virtual perfctr illegal sleep
PATCH 1/1 virtual perfctr illegal sleep
Andrew,
This patch fixes an illegal sleep issue in perfctr's virtualised per-process counters: a spinlock is taken around calls to perfctr_cpu_{reserve,release}() which sleeps on a mutex. Change the spinlock to a mutex too.
The problem was reported by Sami Farin.
Strangely enough, DEBUG_SPINLOCK_SLEEP only triggers if I also have PREEMPT enabled. Is it supposed to be like that?
Signed-off-by Mikael Pettersson <[email protected]>
drivers/perfctr/virtual.c +5 -5
@@ -91,7 +91,7 @@ static inline void vperfctr_init_bad_cpu****************************************************************/ /* XXX: perhaps relax this to number of _live_ perfctrs */ -static spinlock_t nrctrs_lock =SPIN_LOCK_UNLOCKED ; +static DECLARE_MUTEX (nrctrs_mutex);static int nrctrs;static const char this_service[] = __FILE__;
@@ -100,13 +100,13 @@ static int inc_nrctrs(void)const char *other; other =NULL ; -spin_lock (&nrctrs_lock); +down (&nrctrs_mutex);if (++nrctrs ==1 ) { other =perfctr_cpu_reserve (this_service);if (other) nrctrs =0 ; } -spin_unlock (&nrctrs_lock); +up (&nrctrs_mutex);if (other) { printk(KERN_ERR __FILE__ ": cannot operate, perfctr hardware taken by '%s'\n",
@@ -119,10 +119,10 @@ static int inc_nrctrs(void)static void dec_nrctrs (void ) { -spin_lock (&nrctrs_lock); +down (&nrctrs_mutex);if (--nrctrs ==0 )perfctr_cpu_release (this_service); -spin_unlock (&nrctrs_lock); +up (&nrctrs_mutex); }/* Allocate a `struct vperfctr'. Claim and reserve