kprobes: rethook,ARM,arm64: Replace kretprobe trampoline with rethook
Subsystems (6)
Masami Hiramatsu
maintainerv2 cover kprobes: rethook,ARM,arm64: Replace kretprobe trampoline with rethook
Hi,
Here is the 2nd version of the series for replacing kretprobe trampoline with rethook on ARM/arm64. I fixed some compiler warnings in this version. The previous version is here[1];
[1] https://lore.kernel.org/all/164915121498.982637.12787715964983738566.stgit@devnote2/T/#u
This series includes a trivial bugfix for the arm unwinder to initialize an internal data structure([1/4]). This is not critical for stack trace, but required for rethook to find the LR register from the stack. This also have an update for the rethook interface, which allows us to check the rethook_hook() failure ([2/4]). This is also required for the rethook on arm because unwinder is able to fail.
The rest of patches are replacing kretprobe trampoline with rethook on ARM ([3/4]) and arm64 ([4/4]).
Background:
This rethook came from Jiri's request of multiple kprobe for bpf[2]. He tried to solve an issue that starting bpf with multiple kprobe will take a long time because bpf-kprobe will wait for RCU grace period for sync rcu events.
Jiri wanted to attach a single bpf handler to multiple kprobes and he tried to introduce multiple-probe interface to kprobe. So I asked him to use ftrace and kretprobe-like hook if it is only for the function entry and exit, instead of adding ad-hoc interface to kprobes.
For this purpose, I introduced the fprobe (kprobe like interface for ftrace) with the rethook (this is a generic return hook feature for fprobe exit handler)[3].
[2] https://lore.kernel.org/all/[email protected]/T/#u
[3] https://lore.kernel.org/all/164191321766.806991.7930388561276940676.stgit@devnote2/T/#u
The rethook is basically same as the kretprobe trampoline. I just made it decoupled from kprobes. Eventually, the all arch dependent kretprobe trampolines will be replaced with the rethook trampoline instead of cloning the code.
When I port the rethook for all arch which supports kretprobe, the legacy kretprobe specific code (which is for CONFIG_KRETPROBE_ON_RETHOOK=n) will be removed eventually.
BTW, the arm Clang support for rethook is for kretprobes only. fprobe and ftrace seems not working with Clang yet.
Thank you,
---
Masami Hiramatsu (4):
ARM: unwind: Initialize the lr_addr field of unwind_ctrl_block rethook,fprobe,kprobes: Check a failure in the rethook_hook()
ARM: rethook: Replace kretprobe trampoline with rethook
arm64: rethook: Replace kretprobe trampoline with rethook
arch/arm/Kconfig | 1 arch/arm/include/asm/stacktrace.h | 5 + arch/arm/kernel/stacktrace.c | 13 +-- arch/arm/kernel/unwind.c | 1 arch/arm/probes/Makefile | 1 arch/arm/probes/kprobes/core.c | 62 ------------ arch/arm/probes/rethook.c | 127 +++++++++++++++++++++++++ arch/arm64/Kconfig | 1 arch/arm64/include/asm/kprobes.h | 2 arch/arm64/include/asm/stacktrace.h | 2 arch/arm64/kernel/Makefile | 1 arch/arm64/kernel/probes/Makefile | 1 arch/arm64/kernel/probes/kprobes.c | 15 --- arch/arm64/kernel/probes/kprobes_trampoline.S | 86 ----------------- arch/arm64/kernel/rethook.c | 28 ++++++ arch/arm64/kernel/rethook_trampoline.S | 87 +++++++++++++++++ arch/arm64/kernel/stacktrace.c | 9 +- arch/x86/kernel/rethook.c | 4 + include/linux/rethook.h | 4 - kernel/kprobes.c | 8 +- kernel/trace/fprobe.c | 5 + kernel/trace/rethook.c | 12 ++ 22 files changed, 287 insertions(+), 188 deletions(-) create mode 100644 arch/arm/probes/rethook.c delete mode 100644 arch/arm64/kernel/probes/kprobes_trampoline.S create mode 100644 arch/arm64/kernel/rethook.c create mode 100644 arch/arm64/kernel/rethook_trampoline.S
--
Masami Hiramatsu (Linaro) <[email protected]>
Masami Hiramatsu
maintainerPATCH v2 1/4 ARM: unwind: Initialize the lr_addr field of unwind_ctrl_block
Since the unwind_ctrl_block::lr_addr is finally passed to stackframe::lr_addr, that value will be exposed unconditionally. Thus it should be initialized.
Without this fix, when unwind_frame() doesn't update the unwind_ctrl_block::lr_addr (e.g. 'lr' register is not saved in the target function), stackframe::lr_addr will contain a wrong value.
---
Changes in v2:
- Initialize pointer by NULL instead of 0.
--- arch/arm/kernel/unwind.c | 1 + 1 file changed, 1 insertion(+)
arch/arm/kernel/unwind.c +1 -0
@@ -404,6 +404,7 @@ int unwind_frame(struct stackframe *frame) ctrl.vrs[SP ] = frame->sp; ctrl.vrs[LR ] = frame->lr; ctrl.vrs[PC ] =0 ; + ctrl.lr_addr =NULL ;if (idx->insn ==1 )/* can't unwind */
Masami Hiramatsu
maintainerPATCH v2 2/4 rethook,fprobe,kprobes: Check a failure in the rethook_hook()
In reply to Masami HiramatsuSince there are possible to fail to hook the function return (depends on archtecutre implememtation), rethook_hook() should return the error in that case and caller must check it.
