Re: [PATCH] HWPOISON: add a pr_err message when forcibly send a sigbus

[Will Deacon]

On Mon, Aug 28, 2023 at 09:41:55AM +0800, Shuai Xue wrote:
> On 2023/8/22 09:15, Shuai Xue wrote:
> > On 2023/8/21 18:50, Will Deacon wrote:
> >>> diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
> >>> index 3fe516b32577..38e2186882bd 100644
> >>> --- a/arch/arm64/mm/fault.c
> >>> +++ b/arch/arm64/mm/fault.c
> >>> @@ -679,6 +679,8 @@ static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
> >>>  	} else if (fault & (VM_FAULT_HWPOISON_LARGE | VM_FAULT_HWPOISON)) {
> >>>  		unsigned int lsb;
> >>>  
> >>> +		pr_err("MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
> >>> +		       current->comm, current->pid, far);
> >>>  		lsb = PAGE_SHIFT;
> >>>  		if (fault & VM_FAULT_HWPOISON_LARGE)
> >>>  			lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
> >>
> >> Hmm, I'm not convinced by this. We have 'show_unhandled_signals' already,
> >> and there's plenty of code in memory-failure.c for handling poisoned pages
> >> reported by e.g. GHES. I don't think dumping extra messages in dmesg from
> >> the arch code really adds anything.
> > 
> > I see the show_unhandled_signals() will dump the stack but it rely on
> > /proc/sys/debug/exception-trace be set.
> > 
> > The memory failure is the top issue in our production cloud and also other hyperscalers.
> > We have received complaints from our operations engineers and end users that processes
> > are being inexplicably killed :(. Could you please consider add a message?

I don't have any objection to logging this stuff somehow, I'm just not
convinced that the console is the best place for that information in 2023.
Is there really nothing better?

Will