riscv: Implement arch_cmpxchg128() using Zacas

Now that Zacas is supported in the kernel, let's use the double word
atomic version of amocas to improve the SLUB allocator.

Note that we have to select fixed registers, otherwise gcc fails to pick
even registers and then produces a reserved encoding which fails to
assemble.

Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com>
Reviewed-by: Andrew Jones <ajones@ventanamicro.com>
Reviewed-by: Andrea Parri <parri.andrea@gmail.com>
Link: https://lore.kernel.org/r/20241103145153.105097-8-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>

Author: Alexandre Ghiti

arch/riscv/Kconfig

@@ -115,6 +115,7 @@ config RISCV
 	select GENERIC_VDSO_TIME_NS if HAVE_GENERIC_VDSO
 	select HARDIRQS_SW_RESEND
 	select HAS_IOPORT if MMU
+	select HAVE_ALIGNED_STRUCT_PAGE
 	select HAVE_ARCH_AUDITSYSCALL
 	select HAVE_ARCH_HUGE_VMALLOC if HAVE_ARCH_HUGE_VMAP
 	select HAVE_ARCH_HUGE_VMAP if MMU && 64BIT

arch/riscv/include/asm/cmpxchg.h

@@ -296,6 +296,44 @@
 	arch_cmpxchg_release((ptr), (o), (n));				\
 })
 
+#if defined(CONFIG_64BIT) && defined(CONFIG_RISCV_ISA_ZACAS)
+
+#define system_has_cmpxchg128()        riscv_has_extension_unlikely(RISCV_ISA_EXT_ZACAS)
+
+union __u128_halves {
+	u128 full;
+	struct {
+		u64 low, high;
+	};
+};
+
+#define __arch_cmpxchg128(p, o, n, cas_sfx)					\
+({										\
+	__typeof__(*(p)) __o = (o);                                             \
+	union __u128_halves __hn = { .full = (n) };				\
+	union __u128_halves __ho = { .full = (__o) };				\
+	register unsigned long t1 asm ("t1") = __hn.low;			\
+	register unsigned long t2 asm ("t2") = __hn.high;			\
+	register unsigned long t3 asm ("t3") = __ho.low;			\
+	register unsigned long t4 asm ("t4") = __ho.high;			\
+										\
+	__asm__ __volatile__ (							\
+		 "       amocas.q" cas_sfx " %0, %z3, %2"			\
+		 : "+&r" (t3), "+&r" (t4), "+A" (*(p))				\
+		 : "rJ" (t1), "rJ" (t2)						\
+		 : "memory");							\
+										\
+		 ((u128)t4 << 64) | t3;						\
+})
+
+#define arch_cmpxchg128(ptr, o, n)						\
+	__arch_cmpxchg128((ptr), (o), (n), ".aqrl")
+
+#define arch_cmpxchg128_local(ptr, o, n)					\
+	__arch_cmpxchg128((ptr), (o), (n), "")
+
+#endif /* CONFIG_64BIT && CONFIG_RISCV_ISA_ZACAS */
+
 #ifdef CONFIG_RISCV_ISA_ZAWRS
 /*
  * Despite wrs.nto being "WRS-with-no-timeout", in the absence of changes to