Commit c1f64a58 authored by Linus Torvalds's avatar Linus Torvalds Committed by Ingo Molnar

x86: MMIO and gcc re-ordering issue

On Tue, 27 May 2008, Linus Torvalds wrote:
>
> Expecting people to fix up all drivers is simply not going to happen. And
> serializing things shouldn't be *that* expensive. People who cannot take
> the expense can continue to use the magic __raw_writel() etc stuff.

Of course, for non-x86, you kind of have to expect drivers to be
well-behaved, so non-x86 can probably avoid this simply because there are
less relevant drivers involved.

Here's a UNTESTED patch for x86 that may or may not compile and work, and
which serializes (on a compiler level) the IO accesses against regular
memory accesses.

__read[bwlq]()/__write[bwlq]() are not serialized with a :"memory"
barrier, although since they still use "asm volatile" I suspect that i
practice they are probably serial too. Did not look very closely at any
generated code (only did a trivial test to see that the code looks
*roughly* correct).
Signed-off-by: default avatarIngo Molnar <mingo@elte.hu>
parent 1beee8dc
...@@ -3,6 +3,62 @@ ...@@ -3,6 +3,62 @@
#define ARCH_HAS_IOREMAP_WC #define ARCH_HAS_IOREMAP_WC
#include <linux/compiler.h>
#define build_mmio_read(name, size, type, reg, barrier) \
static inline type name(const volatile void __iomem *addr) \
{ type ret; asm volatile("mov" size " %1,%0":"=" reg (ret) \
:"m" (*(volatile type __force *)addr) barrier); return ret; }
#define build_mmio_write(name, size, type, reg, barrier) \
static inline void name(type val, volatile void __iomem *addr) \
{ asm volatile("mov" size " %0,%1": :reg (val), \
"m" (*(volatile type __force *)addr) barrier); }
build_mmio_read(readb, "b", unsigned char, "q", :"memory")
build_mmio_read(readw, "w", unsigned short, "r", :"memory")
build_mmio_read(readl, "l", unsigned int, "r", :"memory")
build_mmio_read(__readb, "b", unsigned char, "q", )
build_mmio_read(__readw, "w", unsigned short, "r", )
build_mmio_read(__readl, "l", unsigned int, "r", )
build_mmio_write(writeb, "b", unsigned char, "q", :"memory")
build_mmio_write(writew, "w", unsigned short, "r", :"memory")
build_mmio_write(writel, "l", unsigned int, "r", :"memory")
build_mmio_write(__writeb, "b", unsigned char, "q", )
build_mmio_write(__writew, "w", unsigned short, "r", )
build_mmio_write(__writel, "l", unsigned int, "r", )
#define readb_relaxed(a) __readb(a)
#define readw_relaxed(a) __readw(a)
#define readl_relaxed(a) __readl(a)
#define __raw_readb __readb
#define __raw_readw __readw
#define __raw_readl __readl
#define __raw_writeb __writeb
#define __raw_writew __writew
#define __raw_writel __writel
#define mmiowb() barrier()
#ifdef CONFIG_X86_64
build_mmio_read(readq, "q", unsigned long, "r", :"memory")
build_mmio_read(__readq, "q", unsigned long, "r", )
build_mmio_write(writeq, "q", unsigned long, "r", :"memory")
build_mmio_write(__writeq, "q", unsigned long, "r", )
#define readq_relaxed(a) __readq(a)
#define __raw_readq __readq
#define __raw_writeq writeq
/* Let people know we have them */
#define readq readq
#define writeq writeq
#endif
#ifdef CONFIG_X86_32 #ifdef CONFIG_X86_32
# include "io_32.h" # include "io_32.h"
#else #else
......
...@@ -149,55 +149,6 @@ extern void __iomem *fix_ioremap(unsigned idx, unsigned long phys); ...@@ -149,55 +149,6 @@ extern void __iomem *fix_ioremap(unsigned idx, unsigned long phys);
#define virt_to_bus virt_to_phys #define virt_to_bus virt_to_phys
#define bus_to_virt phys_to_virt #define bus_to_virt phys_to_virt
/*
* readX/writeX() are used to access memory mapped devices. On some
* architectures the memory mapped IO stuff needs to be accessed
