Rui Sousa wrote:
>
> Jeff Garzik wrote:
> >
> > In 2.3.99-pre9-pre1, there are many changes to the struct pci_device_id
> > PCI probe tables. Unfortunately these changes were made rather blindly
> > and some are not correct. (If these are Rui's changes, I fear I did not
> > speak up soon enough, causing these errors)
> >
> > __devinitdata should NOT be used unless you are certain the driver is a
> > hotplug driver. Otherwise __initdata should be used. Similar rules for
> > __init and __devinit.
> >
> > The patch in 2.3.99-pre9-pre1 blindly moves everything to __devinitdata,
> > which causes memory wastage in non-hotplug kernels (ie. the majority of
> > the kernels out there).
> >
> > Each driver author should carefully consider their __initdata markers
> > versus their __devinitdata markers, and doing a global conversion like
> > that in patch-2.3.99-pre9-pre1 is bound to "undo" carefully thought-out
> > changes by driver authors and maintainers.
> >
>
> Attached is a patch against pre9-1 that solves the mess I made.
> Please check that I got it right this time and that the text in
> Documentation/pci.txt is correct.
Looks ok, thanks. Good catch on the sis900 missing null terminator
BTW. Also, considering emu10k1, you might want to add a pci.txt tip
that says something like:
If you have code which is always called on module_exit(), and
potentially called from module_init() or probe(), then it should not be
marked with '__exit', as it could potentially be dropped at the linker
stage, and not be present at all in the kernel image. Data under
__exitdata should be treated likewise.
A sneaky (and undocumented) thing to do is marking such code __init
instead of __exit. That will wind up with the optimal behavior on
current kernels, but doing so may invite broken code down the road if
semantics change...
Jeff
-- Jeff Garzik | Liberty is always dangerous, but Building 1024 | it is the safest thing we have. MandrakeSoft, Inc. | -- Harry Emerson Fosdick- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon May 15 2000 - 21:00:23 EST