For reasons that are not apparent to anybody, 990bbdad (drm/i915:
Group the GT routines together in both code and vtable) breaks the use
of the BitStream Decoder ring on SandyBridge. The active ingredient of
that patch is the conversion from a udelay(10) to a udelay(1) in the
busy-wait loop of waiting for the forcewake acknowledge. If we restore
that udelay(10) or insert another udelay(1) afterwards (or any wait
longer than 250ns) everything works again. An alternative is also to
remove any delay from the busy-wait loop.

Given that in the atomic sections we want to complete the wait as quick
as possible to avoid blocking the CPU for too long, it makes sense to
remove the delay altogether and simply spin on the exit condition until
it completes. So we replace the udelay(1) with cpu_relax().

Papers over regression from

commit 990bbdad
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Jul 2 11:51:02 2012 -0300

    drm/i915: Group the GT routines together in both code and vtable

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51738Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This function isn't used outside of intel_panel.c, so make it static.
Signed-off-by: NStéphane Marchesin <marcheu@chromium.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The exporter should have given us pages in the correct place, avoid
the prepare object mapping phase on dmar systems.

This fixes an oops on a GM45/R600 machine, when running the intel/radeon
tests.
Signed-off-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

the following warning was produced,

drivers/gpu/drm/i915/i915_gem_context.c: In function ‘i915_switch_context’:
drivers/gpu/drm/i915/i915_gem_context.c:454:6: warning: unused variable ‘ret’ [-Wunused-variable]

fix up by removing it
Signed-off-by: NDevendra Naga <devendra.aaru@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If a buffer that was the target of a PIPE_CONTROL from userland was a
reused one that hadn't been evicted which had not previously had this
workaround applied, then the early return for a correct
presumed_offset in this function meant we would not bind it into the
GTT and the write would land somewhere else.

Fixes reproducible failures with GL_EXT_timer_query usage in apitrace,
and I also expect it to fix the intermittent OQ issues on snb that
danvet's been working on.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=48019
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=52932Signed-off-by: NEric Anholt <eric@anholt.net>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NCarl Worth <cworth@cworth.org>
Tested-by: NCarl Worth <cworth@cworth.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

... by adding seemingly redudant posting reads.

This little dragon lair exploded the first time around when we've
refactored the code a bit to use the common wait_for_atomic_us in
"drm/i915: Group the GT routines together in both code and vtable",
which caused QA to file fdo bug #51738.

Chris Wilson entertained a few approaches to fixing #51738: Replacing
the udelay(1) with the previously-used udelay(10) (or any other
"sufficiently larger" delay), adding a posting read, or ditching the
delay completely and using cpu_relax. We went with the cpu_relax and
"915: Workaround hang with BSD and forcewake on SandyBridge". Which
blew up in fdo bug #52424, but adding the posting read while still
using cpu_relax seems to also fix that, it looks like the
posting read is the important ingriedient to fix these rc6 related
hangs on snb.

Popular theories as to why this is like it is include:
- A herd of pink elephants got royally angered somehow.

- The gpu has internally different functional units and judging by the
  register offsets, the forcewake request register and the forcewake
  ack registers are _not_ in the same functional unit (or at least
  aren't reached through the same routes). Hence the posting read
  syncs up with the wrong block and gets the entire gpu confused.

- ...

As a minimal ducttape fix for 3.6, let's just put these posting reads
into place again. We can try fancier approaches (like adding back the
cpu_relax instead of the udelay) in -next.

This (re-)fixes a regression introduced in

commit 990bbdad
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Mon Jul 2 11:51:02 2012 -0300

    drm/i915: Group the GT routines together in both code and vtable

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: NChris Wilson <chris@chris-wilson.co.uk>
Tested-by: NDu Yan <yanx.du@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=52424
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51738uSigned-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

drv_priv->gmbus is an array. Comparing it with NULL is somewhat less useful
than a chocolate teapot.

