The existing arch_xchg/arch_cmpxchg operations are implemented by inserting fence instructions before or after atomic instructions. This commit replaces them with real acquire/release semantics.
|----------------------------------------------------------------|
| | arch_xchg_release | arch_cmpxchg_release |
| |-----------------------------------------------------------|
| | zabha | !zabha | zabha+zacas | !(zabha+zacas) |
| rl |-----------------------------------------------------------|
| | | (fence rw, w) | | (fence rw, w) |
| | amoswap.rl | lr.w | amocas.rl | lr.w |
| | | sc.w.rl | | sc.w.rl |
|----------------------------------------------------------------|
| | arch_xchg_acquire | arch_cmpxchg_acquire |
| |-----------------------------------------------------------|
| | zabha | !zabha | zabha+zacas | !(zabha+zacas) |
| aq |-----------------------------------------------------------|
| | | lr.w.aq | | lr.w.aq |
| | amoswap.aq | sc.w | amocas.aq | sc.w |
| | | (fence r, rw) | | (fence r, rw) |
|----------------------------------------------------------------|
(fence rw, w), (fence r, rw) here means such instructions will only
be inserted when zalasr is not implemented.
Signed-off-by: Xu Lu <luxu.ker...@bytedance.com>
---
arch/riscv/include/asm/atomic.h | 6 --
arch/riscv/include/asm/cmpxchg.h | 136 ++++++++++++++-----------------
2 files changed, 63 insertions(+), 79 deletions(-)
diff --git a/arch/riscv/include/asm/atomic.h b/arch/riscv/include/asm/atomic.h
index 5b96c2f61adb5..b79a4f889f339 100644
--- a/arch/riscv/include/asm/atomic.h
+++ b/arch/riscv/include/asm/atomic.h
@@ -18,12 +18,6 @@
#include <asm/cmpxchg.h>
-#define __atomic_acquire_fence() \
- __asm__ __volatile__(RISCV_ACQUIRE_BARRIER "" ::: "memory")
-
-#define __atomic_release_fence() \
- __asm__ __volatile__(RISCV_RELEASE_BARRIER "" ::: "memory");
-
static __always_inline int arch_atomic_read(const atomic_t *v)
{
return READ_ONCE(v->counter);
diff --git a/arch/riscv/include/asm/cmpxchg.h b/arch/riscv/include/asm/cmpxchg.h
index 0b749e7102162..207fdba38d1fc 100644
--- a/arch/riscv/include/asm/cmpxchg.h
+++ b/arch/riscv/include/asm/cmpxchg.h
@@ -15,15 +15,23 @@
#include <asm/cpufeature-macros.h>
#include <asm/processor.h>
-#define __arch_xchg_masked(sc_sfx, swap_sfx, prepend, sc_append,
\
- swap_append, r, p, n)
\
+/*
+ * These macros are here to improve the readability of the arch_xchg_XXX()
+ * and arch_cmpxchg_XXX() macros.
