Re: [PATCH v3 8/9] riscv: introduce interfaces to patch kernel code

[Zong Li]

On Sat, Apr 4, 2020 at 8:12 PM Zong Li <zong.li@xxxxxxxxxx> wrote:
>
> On Sat, Apr 4, 2020 at 11:14 AM Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> >
> > Hi Zong,
> >
> > On Fri, 3 Apr 2020 17:04:51 +0800
> > Zong Li <zong.li@xxxxxxxxxx> wrote:
> >
> > > > > > > +{
> > > > > > > +     void *waddr = addr;
> > > > > > > +     bool across_pages = (((uintptr_t) addr & ~PAGE_MASK) + len) > PAGE_SIZE;
> > > > > > > +     unsigned long flags = 0;
> > > > > > > +     int ret;
> > > > > > > +
> > > > > > > +     raw_spin_lock_irqsave(&patch_lock, flags);
> > > > > >
> > > > > > This looks a bit odd since stop_machine() is protected by its own mutex,
> > > > > > and also the irq is already disabled here.
> > > > >
> > > > > We need it because we don't always enter the riscv_patch_text_nosync()
> > > > > through stop_machine mechanism. If we call the
> > > > > riscv_patch_text_nosync() directly, we need a lock to protect the
> > > > > page.
> > > >
> > > > Oh, OK, but it leads another question. Is that safe to patch the
> > > > text without sync? Would you use it for UP system?
> > > > I think it is better to clarify "in what case user can call _nosync()"
> > > > and add a comment on it.
> > >
> > > The ftrace is one of the cases, as documentation of ftrace said, when
> > > dynamic ftrace is initialized, it calls kstop_machine to make the
> > > machine act like a uniprocessor so that it can freely modify code
> > > without worrying about other processors executing that same code. So
> > > the ftrace called the _nosync interface here directly.
> >
> > Hmm, even though, since it already running under kstop_machine(), no
> > other thread will run.
> > Could you consider to use text_mutex instead of that? The text_mutex
> > is already widely used in x86 and kernel/kprobes.c etc.
> >
> > (Hmm, it seems except for x86, alternative code don't care about
> >  racing...)
> >

The mutex_lock doesn't seem to work in ftrace context, I think it
might be the reason why other architectures didn't use text_mutex in
somewhere.

# echo function > current_tracer
[   28.198070] BUG: sleeping function called from invalid context at
kernel/locking/mutex.c:281
[   28.198663] in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid:
11, name: migration/0
[   28.199491] CPU: 0 PID: 11 Comm: migration/0 Not tainted
5.6.0-00012-gd6f56a7a4be2-dirty #10
[   28.200330] Call Trace:
[   28.200798] [<ffffffe00060319a>] walk_stackframe+0x0/0xcc
[   28.201395] [<ffffffe000603442>] show_stack+0x3c/0x46
[   28.200798] [<ffffffe00060319a>] walk_stackframe+0x0/0xcc
[   28.201395] [<ffffffe000603442>] show_stack+0x3c/0x46
[   28.201898] [<ffffffe000d498b0>] dump_stack+0x76/0x90
[   28.202329] [<ffffffe00062c3f0>] ___might_sleep+0x100/0x10e
[   28.202720] [<ffffffe00062c448>] __might_sleep+0x4a/0x78
[   28.203033] [<ffffffe000d61622>] mutex_lock+0x2c/0x54
[   28.203397] [<ffffffe00060393e>] patch_insn_write+0x32/0xd8
[   28.203780] [<ffffffe000603a94>] patch_text_nosync+0x10/0x32
[   28.204139] [<ffffffe0006051b0>] __ftrace_modify_call+0x5c/0x6c
[   28.204497] [<ffffffe0006052c6>] ftrace_update_ftrace_func+0x20/0x4a
[   28.204919] [<ffffffe000697742>] ftrace_modify_all_code+0xa0/0x148
[   28.205378] [<ffffffe0006977fc>] __ftrace_modify_code+0x12/0x1c
[   28.205793] [<ffffffe0006924b6>] multi_cpu_stop+0xa2/0x158
[   28.206147] [<ffffffe0006921b0>] cpu_stopper_thread+0xa4/0x13a
[   28.206510] [<ffffffe000629f38>] smpboot_thread_fn+0xf8/0x1da
[   28.206868] [<ffffffe000625f36>] kthread+0xfa/0x12a
[   28.207201] [<ffffffe0006017e2>] ret_from_exception+0x0/0xc

>
> Yes, text_mutex seems to be great. I'll change to use text_mutex in
> the next version if it works fine after testing. Thanks.
>
> > Thank you,
> > --
> > Masami Hiramatsu <mhiramat@xxxxxxxxxx>