Possibly we should be testing bus != NULL each iteration of the loop
instead ?

gcc could help by warning too!
Signed-off-by: NAlan Cox <alan@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Signed-off-by: NAlan Cox <alan@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Otherwise our initial behaviour is "randomly save a bogus PLL
choice" as far as I can see.
Signed-off-by: NAlan Cox <alan@linux.intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We now refuse to load on gen6+ if kms is not enabled:

commit 26394d92
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Mon Mar 26 21:33:18 2012 +0200

    drm/i915: refuse to load on gen6+ without kms

Which results in the drm core calling our lastclose function to clean
up the mess, but that one is neatly broken for such failure cases
since kms has been introduced in

commit 79e53945
Author: Jesse Barnes <jbarnes@virtuousgeek.org>
Date:   Fri Nov 7 14:24:08 2008 -0800

    DRM: i915: add mode setting support
Reported-and-tested-by: NPaulo Zanoni <przanoni@gmail.com>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Fixes failures in transform feedback on gen7 because our SOL_RESET
flag was setting the transform feedback offsets in the old context
(occasionally happened to be ours) instead of the new context.
Signed-off-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Laurent Pinchart missed this when sending in is giant constify patch:

commit e811f5ae
Author: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Date:   Tue Jul 17 17:56:50 2012 +0200

    drm: Make the .mode_fixup() operations mode argument a const pointer

Acked-by; Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

LVDS is the first output where dpms on/off and prepare/commit don't
perfectly match. Now the idea behind this special case seems to be
that for simple resolution changes on the LVDS we don't need to stop
the pipe, because (at least on newer chips) we can adjust the panel
fitter on the fly.

There are a few problems with the current code though:
- We still stop and restart the pipe unconditionally, because the crtc
  helper code isn't flexible enough.
- We show some ugly flickering, especially when changing crtcs (this
  the crtc helper would actually take into account, but we don't
  implement the encoder->get_crtc callback required to make this work
  properly).

So it doesn't even work as advertised. I agree that it would be nice
to do resolution changes on LVDS (and also eDP) whithout blacking the
screen where the panel fitter allows to do that. But imo we should
implement this as a special case a few layers up in the mode set code,
akin to how we already detect simple framebuffer changes (and only
update the required registers with ->mode_set_base).

Until this is all in place, make our lives easier and just rip it out.

Also note that this seems to fix actual bugs with enabling the lvds
output, see:

http://lists.freedesktop.org/archives/intel-gfx/2012-July/018614.html

Cc: Takashi Iwai <tiwai@suse.de>
Cc: Giacomo Comes <comes@naic.edu>
Acked-by: NChris Wilson <chris@chris-wilson.co.uk>
Tested-by: NTakashi Iwai <tiwai@suse.de>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We need to check that "ctx" is a valid pointer before dereferencing it.
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Otherwise we end up trying to unpin a freed object and BUG.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This prevents a WARN introduced with

  commit de2b9985
  Author: Daniel Vetter <daniel.vetter@ffwll.ch>
  Date:   Wed Jul 4 22:52:50 2012 +0200

      drm/i915: don't return a spurious -EIO from intel_ring_begin
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The issue is that we stale data in the CPU caches, when we come to
swap-out the object, the CPU may short-circuit the reads from those
cacheline and so corrupt the context object.

Secondary, leaving the context object as being marked in the CPU write
domain whilst on the GPU active list is a bad idea and will throw
warnings later.

Note: Thanks to calling set_to_gtt_domain with write = false and not
setting any gpu write domain when putting a context object onto the
active list (when we switch away from it) the set_to_gtt_domain call
won't block.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
[danvet: Added a note to the commit message and a comment in the code
to explain the clever non-blocking trick.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

As we take the struct_mutex lock to access the command-stream, there is
a possibility that we may need to wait for a GPU hang and so should make
the lock both interruptible and error-checking.

References: https://bugs.freedesktop.org/show_bug.cgi?id=50069Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

When we enable/disable the CPU backlight registers we can't forget to
enable/disable the PCH backlight registers. Since we're using the CPU
registers we should also unset the override bit.

