[PATCH] arm64: traps: Mark kernel as tainted on SError panic

[Will Deacon]

On Mon, Jul 14, 2025 at 05:26:43AM -0700, Breno Leitao wrote:
> On Sun, Jul 13, 2025 at 11:46:06PM +0100, Will Deacon wrote:
> > On Thu, Jul 10, 2025 at 03:46:35AM -0700, Breno Leitao wrote:
> 
> > > --- a/arch/arm64/kernel/traps.c
> > > +++ b/arch/arm64/kernel/traps.c
> > > @@ -931,6 +931,7 @@ void __noreturn panic_bad_stack(struct pt_regs *regs, unsigned long esr, unsigne
> > >  
> > >  void __noreturn arm64_serror_panic(struct pt_regs *regs, unsigned long esr)
> > >  {
> > > +	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_STILL_OK);
> > >  	console_verbose();
> > >  
> > >  	pr_crit("SError Interrupt on CPU%d, code 0x%016lx -- %s\n",
> > 
> > If we're going to taint for SError, shouldn't we also taint for an
> > unclaimed SEA?
> 
> Yes. I was not very familiar with SEA errors, given I haven't seen on in
> production yet, but, reading about it, that is another seems to crash
> the system due to hardware errors, thus, we want to taint MACHINE_CHECK.
> 
> What about this?
> 
> 	Author: Breno Leitao <leitao at debian.org>
> 	Date:   Mon Jul 14 05:16:55 2025 -0700
> 
> 	arm64: Taint kernel on fatal hardware error in do_sea()
> 
> 	This patch updates the do_sea() handler to taint the kernel with
> 	TAINT_MACHINE_CHECK when a fatal hardware error is detected and
> 	reported through Synchronous External Abort (SEA). By marking
> 	the kernel as tainted at the point of error, we improve
> 	post-mortem diagnostics and make it clear that a machine check
> 	or unrecoverable hardware fault has occurred.
> 
> 	Suggested-by: Will Deacon <will at kernel.org>
> 	Signed-off-by: Breno Leitao <leitao at debian.org>
> 
> 	diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
> 	index 11eb8d1adc84..f590dc71ce99 100644
> 	--- a/arch/arm64/mm/fault.c
> 	+++ b/arch/arm64/mm/fault.c
> 	@@ -838,6 +838,7 @@ static int do_sea(unsigned long far, unsigned long esr, struct pt_regs *regs)
> 			*/
> 			siaddr  = untagged_addr(far);
> 		}
> 	+	add_taint(TAINT_MACHINE_CHECK, LOCKDEP_STILL_OK);
> 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, siaddr, esr);
> 
> 		return 0;

Yeah, I reckon so. Probably just fold these into a single patch, though.

Cheers,

Will