+ */
+#define LR_SFX(x) x
+#define SC_SFX(x) x
+#define CAS_SFX(x) x
+#define SC_PREPEND(x) x
+#define SC_APPEND(x) x
+
+#define __arch_xchg_masked(lr_sfx, sc_sfx, swap_sfx, sc_prepend, sc_append,
\
+ r, p, n)
\
({
\
if (IS_ENABLED(CONFIG_RISCV_ISA_ZABHA) &&
\
riscv_has_extension_unlikely(RISCV_ISA_EXT_ZABHA)) {
\
__asm__ __volatile__ (
\
- prepend
\
" amoswap" swap_sfx " %0, %z2, %1\n"
\
- swap_append
\
: "=&r" (r), "+A" (*(p))
\
: "rJ" (n)
\
: "memory");
\
@@ -37,14 +45,16 @@
ulong __rc;
\
\
__asm__ __volatile__ (
\
- prepend
\
PREFETCHW_ASM(%5)
\
+ ALTERNATIVE(__nops(1), sc_prepend,
\
+ 0, RISCV_ISA_EXT_ZALASR, 1)
\
"0: lr.w %0, %2\n"
\
" and %1, %0, %z4\n"
\
" or %1, %1, %z3\n"
\
" sc.w" sc_sfx " %1, %1, %2\n"
\
" bnez %1, 0b\n"
\
- sc_append
\
+ ALTERNATIVE(__nops(1), sc_append,
\
+ 0, RISCV_ISA_EXT_ZALASR, 1)
\
: "=&r" (__retx), "=&r" (__rc), "+A" (*(__ptr32b))
\
: "rJ" (__newx), "rJ" (~__mask), "rJ" (__ptr32b)
\
: "memory");
\
@@ -53,19 +63,17 @@
}
\
})
-#define __arch_xchg(sfx, prepend, append, r, p, n) \
+#define __arch_xchg(sfx, r, p, n) \
({ \
__asm__ __volatile__ ( \
- prepend \
" amoswap" sfx " %0, %2, %1\n" \
- append \
: "=r" (r), "+A" (*(p)) \
: "r" (n) \
: "memory"); \
})
-#define _arch_xchg(ptr, new, sc_sfx, swap_sfx, prepend,
\
- sc_append, swap_append) \
+#define _arch_xchg(ptr, new, lr_sfx, sc_sfx, swap_sfx, \
+ sc_prepend, sc_append) \
({ \
__typeof__(ptr) __ptr = (ptr); \
__typeof__(*(__ptr)) __new = (new); \
@@ -73,22 +81,20 @@
\
switch (sizeof(*__ptr)) { \
case 1: \
- __arch_xchg_masked(sc_sfx, ".b" swap_sfx, \
- prepend, sc_append, swap_append, \
+ __arch_xchg_masked(lr_sfx, sc_sfx, ".b" swap_sfx, \
+ sc_prepend, sc_append, \
__ret, __ptr, __new); \
break; \
case 2: \
- __arch_xchg_masked(sc_sfx, ".h" swap_sfx, \
- prepend, sc_append, swap_append, \
+ __arch_xchg_masked(lr_sfx, sc_sfx, ".h" swap_sfx, \
+ sc_prepend, sc_append, \
__ret, __ptr, __new); \
break; \
case 4: \
- __arch_xchg(".w" swap_sfx, prepend, swap_append, \
- __ret, __ptr, __new); \
+ __arch_xchg(".w" swap_sfx, __ret, __ptr, __new); \
break; \
case 8: \
- __arch_xchg(".d" swap_sfx, prepend, swap_append, \
- __ret, __ptr, __new); \
+ __arch_xchg(".d" swap_sfx, __ret, __ptr, __new); \
break; \
default: \
BUILD_BUG(); \
@@ -97,17 +103,23 @@
})
#define arch_xchg_relaxed(ptr, x) \
- _arch_xchg(ptr, x, "", "", "", "", "")
+ _arch_xchg(ptr, x, LR_SFX(""), SC_SFX(""), CAS_SFX(""), \
+ SC_PREPEND(__nops(1)), SC_APPEND(__nops(1)))
#define arch_xchg_acquire(ptr, x) \
- _arch_xchg(ptr, x, "", "", "", \
- RISCV_ACQUIRE_BARRIER, RISCV_ACQUIRE_BARRIER)
+ _arch_xchg(ptr, x, LR_SFX(".aq"), SC_SFX(""), CAS_SFX(".aq"), \
+ SC_PREPEND(__nops(1)), \
+ SC_APPEND(RISCV_ACQUIRE_BARRIER))
#define arch_xchg_release(ptr, x) \
- _arch_xchg(ptr, x, "", "", RISCV_RELEASE_BARRIER, "", "")
+ _arch_xchg(ptr, x, LR_SFX(""), SC_SFX(".rl"), CAS_SFX(".rl"), \
+ SC_PREPEND(RISCV_RELEASE_BARRIER), \
+ SC_APPEND(__nops(1)))
#define arch_xchg(ptr, x) \
- _arch_xchg(ptr, x, ".rl", ".aqrl", "", RISCV_FULL_BARRIER, "")
+ _arch_xchg(ptr, x, LR_SFX(""), SC_SFX(".aqrl"), \
+ CAS_SFX(".aqrl"), SC_PREPEND(__nops(1)), \
+ SC_APPEND(__nops(1)))
#define xchg32(ptr, x) \
({ \
@@ -126,9 +138,7 @@
* store NEW in MEM. Return the initial value in MEM. Success is
* indicated by comparing RETURN with OLD.
