Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753895AbcCHFnI (ORCPT ); Tue, 8 Mar 2016 00:43:08 -0500 Received: from mail-pa0-f43.google.com ([209.85.220.43]:34830 "EHLO mail-pa0-f43.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753488AbcCHFnA (ORCPT ); Tue, 8 Mar 2016 00:43:00 -0500 From: David Long Subject: Re: [PATCH v10 7/9] arm64: Add trampoline code for kretprobes To: William Cohen , Marc Zyngier , Catalin Marinas , Will Deacon , Sandeepa Prabhu , Pratyush Anand , Steve Capper , linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org References: <1456801047-29014-1-git-send-email-dave.long@linaro.org> <1456801047-29014-8-git-send-email-dave.long@linaro.org> <56D5DD27.6040404@arm.com> <56D7592F.70002@redhat.com> Cc: Dave P Martin , Mark Rutland , Robin Murphy , Ard Biesheuvel , Jens Wiklander , Christoffer Dall , =?UTF-8?Q?Alex_Benn=c3=a9e?= , Yang Shi , Greg Kroah-Hartman , Viresh Kumar , "Suzuki K. Poulose" , Kees Cook , Zi Shen Lim , John Blackwood , Feng Kan , Balamurugan Shanmugam , James Morse , Vladimir Murzin , Mark Salyzyn , Petr Mladek , Andrew Morton , Mark Brown Message-ID: <56DE665B.8080209@linaro.org> Date: Tue, 8 Mar 2016 00:42:51 -0500 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101 Thunderbird/38.3.0 MIME-Version: 1.0 In-Reply-To: <56D7592F.70002@redhat.com> Content-Type: text/plain; charset=windows-1252; format=flowed Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 8348 Lines: 247 On 03/02/2016 04:20 PM, William Cohen wrote: > On 03/01/2016 01:19 PM, Marc Zyngier wrote: >> On 01/03/16 02:57, David Long wrote: >>> From: William Cohen >>> >>> The trampoline code is used by kretprobes to capture a return from a probed >>> function. This is done by saving the registers, calling the handler, and >>> restoring the registers. The code then returns to the original saved caller >>> return address. It is necessary to do this directly instead of using a >>> software breakpoint because the code used in processing that breakpoint >>> could itself be kprobe'd and cause a problematic reentry into the debug >>> exception handler. >>> >>> Signed-off-by: William Cohen >>> Signed-off-by: David A. Long >>> --- >>> arch/arm64/include/asm/kprobes.h | 2 + >>> arch/arm64/kernel/Makefile | 1 + >>> arch/arm64/kernel/asm-offsets.c | 22 +++++++++++ >>> arch/arm64/kernel/kprobes.c | 5 +++ >>> arch/arm64/kernel/kprobes_trampoline.S | 67 ++++++++++++++++++++++++++++++++++ >>> 5 files changed, 97 insertions(+) >>> create mode 100644 arch/arm64/kernel/kprobes_trampoline.S >>> >>> diff --git a/arch/arm64/include/asm/kprobes.h b/arch/arm64/include/asm/kprobes.h >>> index 79c9511..61b4915 100644 >>> --- a/arch/arm64/include/asm/kprobes.h >>> +++ b/arch/arm64/include/asm/kprobes.h >>> @@ -56,5 +56,7 @@ int kprobe_exceptions_notify(struct notifier_block *self, >>> unsigned long val, void *data); >>> int kprobe_breakpoint_handler(struct pt_regs *regs, unsigned int esr); >>> int kprobe_single_step_handler(struct pt_regs *regs, unsigned int esr); >>> +void kretprobe_trampoline(void); >>> +void __kprobes *trampoline_probe_handler(struct pt_regs *regs); >>> >>> #endif /* _ARM_KPROBES_H */ >>> diff --git a/arch/arm64/kernel/Makefile b/arch/arm64/kernel/Makefile >>> index 08325e5..f192b7d 100644 >>> --- a/arch/arm64/kernel/Makefile >>> +++ b/arch/arm64/kernel/Makefile >>> @@ -37,6 +37,7 @@ arm64-obj-$(CONFIG_CPU_IDLE) += cpuidle.o >>> arm64-obj-$(CONFIG_JUMP_LABEL) += jump_label.o >>> arm64-obj-$(CONFIG_KGDB) += kgdb.o >>> arm64-obj-$(CONFIG_KPROBES) += kprobes.o kprobes-arm64.o \ >>> + kprobes_trampoline.o \ >>> probes-simulate-insn.o >>> arm64-obj-$(CONFIG_EFI) += efi.o efi-entry.stub.o >>> arm64-obj-$(CONFIG_PCI) += pci.o >>> diff --git a/arch/arm64/kernel/asm-offsets.c b/arch/arm64/kernel/asm-offsets.c >>> index fffa4ac6..460b54c 100644 >>> --- a/arch/arm64/kernel/asm-offsets.c >>> +++ b/arch/arm64/kernel/asm-offsets.c >>> @@ -50,6 +50,28 @@ int main(void) >>> DEFINE(S_X5, offsetof(struct pt_regs, regs[5])); >>> DEFINE(S_X6, offsetof(struct pt_regs, regs[6])); >>> DEFINE(S_X7, offsetof(struct pt_regs, regs[7])); >>> + DEFINE(S_X8, offsetof(struct pt_regs, regs[8])); >>> + DEFINE(S_X9, offsetof(struct pt_regs, regs[9])); >>> + DEFINE(S_X10, offsetof(struct pt_regs, regs[10])); >>> + DEFINE(S_X11, offsetof(struct