LoongArch: Support dbar with different hints

Traditionally, LoongArch uses "dbar 0" (full completion barrier) for
everything. But the full completion barrier is a performance killer, so
Loongson-3A6000 and newer processors have made finer granularity hints
available:

Bit4: ordering or completion (0: completion, 1: ordering)
Bit3: barrier for previous read (0: true, 1: false)
Bit2: barrier for previous write (0: true, 1: false)
Bit1: barrier for succeeding read (0: true, 1: false)
Bit0: barrier for succeeding write (0: true, 1: false)

Hint 0x700: barrier for "read after read" from the same address, which
is needed by LL-SC loops on old models (dbar 0x700 behaves the same as
nop if such reordering is disabled on new models).

This patch makes use of the various new hints for different kinds of
memory barriers. It brings performance improvements on Loongson-3A6000
series, while not affecting the existing models because all variants are
treated as 'dbar 0' there.

Why override queued_spin_unlock()?
After commit 01e3b958efe85a26d9b ("drivers: Remove explicit invocations
of mmiowb()") we need a completion barrier in queued_spin_unlock(), but
the generic implementation use smp_store_release() which only provide an
ordering barrier.

Signed-off-by: Jun Yi <yijun@loongson.cn>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>

6 files changed, 78 insertions, 81 deletions

diff --git a/arch/loongarch/include/asm/Kbuild b/arch/loongarch/include/asm/Kbuild
index 77ad8e6f0906..6b222f227342 100644
--- a/arch/loongarch/include/asm/Kbuild
+++ b/arch/loongarch/include/asm/Kbuild
@@ -5,7 +5,6 @@ generic-y += mcs_spinlock.h
 generic-y += parport.h
 generic-y += early_ioremap.h
 generic-y += qrwlock.h
-generic-y += qspinlock.h
 generic-y += rwsem.h
 generic-y += segment.h
 generic-y += user.h
diff --git a/arch/loongarch/include/asm/barrier.h b/arch/loongarch/include/asm/barrier.h
index cda977675854..4b663f197706 100644
--- a/arch/loongarch/include/asm/barrier.h
+++ b/arch/loongarch/include/asm/barrier.h
@@ -5,27 +5,56 @@
 #ifndef __ASM_BARRIER_H
 #define __ASM_BARRIER_H
 
-#define __sync()	__asm__ __volatile__("dbar 0" : : : "memory")
+/*
+ * Hint encoding:
+ *
+ * Bit4: ordering or completion (0: completion, 1: ordering)
+ * Bit3: barrier for previous read (0: true, 1: false)
+ * Bit2: barrier for previous write (0: true, 1: false)
+ * Bit1: barrier for succeeding read (0: true, 1: false)
+ * Bit0: barrier for succeeding write (0: true, 1: false)
+ *
+ * Hint 0x700: barrier for "read after read" from the same address
+ */
+
+#define DBAR(hint) __asm__ __volatile__("dbar %0 " : : "I"(hint) : "memory")
+
+#define crwrw		0b00000
+#define cr_r_		0b00101
+#define c_w_w		0b01010
 
-#define fast_wmb()	__sync()
-#define fast_rmb()	__sync()
-#define fast_mb()	__sync()
-#define fast_iob()	__sync()
-#define wbflush()	__sync()
+#define orwrw		0b10000
+#define or_r_		0b10101
+#define o_w_w		0b11010
 
-#define wmb()		fast_wmb()
-#define rmb()		fast_rmb()
-#define mb()		fast_mb()
-#define iob()		fast_iob()
+#define orw_w		0b10010
+#define or_rw		0b10100
 
