mirror/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-01-15 19:42:28 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions
mariadb/include/my_atomic.h
Davi Arnaut 93e38e8a3e Bug#22320: my_atomic-t unit test fails 
Bug#52261: 64 bit atomic operations do not work on Solaris i386
           gcc in debug compilation

One of the various problems was that the source operand to
CMPXCHG8b was marked as a input/output operand, causing GCC
to use the EBX register as the destination register for the
CMPXCHG8b instruction. This could lead to crashes as the EBX
register is also implicitly used by the instruction, causing
the value to be potentially garbaged and a protection fault
once the value is used to access a position in memory.

Another problem was the lack of proper clobbers for the atomic
operations and, also, a discrepancy between the implementations
for the Compare and Set operation. The specific problems are
described and fixed by Kristian Nielsen patches:

Patch: 1

Fix bugs in my_atomic_cas*(val,cmp,new) that *cmp is accessed
after CAS succeds.

In the gcc builtin implementation, problem was that *cmp was
read again after atomic CAS to check if old *val == *cmp;
this fails if CAS is successful and another thread modifies
*cmp in-between.

In the x86-gcc implementation, problem was that *cmp was set
also in the case of successful CAS; this means there is a
window where it can clobber a value written by another thread
after successful CAS.

Patch 2:

Add a GCC asm "memory" clobber to primitives that imply a
memory barrier.

This signifies to GCC that any potentially aliased memory
must be flushed before the operation, and re-read after the
operation, so that read or modification in other threads of
such memory values will work as intended.

In effect, it makes these primitives work as memory barriers
for the compiler as well as the CPU. This is better and more
correct than adding "volatile" to variables.

include/atomic/gcc_builtins.h:
  Do not read from *cmp after the operation as it might be
  already gone if the operation was successful.
include/atomic/nolock.h:
  Prefer system provided atomics over the broken x86 asm.
include/atomic/x86-gcc.h:
  Do not mark source operands as input/output operands.
  Add proper memory clobbers.
include/my_atomic.h:
  Add notes about my_atomic_add and my_atomic_cas behaviors.
unittest/mysys/my_atomic-t.c:
  Remove work around, if it fails, there is either a problem
  with the atomic operations code or the specific compiler
  version should be black-listed.
2010-07-23 09:37:10 -03:00

