Nouveau - Jun 2013 - [Bug 66177] New: pwm1 value not restored during hibernate/restore cycle in the event of manual fan management mode

https://bugs.freedesktop.org/show_bug.cgi?id=66177

          Priority: medium
            Bug ID: 66177
          Assignee: nouveau at lists.freedesktop.org
           Summary: pwm1 value not restored during hibernate/restore cycle
                    in the event of manual fan management mode
        QA Contact: xorg-team at lists.x.org
          Severity: normal
    Classification: Unclassified
                OS: All
          Reporter: mr.dash.four at googlemail.com
          Hardware: Other
            Status: NEW
           Version: unspecified
         Component: Driver/nouveau
           Product: xorg

When I have done the following:

echo 1 > /sys/class/hwmon/hwmon0/pwm1_enable 
echo 30 > /sys/class/hwmon/hwmon0/pwm1 

pwm1=30
pwm1_enable=1
pwm1_max=100
pwm1_min=0
temp1_auto_point1_pwm=100
temp1_auto_point1_temp=90000
temp1_auto_point1_temp_hyst=3000
temp1_crit=115000
temp1_crit_hyst=2000
temp1_emergency=135000
temp1_emergency_hyst=5000
temp1_input=41000
temp1_max=95000
temp1_max_hyst=3000
update_rate=1000

and then do hibernate/restore, I get this:

pwm1=100
pwm1_enable=1
pwm1_max=100
pwm1_min=0
temp1_auto_point1_pwm=100
temp1_auto_point1_temp=90000
temp1_auto_point1_temp_hyst=3000
temp1_crit=115000
temp1_crit_hyst=2000
temp1_emergency=135000
temp1_emergency_hyst=5000
temp1_input=37000
temp1_max=95000
temp1_max_hyst=3000
update_rate=1000

The "pwm1_enable" value is restored, thanks to the patch proposed by
Emil
Velikov in Bug 66022, but that doesn't restore the pwm1 value in case of
manual
fan management. I am attaching 2 patches to this report:

1. which fixes this bug and restores everything ("pwm1_enable", as
well as
"pwm1"); and

2. a separate patch, allowing fan_mode and fan duty (percentage) to be
specified at the command line, allowing me to set my fan speed immediately and
not waiting for OS scripts to do that for me as was the case up until now.

Since the automatic fan management is, well, lets just say incomplete,
that's
currently the best option to not have my fan wail like a jet engine every time
I boot up and use my machine.

Patches follow...

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130625/9c82b4ab/attachment-0001.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #1 from Mr-4 <mr.dash.four at googlemail.com> ---
Created attachment 81430
  --> https://bugs.freedesktop.org/attachment.cgi?id=81430&action=edit
Restore the pwm1_enable and pwm1 (in case of manual fan management) after
hibernate/restore

This patch restores pwm1_enable as well as pwm1 (in case of manual fan
management) values after hibernate/restore.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130625/9c6ba3cc/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #2 from Mr-4 <mr.dash.four at googlemail.com> ---
Created attachment 81431
  --> https://bugs.freedesktop.org/attachment.cgi?id=81431&action=edit
set fan_mode and fan_duty at boot up

The following patch allows fan_mode and fan_duty (percentage) to be specified
at the kernel command line

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130625/050b660f/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #3 from Martin Peres <martin.peres at ensi-bourges.fr> ---
The real solution would be to restore all sysfs attributes after reboot. The
proposed solution doesn't seem scalable :s However, I fully agree something
should be done here!

I would rather see an automatic fan management by default and some overrides to
disable it though. Please use the "config" parameter to specify them
instead of
adding parameters to nouveau, it seems to be how darktama wants things to work.