--- arch/x86/kernel/rethook.c | 4 +++- include/linux/rethook.h | 4 ++-- kernel/kprobes.c | 8 +++++--- kernel/trace/fprobe.c | 5 ++++- kernel/trace/rethook.c | 12 ++++++++++-- 5 files changed, 24 insertions(+), 9 deletions(-)
arch/x86/kernel/rethook.c +3 -1
@@ -114,7 +114,7 @@ void arch_rethook_fixup_return(struct pt_regs *regs, }NOKPROBE_SYMBOL (arch_rethook_fixup_return); -void arch_rethook_prepare (struct rethook_node *rh,struct pt_regs *regs,bool mcount) +int arch_rethook_prepare (struct rethook_node *rh,struct pt_regs *regs,bool mcount) {unsigned long *stack = (unsigned long *)regs->sp;
@@ -123,5 +123,7 @@ void arch_rethook_prepare(struct rethook_node *rh, struct pt_regs *regs, bool mc/* Replace the return addr with trampoline addr */ stack[0 ] = (unsigned long ) arch_rethook_trampoline; + +return 0 ; }NOKPROBE_SYMBOL (arch_rethook_prepare);
include/linux/rethook.h +2 -2
@@ -63,12 +63,12 @@ void rethook_free(struct rethook *rh);void rethook_add_node (struct rethook *rh,struct rethook_node *node);struct rethook_node *rethook_try_get (struct rethook *rh);void rethook_recycle (struct rethook_node *node); -void rethook_hook (struct rethook_node *node,struct pt_regs *regs,bool mcount); +int rethook_hook (struct rethook_node *node,struct pt_regs *regs,bool mcount);unsigned long rethook_find_ret_addr (struct task_struct *tsk,unsigned long frame,struct llist_node **cur);/* Arch dependent code must implement arch_* and trampoline code */ -void arch_rethook_prepare (struct rethook_node *node,struct pt_regs *regs,bool mcount); +int arch_rethook_prepare (struct rethook_node *node,struct pt_regs *regs,bool mcount);void arch_rethook_trampoline (void );/**
kernel/kprobes.c +5 -3
@@ -2109,10 +2109,12 @@ static int pre_handler_kretprobe(struct kprobe *p, struct pt_regs *regs) ri = container_of(rhn, struct kretprobe_instance, node); -if (rp->entry_handler && rp->entry_handler (ri, regs)) +if (rp->entry_handler && rp->entry_handler (ri, regs)) {rethook_recycle (rhn); -else -rethook_hook (rhn, regs,kprobe_ftrace (p)); + }else if (rethook_hook (rhn, regs,kprobe_ftrace (p)) <0 ) { +rethook_recycle (rhn); + rp->nmissed++; + }return 0 ; }
kernel/trace/fprobe.c +4 -1
@@ -48,7 +48,10 @@ static void fprobe_handler(unsigned long ip, unsigned long parent_ip, } fpr = container_of(rh, struct fprobe_rethook_node, node); fpr->entry_ip = ip; -rethook_hook (rh,ftrace_get_regs (fregs), true); +if (rethook_hook (rh,ftrace_get_regs (fregs), true) <0 ) { +rethook_recycle (rh); + fp->nmissed++; + } } out:
kernel/trace/rethook.c +10 -2
@@ -174,11 +174,19 @@ NOKPROBE_SYMBOL(rethook_try_get);* from ftrace (mcount) callback, @mcount must be set true. If this is called * from the real function entry (e.g. kprobes) @mcount must be set false. * This is because the way to hook the function return depends on the context. +* This returns 0 if succeeded to hook the function return, or -errno if +* failed. */ -void rethook_hook (struct rethook_node *node,struct pt_regs *regs,bool mcount) +int rethook_hook (struct rethook_node *node,struct pt_regs *regs,bool mcount) { -arch_rethook_prepare (node, regs, mcount); +int ret; + + ret =arch_rethook_prepare (node, regs, mcount); +if (ret <0 ) +return ret; +__llist_add (&node->llist, ¤t->rethooks); +return 0 ; }NOKPROBE_SYMBOL (rethook_hook);
Masami Hiramatsu
maintainerPATCH v2 3/4 ARM: rethook: Replace kretprobe trampoline with rethook
In reply to Masami HiramatsuReplace the kretprob's trampoline code with the rethook on arm. This also enables rethook support on arm. Most of the code has been copied from kretprobe on arm.
The significant difference is that the rethook on mcount (ftrace) support is added. If the rethook is called from the kprobes for kretprobe, there is no problem to replace the LR register with trampoline address because the LR register will be saved after kprobe probed. However, the mcount call will be placed right after making a stack frame for the function. This means we have to decode the stackframe to find where the LR register is saved. With the CONFIG_FRAME_POINTER, the frame pointer (FP register) is used. Without that, rethook has to unwind one stack frame to find it.
--- arch/arm/Kconfig | 1 arch/arm/include/asm/stacktrace.h | 5 + arch/arm/kernel/stacktrace.c | 13 ++-- arch/arm/probes/Makefile | 1 arch/arm/probes/kprobes/core.c | 62 ------------------ arch/arm/probes/rethook.c | 127 +++++++++++++++++++++++++++++++++++++ 6 files changed, 139 insertions(+), 70 deletions(-) create mode 100644 arch/arm/probes/rethook.c
arch/arm/Kconfig +1 -0
@@ -110,6 +110,7 @@ config ARM select HAVE_MOD_ARCH_SPECIFIC select HAVE_NMI select HAVE_OPTPROBES if !THUMB2_KERNEL + select HAVE_RETHOOK select HAVE_PERF_EVENTS select HAVE_PERF_REGS select HAVE_PERF_USER_STACK_DUMP
arch/arm/include/asm/stacktrace.h +3 -2
@@ -17,7 +17,8 @@ struct stackframe {/* address of the LR value on the stack */ unsigned long *lr_addr; -#ifdef CONFIG_KRETPROBES + +#if defined(CONFIG_RETHOOK )struct llist_node *kr_cur;struct task_struct *tsk;#endif
@@ -30,7 +31,7 @@ void arm_get_current_stackframe(struct pt_regs *regs, struct stackframe *frame) frame->sp = regs->ARM_sp; frame->lr = regs->ARM_lr; frame->pc = regs->ARM_pc; -#ifdef CONFIG_KRETPROBES +#if defined(CONFIG_RETHOOK ) frame->kr_cur =NULL ; frame->tsk = current;#endif
arch/arm/kernel/stacktrace.c +7 -6
@@ -1,6 +1,7 @@// SPDX-License-Identifier: GPL-2.0-only #include <linux/export.h> #include <linux/kprobes.h> +#include <linux/rethook.h> #include <linux/sched.h> #include <linux/sched/debug.h> #include <linux/stacktrace.h>
@@ -66,10 +67,10 @@ int notrace unwind_frame(struct stackframe *frame) frame->sp = *(unsigned long *)(fp -8 ); frame->pc = *(unsigned long *)(fp -4 );#endif -#ifdef CONFIG_KRETPROBES -if (is_kretprobe_trampoline (frame->pc)) - frame->pc =kretprobe_find_ret_addr (frame->tsk, - (void *)frame->fp, &frame->kr_cur); +#ifdef CONFIG_RETHOOK +if (is_rethook_trampoline (frame->pc)) + frame->pc =rethook_find_ret_addr (frame->tsk, frame->fp, + &frame->kr_cur);#endif return 0 ;
@@ -163,7 +164,7 @@ static noinline void __save_stack_trace(struct task_struct *tsk, here: frame.pc = (unsigned long )&&here; } -#ifdef CONFIG_KRETPROBES +#ifdef CONFIG_RETHOOK frame.kr_cur =NULL ; frame.tsk = tsk;#endif
@@ -184,7 +185,7 @@ void save_stack_trace_regs(struct pt_regs *regs, struct stack_trace *trace) frame.sp = regs->ARM_sp; frame.lr = regs->ARM_lr; frame.pc = regs->ARM_pc; -#ifdef CONFIG_KRETPROBES +#ifdef CONFIG_RETHOOK frame.kr_cur =NULL ; frame.tsk = current;#endif
arch/arm/probes/Makefile +1 -0
@@ -6,3 +6,4 @@ obj-$(CONFIG_KPROBES) += decode-thumb.o else obj-$(CONFIG_KPROBES) += decode-arm.o endif +obj-$(CONFIG_RETHOOK) += rethook.o
arch/arm/probes/kprobes/core.c +0 -62
68 lines not on this page · raw message
arch/arm/probes/rethook.c new file +127 -0
127 lines not on this page · raw message
Masami Hiramatsu
maintainerPATCH v2 4/4 arm64: rethook: Replace kretprobe trampoline with rethook
In reply to Masami HiramatsuReplace the kretprobe's trampoline code with the rethook on arm64. The rethook on arm64 is almost renamed from kretprobe trampoline code. The mechanism is completely same.
