kernel: fiq_fsm: Handle HC babble errors

See: raspberrypi/linux#588

kernel: Perform I2C combined transactions when possible
See: raspberrypi/linux#318

kernel: V4L2: Increase the MMAL timeout to 3sec
See: raspberrypi/linux#592

kernel: i2c-bcm2708: fixed baudrate
See: raspberrypi/linux#592

firmware: Avoid calling hdcp_shutdown() when hotplug deasserted
See: http://forum.stmlabs.com/showthread.php?tid=11003&pid=101848#pid101848

firmware: hello_fft: Update readme to mention use with GL
See: http://www.raspberrypi.org/forums/viewtopic.php?f=68&t=76189

firmware: Clear unused pixels in the subsample image in H264 codec
See: http://www.raspberrypi.org/forums/viewtopic.php?f=43&t=76845&p=551515#p551515

Author: Dom Cobley

bootcode.bin

@@ -1 +1 @@
-c3db7205bcd8988cf7c185e50c8849542554b1f5
+4d496bf82c6e4546de64061134bb902bd64f07a7

fixup.dat

@@ -104,3 +104,26 @@ It accepts three optional command-line arguments: <log2_N> <batch> <loops>
 
 The special character device is required for the ioctl mailbox through which
 the ARM communicates with the Videocore GPU.
+
+
+*** With Open GL ***
+
+GPU_FFT and Open GL will run concurrently if the GPU_FFT_MEM_* defines in
+file gpu_fft.c are changed as follows:
+
+#define GPU_FFT_MEM_FLG 0x4        // cached=0xC; direct=0x4
+#define GPU_FFT_MEM_MAP 0x20000000 // cached=0x0; direct=0x20000000
+
+Overall performance will probably be higher if GPU_FFT and Open GL take turns
+at using the 3D hardware.  Since eglSwapBuffers() returns immediately without
+waiting for rendering, call glFlush() and glFinish() afterwards as follows:
+
+    for (;;) {
+        ....
+        eglSwapBuffers(....); // non-blocking call returns immediately
+        glFlush();
+        glFinish(); // wait until V3D hardware is idle
+        ....
+        gpu_fft_execute(....); // blocking call
+        ....
+    }