[PATCH v4 3/4] locking/qspinlock: Add ARCH_USE_QUEUED_SPINLOCKS_XCHG32

[Christoph Müllner]

On Wed, Apr 7, 2021 at 6:00 PM Peter Zijlstra <peterz at infradead.org> wrote:
>
> On Wed, Apr 07, 2021 at 04:29:12PM +0200, Christoph Müllner wrote:
> > RISC-V defines LR/SC loops consisting of up to 16 instructions as
> > constrained LR/SC loops.  Such constrained LR/SC loops provide the
> > required forward guarantees, that are expected (similar to what other
> > architectures, like AArch64, have).
>
> The text quoted by others didn't seem to say such a thing, but whatever.

The RISC-V unpriv spec is public can be found here:
https://riscv.org/technical/specifications/
Version 20191213 discusses LR/SC-loops in section 8.3 (page 51).
So in case you are interested in the exact wording, you can find it there.

> > What RISC-V does not have is sub-word atomics and if required, we
> > would have to implement them as LL/SC sequences. And yes, using atomic
> > instructions is preferred over using LL/SC,
>
> (psudo asm, can't be bothered to figure out the actual syntax)
>
>         # setup r_and_mask, r_or_mask
>
> .L1
>         LL r, [word]
>         AND r, r, r_and_mask
>         OR r, r, r_or_mask
>         SC r, [word]
>         JNE .L1

I fully agree with this.
I've implemented a patch for that two weeks ago using the following helper:

+/*
+ * Mask and set given bits at a given address atomically.
+ * The masked old value will be returned.
+ */
+static inline u32 atomic_mask_and_set(u32* p, u32 mask, u32 val)
+{
+       u32 ret, tmp;
+       __asm__ __volatile__ (
+               "0:     lr.w %0, %2\n"
+               "       and  %0, %0, %3\n"
+               "       or   %1, %0, %4\n"
+               "       sc.w %1, %1, %2\n"
+               "       bnez %1, 0b\n"
+               : "+&r"(ret), "=&r" (tmp), "+A"(*p)
+               : "r" (mask), "rJ"(val)
+               : "memory");
+       return ret;
+}

However, Guo pushed out a new patchset in between and I decided to not continue
my approach to not undermine his approach.

I will sync up with Guo to provide a common patchset.

Thanks,
Christoph

> is what you need for LL/SC based xchg16, that's less than 16
> instructions. If RISC-V guarantees fwd progress on that, good, write it
> like that and lets end this thread.
>
> The fact that this is apparently hard, is not good.