---
Changes in v2:
- Fix a compile warning about no prototype.
--- arch/arm64/Kconfig | 1 arch/arm64/include/asm/kprobes.h | 2 - arch/arm64/include/asm/stacktrace.h | 2 - arch/arm64/kernel/Makefile | 1 arch/arm64/kernel/probes/Makefile | 1 arch/arm64/kernel/probes/kprobes.c | 15 ---- arch/arm64/kernel/probes/kprobes_trampoline.S | 86 ------------------------- arch/arm64/kernel/rethook.c | 28 ++++++++ arch/arm64/kernel/rethook_trampoline.S | 87 +++++++++++++++++++++++++ arch/arm64/kernel/stacktrace.c | 9 +-- 10 files changed, 123 insertions(+), 109 deletions(-) delete mode 100644 arch/arm64/kernel/probes/kprobes_trampoline.S create mode 100644 arch/arm64/kernel/rethook.c create mode 100644 arch/arm64/kernel/rethook_trampoline.S
arch/arm64/Kconfig +1 -0
@@ -204,6 +204,7 @@ config ARM64 select HAVE_SYSCALL_TRACEPOINTS select HAVE_KPROBES select HAVE_KRETPROBES + select HAVE_RETHOOK select HAVE_GENERIC_VDSO select IOMMU_DMA if IOMMU_SUPPORT select IRQ_DOMAIN
arch/arm64/include/asm/kprobes.h +0 -2
@@ -39,8 +39,6 @@ void arch_remove_kprobe(struct kprobe *);int kprobe_fault_handler (struct pt_regs *regs,unsigned int fsr);int kprobe_exceptions_notify (struct notifier_block *self,unsigned long val,void *data); -void __kretprobe_trampoline (void ); -void __kprobes *trampoline_probe_handler (struct pt_regs *regs);#endif /* CONFIG_KPROBES */ #endif /* _ARM_KPROBES_H */
arch/arm64/include/asm/stacktrace.h +1 -1
@@ -58,7 +58,7 @@ struct stackframe {DECLARE_BITMAP (stacks_done, __NR_STACK_TYPES);unsigned long prev_fp;enum stack_type prev_type; -#ifdef CONFIG_KRETPROBES +#if defined(CONFIG_RETHOOK )struct llist_node *kr_cur;#endif };
arch/arm64/kernel/Makefile +1 -0
@@ -60,6 +60,7 @@ obj-$(CONFIG_ACPI_NUMA) += acpi_numa.o obj-$(CONFIG_ARM64_ACPI_PARKING_PROTOCOL) += acpi_parking_protocol.o obj-$(CONFIG_PARAVIRT) += paravirt.o obj-$(CONFIG_RANDOMIZE_BASE) += kaslr.o +obj-$(CONFIG_RETHOOK) += rethook.o rethook_trampoline.o obj-$(CONFIG_HIBERNATION) += hibernate.o hibernate-asm.o obj-$(CONFIG_ELF_CORE) += elfcore.o obj-$(CONFIG_KEXEC_CORE) += machine_kexec.o relocate_kernel.o \
arch/arm64/kernel/probes/Makefile +0 -1
@@ -1,6 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 obj-$(CONFIG_KPROBES) += kprobes.o decode-insn.o \ - kprobes_trampoline.o \ simulate-insn.o obj-$(CONFIG_UPROBES) += uprobes.o decode-insn.o \ simulate-insn.o
arch/arm64/kernel/probes/kprobes.c +0 -15
@@ -399,21 +399,6 @@ int __init arch_populate_kprobe_blacklist(void)return ret; } -void __kprobes __used *trampoline_probe_handler (struct pt_regs *regs) -{ -return (void *)kretprobe_trampoline_handler (regs, (void *)regs->regs[29 ]); -} - -void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri, -struct pt_regs *regs) -{ - ri->ret_addr = (kprobe_opcode_t *)regs->regs[30 ]; - ri->fp = (void *)regs->regs[29 ]; - -/* replace return addr (x30) with trampoline */ - regs->regs[30 ] = (long )&__kretprobe_trampoline; -} -int __kprobes arch_trampoline_kprobe (struct kprobe *p) {return 0 ;
arch/arm64/kernel/probes/kprobes_trampoline.S deleted +0 -86
86 lines not on this page · raw message
arch/arm64/kernel/rethook.c new file +28 -0
@@ -0,0 +1,28 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Generic return hook for arm64. +* Most of the code is copied from arch/arm64/kernel/probes/kprobes.c +*/ + +#include <linux/kprobes.h> +#include <linux/rethook.h> + +/* This is called from arch_rethook_trampoline() */ +unsigned long __used arch_rethook_trampoline_callback (struct pt_regs *regs); + +unsigned long __used arch_rethook_trampoline_callback (struct pt_regs *regs) +{ +return rethook_trampoline_handler (regs, regs->regs[29 ]); +} +NOKPROBE_SYMBOL (arch_rethook_trampoline_callback); + +int arch_rethook_prepare (struct rethook_node *rhn,struct pt_regs *regs,bool mcount) +{ + rhn->ret_addr = regs->regs[30 ]; + rhn->frame = regs->regs[29 ]; + +/* replace return addr (x30) with trampoline */ + regs->regs[30 ] = (u64 )arch_rethook_trampoline; +return 0 ; +} +NOKPROBE_SYMBOL (arch_rethook_prepare);
arch/arm64/kernel/rethook_trampoline.S new file +87 -0
87 lines not on this page · raw message
arch/arm64/kernel/stacktrace.c +5 -4
@@ -8,6 +8,7 @@#include <linux/export.h> #include <linux/ftrace.h> #include <linux/kprobes.h> +#include <linux/rethook.h> #include <linux/sched.h> #include <linux/sched/debug.h> #include <linux/sched/task_stack.h>
@@ -38,7 +39,7 @@ static notrace void start_backtrace(struct stackframe *frame, unsigned long fp, { frame->fp = fp; frame->pc = pc; -#ifdef CONFIG_KRETPROBES +#if defined(CONFIG_RETHOOK ) frame->kr_cur =NULL ;#endif
@@ -134,9 +135,9 @@ static int notrace unwind_frame(struct task_struct *tsk, frame->pc = orig_pc; }#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ -#ifdef CONFIG_KRETPROBES -if (is_kretprobe_trampoline (frame->pc)) - frame->pc =kretprobe_find_ret_addr (tsk, (void *)frame->fp, &frame->kr_cur); +#ifdef CONFIG_RETHOOK +if (is_rethook_trampoline (frame->pc)) + frame->pc =rethook_find_ret_addr (tsk, frame->fp, &frame->kr_cur);#endif return 0 ;
Mark Rutland
reviewerRe: [PATCH bpf v2 4/4] arm64: rethook: Replace kretprobe trampoline with rethook
Quoted 218 lines · Masami Hiramatsu
Replace the kretprobe's trampoline code with the rethook on arm64. The rethook on arm64 is almost renamed from kretprobe trampoline code. The mechanism is completely same.
