firefly-linux-kernel-4.4.55.git
10 years agodrm/i915: Introduce new intel_output_name()
Damien Lespiau [Wed, 8 Jan 2014 14:18:23 +0000 (14:18 +0000)]
drm/i915: Introduce new intel_output_name()

That we can use for debugging purposes.

v2: Use designated initializers for the 'names' array (Paulo Zanoni,
    Jani Nikula).
    Add a check in case the array has a hole (which can now remain
    unnoticed with designated initializers) (Jani Nikula)

Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com> (for v1)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't set modes for 2 connectors on the same encoder
Paulo Zanoni [Tue, 7 Jan 2014 16:55:53 +0000 (14:55 -0200)]
drm/i915: don't set modes for 2 connectors on the same encoder

In some cases we have more than 1 connector associated to an encoder
(e.g., SDVO, Haswell DP/HDMI) and we can only set a mode for one of
these connectors. If we only allowed modesets for connected connectors
we would never need this patch, but since we do allow modeset for
disconnected connectors we may see user space trying to set modes on
the two connectors attached to the same encoder, so we need to forbid
that.

This problem can be reproduced by running the following
intel-gpu-tools test case:
  ./kms_setmode --run-subtest clone-exclusive-crtc

Thanks to Daniel Vetter for providing a version of this patch on
pastebin.

Credits-to: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: i830M has watermarks like i855
Daniel Vetter [Sat, 14 Dec 2013 22:38:30 +0000 (20:38 -0200)]
drm/i915: i830M has watermarks like i855

So shuffle the checks around a bit. Also give all the structs and
functions proper prefixes: i830_ for the dual-pipe mobile platforms
and i845_ for the two single-pipe desktop platforms.

Note that the max fifo value isn't actually correct for the i830M, but
since we don't frob the fifo split we don't actually need it. This is
different for some gen3 devices where we need the full fifo for self
refresh mode.

Cc: Thomas Richter <richter@rus.uni-stuttgart.de>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop I915_ prefix from HAS_FBC
Daniel Vetter [Fri, 10 Jan 2014 07:50:12 +0000 (08:50 +0100)]
drm/i915: Drop I915_ prefix from HAS_FBC

My OCD just couldn't let this slide. Spotted while reviewing Ville's
patch to only flip planes when we have FBC.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't swap planes on 830M
Ville Syrjälä [Tue, 7 Jan 2014 14:15:37 +0000 (16:15 +0200)]
drm/i915: Don't swap planes on 830M

Looks like 830M doesn't quite like it when you try to move a plane from
one pipe to another. It seems that the plane's old pipe has to be active
even if the plane is already disabled, otherwise the relevant register
just won't accept new values.

The following commit:

 commit 1f1c2e2468f937cefd6bcb645c959c7b5d9821df
 Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
 Date:   Thu Nov 28 17:30:01 2013 +0200

    drm/i915: Swap primary planes on gen2 for FBC

caused a regression on 830M. It will attempt to swap the planes when the
driver is loaded, but at that time only pipe A might be active, so plane
A gets disabled, but plane B won't get enabled since pipe B is not
active when we try to move the plane over to pipe A.

There's no reason to swap planes on 830M since it doesn't support
FBC. Change the logic a bit to limit the plane swapping to platforms
which actually support FBC. This should avoid getting a black screen on
830M.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix wrong PLL debug messages.
Paulo Zanoni [Wed, 8 Jan 2014 13:12:28 +0000 (11:12 -0200)]
drm/i915: fix wrong PLL debug messages.

LPT does have PCH refclk, but it's different form the IBX/CPT/PPT one
and doesn't use the same structs. It is wrong to have a message saying
that "LPT does not has PCH refclk" (sic). While at it, signal that we
only want this function on IBX/CPT/PPT by renaming it and adding a
WARN.

On HSW we also print "0 shared PLLs initialized", but we *do* have
shared PLLs on HSW (LCPLL, WRPLL, SPLL) and we *do* initialize them.
We just don't use "struct intel_shared_dpll". So remove the debug
message.

In the future we may want to rename all that "intel shared pll" code
to "ibx shared pll", but I'll leave this to another patch.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: don't try to check IPS state on BDW v2
Jesse Barnes [Tue, 7 Jan 2014 21:30:45 +0000 (13:30 -0800)]
drm/i915/bdw: don't try to check IPS state on BDW v2

According to Art, we don't have a way to read back the state reliably at
runtime, through the control reg or the mailbox, at least not without risking
disabling it again.  So drop the readout and checking on BDW.

v2: drop TODO comment (Paulo)
    move POSTING_READ of control reg under HSW branch in disable (Paulo)
    always report IPS as enabled on BDW (Paulo)

References: https://bugs.freedesktop.org/show_bug.cgi?id=71906
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: use crtc_htotal when calculating ilk watermarks
Jesse Barnes [Fri, 20 Dec 2013 21:08:00 +0000 (13:08 -0800)]
drm/i915: use crtc_htotal when calculating ilk watermarks

This was introduced in:

commit 7c4a395ff8f441acb7876281c6777624e6410349
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Wed Oct 9 19:17:56 2013 +0300

    drm/i915: Don't re-compute pipe watermarks except for the affected pipe

and I missed fixing it in:

commit fec8cba306f974f3a4491176994de5d821273643
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Wed Nov 27 11:10:26 2013 -0800

    drm/i915: use crtc_htotal in watermark calculations to match fastboot v2

It's needed for ILK+ platforms to fastboot without crashing on a divide
by 0 after a DPMS on action.

Note: Ville mentioned in his review that this confusion seems to go
down to the original introduction of this code in

commit 801bcfffbb0721d7131e930f9a46103e539c43a4
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri May 31 10:08:35 2013 -0300

    drm/i915: properly set HSW WM_PIPE registers

So it seems to have been missed both in the fastboot patch and in the
3d mode suppport (where only crtc_htotal reflects the real pipe
width).

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Add note based on Ville's review.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix fastboot pfit disable hack to update pipe w/h
Jesse Barnes [Thu, 19 Dec 2013 18:48:01 +0000 (10:48 -0800)]
drm/i915: fix fastboot pfit disable hack to update pipe w/h

When fastbooting, we read out the pipe timings early on, and then in a
panel fitted config, disable the fitter later.  But we weren't updating
the pipe src h/w, which meant the mouse cursor was clipped to the
pfitted size rather than the native size set later.  Fix that up so the
cursor is visible in the new mode.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: check modeset state after a pipe_set_base if using fastboot
Jesse Barnes [Tue, 7 Jan 2014 21:50:49 +0000 (13:50 -0800)]
drm/i915: check modeset state after a pipe_set_base if using fastboot

Otherwise we won't check the state until the next DPMS transition, which
may never happen.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Simplify watermark/init_clock_gating setup
Ville Syrjälä [Tue, 7 Jan 2014 14:14:10 +0000 (16:14 +0200)]
drm/i915: Simplify watermark/init_clock_gating setup

Avoid duplicating the same piece of code several times by separating
the watemark vfunc setup from the init_clock_gating vfunc setup on PCH
platforms.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Enable watermarks for BDW
Ville Syrjälä [Tue, 7 Jan 2014 14:14:09 +0000 (16:14 +0200)]
drm/i915: Enable watermarks for BDW

We forgot to intialize the watermark vfuncs for BDW, and hence the
watermarks were never updated.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix watermark code for BDW
Ville Syrjälä [Tue, 7 Jan 2014 14:14:08 +0000 (16:14 +0200)]
drm/i915: Fix watermark code for BDW

Looks like I forgot to update the ILK/SNB/IVB watermark patches to deal
with BDW. Add the relevant BDW checks to make sure we take the HSW
codepaths on BDW as well.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: avoid unclaimed registers when capturing the error state
Paulo Zanoni [Fri, 20 Dec 2013 17:09:41 +0000 (15:09 -0200)]
drm/i915: avoid unclaimed registers when capturing the error state

We're iterating over the CPU transcoders, so check for the correct
power domain.

