[PATCH 3/5] openrisc: Move to ticket-spinlock

[Guo Ren]

On Tue, Mar 22, 2022 at 12:10 PM Stafford Horne <shorne at gmail.com> wrote:
>
> On Tue, Mar 22, 2022 at 11:29:13AM +0800, Guo Ren wrote:
> > On Tue, Mar 22, 2022 at 7:23 AM Stafford Horne <shorne at gmail.com> wrote:
> > >
> > > On Wed, Mar 16, 2022 at 04:25:58PM -0700, Palmer Dabbelt wrote:
> > > > From: Peter Zijlstra <peterz at infradead.org>
> > > >
> > > > We have no indications that openrisc meets the qspinlock requirements,
> > > > so move to ticket-spinlock as that is more likey to be correct.
> > > >
> > > > Signed-off-by: Palmer Dabbelt <palmer at rivosinc.com>
> > > >
> > > > ---
> > > >
> > > > I have specifically not included Peter's SOB on this, as he sent his
> > > > original patch
> > > > <https://lore.kernel.org/lkml/YHbBBuVFNnI4kjj3@hirez.programming.kicks-ass.net/>
> > > > without one.
> > > > ---
> > > >  arch/openrisc/Kconfig                      | 1 -
> > > >  arch/openrisc/include/asm/Kbuild           | 5 ++---
> > > >  arch/openrisc/include/asm/spinlock.h       | 3 +--
> > > >  arch/openrisc/include/asm/spinlock_types.h | 2 +-
> > > >  4 files changed, 4 insertions(+), 7 deletions(-)
> > >
> > > Hello,
> > >
> > > This series breaks SMP support on OpenRISC.  I haven't traced it down yet, it
> > > seems trivial but I have a few places to check.
> > >
> > > I replied to this on a kbuild warning thread, but also going to reply here with
> > > more information.
> > >
> > >  https://lore.kernel.org/lkml/YjeY7CfaFKjr8IUc@antec/#R
> > >
> > > So far this is what I see:
> > >
> > >   * ticket_lock is stuck trying to lock console_sem
> > >   * it is stuck on atomic_cond_read_acquire
> > >     reading lock value: returns 0    (*lock is 0x10000)
> > >     ticket value: is 1
> > >   * possible issues:
> > >     - OpenRISC is big endian, that seems to impact ticket_unlock, it looks
> > All csky & riscv are little-endian, it seems the series has a bug with
> > big-endian. Is that all right for qemu? (If qemu was all right, but
> > real hardware failed.)
>
> Hi Guo Ren,
>
> OpenRISC real hardware and QEMU are both big-endian.  It fails on both.
>
> I replied on patch 1/5 with a suggested patch which fixes the issue for me.
> Please have a look.
Great, I saw that:

static __always_inline void ticket_unlock(arch_spinlock_t *lock)
{
     u16 *ptr = (u16 *)lock + __is_defined(__BIG_ENDIAN);
//__is_defined(__BIG_ENDIAN) would be zero for openrisc.

Using CONFIG_CPU_BIG_ENDIAN is correct, Arnd has also asked me using
CONFIG_CPU_BIG_ENDIAN in compat.h:

diff --git a/include/asm-generic/compat.h b/include/asm-generic/compat.h
index 11653d6846cc..d06308a2a7a8 100644
--- a/include/asm-generic/compat.h
+++ b/include/asm-generic/compat.h
@@ -14,6 +14,13 @@
 #define COMPAT_OFF_T_MAX       0x7fffffff
 #endif

+#if !defined(compat_arg_u64) && !defined(CONFIG_CPU_BIG_ENDIAN)
+#define compat_arg_u64(name)           u32  name##_lo, u32  name##_hi
+#define compat_arg_u64_dual(name)      u32, name##_lo, u32, name##_hi
+#define compat_arg_u64_glue(name)      (((u64)name##_lo & 0xffffffffUL) | \
+                                        ((u64)name##_hi << 32))
+#endif
+
 /* These types are common across all compat ABIs */
 typedef u32 compat_size_t;
 typedef s32 compat_ssize_t;



> BTW. now I can look into the sparse warnings.
>
> -Stafford
>
--
Best Regards
 Guo Ren

ML: https://lore.kernel.org/linux-csky/