-#define __smp_mb()	__asm__ __volatile__("dbar 0" : : : "memory")
-#define __smp_rmb()	__asm__ __volatile__("dbar 0" : : : "memory")
-#define __smp_wmb()	__asm__ __volatile__("dbar 0" : : : "memory")
+#define c_sync()	DBAR(crwrw)
+#define c_rsync()	DBAR(cr_r_)
+#define c_wsync()	DBAR(c_w_w)
+
+#define o_sync()	DBAR(orwrw)
+#define o_rsync()	DBAR(or_r_)
+#define o_wsync()	DBAR(o_w_w)
+
+#define ldacq_mb()	DBAR(or_rw)
+#define strel_mb()	DBAR(orw_w)
+
+#define mb()		c_sync()
+#define rmb()		c_rsync()
+#define wmb()		c_wsync()
+#define iob()		c_sync()
+#define wbflush()	c_sync()
+
+#define __smp_mb()	o_sync()
+#define __smp_rmb()	o_rsync()
+#define __smp_wmb()	o_wsync()
 
 #ifdef CONFIG_SMP
-#define __WEAK_LLSC_MB		"	dbar 0  \n"
+#define __WEAK_LLSC_MB		"	dbar 0x700	\n"
 #else
-#define __WEAK_LLSC_MB		"		\n"
+#define __WEAK_LLSC_MB		"			\n"
 #endif
 
 #define __smp_mb__before_atomic()	barrier()
@@ -59,68 +88,19 @@ static inline unsigned long array_index_mask_nospec(unsigned long index,
 	return mask;
 }
 
-#define __smp_load_acquire(p)							\
-({										\
-	union { typeof(*p) __val; char __c[1]; } __u;				\
-	unsigned long __tmp = 0;							\
-	compiletime_assert_atomic_type(*p);					\
-	switch (sizeof(*p)) {							\
-	case 1:									\
-		*(__u8 *)__u.__c = *(volatile __u8 *)p;				\
-		__smp_mb();							\
-		break;								\
-	case 2:									\
-		*(__u16 *)__u.__c = *(volatile __u16 *)p;			\
-		__smp_mb();							\
-		break;								\
-	case 4:									\
-		__asm__ __volatile__(						\
-		"amor_db.w %[val], %[tmp], %[mem]	\n"				\
-		: [val] "=&r" (*(__u32 *)__u.__c)				\
-		: [mem] "ZB" (*(u32 *) p), [tmp] "r" (__tmp)			\
-		: "memory");							\
-		break;								\
-	case 8:									\
-		__asm__ __volatile__(						\
-		"amor_db.d %[val], %[tmp], %[mem]	\n"				\
-		: [val] "=&r" (*(__u64 *)__u.__c)				\
-		: [mem] "ZB" (*(u64 *) p), [tmp] "r" (__tmp)			\
-		: "memory");							\
-		break;								\
-	}									\
-	(typeof(*p))__u.__val;								\
+#define __smp_load_acquire(p)				\
+({							\
+	typeof(*p) ___p1 = READ_ONCE(*p);		\
+	compiletime_assert_atomic_type(*p);		\
+	ldacq_mb();					\
+	___p1;						\
 })
 
-#define __smp_store_release(p, v)						\
-do {										\
-	union { typeof(*p) __val; char __c[1]; } __u =				\
-		{ .__val = (__force typeof(*p)) (v) };				\
-	unsigned long __tmp;							\
-	compiletime_assert_atomic_type(*p);					\
-	switch (sizeof(*p)) {							\
-	case 1:									\
-		__smp_mb();							\
-		*(volatile __u8 *)p = *(__u8 *)__u.__c;				\
-		break;								\
-	case 2:									\
-		__smp_mb();							\
-		*(volatile __u16 *)p = *(__u16 *)__u.__c;			\
-		break;								\
-	case 4:									\
-		__asm__ __volatile__(						\
-		"amswap_db.w %[tmp], %[val], %[mem]	\n"			\
-		: [mem] "+ZB" (*(u32 *)p), [tmp] "=&r" (__tmp)			\
-		: [val] "r" (*(__u32 *)__u.__c)					\
-		: );								\
-		break;								\
-	case 8:									\
-		__asm__ __volatile__(						\
-		"amswap_db.d %[tmp], %[val], %[mem]	\n"			\
-		: [mem] "+ZB" (*(u64 *)p), [tmp] "=&r" (__tmp)			\
-		: [val] "r" (*(__u64 *)__u.__c)					\
-		: );								\
-		break;								\
-	}									\
+#define __smp_store_release(p, v)			\
+do {							\
+	compiletime_assert_atomic_type(*p);		\
+	strel_mb();					\
+	WRITE_ONCE(*p, v);				\
 } while (0)
 