Automatic fan management is supposed to work on all cards < NVC0 (which has
it
at boot time, by default). I discussed enabling it by default for Linux 3.11
but it seems we'll need to wait until 3.12.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130626/b9002c36/attachment-0001.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #4 from Mr-4 <mr.dash.four at googlemail.com> ---
(In reply to comment #3)
> The real solution would be to restore all sysfs attributes after reboot.
Indeed, I agree.
> The proposed solution doesn't seem scalable :s
I never claimed that it is - the patch was submitted (note that it wasn't
formal submission either) barely as a guidance - it-works-for-me (tm) (;
> However, I fully agree something should be done here!
Yep.
> I would rather see an automatic fan management by default and some
overrides
> to disable it though. Please use the "config" parameter to
specify them
> instead of adding parameters to nouveau, it seems to be how darktama wants
> things to work.
How do I "use the config" exactly? And who is "darktama" -
is that Ben Skeggs?
> Automatic fan management is supposed to work on all cards < NVC0 (which
has
> it at boot time, by default). I discussed enabling it by default for Linux
> 3.11 but it seems we'll need to wait until 3.12.
I am, quite frankly, amazed that something so raw and incomplete as the
automatic fan management could make it into the mainstream kernel code - Ben is
usually thorough as far as these things are concerned and I don't know how
this
slipped through.

The "automatic" mode doesn't work - far from it, as evident from
my other
submission (and an additional feedback provided) - see Bug #66022 (and Comment
#5 in particular).

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130626/c4badac9/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #5 from Martin Peres <martin.peres at ensi-bourges.fr> ---
(In reply to comment #4)
> (In reply to comment #3)
> > The real solution would be to restore all sysfs attributes after
reboot.
> Indeed, I agree.
> 
> > The proposed solution doesn't seem scalable :s
> I never claimed that it is - the patch was submitted (note that it
wasn't
> formal submission either) barely as a guidance - it-works-for-me (tm) (;
Ok.
> How do I "use the config" exactly? And who is
"darktama" - is that Ben
> Skeggs?
Yes, Darktama is Ben Skeggs. The config option is a string that is used to pass
multiple parameters.
> > Automatic fan management is supposed to work on all cards < NVC0
(which has
> > it at boot time, by default). I discussed enabling it by default for
Linux
> > 3.11 but it seems we'll need to wait until 3.12.
> I am, quite frankly, amazed that something so raw and incomplete as the
> automatic fan management could make it into the mainstream kernel code -
Ben
> is usually thorough as far as these things are concerned and I don't
know
> how this slipped through.
> 
> The "automatic" mode doesn't work - far from it, as evident
from my other
> submission (and an additional feedback provided) - see Bug #66022 (and
> Comment #5 in particular).
Well, I agree that it is my fault that I may not have thought about the
suspend/resume cycle. Thanks for reporting that. However, I'm afraid you are
hit by another bug, a bug in the ALARM mechanism, which explains why sometimes,
the fan speed isn't updated anymore. This is very worrying and may force us
to
move to using Linux's periodic tasks instead.

Thanks for this review. I'll work on fixing this ASAP. I can work on the
sysfs
restoration but I have no card with a PWM fan with me here. I may then require
you to test some stuff.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130628/97dc8550/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #6 from Mr-4 <mr.dash.four at googlemail.com> ---
(In reply to comment #5)
> Well, I agree that it is my fault that I may not have thought about the
> suspend/resume cycle. Thanks for reporting that. However, I'm afraid
you are
> hit by another bug, a bug in the ALARM mechanism, which explains why
> sometimes, the fan speed isn't updated anymore. This is very worrying
and
> may force us to move to using Linux's periodic tasks instead.
It is indeed a serious bug as the average user may not be as careful as I was
or may not necessarily pay attention to the card temperature after
hibernate/restore, particularly if the "auto" fan management was
working prior
to that. If that happens, the video card is almost certain to burn out!
> Thanks for this review. I'll work on fixing this ASAP. I can work on
the
> sysfs restoration but I have no card with a PWM fan with me here. I may
then
> require you to test some stuff.
Have I ever let you down? (;

When you've made your first call for testing of the therm features over 18
months ago, I responded and did quite a lot of some testing (you can't deny
that, can you?), while also benefiting from the "auto" fan management
which
was, at the time, working perfectly. It was simply sublime and was a great pity
that it didn't make it into the main nouveau repository at the end (I used
your
git thermal fork for absolute ages).

So, I'll be able to help you here with the testing as well, as long as there
is
willingness on your and Ben's part to fix these, in my view, very serious
bugs.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130628/f99fe83a/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #7 from Mr-4 <mr.dash.four at googlemail.com> ---
Sorry, one other thing I forgot to mention, which should be pretty obvious,
given what I reported so far:

The "auto" fan management seems to work as expected when I do *not*
use it
after hibernate/restore.

