When doing our global park, we like to be a good citizen and shrink our slab caches (of which we have quite a few now), but each kmem_cache_shrink() incurs a stop_machine() and so ends up being quite expensive, causing machine-wide stalls. While ideally we would like to throw away unused pages in our slab caches whenever it appears that we are idling, doing so will require a much cheaper mechanism. In the meantime use a delayed worked to impose a rate-limit that means we have to have been idle for more than 2 seconds before we start shrinking.
References: https://gitlab.freedesktop.org/drm/intel/issues/848 Signed-off-by: Chris Wilson <ch...@chris-wilson.co.uk> Cc: Tvrtko Ursulin <tvrtko.ursu...@intel.com> --- drivers/gpu/drm/i915/i915_globals.c | 31 +++++++++++++++++++++++------ 1 file changed, 25 insertions(+), 6 deletions(-) diff --git a/drivers/gpu/drm/i915/i915_globals.c b/drivers/gpu/drm/i915/i915_globals.c index be127cd28931..06d2be2daca2 100644 --- a/drivers/gpu/drm/i915/i915_globals.c +++ b/drivers/gpu/drm/i915/i915_globals.c @@ -20,7 +20,8 @@ static LIST_HEAD(globals); static atomic_t active; static atomic_t epoch; static struct park_work { - struct rcu_work work; + struct delayed_work work; + struct rcu_head rcu; int epoch; } park; @@ -37,11 +38,27 @@ static void i915_globals_shrink(void) global->shrink(); } +static void __i915_globals_grace(struct rcu_head *rcu) +{ + /* Ratelimit parking as shrinking is quite slow */ + mod_delayed_work(system_wq, &park.work, + round_jiffies_up_relative(2 * HZ)); +} + +static void __i915_globals_queue_rcu(void) +{ + park.epoch = atomic_inc_return(&epoch); + if (!atomic_read(&active)) + call_rcu(&park.rcu, __i915_globals_grace); +} + static void __i915_globals_park(struct work_struct *work) { /* Confirm nothing woke up in the last grace period */ if (park.epoch == atomic_read(&epoch)) i915_globals_shrink(); + else + __i915_globals_queue_rcu(); } void __init i915_global_register(struct i915_global *global) @@ -85,7 +102,7 @@ int __init i915_globals_init(void) } } - INIT_RCU_WORK(&park.work, __i915_globals_park); + INIT_DELAYED_WORK(&park.work, __i915_globals_park); return 0; } @@ -103,8 +120,9 @@ void i915_globals_park(void) if (!atomic_dec_and_test(&active)) return; - park.epoch = atomic_inc_return(&epoch); - queue_rcu_work(system_wq, &park.work); + /* Queue cleanup after the next RCU grace period has freed slabs */ + if (!work_pending(&park.work.work)) + __i915_globals_queue_rcu(); } void i915_globals_unpark(void) @@ -116,8 +134,9 @@ void i915_globals_unpark(void) void __exit i915_globals_exit(void) { /* Flush any residual park_work */ - atomic_inc(&epoch); - flush_rcu_work(&park.work); + atomic_inc(&epoch); /* skip shrinking */ + rcu_barrier(); /* wait for the work to be queued */ + flush_delayed_work(&park.work); __i915_globals_cleanup(); -- 2.24.1 _______________________________________________ Intel-gfx mailing list Intel-gfx@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/intel-gfx