Signed-off-by Masami Hiramatsu <[email protected]>---
Changes in v2:
- Fix a compile warning about no prototype.--- arch/arm64/Kconfig | 1 arch/arm64/include/asm/kprobes.h | 2 - arch/arm64/include/asm/stacktrace.h | 2 - arch/arm64/kernel/Makefile | 1 arch/arm64/kernel/probes/Makefile | 1 arch/arm64/kernel/probes/kprobes.c | 15 ---- arch/arm64/kernel/probes/kprobes_trampoline.S | 86 ------------------------- arch/arm64/kernel/rethook.c | 28 ++++++++ arch/arm64/kernel/rethook_trampoline.S | 87 +++++++++++++++++++++++++ arch/arm64/kernel/stacktrace.c | 9 +-- 10 files changed, 123 insertions(+), 109 deletions(-) delete mode 100644 arch/arm64/kernel/probes/kprobes_trampoline.S create mode 100644 arch/arm64/kernel/rethook.c create mode 100644 arch/arm64/kernel/rethook_trampoline.Sarch/arm64/Kconfig +1
-0@@ -204,6 +204,7 @@ config ARM64 select HAVE_SYSCALL_TRACEPOINTS select HAVE_KPROBES select HAVE_KRETPROBES + select HAVE_RETHOOK select HAVE_GENERIC_VDSO select IOMMU_DMA if IOMMU_SUPPORT select IRQ_DOMAINarch/arm64/include/asm/kprobes.h +0
-2@@ -39,8 +39,6 @@ void arch_remove_kprobe(struct kprobe *);int kprobe_fault_handler (struct pt_regs *regs,unsigned int fsr);int kprobe_exceptions_notify (struct notifier_block *self,unsigned long val,void *data); -void __kretprobe_trampoline (void ); -void __kprobes *trampoline_probe_handler (struct pt_regs *regs);#endif /* CONFIG_KPROBES */ #endif /* _ARM_KPROBES_H */ arch/arm64/include/asm/stacktrace.h +1
-1@@ -58,7 +58,7 @@ struct stackframe {DECLARE_BITMAP (stacks_done, __NR_STACK_TYPES);unsigned long prev_fp;enum stack_type prev_type; -#ifdef CONFIG_KRETPROBES +#if defined(CONFIG_RETHOOK )struct llist_node *kr_cur;#endif };arch/arm64/kernel/Makefile +1
-0@@ -60,6 +60,7 @@ obj-$(CONFIG_ACPI_NUMA) += acpi_numa.o obj-$(CONFIG_ARM64_ACPI_PARKING_PROTOCOL) += acpi_parking_protocol.o obj-$(CONFIG_PARAVIRT) += paravirt.o obj-$(CONFIG_RANDOMIZE_BASE) += kaslr.o +obj-$(CONFIG_RETHOOK) += rethook.o rethook_trampoline.o obj-$(CONFIG_HIBERNATION) += hibernate.o hibernate-asm.o obj-$(CONFIG_ELF_CORE) += elfcore.o obj-$(CONFIG_KEXEC_CORE) += machine_kexec.o relocate_kernel.o \arch/arm64/kernel/probes/Makefile +0
-1@@ -1,6 +1,5 @@ # SPDX-License-Identifier: GPL-2.0 obj-$(CONFIG_KPROBES) += kprobes.o decode-insn.o \ - kprobes_trampoline.o \ simulate-insn.o obj-$(CONFIG_UPROBES) += uprobes.o decode-insn.o \ simulate-insn.oarch/arm64/kernel/probes/kprobes.c +0
-15@@ -399,21 +399,6 @@ int __init arch_populate_kprobe_blacklist(void)return ret; } -void __kprobes __used *trampoline_probe_handler (struct pt_regs *regs) -{ -return (void *)kretprobe_trampoline_handler (regs, (void *)regs->regs[29 ]); -} - -void __kprobes arch_prepare_kretprobe (struct kretprobe_instance *ri, -struct pt_regs *regs) -{ - ri->ret_addr = (kprobe_opcode_t *)regs->regs[30 ]; - ri->fp = (void *)regs->regs[29 ]; - -/* replace return addr (x30) with trampoline */ - regs->regs[30 ] = (long )&__kretprobe_trampoline; -} -int __kprobes arch_trampoline_kprobe (struct kprobe *p) {return 0 ;arch/arm64/kernel/probes/kprobes_trampoline.S deleted +0
-86@@ -1,86 +0,0 @@ -/* SPDX-License-Identifier: GPL-2.0 */ -/* - * trampoline entry and return code for kretprobes. - */ - -#include <linux/linkage.h> -#include <asm/asm-offsets.h> -#include <asm/assembler.h> - - .text - - .macro save_all_base_regs - stp x0, x1, [sp, #S_X0] - stp x2, x3, [sp, #S_X2] - stp x4, x5, [sp, #S_X4] - stp x6, x7, [sp, #S_X6] - stp x8, x9, [sp, #S_X8] - stp x10, x11, [sp, #S_X10] - stp x12, x13, [sp, #S_X12] - stp x14, x15, [sp, #S_X14] - stp x16, x17, [sp, #S_X16] - stp x18, x19, [sp, #S_X18] - stp x20, x21, [sp, #S_X20] - stp x22, x23, [sp, #S_X22] - stp x24, x25, [sp, #S_X24] - stp x26, x27, [sp, #S_X26] - stp x28, x29, [sp, #S_X28] - add x0, sp, #PT_REGS_SIZE - stp lr, x0, [sp, #S_LR] - /* - * Construct a useful saved PSTATE - */ - mrs x0, nzcv - mrs x1, daif - orr x0, x0, x1 - mrs x1, CurrentEL - orr x0, x0, x1 - mrs x1, SPSel - orr x0, x0, x1 - stp xzr, x0, [sp, #S_PC] - .endm - - .macro restore_all_base_regs - ldr x0, [sp, #S_PSTATE] - and x0, x0, #(PSR_N_BIT | PSR_Z_BIT | PSR_C_BIT | PSR_V_BIT) - msr nzcv, x0 - ldp x0, x1, [sp, #S_X0] - ldp x2, x3, [sp, #S_X2] - ldp x4, x5, [sp, #S_X4] - ldp x6, x7, [sp, #S_X6] - ldp x8, x9, [sp, #S_X8] - ldp x10, x11, [sp, #S_X10] - ldp x12, x13, [sp, #S_X12] - ldp x14, x15, [sp, #S_X14] - ldp x16, x17, [sp, #S_X16] - ldp x18, x19, [sp, #S_X18] - ldp x20, x21, [sp, #S_X20] - ldp x22, x23, [sp, #S_X22] - ldp x24, x25, [sp, #S_X24] - ldp x26, x27, [sp, #S_X26] - ldp x28, x29, [sp, #S_X28] - .endm - -SYM_CODE_START(__kretprobe_trampoline) - sub sp, sp, #PT_REGS_SIZE - - save_all_base_regs - - /* Setup a frame pointer. */ - add x29, sp, #S_FP - - mov x0, sp - bl trampoline_probe_handler - /* - * Replace trampoline address in lr with actual orig_ret_addr return - * address. - */ - mov lr, x0 - - /* The frame pointer (x29) is restored with other registers. */ - restore_all_base_regs - - add sp, sp, #PT_REGS_SIZE - ret - -SYM_CODE_END(__kretprobe_trampoline)arch/arm64/kernel/rethook.c new file +9
-0@@ -0,0 +1,28 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Generic return hook for arm64. +* Most of the code is copied from arch/arm64/kernel/probes/kprobes.c +*/ + +#include <linux/kprobes.h> +#include <linux/rethook.h> +
arch/arm64/kernel/rethook.c line 19
Quoted 10 lines
arch/arm64/kernel/rethook.c new file +10
-0+/* This is called from arch_rethook_trampoline() */ +unsigned long __used arch_rethook_trampoline_callback (struct pt_regs *regs); + +unsigned long __used arch_rethook_trampoline_callback (struct pt_regs *regs) +{ +return rethook_trampoline_handler (regs, regs->regs[29 ]); +} +NOKPROBE_SYMBOL (arch_rethook_trampoline_callback); + +int arch_rethook_prepare (struct rethook_node *rhn,struct pt_regs *regs,bool mcount)
What's the mcount paramater for, and why don't we seem to use it below?
