Re: [PATCH v12 05/10] arm64: Kprobes with single stepping support

[Masami Hiramatsu <mhiramat@kernel.org>]

On Thu, 12 May 2016 16:01:54 +0100
James Morse <james.morse@arm.com> wrote:

> Hi David, Sandeepa,
> 
> On 27/04/16 19:53, David Long wrote:
> > From: Sandeepa Prabhu <sandeepa.s.prabhu@gmail.com>
> 
> > diff --git a/arch/arm64/kernel/kprobes.c b/arch/arm64/kernel/kprobes.c
> > new file mode 100644
> > index 0000000..dfa1b1f
> > --- /dev/null
> > +++ b/arch/arm64/kernel/kprobes.c
> > @@ -0,0 +1,520 @@
> > +/*
> > + * arch/arm64/kernel/kprobes.c
> > + *
> > + * Kprobes support for ARM64
> > + *
> > + * Copyright (C) 2013 Linaro Limited.
> > + * Author: Sandeepa Prabhu <sandeepa.prabhu@linaro.org>
> > + *
> > + * This program is free software; you can redistribute it and/or modify
> > + * it under the terms of the GNU General Public License version 2 as
> > + * published by the Free Software Foundation.
> > + *
> > + * This program is distributed in the hope that it will be useful,
> > + * but WITHOUT ANY WARRANTY; without even the implied warranty of
> > + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
> > + * General Public License for more details.
> > + *
> > + */
> > +#include <linux/kernel.h>
> > +#include <linux/kprobes.h>
> > +#include <linux/module.h>
> > +#include <linux/slab.h>
> > +#include <linux/stop_machine.h>
> > +#include <linux/stringify.h>
> > +#include <asm/traps.h>
> > +#include <asm/ptrace.h>
> > +#include <asm/cacheflush.h>
> > +#include <asm/debug-monitors.h>
> > +#include <asm/system_misc.h>
> > +#include <asm/insn.h>
> > +#include <asm/uaccess.h>
> > +
> > +#include "kprobes-arm64.h"
> > +
> > +#define MIN_STACK_SIZE(addr)	min((unsigned long)MAX_STACK_SIZE,	\
> > +	(unsigned long)current_thread_info() + THREAD_START_SP - (addr))
> 
> What if we probe something called on the irq stack?
> This needs the on_irq_stack() checks too, the start/end can be found from the
> per-cpu irq_stack value.
> 
> [ ... ]
> 
> > +int __kprobes setjmp_pre_handler(struct kprobe *p, struct pt_regs *regs)
> > +{
> > +	struct jprobe *jp = container_of(p, struct jprobe, kp);
> > +	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
> > +	long stack_ptr = kernel_stack_pointer(regs);
> > +
> > +	kcb->jprobe_saved_regs = *regs;
> > +	memcpy(kcb->jprobes_stack, (void *)stack_ptr,
> > +	       MIN_STACK_SIZE(stack_ptr));
> 
> I wonder if we need this stack save/restore?
> 
> The comment next to the equivalent code for x86 says:
> > gcc assumes that the callee owns the argument space and could overwrite it,
> > e.g. tailcall optimization. So, to be absolutely safe we also save and
> > restore enough stack bytes to cover the argument area.
> 
> On arm64 the first eight arguments are passed in registers, so we might not need
> this stack copy. (sparc and powerpc work like this too, their versions of this
> function don't copy chunks of the stack).

Hmm, maybe sparc and powerpc implementation should also be fixed...

> ... then I went looking for functions with >8 arguments...
> 
> Looking at the arm64 defconfig dwarf debug data, there are 71 of these that
> don't get inlined, picking at random:
> > rockchip_clk_register_pll() has 13
> > fib_dump_info() has 11
> > vma_merge() has 10
> > vring_create_virtqueue() has 10
> etc...
> 
> So we do need this stack copying, so that we can probe these function without
> risking the arguments being modified.
> 
> It may be worth including a comment to the effect that this stack save/restore
> is needed for functions that pass >8 arguments where the pre-handler may change
> these values on the stack.

Indeed, commenting on this code can help us to understand the reason why.

Thank you!

> 
> 
> > +	preempt_enable_no_resched();
> > +	return 1;
> > +}
> > +
> 
> 
> Thanks,
> 
> James


-- 
Masami Hiramatsu <mhiramat@kernel.org>