This fixes many "unclaimed register" error messages.

This can be reproduced by the IGT test mentioned below, but we still
get a FAIL when we run it.

Testcase: igt/kms_lip/flip-vs-panning-vs-hang
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Flush outstanding requests before allocating new seqno
Chris Wilson [Thu, 2 Jan 2014 14:32:35 +0000 (14:32 +0000)]
drm/i915: Flush outstanding requests before allocating new seqno

In very rare cases (such as a memory failure stress test) it is possible
to fill the entire ring without emitting a request. Under this
circumstance, the outstanding request is flushed and waited upon. After
space on the ring is cleared, we return to emitting the new command -
except that we just cleared the seqno allocated for this operation and
trigger the sanity check that a request is only ever emitted with a
valid seqno. The fix is to rearrange the code to make sure the
allocation of the seqno for this operation is after any required flushes
of outstanding operations.

The bug exists since the preallocation was introduced in
commit 9d7730914f4cd496e356acfab95b41075aa8eae8
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Tue Nov 27 16:22:52 2012 +0000

    drm/i915: Preallocate next seqno before touching the ring

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@vger.kernel.org
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove duplicate MODULE_LICENSE definition
Jani Nikula [Tue, 7 Jan 2014 09:15:34 +0000 (11:15 +0200)]
drm/i915: remove duplicate MODULE_LICENSE definition

Multiple definitions show up multiple times in modinfo output.

There's already an identical one in i915_drv.c along with other MODULE_*
definitions, so drop the lone one in intel_fbdev.c.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Avoid dereference past end of page array in gen8_ppgtt_insert_entries()
Chris Wilson [Tue, 31 Dec 2013 15:50:31 +0000 (15:50 +0000)]
drm/i915: Avoid dereference past end of page array in gen8_ppgtt_insert_entries()

The bug from gen6_ppgtt_insert_entries() was replicated into
gen8_ppgtt_insert_entries(). This applies the fix for the OOPS from the
previous patch to the gen8 routine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries()
Chris Wilson [Tue, 31 Dec 2013 15:50:30 +0000 (15:50 +0000)]
drm/i915: Avoid dereference past end of page array in gen6_ppgtt_insert_entries()

[   89.237347] BUG: unable to handle kernel paging request at ffff880096326000
[   89.237369] IP: [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170
[   89.237382] PGD 2272067 PUD 25df0e067 PMD 25de5c067 PTE 8000000096326060
[   89.237394] Oops: 0000 [#1] SMP DEBUG_PAGEALLOC
[   89.237404] CPU: 1 PID: 1981 Comm: gem_concurrent_ Not tainted 3.13.0-rc4+ #639
[   89.237411] Hardware name: Intel Corporation 2012 Client Platform/Emerald Lake 2, BIOS ACRVMBY1.86C.0078.P00.1201161002 01/16/2012
[   89.237420] task: ffff88024c038030 ti: ffff88024b130000 task.ti: ffff88024b130000
[   89.237425] RIP: 0010:[<ffffffff81347227>]  [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170
[   89.237435] RSP: 0018:ffff88024b131ae0  EFLAGS: 00010286
[   89.237440] RAX: ffff880096325000 RBX: 0000000000000400 RCX: 0000000000001000
[   89.237445] RDX: 0000000000000200 RSI: 0000000000000001 RDI: 0000000000000010
[   89.237451] RBP: ffff88024b131b30 R08: ffff88024cc3aef0 R09: 0000000000000000
[   89.237456] R10: 0000000000000000 R11: 0000000000000000 R12: ffff88024cc3ae00
[   89.237462] R13: ffff88024a578000 R14: 0000000000000001 R15: ffff88024a578ffc
[   89.237469] FS:  00007ff5475d8900(0000) GS:ffff88025d020000(0000) knlGS:0000000000000000
[   89.237475] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   89.237480] CR2: ffff880096326000 CR3: 000000024d531000 CR4: 00000000001407e0
[   89.237485] Stack:
[   89.237488]  ffff880000000000 0000020000000000 ffff88024b23f2c0 0000000100000000
[   89.237499]  0000000000000001 000000000007ffff ffff8801e7bf5ac0 ffff8801e7bf5ac0
[   89.237510]  ffff88024cc3ae00 ffff880248a2ee40 ffff88024b131b58 ffffffff813455ed
[   89.237521] Call Trace:
[   89.237528]  [<ffffffff813455ed>] ppgtt_bind_vma+0x3d/0x60
[   89.237534]  [<ffffffff8133d8dc>] i915_gem_object_pin+0x55c/0x6a0
[   89.237541]  [<ffffffff8134275b>] i915_gem_execbuffer_reserve_vma.isra.14+0x5b/0x110
[   89.237548]  [<ffffffff81342a88>] i915_gem_execbuffer_reserve+0x278/0x2c0
[   89.237555]  [<ffffffff81343d29>] i915_gem_do_execbuffer.isra.22+0x699/0x1250
[   89.237562]  [<ffffffff81344d91>] ? i915_gem_execbuffer2+0x51/0x290
[   89.237569]  [<ffffffff81344de6>] i915_gem_execbuffer2+0xa6/0x290
[   89.237575]  [<ffffffff813014f2>] drm_ioctl+0x4d2/0x610
[   89.237582]  [<ffffffff81080bf1>] ? cpuacct_account_field+0xa1/0xc0
[   89.237588]  [<ffffffff81080b55>] ? cpuacct_account_field+0x5/0xc0
[   89.237597]  [<ffffffff811371c0>] do_vfs_ioctl+0x300/0x520
[   89.237603]  [<ffffffff810757a1>] ? vtime_account_user+0x91/0xa0
[   89.237610]  [<ffffffff810e40eb>] ?  context_tracking_user_exit+0x9b/0xe0
[   89.237617]  [<ffffffff81083d7d>] ? trace_hardirqs_on+0xd/0x10
[   89.237623]  [<ffffffff81137425>] SyS_ioctl+0x45/0x80
[   89.237630]  [<ffffffff815afffa>] tracesys+0xd4/0xd9
[   89.237634] Code: 5b 41 5c 41 5d 41 5e 41 5f 5d c3 66 0f 1f 84 00 00 00 00 00 83 45 bc 01 49 8b 84 24 78 01 00 00 65 ff 0c 25 e0 b8 00 00 8b 55 bc <4c> 8b 2c d0 65 ff 04 25 e0 b8 00 00 49 8b 45 00 48 c1 e8 2d 48
[   89.237741] RIP  [<ffffffff81347227>] gen6_ppgtt_insert_entries+0x117/0x170
[   89.237749]  RSP <ffff88024b131ae0>
[   89.237753] CR2: ffff880096326000
[   89.237758] ---[ end trace 27416ba8b18d496c ]---

This bug dates back to the original introduction of the
gen6_ppgtt_insert_entries()

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Dropped cc: stable since without full ppgtt there's no way
we'll access the last page directory with this function since that
range is occupied (only in the allocator) with the ppgtt pdes. Without
aliasing we can start to use that range and blow up.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Mention when we enable the Ironlake iommu workarounds
Chris Wilson [Mon, 30 Dec 2013 12:16:15 +0000 (12:16 +0000)]
drm/i915: Mention when we enable the Ironlake iommu workarounds

The iommu and gfx on Ironlake do not like each other and require a
big hammer to prevent hard machine hangs. In

commit 5c0422878fcdc279ae9a8e8b66972a15b5efb67f
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Mon Oct 17 15:51:55 2011 -0700

    drm/i915: ILK + VT-d workaround

we added the workaround, but never emitted any debug message that it was
active. Doing so should help identify known performance regressions.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix refcount leak and possible NULL pointerdereference.
Tetsuo Handa [Fri, 3 Jan 2014 11:42:18 +0000 (20:42 +0900)]
drm/i915: Fix refcount leak and possible NULL pointerdereference.

