Revision e91467ecd1ef381377fd327c0ded922835ec52ab authored by Christian Borntraeger on 05 August 2006, 19:13:52 UTC, committed by Linus Torvalds on 06 August 2006, 15:57:46 UTC
This patch adds a barrier() in futex unqueue_me to avoid aliasing of two pointers. On my s390x system I saw the following oops: Unable to handle kernel pointer dereference at virtual kernel address 0000000000000000 Oops: 0004 [#1] CPU: 0 Not tainted Process mytool (pid: 13613, task: 000000003ecb6ac0, ksp: 00000000366bdbd8) Krnl PSW : 0704d00180000000 00000000003c9ac2 (_spin_lock+0xe/0x30) Krnl GPRS: 00000000ffffffff 000000003ecb6ac0 0000000000000000 0700000000000000 0000000000000000 0000000000000000 000001fe00002028 00000000000c091f 000001fe00002054 000001fe00002054 0000000000000000 00000000366bddc0 00000000005ef8c0 00000000003d00e8 0000000000144f91 00000000366bdcb8 Krnl Code: ba 4e 20 00 12 44 b9 16 00 3e a7 84 00 08 e3 e0 f0 88 00 04 Call Trace: ([<0000000000144f90>] unqueue_me+0x40/0xe4) [<0000000000145a0c>] do_futex+0x33c/0xc40 [<000000000014643e>] sys_futex+0x12e/0x144 [<000000000010bb00>] sysc_noemu+0x10/0x16 [<000002000003741c>] 0x2000003741c The code in question is: static int unqueue_me(struct futex_q *q) { int ret = 0; spinlock_t *lock_ptr; /* In the common case we don't take the spinlock, which is nice. */ retry: lock_ptr = q->lock_ptr; if (lock_ptr != 0) { spin_lock(lock_ptr); /* * q->lock_ptr can change between reading it and * spin_lock(), causing us to take the wrong lock. This * corrects the race condition. [...] and my compiler (gcc 4.1.0) makes the following out of it: 00000000000003c8 <unqueue_me>: 3c8: eb bf f0 70 00 24 stmg %r11,%r15,112(%r15) 3ce: c0 d0 00 00 00 00 larl %r13,3ce <unqueue_me+0x6> 3d0: R_390_PC32DBL .rodata+0x2a 3d4: a7 f1 1e 00 tml %r15,7680 3d8: a7 84 00 01 je 3da <unqueue_me+0x12> 3dc: b9 04 00 ef lgr %r14,%r15 3e0: a7 fb ff d0 aghi %r15,-48 3e4: b9 04 00 b2 lgr %r11,%r2 3e8: e3 e0 f0 98 00 24 stg %r14,152(%r15) 3ee: e3 c0 b0 28 00 04 lg %r12,40(%r11) /* write q->lock_ptr in r12 */ 3f4: b9 02 00 cc ltgr %r12,%r12 3f8: a7 84 00 4b je 48e <unqueue_me+0xc6> /* if r12 is zero then jump over the code.... */ 3fc: e3 20 b0 28 00 04 lg %r2,40(%r11) /* write q->lock_ptr in r2 */ 402: c0 e5 00 00 00 00 brasl %r14,402 <unqueue_me+0x3a> 404: R_390_PC32DBL _spin_lock+0x2 /* use r2 as parameter for spin_lock */ So the code becomes more or less: if (q->lock_ptr != 0) spin_lock(q->lock_ptr) instead of if (lock_ptr != 0) spin_lock(lock_ptr) Which caused the oops from above. After adding a barrier gcc creates code without this problem: [...] (the same) 3ee: e3 c0 b0 28 00 04 lg %r12,40(%r11) 3f4: b9 02 00 cc ltgr %r12,%r12 3f8: b9 04 00 2c lgr %r2,%r12 3fc: a7 84 00 48 je 48c <unqueue_me+0xc4> 400: c0 e5 00 00 00 00 brasl %r14,400 <unqueue_me+0x38> 402: R_390_PC32DBL _spin_lock+0x2 As a general note, this code of unqueue_me seems a bit fishy. The retry logic of unqueue_me only works if we can guarantee, that the original value of q->lock_ptr is always a spinlock (Otherwise we overwrite kernel memory). We know that q->lock_ptr can change. I dont know what happens with the original spinlock, as I am not an expert with the futex code. Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Rusty Russell <rusty@rustcorp.com.au> Acked-by: Ingo Molnar <mingo@redhat.com> Cc: Thomas Gleixner <tglx@timesys.com> Signed-off-by: Christian Borntraeger <borntrae@de.ibm.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
1 parent 72f0b4e
jensen.h
#ifndef __ALPHA_JENSEN_H
#define __ALPHA_JENSEN_H
#include <asm/compiler.h>
/*
* Defines for the AlphaPC EISA IO and memory address space.
*/
/*
* NOTE! The memory operations do not set any memory barriers, as it's
* not needed for cases like a frame buffer that is essentially memory-like.
* You need to do them by hand if the operations depend on ordering.
*
* Similarly, the port IO operations do a "mb" only after a write operation:
* if an mb is needed before (as in the case of doing memory mapped IO
* first, and then a port IO operation to the same device), it needs to be
* done by hand.
*
* After the above has bitten me 100 times, I'll give up and just do the
* mb all the time, but right now I'm hoping this will work out. Avoiding
* mb's may potentially be a noticeable speed improvement, but I can't
* honestly say I've tested it.
*
* Handling interrupts that need to do mb's to synchronize to non-interrupts
* is another fun race area. Don't do it (because if you do, I'll have to
* do *everything* with interrupts disabled, ugh).
*/
/*
* EISA Interrupt Acknowledge address
*/
#define EISA_INTA (IDENT_ADDR + 0x100000000UL)
/*
* FEPROM addresses
*/
#define EISA_FEPROM0 (IDENT_ADDR + 0x180000000UL)
#define EISA_FEPROM1 (IDENT_ADDR + 0x1A0000000UL)
/*
* VL82C106 base address
*/
#define EISA_VL82C106 (IDENT_ADDR + 0x1C0000000UL)
/*
* EISA "Host Address Extension" address (bits 25-31 of the EISA address)
*/
#define EISA_HAE (IDENT_ADDR + 0x1D0000000UL)
/*
* "SYSCTL" register address
*/
#define EISA_SYSCTL (IDENT_ADDR + 0x1E0000000UL)
/*
* "spare" register address
*/
#define EISA_SPARE (IDENT_ADDR + 0x1F0000000UL)
/*
* EISA memory address offset
*/
#define EISA_MEM (IDENT_ADDR + 0x200000000UL)
/*
* EISA IO address offset
*/
#define EISA_IO (IDENT_ADDR + 0x300000000UL)
#ifdef __KERNEL__
#ifndef __EXTERN_INLINE
#define __EXTERN_INLINE extern inline
#define __IO_EXTERN_INLINE
#endif
/*
* Handle the "host address register". This needs to be set
* to the high 7 bits of the EISA address. This is also needed
* for EISA IO addresses, which are only 16 bits wide (the
* hae needs to be set to 0).
*
* HAE isn't needed for the local IO operations, though.
*/
#define JENSEN_HAE_ADDRESS EISA_HAE
#define JENSEN_HAE_MASK 0x1ffffff
__EXTERN_INLINE void jensen_set_hae(unsigned long addr)
{
/* hae on the Jensen is bits 31:25 shifted right */
addr >>= 25;
if (addr != alpha_mv.hae_cache)
set_hae(addr);
}
#define vuip volatile unsigned int *
/*
* IO functions
*
* The "local" functions are those that don't go out to the EISA bus,
* but instead act on the VL82C106 chip directly.. This is mainly the
* keyboard, RTC, printer and first two serial lines..
*
* The local stuff makes for some complications, but it seems to be
* gone in the PCI version. I hope I can get DEC suckered^H^H^H^H^H^H^H^H
* convinced that I need one of the newer machines.
*/
static inline unsigned int jensen_local_inb(unsigned long addr)
{
return 0xff & *(vuip)((addr << 9) + EISA_VL82C106);
}
static inline void jensen_local_outb(u8 b, unsigned long addr)
{
*(vuip)((addr << 9) + EISA_VL82C106) = b;
mb();
}
static inline unsigned int jensen_bus_inb(unsigned long addr)
{
long result;
jensen_set_hae(0);
result = *(volatile int *)((addr << 7) + EISA_IO + 0x00);
return __kernel_extbl(result, addr & 3);
}
static inline void jensen_bus_outb(u8 b, unsigned long addr)
{
jensen_set_hae(0);
*(vuip)((addr << 7) + EISA_IO + 0x00) = b * 0x01010101;
mb();
}
/*
* It seems gcc is not very good at optimizing away logical
* operations that result in operations across inline functions.
* Which is why this is a macro.