The presence of that parameter suggests there is a subtle interaction with ftrace, but the commit message doesn't mention anything of the sort. I really worried that has implications for the unwinder.
I also see that (with these patches applied) it's possible to select CONFIG_FPROBE, but the commit message doesn't describe that either, and I'm not sure how to test any of that.
Quoted 45 lines
arch/arm64/kernel/rethook.c new file +9
-0+{ + rhn->ret_addr = regs->regs[30 ]; + rhn->frame = regs->regs[29 ]; + +/* replace return addr (x30) with trampoline */ + regs->regs[30 ] = (u64 )arch_rethook_trampoline; +return 0 ; +} +NOKPROBE_SYMBOL (arch_rethook_prepare);arch/arm64/kernel/rethook_trampoline.S new file +30
-0@@ -0,0 +1,87 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * trampoline entry and return code for rethook. + * Renamed from arch/arm64/kernel/probes/kprobes_trampoline.S + */ + +#include <linux/linkage.h> +#include <asm/asm-offsets.h> +#include <asm/assembler.h> + + .text + + .macro save_all_base_regs + stp x0, x1, [sp, #S_X0] + stp x2, x3, [sp, #S_X2] + stp x4, x5, [sp, #S_X4] + stp x6, x7, [sp, #S_X6] + stp x8, x9, [sp, #S_X8] + stp x10, x11, [sp, #S_X10] + stp x12, x13, [sp, #S_X12] + stp x14, x15, [sp, #S_X14] + stp x16, x17, [sp, #S_X16] + stp x18, x19, [sp, #S_X18] + stp x20, x21, [sp, #S_X20] + stp x22, x23, [sp, #S_X22] + stp x24, x25, [sp, #S_X24] + stp x26, x27, [sp, #S_X26] + stp x28, x29, [sp, #S_X28] + add x0, sp, #PT_REGS_SIZE + stp lr, x0, [sp, #S_LR]
arch/arm64/kernel/rethook_trampoline.S line 40
Quoted 10 lines
arch/arm64/kernel/rethook_trampoline.S new file +10
-0+ /* + * Construct a useful saved PSTATE + */ + mrs x0, nzcv + mrs x1, daif + orr x0, x0, x1 + mrs x1, CurrentEL + orr x0, x0, x1 + mrs x1, SPSel + orr x0, x0, x1
I realise this is just a copy of the existing kretprobe code, but it would be *really* nice if we could avoid faking the regs when we don't take an exception, like the FTRACE_WITH_ARGS thing.
The "useful saved PSTATE" is getting increasingly bogus these days, since it doesn't contain a bunch of values that are in the real PSTATE, and we don't restore anything other than NZCV anyway.
Quoted 69 lines
arch/arm64/kernel/rethook_trampoline.S new file +47
-0+ stp xzr, x0, [sp, #S_PC] + .endm + + .macro restore_all_base_regs + ldr x0, [sp, #S_PSTATE] + and x0, x0, #(PSR_N_BIT | PSR_Z_BIT | PSR_C_BIT | PSR_V_BIT) + msr nzcv, x0 + ldp x0, x1, [sp, #S_X0] + ldp x2, x3, [sp, #S_X2] + ldp x4, x5, [sp, #S_X4] + ldp x6, x7, [sp, #S_X6] + ldp x8, x9, [sp, #S_X8] + ldp x10, x11, [sp, #S_X10] + ldp x12, x13, [sp, #S_X12] + ldp x14, x15, [sp, #S_X14] + ldp x16, x17, [sp, #S_X16] + ldp x18, x19, [sp, #S_X18] + ldp x20, x21, [sp, #S_X20] + ldp x22, x23, [sp, #S_X22] + ldp x24, x25, [sp, #S_X24] + ldp x26, x27, [sp, #S_X26] + ldp x28, x29, [sp, #S_X28] + .endm + +SYM_CODE_START(arch_rethook_trampoline) + sub sp, sp, #PT_REGS_SIZE + + save_all_base_regs + + /* Setup a frame pointer. */ + add x29, sp, #S_FP + + mov x0, sp + bl arch_rethook_trampoline_callback + /* + * Replace trampoline address in lr with actual orig_ret_addr return + * address. + */ + mov lr, x0 + + /* The frame pointer (x29) is restored with other registers. */ + restore_all_base_regs + + add sp, sp, #PT_REGS_SIZE + ret + +SYM_CODE_END(arch_rethook_trampoline)arch/arm64/kernel/stacktrace.c +2
-1@@ -8,6 +8,7 @@#include <linux/export.h> #include <linux/ftrace.h> #include <linux/kprobes.h> +#include <linux/rethook.h> #include <linux/sched.h> #include <linux/sched/debug.h> #include <linux/sched/task_stack.h> @@ -38,7 +39,7 @@ static notrace void start_backtrace(struct stackframe *frame, unsigned long fp, { frame->fp = fp; frame->pc = pc; -#ifdef CONFIG_KRETPROBES +#if defined(CONFIG_RETHOOK ) frame->kr_cur =NULL ;#endif
arch/arm64/kernel/stacktrace.c static int notrace unwind_frame(struct task_struct *tsk, · line 141
Quoted 10 lines
arch/arm64/kernel/stacktrace.c +3
-3@@ -134,9 +135,9 @@ static int notrace unwind_frame(struct task_struct *tsk, frame->pc = orig_pc; }#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ -#ifdef CONFIG_KRETPROBES -if (is_kretprobe_trampoline (frame->pc)) - frame->pc =kretprobe_find_ret_addr (tsk, (void *)frame->fp, &frame->kr_cur); +#ifdef CONFIG_RETHOOK +if (is_rethook_trampoline (frame->pc)) + frame->pc =rethook_find_ret_addr (tsk, frame->fp, &frame->kr_cur);#endif
Is it possible to have an fprobe and a regular ftrace graph caller call on the same ftrace callsite? ... or are those mutually exclusive?
The existing code here depends on kretprobes and ftrace graph caller instrumentation nesting in a specific order, and I'm worried that might have changed.
Thanks,
Mark.