Since get_pid_task() grabs a reference on the task_struct, we have to drop the
refcount after reading that task's comm name. Use pid_task() with RCU instead.

Also, avoid directly reading like pid_task()->comm because
pid_task() will return NULL if the task have already exit()ed.

This patch fixes both problems.

Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: grab a pages pin count for preallocate stolen
Daniel Vetter [Tue, 17 Dec 2013 22:42:11 +0000 (23:42 +0100)]
drm/i915: grab a pages pin count for preallocate stolen

But only when we indeed set up a gtt mapping. We need this since the
vma also holds a pages_pin_count, on top of the unconditional
pages_pin_count we grab for all stolen objects (to avoid swap-out).

This should avoid a pages_pin_count underrun when cleaning up
framebuffers objects taken over from the BIOS.

Chris mentioned in his review that this bug even predates the vma
conversion.

Reported-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: s/haswell_update_wm/ilk_update_wm/
Imre Deak [Tue, 17 Dec 2013 12:46:36 +0000 (14:46 +0200)]
drm/i915: s/haswell_update_wm/ilk_update_wm/

We use this hook starting from ILK onwards, so change the prefix
accordingly. Also rename functions/struct names used from
haswell_update_wm that are relevant to ILK already.

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove unused WM defines
Imre Deak [Tue, 17 Dec 2013 12:46:35 +0000 (14:46 +0200)]
drm/i915: remove unused WM defines

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: simplify platform specific code in hsw_write_wm_values
Imre Deak [Tue, 17 Dec 2013 12:46:34 +0000 (14:46 +0200)]
drm/i915: simplify platform specific code in hsw_write_wm_values

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoRevert "drm/i915/sprite: Always enable the scaler on IronLake"
Ville Syrjälä [Thu, 5 Dec 2013 13:51:31 +0000 (15:51 +0200)]
Revert "drm/i915/sprite: Always enable the scaler on IronLake"

Apparently always enabling the sprite scaler magically made
sprites work on ILK in the past.

I think the real reason for the failure was missing sprite
watermark programming, and enabling the scaler effectively
disabled LP1+ watermarks, which was enough to keep things going.
Or it might be that the hardware more or less ignores watermarks
for scaled sprites since things seem to work even if I leave
sprite watermarks at 0 and disable all other planes except the
sprite.

In any case, we left the scaler always on but then failed to
check whether we might be exceeding the scaler's source size
limits. That caused the sprite to fail when a sufficiently
large unscaled image was being displayed.

Now that we're getting proper watermark programming for ILK, we
can keep the scaler disabled unless we need to do actual scaling.

This reverts commit 8aaa81a166d80ac9bf2813984e5b4c2503d0fe08.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Avoid underruns when disabling sprites
Ville Syrjälä [Thu, 5 Dec 2013 13:51:41 +0000 (15:51 +0200)]
drm/i915: Avoid underruns when disabling sprites

As the watermark registers aren't double bufferd, clearing the
watermarks immediately after writing the sprite registers can be
hazardous.

Until we have something better, add a wait for vblank between the
two steps to make sure the sprite no longer needs the watermark
levels before we clear them.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't disable primary when color keying is used
Ville Syrjälä [Thu, 5 Dec 2013 13:51:40 +0000 (15:51 +0200)]
drm/i915: Don't disable primary when color keying is used

When color keying is used, the primary may not be invisible even though
the sprite fully covers it. So check for color keying before deciding to
disable the primary plane.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Try to fix the messy IVB sprite scaling workaround
Ville Syrjälä [Thu, 5 Dec 2013 13:51:39 +0000 (15:51 +0200)]
drm/i915: Try to fix the messy IVB sprite scaling workaround

We now have a very clear method of disabling LP1+ wartermarks,
and we can actually detect if we actually did disable them, or
if they were already disabled. Use that to clean up the
WaCxSRDisabledForSpriteScaling:ivb handling.

I was hoping to apply the workaround in a way that wouldn't
require a blocking wait, but sadly IVB really does appear to
require LP1+ watermarks to be off for an entire frame before
enabling sprite scaling. Simply disabling LP1+ watermarks
during the previous frame is not enough, no matter how early
in the frame we do it :(

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move ILK/SNB/IVB over to the HSW WM code
Ville Syrjälä [Thu, 5 Dec 2013 13:51:38 +0000 (15:51 +0200)]
drm/i915: Move ILK/SNB/IVB over to the HSW WM code

The new HSW watermark code can now handle ILK/SNB/IVB as well, so
switch them over. Kill the old code.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Disable LP1+ watermarks safely in init
Ville Syrjälä [Thu, 5 Dec 2013 13:51:37 +0000 (15:51 +0200)]
drm/i915: Disable LP1+ watermarks safely in init

ILK doesn't like if we just write the LP1+ watermarks registers with 0.
We need to just disable the watermarks by clearing the enable bit. Use
that method also when disabling LP1+ watermarks in init_clock_gating.

It looks like disabling the sprite LP1 watermarks can cause underruns
even if we just toggle the WM1S_LP_EN bit. So treat that bit like the
actual watermark numbers and avoid setting it to 0 immediately.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Linetime watermarks are a HSW feature
Ville Syrjälä [Thu, 5 Dec 2013 13:51:36 +0000 (15:51 +0200)]
drm/i915: Linetime watermarks are a HSW feature

Linetime watermarks don't exist on ILK/SNB/IVB, so don't compute them
except on HSW.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Disable FBC WM on ILK, and disable LP2+ when FBC is enabled
Ville Syrjälä [Thu, 5 Dec 2013 13:51:35 +0000 (15:51 +0200)]
drm/i915: Disable FBC WM on ILK, and disable LP2+ when FBC is enabled

ILK has a bunch of issues with FBC. First of all, BSpec tells us that
FBC WM should never be enabled. Secondly when FBC is enabled
with FBC WM disabled, LP2+ watermarks must be disabled.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't merge LP1+ watermarks on ILK/SNB/IVB when multiple pipes are enabled
Ville Syrjälä [Thu, 5 Dec 2013 13:51:34 +0000 (15:51 +0200)]
drm/i915: Don't merge LP1+ watermarks on ILK/SNB/IVB when multiple pipes are enabled

Multi-pipe LP1+ watermarks are a HSW+ feature, so let's not do it on
earlier generations.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix LP1+ watermark disabling ILK
Ville Syrjälä [Thu, 5 Dec 2013 13:51:33 +0000 (15:51 +0200)]
drm/i915: Fix LP1+ watermark disabling ILK

On ILK disabling LP1+ watermarks must be done carefully to avoid
underruns. If we just write 0 to the register in the middle of the scan
cycle we often get an underrun. So instead we have to leave the actual
watermark levels in the register intact, and just toggle the enable bit.

Presumably the hardware takes a while to get out of low power mode, and
so the watermark level need to stay valid until that time.

