LoongArch: Add subword xchg/cmpxchg emulation

LoongArch only support 32-bit/64-bit xchg/cmpxchg in native. But percpu
operation, qspinlock and some drivers need 8-bit/16-bit xchg/cmpxchg. We
add subword xchg/cmpxchg emulation in this patch because the emulation
has better performance than the generic implementation (on NUMA system),
and it can fix some build errors meanwhile [1].

LoongArch's guarantee for forward progress (avoid many ll/sc happening
at the same time and no one succeeds):

We have the "exclusive access (with timeout) of ll" feature to avoid
simultaneous ll (which also blocks other memory load/store on the same
address), and the "random delay of sc" feature to avoid simultaneous
sc. It is a mandatory requirement for multi-core LoongArch processors
to implement such features, only except those single-core and dual-core
processors (they also don't support multi-chip interconnection).

Feature bits are introduced in CPUCFG3, bit 3 and bit 4 [2].

[1] https://lore.kernel.org/loongarch/CAAhV-H6vvkuOzy8OemWdYK3taj5Jn3bFX0ZTwE=twM8ywpBUYA@mail.gmail.com/T/#t
[2] https://loongson.github.io/LoongArch-Documentation/LoongArch-Vol1-EN.html#_cpucfg

Reported-by: Sudip Mukherjee (Codethink) <sudipm.mukherjee@gmail.com>
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Rui Wang <wangrui@loongson.cn>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>

arch/loongarch/include/asm/cmpxchg.h

@@ -5,8 +5,9 @@
 #ifndef __ASM_CMPXCHG_H
 #define __ASM_CMPXCHG_H
 
-#include <asm/barrier.h>
+#include <linux/bits.h>
 #include <linux/build_bug.h>
+#include <asm/barrier.h>
 
 #define __xchg_asm(amswap_db, m, val)		\
 ({						\
@@ -21,10 +22,53 @@
 		__ret;				\
 })
 
+static inline unsigned int __xchg_small(volatile void *ptr, unsigned int val,
+					unsigned int size)
+{
+	unsigned int shift;
+	u32 old32, mask, temp;
+	volatile u32 *ptr32;
+
+	/* Mask value to the correct size. */
+	mask = GENMASK((size * BITS_PER_BYTE) - 1, 0);
+	val &= mask;
+
+	/*
+	 * Calculate a shift & mask that correspond to the value we wish to
+	 * exchange within the naturally aligned 4 byte integerthat includes
+	 * it.
+	 */
+	shift = (unsigned long)ptr & 0x3;
+	shift *= BITS_PER_BYTE;
+	mask <<= shift;
+
+	/*
+	 * Calculate a pointer to the naturally aligned 4 byte integer that
+	 * includes our byte of interest, and load its value.
+	 */
+	ptr32 = (volatile u32 *)((unsigned long)ptr & ~0x3);
+
+	asm volatile (
+	"1:	ll.w		%0, %3		\n"
+	"	andn		%1, %0, %z4	\n"
+	"	or		%1, %1, %z5	\n"
+	"	sc.w		%1, %2		\n"
+	"	beqz		%1, 1b		\n"
+	: "=&r" (old32), "=&r" (temp), "=ZC" (*ptr32)
+	: "ZC" (*ptr32), "Jr" (mask), "Jr" (val << shift)
+	: "memory");
+
+	return (old32 & mask) >> shift;
+}
+
 static inline unsigned long __xchg(volatile void *ptr, unsigned long x,
 				   int size)
 {
 	switch (size) {
+	case 1:
+	case 2:
+		return __xchg_small(ptr, x, size);
+
 	case 4:
 		return __xchg_asm("amswap_db.w", (volatile u32 *)ptr, (u32)x);
 
@@ -67,10 +111,62 @@ static inline unsigned long __xchg(volatile void *ptr, unsigned long x,
 	__ret;								\
 })
 
+static inline unsigned int __cmpxchg_small(volatile void *ptr, unsigned int old,
+					   unsigned int new, unsigned int size)
+{
+	unsigned int shift;
+	u32 old32, mask, temp;
+	volatile u32 *ptr32;
+
+	/* Mask inputs to the correct size. */
+	mask = GENMASK((size * BITS_PER_BYTE) - 1, 0);
+	old &= mask;
+	new &= mask;
+
+	/*
+	 * Calculate a shift & mask that correspond to the value we wish to
+	 * compare & exchange within the naturally aligned 4 byte integer
+	 * that includes it.
+	 */
+	shift = (unsigned long)ptr & 0x3;
+	shift *= BITS_PER_BYTE;
+	old <<= shift;
+	new <<= shift;
+	mask <<= shift;
+
+	/*
+	 * Calculate a pointer to the naturally aligned 4 byte integer that
+	 * includes our byte of interest, and load its value.
+	 */
+	ptr32 = (volatile u32 *)((unsigned long)ptr & ~0x3);
+
+	asm volatile (
+	"1:	ll.w		%0, %3		\n"
+	"	and		%1, %0, %z4	\n"
+	"	bne		%1, %z5, 2f	\n"
+	"	andn		%1, %0, %z4	\n"
+	"	or		%1, %1, %z6	\n"
+	"	sc.w		%1, %2		\n"
+	"	beqz		%1, 1b		\n"
+	"	b		3f		\n"
+	"2:					\n"
+	__WEAK_LLSC_MB
+	"3:					\n"
+	: "=&r" (old32), "=&r" (temp), "=ZC" (*ptr32)
+	: "ZC" (*ptr32), "Jr" (mask), "Jr" (old), "Jr" (new)
+	: "memory");
+
+	return (old32 & mask) >> shift;
+}
+
 static inline unsigned long __cmpxchg(volatile void *ptr, unsigned long old,
 				      unsigned long new, unsigned int size)
 {
 	switch (size) {
+	case 1:
+	case 2:
+		return __cmpxchg_small(ptr, old, new, size);
+
 	case 4:
 		return __cmpxchg_asm("ll.w", "sc.w", (volatile u32 *)ptr,
 				     (u32)old, new);

arch/loongarch/include/asm/percpu.h

@@ -123,6 +123,10 @@ static inline unsigned long __percpu_xchg(void *ptr, unsigned long val,
 						int size)
 {
 	switch (size) {
+	case 1:
+	case 2:
+		return __xchg_small((volatile void *)ptr, val, size);
+
 	case 4:
 		return __xchg_asm("amswap.w", (volatile u32 *)ptr, (u32)val);
 
@@ -204,9 +208,13 @@ do {									\
 #define this_cpu_write_4(pcp, val) _percpu_write(pcp, val)
 #define this_cpu_write_8(pcp, val) _percpu_write(pcp, val)
 
+#define this_cpu_xchg_1(pcp, val) _percpu_xchg(pcp, val)
+#define this_cpu_xchg_2(pcp, val) _percpu_xchg(pcp, val)
 #define this_cpu_xchg_4(pcp, val) _percpu_xchg(pcp, val)
 #define this_cpu_xchg_8(pcp, val) _percpu_xchg(pcp, val)
 
+#define this_cpu_cmpxchg_1(ptr, o, n) _protect_cmpxchg_local(ptr, o, n)
+#define this_cpu_cmpxchg_2(ptr, o, n) _protect_cmpxchg_local(ptr, o, n)
 #define this_cpu_cmpxchg_4(ptr, o, n) _protect_cmpxchg_local(ptr, o, n)
 #define this_cpu_cmpxchg_8(ptr, o, n) _protect_cmpxchg_local(ptr, o, n)