Masami Hiramatsu
maintainerRe: [PATCH bpf v2 4/4] arm64: rethook: Replace kretprobe trampoline with rethook
Hi Mark,
Thanks for the comment, and sorry about late reply.
Quoted 28 lines · On Tue, 12 Apr 2022 20:04:28 +0100 Mark Rutland <[email protected]> wrote:
Quoted 15 lines
arch/arm64/kernel/rethook.c new file +9
-0@@ -0,0 +1,28 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Generic return hook for arm64. +* Most of the code is copied from arch/arm64/kernel/probes/kprobes.c +*/ + +#include <linux/kprobes.h> +#include <linux/rethook.h> +arch/arm64/kernel/rethook.c line 19
Quoted 10 lines
arch/arm64/kernel/rethook.c new file +10
-0+/* This is called from arch_rethook_trampoline() */ +unsigned long __used arch_rethook_trampoline_callback (struct pt_regs *regs); + +unsigned long __used arch_rethook_trampoline_callback (struct pt_regs *regs) +{ +return rethook_trampoline_handler (regs, regs->regs[29 ]); +} +NOKPROBE_SYMBOL (arch_rethook_trampoline_callback); + +int arch_rethook_prepare (struct rethook_node *rhn,struct pt_regs *regs,bool mcount)What's the
mcountparamater for, and why don't we seem to use it below?
Quoted 3 lines
The presence of that parameter suggests there is a subtle interaction with ftrace, but the commit message doesn't mention anything of the sort. I really worried that has implications for the unwinder.
mcount parameter is introduced only for arm arch at this point. Other arch will ignore this parameter. See [3/4] for details. When the rethook caller (ftrace or kprobe) hooks the function entry, how to get the real return address depends on whether it is called from ftrace (mcount) or kprobe.
By kprobe, that is called from the first instruction of the function. Thus the real return address can be found from regs->lr. But from the ftrace (mcount) context, we can not find it from regs->lr (this LR register is used for saving the return address of mcount itself) on arm. On arm64, this seems to be fixed. So I'm not sure this is a real issue on arm or known limitation.
Quoted 4 lines
I also see that (with these patches applied) it's possible to select
CONFIG_FPROBE, but the commit message doesn't describe that either, and I'm not sure how to test any of that.
Ah, Sorry. Fprobe requires this feature, so HAVE_RETHOOK=y, CONFIG_FPROBE is available on this architecture. If you enable CONFIG_FPROBE_SANITY_TEST=y, this feature is tested by fprobe.
Quoted 52 lines
Quoted 52 lines
arch/arm64/kernel/rethook.c new file +9
-0+{ + rhn->ret_addr = regs->regs[30 ]; + rhn->frame = regs->regs[29 ]; + +/* replace return addr (x30) with trampoline */ + regs->regs[30 ] = (u64 )arch_rethook_trampoline; +return 0 ; +} +NOKPROBE_SYMBOL (arch_rethook_prepare);arch/arm64/kernel/rethook_trampoline.S new file +36
-0@@ -0,0 +1,87 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * trampoline entry and return code for rethook. + * Renamed from arch/arm64/kernel/probes/kprobes_trampoline.S + */ + +#include <linux/linkage.h> +#include <asm/asm-offsets.h> +#include <asm/assembler.h> + + .text + + .macro save_all_base_regs + stp x0, x1, [sp, #S_X0] + stp x2, x3, [sp, #S_X2] + stp x4, x5, [sp, #S_X4] + stp x6, x7, [sp, #S_X6] + stp x8, x9, [sp, #S_X8] + stp x10, x11, [sp, #S_X10] + stp x12, x13, [sp, #S_X12] + stp x14, x15, [sp, #S_X14] + stp x16, x17, [sp, #S_X16] + stp x18, x19, [sp, #S_X18] + stp x20, x21, [sp, #S_X20] + stp x22, x23, [sp, #S_X22] + stp x24, x25, [sp, #S_X24] + stp x26, x27, [sp, #S_X26] + stp x28, x29, [sp, #S_X28] + add x0, sp, #PT_REGS_SIZE + stp lr, x0, [sp, #S_LR] + /* + * Construct a useful saved PSTATE + */ + mrs x0, nzcv + mrs x1, daif + orr x0, x0, x1
Quoted 8 lines
arch/arm64/kernel/rethook_trampoline.S line 40
Quoted 4 lines
arch/arm64/kernel/rethook_trampoline.S new file +4
-0+ mrs x1, CurrentEL + orr x0, x0, x1 + mrs x1, SPSel + orr x0, x0, x1I realise this is just a copy of the existing kretprobe code, but it would be *really* nice if we could avoid faking the regs when we don't take an exception, like the
FTRACE_WITH_ARGSthing.
No, since this is not a "function call". This is function exit, thus the all registers can be involved.
Quoted 4 lines
The "useful saved PSTATE" is getting increasingly bogus these days, since it doesn't contain a bunch of values that are in the real PSTATE, and we don't restore anything other than NZCV anyway.
Hmm, that is a real bug. It should be fixed even without this patch, on kretprobes. Are there any good way to store those flags in the real PSTATE?
I need arm64 maintainer's help to fix that. If there is a limitation of getting the PSTATE, I need to decide to
- use kprobe instead of hand-assembly code (like powerpc) to hook the trampoline. or
- just notice users this is a limitation of kretprobe on arm64.
Quoted 69 lines
Quoted 69 lines
arch/arm64/kernel/rethook_trampoline.S new file +47
-0+ stp xzr, x0, [sp, #S_PC] + .endm + + .macro restore_all_base_regs + ldr x0, [sp, #S_PSTATE] + and x0, x0, #(PSR_N_BIT | PSR_Z_BIT | PSR_C_BIT | PSR_V_BIT) + msr nzcv, x0 + ldp x0, x1, [sp, #S_X0] + ldp x2, x3, [sp, #S_X2] + ldp x4, x5, [sp, #S_X4] + ldp x6, x7, [sp, #S_X6] + ldp x8, x9, [sp, #S_X8] + ldp x10, x11, [sp, #S_X10] + ldp x12, x13, [sp, #S_X12] + ldp x14, x15, [sp, #S_X14] + ldp x16, x17, [sp, #S_X16] + ldp x18, x19, [sp, #S_X18] + ldp x20, x21, [sp, #S_X20] + ldp x22, x23, [sp, #S_X22] + ldp x24, x25, [sp, #S_X24] + ldp x26, x27, [sp, #S_X26] + ldp x28, x29, [sp, #S_X28] + .endm + +SYM_CODE_START(arch_rethook_trampoline) + sub sp, sp, #PT_REGS_SIZE + + save_all_base_regs + + /* Setup a frame pointer. */ + add x29, sp, #S_FP + + mov x0, sp + bl arch_rethook_trampoline_callback + /* + * Replace trampoline address in lr with actual orig_ret_addr return + * address. + */ + mov lr, x0 + + /* The frame pointer (x29) is restored with other registers. */ + restore_all_base_regs + + add sp, sp, #PT_REGS_SIZE + ret + +SYM_CODE_END(arch_rethook_trampoline)arch/arm64/kernel/stacktrace.c +2
-1@@ -8,6 +8,7 @@#include <linux/export.h> #include <linux/ftrace.h> #include <linux/kprobes.h> +#include <linux/rethook.h> #include <linux/sched.h> #include <linux/sched/debug.h> #include <linux/sched/task_stack.h> @@ -38,7 +39,7 @@ static notrace void start_backtrace(struct stackframe *frame, unsigned long fp, { frame->fp = fp; frame->pc = pc; -#ifdef CONFIG_KRETPROBES +#if defined(CONFIG_RETHOOK ) frame->kr_cur =NULL ;#endif
Quoted 14 lines
arch/arm64/kernel/stacktrace.c static int notrace unwind_frame(struct task_struct *tsk, · line 141
Quoted 11 lines
arch/arm64/kernel/stacktrace.c +3
-3@@ -134,9 +135,9 @@ static int notrace unwind_frame(struct task_struct *tsk, frame->pc = orig_pc; }#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ -#ifdef CONFIG_KRETPROBES -if (is_kretprobe_trampoline (frame->pc)) - frame->pc =kretprobe_find_ret_addr (tsk, (void *)frame->fp, &frame->kr_cur); +#ifdef CONFIG_RETHOOK +if (is_rethook_trampoline (frame->pc)) + frame->pc =rethook_find_ret_addr (tsk, frame->fp, &frame->kr_cur);#endif Is it possible to have an fprobe and a regular ftrace graph caller call on the same ftrace callsite? ... or are those mutually exclusive?
