patch-2.4.19-rmk7-ds1

The latest patch for IOP3xx, IXP2xxx, IXP4xx, and IXP12xx systems is
available at:

  ftp://source.mvista.com/pub/ds-patches/2.4/patch-2.4.19-rmk7-ds1.gz

An incremental patch from ds1 to ds2 is also avaialable:

  ftp://source.mvista.com/pub/ds-patches/2.4/inc-2.4.19-rmk6-ds2-rmk7-ds1.gz

(If anyone can think of a better naming scheme for incremental patches,
 please send me your suggestion).

To access my working tree with BK:

   bk clone http://source.mvista.com:14690/linux-arm-2.4-ds

Build Documentation:

  http://www.arm.linux.org.uk/docs/kerncomp.shtml

Summary of Changes:

IXP2xxx:

- Added IXP2000 driver to access slave NPUs over PCI (Me)

- Updated default config so default kernel works with IXA SDK (Naeem Afzal)

IO3xx:

- IQ80321 E1000 driver "hook" to make soft reset work (Drew Moseley)

- Fix AIC7xxx driver to not consistent_alloc memory in IRQ context (Dave Jiang)

- Updated AUU driver (Dave Jiang, David Griego, Me)

  As far as I can tell nobody was using the old AAU driver, so Dave has
  written a new driver with the specific purpose of accelerating the
  Linux RAID stack. For those not super familiar with the AAU, it's a
  HW XOR unit designed specifically for that purpose. The code has only
  been tested on IOP321, but is written to dynamically determine what
  IOP it's running on and access the correct registers. Please play with
  it and send us your feedback. Note that the XOR test that's done
  in the beggining will actually result in lower performance with
  the AAU acceleration since it only tests with 2 disks and the overhead
  for just two buffers is a bit large. Since 2 disks/buffers is not a
  realistic RAID5 scenario, we disable the SW routines when the AAU 
  acceleration is turned on. Dave's performance testing has shown that in
  the real world, we get a huge drop in CPU utilization:

  4 disk RAID 5 (slow drives, I can't find any U160s, they are about 9GB each)

  RAID resync (build at mkraid)

  8 min 40s for AAU with ~70% CPU idle 

  8 min 50s for CPU with ~40% CPU idle

  Given that these are slow disks, we're probably hitting the disk speed
  wall.

  The AAU supports 32 buffers at once, but the Linux RAID stack can
  only do 5 max.  If someone has the free time to clean up the RAID
  code to support > 5 buffers, we could see some massive performance
  gains on large RAID5 applications. :)

  If someone was actually using the old AAU driver, the old code is
  still available in my old patches and you are welcome to maintain it
  since both I and Intel folks don't plans to do so.

IXP4xx:

- Added support for B0 silicon to IXP425 PCI port (Me)

- Reverted to old static I/O mappings so that Intel CSR works (Me)

Enjoy,
~Deepak

-- 
Deepak Saxena
MontaVista Software - Powering the Embedded Revolution - www.mvista.com


_______________________________________________
http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-announce
Please visit the above address for information on this list.