* differently. On the x86 architecture, we just read/write the
* memory location directly.
*/
static inline unsigned char readb(const volatile void __iomem *addr)
{
return *(volatile unsigned char __force *)addr;
}
static inline unsigned short readw(const volatile void __iomem *addr)
{
return *(volatile unsigned short __force *)addr;
}
static inline unsigned int readl(const volatile void __iomem *addr)
{
return *(volatile unsigned int __force *) addr;
}
#define readb_relaxed(addr) readb(addr)
#define readw_relaxed(addr) readw(addr)
#define readl_relaxed(addr) readl(addr)
#define __raw_readb readb
#define __raw_readw readw
#define __raw_readl readl
static inline void writeb(unsigned char b, volatile void __iomem *addr)
{
*(volatile unsigned char __force *)addr = b;
}
static inline void writew(unsigned short b, volatile void __iomem *addr)
{
*(volatile unsigned short __force *)addr = b;
}
static inline void writel(unsigned int b, volatile void __iomem *addr)
{
*(volatile unsigned int __force *)addr = b;
}
#define __raw_writeb writeb
#define __raw_writew writew
#define __raw_writel writel
#define mmiowb()
static inline void static inline void
memset_io(volatile void __iomem *addr, unsigned char val, int count) memset_io(volatile void __iomem *addr, unsigned char val, int count)
{ {
......
...@@ -204,77 +204,6 @@ extern void __iomem *fix_ioremap(unsigned idx, unsigned long phys); ...@@ -204,77 +204,6 @@ extern void __iomem *fix_ioremap(unsigned idx, unsigned long phys);
#define virt_to_bus virt_to_phys #define virt_to_bus virt_to_phys
#define bus_to_virt phys_to_virt #define bus_to_virt phys_to_virt
/*
* readX/writeX() are used to access memory mapped devices. On some
* architectures the memory mapped IO stuff needs to be accessed
* differently. On the x86 architecture, we just read/write the
* memory location directly.
*/
static inline __u8 __readb(const volatile void __iomem *addr)
{
return *(__force volatile __u8 *)addr;
}
static inline __u16 __readw(const volatile void __iomem *addr)
{
return *(__force volatile __u16 *)addr;
}
static __always_inline __u32 __readl(const volatile void __iomem *addr)
{
return *(__force volatile __u32 *)addr;
}
static inline __u64 __readq(const volatile void __iomem *addr)
{
return *(__force volatile __u64 *)addr;
}
#define readb(x) __readb(x)
#define readw(x) __readw(x)
#define readl(x) __readl(x)
#define readq(x) __readq(x)
#define readb_relaxed(a) readb(a)
#define readw_relaxed(a) readw(a)
#define readl_relaxed(a) readl(a)
#define readq_relaxed(a) readq(a)
#define __raw_readb readb
#define __raw_readw readw
#define __raw_readl readl
#define __raw_readq readq
#define mmiowb()
static inline void __writel(__u32 b, volatile void __iomem *addr)
{
*(__force volatile __u32 *)addr = b;
}
static inline void __writeq(__u64 b, volatile void __iomem *addr)
{
*(__force volatile __u64 *)addr = b;
}
static inline void __writeb(__u8 b, volatile void __iomem *addr)
{
*(__force volatile __u8 *)addr = b;
}
static inline void __writew(__u16 b, volatile void __iomem *addr)
{
*(__force volatile __u16 *)addr = b;
}
#define writeq(val, addr) __writeq((val), (addr))
#define writel(val, addr) __writel((val), (addr))
#define writew(val, addr) __writew((val), (addr))
#define writeb(val, addr) __writeb((val), (addr))
#define __raw_writeb writeb
#define __raw_writew writew
#define __raw_writel writel
#define __raw_writeq writeq
void __memcpy_fromio(void *, unsigned long, unsigned); void __memcpy_fromio(void *, unsigned long, unsigned);
void __memcpy_toio(unsigned long, const void *, unsigned); void __memcpy_toio(unsigned long, const void *, unsigned);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment