diff mbox series

[v4,3/4] riscv/barrier: Consolidate fence definitions

Message ID 20240213201210.2549488-1-ericchancf@google.com
State New
Headers
Series riscv/barrier: tidying up barrier-related macro |

Commit Message

Eric Chan Feb. 13, 2024, 8:12 p.m. UTC 
  Disparate fence implementations are consolidated into fence.h.
Also introduce RISCV_FENCE_ASM to make fence macro more reusable.

Signed-off-by: Eric Chan <ericchancf@google.com>
---
v3 -> v4: fix the form that can pass the checking of checkpatch.pl.
 arch/riscv/include/asm/atomic.h  |  8 ++------
 arch/riscv/include/asm/barrier.h |  3 +--
 arch/riscv/include/asm/cmpxchg.h |  1 -
 arch/riscv/include/asm/fence.h   | 10 +++++++---
 arch/riscv/include/asm/io.h      |  8 ++++----
 arch/riscv/include/asm/mmio.h    |  5 +++--
 arch/riscv/include/asm/mmiowb.h  |  2 +-
 7 files changed, 18 insertions(+), 19 deletions(-)

--
2.43.0.687.g38aa6559b0-goog

Comments

Samuel Holland Feb. 13, 2024, 9:31 p.m. UTC | #1 
Hi Eric,

On 2024-02-13 2:12 PM, Eric Chan wrote:
> Disparate fence implementations are consolidated into fence.h.
> Also introduce RISCV_FENCE_ASM to make fence macro more reusable.
> 
> Signed-off-by: Eric Chan <ericchancf@google.com>
> ---
> v3 -> v4: fix the form that can pass the checking of checkpatch.pl.
>  arch/riscv/include/asm/atomic.h  |  8 ++------
>  arch/riscv/include/asm/barrier.h |  3 +--
>  arch/riscv/include/asm/cmpxchg.h |  1 -
>  arch/riscv/include/asm/fence.h   | 10 +++++++---
>  arch/riscv/include/asm/io.h      |  8 ++++----
>  arch/riscv/include/asm/mmio.h    |  5 +++--
>  arch/riscv/include/asm/mmiowb.h  |  2 +-
>  7 files changed, 18 insertions(+), 19 deletions(-)
> 
> diff --git a/arch/riscv/include/asm/atomic.h b/arch/riscv/include/asm/atomic.h
> index 31e6e2e7cc18..19050d13b6c1 100644
> --- a/arch/riscv/include/asm/atomic.h
> +++ b/arch/riscv/include/asm/atomic.h
> @@ -17,13 +17,9 @@
>  #endif
> 
>  #include <asm/cmpxchg.h>
> -#include <asm/barrier.h>
> 
> -#define __atomic_acquire_fence()					\
> -	__asm__ __volatile__(RISCV_ACQUIRE_BARRIER "" ::: "memory")
> -
> -#define __atomic_release_fence()					\
> -	__asm__ __volatile__(RISCV_RELEASE_BARRIER "" ::: "memory");
> +#define __atomic_acquire_fence()	RISCV_FENCE(r, rw)
> +#define __atomic_release_fence()	RISCV_FENCE(rw, r)

This change removes the omit-the-fence-on-uniprocessor optimization, like you
added some instances of in your previous patch. Should we keep this optimization?

Regards,
Samuel
Eric Chan Feb. 13, 2024, 10:43 p.m. UTC | #2 
Hi Samuel,

  Thanks for pointing out the negligence, I think we should keep the optimization.
  I've updated patch v5 at [0].
  Thank for the reviewing and suggestions.

[0] [https://lore.kernel.org/lkml/20240213223810.2595804-1-ericchancf@google.com/]

Sincerely,
  Eric Chan
diff mbox series

Patch

diff --git a/arch/riscv/include/asm/atomic.h b/arch/riscv/include/asm/atomic.h
index 31e6e2e7cc18..19050d13b6c1 100644
--- a/arch/riscv/include/asm/atomic.h
+++ b/arch/riscv/include/asm/atomic.h
@@ -17,13 +17,9 @@ 
 #endif

 #include <asm/cmpxchg.h>
-#include <asm/barrier.h>

-#define __atomic_acquire_fence()					\
-	__asm__ __volatile__(RISCV_ACQUIRE_BARRIER "" ::: "memory")
-
-#define __atomic_release_fence()					\
-	__asm__ __volatile__(RISCV_RELEASE_BARRIER "" ::: "memory");
+#define __atomic_acquire_fence()	RISCV_FENCE(r, rw)
+#define __atomic_release_fence()	RISCV_FENCE(rw, r)

 static __always_inline int arch_atomic_read(const atomic_t *v)
 {
diff --git a/arch/riscv/include/asm/barrier.h b/arch/riscv/include/asm/barrier.h
index 4c49a8ff2c68..4f4743d7440d 100644
--- a/arch/riscv/include/asm/barrier.h
+++ b/arch/riscv/include/asm/barrier.h
@@ -11,13 +11,12 @@ 
 #define _ASM_RISCV_BARRIER_H

 #ifndef __ASSEMBLY__
+#include <asm/fence.h>

 #define nop()		__asm__ __volatile__ ("nop")
 #define __nops(n)	".rept	" #n "\nnop\n.endr\n"
 #define nops(n)		__asm__ __volatile__ (__nops(n))