We also have to be careful with the WM1S_LP_EN bit. It seems the
hardware more or less treats it like the actual watermarks numbers, and
so we must not toggle it too soon. Just leave it alone when disabling
the LP1+ watermarks.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix LP1 sprite watermarks for ILK/SNB
Ville Syrjälä [Thu, 5 Dec 2013 13:51:32 +0000 (15:51 +0200)]
drm/i915: Fix LP1 sprite watermarks for ILK/SNB

ILK/SNB don't have LP2+ watermarks for sprites. Also the LP1 sprite
watermark register has its own enable bit. Take these differences
into account when programming the LP1+ registers.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Avoid computing invalid WM levels when sprites/scaling is enabled
Ville Syrjälä [Thu, 5 Dec 2013 13:51:30 +0000 (15:51 +0200)]
drm/i915: Avoid computing invalid WM levels when sprites/scaling is enabled

On ILK/SNB only LP0/1 watermarks can be enabled when sprites are
enabled, and on ILK/SNB/IVB sprite scaling is limited to LP0 only.

So we can avoid computing the extra levels we're never going to use.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add ILK/SNB/IVB WM latency field support
Ville Syrjälä [Thu, 5 Dec 2013 13:51:29 +0000 (15:51 +0200)]
drm/i915: Add ILK/SNB/IVB WM latency field support

Add a new function  ilk_wm_lp_latency() which will tell us what to write
into the WM_LPx register latency field. HSW is different from erlier
gens in this regard.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add IVB DDB partitioning control
Ville Syrjälä [Thu, 5 Dec 2013 13:51:28 +0000 (15:51 +0200)]
drm/i915: Add IVB DDB partitioning control

On IVB the display data buffer partitioning control lives in the
DISP_ARB_CTL2 register. Add the relevant defines/code for it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: kick firmware fbs even when i915 fbdev is disabled
Daniel Vetter [Tue, 17 Dec 2013 09:00:54 +0000 (10:00 +0100)]
drm/i915: kick firmware fbs even when i915 fbdev is disabled

Otherwise we don't kick out firmware framebuffers like vesafb and
efifb when CONFIG_DRM_I915_FBDEV=n but CONFIG_FB=y.

There's still the pesky issue with vgacon which we should somehow
replace with the dummy console at least. We have a similar issue at
module un/reload, since vgacon state is terminally botched after
i915.ko has loaded in modeset mode. But this gets us a step further at
least.

v2: Use IS_ENABLED - I always get this wrong for tristates. Spotted by
Jani.

Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use IS_VALLEYVIEW() to test the is_valleyview flag
Damien Lespiau [Thu, 12 Dec 2013 14:36:36 +0000 (14:36 +0000)]
drm/i915: Use IS_VALLEYVIEW() to test the is_valleyview flag

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use symbolic names for booleans in i915_semaphore_is_enabled
Daniel Vetter [Tue, 17 Dec 2013 08:56:53 +0000 (09:56 +0100)]
drm/i915: Use symbolic names for booleans in i915_semaphore_is_enabled

Noticed while reviewing a patch and couldn't resist the OCD.

Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't emit mbox updates without semaphores
Ben Widawsky [Tue, 17 Dec 2013 04:50:38 +0000 (20:50 -0800)]
drm/i915: Don't emit mbox updates without semaphores

Aside from the fact that it leaves confusing dumps on error capture, it
is entirely unnecessary, and potentially harmful in cases like BDW,
where the instruction has changed.

In reality (seemingly), this will have no behavioral impact.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Reorder/respace MI instruction definition
Ben Widawsky [Tue, 17 Dec 2013 04:50:37 +0000 (20:50 -0800)]
drm/i915: Reorder/respace MI instruction definition

A few command were out of numerical order and had different spacing. Put
them back in numerical order, with proper spacing.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/vlv: split DPIO init and reset
Jesse Barnes [Tue, 17 Dec 2013 00:34:24 +0000 (16:34 -0800)]
drm/i915/vlv: split DPIO init and reset

We only need to init the reg offset for DPIO once, but we need to reset
DPIO at resume time and at init time.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/vlv: add early DPIO init v3
Jesse Barnes [Tue, 17 Dec 2013 00:34:23 +0000 (16:34 -0800)]
drm/i915/vlv: add early DPIO init v3

Just add an early init since we may need to access DPIO regs early on.
The init call in modeset_init_hw is also needed for the resume case,
when we need to reset DPIO to keep things happy.

v2: split reset and reg init
v3: split patches (Daniel)

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: only build i915_debugfs.c when CONFIG_DEBUG_FS is enabled
Jani Nikula [Mon, 16 Dec 2013 11:10:36 +0000 (13:10 +0200)]
drm/i915: only build i915_debugfs.c when CONFIG_DEBUG_FS is enabled

The whole file is wrapped around in #if defined(CONFIG_DEBUG_FS) anyway,
so skip the file at the build level already.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix return value check of debugfs_create_file()
Wei Yongjun [Mon, 16 Dec 2013 06:13:25 +0000 (14:13 +0800)]
drm/i915: fix return value check of debugfs_create_file()

In case of error, the function debugfs_create_file() returns NULL
pointer not ERR_PTR() if debugfs is enabled. The IS_ERR() test in
the return value check should be replaced with NULL test.

Signed-off-by: Wei Yongjun <yongjun_wei@trendmicro.com.cn>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: parse backlight modulation frequency from the BIOS VBT
Jani Nikula [Sat, 14 Dec 2013 22:38:29 +0000 (20:38 -0200)]
drm/i915: parse backlight modulation frequency from the BIOS VBT

We don't actually do anything with the information yet, but parse and
log what's in the VBT.

Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add runtime PM support on Haswell
Paulo Zanoni [Fri, 13 Dec 2013 17:22:32 +0000 (15:22 -0200)]
drm/i915: add runtime PM support on Haswell

Now that all the infrastructure is in place and all the tests from
pm_pc8 pass, we can finally enable the feature.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: release the GTT mmaps when going into D3
Paulo Zanoni [Fri, 13 Dec 2013 17:22:31 +0000 (15:22 -0200)]
drm/i915: release the GTT mmaps when going into D3

So we'll get a fault when someone tries to access the mmap, then we'll
wake up from D3.

v2: - Rebase
v3: - Use gtt active/inactive

Testcase: igt/pm_pc8/gem-mmap-gtt
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[danvet: Add comment + WARN as discussed with Paulo on irc.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: cancel the hangcheck before runtime suspend
Paulo Zanoni [Fri, 13 Dec 2013 17:22:30 +0000 (15:22 -0200)]
drm/i915: cancel the hangcheck before runtime suspend

The hangcheck function requires the hardware to be working, and if
we're suspending we're going to put the HW in D3 state.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: Implement ff workarounds
Ben Widawsky [Thu, 12 Dec 2013 23:28:04 +0000 (15:28 -0800)]
drm/i915/bdw: Implement ff workarounds

WaVSRefCountFullforceMissDisable and
WaDSRefCountFullforceMissDisable

VS is a carry-over from HSW, and DS is likely not used by anyone yet.

Cc: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Line of 106 chars is too long. Really.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: Force all Data Cache Data Port access to be Non-Coherent
Ben Widawsky [Fri, 13 Dec 2013 01:26:03 +0000 (17:26 -0800)]
drm/i915/bdw: Force all Data Cache Data Port access to be Non-Coherent

I stumbled on to some unimplemented errata. To be honest, I am not
really sure of the impact, just that the docs say to do.