308 lines
9.6 KiB
C
Raw Blame History

#ifndef MY_ATOMIC_INCLUDED
#define MY_ATOMIC_INCLUDED
/* Copyright (C) 2006 MySQL AB
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
/*
This header defines five atomic operations:
my_atomic_add#(&var, what)
'Fetch and Add'
add 'what' to *var, and return the old value of *var
my_atomic_fas#(&var, what)
'Fetch And Store'
store 'what' in *var, and return the old value of *var
my_atomic_cas#(&var, &old, new)
An odd variation of 'Compare And Set/Swap'
if *var is equal to *old, then store 'new' in *var, and return TRUE
otherwise store *var in *old, and return FALSE
Usually, &old should not be accessed if the operation is successful.
my_atomic_load#(&var)
return *var
my_atomic_store#(&var, what)
store 'what' in *var
'#' is substituted by a size suffix - 8, 16, 32, 64, or ptr
(e.g. my_atomic_add8, my_atomic_fas32, my_atomic_casptr).
NOTE This operations are not always atomic, so they always must be
enclosed in my_atomic_rwlock_rdlock(lock)/my_atomic_rwlock_rdunlock(lock)
or my_atomic_rwlock_wrlock(lock)/my_atomic_rwlock_wrunlock(lock).
Hint: if a code block makes intensive use of atomic ops, it make sense
to take/release rwlock once for the whole block, not for every statement.
On architectures where these operations are really atomic, rwlocks will
be optimized away.
8- and 16-bit atomics aren't implemented for windows (see generic-msvc.h),
but can be added, if necessary.
*/
#ifndef my_atomic_rwlock_init
#define intptr void *
/**
Currently we don't support 8-bit and 16-bit operations.
It can be added later if needed.
*/
#undef MY_ATOMIC_HAS_8_16
#ifndef MY_ATOMIC_MODE_RWLOCKS
/*
* Attempt to do atomic ops without locks
*/
#include "atomic/nolock.h"
#endif
#ifndef make_atomic_cas_body
/* nolock.h was not able to generate even a CAS function, fall back */
#include "atomic/rwlock.h"
#endif
/* define missing functions by using the already generated ones */
#ifndef make_atomic_add_body
#define make_atomic_add_body(S) \
int ## S tmp=*a; \
while (!my_atomic_cas ## S(a, &tmp, tmp+v)) ; \
v=tmp;
#endif
#ifndef make_atomic_fas_body
#define make_atomic_fas_body(S) \
int ## S tmp=*a; \
while (!my_atomic_cas ## S(a, &tmp, v)) ; \
v=tmp;
#endif
#ifndef make_atomic_load_body
#define make_atomic_load_body(S) \
ret= 0; /* avoid compiler warning */ \
(void)(my_atomic_cas ## S(a, &ret, ret));
#endif
#ifndef make_atomic_store_body
#define make_atomic_store_body(S) \
(void)(my_atomic_fas ## S (a, v));
#endif
/*
transparent_union doesn't work in g++
Bug ?
Darwin's gcc doesn't want to put pointers in a transparent_union
when built with -arch ppc64. Complains:
warning: 'transparent_union' attribute ignored
*/
#if defined(__GNUC__) && !defined(__cplusplus) && \
! (defined(__APPLE__) && (defined(_ARCH_PPC64) ||defined (_ARCH_PPC)))
/*
we want to be able to use my_atomic_xxx functions with
both signed and unsigned integers. But gcc will issue a warning
"passing arg N of `my_atomic_XXX' as [un]signed due to prototype"
if the signedness of the argument doesn't match the prototype, or
"pointer targets in passing argument N of my_atomic_XXX differ in signedness"
if int* is used where uint* is expected (or vice versa).
Let's shut these warnings up
*/
#define make_transparent_unions(S) \
typedef union { \
int ## S i; \
uint ## S u; \
} U_ ## S __attribute__ ((transparent_union)); \
typedef union { \
int ## S volatile *i; \
uint ## S volatile *u; \
} Uv_ ## S __attribute__ ((transparent_union));
#define uintptr intptr
make_transparent_unions(8)
make_transparent_unions(16)
make_transparent_unions(32)
make_transparent_unions(64)
make_transparent_unions(ptr)
#undef uintptr
#undef make_transparent_unions
#define a U_a.i
#define cmp U_cmp.i
#define v U_v.i
#define set U_set.i
#else
#define U_8 int8
#define U_16 int16
#define U_32 int32
#define U_64 int64
#define U_ptr intptr
#define Uv_8 int8
#define Uv_16 int16
#define Uv_32 int32
#define Uv_64 int64
#define Uv_ptr intptr
#define U_a volatile *a
#define U_cmp *cmp
#define U_v v
#define U_set set
#endif /* __GCC__ transparent_union magic */
#ifdef HAVE_INLINE
#define make_atomic_cas(S) \
STATIC_INLINE int my_atomic_cas ## S(Uv_ ## S U_a, \
Uv_ ## S U_cmp, U_ ## S U_set) \
{ \
int8 ret; \
make_atomic_cas_body(S); \
return ret; \
}
#define make_atomic_add(S) \
STATIC_INLINE int ## S my_atomic_add ## S( \
Uv_ ## S U_a, U_ ## S U_v) \
{ \
make_atomic_add_body(S); \
return v; \
}
#define make_atomic_fas(S) \
STATIC_INLINE int ## S my_atomic_fas ## S( \
Uv_ ## S U_a, U_ ## S U_v) \
{ \
make_atomic_fas_body(S); \
return v; \
}
#define make_atomic_load(S) \
STATIC_INLINE int ## S my_atomic_load ## S(Uv_ ## S U_a) \
{ \
int ## S ret; \
make_atomic_load_body(S); \
return ret; \
}
#define make_atomic_store(S) \
STATIC_INLINE void my_atomic_store ## S( \
Uv_ ## S U_a, U_ ## S U_v) \
{ \
make_atomic_store_body(S); \
}
#else /* no inline functions */
#define make_atomic_add(S) \
extern int ## S my_atomic_add ## S(Uv_ ## S U_a, U_ ## S U_v);
#define make_atomic_fas(S) \
extern int ## S my_atomic_fas ## S(Uv_ ## S U_a, U_ ## S U_v);
#define make_atomic_cas(S) \
extern int my_atomic_cas ## S(Uv_ ## S U_a, Uv_ ## S U_cmp, U_ ## S U_set);
#define make_atomic_load(S) \
extern int ## S my_atomic_load ## S(Uv_ ## S U_a);
#define make_atomic_store(S) \
extern void my_atomic_store ## S(Uv_ ## S U_a, U_ ## S U_v);
#endif /* HAVE_INLINE */
#ifdef MY_ATOMIC_HAS_8_16
make_atomic_cas(8)
make_atomic_cas(16)
#endif
make_atomic_cas(32)
make_atomic_cas(64)
make_atomic_cas(ptr)
#ifdef MY_ATOMIC_HAS_8_16
make_atomic_add(8)
make_atomic_add(16)
#endif
make_atomic_add(32)
make_atomic_add(64)
#ifdef MY_ATOMIC_HAS_8_16
make_atomic_load(8)
make_atomic_load(16)
#endif
make_atomic_load(32)
make_atomic_load(64)
make_atomic_load(ptr)
#ifdef MY_ATOMIC_HAS_8_16
make_atomic_fas(8)
make_atomic_fas(16)
#endif
make_atomic_fas(32)
make_atomic_fas(64)
make_atomic_fas(ptr)
#ifdef MY_ATOMIC_HAS_8_16
make_atomic_store(8)
make_atomic_store(16)
#endif
make_atomic_store(32)
make_atomic_store(64)
make_atomic_store(ptr)
#ifdef _atomic_h_cleanup_
#include _atomic_h_cleanup_
#undef _atomic_h_cleanup_
#endif
#undef U_8
#undef U_16
#undef U_32
#undef U_64
#undef U_ptr
#undef Uv_8
#undef Uv_16
#undef Uv_32
#undef Uv_64
#undef Uv_ptr
#undef a
#undef cmp
#undef v
#undef set
#undef U_a
#undef U_cmp
#undef U_v
#undef U_set
#undef make_atomic_add
#undef make_atomic_cas
#undef make_atomic_load
#undef make_atomic_store
#undef make_atomic_fas
#undef make_atomic_add_body
#undef make_atomic_cas_body
#undef make_atomic_load_body
#undef make_atomic_store_body
#undef make_atomic_fas_body
#undef intptr
/*
the macro below defines (as an expression) the code that
will be run in spin-loops. Intel manuals recummend to have PAUSE there.
It is expected to be defined in include/atomic/ *.h files
*/
#ifndef LF_BACKOFF
#define LF_BACKOFF (1)
#endif
#define MY_ATOMIC_OK 0
#define MY_ATOMIC_NOT_1CPU 1
extern int my_atomic_initialize();
#endif
#endif /* MY_ATOMIC_INCLUDED */
Reference in a new issue View git blame Copy permalink