DMA in B2Test board?

Jim Gettys jg at laptop.org
Sat Mar 10 14:25:06 EST 2007


Note: I recommend only reasonably adventurous people do this on their
personal machines: build 299 has a crash problem that has prevented it
from being declared stable for widespread release, in IRC discussions of
the last hour.  So don't deploy this build widely. We'll spin a build
300 Monday, which we hope will become the next stable build.

1) Download the Q2B76 Olpc-Q2B76.rom image from:
http://wiki.laptop.org/go/OLPC_Firmware_Q2B76

2) Rename the downloaded file to Q2B76.rom.

3) Download the build 299:

http://olpc.download.redhat.com/olpc/streams/development/build299/devel_jffs2/olpc-redhat-stream-development-devel_jffs2.img  

4) Rename this to Build299.img

5) Put these files onto a USB key.

6) Reboot your system.

7) Press the space bar during the OFW startup to get an
"ok" prompt.

8) copy-nand disk:\Build299.img

9) flash disk:\Q2B76.rom
                                     - Jim

The machine will power off.
Olpc-Q2BOlpc-Q2B76.rom76.rom
Power the machine back on.
On Sat, 2007-03-10 at 14:07 -0500, Owen Williams wrote:
> How do I install 299 on the main flash without an autoreinstallation
> image?
> 
> owen
> 
> 
> On Sat, 2007-03-10 at 12:46 -0500, Jim Gettys wrote:
> > On Sat, 2007-03-10 at 22:58 +0545, Bipin Gautam wrote:
> > > hello,
> > > I experimented with various block size for file I/O. I experimented
> > > with vfat and ext2 filesystem in the pen drive. 
> > 
> > Without specifications on the Pen drive, it's hard predict what would
> > happen.  Write performance of file systems on flash depends on whether
> > the block size of the file system matches well the underlying flash
> > block size.
> > 
> > > i also experimented
> > > with file write within nand storage; like
> > > 
> > > date
> > > dd if=/dev/zero of=/testfile bs=6144 count=102400
> > > date
> > 
> > On the NAND flash:
> > 
> > JFFS2 does data compression.
> > 
> > If you are writing zeros into a file, the actual write will extremely
> > highly compressed, and so you are appending small writes to the log; to
> > rewrite blocks on the log, you have to read the blocks, and might
> > encounter errors.  This can have significant performance consequences,
> > which I think are fixed in later builds.
> > 
> > 
> > > 
> > > The maximum preformance of XO nand storage (alone) was ~1.8 mb/sec
> > > (jffs2) while the preformance of external usb drive is ~ 3-11 mb /sec
> > > (at various block size)
> > 
> > Please install build 299, rather than 239.  There has been work on the
> > NAND driver to highly optimize the error correction path in the NAND
> > driver, which I believe was not in build 239, but IIRC, is in build 299.
> > 
> > So if you had NAND with errors, the error correction path was a possible
> > source of CPU usage.  I believe the error correction code is now 100 or
> > more times faster, courtesy of a community developer whose name slips my
> > mind.
> > 
> > 
> > > 
> > > i thought 100% cpu use is because direct memory access(DMA) for disk
> > > I/O has some problems (or disabled). Also, running  XO OS[1] in
> > > virtual machine inside Parallels Workstation 2.2 with host OS xp sp2
> > > shows the same problem of 100% cpu use during disk i/o
> > > 
> > > 
> > > i cant give you a detailed stats. right now as my B2 OS is dead and
> > > i'm downloading the latest build. (maybe tomorrow... if you want to
> > > hear it)
> > 
> > Please install build 299.
> > 
> > > 
> > > with regards,
> > > -bipin
> > > [1]olpc-redhat-stream-development-build-239-20070118_1416-devel_ext
> 
-- 
Jim Gettys
One Laptop Per Child





More information about the Devel mailing list