From: Peter Hessler Subject: Re: Plan to enable parallel faults handling To: tech@openbsd.org Date: Wed, 12 Nov 2025 17:02:15 +0100 On 2025 Nov 12 (Wed) at 17:00:14 +0100 (+0100), Jeremie Courreges-Anglas wrote: :On Wed, Nov 12, 2025 at 04:23:39PM +0100, Claudio Jeker wrote: :> On Wed, Nov 12, 2025 at 02:40:15PM +0000, Martin Pieuchot wrote: :> > I'd like to move forward with parallel fault handling. This has been :> > properly tested on amd64, i386 and arm64. sparc64 pmap needs some love :> > and enabling this will expose the existing bugs. :> > :> > My plan is to enable parallel faults on amd64 and arm64 as a first step :> > then i386 because it is where swapping is properly exercised. :> > :> > Does that sound sane, do you agree with this approach? :> :> I'm fine with this approach if we don't leave it like this forever. I'm :> not sure if it is only the sparc64 pmap that needs love. : :Indeed. After that diff goes in I'd like to test riscv64 again and :add it to the list. But for the rest of the MP architectures we'll :need to prod people to test and report. : Yea, I'm happy with a restricted list for now that expands (quickly ;) ) over time. :> In any case this is the right time to push forward. OK claudio@ : :I'd also be happy to see this make progress. ok jca@ : OK :> > Index: uvm/uvm_fault.c :> > =================================================================== :> > RCS file: /cvs/src/sys/uvm/uvm_fault.c,v :> > diff -u -p -r1.171 uvm_fault.c :> > --- uvm/uvm_fault.c 11 Sep 2025 17:04:35 -0000 1.171 :> > +++ uvm/uvm_fault.c 12 Nov 2025 14:09:52 -0000 :> > @@ -618,7 +618,7 @@ uvm_fault(vm_map_t orig_map, vaddr_t vad :> > flt.access_type = access_type; :> > flt.narrow = FALSE; /* assume normal fault for now */ :> > flt.wired = FALSE; /* assume non-wired fault for now */ :> > -#if notyet :> > +#if defined(__amd64__) || defined(__arm64__) :> > flt.upper_lock_type = RW_READ; :> > flt.lower_lock_type = RW_READ; /* shared lock for now */ :> > #else :> > :> > :> :> -- :> :wq Claudio :> : :-- :jca :