AMS: Return ascodec to interrupt-based I2C2 driver

1. Slightly revised and regularized internal interface. Callback is used
for read and write to provide completion signal instead of having two
mechanisms.

2. Lower overhead for asynchronous or alterate completion callbacks. We
now only init what is required by the transfer. A couple unneeded
structure members were also nixed.

3. Fixes a bug that would neglect a semaphore wait if pumping the I2C
interrupts in a loop when not in thread state or interrupts are masked.

4. Corrects broken initialization order by defining KDEV_INIT, which
makes kernel_init() call kernel_device_init() to initialize additional
devices _after_ the kernel, threading and synchronization objects are
safe to use.

5. Locking set_cpu_frequency has to be done at the highest level in
system.c to ensure the boost counter and the frequency are both set in
agreement. Reconcile the locking inteface between PP and AMS (the only
two currently using locking there) to keep it clean.

Now works fine with voltages in GIT HEAD on my Fuze v2, type 0.
Previously, everything crashed and died instantly. action.c calling
set_cpu_frequency from a tick was part of it. The rest may have been
related to 3. and 4. Honestly, I'm not certain!

Testing by Mihail Zenkov indicates it solves our problems. This will
get the developer builds running again after the kernel assert code
push.

Change-Id: Ie245994fb3e318dd5ef48e383ce61fdd977224d4

1 files changed, 19 insertions, 2 deletions

diff --git a/firmware/drivers/adc-as3514.c b/firmware/drivers/adc-as3514.c
index 3b411a379d..9a81a52cc7 100644
--- a/firmware/drivers/adc-as3514.c
+++ b/firmware/drivers/adc-as3514.c
@@ -26,18 +26,35 @@
 /* Read 10-bit channel data */
 unsigned short adc_read(int channel)
 {
-    unsigned char buf[2];
+    unsigned short data = 0;
+
+    if ((unsigned)channel >= NUM_ADC_CHANNELS)
+        return 0;
 
     ascodec_lock();
 
     /* Select channel */
     ascodec_write(AS3514_ADC_0, (channel << 4));
+    /*
+     * The AS3514 ADC will trigger an interrupt when the conversion
+     * is finished, if the corresponding enable bit in IRQ_ENRD2
+     * is set.
+     * Previously the code did not wait and this apparently did
+     * not pose any problems, but this should be more correct.
+     * Without the wait the data read back may be completely or
+     * partially (first one of the two bytes) stale.
+     */
+    ascodec_wait_adc_finished();
+
 
+    /* Read data */
+    unsigned char buf[2] = { 0, 0 };
     ascodec_readbytes(AS3514_ADC_0, 2, buf);
+    data = (((buf[0] & 0x3) << 8) | buf[1]);
 
     ascodec_unlock();
 
-    return (((buf[0] & 0x3) << 8) | buf[1]);
+    return data;
 }
 
 void adc_init(void)