We must use 'volatile' in C-version read/write IO accessors implementation to avoid merging several reads (writes) into one read (write), or optimizing them out by compiler.

Fixes commit 07906b3dad15 ("ARC: Switch to generic accessors")

Signed-off-by: Eugeniy Paltsev Eugeniy.Paltsev@synopsys.com --- arch/arc/include/asm/io.h | 24 +++++++++++++++--------- 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/arch/arc/include/asm/io.h b/arch/arc/include/asm/io.h index 70d050590de..1f1ae889f3a 100644 --- a/arch/arc/include/asm/io.h +++ b/arch/arc/include/asm/io.h @@ -54,15 +54,21 @@ static inline void sync(void) /* Not yet implemented */ }

-#define __arch_getb(a) (*(unsigned char *)(a)) -#define __arch_getw(a) (*(unsigned short *)(a)) -#define __arch_getl(a) (*(unsigned int *)(a)) -#define __arch_getq(a) (*(unsigned long long *)(a)) - -#define __arch_putb(v, a) (*(unsigned char *)(a) = (v)) -#define __arch_putw(v, a) (*(unsigned short *)(a) = (v)) -#define __arch_putl(v, a) (*(unsigned int *)(a) = (v)) -#define __arch_putq(v, a) (*(unsigned long long *)(a) = (v)) +/* + * We must use 'volatile' in C-version read/write IO accessors implementation + * to avoid merging several reads (writes) into one read (write), or optimizing + * them out by compiler. + */ +#define __arch_getb(a) (*(volatile u8 *)(a)) +#define __arch_getw(a) (*(volatile u16 *)(a)) +#define __arch_getl(a) (*(volatile u32 *)(a)) +#define __arch_getq(a) (*(volatile u64 *)(a)) + +#define __arch_putb(v, a) (*(volatile u8 *)(a) = (v)) +#define __arch_putw(v, a) (*(volatile u16 *)(a) = (v)) +#define __arch_putl(v, a) (*(volatile u32 *)(a) = (v)) +#define __arch_putq(v, a) (*(volatile u64 *)(a) = (v)) +

#define __raw_writeb(v, a) __arch_putb(v, a) #define __raw_writew(v, a) __arch_putw(v, a)

We add memory barriers for __raw_readX / __raw_writeX accessors same way as it is done for readX and writeX accessors as lots of U-boot driver uses __raw_readX / __raw_writeX instead of proper accessor with barrier.

It will save us from lot's of debugging in the future and it is OK as U-Boot is not that performance oriented as real run-time software like OS or user bare-metal app so we may afford being not super fast as we only being executed once.

Signed-off-by: Eugeniy Paltsev Eugeniy.Paltsev@synopsys.com --- arch/arc/include/asm/io.h | 58 +++++++++++++++++++++++---------------- 1 file changed, 35 insertions(+), 23 deletions(-)

diff --git a/arch/arc/include/asm/io.h b/arch/arc/include/asm/io.h index 50c88456002..e770a1fe4b8 100644 --- a/arch/arc/include/asm/io.h +++ b/arch/arc/include/asm/io.h @@ -79,21 +79,29 @@ static inline void sync(void) #define __arch_putq(v, a) ({ __comp_b(); *(volatile u64 *)(a) = (v); __comp_b(); })

-#define __raw_writeb(v, a) __arch_putb(v, a) -#define __raw_writew(v, a) __arch_putw(v, a) -#define __raw_writel(v, a) __arch_putl(v, a) -#define __raw_writeq(v, a) __arch_putq(v, a) +/* + * We add memory barriers for __raw_readX / __raw_writeX accessors same way as + * it is done for readX and writeX accessors as lots of U-boot driver uses + * __raw_readX / __raw_writeX instead of proper accessor with barrier. + */ +#define __raw_writeb(v, c) ({ __iowmb(); __arch_putb(v, c); }) +#define __raw_writew(v, c) ({ __iowmb(); __arch_putw(v, c); }) +#define __raw_writel(v, c) ({ __iowmb(); __arch_putl(v, c); }) +#define __raw_writeq(v, c) ({ __iowmb(); __arch_putq(v, c); }) + +#define __raw_readb(c) ({ u8 __v = __arch_getb(c); __iormb(); __v; }) +#define __raw_readw(c) ({ u16 __v = __arch_getw(c); __iormb(); __v; }) +#define __raw_readl(c) ({ u32 __v = __arch_getl(c); __iormb(); __v; }) +#define __raw_readq(c) ({ u64 __v = __arch_getq(c); __iormb(); __v; })

-#define __raw_readb(a) __arch_getb(a) -#define __raw_readw(a) __arch_getw(a) -#define __raw_readl(a) __arch_getl(a) -#define __raw_readq(a) __arch_getq(a)

static inline void __raw_writesb(unsigned long addr, const void *data, int bytelen) { u8 *buf = (uint8_t *)data;

+ __iowmb(); + while (bytelen--) __arch_putb(*buf++, addr); } @@ -103,6 +111,8 @@ static inline void __raw_writesw(unsigned long addr, const void *data, { u16 *buf = (uint16_t *)data;

+ __iowmb(); + while (wordlen--) __arch_putw(*buf++, addr); } @@ -112,6 +122,8 @@ static inline void __raw_writesl(unsigned long addr, const void *data, { u32 *buf = (uint32_t *)data;

+ __iowmb(); + while (longlen--) __arch_putl(*buf++, addr); } @@ -122,6 +134,8 @@ static inline void __raw_readsb(unsigned long addr, void *data, int bytelen)

while (bytelen--) *buf++ = __arch_getb(addr); + + __iormb(); }

static inline void __raw_readsw(unsigned long addr, void *data, int wordlen) @@ -130,6 +144,8 @@ static inline void __raw_readsw(unsigned long addr, void *data, int wordlen)

while (wordlen--) *buf++ = __arch_getw(addr); + + __iormb(); }

static inline void __raw_readsl(unsigned long addr, void *data, int longlen) @@ -138,6 +154,8 @@ static inline void __raw_readsl(unsigned long addr, void *data, int longlen)

while (longlen--) *buf++ = __arch_getl(addr); + + __iormb(); }

/* @@ -145,21 +163,15 @@ static inline void __raw_readsl(unsigned long addr, void *data, int longlen) * ordering rules but do not guarantee any ordering relative to Normal memory * accesses. */ -#define readb_relaxed(c) ({ u8 __r = __raw_readb(c); __r; }) -#define readw_relaxed(c) ({ u16 __r = le16_to_cpu((__force __le16) \ - __raw_readw(c)); __r; }) -#define readl_relaxed(c) ({ u32 __r = le32_to_cpu((__force __le32) \ - __raw_readl(c)); __r; }) -#define readq_relaxed(c) ({ u64 __r = le64_to_cpu((__force __le64) \ - __raw_readq(c)); __r; }) - -#define writeb_relaxed(v, c) ((void)__raw_writeb((v), (c))) -#define writew_relaxed(v, c) ((void)__raw_writew((__force u16) \ - cpu_to_le16(v), (c))) -#define writel_relaxed(v, c) ((void)__raw_writel((__force u32) \ - cpu_to_le32(v), (c))) -#define writeq_relaxed(v, c) ((void)__raw_writeq((__force u64) \ - cpu_to_le64(v), (c))) +#define readb_relaxed(c) ({ u8 __r = __arch_getb(c); __r; }) +#define readw_relaxed(c) ({ u16 __r = le16_to_cpu((__force __le16)__arch_getw(c)); __r; }) +#define readl_relaxed(c) ({ u32 __r = le32_to_cpu((__force __le32)__arch_getl(c)); __r; }) +#define readq_relaxed(c) ({ u64 __r = le64_to_cpu((__force __le64)__arch_getq(c)); __r; }) + +#define writeb_relaxed(v, c) ((void)__arch_putb((v), (c))) +#define writew_relaxed(v, c) ((void)__arch_putw((__force u16)cpu_to_le16(v), (c))) +#define writel_relaxed(v, c) ((void)__arch_putl((__force u32)cpu_to_le32(v), (c))) +#define writeq_relaxed(v, c) ((void)__arch_putq((__force u64)cpu_to_le64(v), (c)))

/* * MMIO can also get buffered/optimized in micro-arch, so barriers needed