bmcsensors port to 2.6
yani.ioannou at gmail.com
Sun Mar 27 06:16:59 CEST 2005
As suggested after some cleanup of the code and waiting on responses
from testers I'm now submitting the latest patch against the 188.8.131.52
kernel for inclusion into the kernel/lm_sensors. Due to the size of
the uncompressed patch I'll have to link to the sourceforge download
page rather than inline it:
The driver is a direct port over of the 2.4 bmcsensors and i2c-ipmi
drivers. I've merged all the patches that were applied to the 2.4
version since the original port a year ago, and applied a patch
supporting ipmi 0.9 (which was also back-ported to the 2.4 version),
as well as a patch moving initialization to a kernel thread. Finally I
cleaned up the code a bit to better meet the guidelines in
Signed-off-by: Yani Ioannou <yani.ioannou at gmail.com>
PS: I have a patch adding support for dynamic callbacks in device.h,
but I'm porting over the bmcsensors driver to make sure that all works
as expected before posting it here. I'll let post my results when I
have it working.
On Tue, 15 Mar 2005 17:36:59 -0500, Mark Studebaker <mds at mds.gotdns.com> wrote:
> I suggest you submit it to us now (email the patch with an explanation and a signed-off-by line)
> and work on the dymaic improvements as a phase 2, after the patch is hopefully accepted.
> Yani Ioannou wrote:
> > Just to follow up on Martin's message, the problem was discussed in
> > more detail in this mailing list before (see
> > http://archives.andrew.net.au/lm-sensors/msg26067.html), however
> > nothing has addressed this problem since.
> > If I was to create a patch that allowed a more functional device sysfs
> > callback, how likely would it ever be accepted into the mainstream
> > kernel?
> > To highlight the problem, having a maximum of 150 sensors in the
> > bmcsensors drivers results in a kernel module 110kb in size, compared
> > to a size of ~50kb for the driver with a limit of 50 drivers (due to
> > the extra callback functions and pointers). Aside from the kludge that
> > is the non-dynamic callbacks the driver uses currently it causes a
> > real and unacceptable (in my opinion) inefficiency.
> > If there really is going to be no way around this then I would suggest
> > merging the bmcsensors driver into the mainstream kernel, I believe it
> > is stable enough.
> > Yani
> > On Fri, 11 Mar 2005 15:21:17 +0100, Bene Martin
> > <martin.bene at icomedias.com> wrote:
> >>the bmcsensors port for kernel 2.6 by Yani Ioannou (available on
> >>sourceforge, http://bmcsensors-26.sourceforge.net/) looks quite
> >>functional now.
> >>As also noted by Yani, one thing is really ugly right now: at compile
> >>time, bmcsesnors has no idea how many sensors a board provides; this
> >>seems to range from ~20 Sesnors for simple boards up to > 60 for a dell
> >>The sysfs interface requires a seperate callback function for read, min,
> >>max, label, write min and write max; currently the driver assumes 100
> >>sensors max and just declares callback functions for all of them; while
> >>this works ist neither nice code nor especially efficient since most of
> >>those functions generaly aren't required.
> >>Anyone got an idea how to better handle an unknown number of sensors in
> >>the driver?
> >>Thanks, Martin
More information about the lm-sensors