The current trigger() has two crucial problems:
1) The DMA request enabling operations (FSL_SAI_CSR_FRDE) for Tx and Rx are
   now totally exclusive: It would fail to run simultaneous Tx-Rx cases.
2) The TERE disabling operation depends on an incorrect condition -- active
   reference count that only gets increased in snd_pcm_open() and decreased
   in snd_pcm_close(): The TERE would never get cleared.

So this patch overwrites the trigger function by following these rules:
A) We continue to support tx-async-while-rx-sync-to-tx case alone, which's
   originally limited by this fsl_sai driver, but we make the code easy to
   modify for the further support of the opposite case.
B) We enable both TE and RE for PLAYBACK stream or CAPTURE stream but only
   enabling the DMA request bit (FSL_SAI_CSR_FRDE) of the current direction
   due to the requirement of SAI -- For tx-async-while-rx-sync-to-tx case,
   the receiver is enabled only when both the transmitter and receiver are
   enabled.

Tested cases:
a) aplay test.wav -d5
b) arecord -r44100 -c2 -fS16_LE test.wav -d5
c) arecord -r44100 -c2 -fS16_LE -d5 | aplay
d) (aplay test2.wav &); sleep 1; arecord -r44100 -c2 -fS16_LE test.wav -d1
e) (arecord -r44100 -c2 -fS16_LE test.wav -d5 &); sleep 1; aplay test.wav -d1
Signed-off-by: Nicolin Chen <Guangyu.Chen@freescale.com>
Acked-by: Xiubo Li <Li.Xiubo@freescale.com>
Signed-off-by: Mark Brown <broonie@linaro.org>