drivers/timer/cavs_timer: Fix race in k_cycle_get_64()
In commit 918a574c88c3 ("clock: add k_cycle_get_64") this driver was
augmented with a count64() method to get a 64 bit cycle output from
the two-32-bit-word device registers.
Unfortunately it appeared to be trying to use a spinlock around the
two (low/high) reads to protect against overflow. But that doesn't
work: spinlocks protect against other CPU code using the same
spinlock, not against a hardware counter that is incrementing in real
time!
Thankfully there was already a count() routine in place that does a
detect-overflow-and-retry loop to solve this. Use that.
Signed-off-by: Andy Ross <andrew.j.ross@intel.com>
diff --git a/drivers/timer/cavs_timer.c b/drivers/timer/cavs_timer.c
index 3c2f251..9d824b4 100644
--- a/drivers/timer/cavs_timer.c
+++ b/drivers/timer/cavs_timer.c
@@ -77,17 +77,6 @@
return shim_regs->walclk32_lo;
}
-static uint64_t count64(void)
-{
- k_spinlock_key_t key = k_spin_lock(&lock);
- uint64_t ret = shim_regs->walclk32_lo;
-
- ret |= (uint64_t)shim_regs->walclk32_hi << 32;
-
- k_spin_unlock(&lock, key);
- return ret;
-}
-
static void compare_isr(const void *arg)
{
ARG_UNUSED(arg);
@@ -195,7 +184,7 @@
uint64_t sys_clock_cycle_get_64(void)
{
- return count64();
+ return count();
}
/* Runs on secondary cores */