No w/a name for this one.

v2: v1 was a stale thing which should have never seen the light of day.
(Haihao)

Cc: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: Don't use forcewake needlessly
Ville Syrjälä [Fri, 13 Dec 2013 11:26:20 +0000 (13:26 +0200)]
drm/i915/bdw: Don't use forcewake needlessly

Not all registers need forcewake even if they're not shadowed.
Add the missing check to gen8_writeX() to avoid needless forcewake
usage when writing eg. display registers.

v2: Use straight up <0x40000 check instead of NEEDS_FORCE_WAKE()

Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Clear out old GT FIFO errors in intel_uncore_early_sanitize()
Ville Syrjälä [Tue, 3 Dec 2013 09:30:09 +0000 (11:30 +0200)]
drm/i915: Clear out old GT FIFO errors in intel_uncore_early_sanitize()

The BIOS or someone else might have done something bad and there
might be old GT FIFO erros reported in GTFIFODBG. Clear those out
in intel_uncore_early_sanitize() to make sure we don't mistake them
for our problems.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: dont call irq_put when irq test is on
Mika Kuoppala [Thu, 12 Dec 2013 15:54:42 +0000 (17:54 +0200)]
drm/i915: dont call irq_put when irq test is on

If test is running, irq_get was not called so we should gain
balance by not doing irq_put

"So the rule is: if you access unlocked values, you use ACCESS_ONCE().
You don't say "but it can't matter". Because you simply don't know."
-- Linus

v2: use local variable so it can't change during test (Chris)

v3: update commit msg and use ACCESS_ONCE (Ville)

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Rework the FBC interval/stall stuff a bit
Ville Syrjälä [Thu, 12 Dec 2013 15:27:40 +0000 (17:27 +0200)]
drm/i915: Rework the FBC interval/stall stuff a bit

Don't touch DPFC_RECOMP_CTL on FBC2, use RMW to update
the FBC_CONTROL on FBC1 to make it easier for people to
experiment with different numbers. Also fix the interval
mask for FBC1.

v2: Rebased

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Enable FBC for all mobile gen2 and gen3 platforms
Ville Syrjälä [Thu, 28 Nov 2013 15:30:02 +0000 (17:30 +0200)]
drm/i915: Enable FBC for all mobile gen2 and gen3 platforms

All mobile gen2 and gen3 chipsets should have FBC1, and the code
should now handle them all. So just set has_fbc=true for all such
chipsets.

Note that fbc is still disabled by default for now.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: FBC_CONTROL2 is gen4 only
Ville Syrjälä [Thu, 28 Nov 2013 15:29:57 +0000 (17:29 +0200)]
drm/i915: FBC_CONTROL2 is gen4 only

Gen2 and gen3 don't have the FBC_CONTROL2 register, so don't
touch it.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Gen2 FBC1 CFB pitch wants 32B units
Ville Syrjälä [Thu, 28 Nov 2013 15:29:56 +0000 (17:29 +0200)]
drm/i915: Gen2 FBC1 CFB pitch wants 32B units

On gen2 the compressed frame buffer pitch is specified in 32B units
rather than the 64B units used on gen3+.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: split intel_ddi_pll_mode_set in 2 pieces
Paulo Zanoni [Mon, 25 Nov 2013 17:27:08 +0000 (15:27 -0200)]
drm/i915: split intel_ddi_pll_mode_set in 2 pieces

The first piece, intel_ddi_pll_select, finds a PLL and assigns it to
the CRTC, but doesn't write any register. It can also fail in case it
doesn't find a PLL.

The second piece, intel_ddi_pll_enable, uses the information stored by
intel_ddi_pll_select to actually enable the PLL by writing to its
register. This function can't fail. We also have some refcount sanity
checks here.

The idea is that one day we'll remove all the functions that touch
registers from haswell_crtc_mode_set to haswell_crtc_enable, so we'll
call intel_ddi_pll_select at haswell_crtc_mode_set and then call
intel_ddi_pll_enable at haswell_crtc_enable. Since I'm already
touching this code, let's take care of this particular split today.

v2: - Clock on the debug message is in KHz
    - Add missing POSTING_READ

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Bikeshed comments.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix timeout with missed interrupts in __wait_seqno
Mika Kuoppala [Tue, 10 Dec 2013 15:02:43 +0000 (17:02 +0200)]
drm/i915: Fix timeout with missed interrupts in __wait_seqno

Commit 094f9a54e355 ("drm/i915: Fix __wait_seqno to use true infinite
timeouts") added support for __wait_seqno to detect missing interrupts and
go around them by polling. As there is also timeout detection in
__wait_seqno, the polling and timeout detection were done with the same
timer.

When there has been missed interrupts and polling is needed, the timer is
set to trigger in (now + 1) jiffies in future, instead of the caller
specified timeout.

Now when io_schedule() returns, we calculate the jiffies left to timeout
using the timer expiration value. As the current jiffies is now bound to be
always equal or greater than the expiration value, the timeout_jiffies will
become zero or negative and we return -ETIME to caller even tho the
timeout was never reached.

Fix this by decoupling timeout calculation from timer expiration.

v2: Commit message with some sense in it (Chris Wilson)

v3: add parenthesis on timeout_expire calculation

v4: don't read jiffies without timeout (Chris Wilson)

Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: touch VGA MSR after we enable the power well
Paulo Zanoni [Wed, 11 Dec 2013 20:50:10 +0000 (18:50 -0200)]
drm/i915: touch VGA MSR after we enable the power well

Fixes regression introduced by:
    commit bf51d5e2cda5d36d98e4b46ac7fca9461e512c41
    Author: Paulo Zanoni <paulo.r.zanoni at intel.com>
    Date:   Wed Jul 3 17:12:13 2013 -0300
        drm/i915: switch disable_power_well default value to 1

The bug I'm seeing can be reproduced with:
  - Have vgacon configured/enabled
  - Make sure the power well gets disabled, then enabled. You can
    check this by seeing the messages print by hsw_set_power_well
  - Stop your display manager
  - echo 0 > /sys/class/vtconsole/vtcon1/bind

I can easily reproduce this by blacklising snd_hda_intel and booting
with eDP+HDMI.

If you do this and then look at dmesg, you'll see we're printing
infinite "Unclaimed register" messages. This is happening because
we're stuck on an infinite loop inside console_unlock(), which is
calling many functions from vgacon.c. And the code that's triggering
the error messages is from vgacon_set_cursor_size().

After we re-enable the power well, every time we read/write the VGA
address 0x3d5 we get an "unclaimed register" interrupt (ERR_INT) and
print error messages. If we write anything to the VGA MSR register (it
doesn't really matter which value you write to bit 0), any
reads/writes to 0x3d5 _don't_ trigger the "unclaimed register" errors
anymore (even if MSR bit 0 is zero). So what happens with the current
code is that when we unbind i915 and bind vgacon, we call
console_unlock(). Function console_unlock() is responsible for
printing any messages that were supposed to be print when the console
was locked, so it calls the TTY layer, which calls the console layer,
which calls vgacon to print the messages. At this point, vgacon
eventually calls vgacon_set_cursor_size(), which touches 0x3d5, which
triggers unclaimed register interrupts. The problem is that when we
get these interrupts, we print the error messages, so we add more work
to console_unlock(), which will try to print it again, and then call
vgacon again, trigger a new interrupt, which will put more stuff to
the buffer, and then we'll be stuck at console_unlock() forever.

If you patch intel_uncore.c to not print anything when we detect
unclaimed registers, we won't get into the console_unlock() infinite
loop and the driver unbind will work just fine. We will still be
getting interrupts every time vgacon touches those registers, but we
will survive. This is a valid experiment, but IMHO it's not the real
fix: if we don't print any error messages we will still keep getting
the interrupts, and if we disable ERR_INT we won't get the interrupt
anymore, but we will also stop getting all the other error interrupts.