In other words, when I boot up my machine (which, by default, has fan
management disabled) and then select "auto" fan management with
"echo 2 >
../pwm1_enable", then things are OK - the fan duty (pwm1) constantly hovers
between 20 and 28, depending on my card temperature, keeping that temperature
under control. No problems there.

Things are then *really* screwed up when I do hibernate/restore as reported
previously. So, I suspect the hibernate/restore does something to the nouveau
driver which prevents the auto mode from working properly.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130628/0ed269a5/attachment-0001.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #8 from Martin Peres <martin.peres at ensi-bourges.fr> ---
(In reply to comment #6)
> (In reply to comment #5)
> > Well, I agree that it is my fault that I may not have thought about
the
> > suspend/resume cycle. Thanks for reporting that. However, I'm
afraid you are
> > hit by another bug, a bug in the ALARM mechanism, which explains why
> > sometimes, the fan speed isn't updated anymore. This is very
worrying and
> > may force us to move to using Linux's periodic tasks instead.
> It is indeed a serious bug as the average user may not be as careful as I
> was or may not necessarily pay attention to the card temperature after
> hibernate/restore, particularly if the "auto" fan management was
working
> prior to that. If that happens, the video card is almost certain to burn
out!
Well, luckily, this isn't true :) PTHERM sends the kernel some IRQs whenever
temperature reaches some threshold. The first one is fan-boost which puts the
fan to 100%. The last one turns off the computer.

This feature shouldn't be affected by resume/suspend.
> 
> > Thanks for this review. I'll work on fixing this ASAP. I can work
on the
> > sysfs restoration but I have no card with a PWM fan with me here. I
may then
> > require you to test some stuff.
> Have I ever let you down? (;
> 
> When you've made your first call for testing of the therm features over
18
> months ago, I responded and did quite a lot of some testing (you can't
deny
> that, can you?), while also benefiting from the "auto" fan
management which
> was, at the time, working perfectly. It was simply sublime and was a great
> pity that it didn't make it into the main nouveau repository at the end
(I
> used your git thermal fork for absolute ages).
Hey hey, thanks for being supportive. I indeed recognized your nick straight
away.

The new model is nicer than it used to be from a dev point of view. As a user,
I agree the previous system was better. I'll try to improve on that as time
permits :)
> 
> So, I'll be able to help you here with the testing as well, as long as
there
> is willingness on your and Ben's part to fix these, in my view, very
serious
> bugs.
I agree. I am away from France for a few more months, but I can use my laptop
and fake a fan to check everything is right. I'll see what I can do in a
timely
fashion.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130628/abe75ece/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

--- Comment #9 from Martin Peres <martin.peres at ensi-bourges.fr> ---
(In reply to comment #7)
> Sorry, one other thing I forgot to mention, which should be pretty obvious,
> given what I reported so far:
> 
> The "auto" fan management seems to work as expected when I do
*not* use it
> after hibernate/restore.
> 
> In other words, when I boot up my machine (which, by default, has fan
> management disabled) and then select "auto" fan management with
"echo 2 >
> ../pwm1_enable", then things are OK - the fan duty (pwm1) constantly
hovers
> between 20 and 28, depending on my card temperature, keeping that
> temperature under control. No problems there.
That's what I tested and it indeed worked beautifully. But I only tested
that
on my desktop PCs and I never suspend them. Silly me...
> 
> Things are then *really* screwed up when I do hibernate/restore as reported
> previously. So, I suspect the hibernate/restore does something to the
> nouveau driver which prevents the auto mode from working properly.
Again, what worries me the most is the fact that the fan stop being updated
randomly. That's a SERIOUS bug! For the sysfs bug, I'll try to come up
with an
acceptable solution for both users and devs.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130628/cc6f4c47/attachment.html>

https://bugs.freedesktop.org/show_bug.cgi?id=66177

Martin Peres <martin.peres at ensi-bourges.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|nouveau at lists.freedesktop.o |martin.peres at
ensi-bourges.f
                   |rg                          |r
  Attachment #81430|0                           |1
        is obsolete|                            |

--- Comment #10 from Martin Peres <martin.peres at ensi-bourges.fr> ---
Created attachment 83879
  --> https://bugs.freedesktop.org/attachment.cgi?id=83879&action=edit
[PATCH 1/5] drm/nouveau/therm: Set the correct pwm_mode upon resume

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/nouveau/attachments/20130809/4d013b13/attachment.html>