Fixes a regression on the following commit:
  drm/i915: properly enable the blc controller on the right pipe

The commit just deleted the code that sets the PCH registers, so it
was relying on the values set by the BIOS. I told my BIOS to boot on
the DVI monitor instead of the LVDS panel, so I noticed the bug.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

If we drop the breadcrumb request after a batch due to a signal for
example we aim to fix it up at the next opportunity. In this case we
emit a second batchbuffer with no waits upon the first and so no
opportunity to insert the missing request, so we need to emit the
missing flush for coherency. (Note that that invalidating the render
cache is the same as flushing it, so there should have been no
observable corruption.)

Note that beside simply adding the missing flush, avoiding potential
render corruption, this will also fix at least parts of the problem
introduced by some funny interaction of these two commits:

commit de2b9985
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Jul 4 22:52:50 2012 +0200

    drm/i915: don't return a spurious -EIO from intel_ring_begin

which allowed intel_ring_begin to return -ERESTARTSYS and

commit cc889e0f
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Wed Jun 13 20:45:19 2012 +0200

    drm/i915: disable flushing_list/gpu_write_list

which essentially disabled the flushing list.

The issue happens when we submit a batch & emit it, but get
interrupted (thanks to the first patch) while trying to emit the
flush. On the next batch we still assume that the full gpu domain
handling is in effect and hence compute the invalidate&flushing
domains. But thanks to the 2nd patch we totally ignore these and only
invalidate all gpu domains, presuming that any required flushes have
been issued already.  Which is wrong and eventually results in us
updating the new write_domain values with the computed
pending_write_domain values, which leaves an object with write_domain
== 0 on the gpu_write_list.

As soon as we try to unbind that object, things blow up.

Fix this by emitting the missing flush according to the new
ring->gpu_caches_dirty flag.

Note that this does _not_ fix all the current cases where we end up
with an object on the flushing_list that can't be flushed.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=52040Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
[danvet: Add bug explanation to commit message.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Or going from tiled to untiled may break.
Signed-off-by: NAnder Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

While debugging Haswell link train failures I observed that we never
try the maximum voltage configuration more than once consecutively. We
start the training, the monitor keeps telling us to increase the
voltage, then when we reach the maximum we just go back to the start
(because of the "memset" above "voltage_tries = 0"). When we reach
this point, we keep alternating between the maximum and the minimum
voltages until we give up.

The DP spec suggests that we should try the same voltage 5 times
before giving up. This patch makes us try the maximum voltage at
least 5 times before going back to the minimum voltages.

This patch does not fix any particular bug I'm aware of.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We have way too much lying hardware to rely on a simple "does someone
answer on the ddc i2c address?" check. And now it's unused, so just
kill it.
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Somehow detect_ddc manages to fall through all checks when we think
that something responds on the ddc i2c address, but the edid read
failed. Fix this up by explicitly checking for this case.

This fixes a regression on newer chips because since

commit aaa37730
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Sat Jun 16 15:30:32 2012 +0200

    drm/i915/crt: Do not rely upon the HPD presence pin

we use ddc detection also on hotplug capable platforms. And one of
these reads all 0s for any i2c transaction if nothing is connected to
the vga port.

v2: Implement Chris Wilson's review:
- simplify logic, default to "nothing detected"
- kill stale comment
- BUG_ON(!crt->type != ANALOG)

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51900Tested-by: NYang Guang <guang.a.yang@intel.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

I so totally suck.

This can cause a black screen if (for whatever reason) the bios
hasn't set this bit itself.

This regression has been introduced in

commit 7cf41601
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Jun 5 10:07:09 2012 +0200

    drm/i915: clear up backlight #define confusion on gen4+
Tested-by: NKenneth Graunke <kenneth@whitecape.org>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Having had to dive into the bspec to understand what each stage of the
workaround meant, and how that the ring broadcasting IDLE corresponded
with the GT powering down the ring (i.e. rc6) add comments to aide
the next reader.