*/
-#define __arch_cmpxchg_masked(sc_sfx, cas_sfx,
\
- sc_prepend, sc_append,
\
- cas_prepend, cas_append,
\
+#define __arch_cmpxchg_masked(lr_sfx, sc_sfx, cas_sfx, sc_prepend, sc_append,
\
r, p, o, n)
\
({
\
if (IS_ENABLED(CONFIG_RISCV_ISA_ZABHA) &&
\
@@ -138,9 +148,7 @@
r = o;
\
\
__asm__ __volatile__ (
\
- cas_prepend
\
" amocas" cas_sfx " %0, %z2, %1\n"
\
- cas_append
\
: "+&r" (r), "+A" (*(p))
\
: "rJ" (n)
\
: "memory");
\
@@ -155,15 +163,17 @@
ulong __rc;
\
\
__asm__ __volatile__ (
\
- sc_prepend
\
- "0: lr.w %0, %2\n"
\
+ ALTERNATIVE(__nops(1), sc_prepend,
\
+ 0, RISCV_ISA_EXT_ZALASR, 1)
\
+ "0: lr.w" lr_sfx " %0, %2\n"
\
" and %1, %0, %z5\n"
\
" bne %1, %z3, 1f\n"
\
" and %1, %0, %z6\n"
\
" or %1, %1, %z4\n"
\
" sc.w" sc_sfx " %1, %1, %2\n"
\
" bnez %1, 0b\n"
\
- sc_append
\
+ ALTERNATIVE(__nops(1), sc_append,
\
+ 0, RISCV_ISA_EXT_ZALASR, 1)
\
"1:\n"
\
: "=&r" (__retx), "=&r" (__rc), "+A" (*(__ptr32b))
\
: "rJ" ((long)__oldx), "rJ" (__newx),
\
@@ -174,9 +184,7 @@
}
\
})
-#define __arch_cmpxchg(lr_sfx, sc_sfx, cas_sfx,
\
- sc_prepend, sc_append, \
- cas_prepend, cas_append, \
+#define __arch_cmpxchg(lr_sfx, sc_sfx, cas_sfx, sc_prepend, sc_append,
\
r, p, co, o, n) \
({ \
if (IS_ENABLED(CONFIG_RISCV_ISA_ZACAS) && \
@@ -184,9 +192,7 @@
r = o; \
\
__asm__ __volatile__ ( \
- cas_prepend \
" amocas" cas_sfx " %0, %z2, %1\n" \
- cas_append \
: "+&r" (r), "+A" (*(p)) \
: "rJ" (n) \
: "memory"); \
@@ -194,12 +200,14 @@
register unsigned int __rc; \
\
__asm__ __volatile__ ( \
- sc_prepend \
+ ALTERNATIVE(__nops(1), sc_prepend, \
+ 0, RISCV_ISA_EXT_ZALASR, 1) \
"0: lr" lr_sfx " %0, %2\n" \
" bne %0, %z3, 1f\n" \
" sc" sc_sfx " %1, %z4, %2\n" \
" bnez %1, 0b\n" \
- sc_append \
+ ALTERNATIVE(__nops(1), sc_append, \
+ 0, RISCV_ISA_EXT_ZALASR, 1) \
"1:\n" \
: "=&r" (r), "=&r" (__rc), "+A" (*(p)) \
: "rJ" (co o), "rJ" (n) \
@@ -207,9 +215,8 @@
} \
})
-#define _arch_cmpxchg(ptr, old, new, sc_sfx, cas_sfx, \
- sc_prepend, sc_append, \
- cas_prepend, cas_append) \
+#define _arch_cmpxchg(ptr, old, new, lr_sfx, sc_sfx, cas_sfx, \
+ sc_prepend, sc_append) \
({ \
__typeof__(ptr) __ptr = (ptr); \
__typeof__(*(__ptr)) __old = (old); \
@@ -218,27 +225,23 @@
\
switch (sizeof(*__ptr)) { \
case 1: \
- __arch_cmpxchg_masked(sc_sfx, ".b" cas_sfx, \
+ __arch_cmpxchg_masked(lr_sfx, sc_sfx, ".b" cas_sfx, \
sc_prepend, sc_append, \
