mirrors/pipewire
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/pipewire/pipewire.git synced 2025-11-01 22:58:50 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

impl-node: Do xrun check a bit better

Browse source 
Check if the node is FINISHED instead of checking the refcounts. It's
possible that the refcounts are 0 but the node was not scheduled or
finished yet.

If the node is not FINISHED but TRIGGERED, we can run the recover
without reporting an error.

Any other state is an error and we need to log this and recover.

See #4182
This commit is contained in:
Wim Taymans 2024-08-08 15:17:27 +02:00
parent f1b6a62d8b
commit 074ad04f16
1 changed files with 19 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

29
src/pipewire/impl-node.c
View file

@ -1344,7 +1344,7 @@ static inline void debug_xrun_target(struct pw_impl_node *driver,
str_status(status), suppressed);
}
static inline void debug_xrun_graph(struct pw_impl_node *driver, uint64_t nsec)
static inline void debug_xrun_graph(struct pw_impl_node *driver, uint64_t nsec, uint32_t old_status)
{
int suppressed;
enum spa_log_level level = SPA_LOG_LEVEL_DEBUG;
@ -1353,8 +1353,8 @@ static inline void debug_xrun_graph(struct pw_impl_node *driver, uint64_t nsec)
if ((suppressed = spa_ratelimit_test(&driver->rt.rate_limit, nsec)) >= 0)
level = SPA_LOG_LEVEL_WARN;
pw_log(level, "(%s-%u) graph xrun (%d suppressed)",
driver->name, driver->info.id, suppressed);
pw_log(level, "(%s-%u) graph xrun %s (%d suppressed)",
driver->name, driver->info.id, str_status(old_status), suppressed);
spa_list_for_each(t, &driver->rt.target_list, link) {
struct pw_node_activation *a = t->activation;
@ -2010,13 +2010,12 @@ static int node_ready(void *data, int status)
struct pw_impl_node *node = data;
struct pw_impl_node *driver = node->driver_node;
struct pw_node_activation *a = node->rt.target.activation;
struct pw_node_activation_state *state = &a->state[0];
struct spa_system *data_system = node->rt.target.system;
struct pw_node_target *t, *reposition_target = NULL;;
struct pw_impl_port *p;
struct spa_io_clock *cl = &node->rt.position->clock;
int sync_type, all_ready, update_sync, target_sync, old_status;
uint32_t owner[2], reposition_owner, pending;
uint32_t owner[2], reposition_owner;
uint64_t min_timeout = UINT64_MAX, nsec;
pw_log_trace_fp("%p: ready driver:%d exported:%d %p status:%d prepared:%d", node,
@ -2037,15 +2036,25 @@ static int node_ready(void *data, int status)
nsec = get_time_ns(data_system);
if (SPA_UNLIKELY((pending = pw_node_activation_state_xchg(state)) > 0)) {
pw_impl_node_rt_emit_incomplete(driver);
while (true) {
old_status = SPA_ATOMIC_LOAD(a->status);
if (old_status != PW_NODE_ACTIVATION_FINISHED) {
debug_xrun_graph(node, nsec);
SPA_ATOMIC_STORE(a->status, PW_NODE_ACTIVATION_TRIGGERED);
if (SPA_LIKELY(old_status == PW_NODE_ACTIVATION_FINISHED))
/* all good, graph completed */
break;
if (SPA_ATOMIC_CAS(a->status, old_status, PW_NODE_ACTIVATION_TRIGGERED)) {
/* if we got triggered but did not run the processing yet we don't
* really have an error so we can skip the error reporting. We need
* to run recovery anyway because the ready callback is already
* emitted */
if (old_status != PW_NODE_ACTIVATION_TRIGGERED) {
/* otherwise, something was wrong and we debug */
debug_xrun_graph(node, nsec, old_status);
pw_impl_node_rt_emit_incomplete(driver);
}
SPA_FLAG_SET(cl->flags, SPA_IO_CLOCK_FLAG_XRUN_RECOVER);
process_node(node, nsec);
SPA_FLAG_CLEAR(cl->flags, SPA_IO_CLOCK_FLAG_XRUN_RECOVER);
break;
}
}