Re: [PATCH 2/4] arm64: Extract kprobes_save_local_irqflag() and kprobes_restore_local_irqflag()

From: Masami Hiramatsu
Date: Mon May 18 2020 - 11:14:55 EST


On Sat, 16 May 2020 09:17:21 -0700
Doug Anderson <dianders@xxxxxxxxxxxx> wrote:

> Hi,
>
> On Sat, May 16, 2020 at 1:47 AM liwei (GF) <liwei391@xxxxxxxxxx> wrote:
> >
> > >> - kprobes_save_local_irqflag(kcb, regs);
> > >> + kernel_prepare_single_step(&kcb->saved_irqflag, regs);
> > >
> > > Is there some reason to have two functions? It seems like every time
> > > you call kernel_enable_single_step() you'd want to call
> > > kernel_prepare_single_step(). ...and every time you call
> > > kernel_disable_single_step() you'd want to call
> > > kernel_cleanup_single_step().
> > >
> > > Maybe you can just add the flags parameter to
> > > kernel_enable_single_step() / kernel_disable_single_step() and put the
> > > code in there?
> > >
> >
> > As kernel_enable_single_step() / kernel_disable_single_step() are also called in
> > breakpoint_handler() and watchpoint_handler(), i am not sure it's a right thing
> > to put the daif flag prepare/cleanup into them, especially we don't have a context
> > to save the flags.
>
> I think you misunderstood what I was suggesting. Maybe better with
> examples? I was suggesting doing this:
>
> kcb->saved_irqflag = kernel_enable_single_step(regs);
> ...
> kernel_disable_single_step(kcb->saved_irqflag, regs);
>
> To me that seems better than what you have now:
>
> kcb->saved_irqflag = kernel_prepare_single_step(regs);
> kernel_enable_single_step(regs);
> ...
> kernel_cleanup_single_step(kcb->saved_irqflag, regs);
> kernel_disable_single_step();
>
> ...or am I confused?

+1, this sounds good to me. Currently arch/arm64/kernel/probes/kprobes.c
has a code which sololy use kernel_disable_single_step() without regs
restoring, but it looks like a bug there. So maybe you need following patch.

Thank you,

-----
From: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
Date: Mon, 18 May 2020 23:08:28 +0900
Subject: [PATCH] arm64: kprobes: Restore saved interrupt flag before disabling
single step

Restore the saved interrupt flag in kprobe_ctlblk to regs->pstate
when a page fault happens on single-stepping instruction.
Without this fix, we will lose the flag if it happens because
kcb->saved_irqflag only knows the previous flag.

Fixes: 2dd0e8d2d2a1 ("arm64: Kprobes with single stepping support")
Signed-off-by: Masami Hiramatsu <mhiramat@xxxxxxxxxx>
Cc: stable@xxxxxxxxxxxxxxx
---
arch/arm64/kernel/probes/kprobes.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/arch/arm64/kernel/probes/kprobes.c b/arch/arm64/kernel/probes/kprobes.c
index d1c95dcf1d78..73fb99770f69 100644
--- a/arch/arm64/kernel/probes/kprobes.c
+++ b/arch/arm64/kernel/probes/kprobes.c
@@ -308,6 +308,7 @@ int __kprobes kprobe_fault_handler(struct pt_regs *regs, unsigned int fsr)
if (!instruction_pointer(regs))
BUG();

+ kprobes_restore_local_irqflag(kcb, regs);
kernel_disable_single_step();

if (kcb->kprobe_status == KPROBE_REENTER)
--
2.25.1




--
Masami Hiramatsu <mhiramat@xxxxxxxxxx>