In order to reduce the penalty of fallbacks under memory pressure and to
avoid a potential immediate ping-pong of evicting a mmaped buffer, we
move the object to the tail of the inactive list when a page is freshly
faulted or the object is moved into the CPU domain.
We choose not to protect the CPU objects from casual eviction,
preferring to keep the GPU active for as long as possible.
v2: Daniel Vetter found a bug where I forgot that pinned objects are
kept off the inactive list.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Based in a large part upon Daniel Vetter's implementation and adapted
for handling multiple rings in a single pass.
This should lead to better gtt usage and fixes the page-fault-of-doom
triggered. The fairness is provided by scanning through the GTT space
amalgamating space in rendering order. As soon as we have a contiguous
space in the GTT large enough for the new object (and its alignment),
evict any object which lies within that space. This should keep more
objects resident in the GTT.
Doing throughput testing on a PineView machine with cairo-perf-trace
indicates that there is very little difference with the new LRU scan,
perhaps a small improvement... Except oddly for the poppler trace.
Reference:
Bug 15911 - Intermittent X crash (freeze)
https://bugzilla.kernel.org/show_bug.cgi?id=15911
Bug 20152 - cannot view JPG in firefox when running UXA
https://bugs.freedesktop.org/show_bug.cgi?id=20152
Bug 24369 - Hang when scrolling firefox page with window in front
https://bugs.freedesktop.org/show_bug.cgi?id=24369
Bug 28478 - Intermittent graphics lockups due to overflow/loop
https://bugs.freedesktop.org/show_bug.cgi?id=28478
v2: Attempt to clarify the logic and order of eviction through the use
of comments and macros.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
The eviction code is the gnarly underbelly of memory management, and is
clearer if kept separated from the normal domain management in GEM.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
This will be used by the eviction logic to maintain fairness between the
rings.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
This does two little changes:
- Add an alignment parameter for evict_something. It's not really great to
whack a carefully sized hole into the gtt with the wrong alignment.
Especially since the fallback path is a full evict.
- With the inactive scan stuff we need to evict more that one object, so
move the unbind call into the helper function that scans for the object
to be evicted, too. And adjust its name.
No functional changes in this patch, just preparation.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
In order to properly track bound objects, they need to exist on one of
the inactive/active lists or be pinned. As this is a requirement, do the
work inside i915_gem_bind_to_gtt() rather than dotted around the
callsites.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
v2: Add the interrupt status and address.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
Using dev_priv__ avoids sparse complaining about shadowed variables in
the *LP_RING() macros.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
As the function has been reduced to a store plus increment, the body is
now smaller than the call so inline it.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
As we check that the ringbuffer will not wrap upon emission, we do not
need to check that incrementing the tail wrapped every time. However, we
do upon advancing just in case the tail is now pointing at the very end
of the ring.
Likewise we can account for the space used during emission in begin()
and avoid decrementing it for every emit.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
The tail is quadword aligned, so we can add two MI_NOOP as a time.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
This debugging trace was useful for finding the fbcon regression on
i965, and it may prove useful again in future.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
The comments have long desired that we should switch off the cursor
along with the display plane, make it so.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
My i855GM suffers from a 80k/s interrupt storm without this.
So add 2nd gen to the list of things that don't like more than
one outstanding pageflip request.
Furthermore I've changed the busy loop into a ringbuffer wait.
Busy-loops that don't check whether the chip died are simply evil.
And performance should actually improve, because there's usually
a decent amount of rendering queued on the gpu, hopefully rendering
that MI_WAIT into a noop by the time it's executed.
The current code holds dev->struct_mutex while executing this loop,
hence stalling all other gem activity anyway.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@kernel.org
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[anholt: resolved against conflict]
Signed-off-by: Eric Anholt <eric@anholt.net>
Add a new path for 2nd gen chips that uses the commands for i81x
chips (where public docs do exist) augmented with the plane bits
from i915. It seems to work and doesn't result in a black screen
like before.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: stable@kernel.org
[anholt: resolved against conflict]
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Eric Anholt <eric@anholt.net>
Incorporates a similar patch by Daniel Vetter, the alteration being to
report the current busy state after retiring.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>
This avoids the excess flush and requests on idle rings (and spamming
the debug log ;-)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
This property is slightly unusual in that it is a boolean and so has no
GET_MAX command.
Reference:
Bug 28636 - missing TV parameter "Dot Crawl freeze"
https://bugs.freedesktop.org/show_bug.cgi?id=28636
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Reference:
Bug 28634 - missing TV parameter "Flicker Filter"
https://bugs.freedesktop.org/show_bug.cgi?id=28634
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Make the code that tiny bit clearer by reducing the pointer dance.
