OpenE2K/qemu-e2k
13
4
Fork 0
Code Issues 4 Pull Requests Projects 2 Releases Activity

trace: enable tracing of TCG atomics

Browse Source 
We do not trace guest atomic accesses. Fix it.

Tested with a modified atomic_add-bench so that it executes
a deterministic number of instructions, i.e. fixed seeding,
no threading and fixed number of loop iterations instead
of running for a certain time.

Before:
- With parallel_cpus = false (no clone syscall so it is never set to true):
  220070 memory accesses
- With parallel_cpus = true (hard-coded):
  212105 memory accesses <-- we're not tracing the atomics!

After:
  220070 memory accesses regardless of parallel_cpus.

Signed-off-by: Emilio G. Cota <cota@braap.org>
Message-id: 1527028012-21888-6-git-send-email-cota@braap.org
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
This commit is contained in:
Emilio G. Cota 2018-05-22 18:26:52 -04:00 committed by Stefan Hajnoczi
parent f9b47999af
commit d071f4cd55
1 changed files with 81 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

87
accel/tcg/atomic_template.h
View File

@ -18,30 +18,37 @@
* License along with this library; if not, see <http://www.gnu.org/licenses/>. * License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/ */
#include "trace/mem.h"
#if DATA_SIZE == 16 #if DATA_SIZE == 16
# define SUFFIX o # define SUFFIX o
# define DATA_TYPE Int128 # define DATA_TYPE Int128
# define BSWAP bswap128 # define BSWAP bswap128
# define SHIFT 4
#elif DATA_SIZE == 8 #elif DATA_SIZE == 8
# define SUFFIX q # define SUFFIX q
# define DATA_TYPE uint64_t # define DATA_TYPE uint64_t
# define SDATA_TYPE int64_t # define SDATA_TYPE int64_t
# define BSWAP bswap64 # define BSWAP bswap64
# define SHIFT 3
#elif DATA_SIZE == 4 #elif DATA_SIZE == 4
# define SUFFIX l # define SUFFIX l
# define DATA_TYPE uint32_t # define DATA_TYPE uint32_t
# define SDATA_TYPE int32_t # define SDATA_TYPE int32_t
# define BSWAP bswap32 # define BSWAP bswap32
# define SHIFT 2
#elif DATA_SIZE == 2 #elif DATA_SIZE == 2
# define SUFFIX w # define SUFFIX w
# define DATA_TYPE uint16_t # define DATA_TYPE uint16_t
# define SDATA_TYPE int16_t # define SDATA_TYPE int16_t
# define BSWAP bswap16 # define BSWAP bswap16
# define SHIFT 1
#elif DATA_SIZE == 1 #elif DATA_SIZE == 1
# define SUFFIX b # define SUFFIX b
# define DATA_TYPE uint8_t # define DATA_TYPE uint8_t
# define SDATA_TYPE int8_t # define SDATA_TYPE int8_t
# define BSWAP # define BSWAP
# define SHIFT 0
#else #else
# error unsupported data size # error unsupported data size
#endif #endif
@ -52,14 +59,37 @@
# define ABI_TYPE uint32_t # define ABI_TYPE uint32_t
#endif #endif
#define ATOMIC_TRACE_RMW do { \
uint8_t info = glue(trace_mem_build_info_no_se, MEND)(SHIFT, false); \
\
trace_guest_mem_before_exec(ENV_GET_CPU(env), addr, info); \
trace_guest_mem_before_exec(ENV_GET_CPU(env), addr, \
info | TRACE_MEM_ST); \
} while (0)
#define ATOMIC_TRACE_LD do { \
uint8_t info = glue(trace_mem_build_info_no_se, MEND)(SHIFT, false); \
\
trace_guest_mem_before_exec(ENV_GET_CPU(env), addr, info); \
} while (0)
# define ATOMIC_TRACE_ST do { \
uint8_t info = glue(trace_mem_build_info_no_se, MEND)(SHIFT, true); \
\
trace_guest_mem_before_exec(ENV_GET_CPU(env), addr, info); \
} while (0)
/* Define host-endian atomic operations. Note that END is used within /* Define host-endian atomic operations. Note that END is used within
the ATOMIC_NAME macro, and redefined below. */ the ATOMIC_NAME macro, and redefined below. */
#if DATA_SIZE == 1 #if DATA_SIZE == 1
# define END # define END
# define MEND _be /* either le or be would be fine */
#elif defined(HOST_WORDS_BIGENDIAN) #elif defined(HOST_WORDS_BIGENDIAN)
# define END _be # define END _be
# define MEND _be
#else #else
