IS_ERR(part->pool))
+		return PTR_ERR(part->pool);
+
+	ret = gen_pool_add_virt(part->pool, (unsigned long)part->base, start,
+				block->size, NUMA_NO_NODE);
+	if (ret < 0) {
+		dev_err(sram->dev, "failed to register subpool: %d\n", ret);
+		return ret;
+	}
+
+	return 0;
+}
+
+static int sram_add_export(struct sram_dev *sram, struct sram_reserve *block,
+			   phys_addr_t start, struct sram_partition *part)
+{
+	sysfs_bin_attr_init(&part->battr);
+	part->battr.attr.name = devm_kasprintf(sram->dev, GFP_KERNEL,
+					       "%llx.sram",
+					       (unsigned long long)start);
+	if (!part->battr.attr.name)
+		return -ENOMEM;
+
+	part->battr.attr.mode = S_IRUSR | S_IWUSR;
+	part->battr.read = sram_read;
+	part->battr.write = sram_write;
+	part->battr.size = block->size;
+
+	return device_create_bin_file(sram->dev, &part->battr);
+}
+
+static int sram_add_partition(struct sram_dev *sram, struct sram_reserve *block,
+			      phys_addr_t start)
+{
+	int ret;
+	struct sram_partition *part = &sram->partition[sram->partitions];
+
+	mutex_init(&part->lock);
+
+	if (sram->config && sram->config->map_only_reserved) {
+		void __iomem *virt_base;
+
+		if (sram->no_memory_wc)
+			virt_base = devm_ioremap_resource(sram->dev, &block->res);
+		else
+			virt_base = devm_ioremap_resource_wc(sram->dev, &block->res);
+
+		if (IS_ERR(virt_base)) {
+			dev_err(sram->dev, "could not map SRAM at %pr\n", &block->res);
+			return PTR_ERR(virt_base);
+		}
+
+		part->base = virt_base;
+	} else {
+		part->base = sram->virt_base + block->start;
+	}
+
+	if (block->pool) {
+		ret = sram_add_pool(sram, block, start, part);
+		if (ret)
+			return ret;
+	}
+	if (block->export) {
+		ret = sram_add_export(sram, block, start, part);
+		if (ret)
+			return ret;
+	}
+	if (block->protect_exec) {
+		ret = sram_check_protect_exec(sram, block, part);
+		if (ret)
+			return ret;
+
+		ret = sram_add_pool(sram, block, start, part);
+		if (ret)
+			return ret;
+
+		sram_add_protect_exec(part);
+	}
+
+	sram->partitions++;
+
+	return 0;
+}
+
+static void sram_free_partitions(struct sram_dev *sram)
+{
+	struct sram_partition *part;
+
+	if (!sram->partitions)
+		return;
+
+	part = &sram->partition[sram->partitions - 1];
+	for (; sram->partitions; sram->partitions--, part--) {
+		if (part->battr.size)
+			device_remove_bin_file(sram->dev, &part->battr);
+
+		if (part->pool &&
+		    gen_pool_avail(part->pool) < gen_pool_size(part->pool))
+			dev_err(sram->dev, "removed pool while SRAM allocated\n");
+	}
+}
+
+static int sram_reserve_cmp(void *priv, const struct list_head *a,
+					const struct list_head *b)
+{
+	const struct sram_reserve *ra = list_entry(a, struct sram_reserve, list);
+	const struct sram_reserve *rb = list_entry(b, struct sram_reserve, list);
+
+	return ra->start - rb->start;
+}
+
+static int sram_reserve_regions(struct sram_dev *sram, struct resource *res)
+{
+	struct device_node *np = sram->dev->of_node, *child;
+	unsigned long size, cur_start, cur_size;
+	struct sram_reserve *rblocks, *block;
+	struct list_head reserve_list;
+	unsigned int nblocks, exports = 0;
+	const char *label;
+	int ret = 0;
+
+	INIT_LIST_HEAD(&reserve_list);
+
+	size = resource_size(res);
+
+	/*
+	 * We need an additional block to mark the end of the memory region
+	 * after the reserved blocks from the dt are processed.