I talked about this problem with the HW engineer and his
recommendation is "So don't do any VGA I/O or memory access while the
power well is disabled, and make to re-program MSR after enabling the
power well and before using VGA I/O or memory accesses.".

Notice that this is just a partial fix to fd.o #67813. This fixes the
case where the power well is already enabled when we unbind, not when
it's disabled when we unbind.

V2: - Rebase (first version was sent in September).
V3: - Complete rewrite of the same fix: smaller implementation,
      improved commit message.

Testcase: igt/drv_module_reload
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=67813
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: extract hsw_power_well_post_{enable, disable}
Paulo Zanoni [Wed, 11 Dec 2013 20:50:09 +0000 (18:50 -0200)]
drm/i915: extract hsw_power_well_post_{enable, disable}

I want to add more code to the post_enable function.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: remove i915_disable_vga_mem declaration
Paulo Zanoni [Wed, 11 Dec 2013 20:50:08 +0000 (18:50 -0200)]
drm/i915: remove i915_disable_vga_mem declaration

It was supposed to have been killed on the same commit that killed the
function, e1264ebe9ff48e1b3e1dd11805eec9f5b143ab7c, but I guess the
intel_drv.h reorganization accidentally brought it back.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Parametrize the dphy and other spec specific parameters
Shobhit Kumar [Tue, 10 Dec 2013 06:45:00 +0000 (12:15 +0530)]
drm/i915: Parametrize the dphy and other spec specific parameters

The values of these parameters will be different for differnet panel
based on dsi rate, lane count, etc. Remove the hardcodings and make
these as parameters whch will be initialized in panel specific
sub-encoder implementaion.

This will also form groundwork for planned generic panel sub-encoder
implemntation based on VBT design enhancments to support multiple panels

v2: Mask away the port_bits before use

Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove redundant DSI PLL enabling
Shobhit Kumar [Tue, 10 Dec 2013 06:44:59 +0000 (12:14 +0530)]
drm/i915: Remove redundant DSI PLL enabling

DSI PLL will get configured during crtc_enable using ->pre_pll_enable
and no need to do in ->mode_set

Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Reorganize the DSI enable/disable sequence
Shobhit Kumar [Wed, 11 Dec 2013 12:22:05 +0000 (17:52 +0530)]
drm/i915: Reorganize the DSI enable/disable sequence

Basically ULPS handling during enable/disable has been moved to
pre_enable and post_disable phases. PLL and panel power disable
also has been moved to post_disable phase. The ULPS entry/exit
sequneces as suggested by HW team is as follows -

During enable time -
set DEVICE_READY --> Clear DEVICE_READY --> set DEVICE_READY

And during disable time to flush all FIFOs -
set ENTER_SLEEP --> EXIT_SLEEP --> ENTER_SLEEP

Also during disbale sequnece sub-encoder disable is moved to the end
after port is disabled.

v2: Based on comments from Ville
    - Detailed epxlaination in the commit messgae
    - Moved parameter changes out into another patch
    - Backlight enabling will be a new patch

v3: Updated as per Jani's comments
    - Removed the I915_WRITE_BITS as it is not needed
    - Moved panel_reset and send_otp_cmds hooks to dsi_pre_enable
    - Moved disable_panel_power hook to dsi_post_disable
    - Replace hardcoding with AFE_LATCHOUT

v4: Make intel_dsi_device_ready and intel_dsi_clear_device_ready static

Signed-off-by: Yogesh Mohan Marimuthu <yogesh.mohan.marimuthu@intel.com>
Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Try harder to get best m, n, p values with minimal error
Shobhit Kumar [Tue, 10 Dec 2013 06:44:57 +0000 (12:14 +0530)]
drm/i915: Try harder to get best m, n, p values with minimal error

Basically check for both +ive and -ive deviation from target clock and
pick the one with minimal error. If we get a direct match, break from
loop to acheive some optimization.

v2: Use signed variable for target and calculated dsi clock values

Signed-off-by: Vijayakumar Balakrishnan <vijayakumar.balakrishnan@intel.com>
Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Compute dsi_clk from pixel clock
Shobhit Kumar [Tue, 10 Dec 2013 06:44:56 +0000 (12:14 +0530)]
drm/i915: Compute dsi_clk from pixel clock

Pixel clock based calculation is recommended in the MIPI host controller
documentation

v2: Based on review comments from Jani and Ville
    - Use dsi_clk in KHz rather than converting in Hz and back to MHz
    - RR formula is retained though not used but return dsi_clk in KHz now
    - Moved the m-n-p changes into a separate patch
    - Removed the parameter check for intel_dsi->dsi_clock_freq. This will be
      bought back in if needed when appropriate panel drivers are done

v3: Removed the unused mnp calculation from static table

Signed-off-by: Vijayakumar Balakrishnan <vijayakumar.balakrishnan@intel.com>
Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use FLISDSI interface for band gap reset
Shobhit Kumar [Tue, 10 Dec 2013 06:44:55 +0000 (12:14 +0530)]
drm/i915: Use FLISDSI interface for band gap reset

v2: Rebased on latest code

Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Signed-off-by: Yogesh Mohan Marimuthu <yogesh.mohan.marimuthu@intel.com>
Reviewed-by: Jani Nikula<jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add more dev ops for MIPI sub encoder
Shobhit Kumar [Tue, 10 Dec 2013 06:44:54 +0000 (12:14 +0530)]
drm/i915: Add more dev ops for MIPI sub encoder

Some panels require one time programming if they do not contain their
own eeprom for basic register initialization. The sequence is

Panel Reset --> Send OTP --> Enable Pixel Stream --> Enable the panel

v2: Based on review comments from Jani and Ville
    - Updated the commit message with more details
    - Move the new parameters out of this patch

Signed-off-by: Yogesh Mohan Marimuthu <yogesh.mohan.marimuthu@intel.com>
Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Decrease gen2 vco frequency minimum to 908 MHz
Ville Syrjälä [Mon, 9 Dec 2013 16:54:17 +0000 (18:54 +0200)]
drm/i915: Decrease gen2 vco frequency minimum to 908 MHz

On my 855 machine the BIOS uses the following DPLL settings:
DPLL 0x90016000
FP0 = 0x61207
FP1 = 0x21207

With the 66MHz SSC refclock, that puts the BIOS generated VCO
frequency at ~908 MHz, which is lower than the 930 MHz limit
we have currently. This also results in the pixel clock coming
out significantly higher than the requested 65 MHz when we try
to recompute it.

Reduce the the VCO limit to 908 MHz. Combined with the earlier
SSC reference clock accuracy fix, this results in the pixel clock
coming out as 65.08 MHz which is quite close to the target. For
some reason the BIOS uses 64.881 MHz, which isn't quite as close.

This makes kms_flip wf_vblank-ts-check pass for the first time
on this machine \o/

Cc: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix 66 MHz LVDS SSC freq for gen2
Ville Syrjälä [Mon, 9 Dec 2013 16:54:16 +0000 (18:54 +0200)]
drm/i915: Fix 66 MHz LVDS SSC freq for gen2

Store the SSC refclock frequency in kHz to get more accuracy. Currently
we're pretending that 66 MHz is ~66000 kHz, when in fact it is actually
~66667 kHz. By storing the less rounded kHz value we get a much better
accuracy for out pixel clock calculations.

Cc: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Increase gen2 vco frequency limit to 1512 MHz
Ville Syrjälä [Mon, 9 Dec 2013 16:54:15 +0000 (18:54 +0200)]
drm/i915: Increase gen2 vco frequency limit to 1512 MHz

Bruno Prémont has a 855 machine with a 1400x1050 LVDS screen.

