Re: [PATCH -next V6 1/7] riscv: ftrace: Fixup panic by disabling preemption

[Guo Ren]

On Tue, Jan 10, 2023 at 1:20 AM Mark Rutland <mark.rutland@xxxxxxx> wrote:
>
> On Sat, Jan 07, 2023 at 08:35:43AM -0500, guoren@xxxxxxxxxx wrote:
> > From: Andy Chiu <andy.chiu@xxxxxxxxxx>
> >
> > In RISCV, we must use an AUIPC + JALR pair to encode an immediate,
> > forming a jump that jumps to an address over 4K. This may cause errors
> > if we want to enable kernel preemption and remove dependency from
> > patching code with stop_machine(). For example, if a task was switched
> > out on auipc. And, if we changed the ftrace function before it was
> > switched back, then it would jump to an address that has updated 11:0
> > bits mixing with previous XLEN:12 part.
> >
> > p: patched area performed by dynamic ftrace
> > ftrace_prologue:
> > p|      REG_S   ra, -SZREG(sp)
> > p|      auipc   ra, 0x? ------------> preempted
> >                                       ...
> >                               change ftrace function
> >                                       ...
> > p|      jalr    -?(ra) <------------- switched back
> > p|      REG_L   ra, -SZREG(sp)
> > func:
> >       xxx
> >       ret
>
> What happens on SMP but not !PREEMPTION; can't a CPU be in the middle of this
> while you're patching the sequence?
Yes, when PREEMPTION, a timer interrupt between auipc & jalr may cause
context_switch. And riscv uses stop_machine for patch_text. Then, we
may modify auipc part, but only execute the jalr part when return.

>
> Do you have any guarantee as to the atomicity and ordering of instruction
> fetches?
Not yet. If the region is short, we could use nop + jalr pair instead.
Only one jalr instruction makes the entry atomicity.

There are already several proposed solutions:
1. Make stop_machine guarantee all CPU out of preemption point.
2. Expand -fpatchable-function-entry from 4 to 24, and make detour
codes atomicity.
3. We want to propose a solution to make auipc by hardware mask_irq.
For more details, see:
https://www.youtube.com/watch?v=4JkkkXuEvCw


>
> Thanks,
> Mark.
>
> >
> > Fixes: afc76b8b8011 ("riscv: Using PATCHABLE_FUNCTION_ENTRY instead of MCOUNT")
> > Signed-off-by: Andy Chiu <andy.chiu@xxxxxxxxxx>
> > Signed-off-by: Guo Ren <guoren@xxxxxxxxxx>
> > ---
> >  arch/riscv/Kconfig | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> > index e2b656043abf..ee0d39b26794 100644
> > --- a/arch/riscv/Kconfig
> > +++ b/arch/riscv/Kconfig
> > @@ -138,7 +138,7 @@ config RISCV
> >       select HAVE_DYNAMIC_FTRACE_WITH_REGS if HAVE_DYNAMIC_FTRACE
> >       select HAVE_FTRACE_MCOUNT_RECORD if !XIP_KERNEL
> >       select HAVE_FUNCTION_GRAPH_TRACER
> > -     select HAVE_FUNCTION_TRACER if !XIP_KERNEL
> > +     select HAVE_FUNCTION_TRACER if !XIP_KERNEL && !PREEMPTION
> >
> >  config ARCH_MMAP_RND_BITS_MIN
> >       default 18 if 64BIT
> > --
> > 2.36.1
> >



--
Best Regards
 Guo Ren