V4L/DVB (6027): Get rid of an ill-behaved msleep in i2c write

Configuring the OLPC camera requires something over 150 register
writes.  Unfortunately, querying the CAFE i2c controller too
soon after a write causes the hardware to flake.  The problem had
been "solved" with an msleep() call, but, between the number of
registers and how msleep() behaves, that resulted in a 3-second
delay on camera initialization.  Instead, we hand-code a wait for
the completion interrupt which avoids reading the status registers.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>

1 files changed, 21 insertions, 1 deletions

diff --git a/drivers/media/video/cafe_ccic.c b/drivers/media/video/cafe_ccic.c
index 88090107cd4..acf64b19730 100644
--- a/drivers/media/video/cafe_ccic.c
+++ b/drivers/media/video/cafe_ccic.c
@@ -356,6 +356,7 @@ static int cafe_smbus_write_data(struct cafe_camera *cam,
 {
         unsigned int rval;
         unsigned long flags;
+        DEFINE_WAIT(the_wait);
 
         spin_lock_irqsave(&cam->dev_lock, flags);
         rval = TWSIC0_EN | ((addr << TWSIC0_SID_SHIFT) & TWSIC0_SID);
@@ -369,10 +370,29 @@ static int cafe_smbus_write_data(struct cafe_camera *cam,
         rval = value | ((command << TWSIC1_ADDR_SHIFT) & TWSIC1_ADDR);
         cafe_reg_write(cam, REG_TWSIC1, rval);
         spin_unlock_irqrestore(&cam->dev_lock, flags);
-        msleep(2); /* Required or things flake */
 
+        /*
+         * Time to wait for the write to complete.  THIS IS A RACY
+         * WAY TO DO IT, but the sad fact is that reading the TWSIC1
+         * register too quickly after starting the operation sends
+         * the device into a place that may be kinder and better, but
+         * which is absolutely useless for controlling the sensor.  In
+         * practice we have plenty of time to get into our sleep state
+         * before the interrupt hits, and the worst case is that we
+         * time out and then see that things completed, so this seems
+         * the best way for now.
+         */
+        do {
+                prepare_to_wait(&cam->smbus_wait, &the_wait,
+                                TASK_UNINTERRUPTIBLE);
+                schedule_timeout(1); /* even 1 jiffy is too long */
+                finish_wait(&cam->smbus_wait, &the_wait);
+        } while (!cafe_smbus_write_done(cam));
+
+#ifdef IF_THE_CAFE_HARDWARE_WORKED_RIGHT
         wait_event_timeout(cam->smbus_wait, cafe_smbus_write_done(cam),
                         CAFE_SMBUS_TIMEOUT);
+#endif
         spin_lock_irqsave(&cam->dev_lock, flags);
         rval = cafe_reg_read(cam, REG_TWSIC1);
         spin_unlock_irqrestore(&cam->dev_lock, flags);