From: Mathieu Desnoyers Date: Wed, 17 Mar 2010 13:31:26 +0000 (-0400) Subject: Update urcu and urcu-qsbr, make compiler barriers explicit X-Git-Tag: v0.4.4~6 X-Git-Url: https://git.lttng.org./?a=commitdiff_plain;h=d40fde2c4f6170c3ace770354d99acc9e9732ec5;p=userspace-rcu.git Update urcu and urcu-qsbr, make compiler barriers explicit Signed-off-by: Mathieu Desnoyers --- diff --git a/urcu-qsbr.c b/urcu-qsbr.c index c678d38..b043b85 100644 --- a/urcu-qsbr.c +++ b/urcu-qsbr.c @@ -121,10 +121,14 @@ static void update_counter_and_wait(void) #endif /* !(BITS_PER_LONG < 64) */ /* - * Enforce compiler-order of store to rcu_gp_ctr before before - * load rcu_reader ctr. - * This ensures synchronize_rcu() cannot be starved by readers. - * + * Must commit rcu_gp_ctr update to memory before waiting for quiescent + * state. Failure to do so could result in the writer waiting forever + * while new readers are always accessing data (no progress). Enforce + * compiler-order of store to rcu_gp_ctr before load rcu_reader ctr. + */ + barrier(); + + /* * Adding a smp_mb() which is _not_ formally required, but makes the * model easier to understand. It does not have a big performance impact * anyway, given this is the write-side. @@ -208,11 +212,12 @@ void synchronize_rcu(void) /* * Must finish waiting for quiescent state for parity 0 before - * committing qparity update to memory. Failure to do so could result in - * the writer waiting forever while new readers are always accessing - * data (no progress). - * Ensured by STORE_SHARED and LOAD_SHARED. + * committing next rcu_gp_ctr update to memory. Failure to do so could + * result in the writer waiting forever while new readers are always + * accessing data (no progress). Enforce compiler-order of load + * rcu_reader ctr before store to rcu_gp_ctr. */ + barrier(); /* * Adding a smp_mb() which is _not_ formally required, but makes the diff --git a/urcu.c b/urcu.c index bb6629c..c6d7dbc 100644 --- a/urcu.c +++ b/urcu.c @@ -208,16 +208,14 @@ void update_counter_and_wait(void) STORE_SHARED(rcu_gp_ctr, rcu_gp_ctr ^ RCU_GP_CTR_PHASE); /* - * Must commit qparity update to memory before waiting for other parity - * quiescent state. Failure to do so could result in the writer waiting - * forever while new readers are always accessing data (no progress). - * Ensured by STORE_SHARED and LOAD_SHARED. + * Must commit rcu_gp_ctr update to memory before waiting for quiescent + * state. Failure to do so could result in the writer waiting forever + * while new readers are always accessing data (no progress). Enforce + * compiler-order of store to rcu_gp_ctr before load rcu_reader ctr. */ + barrier(); /* - * Enforce compiler-order of store to rcu_gp_ctr before before - * load rcu_reader ctr. - * This ensures synchronize_rcu() cannot be starved by readers. * * Adding a smp_mb() which is _not_ formally required, but makes the * model easier to understand. It does not have a big performance impact @@ -306,11 +304,12 @@ void synchronize_rcu(void) /* * Must finish waiting for quiescent state for parity 0 before - * committing qparity update to memory. Failure to do so could result in - * the writer waiting forever while new readers are always accessing - * data (no progress). - * Ensured by STORE_SHARED and LOAD_SHARED. + * committing next rcu_gp_ctr update to memory. Failure to do so could + * result in the writer waiting forever while new readers are always + * accessing data (no progress). Enforce compiler-order of load + * rcu_reader ctr before store to rcu_gp_ctr. */ + barrier(); /* * Adding a smp_mb() which is _not_ formally required, but makes the