It's required by the specs, but we don't know why. Let's not find out
why.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Add an interrupt handler for switching graphics frequencies and handling
PM interrupts. This should allow for increased performance when busy
and lower power consumption when idle.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This patch changes the strategy for pageflip completion
timestamping. It detects if the pageflip completion
routine gets executed before or after drm_handle_vblank,
and thereby decides if the returned vblank count and
timestamp must be incremented by 1 frame(duration) or
not. It compares the current system time at invocation
against the current vblank timestamp. If the difference
is more than 0.9 video refresh interval durations then
it assumes the vblank timestamp and count are outdated
and need to be incremented and does so. Otherwise it
assumes a delayed pageflip irq and doesn't correct
the timestamp and count.
Advantage of this patch: Pageflip timestamping becomes
more robust against implementation errors and is
maintenance free for future GPU's.
Disadvantage: A few dozen (hundred?) nsecs extra
time spent in pageflip irq handler for each flip,
compared to hard-coded per-gpu settings?
Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
v2: Change IS_IRONLAKE to IS_GEN5 to adapt to 2.6.37
This patch adds new functions for use by the drm core:
.get_vblank_timestamp() provides a precise timestamp
for the end of the most recent (or current) vblank
interval of a given crtc, as needed for the DRI2
implementation of the OML_sync_control extension.
It is a thin wrapper around the drm function
drm_calc_vbltimestamp_from_scanoutpos() which does
almost all the work.
.get_scanout_position() provides the current horizontal
and vertical video scanout position and "in vblank"
status of a given crtc, as needed by the drm for use by
drm_calc_vbltimestamp_from_scanoutpos().
The patch modifies the pageflip completion routine
to use these precise vblank timestamps as the timestamps
for pageflip completion events.
This code has been only tested on a HP-Mini Netbook with
Atom processor and Intel 945GME gpu. The codepath for
(IS_G4X(dev) || IS_GEN5(dev) || IS_GEN6(dev)) gpu's
has not been tested so far due to lack of hardware.
Signed-off-by: Mario Kleiner <mario.kleiner@tuebingen.mpg.de>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Add frame buffer compression on Sandybridge. The method is similar to
Ironlake, except that two new registers of type GTTMMADR must be written
with the right fence info.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Add the support of memory self-refresh on Sandybridge, which is now
support 3 levels of watermarks and the source of the latency values
for watermarks has changed.
On Sandybridge, the LP0 WM value is not hardcoded any more. All the
latency value is now should be extracted from MCHBAR SSKPD register.
And the MCHBAR base address is changed, too.
For the WM values, if any calculated watermark values is larger than
the maximum value that can be programmed into the associated watermark
register, that watermark must be disabled.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
[ickle: remove duplicate compute routines and fixup for checkpatch]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Be paranoid and ensure that the vblank has passed and the scanout has
switched to the new fb, before unpinning the old one and possibly
tearing down its PTEs.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
As we already know the limits for the hardware clock, pass it down
rather than recomputing them for each match.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Don't post a downclocking task if the device is still active when the
idle timer fires. A pathological process could queue up several seconds
worth of processing and then go to sleep, during which time the idle
timer would kick in and downclock the GPU.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The ability to save the hardware context upon powering down the render
clock through PWRCTXA is only available on a couple of gen4 chipsets.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The bulk of the change is to convert the growing list of rings into an
array so that the relationship between the rings and the semaphore sync
registers can be easily computed.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Magic numbers from the specs. This is supposed to allow the PLL some
variance to improve jitter performance and VCO headroom across
manufacturing and environmental variations.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
... it's because setting the Pixel Multiply bits only takes effect once
the PLL is enabled and stable.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Fixes the modesetting on the secondary panel of the Libretto W100 and
presumably many more Ironlake laptops with SDVO LVDS displays.
Reported-and-tested-by: Matthew Willoughby <mattfredwill@gmail.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
Use the hardware DDA to calculate the ratio with as much accuracy as is
possible.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
If we leave the registers in a conflicting state then when we attempt
to teardown the active mode, we will not disable the pipes and planes
in the correct order -- leaving a plane reading from a disabled pipe and
possibly leading to undefined behaviour.
Reported-and-tested-by: Andy Whitcroft <apw@canonical.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=32078
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
The pipe is always set to 8BPC, but here we were leaving whatever
previous bits were set by the BIOS in place.
Signed-off-by: Eric Anholt <eric@anholt.net>
Tested-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
With this change, every batchbuffer can use all available fences (save
pinned and scanout, of course) without ever stalling the gpu!
In theory. Currently the actual pipelined update of the register is
disabled due to some stability issues. However, just the deferred update
is a significant win.
Based on a series of patches by Daniel Vetter.
The premise is that before every access to a buffer through the GTT we
have to declare whether we need a register or not. If the access is by
the GPU, a pipelined update to the register is made via the ringbuffer,
and we track the last seqno of the batches that access it. If by the
CPU we wait for the last GPU access and update the register (either
to clear or to set it for the current buffer).
One advantage of being able to pipeline changes is that we can defer the
actual updating of the fence register until we first need to access the
object through the GTT, i.e. we can eliminate the stall on set_tiling.
This is important as the userspace bo cache does not track the tiling
status of active buffers which generate frequent stalls on gen3 when
enabling tiling for an already bound buffer.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
... otherwise the panel-fitter may be left enabled with random settings
and cause unintended filtering (i.e. blurring of native modes on external
panels).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=31942
Reported-and-tested-by: Ben Kohler <bkohler@gmail.com>
Tested-by: Ciprian Docan <docan@eden.rutgers.edu>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When trying to diagnose mysterious errors on resume, capture the
display register contents as well.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
An old and oft reported bug, is that of the GPU hanging on a
MI_WAIT_FOR_EVENT following a mode switch. The cause is that the GPU is
waiting on a scanline counter on an inactive pipe, and so waits for a
very long time until eventually the user reboots his machine.