+	 */
+	nblocks = (np) ? of_get_available_child_count(np) + 1 : 1;
+	rblocks = kzalloc_objs(*rblocks, nblocks);
+	if (!rblocks)
+		return -ENOMEM;
+
+	block = &rblocks[0];
+	for_each_available_child_of_node(np, child) {
+		struct resource child_res;
+
+		ret = of_address_to_resource(child, 0, &child_res);
+		if (ret < 0) {
+			dev_err(sram->dev,
+				"could not get address for node %pOF\n",
+				child);
+			goto err_chunks;
+		}
+
+		if (child_res.start < res->start || child_res.end > res->end) {
+			dev_err(sram->dev,
+				"reserved block %pOF outside the sram area\n",
+				child);
+			ret = -EINVAL;
+			goto err_chunks;
+		}
+
+		block->start = child_res.start - res->start;
+		block->size = resource_size(&child_res);
+		block->res = child_res;
+		list_add_tail(&block->list, &reserve_list);
+
+		block->export = of_property_read_bool(child, "export");
+		block->pool = of_property_read_bool(child, "pool");
+		block->protect_exec = of_property_read_bool(child, "protect-exec");
+
+		if ((block->export || block->pool || block->protect_exec) &&
+		    block->size) {
+			exports++;
+
+			label = NULL;
+			ret = of_property_read_string(child, "label", &label);
+			if (ret && ret != -EINVAL) {
+				dev_err(sram->dev,
+					"%pOF has invalid label name\n",
+					child);
+				goto err_chunks;
+			}
+			if (!label)
+				block->label = devm_kasprintf(sram->dev, GFP_KERNEL,
+							      "%s", of_node_full_name(child));
+			else
+				block->label = devm_kstrdup(sram->dev,
+							    label, GFP_KERNEL);
+			if (!block->label) {
+				ret = -ENOMEM;
+				goto err_chunks;
+			}
+
+			dev_dbg(sram->dev, "found %sblock '%s' 0x%x-0x%x\n",
+				block->export ? "exported " : "", block->label,
+				block->start, block->start + block->size);
+		} else {
+			dev_dbg(sram->dev, "found reserved block 0x%x-0x%x\n",
+				block->start, block->start + block->size);
+		}
+
+		block++;
+	}
+	child = NULL;
+
+	/* the last chunk marks the end of the region */
+	rblocks[nblocks - 1].start = size;
+	rblocks[nblocks - 1].size = 0;
+	list_add_tail(&rblocks[nblocks - 1].list, &reserve_list);
+
+	list_sort(NULL, &reserve_list, sram_reserve_cmp);
+
+	if (exports) {
+		sram->partition = devm_kcalloc(sram->dev,
+				       exports, sizeof(*sram->partition),
+				       GFP_KERNEL);
+		if (!sram->partition) {
+			ret = -ENOMEM;
+			goto err_chunks;
+		}
+	}
+
+	cur_start = 0;
+	list_for_each_entry(block, &reserve_list, list) {
+		/* can only happen if sections overlap */
+		if (block->start < cur_start) {
+			dev_err(sram->dev,
+				"block at 0x%x starts after current offset 0x%lx\n",
+				block->start, cur_start);
+			ret = -EINVAL;
+			sram_free_partitions(sram);
+			goto err_chunks;
+		}
+
+		if ((block->export || block->pool || block->protect_exec) &&
+		    block->size) {
+			ret = sram_add_partition(sram, block,
+						 res->start + block->start);
+			if (ret) {
+				sram_free_partitions(sram);
+				goto err_chunks;
+			}
+		}
+
+		/* current start is in a reserved block, so continue after it */
+		if (block->start == cur_start) {
+			cur_start = block->start + block->size;
+			continue;
+		}
+
+		/*
+		 * allocate the space between the current starting
+		 * address and the following reserved block, or the
+		 * end of the region.
+		 */
+		cur_size = block->start - cur_start;
+
+		if (sram->pool) {
+			dev_dbg(sram->dev, "adding chunk 0x%lx-0x%lx\n",
+				cur_start, cur_start + cur_size);
+
+			ret = gen_pool_add_virt(sram->pool,
+					(unsigned long)sram->virt_base + cur_start,
+					res->start + cur_start, cur_size, -1);
+			if (ret < 0) {
+				sram_free_partitions(sram);
+				goto err_chunks;
+			}
+		}
+
+		/* next allocation after this reserved block */
+		cur_start = block->start + block->size;
+	}
+
+err_chunks:
+	of_node_put(child);
+	kfree(rblocks);
+
+	return ret;
+}
+
+static int atmel_securam_wait(void)
+{
+	struct regmap *regmap;
+	u32 val;
+
+	regmap = syscon_regmap_lookup_by_compatible("atmel,sama5d2-secumod");
+	if (IS_ERR(regmap))
+		return -ENODEV;
+
+	return regmap_read_poll_timeout(regmap, AT91_SECUMOD_RAMRDY, val,
+					val & AT91_SECUMOD_RAMRDY_READY,
+					10000, 500000);
+}
+
+static const struct sram_config atmel_securam_config = {
+	.init = atmel_securam_wait,
+};
+
+/*
+ * SYSRAM contains areas that are not accessible by the
+ * kernel, such as the first 256K that is reserved for TZ.