Hmm, those are too different at this moment, but I would like to do that. What I would like to do is
- if the arch has no ftrace-graph implementation, rethook will use the kretprobe trampoline based implementation. This can drop some events because of the resource limitation.
- else, it will use ftrace-graph implementation. This can be very effective in some case (like tracing frequently called function) because it uses par-task shadow stack. But not many arch implemented it.
- Also, the interface needs to be unified. Rethook API is intended to be used for the integration. kretprobe and fprobe will use the rethook, thus the pt_regs is required.
Quoted 4 lines
The existing code here depends on kretprobes and ftrace graph caller instrumentation nesting in a specific order, and I'm worried that might have changed.
Would you mean that the "nesting" is using kretprobe and ftrace-graph on same function, correct?
At this moment, this rethook (kretprobe and fprobe) does not use the same shadow stack of the ftrace graph caller, so the nesting must work correctly.
Thank you,
Quoted 3 lines
Thanks,
Mark.
Signature · 1 line
Masami Hiramatsu <[email protected]>
Mark Rutland
reviewerQuoted 1 line · Masami Hiramatsu
Hi,
Hi,
Quoted 2 lines
Here is the 2nd version of the series for replacing kretprobe trampoline with rethook on ARM/arm64. I fixed some compiler warnings in this version.
What tree is this based on? It doesn't cleanly apply atop v5.18-rc1:
| [mark@lakrids:~/src/linux]% git am v2_20220408_mhiramat_kprobes_rethook_arm_arm64_replace_kretprobe_trampoline_with_rethook.mbx | Applying: ARM: unwind: Initialize the lr_addr field of unwind_ctrl_block | Applying: rethook,fprobe,kprobes: Check a failure in the rethook_hook() | Applying: ARM: rethook: Replace kretprobe trampoline with rethook | error: patch failed: arch/arm/kernel/stacktrace.c:66 | error: arch/arm/kernel/stacktrace.c: patch does not apply | Patch failed at 0003 ARM: rethook: Replace kretprobe trampoline with rethook | hint: Use 'git am --show-current-patch=diff' to see the failed patch | When you have resolved this problem, run "git am --continue". | If you prefer to skip this patch, run "git am --skip" instead. | To restore the original branch and stop patching, run "git am --abort".
I've done a git am -3 locally to make that work for now.
Quoted 4 lines
The previous version is here[1];
[1] https://lore.kernel.org/all/164915121498.982637.12787715964983738566.stgit@devnote2/T/#u
Quoted 8 lines
This series includes a trivial bugfix for the arm unwinder to initialize an internal data
structure([1/4]). This is not critical for stack trace, but required for rethook to find the LR register from the stack. This also have an update for the rethook interface, which allows us to check therethook_hook()failure ([2/4]). This is also required for the rethook on arm because unwinder is able to fail.
The rest of patches are replacing kretprobe trampoline with rethook on ARM ([3/4]) and arm64 ([4/4]).
Generally, the arm and arm64 bits go via different trees, and for unwinding the two are quite different.
IIUC the dependency between the two is just because patch 2 changes the prototypes of some functions. Is that right?
Quoted 24 lines
Background:
This rethook came from Jiri's request of multiple kprobe for bpf[2]. He tried to solve an issue that starting bpf with multiple kprobe will take a long time because bpf-kprobe will wait for RCU grace period for sync rcu events.
Jiri wanted to attach a single bpf handler to multiple kprobes and he tried to introduce multiple-probe interface to kprobe. So I asked him to use ftrace and kretprobe-like hook if it is only for the function entry and exit, instead of adding ad-hoc interface to kprobes.
For this purpose, I introduced the fprobe (kprobe like interface for ftrace) with the rethook (this is a generic return hook feature for fprobe exit handler)[3].[2] https://lore.kernel.org/all/[email protected]/T/#u
[3] https://lore.kernel.org/all/164191321766.806991.7930388561276940676.stgit@devnote2/T/#uThe rethook is basically same as the kretprobe trampoline. I just made it decoupled from kprobes. Eventually, the all arch dependent kretprobe trampolines will be replaced with the rethook trampoline instead of cloning the code.
Quoted 6 lines
When I port the rethook for all arch which supports kretprobe, the legacy kretprobe specific code (which is for
CONFIG_KRETPROBE_ON_RETHOOK=n) will be removed eventually.BTW, the arm Clang support for rethook is for kretprobes only. fprobe and ftrace seems not working with Clang yet.
Do you mean that's an existing issue?
Thanks,
Mark.
Quoted 42 lines
Thank you,
---
Masami Hiramatsu (4):
ARM: unwind: Initialize thelr_addrfield ofunwind_ctrl_blockrethook,fprobe,kprobes: Check a failure in therethook_hook()
ARM: rethook: Replace kretprobe trampoline with rethook
arm64: rethook: Replace kretprobe trampoline with rethookarch/arm/Kconfig | 1 arch/arm/include/asm/stacktrace.h | 5 + arch/arm/kernel/stacktrace.c | 13 +-- arch/arm/kernel/unwind.c | 1 arch/arm/probes/Makefile | 1 arch/arm/probes/kprobes/core.c | 62 ------------ arch/arm/probes/rethook.c | 127 +++++++++++++++++++++++++ arch/arm64/Kconfig | 1 arch/arm64/include/asm/kprobes.h | 2 arch/arm64/include/asm/stacktrace.h | 2 arch/arm64/kernel/Makefile | 1 arch/arm64/kernel/probes/Makefile | 1 arch/arm64/kernel/probes/kprobes.c | 15 --- arch/arm64/kernel/probes/kprobes_trampoline.S | 86 ----------------- arch/arm64/kernel/rethook.c | 28 ++++++ arch/arm64/kernel/rethook_trampoline.S | 87 +++++++++++++++++ arch/arm64/kernel/stacktrace.c | 9 +- arch/x86/kernel/rethook.c | 4 + include/linux/rethook.h | 4 - kernel/kprobes.c | 8 +- kernel/trace/fprobe.c | 5 + kernel/trace/rethook.c | 12 ++ 22 files changed, 287 insertions(+), 188 deletions(-) create mode 100644 arch/arm/probes/rethook.c delete mode 100644 arch/arm64/kernel/probes/kprobes_trampoline.S create mode 100644 arch/arm64/kernel/rethook.c create mode 100644 arch/arm64/kernel/rethook_trampoline.S--
Masami Hiramatsu (Linaro) <[email protected]>
Masami Hiramatsu
maintainerHi Mark,
Sorry for replying this late.