-#define RISCV_FENCE(p, s) \
-	__asm__ __volatile__ ("fence " #p "," #s : : : "memory")

 /* These barriers need to enforce ordering on both devices or memory. */
 #define __mb()		RISCV_FENCE(iorw,iorw)
diff --git a/arch/riscv/include/asm/cmpxchg.h b/arch/riscv/include/asm/cmpxchg.h
index a608e4d1a0a4..2fee65cc8443 100644
--- a/arch/riscv/include/asm/cmpxchg.h
+++ b/arch/riscv/include/asm/cmpxchg.h
@@ -8,7 +8,6 @@ 

 #include <linux/bug.h>

-#include <asm/barrier.h>
 #include <asm/fence.h>

 #define __xchg_relaxed(ptr, new, size)					\
diff --git a/arch/riscv/include/asm/fence.h b/arch/riscv/include/asm/fence.h
index 6c26c44dfcd6..5b46f96a3ec8 100644
--- a/arch/riscv/include/asm/fence.h
+++ b/arch/riscv/include/asm/fence.h
@@ -1,10 +1,14 @@ 
 #ifndef _ASM_RISCV_FENCE_H
 #define _ASM_RISCV_FENCE_H

+#define RISCV_FENCE_ASM(p, s)		"\tfence " #p "," #s "\n"
+#define RISCV_FENCE(p, s) \
+	({ __asm__ __volatile__ (RISCV_FENCE_ASM(p, s) : : : "memory"); })
+
 #ifdef CONFIG_SMP
-#define RISCV_ACQUIRE_BARRIER		"\tfence r , rw\n"
-#define RISCV_RELEASE_BARRIER		"\tfence rw,  w\n"
-#define RISCV_FULL_BARRIER		"\tfence rw, rw\n"
+#define RISCV_ACQUIRE_BARRIER		RISCV_FENCE_ASM(r, rw)
+#define RISCV_RELEASE_BARRIER		RISCV_FENCE_ASM(rw, r)
+#define RISCV_FULL_BARRIER		RISCV_FENCE_ASM(rw, rw)
 #else
 #define RISCV_ACQUIRE_BARRIER
 #define RISCV_RELEASE_BARRIER
diff --git a/arch/riscv/include/asm/io.h b/arch/riscv/include/asm/io.h
index 42497d487a17..1c5c641075d2 100644
--- a/arch/riscv/include/asm/io.h
+++ b/arch/riscv/include/asm/io.h
@@ -47,10 +47,10 @@ 
  * sufficient to ensure this works sanely on controllers that support I/O
  * writes.
  */
-#define __io_pbr()	__asm__ __volatile__ ("fence io,i"  : : : "memory");
-#define __io_par(v)	__asm__ __volatile__ ("fence i,ior" : : : "memory");
-#define __io_pbw()	__asm__ __volatile__ ("fence iow,o" : : : "memory");
-#define __io_paw()	__asm__ __volatile__ ("fence o,io"  : : : "memory");
+#define __io_pbr()	RISCV_FENCE(io, i)
+#define __io_par(v)	RISCV_FENCE(i, ior)
+#define __io_pbw()	RISCV_FENCE(iow, o)
+#define __io_paw()	RISCV_FENCE(o, io)

 /*
  * Accesses from a single hart to a single I/O address must be ordered.  This
diff --git a/arch/riscv/include/asm/mmio.h b/arch/riscv/include/asm/mmio.h
index 4c58ee7f95ec..06cadfd7a237 100644
--- a/arch/riscv/include/asm/mmio.h
+++ b/arch/riscv/include/asm/mmio.h
@@ -12,6 +12,7 @@ 
 #define _ASM_RISCV_MMIO_H

 #include <linux/types.h>
+#include <asm/fence.h>
 #include <asm/mmiowb.h>

 /* Generic IO read/write.  These perform native-endian accesses. */
@@ -131,8 +132,8 @@  static inline u64 __raw_readq(const volatile void __iomem *addr)
  * doesn't define any ordering between the memory space and the I/O space.
  */
 #define __io_br()	do {} while (0)
-#define __io_ar(v)	({ __asm__ __volatile__ ("fence i,ir" : : : "memory"); })
-#define __io_bw()	({ __asm__ __volatile__ ("fence w,o" : : : "memory"); })
+#define __io_ar(v)	RISCV_FENCE(i, ir)
+#define __io_bw()	RISCV_FENCE(w, o)
 #define __io_aw()	mmiowb_set_pending()

 #define readb(c)	({ u8  __v; __io_br(); __v = readb_cpu(c); __io_ar(__v); __v; })
diff --git a/arch/riscv/include/asm/mmiowb.h b/arch/riscv/include/asm/mmiowb.h
index 0b2333e71fdc..52ce4a399d9b 100644
--- a/arch/riscv/include/asm/mmiowb.h
+++ b/arch/riscv/include/asm/mmiowb.h
@@ -7,7 +7,7 @@ 
  * "o,w" is sufficient to ensure that all writes to the device have completed
  * before the write to the spinlock is allowed to commit.
  */
-#define mmiowb()	__asm__ __volatile__ ("fence o,w" : : : "memory");
+#define mmiowb()	RISCV_FENCE(o, w)

 #include <linux/smp.h>
 #include <asm-generic/mmiowb.h>