We can prevent this either by moving the WAIT into the kernel and
thereby incurring considerable cost on every swapbuffers, or by waiting
for the GPU to retire the last batch that accesses the framebuffer
before installing a new one. As mode switches are much rarer than swap
buffers, this looks like an easy choice.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=28964
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=29252
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: stable@kernel.org
We only ever used the PRB0, neglecting the secondary ring buffers, and
now with the advent of multiple engines with separate ring buffers we
need to excise the anachronisms from our code (and be explicit about
which ring we mean where). This is doubly important in light of the
FORCEWAKE required to read ring buffer registers on SandyBridge.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We use i915_gem_object_get_fence_reg() to do LRU tracking of the fence
registers, so stop trying to be too clever when pinning the fb->obj.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This is not known to fix any particular bugs we have, but the spec
says to do it, and the BIOS hadn't already set it up on my system.
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
a00b10c360 "Only enforce fence limits inside the GTT" also
added a fenceable/mappable disdinction when binding/pinning buffers.
This only complicates the code with no pratical gain:
- In execbuffer this matters on for g33/pineview, as this is the only
chip that needs fences and has an unmappable gtt area. But fences
are only possible in the mappable part of the gtt, so need_fence
implies need_mappable. And need_mappable is only set independantly
with relocations which implies (for sane userspace) that the buffer
is untiled.
- The overlay code is only really used on i8xx, which doesn't have
unmappable gtt. And it doesn't support tiled buffers, currently.
- For all other buffers it's a bug to pass in a tiled bo.
In short, this disdinction doesn't have any practical gain.
I've also reverted mapping the overlay and context pages as possibly
unmappable. It's not worth being overtly clever here, all the big
gains from unmappable are for execbuf bos.
Also add a comment for a clever optimization that confused me
while reading the original patch by Chris Wilson.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In a00b10c360 "Only enforce fence limits inside the GTT"
Chris Wilson implemented an optimization to only pin framebuffers
as mappable for crtc_set_base (but not for pageflips). This breaks
the abi, eg: A double buffering mesa client might leave the last
framebuffer in unmappable space on close. A subsequent glReadPix
by a frontbuffer rendering client then goes boom. My pretty anal
mappable/unmappable consistency checking detected this, see
https://bugs.freedesktop.org/show_bug.cgi?id=31286
Chris Wilson tried to fix this in 085ce26437 by pinning
tiled framebuffers into mappable space. This
a) renders the original optimization of not forcing framebuffers
for pageflipping clients into mappable pointless because all our
scanout buffers are tiled by default.
b) doesn't solve the problem for untiled framebuffers.
So kill this. Emperically it's no gain anyway because framebuffers are
being reused by the ddx and hence there's no chance for them to get
constanly bounced between mappable and unmappable.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We should enable FDI normal training on Sandybridge/CPT system
as well.
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>
[ickle: removed unrelated chunks]
Cc: stable@kernel.org
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
When merging Daniel's full-gtt patches I had a set of tweaks which I
thought I had undone. I was half right...
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=31286
Reported-by: jinjin.wang@intel.com
Reported-by: Alexey Fisher <bug-track@fisher-privat.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Immediate merge to resolve conflicts from applying a stability fix to
both branches.
Conflicts:
drivers/gpu/drm/i915/intel_ringbuffer.c
drivers/gpu/drm/i915/intel_ringbuffer.h
Part of the issue here was that Eric slipped in a debug hack for
testing the i915 IPS code before the intel_ips.c driver had landed.
This caused the driver to always use the full range of frequencies,
which is only legal when IPS tells us we have the headroom. Once that
hack was removed, there was confusion about the driver's frequency
clamping variables: max_delay is the driver's current limit on the
highest frequency the IPS driver wants us to use, while dev_priv->fmax
is the hardware-reported limit that the IPS driver can increase up to.
Tested with IPS driver loaded or not. Note that on Ironlake systems
without the IPS driver loaded this will result in a performance
reduction, and the inital warmup of frequency limits can impact
benchmarking on systems with IPS loaded.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
[ickle: demoted a debugging printk]
Cc: stable@kernel.org
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
So long as we adhere to the fence registers rules for alignment and no
overlaps (including with unfenced accesses to linear memory) and account
for the tiled access in our size allocation, we do not have to allocate
the full fenced region for the object. This allows us to fight the bloat
tiling imposed on pre-i965 chipsets and frees up RAM for real use. [Inside
the GTT we still suffer the additional alignment constraints, so it doesn't
magic allow us to render larger scenes without stalls -- we need the
expanded GTT and fence pipelining to overcome those...]
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Like before add a parameter mappable (also to gem_object_pin) and
set it depending upon the context. Only bos that are brought into
the gtt due to an execbuffer call can be put into the unmappable
part of the gtt, everything else (especially pinned objects) need
to be put into the mappable part of the gtt.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Preparing the ringbuffer for adding new commands can fail (a timeout
whilst waiting for the GPU to catch up and free some space). So check
for any potential error before overwriting HEAD with new commands, and
propagate that error back to the user where possible.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Block execbuffer for the fb to be flipped away, not the one that is to
be flipped in.
[ickle: rewritten for -next]
Signed-off-by: Chia-I Wu <olvaffe@gmail.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Kristian Høgsberg <krh@bitplanet.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
