summaryrefslogtreecommitdiff
path: root/libguile/fluids.c
diff options
context:
space:
mode:
authorLudovic Courtès <ludo@gnu.org>2010-03-05 13:38:28 +0100
committerLudovic Courtès <ludo@gnu.org>2010-03-05 13:41:20 +0100
commitbd5a75dcd8436ebe077c9ce52300d013e9519d94 (patch)
treeab0d4f71387069089fb3eec8c41da2b01f1ea483 /libguile/fluids.c
parentc02924d02b4b78998b380deb20a4fd2d5fc3cfbb (diff)
downloadguile-bd5a75dcd8436ebe077c9ce52300d013e9519d94.tar.gz
Recycle fluid numbers.
* libguile/fluids.c: Remove outdated comment on the complexity of fluid GC. (FLUID_GROW): Increase to 128. (allocated_fluids): Change to `void **'. (allocated_fluids_num): Remove. (grow_dynamic_state): Remove race-condition checker. Grow to ALLOCATED_FLUIDS_LEN. (next_fluid_num): Rename to... (new_fluid): ... this. Return a fluid instead of a fluid number. Assume ALLOCATED_FLUIDS is sparse and search for a free number in it. Make ALLOCATED_FLUIDS point to pointerless memory and initialize individual elements with a pointer to the new fluid. Register a disappearing link from there to the fluid. (scm_make_fluid): Use `new_fluid ()'. (scm_fluid_ref, scm_fluid_set_x, scm_i_swap_with_fluids): Remove assertion that wasn't checked in a thread-safe way.
Diffstat (limited to 'libguile/fluids.c')
-rw-r--r--libguile/fluids.c148
1 files changed, 58 insertions, 90 deletions
diff --git a/libguile/fluids.c b/libguile/fluids.c
index 765df7b28..17d67e934 100644
--- a/libguile/fluids.c
+++ b/libguile/fluids.c
@@ -22,7 +22,6 @@
#include <stdio.h>
#include <string.h>
-#include <assert.h>
#include "libguile/_scm.h"
#include "libguile/print.h"
@@ -34,36 +33,17 @@
#include "libguile/deprecation.h"
#include "libguile/lang.h"
#include "libguile/validate.h"
+#include "libguile/bdw-gc.h"
-#define FLUID_GROW 20
-
-/* A lot of the complexity below stems from the desire to reuse fluid
- slots. Normally, fluids should be pretty global and long-lived
- things, so that reusing their slots should not be overly critical,
- but it is the right thing to do nevertheless. The code therefore
- puts the burdon on allocating and collection fluids and keeps
- accessing fluids lock free. This is achieved by manipulating the
- global state of the fluid machinery mostly in single threaded
- sections.
-
- Reusing a fluid slot means that it must be reset to #f in all
- dynamic states. We do this by maintaining a weak list of all
- dynamic states, which is used after a GC to do the resetting.
-
- Also, the fluid vectors in the dynamic states need to grow from
- time to time when more fluids are created. We do this in a single
- threaded section so that threads do not need to lock when accessing
- a fluid in the normal way.
-*/
-
-static scm_i_pthread_mutex_t fluid_admin_mutex = SCM_I_PTHREAD_MUTEX_INITIALIZER;
+/* Number of additional slots to allocate when ALLOCATED_FLUIDS is full. */
+#define FLUID_GROW 128
-/* Protected by fluid_admin_mutex, but also accessed during GC. See
- next_fluid_num for a discussion of this.
