tasklet: proceed with data_done even if no block is there

Now that we are self-timed, the only reason white the active input block is null may be a zio buffer-full situation. We need to keep the loop running anyways (by discarding one sample), because we miss notification when the buffer has new space. We can't change zio to this aim, as it would be extra code in a hot path: we'd better loose one block when the buffer is full, we would loose it soon after anyways, since the system is overloaded. Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

tasklet: proceed with data_done even if no block is there
Now that we are self-timed, the only reason white the active input block is null may be a zio buffer-full situation. We need to keep the loop running anyways (by discarding one sample), because we miss notification when the buffer has new space. We can't change zio to this aim, as it would be extra code in a hot path: we'd better loose one block when the buffer is full, we would loose it soon after anyways, since the system is overloaded. Signed-off-by: Alessandro Rubini <rubini@gnudd.com>
48079bc7 · Alessandro Rubini · 14b5d0da · 48079bc7
Commit 48079bc7 authored Jan 26, 2014 by Alessandro Rubini
Hide whitespace changes
Inline Side-by-side

Showing with 6 additions and 2 deletions

fd-irq.c kernel/fd-irq.c +6 -2

No files found.
--- a/kernel/fd-irq.c
+++ b/kernel/fd-irq.c
@@ -88,10 +88,14 @@ int fd_read_sw_fifo(struct fd_dev *fd, struct zio_channel *chan)
 	struct fd_time t;
 	unsigned long flags;

-	if (!chan->active_block)
-		return -EAGAIN;
 	if (fd->sw_fifo.tail == fd->sw_fifo.head)
 		return -EAGAIN;
+	/*
+	 * Proceed even if no active block is there. The buffer may be
+	 * full, but we need to keep the trigger armed for next time,
+	 * so deal with data and return success. If we -EAGAIN when
+	 * !chan->active_block is null, we'll miss an irq to restar the loop.
+	 */

 	/* Copy the sample to local storage */
 	spin_lock_irqsave(&fd->lock, flags);