The VBT mode is as follows:
0:"1400x1050" 0 108000 1400 1416 1528 1688 1050 1051 1054 1066 0x8 0xa

The BIOS uses the following DPLL settings:
DPLL = 0x90020000
FP0 = 0x2140e
FP1 = 0x21207

That puts the BIOS generated VCO frequency at 1512 MHz, which is
higher than the 1400 MHz limit we have currently.

Let's bump the VCO limit to 1512 MHz and see what happens.

Cc: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Change N divider minimum from 3 to 2 for gen2
Ville Syrjälä [Mon, 9 Dec 2013 16:54:14 +0000 (18:54 +0200)]
drm/i915: Change N divider minimum from 3 to 2 for gen2

Bruno Prémont has a 855 machine with a 1400x1050 LVDS screen.

The VBT mode is as follows:
0:"1400x1050" 0 108000 1400 1416 1528 1688 1050 1051 1054 1066 0x8 0xa

The BIOS uses the following DPLL settings:
DPLL = 0x90020000
FP0 = 0x2140e
FP1 = 0x21207

We can't generate that pixel clock currently as we're limiting the N
divider to at least 3, whereas the BIOS uses a value of 2.

Let's reduce the N minimum to 2 and see what happens.

Cc: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Extract p2 divider correctly for gen2 LVDS dual channel
Ville Syrjälä [Mon, 9 Dec 2013 16:54:13 +0000 (18:54 +0200)]
drm/i915: Extract p2 divider correctly for gen2 LVDS dual channel

In order to determine the correct p2 divider for LVDS on gen2,
we need to check the CLKB mode from the LVDS port register to
determine if we're dealing with single or dual channel LVDS.

Cc: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Bruno Prémont <bonbons@linux-vserver.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Record BB_ADDR for every ring
Ville Syrjälä [Tue, 10 Dec 2013 19:44:43 +0000 (21:44 +0200)]
drm/i915: Record BB_ADDR for every ring

Every ring seems to have a BB_ADDR registers, so include them all in the
error state.

v2: Also include the _UDW on BDW

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use 32bit read for BB_ADDR
Ville Syrjälä [Tue, 10 Dec 2013 18:47:44 +0000 (20:47 +0200)]
drm/i915: Use 32bit read for BB_ADDR

The BB_ADDR register is documented to be 32bits at least since SNB.
Prior to that the high 32bits were listed as MBZ, so using a 64bit read
doesn't seem worth anything. Also the simulator doesn't like the 64bit
read. So just switch to using a 32bit read instead.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix VDD override off wait
Paulo Zanoni [Fri, 6 Dec 2013 19:32:42 +0000 (17:32 -0200)]
drm/i915: fix VDD override off wait

If we're disabling the VDD override bit and the panel is enabled, we
don't need to wait for anything. If the panel is disabled, then we
need to actually wait for panel_power_cycle_delay, not
panel_power_down_delay, because the power down delay was already
respected when we disabled the panel.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't touch the VDD when disabling the panel
Paulo Zanoni [Fri, 6 Dec 2013 19:32:41 +0000 (17:32 -0200)]
drm/i915: don't touch the VDD when disabling the panel

I don't see a reason to touch VDD when we're disabling the panel:
since the panel is enabled, we don't need VDD. This saves a few sleep
calls from the vdd_on and vdd_off functions at every modeset.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=69693
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: Fix the patch mangle wiggle has done ... Spotted by Paulo.
Also drop the runtime_pm_put call which now has to go due to different
patch ordering. Also from Paul.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't enable VDD just to enable the panel
Paulo Zanoni [Fri, 6 Dec 2013 19:32:40 +0000 (17:32 -0200)]
drm/i915: don't enable VDD just to enable the panel

We just don't need this. This saves 250ms from every modeset on my
machine.

Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add runtime PM support on Haswell
Paulo Zanoni [Thu, 21 Nov 2013 15:47:28 +0000 (13:47 -0200)]
drm/i915: add runtime PM support on Haswell

The code to enable/disable PC8 already takes care of saving and
restoring all the registers we need to save/restore, so do a put()
call when we enable PC8 and a get() call when we disable it.

Ideally, in order to make it easier to add runtime PM support to other
platforms, we should move some things from the PC8 code to the runtime
PM code, but let's do this later, since we can make Haswell work right
now.

V2: - Rebase

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[danvet: Don't actually enable runtime pm since I didn't merge all
patches.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: disable interrupts when enabling PC8
Paulo Zanoni [Thu, 21 Nov 2013 15:47:25 +0000 (13:47 -0200)]
drm/i915: disable interrupts when enabling PC8

The plan is to merge PC8 and D3 into a single feature, and when we're
in D3 we won't get any hotplug interrupt anyway, so leaving them
enable doesn't make sense, and it also brings us a problem. The
problem is that we get a hotplug interrupt right when we we wake up
from D3, when we're still waking up everything. If we fully disable
interrupts we won't get this hotplug interrupt, so we won't have
problems.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: do not assert DE_PCH_EVENT_IVB enabled
Paulo Zanoni [Thu, 21 Nov 2013 15:47:24 +0000 (13:47 -0200)]
drm/i915: do not assert DE_PCH_EVENT_IVB enabled

The current code was checking if all bits of "val" were enabled and
DE_PCH_EVENT_IVB was disabled. The new code doesn't care about the
state of DE_PCH_EVENT_IVB: it just checks if everything else is 1.

The goal is that future patches may completely disable interrupts, and
the LCPLL-disabling code shouldn't care about the state of
DE_PCH_EVENT_IVB.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[danvet: I think the commit message is actually wrong in it's
description of what the old test checked, but the new one seems sane.
So meh.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: get a runtime PM reference when the panel VDD is on
Paulo Zanoni [Thu, 21 Nov 2013 15:47:23 +0000 (13:47 -0200)]
drm/i915: get a runtime PM reference when the panel VDD is on

And put it when it's off. Otherwise, when you run pm_pc8 from
intel-gpu-tools, and the delayed function that disables VDD runs,
we'll get some messages saying we're touching registers while the HW
is suspended.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add some runtime PM get/put calls
Paulo Zanoni [Wed, 27 Nov 2013 20:21:54 +0000 (18:21 -0200)]
drm/i915: add some runtime PM get/put calls

These are needed when we cat the debugfs and sysfs files.

V2: - Rebase
V3: - Rebase
V4: - Rebase

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add runtime put/get calls at the basic places
Paulo Zanoni [Wed, 27 Nov 2013 20:20:34 +0000 (18:20 -0200)]
drm/i915: add runtime put/get calls at the basic places

If I add code to enable runtime PM on my Haswell machine, start a
desktop environment, then enable runtime PM, these functions will
complain that they're trying to read/write registers while the
graphics card is suspended.

v2: - Simplify i915_gem_fault changes.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
[danvet: Drop the hunk in i915_hangcheck_elapsed, it's the wrong thing
to do.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: do adapter power state notification at runtime PM
Paulo Zanoni [Fri, 6 Dec 2013 22:34:21 +0000 (20:34 -0200)]
drm/i915: do adapter power state notification at runtime PM

Now that we are actually setting the device to the D3 state, we should
issue the notification.

The opregion spec says we should send the message before the adapter
is about to be placed in a lower power state, and after the adapter is
placed in a higher power state.

Jani originally wrote a similar patch for PC8, but then we discovered
that we were not really changing the PCI D states when
enabling/disabling PC8, so we had to postpone his patch.

v2: - Improve commit message, explaining the expected state.

v3: - Rebase.