*/
#define jensen_is_local(addr) ( \
/* keyboard */ (addr == 0x60 || addr == 0x64) || \
/* RTC */ (addr == 0x170 || addr == 0x171) || \
/* mb COM2 */ (addr >= 0x2f8 && addr <= 0x2ff) || \
/* mb LPT1 */ (addr >= 0x3bc && addr <= 0x3be) || \
/* mb COM2 */ (addr >= 0x3f8 && addr <= 0x3ff))
__EXTERN_INLINE u8 jensen_inb(unsigned long addr)
{
if (jensen_is_local(addr))
return jensen_local_inb(addr);
else
return jensen_bus_inb(addr);
}
__EXTERN_INLINE void jensen_outb(u8 b, unsigned long addr)
{
if (jensen_is_local(addr))
jensen_local_outb(b, addr);
else
jensen_bus_outb(b, addr);
}
__EXTERN_INLINE u16 jensen_inw(unsigned long addr)
{
long result;
jensen_set_hae(0);
result = *(volatile int *) ((addr << 7) + EISA_IO + 0x20);
result >>= (addr & 3) * 8;
return 0xffffUL & result;
}
__EXTERN_INLINE u32 jensen_inl(unsigned long addr)
{
jensen_set_hae(0);
return *(vuip) ((addr << 7) + EISA_IO + 0x60);
}
__EXTERN_INLINE void jensen_outw(u16 b, unsigned long addr)
{
jensen_set_hae(0);
*(vuip) ((addr << 7) + EISA_IO + 0x20) = b * 0x00010001;
mb();
}
__EXTERN_INLINE void jensen_outl(u32 b, unsigned long addr)
{
jensen_set_hae(0);
*(vuip) ((addr << 7) + EISA_IO + 0x60) = b;
mb();
}
/*
* Memory functions.
*/
__EXTERN_INLINE u8 jensen_readb(const volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
long result;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
result = *(volatile int *) ((addr << 7) + EISA_MEM + 0x00);
result >>= (addr & 3) * 8;
return 0xffUL & result;
}
__EXTERN_INLINE u16 jensen_readw(const volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
long result;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
result = *(volatile int *) ((addr << 7) + EISA_MEM + 0x20);
result >>= (addr & 3) * 8;
return 0xffffUL & result;
}
__EXTERN_INLINE u32 jensen_readl(const volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
return *(vuip) ((addr << 7) + EISA_MEM + 0x60);
}
__EXTERN_INLINE u64 jensen_readq(const volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
unsigned long r0, r1;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
addr = (addr << 7) + EISA_MEM + 0x60;
r0 = *(vuip) (addr);
r1 = *(vuip) (addr + (4 << 7));
return r1 << 32 | r0;
}
__EXTERN_INLINE void jensen_writeb(u8 b, volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
*(vuip) ((addr << 7) + EISA_MEM + 0x00) = b * 0x01010101;
}
__EXTERN_INLINE void jensen_writew(u16 b, volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
*(vuip) ((addr << 7) + EISA_MEM + 0x20) = b * 0x00010001;
}
__EXTERN_INLINE void jensen_writel(u32 b, volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
*(vuip) ((addr << 7) + EISA_MEM + 0x60) = b;
}
__EXTERN_INLINE void jensen_writeq(u64 b, volatile void __iomem *xaddr)
{
unsigned long addr = (unsigned long) xaddr;
jensen_set_hae(addr);
addr &= JENSEN_HAE_MASK;
addr = (addr << 7) + EISA_MEM + 0x60;
*(vuip) (addr) = b;
*(vuip) (addr + (4 << 7)) = b >> 32;
}
__EXTERN_INLINE void __iomem *jensen_ioportmap(unsigned long addr)
{
return (void __iomem *)addr;
}
__EXTERN_INLINE void __iomem *jensen_ioremap(unsigned long addr,
unsigned long size)
{
return (void __iomem *)(addr + 0x100000000ul);
}
__EXTERN_INLINE int jensen_is_ioaddr(unsigned long addr)
{
return (long)addr >= 0;
}
__EXTERN_INLINE int jensen_is_mmio(const volatile void __iomem *addr)
{
return (unsigned long)addr >= 0x100000000ul;
}
/* New-style ioread interface. All the routines are so ugly for Jensen
that it doesn't make sense to merge them. */
#define IOPORT(OS, NS) \
__EXTERN_INLINE unsigned int jensen_ioread##NS(void __iomem *xaddr) \
{ \
if (jensen_is_mmio(xaddr)) \
return jensen_read##OS(xaddr - 0x100000000ul); \
else \
return jensen_in##OS((unsigned long)xaddr); \
} \
__EXTERN_INLINE void jensen_iowrite##NS(u##NS b, void __iomem *xaddr) \
{ \
if (jensen_is_mmio(xaddr)) \
jensen_write##OS(b, xaddr - 0x100000000ul); \
else \
jensen_out##OS(b, (unsigned long)xaddr); \
}
IOPORT(b, 8)
IOPORT(w, 16)
IOPORT(l, 32)
#undef IOPORT
#undef vuip
#undef __IO_PREFIX
#define __IO_PREFIX jensen
#define jensen_trivial_rw_bw 0
#define jensen_trivial_rw_lq 0
#define jensen_trivial_io_bw 0
#define jensen_trivial_io_lq 0
#define jensen_trivial_iounmap 1
#include <asm/io_trivial.h>
#ifdef __IO_EXTERN_INLINE
#undef __EXTERN_INLINE
#undef __IO_EXTERN_INLINE
#endif
#endif /* __KERNEL__ */
#endif /* __ALPHA_JENSEN_H */
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...