[U-Boot] [PATCH] ARM: fix write*() I/O accessors

Commit 3c0659b "ARM: Avoid compiler optimization for readb, writeb and friends." introduced I/O accessors with memory barriers. Unfortunately the new write*() accessors introduced a bug:
The problem is that the argument "v" gets evaluated twice. This breaks code like used here (from "drivers/net/dnet.c"):
for (i = 0; i < wrsz; i++) writel(*bufp++, &dnet->regs->TX_DATA_FIFO);
Use auxiliary variables to avoid such problems.
Signed-off-by: Wolfgang Denk wd@denx.de Cc: Albert Aribaud albert.aribaud@free.fr Cc: Alexander Holler holler@ahsoftware.de Cc: Dirk Behme dirk.behme@googlemail.com ---
This patch fixes a pretty nasty problem. Everybody who has experienced strange failures on ARM lately should apply it ASAP and re-test. As far as I can tell at least the following drivers are affected:
drivers/net/dnet.c: writel(*bufp++, &dnet->regs->TX_DATA_FIFO); drivers/usb/musb/musb_core.c: writeb(*data++, &musbr->fifox[ep]); drivers/mmc/pxa_mmc.c: writel(*src++, MMC_TXFIFO); drivers/mmc/mxcmmc.c: writel(*buf++, &host->base->buffer_access); drivers/spi/davinci_spi.c: writel(data1_reg_val | *txp++, &ds->regs->dat1);
Albert, please apply ASAP!
arch/arm/include/asm/io.h | 6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-)
diff --git a/arch/arm/include/asm/io.h b/arch/arm/include/asm/io.h index 3886f15..1fbc531 100644 --- a/arch/arm/include/asm/io.h +++ b/arch/arm/include/asm/io.h @@ -133,9 +133,9 @@ extern inline void __raw_readsl(unsigned int addr, void *data, int longlen) #define __iormb() dmb() #define __iowmb() dmb()
-#define writeb(v,c) ({ __iowmb(); __arch_putb(v,c); v; }) -#define writew(v,c) ({ __iowmb(); __arch_putw(v,c); v; }) -#define writel(v,c) ({ __iowmb(); __arch_putl(v,c); v; }) +#define writeb(v,c) ({ u8 __v = v; __iowmb(); __arch_putb(__v,c); __v; }) +#define writew(v,c) ({ u16 __v = v; __iowmb(); __arch_putw(__v,c); __v; }) +#define writel(v,c) ({ u32 __v = v; __iowmb(); __arch_putl(__v,c); __v; })
#define readb(c) ({ u8 __v = __arch_getb(c); __iormb(); __v; }) #define readw(c) ({ u16 __v = __arch_getw(c); __iormb(); __v; })

Le 11/02/2011 23:25, Wolfgang Denk a écrit :
Commit 3c0659b "ARM: Avoid compiler optimization for readb, writeb and friends." introduced I/O accessors with memory barriers. Unfortunately the new write*() accessors introduced a bug:
The problem is that the argument "v" gets evaluated twice. This breaks code like used here (from "drivers/net/dnet.c"):
for (i = 0; i< wrsz; i++) writel(*bufp++,&dnet->regs->TX_DATA_FIFO);
Use auxiliary variables to avoid such problems.
Signed-off-by: Wolfgang Denkwd@denx.de Cc: Albert Aribaudalbert.aribaud@free.fr Cc: Alexander Hollerholler@ahsoftware.de Cc: Dirk Behmedirk.behme@googlemail.com
This patch fixes a pretty nasty problem. Everybody who has experienced strange failures on ARM lately should apply it ASAP and re-test. As far as I can tell at least the following drivers are affected:
drivers/net/dnet.c: writel(*bufp++,&dnet->regs->TX_DATA_FIFO); drivers/usb/musb/musb_core.c: writeb(*data++,&musbr->fifox[ep]); drivers/mmc/pxa_mmc.c: writel(*src++, MMC_TXFIFO); drivers/mmc/mxcmmc.c: writel(*buf++,&host->base->buffer_access); drivers/spi/davinci_spi.c: writel(data1_reg_val | *txp++,&ds->regs->dat1);
Albert, please apply ASAP!
Applied, available on u-boot-arm/master.
Amicalement,

Dear Albert ARIBAUD,
In message 4D55B8C9.8070605@free.fr you wrote:
This patch fixes a pretty nasty problem. Everybody who has experienced strange failures on ARM lately should apply it ASAP and re-test. As far as I can tell at least the following drivers are affected:
drivers/net/dnet.c: writel(*bufp++,&dnet->regs->TX_DATA_FIFO); drivers/usb/musb/musb_core.c: writeb(*data++,&musbr->fifox[ep]); drivers/mmc/pxa_mmc.c: writel(*src++, MMC_TXFIFO); drivers/mmc/mxcmmc.c: writel(*buf++,&host->base->buffer_access); drivers/spi/davinci_spi.c: writel(data1_reg_val | *txp++,&ds->regs->dat1);
Albert, please apply ASAP!
Applied, available on u-boot-arm/master.
Thanks, highly appreciated.
Best regards,
Wolfgang Denk

Hello,
Am 11.02.2011 23:25, schrieb Wolfgang Denk:
+#define writeb(v,c) ({ u8 __v = v; __iowmb(); __arch_putb(__v,c); __v; }) +#define writew(v,c) ({ u16 __v = v; __iowmb(); __arch_putw(__v,c); __v; }) +#define writel(v,c) ({ u32 __v = v; __iowmb(); __arch_putl(__v,c); __v; })
Thanks a lot, can I have the do {} while() back? I still don't like such macro-making-sense-only constructs. (Just kidding). ;)
Regards,
Alexander
participants (3)
-
Albert ARIBAUD
-
Alexander Holler
-
Wolfgang Denk