Re: [PATCH v3] x86/fault: Send a SIGBUS to user process always for hwpoison page access.

[Aili Yao <yaoaili@kingsoft.com>]

On Wed, 10 Mar 2021 17:28:12 -0800
Andy Lutomirski <luto@amacapital.net> wrote:

> On Wed, Mar 10, 2021 at 5:19 PM Aili Yao <yaoaili@kingsoft.com> wrote:
> >
> > On Mon, 8 Mar 2021 11:00:28 -0800
> > Andy Lutomirski <luto@amacapital.net> wrote:
> >  
> > > > On Mar 8, 2021, at 10:31 AM, Luck, Tony <tony.luck@intel.com> wrote:
> > > >
> > > >   
> > > >>
> > > >> Can you point me at that SIGBUS code in a current kernel?  
> > > >
> > > > It is in kill_me_maybe().  mce_vaddr is setup when we disassemble whatever get_user()
> > > > or copy from user variant was in use in the kernel when the poison memory was consumed.
> > > >
> > > >        if (p->mce_vaddr != (void __user *)-1l) {
> > > >                force_sig_mceerr(BUS_MCEERR_AR, p->mce_vaddr, PAGE_SHIFT);  
> > >
> > > Hmm. On the one hand, no one has complained yet. On the other hand, hardware that supports this isn’t exactly common.
> > >
> > > We may need some actual ABI design here. We also need to make sure that things like io_uring accesses or, more generally, anything using the use_mm / use_temporary_mm ends up either sending no signal or sending a signal to the right target.
> > >  
> > > >
> > > > Would it be any better if we used the BUS_MCEERR_AO code that goes into siginfo?  
> > >
> > > Dunno.  
> >
> > I have one thought here but don't know if it's proper:
> >
> > Previous patch use force_sig_mceerr to the user process for such a scenario; with this method
> > The SIGBUS can't be ignored as force_sig_mceerr() was designed to.
> >
> > If the user process don't want this signal, will it set signal config to ignore?
> > Maybe we can use a send_sig_mceerr() instead of force_sig_mceerr(), if process want to
> > ignore the SIGBUS, then it will ignore that, or it can also process the SIGBUS?  
> 
> I don't think the signal blocking mechanism makes sense for this.
> Blocking a signal is for saying that, if another process sends the
> signal (or an async event like ctrl-C), then the process doesn't want
> it.  Blocking doesn't block synchronous things like faults.
> 
> I think we need to at least fix the existing bug before we add more
> signals.  AFAICS the MCE_IN_KERNEL_COPYIN code is busted for kernel
> threads.

Got this, Thanks!

I read https://man7.org/linux/man-pages/man2/write.2.html, and it seems the write syscall is not expecting
an signal, maybe a specific error code for this scenario is enough.

-- 
Thanks!
Aili Yao