Re: [PATCH 3/3] i8k: Remove laptop specific config data (fan_mult, fan_max) from driver

From: Gabriele Mazzotta
Date: Sat Dec 27 2014 - 09:13:42 EST


On Thursday 25 December 2014 22:54:34 Gabriele Mazzotta wrote:
> On Thursday 18 December 2014 12:08:58 Pali Rohár wrote:
> > On Wednesday 10 December 2014 14:32:16 Gabriele Mazzotta wrote:
> > > On Wednesday 10 December 2014 12:51:30 Pali Rohár wrote:
> > > > On Tuesday 09 December 2014 21:07:01 Pali Rohár wrote:
> > > > > Now we have autodetection code for fan multiplier and
> > > > > maximal fan speed so we do not need to have those
> > > > > constants for each laptop in kernel driver code.
> > > > >
> > > > > Signed-off-by: Pali Rohár <pali.rohar@xxxxxxxxx>
> > > > > ---
> > > > > !!!Please do not apply this patch until all affected
> > > > > machines will be tested!!!
> > > > >
> > > > > I tested autodetection code only on Dell Latitude E6440
> > > > > (where it worked). Other machines which needs to be
> > > > > tested:
> > > > >
> > > > > Dell Latitude D520
> > > > > Dell Latitude E6540
> > > > > Dell Precision WorkStation 490
> > > > > Dell Studio
> > > > > Dell XPS M140 (MXC051)
> > > > > ---
> > > >
> > > > Can somebody else with dell laptops test this patch series?
> > >
> > > i8k_get_fan_nominal_rpm() returns -22 on my XPS13, so nothing
> > > changed with this patch series applied.
> > >
> > > Gabriele
> >
> > So your BIOS cannot report nominal_rpm and because your machine
> > is not in dmi list, all 3 patches do nothing for your machine.
> >
> > But you need to set multiplier to 1, right?
> >
> > What about this patch? (on top of 3/3)
> >
> > --- a/drivers/char/i8k.c
> > +++ b/drivers/char/i8k.c
> > @@ -850,6 +850,10 @@ static int __init i8k_probe(void)
> > */
> > for (fan = 0; fan < I8K_FAN_COUNT; ++fan) {
> > i8k_fan_mult[fan] = I8K_FAN_DEFAULT_MULT;
> > + if (i8k_get_fan_rpm(fan) > I8K_FAN_MAX_RPM) {
> > + i8k_fan_mult[fan] = 1;
> > + continue;
> > + }
> > for (val = 0; val < 256; ++val) {
> > ret = i8k_get_fan_nominal_rpm(fan, val);
> > if (ret > I8K_FAN_MAX_RPM) {
> >
> >
>
> Hi,
>
> I'm sorry for replying only now, but I couldn't follow this thread
> closely and I'm a bit lost now. I haven't tested the suggested
> change, but I don't think it would work in a reliable way. It's not
> rare for the fan to be completely stopped, especially on boot. You
> are right though, 1 is the right multiplier.

I took a better look at the code and my laptop is indeed able to report
the nominal speed. The problem with the original patch was that the
first call of i8k_get_fan_nominal_rpm() correctly returned -22 (the fan
doesn't exist) and the loop was terminated because of that. I tested
the following patch http://www.spinics.net/lists/kernel/msg1892101.html
and the fan multiplier auto detection worked.

I confirm that the suggested change in case the nominal speed can't
be retrieved is not OK as its outcome depends on the current state of
the fans.

> Guenter, while I was trying to catch up, I noticed that the support
> for the XPS 13 [1] will be added. May I ask you which revision of
> the laptop was tested? I own the 9333 one and I'm not sure that
> having i8k automatically loaded is a good idea. The reason is that
> reading and setting the speed of the right (and only) fan causes a
> freeze of the laptop of some milliseconds, enough to be annoying and
> noticeable. I'm worried of the effect this might have in the everyday
> use, users might start noticing random freezes because of one
> application that all the sudden is able to read the speed of the fan.
> I don't if the same happens on all the other Dell systems, this is
> the first and only Dell laptop I owned.
>
> If the tested laptop wasn't the XPS13 9333 and this problem does not
> exists, would it be possible to make the DMI_PRODUCT_NAME string such
> that it matches only the tested revision? The string that identifies
> my laptop is "XPS13 9333".
>
> Gabriele
>
> [1] http://www.spinics.net/lists/kernel/msg1878801.html

I modified i8k so that it prints the time required to execute the
assembly code in i8k_smm(). Here what I got:

[ 4697.485130] i8k_smm asm: 1965800 ns #pwm1
[ 4697.487383] i8k_smm asm: 1375057 ns #temp3_input
[ 4697.489477] i8k_smm asm: 1112493 ns #temp3_label
[ 4697.991687] i8k_smm asm: 500796086 ns #fan1_input
[ 4697.998245] i8k_smm asm: 1957365 ns #fan1_label
[ 4698.009190] i8k_smm asm: 1770247 ns #temp1_input
[ 4698.014416] i8k_smm asm: 749734 ns #temp1_label
[ 4698.019103] i8k_smm asm: 2503962 ns #temp2_input
[ 4698.023490] i8k_smm asm: 1738982 ns #temp2_label

As you can see, the time required to read the speed of the fan is
substantially higher than the combined time of all the other reads.

Is the difference so big for the other systems that have been tested?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/