74fedbe377
With a SPI clock above about 24MHz the APB cannot keep up when doing individual byte transfers. Adjust the driver to use 16-bit reads when it can, to remove this bottleneck. Any transaction which involves writing bytes still uses 8-bit transfers, to simplify the code. These are the transfers that are not time-critical since they tend to be small. The case that really matters is reading from SPI flash. In general we can use 16-bit reads anytime we are transferring an even number of bytes. If the code detects an odd number of bytes, it tries to perform the operation in two steps: once in 16-bit mode with an even number of bytes, and once in 8-bit mode for the final byte. This allow us to use 16-bit reads even if asked to transfer (for example) 0xf423 bytes. The limit on in_now and out_now is adjusted to 0xfffe to avoid an extra transfer when transferring ~>=64KB. CQ-DEPEND=CL:383232 BUG=chrome-os-partner:56556 BRANCH=none TEST=boot on gru and see that things still work correctly. I tested (with extra debugging) that the 16-bit case is being picked when it should be. Change-Id: If5effae9a84e4de06537fd594bedf7f01d6a9c88 Signed-off-by: Patrick Georgi <pgeorgi@chromium.org> Original-Commit-Id: ec250b4931c7d99cc014e32ab597fca948299d08 Original-Change-Id: Idc5b7e5d82cdbdc1e8fe8b2d6da819edf2d5570c Original-Signed-off-by: Simon Glass <sjg@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/381312 Original-Commit-Ready: Julius Werner <jwerner@chromium.org> Original-Tested-by: Julius Werner <jwerner@chromium.org> Original-Reviewed-by: Julius Werner <jwerner@chromium.org> Reviewed-on: https://review.coreboot.org/16712 Tested-by: build bot (Jenkins) Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net> Reviewed-by: Martin Roth <martinroth@google.com> |
||
|---|---|---|
| .. | ||
| include/soc | ||
| cbmem.c | ||
| edp.c | ||
| gpio.c | ||
| i2c.c | ||
| pwm.c | ||
| rk808.c | ||
| spi.c | ||
| uart.c | ||
| vop.c | ||