And since the register "is used to control all aspects of PSMI and power
saving functions" that makes it quite interesting to inspect with
regards to RC6 hangs, so add it to the error-state.

v2: Rediscover the piece of magic, set the RNCID to 0 before waiting for
the ring to wake up.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It never quite worked despite the numerous workarounds, yet I still see
people trying to use this hardware and filing bug reports. As we no
longer even try to implement the workarounds, since 6a233c78
(drm/i915/ringbuffer: kill snb blt workaround), simply disable the ring.

v2: Add a message to inform the user about the limited capabilities of
their pre-production hardware.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NEric Anholt <eric@anholt.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This initializes power wells within the modeset_init_hw routine.
Testing has shown that this works for both driver load time and for
suspend-resume code paths.
Signed-off-by: NEugeni Dodonov <eugeni.dodonov@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

There is little point waking up every 10ms to service an interrupt which
we then promptly ignore. So only program the the PMIER to enable
interrupts for those events which we do handle, not all of them!
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Cc: Eugeni Dodonov <eugeni.dodonov@intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

There were some fields missed. Daniel pointed this out in review, and I
know I fixed it, but something happened somehow and some time.
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

*sigh* the docs had it spelled wrong, corrected it, and then proceeded
to re-do the original error. The original code preserved this history,
and this patch attempts to keep in sync with the current docs.
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Userspace uses long in quite a few places more than the kernel. Which
gives me neat proof that I'm the only guy on this side of the galaxy
who ever tried to run glxgears on a 64bit machine with sis graphics on
linux.

Note that the longs in drm_sis_mem_t aren't aligned properly, so this
won't even work with 32bit userspace on 64bit kernel as-is. Hence the
patch can't break that, either.

Nope, I'm not nuts enough to write the 32bit ioctl compat layer for
this and test it with some wine app. Even though hunting the ebay
dungeons for a sis card actually supported by the mesa drivers casts
some doubts on this ...
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

With the last patch to ditch DMA_QUEUE support, we should be able
to call the dma cleanup uncoditionally, even when the master has
disappeared.

Do so because it just makes more sense.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Absolutely unused. All the values are only ever initialized and
then used at most in some debug printout functions.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

Only one driver (i810) even sets that flag. Now the actual locking
code uncoditionally promotes lock->context to an unsigned int.

Closer inspection of the userspace reveals that the drm lock context
is defined as an unsigned int (at least on linux). I suspect we just
have a strange case of signedness confusion going on.

Tested on my i815, doesn't seem to break anything.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

All leftover users either haven't set DRIVER_HAVE_DMA, in which
case this will never be called, or use the drm_core implementation.

Call that directly in the only callsite.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

The reclaim_buffers function of the savage driver actually wants to run
with the hw_lock held - at least there are printks in the call-chain
to that effect. But the drm core only calls reclaim_buffers as used
by savage _after_ forcefully dropping the hwlock (in case it's still
hold by the closing fd).

So do the same idlelock dance as for the other dma drivers and hope
that papers over any issues.

v2: Don't let the idlelock linger around.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Tested-by: NTormod Volden <debian.tormod@gmail.com>
Signed-off-by: NDave Airlie <airlied@redhat.com>

i810 was the last user of this code, with that gone, kill it.
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>

This reverts commit 6e877b57,
reinstating the original commit:

commit 87499ffd
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue Oct 25 23:51:24 2011 +0200

    drm/i810: cleanup reclaim_buffers

    My dear old i815 always hits the deadlocked on reclaim_buffers
    warning. Switch over to the idlelock duct-tape on hope that
    works better. I've fired up my i815 and now closing glxgears doesn't
    take 5 seconds anymore. \o/

The original problem with that was that I've moved it ahead in the
series so that it could be included despite some patches not being
ready quite yet. The little problem is that this patch required some
of the previous rework to work correctly.

Now that everything is in the right order again, this actually works
on my i810 and does speed up closing gl apps as the original commit
claimed. Without hanging the machine, as the revert says.
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NDave Airlie <airlied@redhat.com>