LogFS for F11-XO-1?

Jörn Engel joern at logfs.org
Thu Nov 26 07:03:47 EST 2009


On Thu, 26 November 2009 12:46:35 +0100, Martin Langhoff wrote:
> On Thu, Nov 26, 2009 at 12:37 PM, Jörn Engel <joern at logfs.org> wrote:
> > On Thu, 26 November 2009 10:54:58 +0100, Martin Langhoff wrote:
> >>
> >> Would be interesting to see what the performance & stability is.
> >
> > I'd love to see results.
> 
> We are currently making F11-based builds, I think w 2.6.30. Does logfs
> Just Work there if we apply the patches?

Assuming the patches apply or you massage them properly to apply, logfs
(the kernel code) should Just Work.  You will also have to create the
filesystem, using mklogfs.  Source can be found here:
http://logfs.org/git?p=logfsprogs/.git;a=summary

I haven't written an image generator.  If I were sufficiently motivated
to write one (I'm not), I'd propably resurrect my userspace vfs
http://logfs.org/git?p=testlogfs;a=summary
and just call the fake_creat and fake_pwrite functions to fill the
image.  Or you could loop-mount on the development host, fill a
filesystem and use that as an image.

> >> If it works, buried in the patchseries announcement there is a hint at
> >> support for async ops. Hard to know how it'd interact with the timings
> >> of our CaFe, but it could be interesting to retry the async patch.
> >
> > I wouldn't expect spectacular results for the X0.  Iirc it has a single
> > flash chip and the controller can queue a total of one command at a
> > time.
> 
> Thanks for the hint. For the record -- and in case anyone wonders --
> this threads records the proposed patch to our CaFe driver and the
> results http://lists.laptop.org/pipermail/devel/2008-December/021677.html

With my device (sorry, driver is not available just yet), tasklets
performed significantly better than workqueues.  When implementing the
asynchronous fio_* methods you don't have to deal with wait_event at
all, which is nice.  But you still need wrapper functions for the
synchronous methods.  Most likely those wrappers can be identical to
those in my driver.

I really should clean it up and see that I can publish the thing.

Jörn

-- 
Simplicity is prerequisite for reliability.
-- Edsger W. Dijkstra



More information about the Devel mailing list