From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions 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) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- lib/logic_iomem.c | 321 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 321 insertions(+) create mode 100644 lib/logic_iomem.c (limited to 'lib/logic_iomem.c') diff --git a/lib/logic_iomem.c b/lib/logic_iomem.c new file mode 100644 index 000000000..42a571d05 --- /dev/null +++ b/lib/logic_iomem.c @@ -0,0 +1,321 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright (C) 2021 Intel Corporation + * Author: Johannes Berg + */ +#include +#include +#include +#include + +struct logic_iomem_region { + const struct resource *res; + const struct logic_iomem_region_ops *ops; + struct list_head list; +}; + +struct logic_iomem_area { + const struct logic_iomem_ops *ops; + void *priv; +}; + +#define AREA_SHIFT 24 +#define MAX_AREA_SIZE (1 << AREA_SHIFT) +#define MAX_AREAS ((1U << 31) / MAX_AREA_SIZE) +#define AREA_BITS ((MAX_AREAS - 1) << AREA_SHIFT) +#define AREA_MASK (MAX_AREA_SIZE - 1) +#ifdef CONFIG_64BIT +#define IOREMAP_BIAS 0xDEAD000000000000UL +#define IOREMAP_MASK 0xFFFFFFFF00000000UL +#else +#define IOREMAP_BIAS 0x80000000UL +#define IOREMAP_MASK 0x80000000UL +#endif + +static DEFINE_MUTEX(regions_mtx); +static LIST_HEAD(regions_list); +static struct logic_iomem_area mapped_areas[MAX_AREAS]; + +int logic_iomem_add_region(struct resource *resource, + const struct logic_iomem_region_ops *ops) +{ + struct logic_iomem_region *rreg; + int err; + + if (WARN_ON(!resource || !ops)) + return -EINVAL; + + if (WARN_ON((resource->flags & IORESOURCE_TYPE_BITS) != IORESOURCE_MEM)) + return -EINVAL; + + rreg = kzalloc_obj(*rreg); + if (!rreg) + return -ENOMEM; + + err = request_resource(&iomem_resource, resource); + if (err) { + kfree(rreg); + return -ENOMEM; + } + + mutex_lock(®ions_mtx); + rreg->res = resource; + rreg->ops = ops; + list_add_tail(&rreg->list, ®ions_list); + mutex_unlock(®ions_mtx); + + return 0; +} +EXPORT_SYMBOL(logic_iomem_add_region); + +#ifndef CONFIG_INDIRECT_IOMEM_FALLBACK +static void __iomem *real_ioremap(phys_addr_t offset, size_t size) +{ + WARN(1, "invalid ioremap(0x%llx, 0x%zx)\n", + (unsigned long long)offset, size); + return NULL; +} + +static void real_iounmap(volatile void __iomem *addr) +{ + WARN(1, "invalid iounmap for addr 0x%llx\n", + (unsigned long long)(uintptr_t __force)addr); +} +#endif /* CONFIG_INDIRECT_IOMEM_FALLBACK */ + +void __iomem *ioremap(phys_addr_t offset, size_t size) +{ + void __iomem *ret = NULL; + struct logic_iomem_region *rreg, *found = NULL; + int i; + + mutex_lock(®ions_mtx); + list_for_each_entry(rreg, ®ions_list, list) { + if (rreg->res->start > offset) + continue; + if (rreg->res->end < offset + size - 1) + continue; + found = rreg; + break; + } + + if (!found) + goto out; + + for (i = 0; i < MAX_AREAS; i++) { + long offs; + + if (mapped_areas[i].ops) + continue; + + offs = rreg->ops->map(offset - found->res->start, + size, &mapped_areas[i].ops, + &mapped_areas[i].priv); + if (offs < 0) { + mapped_areas[i].ops = NULL; + break; + } + + if (WARN_ON(!mapped_areas[i].ops)) { + mapped_areas[i].ops = NULL; + break; + } + + ret = (void __iomem *)(IOREMAP_BIAS + (i << AREA_SHIFT) + offs); + break; + } +out: + mutex_unlock(®ions_mtx); + if (ret) + return ret; + return real_ioremap(offset, size); +} +EXPORT_SYMBOL(ioremap); + +static inline struct logic_iomem_area * +get_area(const volatile void __iomem *addr) +{ + unsigned long a = (unsigned long)addr; + unsigned int idx; + + if (WARN_ON((a & IOREMAP_MASK) != IOREMAP_BIAS)) + return NULL; + + idx = (a & AREA_BITS) >> AREA_SHIFT; + + if (mapped_areas[idx].ops) + return &mapped_areas[idx]; + + return NULL; +} + +void iounmap(volatile void __iomem *addr) +{ + struct logic_iomem_area *area = get_area(addr); + + if (!area) { + real_iounmap(addr); + return; + } + + if (area->ops->unmap) + area->ops->unmap(area->priv); + + mutex_lock(®ions_mtx); + area->ops = NULL; + area->priv = NULL; + mutex_unlock(®ions_mtx); +} +EXPORT_SYMBOL(iounmap); + +#ifndef CONFIG_INDIRECT_IOMEM_FALLBACK +#define MAKE_FALLBACK(op, sz) \ +static u##sz real_raw_read ## op(const volatile void __iomem *addr) \ +{ \ + WARN(1, "Invalid read" #op " at address %llx\n", \ + (unsigned long long)(uintptr_t __force)addr); \ + return (u ## sz)~0ULL; \ +} \ + \ +static void real_raw_write ## op(u ## sz val, \ + volatile void __iomem *addr) \ +{ \ + WARN(1, "Invalid writeq" #op " of 0x%llx at address %llx\n", \ + (unsigned long long)val, \ + (unsigned long long)(uintptr_t __force)addr);\ +} \ + +MAKE_FALLBACK(b, 8); +MAKE_FALLBACK(w, 16); +MAKE_FALLBACK(l, 32); +#ifdef CONFIG_64BIT +MAKE_FALLBACK(q, 64); +#endif + +static void real_memset_io(volatile void __iomem *addr, int value, size_t size) +{ + WARN(1, "Invalid memset_io at address 0x%llx\n", + (unsigned long long)(uintptr_t __force)addr); +} + +static void real_memcpy_fromio(void *buffer, const volatile void __iomem *addr, + size_t size) +{ + WARN(1, "Invalid memcpy_fromio at address 0x%llx\n", + (unsigned long long)(uintptr_t __force)addr); + + memset(buffer, 0xff, size); +} + +static void real_memcpy_toio(volatile void __iomem *addr, const void *buffer, + size_t size) +{ + WARN(1, "Invalid memcpy_toio at address 0x%llx\n", + (unsigned long long)(uintptr_t __force)addr); +} +#endif /* CONFIG_INDIRECT_IOMEM_FALLBACK */ + +#define MAKE_OP(op, sz) \ +u##sz __raw_read ## op(const volatile void __iomem *addr) \ +{ \ + struct logic_iomem_area *area = get_area(addr); \ + \ + if (!area) \ + return real_raw_read ## op(addr); \ + \ + return (u ## sz) area->ops->read(area->priv, \ + (unsigned long)addr & AREA_MASK,\ + sz / 8); \ +} \ +EXPORT_SYMBOL(__raw_read ## op); \ + \ +void __raw_write ## op(u ## sz val, volatile void __iomem *addr) \ +{ \ + struct logic_iomem_area *area = get_area(addr); \ + \ + if (!area) { \ + real_raw_write ## op(val, addr); \ + return; \ + } \ + \ + area->ops->write(area->priv, \ + (unsigned long)addr & AREA_MASK, \ + sz / 8, val); \ +} \ +EXPORT_SYMBOL(__raw_write ## op) + +MAKE_OP(b, 8); +MAKE_OP(w, 16); +MAKE_OP(l, 32); +#ifdef CONFIG_64BIT +MAKE_OP(q, 64); +#endif + +void memset_io(volatile void __iomem *addr, int value, size_t size) +{ + struct logic_iomem_area *area = get_area(addr); + unsigned long offs, start; + + if (!area) { + real_memset_io(addr, value, size); + return; + } + + start = (unsigned long)addr & AREA_MASK; + + if (area->ops->set) { + area->ops->set(area->priv, start, value, size); + return; + } + + for (offs = 0; offs < size; offs++) + area->ops->write(area->priv, start + offs, 1, value); +} +EXPORT_SYMBOL(memset_io); + +void memcpy_fromio(void *buffer, const volatile void __iomem *addr, + size_t size) +{ + struct logic_iomem_area *area = get_area(addr); + u8 *buf = buffer; + unsigned long offs, start; + + if (!area) { + real_memcpy_fromio(buffer, addr, size); + return; + } + + start = (unsigned long)addr & AREA_MASK; + + if (area->ops->copy_from) { + area->ops->copy_from(area->priv, buffer, start, size); + return; + } + + for (offs = 0; offs < size; offs++) + buf[offs] = area->ops->read(area->priv, start + offs, 1); +} +EXPORT_SYMBOL(memcpy_fromio); + +void memcpy_toio(volatile void __iomem *addr, const void *buffer, size_t size) +{ + struct logic_iomem_area *area = get_area(addr); + const u8 *buf = buffer; + unsigned long offs, start; + + if (!area) { + real_memcpy_toio(addr, buffer, size); + return; + } + + start = (unsigned long)addr & AREA_MASK; + + if (area->ops->copy_to) { + area->ops->copy_to(area->priv, start, buffer, size); + return; + } + + for (offs = 0; offs < size; offs++) + area->ops->write(area->priv, start + offs, 1, buf[offs]); +} +EXPORT_SYMBOL(memcpy_toio); -- cgit v1.3.1