[PATCH v6 13/13] riscv: Add qspinlock support

[Will Deacon]

On Tue, Nov 12, 2024 at 09:49:15AM +0800, Guo Ren wrote:
> On Tue, Nov 12, 2024 at 12:43 AM Will Deacon <will at kernel.org> wrote:
> >
> > On Sun, Nov 03, 2024 at 03:51:53PM +0100, Alexandre Ghiti wrote:
> > > In order to produce a generic kernel, a user can select
> > > CONFIG_COMBO_SPINLOCKS which will fallback at runtime to the ticket
> > > spinlock implementation if Zabha or Ziccrse are not present.
> > >
> > > Note that we can't use alternatives here because the discovery of
> > > extensions is done too late and we need to start with the qspinlock
> > > implementation because the ticket spinlock implementation would pollute
> > > the spinlock value, so let's use static keys.
> >
> > I think the static key toggling takes a mutex (jump_label_lock()) which
> > can take a spinlock (lock->wait_lock) internally, so I don't grok how
> > this works:
> >
> > > +static void __init riscv_spinlock_init(void)
> > > +{
> > > +     char *using_ext = NULL;
> > > +
> > > +     if (IS_ENABLED(CONFIG_RISCV_TICKET_SPINLOCKS)) {
> > > +             pr_info("Ticket spinlock: enabled\n");
> > > +             return;
> > > +     }
> > > +
> > > +     if (IS_ENABLED(CONFIG_RISCV_ISA_ZABHA) &&
> > > +         IS_ENABLED(CONFIG_RISCV_ISA_ZACAS) &&
> > > +         riscv_isa_extension_available(NULL, ZABHA) &&
> > > +         riscv_isa_extension_available(NULL, ZACAS)) {
> > > +             using_ext = "using Zabha";
> > > +     } else if (riscv_isa_extension_available(NULL, ZICCRSE)) {
> > > +             using_ext = "using Ziccrse";
> > > +     }
> > > +#if defined(CONFIG_RISCV_COMBO_SPINLOCKS)
> > > +     else {
> > > +             static_branch_disable(&qspinlock_key);
> > > +             pr_info("Ticket spinlock: enabled\n");
> > > +             return;
> > > +     }
> > > +#endif
> >
> > i.e. we've potentially already used the qspinlock at this point.
> Yes, I've used qspinlock here. But riscv_spinlock_init is called with
> irq_disabled and smp_off. That means this qspinlock only performs a
> test-set lock behavior by qspinlock fast-path.

That's... horrendous.

Will