spi: spi-geni-qcom: Use the FIFO even more

In commit 902481a ("spi: spi-geni-qcom: Actually use our FIFO") I
explained that the maximum size we could program the FIFO was
"mas->tx_fifo_depth - 3" but that I chose "mas->tx_fifo_depth()"
because I was worried about decreased bandwidth.

Since that time:
* All the interconnect patches have landed, making things run at the
  proper speed.
* I've done more measurements.

This lets me confirm that there's really no downside of using the FIFO
more.  Specifically I did "flashrom -p ec -r /tmp/foo.bin" on a
Chromebook and averaged over several runs.

Before: It took 6.66 seconds and 59669 interrupts fired.
After:  It took 6.66 seconds and 47992 interrupts fired.

Signed-off-by: Douglas Anderson <dianders@chromium.org>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Link: https://lore.kernel.org/r/20200912140730.1.Ie67fa32009b94702d56232c064f1d89065ee8836@changeid
Signed-off-by: Mark Brown <broonie@kernel.org>

drivers/spi/spi-geni-qcom.c

@@ -308,7 +308,7 @@ static int spi_geni_init(struct spi_geni_master *mas)
 	 * Hardware programming guide suggests to configure
 	 * RX FIFO RFR level to fifo_depth-2.
 	 */
-	geni_se_init(se, mas->tx_fifo_depth / 2, mas->tx_fifo_depth - 2);
+	geni_se_init(se, mas->tx_fifo_depth - 3, mas->tx_fifo_depth - 2);
 	/* Transmit an entire FIFO worth of data per IRQ */
 	mas->tx_wm = 1;
 	ver = geni_se_get_qup_hw_version(se);