- cas_prepend, cas_append, \
__ret, __ptr, __old, __new); \
break; \
case 2: \
- __arch_cmpxchg_masked(sc_sfx, ".h" cas_sfx, \
+ __arch_cmpxchg_masked(lr_sfx, sc_sfx, ".h" cas_sfx, \
sc_prepend, sc_append, \
- cas_prepend, cas_append, \
__ret, __ptr, __old, __new); \
break; \
case 4: \
- __arch_cmpxchg(".w", ".w" sc_sfx, ".w" cas_sfx, \
+ __arch_cmpxchg(".w" lr_sfx, ".w" sc_sfx, ".w" cas_sfx, \
sc_prepend, sc_append, \
- cas_prepend, cas_append, \
__ret, __ptr, (long)(int)(long), __old, __new);
\
break; \
case 8: \
- __arch_cmpxchg(".d", ".d" sc_sfx, ".d" cas_sfx, \
+ __arch_cmpxchg(".d" lr_sfx, ".d" sc_sfx, ".d" cas_sfx, \
sc_prepend, sc_append, \
- cas_prepend, cas_append, \
__ret, __ptr, /**/, __old, __new); \
break; \
default: \
@@ -247,40 +250,27 @@
(__typeof__(*(__ptr)))__ret; \
})
-/*
- * These macros are here to improve the readability of the arch_cmpxchg_XXX()
- * macros.
- */
-#define SC_SFX(x) x
-#define CAS_SFX(x) x
-#define SC_PREPEND(x) x
-#define SC_APPEND(x) x
-#define CAS_PREPEND(x) x
-#define CAS_APPEND(x) x
-
#define arch_cmpxchg_relaxed(ptr, o, n)
\
_arch_cmpxchg((ptr), (o), (n), \
- SC_SFX(""), CAS_SFX(""), \
- SC_PREPEND(""), SC_APPEND(""), \
- CAS_PREPEND(""), CAS_APPEND(""))
+ LR_SFX(""), SC_SFX(""), CAS_SFX(""), \
+ SC_PREPEND(__nops(1)), SC_APPEND(__nops(1)))
#define arch_cmpxchg_acquire(ptr, o, n)
\
_arch_cmpxchg((ptr), (o), (n), \
- SC_SFX(""), CAS_SFX(""), \
- SC_PREPEND(""), SC_APPEND(RISCV_ACQUIRE_BARRIER), \
- CAS_PREPEND(""), CAS_APPEND(RISCV_ACQUIRE_BARRIER))
+ LR_SFX(".aq"), SC_SFX(""), CAS_SFX(".aq"), \
+ SC_PREPEND(__nops(1)), \
+ SC_APPEND(RISCV_ACQUIRE_BARRIER))
#define arch_cmpxchg_release(ptr, o, n)
\
_arch_cmpxchg((ptr), (o), (n), \
- SC_SFX(""), CAS_SFX(""), \
- SC_PREPEND(RISCV_RELEASE_BARRIER), SC_APPEND(""), \
- CAS_PREPEND(RISCV_RELEASE_BARRIER), CAS_APPEND(""))
+ LR_SFX(""), SC_SFX(".rl"), CAS_SFX(".rl"), \
+ SC_PREPEND(RISCV_RELEASE_BARRIER), \
+ SC_APPEND(__nops(1)))
#define arch_cmpxchg(ptr, o, n)
\
_arch_cmpxchg((ptr), (o), (n), \
- SC_SFX(".rl"), CAS_SFX(".aqrl"), \
- SC_PREPEND(""), SC_APPEND(RISCV_FULL_BARRIER), \
- CAS_PREPEND(""), CAS_APPEND(""))
+ LR_SFX(""), SC_SFX(".aqrl"), CAS_SFX(".aqrl"), \
+ SC_PREPEND(__nops(1)), SC_APPEND(__nops(1)))
#define arch_cmpxchg_local(ptr, o, n) \
arch_cmpxchg_relaxed((ptr), (o), (n))
--
2.20.1