- */
+/* Vector of allocated fluids indexed by fluid numbers. Access is protected by
+ FLUID_ADMIN_MUTEX. */
+static void **allocated_fluids = NULL;
static size_t allocated_fluids_len = 0;
-static size_t allocated_fluids_num = 0;
-static char *allocated_fluids = NULL;
+
+static scm_i_pthread_mutex_t fluid_admin_mutex = SCM_I_PTHREAD_MUTEX_INITIALIZER;
#define IS_FLUID(x) SCM_I_FLUID_P (x)
#define FLUID_NUM(x) SCM_I_FLUID_NUM (x)
@@ -74,34 +54,26 @@ static char *allocated_fluids = NULL;
-/* Grow STATE so that it can hold up to ALLOCATED_FLUIDS_NUM fluids. */
+/* Grow STATE so that it can hold up to ALLOCATED_FLUIDS_LEN fluids. This may
+ be more than necessary since ALLOCATED_FLUIDS is sparse and the current
+ thread may not access all the fluids anyway. Memory usage could be improved
+ by using a 2-level array as is done in glibc for pthread keys (TODO). */
static void
grow_dynamic_state (SCM state)
{
SCM new_fluids;
SCM old_fluids = DYNAMIC_STATE_FLUIDS (state);
- size_t i, new_len, old_len = SCM_SIMPLE_VECTOR_LENGTH (old_fluids);
+ size_t i, len, old_len = SCM_SIMPLE_VECTOR_LENGTH (old_fluids);
- retry:
- new_len = allocated_fluids_num;
- new_fluids = scm_c_make_vector (new_len, SCM_BOOL_F);
+ /* Assume the assignment below is atomic. */
+ len = allocated_fluids_len;
- scm_i_pthread_mutex_lock (&fluid_admin_mutex);
- if (new_len != allocated_fluids_num)
- {
- /* We lost the race. */
- scm_i_pthread_mutex_unlock (&fluid_admin_mutex);
- goto retry;
- }
-
- assert (allocated_fluids_num > old_len);
+ new_fluids = scm_c_make_vector (len, SCM_BOOL_F);
for (i = 0; i < old_len; i++)
SCM_SIMPLE_VECTOR_SET (new_fluids, i,
SCM_SIMPLE_VECTOR_REF (old_fluids, i));
SET_DYNAMIC_STATE_FLUIDS (state, new_fluids);
-
- scm_i_pthread_mutex_unlock (&fluid_admin_mutex);
}
void
@@ -128,45 +100,53 @@ scm_i_with_fluids_print (SCM exp, SCM port, scm_print_state *pstate SCM_UNUSED)
scm_putc ('>', port);
}
-static size_t
-next_fluid_num ()
+
+/* Return a new fluid. */
+static SCM
+new_fluid ()
{
- size_t n;
+ SCM fluid;
+ size_t trial, n;
+
+ /* Fluids are pointerless cells: the first word is the type tag; the second
+ word is the fluid number. */
+ fluid = PTR2SCM (scm_gc_malloc_pointerless (sizeof (scm_t_cell), "fluid"));
+ SCM_SET_CELL_TYPE (fluid, scm_tc7_fluid);
scm_dynwind_begin (0);
scm_i_dynwind_pthread_mutex_lock (&fluid_admin_mutex);
- if ((allocated_fluids_len > 0) &&
- (allocated_fluids_num == allocated_fluids_len))
- {
- /* All fluid numbers are in use. Run a GC to try to free some
- up.
- */
- scm_gc ();
- }
-
- if (allocated_fluids_num < allocated_fluids_len)
+ for (trial = 0; trial < 2; trial++)
{
+ /* Look for a free fluid number. */
for (n = 0; n < allocated_fluids_len; n++)
- if (allocated_fluids[n] == 0)
+ /* TODO: Use `__sync_bool_compare_and_swap' where available. */
+ if (allocated_fluids[n] == NULL)
break;
+
+ if (trial == 0 && n >= allocated_fluids_len)
+ /* All fluid numbers are in use. Run a GC and retry. Explicitly
+ running the GC is costly and bad-style. We only do this because
+ dynamic state fluid vectors would grow unreasonably if fluid numbers
+ weren't reused. */
+ scm_i_gc ("fluids");
}
- else
+
+ if (n >= allocated_fluids_len)
{
/* Grow the vector of allocated fluids. */
- /* FIXME: Since we use `scm_malloc ()', ALLOCATED_FLUIDS is scanned by
- the GC; therefore, all fluids remain reachable for the entire
- program lifetime. Hopefully this is not a problem in practice. */
- char *new_allocated_fluids =
- scm_gc_malloc (allocated_fluids_len + FLUID_GROW,
- "allocated fluids");
+ void **new_allocated_fluids =
+ scm_gc_malloc_pointerless ((allocated_fluids_len + FLUID_GROW)
+ * sizeof (*allocated_fluids),
+ "allocated fluids");
/* Copy over old values and initialize rest. GC can not run
during these two operations since there is no safe point in
- them.
- */
- memcpy (new_allocated_fluids, allocated_fluids, allocated_fluids_len);
- memset (new_allocated_fluids + allocated_fluids_len, 0, FLUID_GROW);
+ them. */
+ memcpy (new_allocated_fluids, allocated_fluids,
+ allocated_fluids_len * sizeof (*allocated_fluids));
+ memset (new_allocated_fluids + allocated_fluids_len, 0,
+ FLUID_GROW * sizeof (*allocated_fluids));
n = allocated_fluids_len;
/* Update the vector of allocated fluids. Dynamic states will
@@ -175,12 +155,15 @@ next_fluid_num ()
allocated_fluids = new_allocated_fluids;
allocated_fluids_len += FLUID_GROW;
}
-
- allocated_fluids_num += 1;
- allocated_fluids[n] = 1;
-
+
+ allocated_fluids[n] = SCM2PTR (fluid);
+ SCM_SET_CELL_WORD_1 (fluid, (scm_t_bits) n);
+
+ GC_GENERAL_REGISTER_DISAPPEARING_LINK (&allocated_fluids[n],
+ SCM2PTR (fluid));
+
scm_dynwind_end ();
- return n;
+ return fluid;
}
SCM_DEFINE (scm_make_fluid, "make-fluid", 0, 0, 0,
@@ -194,7 +177,7 @@ SCM_DEFINE (scm_make_fluid, "make-fluid", 0, 0, 0,
"with its own dynamic state, you can use fluids for thread local storage.")
#define FUNC_NAME s_scm_make_fluid
{
- return scm_cell (scm_tc7_fluid, (scm_t_bits) next_fluid_num ());
+ return new_fluid ();
}
#undef FUNC_NAME
@@ -229,11 +212,6 @@ SCM_DEFINE (scm_fluid_ref, "fluid-ref", 1, 0, 0,
if (SCM_UNLIKELY (FLUID_NUM (fluid) >= SCM_SIMPLE_VECTOR_LENGTH (fluids)))
{
- /* We should only get there when the current thread's dynamic state
- turns out to be too small compared to the set of currently allocated
- fluids. */
- assert (SCM_SIMPLE_VECTOR_LENGTH (fluids) < allocated_fluids_num);
-
/* Lazily grow the current thread's dynamic state. */
grow_dynamic_state (SCM_I_CURRENT_THREAD->dynamic_state);
@@ -255,11 +233,6 @@ SCM_DEFINE (scm_fluid_set_x, "fluid-set!", 2, 0, 0,
if (SCM_UNLIKELY (FLUID_NUM (fluid) >= SCM_SIMPLE_VECTOR_LENGTH (fluids)))
{
- /* We should only get there when the current thread's dynamic state
- turns out to be too small compared to the set of currently allocated
- fluids. */
- assert (SCM_SIMPLE_VECTOR_LENGTH (fluids) < allocated_fluids_num);
-
/* Lazily grow the current thread's dynamic state. */
grow_dynamic_state (SCM_I_CURRENT_THREAD->dynamic_state);
@@ -330,11 +303,6 @@ scm_i_swap_with_fluids (SCM wf, SCM dynstate)
if (SCM_UNLIKELY (max >= SCM_SIMPLE_VECTOR_LENGTH (fluids)))
{
- /* We should only get there when the current thread's dynamic state turns
- out to be too small compared to the set of currently allocated
- fluids. */
- assert (SCM_SIMPLE_VECTOR_LENGTH (fluids) < allocated_fluids_num);
-
/* Lazily grow the current thread's dynamic state. */
grow_dynamic_state (dynstate);