bcm2835-i2s regression from asymmetric sample formats change

[HiassofT]

PR #1783 introduced a regression: when both a playback and record stream are opened at the same time several registers for the second stream aren't set up and the second stream fails.

eg use aplay | arecord to reproduce.

At the very beginning bcm2835_i2s_hw_params checks if any stream has been started and if yes hw_params bails out immediately:

	/*
	 * If a stream is already enabled,
	 * the registers are already set properly.
	 */
	regmap_read(dev->i2s_regmap, BCM2835_I2S_CS_A_REG, &csreg);

	if (csreg & (BCM2835_I2S_TXON | BCM2835_I2S_RXON))
		return 0;

This leads to eg BCM2835_I2S_TXC_A_REG/BCM2835_I2S_RXC_A_REG, BCM2835_I2S_MODE_A_REG not being initialized.

A proper fix would need quite some rework (only set up clocks once in hw_params, do separate clear_fifos for RX/TX etc) and also a lot of testing because I2S reconfiguration while another stream is running has never been tested before.

@gtrainavicius could you have a look at this?

@popcornmix @pelwell I suggest it's better to revert #1783 for now

[pelwell]

Reverted.

[HiassofT]

thanks!

[gtrainavicius]

@pelwell @HiassofT Thank you, I will take a look when I have time and submit a new pull request.