I see. As long as amd can still just have kernel hiding unsupported mode by
doing a pre-train I am okay with the proposal. Just to caution you the link
training fallback we implemented on windows in old dal architecture is very
painful to get right. Windows 7, 8.1 and 10 all behave
On Mon, Nov 14, 2016 at 11:31:11AM +, Tvrtko Ursulin wrote:
>
> On 14/11/2016 08:57, Chris Wilson wrote:
> >This emulates execlists on top of the GuC in order to defer submission of
> >requests to the hardware. This deferral allows time for high priority
> >requests to gazump their way to the
I forgot to pass -1 to git send-email, that's why this patch ended up here.
Please ignore.
On Mon, 2016-11-14 at 16:24 +0200, Ander Conselvan de Oliveira wrote:
> v2: Add new 0x3185 ID. (Joonas)
> Signed-off-by: Ander Conselvan de Oliveira com>
> Reviewed-by:
Geminilake is an Intel® Processor containing Intel® HD Graphics
following Broxton.
Let's start by adding the platform definition. PCI IDs and plaform
specific code will follow.
v2: Rebase (don't allow dev to be used with the new macro).
v3: Update ddb size. (Matt)
Rebase on
On Mon, Nov 14, 2016 at 11:48:32AM +, Tvrtko Ursulin wrote:
>
> On 14/11/2016 11:41, Chris Wilson wrote:
> >On Mon, Nov 14, 2016 at 11:15:52AM +, Tvrtko Ursulin wrote:
> >>On 14/11/2016 08:56, Chris Wilson wrote:
> >>>+static void execlists_schedule(struct drm_i915_gem_request *request,
v2: Add new 0x3185 ID. (Joonas)
Signed-off-by: Ander Conselvan de Oliveira
Reviewed-by: Rodrigo Vivi
---
drivers/gpu/drm/i915/i915_pci.c | 1 +
include/drm/i915_pciids.h | 4
2 files changed, 5 insertions(+)
diff --git
From: Rodrigo Vivi
The following LP platform inherits a lot of this platform
So let's simplify here to re-use this later.
v2: Keep ddb_size out of the new macro.
Signed-off-by: Rodrigo Vivi
Signed-off-by: Ander Conselvan de Oliveira
On Mon, Nov 14, 2016 at 08:54:53AM +, Chris Wilson wrote:
> If the LLC is coherent with the object, we do not need to worry about
> whether main memory and cache mismatch when we hand the object back to
> the system.
>
> Signed-off-by: Chris Wilson
Makes sense to
On Mon, Nov 14, 2016 at 02:44:35PM +0200, Petri Latvala wrote:
> Chris, happy with this revision?
Me? No. It still uses a thread instead of events, so I don't think it
qualifies as a good example for anyone else wanting to do the same thing.
Lots of hardcoded expectations (specific sleep
== Series Details ==
Series: drm_crtc.[hc] cleanup and documenation
URL : https://patchwork.freedesktop.org/series/15272/
State : warning
== Summary ==
Series 15272v1 drm_crtc.[hc] cleanup and documenation
https://patchwork.freedesktop.org/api/1.0/series/15272/revisions/1/mbox/
Test
Chris, happy with this revision?
On Fri, Oct 21, 2016 at 09:22:13AM +0530, Nautiyal Ankit wrote:
> From: Ramalingam C
>
> Idleness DRRS:
> By default the DRRS state will be at DRRS_HIGH_RR. When a Display
> content is Idle for more than 1Sec Idleness will be
On to, 2016-11-10 at 14:16 +, Patchwork wrote:
> == Series Details ==
>
> Series: series starting with [v2] drm: Add missing mutex_destroy in
> drm_dev_init/release (rev2)
> URL : https://patchwork.freedesktop.org/series/15110/
> State : success
Daniel applied the other patch and I the
== Series Details ==
Series: dma-buf: Use fence_get_rcu_safe() for retrieving the exclusive fence
URL : https://patchwork.freedesktop.org/series/15269/
State : warning
== Summary ==
Series 15269v1 dma-buf: Use fence_get_rcu_safe() for retrieving the exclusive
fence
On Mon, Nov 14, 2016 at 11:55:40AM +, Chris Wilson wrote:
> The current code is subject to a race where we may try to acquire a
> reference on a stale fence:
From i915.ko pov, this
Fixes: d07f0e59b2c7 ("drm/i915: Move GEM activity tracking into a common struct
reservation_object")
-Chris
--
Put the callback docs into struct drm_driver, and the small overview
into a DOC comment.
Signed-off-by: Daniel Vetter
---
Documentation/gpu/drm-kms.rst | 42 ++---
drivers/gpu/drm/drm_dumb_buffers.c | 46
Just noise.
Signed-off-by: Daniel Vetter
---
include/drm/drm_crtc.h | 14 +++---
1 file changed, 7 insertions(+), 7 deletions(-)
diff --git a/include/drm/drm_crtc.h b/include/drm/drm_crtc.h
index cf96b393091a..bcc1a4d1d1a6 100644
--- a/include/drm/drm_crtc.h
And also put the overview section into the KMS Properties part of the
docs, instead of randomly-placed within the helpers - this is part of
the uabi.
With this patch I think drm_crtc.[hc] is cleaned up and entirely
documented.
Signed-off-by: Daniel Vetter
---
And shuffle the kernel-doc structure a bit since drm_crtc.[hc] now
only contains CRTC-related functions and structures.
Signed-off-by: Daniel Vetter
---
Documentation/gpu/drm-kms.rst | 32 +-
drivers/gpu/drm/Makefile| 2 +-
Just cleans up what's there, still plenty missing.