pt_regs, regs[11])); >>> + DEFINE(S_X12, offsetof(struct pt_regs, regs[12])); >>> + DEFINE(S_X13, offsetof(struct pt_regs, regs[13])); >>> + DEFINE(S_X14, offsetof(struct pt_regs, regs[14])); >>> + DEFINE(S_X15, offsetof(struct pt_regs, regs[15])); >>> + DEFINE(S_X16, offsetof(struct pt_regs, regs[16])); >>> + DEFINE(S_X17, offsetof(struct pt_regs, regs[17])); >>> + DEFINE(S_X18, offsetof(struct pt_regs, regs[18])); >>> + DEFINE(S_X19, offsetof(struct pt_regs, regs[19])); >>> + DEFINE(S_X20, offsetof(struct pt_regs, regs[20])); >>> + DEFINE(S_X21, offsetof(struct pt_regs, regs[21])); >>> + DEFINE(S_X22, offsetof(struct pt_regs, regs[22])); >>> + DEFINE(S_X23, offsetof(struct pt_regs, regs[23])); >>> + DEFINE(S_X24, offsetof(struct pt_regs, regs[24])); >>> + DEFINE(S_X25, offsetof(struct pt_regs, regs[25])); >>> + DEFINE(S_X26, offsetof(struct pt_regs, regs[26])); >>> + DEFINE(S_X27, offsetof(struct pt_regs, regs[27])); >>> + DEFINE(S_X28, offsetof(struct pt_regs, regs[28])); >>> + DEFINE(S_X29, offsetof(struct pt_regs, regs[29])); >> >> Do we need all of these? Specially considering that we're only using the >> even ones? You may want to consider something like >> arch/arm64/kvm/hyp/entry.S does. > > Following what arch/arm64/kvm/hyp/entry.S does would make the patch a bit smaller. > I have cut the defines in half and duplicated the entry.S style using macros. >> >>> DEFINE(S_LR, offsetof(struct pt_regs, regs[30])); >>> DEFINE(S_SP, offsetof(struct pt_regs, sp)); >>> #ifdef CONFIG_COMPAT >>> diff --git a/arch/arm64/kernel/kprobes.c b/arch/arm64/kernel/kprobes.c >>> index ffc5affd..98f4fe5 100644 >>> --- a/arch/arm64/kernel/kprobes.c >>> +++ b/arch/arm64/kernel/kprobes.c >>> @@ -532,6 +532,11 @@ int __kprobes longjmp_break_handler(struct kprobe *p, struct pt_regs *regs) >>> return 1; >>> } >>> >>> +void __kprobes __used *trampoline_probe_handler(struct pt_regs *regs) >>> +{ >>> + return (void *) 0; >> >> Something wrong with NULL? ;-) > > Nothing wrong with using NULL. > Changed. >> >>> +} >>> + >>> int __init arch_init_kprobes(void) >>> { >>> return 0; >>> diff --git a/arch/arm64/kernel/kprobes_trampoline.S b/arch/arm64/kernel/kprobes_trampoline.S >>> new file mode 100644 >>> index 0000000..5a336cf >>> --- /dev/null >>> +++ b/arch/arm64/kernel/kprobes_trampoline.S >>> @@ -0,0 +1,67 @@ >>> +/* >>> + * trampoline entry and return code for kretprobes. >>> + */ >>> + >>> +#include >>> +#include >> >> #include >> Fixed. >>> + >>> + .text >>> + >>> +ENTRY(kretprobe_trampoline) >>> + >>> + sub sp, sp, #S_FRAME_SIZE >>> + >>> + 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] >>> + str x30, [sp, #S_LR] >> >> Might as well call it LR? Ah, that exposed the fact there's a missing include of asm/assembler.h. All fixed. > > Given the code is going to overwrite lr with the value returned by trampoline_probe_handler might eliminate the store to #S_LR above. > It's not only about what's restored though, it's about what the trace/user code is told the register contents are. >> >>> + add x0, sp, #S_FRAME_SIZE >>> + str x0, [sp, #S_SP] >>> + mrs x0, nzcv >>> + mrs x1, daif >>> + orr x0, x0, x1 >>> + /* There seems no easy way to get the mode field so make one up */ >>> + add x0, x0, #5 >> >> Do you mean something like CurrentEL? You could also save SPSel whilst >> you're at it. OK, I've retrieved CurrentEL and SPSel and orr'd them into the saved PSTATE. > > Looking at the my patch again I am wondering if the patch could skip storing daif and the mode field. The nzcv bit are the only thing restored. > Again, want to have realistic saved register contents. >> >>> + str x0, [sp, #S_PSTATE] >>> + >>> + mov x0, sp >>> + bl trampoline_probe_handler >>> + /* Replace trampoline address in lr with actual >>> + orig_ret_addr return address. */ >>> + str x0, [sp, #S_LR] >> >> Why do you need to store it on the stack? You could do a "mov lr, x0", >> and drop the last load of the sequence below... > > Ah, yes, that would save a store/load pair. > Done. >>> + >>> + ldr x0, [sp, #S_PSTATE] >>> + 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] >>> + ldr x30, [sp, #S_LR] >>> + >>> + add sp, sp, #S_FRAME_SIZE >>> + ret >>> + >>> +ENDPROC(kretprobe_trampoline) >>> >> >> Thanks, >> >> M. >> > > Thanks, > > -Will > -dl