2 files changed, 130 insertions(+), 147 deletions(-)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Subclass intel_encoder to reduce the pointer dance through
intel_encoder->dev_priv.
10 files changed, 896 insertions(+), 997 deletions(-)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
As we already have appropriate debug and warnings when we activate and
deactivate the self-refresh FIFO, having a further INFO is just annoying.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/wq: (55 commits)
workqueue: mark init_workqueues() as early_initcall()
workqueue: explain for_each_*cwq_cpu() iterators
fscache: fix build on !CONFIG_SYSCTL
slow-work: kill it
gfs2: use workqueue instead of slow-work
drm: use workqueue instead of slow-work
cifs: use workqueue instead of slow-work
fscache: drop references to slow-work
fscache: convert operation to use workqueue instead of slow-work
fscache: convert object to use workqueue instead of slow-work
workqueue: fix how cpu number is stored in work->data
workqueue: fix mayday_mask handling on UP
workqueue: fix build problem on !CONFIG_SMP
workqueue: fix locking in retry path of maybe_create_worker()
async: use workqueue for worker pool
workqueue: remove WQ_SINGLE_CPU and use WQ_UNBOUND instead
workqueue: implement unbound workqueue
workqueue: prepare for WQ_UNBOUND implementation
libata: take advantage of cmwq and remove concurrency limitations
workqueue: fix worker management invocation without pending works
...
Fixed up conflicts in fs/cifs/* as per Tejun. Other trivial conflicts in
include/linux/workqueue.h, kernel/trace/Kconfig and kernel/workqueue.c
* 'drm-core-next' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6: (204 commits)
agp: intel-agp: do not use PCI resources before pci_enable_device()
agp: efficeon-agp: do not use PCI resources before pci_enable_device()
drm: kill BKL from common code
drm/kms: Simplify setup of the initial I2C encoder config.
drm,io-mapping: Specify slot to use for atomic mappings
drm/radeon/kms: only expose underscan on avivo chips
drm/radeon: add new pci ids
drm: Cleanup after failing to create master->unique and dev->name
drm/radeon: tone down overchatty acpi debug messages.
drm/radeon/kms: enable underscan option for digital connectors
drm/radeon/kms: fix calculation of h/v scaling factors
drm/radeon/kms/igp: sideport is AMD only
drm/radeon/kms: handle the case of no active displays properly in the bandwidth code
drm: move ttm global code to core drm
drm/i915: Clear the Ironlake dithering flags when the pipe doesn't want it.
drm/radeon/kms: make sure HPD is set to NONE on analog-only connectors
drm/radeon/kms: make sure rio_mem is valid before unmapping it
drm/agp/i915: trim stolen space to 32M
drm/i915: Unset cursor if out-of-bounds upon mode change (v4)
drm/i915: Unreference object not handle on creation
...
* 'kms-merge' of git://git.kernel.org/pub/scm/linux/kernel/git/jwessel/linux-2.6-kgdb:
kgdb,docs: Update the kgdb docs to include kms
drm_fb_helper: Preserve capability to use atomic kms
i915: when kgdb is active display compression should be off
drm/i915: use new fb debug hooks
drm: add KGDB/KDB support
fb: add hooks to handle KDB enter/exit
kgdboc: Add call backs to allow kernel mode switching
vt,console,kdb: automatically set kdb LINES variable
vt,console,kdb: implement atomic console enter/leave functions
Commit 5349ef3127 (drm/fb: fix
FBIOGET/PUT_VSCREENINFO pixel clock handling) changed the logic of
when a pixclock was valid vs invalid.
The atomic kernel mode setting used by the kernel debugger relies upon
the drm_fb_helper_check_var() to always return -EINVAL. Until a
better solution exists, this behavior will be restored.
CC: David Airlie <airlied@linux.ie>
CC: Jesse Barnes <jbarnes@virtuousgeek.org>
CC: Clemens Ladisch <clemens@ladisch.de>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
If the HW compression is left on, the call backs from the HW will
crash the kernel. The only time this code is called is when kernel
mode setting is in use with kgdb and the kdb shell.
The atomic display pipe handler callback will reset everything when
kgdb restores kernel to the run state.
