[ECOS] Is it the right forum?
Mon Dec 10 03:46:00 GMT 2001
Jonathan Larmour <email@example.com> writes:
> Dan Conti wrote:
> > Were the mutex optimization patch to be applied, i'd be more than happy
> > to provide ARM hal support.
> > Is the issue that Luoqi is uninterested in doing the copyright
> > assignment, or just one of those never got around to it things?
> Well I know Nick is unhappy with the patch as it stands, but we can work
> through that I would hope. But there's no point even trying unless we know
> we can accept the code. Luoqi has expressed interest in the assignment. No
> doubt it's the usual problem that it's the company lawyers who hold it up.
> Red Hat haven't exactly got a mega-responsive legal department either :-|.
For the record, my reasons for being unhappy with the patch are:
- It conflicts with my own plans for a spinlock based mutex
implementation. A spinlock implementation and the current
implementation can be combined fairly seamlessly, whereas the
compare-and-swap version would have to exist as an alternative, with
the maintenance problems that entails. The mutex code is already
getting to be a maze of twisty little ifdefs, I want to avoid more.
- Compare-And-Swap is not supported on all architectures - the ARM is
one of those. Even where support is available in the architecture,
it is often omitted in the embedded variants that we have to run
on. I am reluctant to introduce changes into generic code that can
only be used on a limited range of targets.
- I am unhappy about some of the asumptions made in the code.
Specifically: that the least significant bit of a pointer is
available for the "want" bit, and that a pointer will fit into the
size of data object controlled by compare-and-swap.
Nick Garnett, eCos Kernel Architect
Red Hat, Cambridge, UK
More information about the Ecos-discuss