+ * Accesses to those areas (including speculative accesses)
+ * trigger SErrors. As such we must map only the areas of
+ * SYSRAM specified in the device tree.
+ */
+static const struct sram_config tegra_sysram_config = {
+	.map_only_reserved = true,
+};
+
+static const struct of_device_id sram_dt_ids[] = {
+	{ .compatible = "mmio-sram" },
+	{ .compatible = "atmel,sama5d2-securam", .data = &atmel_securam_config },
+	{ .compatible = "nvidia,tegra186-sysram", .data = &tegra_sysram_config },
+	{ .compatible = "nvidia,tegra194-sysram", .data = &tegra_sysram_config },
+	{ .compatible = "nvidia,tegra234-sysram", .data = &tegra_sysram_config },
+	{ }
+};
+
+static int sram_probe(struct platform_device *pdev)
+{
+	const struct sram_config *config;
+	struct sram_dev *sram;
+	int ret;
+	struct resource *res;
+	struct clk *clk;
+
+	config = of_device_get_match_data(&pdev->dev);
+
+	sram = devm_kzalloc(&pdev->dev, sizeof(*sram), GFP_KERNEL);
+	if (!sram)
+		return -ENOMEM;
+
+	sram->dev = &pdev->dev;
+	sram->no_memory_wc = of_property_read_bool(pdev->dev.of_node, "no-memory-wc");
+	sram->config = config;
+
+	if (!config || !config->map_only_reserved) {
+		res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+		if (sram->no_memory_wc)
+			sram->virt_base = devm_ioremap_resource(&pdev->dev, res);
+		else
+			sram->virt_base = devm_ioremap_resource_wc(&pdev->dev, res);
+		if (IS_ERR(sram->virt_base)) {
+			dev_err(&pdev->dev, "could not map SRAM registers\n");
+			return PTR_ERR(sram->virt_base);
+		}
+
+		sram->pool = devm_gen_pool_create(sram->dev, ilog2(SRAM_GRANULARITY),
+						  NUMA_NO_NODE, NULL);
+		if (IS_ERR(sram->pool))
+			return PTR_ERR(sram->pool);
+	}
+
+	clk = devm_clk_get_optional_enabled(sram->dev, NULL);
+	if (IS_ERR(clk))
+		return PTR_ERR(clk);
+
+	ret = sram_reserve_regions(sram,
+			platform_get_resource(pdev, IORESOURCE_MEM, 0));
+	if (ret)
+		return ret;
+
+	platform_set_drvdata(pdev, sram);
+
+	if (config && config->init) {
+		ret = config->init();
+		if (ret)
+			goto err_free_partitions;
+	}
+
+	if (sram->pool)
+		dev_dbg(sram->dev, "SRAM pool: %zu KiB @ 0x%p\n",
+			gen_pool_size(sram->pool) / 1024, sram->virt_base);
+
+	return 0;
+
+err_free_partitions:
+	sram_free_partitions(sram);
+
+	return ret;
+}
+
+static void sram_remove(struct platform_device *pdev)
+{
+	struct sram_dev *sram = platform_get_drvdata(pdev);
+
+	sram_free_partitions(sram);
+
+	if (sram->pool && gen_pool_avail(sram->pool) < gen_pool_size(sram->pool))
+		dev_err(sram->dev, "removed while SRAM allocated\n");
+}
+
+static struct platform_driver sram_driver = {
+	.driver = {
+		.name = "sram",
+		.of_match_table = sram_dt_ids,
+	},
+	.probe = sram_probe,
+	.remove = sram_remove,
+};
+
+static int __init sram_init(void)
+{
+	return platform_driver_register(&sram_driver);
+}
+
+postcore_initcall(sram_init);