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
CC: David Airlie <airlied@linux.ie>
Implement atomic kernel mode settings using the fb layer's debug hook
system for supporting debugger interaction.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
Implement the callbacks for KDB entry/exit via the drm helpers.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Jason Wessel <jason.wessel@windriver.com>
This restricts the use of the big kernel lock to the i830 and i810
device drivers. The three remaining users in common code (open, ioctl
and release) get converted to a new mutex, the drm_global_mutex,
making the locking stricter than the big kernel lock.
This may have a performance impact, but only in those cases that
currently don't use DRM_UNLOCKED flag in the ioctl list and would
benefit from that anyway.
The reason why i810 and i830 cannot use drm_global_mutex in their
mmap functions is a lock-order inversion problem between the current
use of the BKL and mmap_sem in these drivers. Since the BKL has
release-on-sleep semantics, it's harmless but it would cause trouble
if we replace the BKL with a mutex.
Instead, these drivers get their own ioctl wrappers that take the
BKL around every ioctl call and then set their own handlers as
DRM_UNLOCKED.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: David Airlie <airlied@linux.ie>
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
In most use cases the driver will be using the same static config all
the time: interpreting i2c_board_info::platform_data as the default
config we can can save the GPU driver a redundant set_config() call.
Signed-off-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is required should we ever attempt to use an io-mapping where
KM_USER0 is verboten, such as inside an IRQ context.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Eric Anholt <eric@anholt.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
R4xx also uses the atom add connector function, but underscan is only
supported on avivo chips.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
v2: Userspace (notably xf86-video-{intel,ati}) became confused when
drmSetInterfaceVersion() started returning -EBUSY as they used a second
call (the first done in drmOpen()) to check their master credentials.
Since userspace wants to be able to repeatedly call
drmSetInterfaceVersion() allow them to do so.
v3: Rebase to drm-core-next.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This connector attribute allows you to enable or disable underscan
on a digital output to compensate for panels that automatically
overscan (e.g., many HDMI TVs). Valid values for the attribute are:
off - forces underscan off
on - forces underscan on
auto - enables underscan if an HDMI TV is connected, off otherwise
default value is auto.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Prior to this patch the code was dividing the src_v by the dst_h
and vice versa, rather than src_v/dst_v and src_h/dst_h.
This could lead to problems in the calculation of the display
watermarks.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
* 'intel/drm-intel-next' of /ssd/git/drm-next: (230 commits)
drm/i915: Clear the Ironlake dithering flags when the pipe doesn't want it.
drm/agp/i915: trim stolen space to 32M
drm/i915: Unset cursor if out-of-bounds upon mode change (v4)
drm/i915: Unreference object not handle on creation
drm/i915: Attempt to uncouple object after catastrophic failure in unbind
drm/i915: Repeat unbinding during free if interrupted (v6)
drm/i915: Refactor i915_gem_retire_requests()
drm/i915: Warn if we run out of FIFO space for a mode
drm/i915: Round up the watermark entries (v3)
drm/i915: Typo in (unused) register mask for overlay.
drm/i915: Check overlay stride errata for i830 and i845
drm/i915: Validate the mode for eDP by using fixed panel size
drm/i915: Always use the fixed panel timing for eDP
drm/i915: Enable panel fitting for eDP
drm/i915: Add fixed panel mode parsed from EDID for eDP without fixed mode in VBT
drm/i915/sdvo: Set sync polarity based on actual mode
drm/i915/hdmi: Set sync polarity based on actual mode
drm/i915/pch: Set transcoder sync polarity for DP based on actual mode
drm/i915: Initialize LVDS and eDP outputs before anything else
drm/i915/dp: Correctly report eDP in the core connector type
...
Intel variants don't support it.
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
Logic was:
if (mode0 && mode1)
else if (mode0)
else
Should be:
if (mode0 && mode1)
else if (mode0)
else if (mode1)
Otherwise we may end up calculating the priority regs with
unitialized values.
Fixes:
https://bugzilla.kernel.org/show_bug.cgi?id=16492
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: stable@kernel.org
Signed-off-by: Dave Airlie <airlied@redhat.com>
I wrote this for the prime sharing work, but I also noticed other external
non-upstream drivers from a large company carrying a similiar patch, so I
may as well ship it in master.
Signed-off-by: Dave Airlie <airlied@redhat.com>
My fine DisplayPort output was getting ST dithering forever after
having had the LVDS enabled at one point.
Signed-off-by: Eric Anholt <eric@anholt.net>
If we were not able to map the io bar in device init, don't attempt
to unmap it in device fini. All radeons should have a io bar, so
I doubt this would ever trigger, but just to be on the safe side...
Pointed out by: Alberto Milone <alberto.milone@canonical.com>
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