 #define __smp_store_mb(p, v)							\
diff --git a/arch/loongarch/include/asm/io.h b/arch/loongarch/include/asm/io.h
index 545e2708fbf7..1c9410220040 100644
--- a/arch/loongarch/include/asm/io.h
+++ b/arch/loongarch/include/asm/io.h
@@ -62,7 +62,7 @@ extern pgprot_t pgprot_wc;
 #define ioremap_cache(offset, size)	\
 	ioremap_prot((offset), (size), pgprot_val(PAGE_KERNEL))
 
-#define mmiowb() asm volatile ("dbar 0" ::: "memory")
+#define mmiowb() wmb()
 
 /*
  * String version of I/O memory access operations.
diff --git a/arch/loongarch/include/asm/qspinlock.h b/arch/loongarch/include/asm/qspinlock.h
new file mode 100644
index 000000000000..34f43f8ad591
--- /dev/null
+++ b/arch/loongarch/include/asm/qspinlock.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _ASM_QSPINLOCK_H
+#define _ASM_QSPINLOCK_H
+
+#include <asm-generic/qspinlock_types.h>
+
+#define queued_spin_unlock queued_spin_unlock
+
+static inline void queued_spin_unlock(struct qspinlock *lock)
+{
+	compiletime_assert_atomic_type(lock->locked);
+	c_sync();
+	WRITE_ONCE(lock->locked, 0);
+}
+
+#include <asm-generic/qspinlock.h>
+
+#endif /* _ASM_QSPINLOCK_H */
diff --git a/arch/loongarch/kernel/smp.c b/arch/loongarch/kernel/smp.c
index 255967ff8c36..8ea1bbcf13a7 100644
--- a/arch/loongarch/kernel/smp.c
+++ b/arch/loongarch/kernel/smp.c
@@ -115,7 +115,7 @@ static u32 ipi_read_clear(int cpu)
 	action = iocsr_read32(LOONGARCH_IOCSR_IPI_STATUS);
 	/* Clear the ipi register to clear the interrupt */
 	iocsr_write32(action, LOONGARCH_IOCSR_IPI_CLEAR);
-	smp_mb();
+	wbflush();
 
 	return action;
 }
diff --git a/arch/loongarch/mm/tlbex.S b/arch/loongarch/mm/tlbex.S
index 244e2f5aeee5..240ced55586e 100644
--- a/arch/loongarch/mm/tlbex.S
+++ b/arch/loongarch/mm/tlbex.S
@@ -184,7 +184,7 @@ tlb_huge_update_load:
 	ertn
 
 nopage_tlb_load:
-	dbar		0
+	dbar		0x700
 	csrrd		ra, EXCEPTION_KS2
 	la_abs		t0, tlb_do_page_fault_0
 	jr		t0
@@ -333,7 +333,7 @@ tlb_huge_update_store:
 	ertn
 
 nopage_tlb_store:
-	dbar		0
+	dbar		0x700
 	csrrd		ra, EXCEPTION_KS2
 	la_abs		t0, tlb_do_page_fault_1
 	jr		t0
@@ -480,7 +480,7 @@ tlb_huge_update_modify:
 	ertn
 
 nopage_tlb_modify:
-	dbar		0
+	dbar		0x700
 	csrrd		ra, EXCEPTION_KS2
 	la_abs		t0, tlb_do_page_fault_1
 	jr		t0