Cc: Jani Nikula <jani.nikula@intel.com>
Credits-to: Jani Nikula <jani.nikula@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com> (v2)
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: add initial Runtime PM functions
Paulo Zanoni [Fri, 6 Dec 2013 22:32:13 +0000 (20:32 -0200)]
drm/i915: add initial Runtime PM functions

This patch adds the initial infrastructure to allow a Runtime PM
implementation that sets the device to its D3 state. The patch just
adds the necessary callbacks and the initial infrastructure.

We still don't have any platform that actually uses this
infrastructure, we still don't call get/put in all the places we need
to, and we don't have any function to save/restore the state of the
registers. This is not a problem since no platform uses the code added
by this patch. We have a few people simultaneously working on runtime
PM, so this initial code could help everybody make their plans.

V2: - Move some functions to intel_pm.c
    - Remove useless pm_runtime_allow() call at init
    - Remove useless pm_runtime_mark_last_busy() call at get
    - Use pm_runtime_get_sync() instead of 2 calls
    - Add a WARN to check if we're really awake

V3: - Rebase.

V4: - Don't need to call pci_{save,restore}_state and
      pci_set_power_sate, since they're already called by the PCI
      layer
    - Remove wrong pm_runtime_enable() call at init_runtime_pm

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: get a PC8 reference when enabling the power well
Paulo Zanoni [Wed, 27 Nov 2013 19:59:22 +0000 (17:59 -0200)]
drm/i915: get a PC8 reference when enabling the power well

In the current code, at haswell_modeset_global_resources, first we
decide if we want to enable/disable the power well, then we decide if
we want to enable/disable PC8. On the case where we're enabling PC8
this works fine, but on the case where we disable PC8 due to a non-eDP
monitor being enabled, we first enable the power well and then disable
PC8. Although wrong, this doesn't seem to be causing any problems now,
and we don't even see anything in dmesg. But the patches for runtime
D3 turn this problem into a real bug, so we need to fix it.

This fixes the "modeset-non-lpsp" subtest from the "pm_pc8" test from
intel-gpu-tools.

v2: - Rebase (i915_disable_power_well).
v3: - More reabase.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: WARN if !HAS_PC8 when enabling/disabling PC8
Paulo Zanoni [Thu, 21 Nov 2013 15:47:15 +0000 (13:47 -0200)]
drm/i915: WARN if !HAS_PC8 when enabling/disabling PC8

We already have some checks and shouldn't be reaching these places on
!HAS_PC8 platforms, but add a WARN,  just in case.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Move VLV PHY CRI clock enable into intel_init_dpio()
Ville Syrjälä [Tue, 10 Dec 2013 12:06:45 +0000 (14:06 +0200)]
drm/i915: Move VLV PHY CRI clock enable into intel_init_dpio()

The CRI clock is related to the display PHY, so the setup belongs
in intel_init_dpio().

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't cast away const from infoframe buffer
Ville Syrjälä [Tue, 10 Dec 2013 13:19:08 +0000 (15:19 +0200)]
drm/i915: Don't cast away const from infoframe buffer

We don't modify the packed infoframe data, so we should keep the
const qualifier in place. Just pass the buffer as 'const void *'
instead of 'const uint8_t *' and we can drop the cast entirely.

v2: Do intel_sdvo_write_infoframe() as well

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make downclock deduction common for all panels
Vandana Kannan [Tue, 10 Dec 2013 08:07:36 +0000 (13:37 +0530)]
drm/i915: Make downclock deduction common for all panels

If one mode of a internal panel has more than one refresh rate, then a reduced
clock is found for the LFP (LVDS/eDP). This enables switching between low
and high frequency dynamically. Moving downclock calculation to intel_panel
so that it is common for LVDS and eDP.

Signed-off-by: Vandana Kannan <vandana.kannan@intel.com>
Signed-off-by: Pradeep Bhat <pradeep.bhat@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove duplicate intel_uncore_forcewake_reset.
Deepak S [Sun, 8 Dec 2013 08:22:59 +0000 (13:52 +0530)]
drm/i915: Remove duplicate intel_uncore_forcewake_reset.

Since early sanitize and uncore sanitize are called one after the other,
I think, we can remove second forcewake reset which was are calling
twice in both the functions.

Note that this is merge fallout between

commit ef46e0d247da0a7a408573aa15870e231bbd4af2
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Sat Nov 16 16:00:09 2013 +0100

    drm/i915: restore the early forcewake cleanup

and

commit 521198a2e7095c8c7daa8d7d3a76a110c346be6f
Author: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Date:   Fri Aug 23 16:52:30 2013 +0300

    drm/i915: sanitize forcewake registers on reset

Signed-off-by: Deepak S <deepak.s@intel.com>
[danvet: Explain how this came to be.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoMerge tag 'v3.13-rc3' into drm-intel-next-queued
Daniel Vetter [Mon, 9 Dec 2013 08:17:02 +0000 (09:17 +0100)]
Merge tag 'v3.13-rc3' into drm-intel-next-queued

Linux 3.13-rc3

I need a backmerge for two reasons:
- For merging the ppgtt patches from Ben I need to pull in the bdw
  support.
- We now have duplicated calls to intel_uncore_forcewake_reset in the
  setup code to due 2 different patches merged into -next and 3.13.
  The conflict is silen so I need the merge to be able to apply
  Deepak's fixup patch.

Conflicts:
drivers/gpu/drm/i915/intel_display.c

Trivial conflict, it doesn't even show up in the merge diff.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: change CRTC assertion on LCPLL disable
Paulo Zanoni [Fri, 6 Dec 2013 22:29:01 +0000 (20:29 -0200)]
drm/i915: change CRTC assertion on LCPLL disable

Currently, PC8 is enabled at modeset_global_resources, which is called
after intel_modeset_update_state. Due to this, there's a small race
condition on the case where we start enabling PC8, then do a modeset
while PC8 is still being enabled. The racing condition triggers a WARN
because intel_modeset_update_state will mark the CRTC as enabled, then
the thread that's still enabling PC8 might look at the data structure
and think that PC8 is being enabled while a pipe is enabled. Despite
the WARN, this is not really a bug since we'll wait for the
PC8-enabling thread to finish when we call modeset_global_resources.

The spec says the CRTC cannot be enabled when we disable LCPLL, so we
had a check for crtc->base.enabled. If we change to crtc->active we
will still prevent disabling LCPLL while the CRTC is enabled, and we
will also prevent the WARN above.

This is a replacement for the previous patch named
    "drm/i915: get/put PC8 when we get/put a CRTC"

Testcase: igt/pm_pc8/modeset-lpsp-stress-no-wait
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoLinux 3.13-rc3
Linus Torvalds [Fri, 6 Dec 2013 17:34:04 +0000 (09:34 -0800)]
Linux 3.13-rc3

10 years agoMerge tag 'trace-fixes-3.13-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Fri, 6 Dec 2013 16:34:16 +0000 (08:34 -0800)]
Merge tag 'trace-fixes-3.13-rc2' of git://git./linux/kernel/git/rostedt/linux-trace

Pull tracing fix from Steven Rostedt:
 "A regression showed up that there's a large delay when enabling all
  events.  This was prevalent when FTRACE_SELFTEST was enabled which
  enables all events several times, and caused the system bootup to
  pause for over a minute.

  This was tracked down to an addition of a synchronize_sched()
  performed when system call tracepoints are unregistered.

  The synchronize_sched() is needed between the unregistering of the
  system call tracepoint and a deletion of a tracing instance buffer.
  But placing the synchronize_sched() in the unreg of *every* system
  call tracepoint is a bit overboard.  A single synchronize_sched()
  before the deletion of the instance is sufficient"

* tag 'trace-fixes-3.13-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt/linux-trace:
  tracing: Only run synchronize_sched() at instance deletion time