# define END _le # define END _le
# define MEND _le
#endif #endif
ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr, ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr,
@ -67,7 +97,10 @@ ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr,
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP;
DATA_TYPE ret = atomic_cmpxchg__nocheck(haddr, cmpv, newv); DATA_TYPE ret;
ATOMIC_TRACE_RMW;
ret = atomic_cmpxchg__nocheck(haddr, cmpv, newv);
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
return ret; return ret;
} }
@ -77,6 +110,8 @@ ABI_TYPE ATOMIC_NAME(ld)(CPUArchState *env, target_ulong addr EXTRA_ARGS)
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE val, *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE val, *haddr = ATOMIC_MMU_LOOKUP;
ATOMIC_TRACE_LD;
__atomic_load(haddr, &val, __ATOMIC_RELAXED); __atomic_load(haddr, &val, __ATOMIC_RELAXED);
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
return val; return val;
@ -87,6 +122,8 @@ void ATOMIC_NAME(st)(CPUArchState *env, target_ulong addr,
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP;
ATOMIC_TRACE_ST;
__atomic_store(haddr, &val, __ATOMIC_RELAXED); __atomic_store(haddr, &val, __ATOMIC_RELAXED);
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
} }
@ -96,7 +133,10 @@ ABI_TYPE ATOMIC_NAME(xchg)(CPUArchState *env, target_ulong addr,
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP;
DATA_TYPE ret = atomic_xchg__nocheck(haddr, val); DATA_TYPE ret;
ATOMIC_TRACE_RMW;
ret = atomic_xchg__nocheck(haddr, val);
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
return ret; return ret;
} }
@ -107,7 +147,10 @@ ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \
{ \ { \
ATOMIC_MMU_DECLS; \ ATOMIC_MMU_DECLS; \
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \ DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \
DATA_TYPE ret = atomic_##X(haddr, val); \ DATA_TYPE ret; \
\
ATOMIC_TRACE_RMW; \
ret = atomic_##X(haddr, val); \
ATOMIC_MMU_CLEANUP; \ ATOMIC_MMU_CLEANUP; \
return ret; \ return ret; \
} }
@ -126,6 +169,9 @@ GEN_ATOMIC_HELPER(xor_fetch)
/* These helpers are, as a whole, full barriers. Within the helper, /* These helpers are, as a whole, full barriers. Within the helper,
* the leading barrier is explicit and the trailing barrier is within * the leading barrier is explicit and the trailing barrier is within
* cmpxchg primitive. * cmpxchg primitive.
*
* Trace this load + RMW loop as a single RMW op. This way, regardless
* of CF_PARALLEL's value, we'll trace just a read and a write.
*/ */
#define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \ #define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \
ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \
@ -134,6 +180,8 @@ ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \
ATOMIC_MMU_DECLS; \ ATOMIC_MMU_DECLS; \
XDATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \ XDATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \
XDATA_TYPE cmp, old, new, val = xval; \ XDATA_TYPE cmp, old, new, val = xval; \
\
ATOMIC_TRACE_RMW; \
smp_mb(); \ smp_mb(); \
cmp = atomic_read__nocheck(haddr); \ cmp = atomic_read__nocheck(haddr); \
do { \ do { \
@ -158,6 +206,7 @@ GEN_ATOMIC_HELPER_FN(umax_fetch, MAX, DATA_TYPE, new)
#endif /* DATA SIZE >= 16 */ #endif /* DATA SIZE >= 16 */
#undef END #undef END
#undef MEND
#if DATA_SIZE > 1 #if DATA_SIZE > 1
@ -165,8 +214,10 @@ GEN_ATOMIC_HELPER_FN(umax_fetch, MAX, DATA_TYPE, new)
within the ATOMIC_NAME macro. */ within the ATOMIC_NAME macro. */
#ifdef HOST_WORDS_BIGENDIAN #ifdef HOST_WORDS_BIGENDIAN
# define END _le # define END _le
# define MEND _le
#else #else
# define END _be # define END _be
# define MEND _be
#endif #endif
ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr, ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr,
