[PATCH v8 17/22] arm64: mm: Add page fault trace points

[Nam Cao]

On Tue, May 20, 2025 at 03:15:01PM +0100, Catalin Marinas wrote:
> On Tue, May 20, 2025 at 02:25:48PM +0200, Nam Cao wrote:
> > On Mon, May 19, 2025 at 03:49:29PM +0100, Catalin Marinas wrote:
> > > On Mon, May 12, 2025 at 12:51:00PM +0200, Nam Cao wrote:
> > > > diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
> > > > index ec0a337891dd..55094030e377 100644
> > > > --- a/arch/arm64/mm/fault.c
> > > > +++ b/arch/arm64/mm/fault.c
> > > > @@ -44,6 +44,9 @@
> > > >  #include <asm/tlbflush.h>
> > > >  #include <asm/traps.h>
> > > >  
> > > > +#define CREATE_TRACE_POINTS
> > > > +#include <trace/events/exceptions.h>
> > > > +
> > > >  struct fault_info {
> > > >  	int	(*fn)(unsigned long far, unsigned long esr,
> > > >  		      struct pt_regs *regs);
> > > > @@ -559,6 +562,11 @@ static int __kprobes do_page_fault(unsigned long far, unsigned long esr,
> > > >  	if (kprobe_page_fault(regs, esr))
> > > >  		return 0;
> > > >  
> > > > +	if (user_mode(regs))
> > > > +		trace_page_fault_user(addr, regs, esr);
> > > > +	else
> > > > +		trace_page_fault_kernel(addr, regs, esr);
> > > 
> > > What are the semantics for these tracepoints? When are they supposed to
> > > be called? In the RV context context I guess you only care about the
> > > benign, recoverable faults that would affect timing. These tracepoints
> > > were generalised from the x86 code but I don't know enough about it to
> > > tell when they would be invoked.
> > > 
> > > For arm64, we also have the do_translation_fault() path for example that
> > > may or may not need to log such trace events.
> > 
> > These tracepoints are invoked for x86 page fault exceptions. Are arm64's
> > translation faults considered equivalent to x86 page faults?
> 
> Probably. We route permission or access flag faults via do_page_fault()
> directly while missing page table entries via do_translation_fault().
> The latter end up in do_page_fault() only if the faulting address is in
> the user address range.
> 
> My point was that we may not always invoke the trace callbacks if, for
> example, the user tries to access the kernel space (and results in a
> SIGSEGV). While that's fine for RV, I wanted to know what is expected of
> these trace points in general. Do we need to log such SIGSEGV-generating
> events? We do log them if there's a permission fault.

I'm not sure. Digging into history, these tracepoints were added for LTTng.
So maybe LTTng's developer could answer this.

Added to the conversation: Mathieu Desnoyers <mathieu.desnoyers at efficios.com>
Can you please give some insight to the above question?

Best regards,
Nam