[PATCH v3 0/3] Add PATA host controller support for Cirrus Logic EP93xx CPU

Rafal Prylowski prylowski at metasoft.pl
Thu Apr 12 08:10:36 EDT 2012


Hi,

this is a third version of EP93xx PATA driver (PATCH v3 1/3),
IDE platform code for ep93xx (PATCH v3 2/3) and IDE support
for EDB93xx boards (PATCH v3 3/3).
Patches apply to current mainline (v3.4rc2).

v2:
- don't BUG() in ep93xx_pata_dma_start(), ep93xx_pata_dma_setup(),
- don't use ata_ioports structure for EP93xx IDECTRL register values,
- IORDY timeout handling using jiffies (ep93xx_pata_wait_for_iordy),
- IDE/GPIO pin muxing solved by adding ep93xx_ide_acquire/release_gpio,
- register the driver also for EDB9312 and EDB9315 boards,
- corrected many coding style issues,
- dma initialization changed: request and configure channels in
  ep93xx_pata_dma_init (was: request in ep93xx_pata_dma_init,
  configuration in ep93xx_pata_dma_setup - channels were needlessly
  configured to the same values before each transfer).

v3:
- use custom delay loop (ep93xx_pata_delay()) instead of ndelay(),
- use ata_timing_compute() and ata_timing_merge() to properly calculate
  timings for the delay loop,
- use single ata_timing struct in ep93xx_pata_data for both IDE register
  and PIO data transfers,
- IORDY timeout handling using custom delay loop,
- move common parts of code from ep93xx_pata_{read/write}() to
  ep93xx_pata_rw_begin(), ep93xx_pata_rw_end(),
- introduce ep93xx_pata_{read/write}_{reg/data}() - wrappers which call
  ep93xx_pata_{read/write}() with proper timings selected,
- corrected delays in ep93xx_pata_{read/write}: time waited for IORDY
  is added to t2; wait (t0 - t2) or t2i (whichever is greater) after
  deassertion of DIOW-/DIOR-,
- removed DPRINTK, VPRINTK.

The reason for using custom delay loop is that currently ndelay() has
only 1us resolution on ep93xx. Using ndelay() gives us about 6 - 7us
total execution time for each IDE register/PIO data transfer. This
results in ~250kB/s transfers, no matter what PIO mode is selected.
Using custom delay loop, this driver transfers data at ~2,1MB/s (PIO4).
Is using such delay loop in the driver acceptable?

Thanks,
RP

[PATCH v3 1/3] PATA host controller driver for ep93xx
[PATCH v3 2/3] ep93xx: IDE driver platform support code
[PATCH v3 3/3] ep93xx: Add IDE support to edb93xx boards

 arch/arm/mach-ep93xx/core.c                  |   96 +
 arch/arm/mach-ep93xx/edb93xx.c               |   24
 arch/arm/mach-ep93xx/include/mach/platform.h |    3
 arch/arm/mach-ep93xx/soc.h                   |    1
 drivers/ata/Kconfig                          |    9
 drivers/ata/Makefile                         |    1
 drivers/ata/pata_ep93xx.c                    | 1044 +++++++++++++++++
 7 files changed, 1178 insertions(+)



More information about the linux-arm-kernel mailing list