@ -174,7 +225,10 @@ ABI_TYPE ATOMIC_NAME(cmpxchg)(CPUArchState *env, target_ulong addr,
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP;
DATA_TYPE ret = atomic_cmpxchg__nocheck(haddr, BSWAP(cmpv), BSWAP(newv)); DATA_TYPE ret;
ATOMIC_TRACE_RMW;
ret = atomic_cmpxchg__nocheck(haddr, BSWAP(cmpv), BSWAP(newv));
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
return BSWAP(ret); return BSWAP(ret);
} }
@ -184,6 +238,8 @@ ABI_TYPE ATOMIC_NAME(ld)(CPUArchState *env, target_ulong addr EXTRA_ARGS)
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE val, *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE val, *haddr = ATOMIC_MMU_LOOKUP;
ATOMIC_TRACE_LD;
__atomic_load(haddr, &val, __ATOMIC_RELAXED); __atomic_load(haddr, &val, __ATOMIC_RELAXED);
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
return BSWAP(val); return BSWAP(val);
@ -194,6 +250,8 @@ void ATOMIC_NAME(st)(CPUArchState *env, target_ulong addr,
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP;
ATOMIC_TRACE_ST;
val = BSWAP(val); val = BSWAP(val);
__atomic_store(haddr, &val, __ATOMIC_RELAXED); __atomic_store(haddr, &val, __ATOMIC_RELAXED);
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
@ -204,7 +262,10 @@ ABI_TYPE ATOMIC_NAME(xchg)(CPUArchState *env, target_ulong addr,
{ {
ATOMIC_MMU_DECLS; ATOMIC_MMU_DECLS;
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP;
ABI_TYPE ret = atomic_xchg__nocheck(haddr, BSWAP(val)); ABI_TYPE ret;
ATOMIC_TRACE_RMW;
ret = atomic_xchg__nocheck(haddr, BSWAP(val));
ATOMIC_MMU_CLEANUP; ATOMIC_MMU_CLEANUP;
return BSWAP(ret); return BSWAP(ret);
} }
@ -215,7 +276,10 @@ ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \
{ \ { \
ATOMIC_MMU_DECLS; \ ATOMIC_MMU_DECLS; \
DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \ DATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \
DATA_TYPE ret = atomic_##X(haddr, BSWAP(val)); \ DATA_TYPE ret; \
\
ATOMIC_TRACE_RMW; \
ret = atomic_##X(haddr, BSWAP(val)); \
ATOMIC_MMU_CLEANUP; \ ATOMIC_MMU_CLEANUP; \
return BSWAP(ret); \ return BSWAP(ret); \
} }
@ -232,6 +296,9 @@ GEN_ATOMIC_HELPER(xor_fetch)
/* These helpers are, as a whole, full barriers. Within the helper, /* These helpers are, as a whole, full barriers. Within the helper,
* the leading barrier is explicit and the trailing barrier is within * the leading barrier is explicit and the trailing barrier is within
* cmpxchg primitive. * cmpxchg primitive.
*
* Trace this load + RMW loop as a single RMW op. This way, regardless
* of CF_PARALLEL's value, we'll trace just a read and a write.
*/ */
#define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \ #define GEN_ATOMIC_HELPER_FN(X, FN, XDATA_TYPE, RET) \
ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \ ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \
@ -240,6 +307,8 @@ ABI_TYPE ATOMIC_NAME(X)(CPUArchState *env, target_ulong addr, \
ATOMIC_MMU_DECLS; \ ATOMIC_MMU_DECLS; \
XDATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \ XDATA_TYPE *haddr = ATOMIC_MMU_LOOKUP; \
XDATA_TYPE ldo, ldn, old, new, val = xval; \ XDATA_TYPE ldo, ldn, old, new, val = xval; \
\
ATOMIC_TRACE_RMW; \
smp_mb(); \ smp_mb(); \
ldn = atomic_read__nocheck(haddr); \ ldn = atomic_read__nocheck(haddr); \
do { \ do { \
@ -271,11 +340,17 @@ GEN_ATOMIC_HELPER_FN(add_fetch, ADD, DATA_TYPE, new)
#endif /* DATA_SIZE >= 16 */ #endif /* DATA_SIZE >= 16 */
#undef END #undef END
#undef MEND
#endif /* DATA_SIZE > 1 */ #endif /* DATA_SIZE > 1 */
#undef ATOMIC_TRACE_ST
#undef ATOMIC_TRACE_LD
#undef ATOMIC_TRACE_RMW
#undef BSWAP #undef BSWAP
#undef ABI_TYPE #undef ABI_TYPE
#undef DATA_TYPE #undef DATA_TYPE
#undef SDATA_TYPE #undef SDATA_TYPE
#undef SUFFIX #undef SUFFIX
#undef DATA_SIZE #undef DATA_SIZE
#undef SHIFT