Re: [PATCH 1/1] Revert "drm/i915: drop i915_ prefix from enable_rc6, enable_fbc, enable_ppgtt parameters"

[Amit Shah <amit.shah@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 4827 bytes --]

On (Thu) 17 Jul 2014 [14:50:12], Amit Shah wrote:
> On (Thu) 17 Jul 2014 [11:11:15], Daniel Vetter wrote:
> > On Thu, Jul 17, 2014 at 02:32:41PM +0530, Amit Shah wrote:
> > > On (Thu) 17 Jul 2014 [09:35:20], Daniel Vetter wrote:
> > > > On Wed, Jul 16, 2014 at 9:54 PM, Linus Torvalds
> > > > <torvalds@linux-foundation.org> wrote:
> > > > > Sorry for the top post, I'm on the road..
> > > > >
> > > > > In wondering if we couldn't just keep both the old an the new names and have
> > > > > them both point at the same variable? Remove the description for the old
> > > > > name, but keep it working?
> > > > 
> > > > I'm really surprised here ... We have rc6 enabled by default
> > > > everywhere, and all the additional rc6 levels that users try to enable
> > > > are known to hard-hang machines.
> > > 
> > > I haven't had this problem on my hardware (ThinkPad T420s, lspci
> > > below) for a few kernel versions.  I think I added the enable_rc6=
> > > setting back from the time the deeper states were enabled and then
> > > reverted for SandyBridge.
> > > 
> > > Nevertheless, with the current state, RC6p and RC6pp states are not
> > > used.
> > 
> > Yeah, on snb they cause crashes and instability and also don't provide
> > measurable power benefits (afaik). So I recommend you drop that one.
> 
> Not for me -- there have been no crashes / hangs / lockups as I
> mentioned.
> 
> > > > I actually have plans to taint the
> > > > kernel if you set any of them since I'm fed up with the random crash
> > > > reports. Same for fbc, even more so or the ppgtt knob. My stance is
> > > > that if you know about these knobs you _really_ should know the driver
> > > > to its depths and so also be able to follow module parameter
> > > > renamings.
> > > 
> > > I also remember there being bugzillas about power consumption, and
> > > using this setting was recommended (for Fedora, I think).  I know
> > > a few people are using this setting.
> > 
> > I know, google is littered with such entries. Unfortunately by the time
> > google thinks something is important (which usually takes a few months)
> > it's already badly outdated: i915 graphics developement is charging ahead
> > at a really brisk pace - we merge a few hundred patches per release for
> > i915 alone.
> 
> But for SNB, there's really no "improvement" for the RC6 states, is
> there?
> 
> > > > > On Jul 16, 2014 8:34 AM, "Amit Shah" <amit.shah@redhat.com> wrote:
> > > > >>
> > > > >> This reverts commit 3adee7a7976012a20f1d3b5a529a3c105e29fef1.
> > > > >>
> > > > >> After upgrading to v3.15, my laptop's battery started draining quite
> > > > >> fast.  Powertop pointed to the deep RC6 states not being used.  The
> > > > >> kernel param I had put to enable them had stopped working the way it
> > > > >> used to; so I disagree with the 'not maintaing ABI' part of the param
> > > > >> name change.
> > > > >>
> > > > >> However weird the names may be, they're in active use and changing them
> > > > >> only causes pain for users.  This also isn't advertised (marked
> > > > >> deprecated, big warning shown, etc.), so just reverting now.
> > > > >>
> > > > >> CC: Daniel Vetter <daniel.vetter@ffwll.ch>
> > > > >> CC: Jani Nikula <jani.nikula@linux.intel.com>
> > > > >> CC: David Airlie <airlied@linux.ie>
> > > > >> CC: <stable@vger.kernel.org> # v3.15+
> > > > >> Signed-off-by: Amit Shah <amit.shah@redhat.com>
> > > > 
> > > > Anyway we need to figure out what went wrong here. Please share your
> > > > exact kernelcmdline and lspci -nn. Also stats for before/after from
> > > > powertop when idle please.
> > > 
> > > Powertop stats for idle are a little difficult -- since this is my
> > > primary laptop.
> > 
> > Now I'm a bit confused: How have you measured that the lack of rc6p/pp is
> > the reason for your power consumption regression?
> > -Daniel
> 
> What I meant was rebooting in the middle of something is a pain
> (usually a week or two between trying these things); and also for a
> fair comparison, the workloads have to be similar for both the
> powertop ratings.

Attached are the two powertop runs with 'powertop --html'.  Both are
taken on a fresh reboot on the same kernel, just the difference is
that powertop-rc6pp.html has i915.enable_rc6=7 in the cmdline.  Dunno
if that's what you wanted, but it shows the laptop fan is spinning
with more rpm, fwiw.  Both runs are taken after about 10 mins of
desktop idling.


> In any case, my daily work doesn't change, and I noticed this
> immediately upon booting into 3.15.  The laptop heats up a bit more,
> that's the first clue; and the battery doesn't provide as much backup
> as it used to.

As I stated, I've used the deeper states on this h/w for a while w/o
any adverse effects.  So please consider picking this revert, or
enable the deeper states for this h/w.



		Amit

[-- Attachment #2: powertop-rc6pp.html --]
[-- Type: text/html, Size: 57891 bytes --]

[-- Attachment #3: powertop-norc6.html --]
[-- Type: text/html, Size: 53393 bytes --]

[-- Attachment #4: Type: text/plain, Size: 159 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx