Files
jemalloc/src/extent_dss.c
Jason Evans 577d4572b0 Make dss operations lockless.
Rather than protecting dss operations with a mutex, use atomic
operations.  This has negligible impact on synchronization overhead
during typical dss allocation, but is a substantial improvement for
extent_in_dss() and the newly added extent_dss_mergeable(), which can be
called multiple times during extent deallocations.

This change also has the advantage of avoiding tsd in deallocation paths
associated with purging, which resolves potential deadlocks during
thread exit due to attempted tsd resurrection.

This resolves #425.
2016-10-13 15:37:00 -07:00

242 lines
5.6 KiB
C

#define JEMALLOC_EXTENT_DSS_C_
#include "jemalloc/internal/jemalloc_internal.h"
/******************************************************************************/
/* Data. */
const char *opt_dss = DSS_DEFAULT;
const char *dss_prec_names[] = {
"disabled",
"primary",
"secondary",
"N/A"
};
/*
* Current dss precedence default, used when creating new arenas. NB: This is
* stored as unsigned rather than dss_prec_t because in principle there's no
* guarantee that sizeof(dss_prec_t) is the same as sizeof(unsigned), and we use
* atomic operations to synchronize the setting.
*/
static unsigned dss_prec_default = (unsigned)DSS_PREC_DEFAULT;
/* Base address of the DSS. */
static void *dss_base;
/* Atomic boolean indicating whether the DSS is exhausted. */
static unsigned dss_exhausted;
/* Atomic current upper limit on DSS addresses. */
static void *dss_max;
/******************************************************************************/
static void *
extent_dss_sbrk(intptr_t increment)
{
#ifdef JEMALLOC_DSS
return (sbrk(increment));
#else
not_implemented();
return (NULL);
#endif
}
dss_prec_t
extent_dss_prec_get(void)
{
dss_prec_t ret;
if (!have_dss)
return (dss_prec_disabled);
ret = (dss_prec_t)atomic_read_u(&dss_prec_default);
return (ret);
}
bool
extent_dss_prec_set(dss_prec_t dss_prec)
{
if (!have_dss)
return (dss_prec != dss_prec_disabled);
atomic_write_u(&dss_prec_default, (unsigned)dss_prec);
return (false);
}
static void *
extent_dss_max_update(void *new_addr)
{
void *max_cur;
spin_t spinner;
/*
* Get the current end of the DSS as max_cur and assure that dss_max is
* up to date.
*/
spin_init(&spinner);
while (true) {
void *max_prev = atomic_read_p(&dss_max);
max_cur = extent_dss_sbrk(0);
if ((uintptr_t)max_prev > (uintptr_t)max_cur) {
/*
* Another thread optimistically updated dss_max. Wait
* for it to finish.
*/
spin_adaptive(&spinner);
continue;
}
if (!atomic_cas_p(&dss_max, max_prev, max_cur))
break;
}
/* Fixed new_addr can only be supported if it is at the edge of DSS. */
if (new_addr != NULL && max_cur != new_addr)
return (NULL);
return (max_cur);
}
void *
extent_alloc_dss(tsdn_t *tsdn, arena_t *arena, void *new_addr, size_t size,
size_t alignment, bool *zero, bool *commit)
{
extent_t *gap;
cassert(have_dss);
assert(size > 0);
assert(alignment > 0);
/*
* sbrk() uses a signed increment argument, so take care not to
* interpret a large allocation request as a negative increment.
*/
if ((intptr_t)size < 0)
return (NULL);
gap = extent_alloc(tsdn, arena);
if (gap == NULL)
return (NULL);
if (!atomic_read_u(&dss_exhausted)) {
/*
* The loop is necessary to recover from races with other
* threads that are using the DSS for something other than
* malloc.
*/
while (true) {
void *ret, *max_cur, *gap_addr, *dss_next, *dss_prev;
size_t gap_size;
intptr_t incr;
max_cur = extent_dss_max_update(new_addr);
if (max_cur == NULL)
goto label_oom;
/*
* Compute how much gap space (if any) is necessary to
* satisfy alignment. This space can be recycled for
* later use.
*/
gap_addr = (void *)(PAGE_CEILING((uintptr_t)max_cur));
ret = (void *)ALIGNMENT_CEILING((uintptr_t)gap_addr,
PAGE_CEILING(alignment));
gap_size = (uintptr_t)ret - (uintptr_t)gap_addr;
if (gap_size != 0) {
extent_init(gap, arena, gap_addr, gap_size,
gap_size, false, false, true, false);
}
dss_next = (void *)((uintptr_t)ret + size);
if ((uintptr_t)ret < (uintptr_t)max_cur ||
(uintptr_t)dss_next < (uintptr_t)max_cur)
goto label_oom; /* Wrap-around. */
incr = gap_size + size;
/*
* Optimistically update dss_max, and roll back below if
* sbrk() fails. No other thread will try to extend the
* DSS while dss_max is greater than the current DSS
* max reported by sbrk(0).
*/
if (atomic_cas_p(&dss_max, max_cur, dss_next))
continue;
/* Try to allocate. */
dss_prev = extent_dss_sbrk(incr);
if (dss_prev == max_cur) {
/* Success. */
if (gap_size != 0)
extent_dalloc_gap(tsdn, arena, gap);
else
extent_dalloc(tsdn, arena, gap);
if (*zero)
memset(ret, 0, size);
if (!*commit)
*commit = pages_decommit(ret, size);
return (ret);
}
/*
* Failure, whether due to OOM or a race with a raw
* sbrk() call from outside the allocator. Try to roll
* back optimistic dss_max update; if rollback fails,
* it's due to another caller of this function having
* succeeded since this invocation started, in which
* case rollback is not necessary.
*/
atomic_cas_p(&dss_max, dss_next, max_cur);
if (dss_prev == (void *)-1) {
/* OOM. */
atomic_write_u(&dss_exhausted, (unsigned)true);
goto label_oom;
}
}
}
label_oom:
extent_dalloc(tsdn, arena, gap);
return (NULL);
}
static bool
extent_in_dss_helper(void *addr, void *max)
{
return ((uintptr_t)addr >= (uintptr_t)dss_base && (uintptr_t)addr <
(uintptr_t)max);
}
bool
extent_in_dss(void *addr)
{
cassert(have_dss);
return (extent_in_dss_helper(addr, atomic_read_p(&dss_max)));
}
bool
extent_dss_mergeable(void *addr_a, void *addr_b)
{
void *max;
cassert(have_dss);
if ((uintptr_t)addr_a < (uintptr_t)dss_base && (uintptr_t)addr_b <
(uintptr_t)dss_base)
return (true);
max = atomic_read_p(&dss_max);
return (extent_in_dss_helper(addr_a, max) ==
extent_in_dss_helper(addr_b, max));
}
void
extent_dss_boot(void)
{
cassert(have_dss);
dss_base = extent_dss_sbrk(0);
dss_exhausted = (unsigned)(dss_base == (void *)-1);
dss_max = dss_base;
}
/******************************************************************************/