Quoted 9 lines · On Tue, 12 Apr 2022 19:03:53 +0100 Mark Rutland <[email protected]> wrote:
Quoted 1 line · Masami Hiramatsu
Hi,
Hi,
Quoted 2 lines
Here is the 2nd version of the series for replacing kretprobe trampoline with rethook on ARM/arm64. I fixed some compiler warnings in this version.
What tree is this based on? It doesn't cleanly apply atop v5.18-rc1:
I worked on bpf tree, but I'll try rebasing on the latest linus tree (or arm/arm64 tree) in the next version.
Quoted 28 lines
| [mark@lakrids:~/src/linux]% git am v2_20220408_mhiramat_kprobes_rethook_arm_arm64_replace_kretprobe_trampoline_with_rethook.mbx | Applying: ARM: unwind: Initialize the lr_addr field of unwind_ctrl_block | Applying: rethook,fprobe,kprobes: Check a failure in the rethook_hook() | Applying: ARM: rethook: Replace kretprobe trampoline with rethook | error: patch failed: arch/arm/kernel/stacktrace.c:66 | error: arch/arm/kernel/stacktrace.c: patch does not apply | Patch failed at 0003 ARM: rethook: Replace kretprobe trampoline with rethook | hint: Use 'git am --show-current-patch=diff' to see the failed patch | When you have resolved this problem, run "git am --continue". | If you prefer to skip this patch, run "git am --skip" instead. | To restore the original branch and stop patching, run "git am --abort".I've done a
git am -3locally to make that work for now.Quoted 4 lines
The previous version is here[1];
[1] https://lore.kernel.org/all/164915121498.982637.12787715964983738566.stgit@devnote2/T/#u
Quoted 8 lines
This series includes a trivial bugfix for the arm unwinder to initialize an internal data
structure([1/4]). This is not critical for stack trace, but required for rethook to find the LR register from the stack. This also have an update for the rethook interface, which allows us to check therethook_hook()failure ([2/4]). This is also required for the rethook on arm because unwinder is able to fail.
The rest of patches are replacing kretprobe trampoline with rethook on ARM ([3/4]) and arm64 ([4/4]).
Quoted 2 lines
Generally, the arm and arm64 bits go via different trees, and for unwinding the two are quite different.
OK, Should I split this into 2 series?
Anyway, I'll send the first one for arm tree because this is a real bug.
Quoted 3 lines
IIUC the dependency between the two is just because patch 2 changes the prototypes of some functions. Is that right?
Yes. I can push it one by one. What is the best way, would you think? If I split this into some series, I will;
- Send [1/4] to arm tree as a bugfix.
- Send [2/4] and [3/4] to arm tree as an enhancement.
- Send [4/4] to arm64 when above 2 series are merged.
Is that good for you?
Quoted 26 lines
Quoted 26 lines
Background:
This rethook came from Jiri's request of multiple kprobe for bpf[2]. He tried to solve an issue that starting bpf with multiple kprobe will take a long time because bpf-kprobe will wait for RCU grace period for sync rcu events.
Jiri wanted to attach a single bpf handler to multiple kprobes and he tried to introduce multiple-probe interface to kprobe. So I asked him to use ftrace and kretprobe-like hook if it is only for the function entry and exit, instead of adding ad-hoc interface to kprobes.
For this purpose, I introduced the fprobe (kprobe like interface for ftrace) with the rethook (this is a generic return hook feature for fprobe exit handler)[3].[2] https://lore.kernel.org/all/[email protected]/T/#u
[3] https://lore.kernel.org/all/164191321766.806991.7930388561276940676.stgit@devnote2/T/#uThe rethook is basically same as the kretprobe trampoline. I just made it decoupled from kprobes. Eventually, the all arch dependent kretprobe trampolines will be replaced with the rethook trampoline instead of cloning the code.
Quoted 8 lines
Quoted 6 lines
When I port the rethook for all arch which supports kretprobe, the legacy kretprobe specific code (which is for
CONFIG_KRETPROBE_ON_RETHOOK=n) will be removed eventually.BTW, the arm Clang support for rethook is for kretprobes only. fprobe and ftrace seems not working with Clang yet.
Do you mean that's an existing issue?
Yes, but the above is not enough, Clang on arm and dynamic ftrace doesn't work. Perhaps, Clang on arm doesn't support -mrecord-mcount?
Thank you,
Quoted 46 lines
Thanks,
Mark.Quoted 42 lines
Thank you,
---
Masami Hiramatsu (4):
ARM: unwind: Initialize thelr_addrfield ofunwind_ctrl_blockrethook,fprobe,kprobes: Check a failure in therethook_hook()
ARM: rethook: Replace kretprobe trampoline with rethook
arm64: rethook: Replace kretprobe trampoline with rethookarch/arm/Kconfig | 1 arch/arm/include/asm/stacktrace.h | 5 + arch/arm/kernel/stacktrace.c | 13 +-- arch/arm/kernel/unwind.c | 1 arch/arm/probes/Makefile | 1 arch/arm/probes/kprobes/core.c | 62 ------------ arch/arm/probes/rethook.c | 127 +++++++++++++++++++++++++ arch/arm64/Kconfig | 1 arch/arm64/include/asm/kprobes.h | 2 arch/arm64/include/asm/stacktrace.h | 2 arch/arm64/kernel/Makefile | 1 arch/arm64/kernel/probes/Makefile | 1 arch/arm64/kernel/probes/kprobes.c | 15 --- arch/arm64/kernel/probes/kprobes_trampoline.S | 86 ----------------- arch/arm64/kernel/rethook.c | 28 ++++++ arch/arm64/kernel/rethook_trampoline.S | 87 +++++++++++++++++ arch/arm64/kernel/stacktrace.c | 9 +- arch/x86/kernel/rethook.c | 4 + include/linux/rethook.h | 4 - kernel/kprobes.c | 8 +- kernel/trace/fprobe.c | 5 + kernel/trace/rethook.c | 12 ++ 22 files changed, 287 insertions(+), 188 deletions(-) create mode 100644 arch/arm/probes/rethook.c delete mode 100644 arch/arm64/kernel/probes/kprobes_trampoline.S create mode 100644 arch/arm64/kernel/rethook.c create mode 100644 arch/arm64/kernel/rethook_trampoline.S--
Masami Hiramatsu (Linaro) <[email protected]>
Signature · 1 line
Masami Hiramatsu <[email protected]>