Signed-off-by: Daniel Vetter
---
Documentation/gpu/drm-internals.rst | 3 +
include/drm/drm_drv.h | 168 +++-
2 files changed, 109 insertions(+), 62 deletions(-)
diff
Would be great if everony could add
$ make DOCBOOKS="" htmldocs
to their build scripts to catch these. 0day should also report them,
not sure why it failed to spot this.
Fixes: b42fe9ca0a1e ("drm/i915: Split out i915_vma.c")
Cc: Tvrtko Ursulin
Cc: Chris Wilson
kerneldoc expects the comment next to definitions, otherwise it can't
pick up exported vs. internal stuff.
This fixes a warning from the doc build done with:
$ make DOCBOOKS="" htmldocs
Fixes: d8187177b0b1 ("drm: add helper for printing to log or seq_file")
Cc: Rob Clark
Just code movement, doc cleanup will follow up later.
Signed-off-by: Daniel Vetter
---
drivers/gpu/drm/Makefile| 3 +-
drivers/gpu/drm/drm_crtc.c | 109 -
drivers/gpu/drm/drm_crtc_internal.h | 18 ++---
I want to move dumb buffer documentation into the right vfuncs, and
for that I first need to be able to pull that into kerneldoc without
having to clean up all of drmP.h. Also, header-splitting is nice.
While at it shuffle all the function declarations for drm_drv.c into
the right spots, and drop
Hi all,
Final bits, finally, leaving really only CRTC related code and structures in
drm_crtc.[hc].
I guess next up is a bit of polish for atomic code, that has grown a lot and I
think it'd be useful to split at least the helpers a bit. And a recent
discussion with Rob Clark showed me that we've
Would be great if everony could add
$ make DOCBOOKS="" htmldocs
to their build scripts to catch these. 0day should also report them,
not sure why it failed to spot this.
Fixes: f54d1867005c ("dma-buf: Rename struct fence to dma_fence")
Cc: Chris Wilson
Cc: Gustavo
The current code is subject to a race where we may try to acquire a
reference on a stale fence:
[13703.335118] WARNING: CPU: 1 PID: 14975 at ./include/linux/kref.h:46
i915_gem_object_wait+0x1a3/0x1c0
[13703.335184] Modules linked in:
[13703.335202] CPU: 1 PID: 14975 Comm: gem_concurrent_ Not
On 14/11/2016 11:41, Chris Wilson wrote:
On Mon, Nov 14, 2016 at 11:15:52AM +, Tvrtko Ursulin wrote:
On 14/11/2016 08:56, Chris Wilson wrote:
+static void execlists_schedule(struct drm_i915_gem_request *request, int prio)
+{
+ struct intel_engine_cs *engine = NULL;
+ struct
Hi Andrew,
On 11/11/2016 08:50, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin
Userptr backing store with SWIOTBL active is currently allocated in the same
inefficient manner, with one sg entry per object page, as what the commit
871dfbd67d4e ("drm/i915: Allow
On Mon, Nov 14, 2016 at 11:15:52AM +, Tvrtko Ursulin wrote:
> On 14/11/2016 08:56, Chris Wilson wrote:
> >+static void execlists_schedule(struct drm_i915_gem_request *request, int
> >prio)
> >+{
> >+struct intel_engine_cs *engine = NULL;
> >+struct i915_dependency *dep, *p;
> >+
On 14/11/2016 08:57, Chris Wilson wrote:
Use a priority stored in the context as the initial value when
submitting a request. This allows us to change the default priority on a
per-context basis, allowing different contexts to be favoured with GPU
time at the expense of lower importance work.
On 14/11/2016 08:57, Chris Wilson wrote:
This emulates execlists on top of the GuC in order to defer submission of
requests to the hardware. This deferral allows time for high priority
requests to gazump their way to the head of the queue, however it nerfs
the GuC by converting it back into a
On the DMA mapping error path, sg may be NULL (it has already been
marked as the last scatterlist entry), and we should avoid dereferencing
it again.
Reported-by: Dan Carpenter
Fixes: e227330223a7 ("drm/i915: avoid leaking DMA mappings")
Signed-off-by: Chris Wilson
On Mon, Nov 14, 2016 at 02:14:56PM +0300, Dan Carpenter wrote:
> Hello Chris Wilson,
>
> The patch 871dfbd67d4e: "drm/i915: Allow compaction upto SWIOTLB max
> segment size" from Oct 11, 2016, leads to the following static
> checker warning:
>
> drivers/gpu/drm/i915/i915_gem.c:2357
On 14/11/2016 08:56, Chris Wilson wrote:
In order to support userspace defining different levels of importance to
different contexts, and in particular the preferred order of execution,
store a priority value on each context. By default, the kernel's
context, which is used for idling and other
== Series Details ==
Series: Add support for GuC-based SLPC (rev6)
URL : https://patchwork.freedesktop.org/series/2691/
State : success
== Summary ==
Series 2691v6 Add support for GuC-based SLPC
https://patchwork.freedesktop.org/api/1.0/series/2691/revisions/6/mbox/
fi-bdw-5557u
On 14/11/2016 08:56, Chris Wilson wrote:
Track the priority of each request and use it to determine the order in
which we submit requests to the hardware via execlists.
The priority of the request is determined by the user (eventually via
the context) but may be overridden at any time by the
Hello Chris Wilson,
The patch 871dfbd67d4e: "drm/i915: Allow compaction upto SWIOTLB max
segment size" from Oct 11, 2016, leads to the following static
checker warning:
drivers/gpu/drm/i915/i915_gem.c:2357 i915_gem_object_get_pages_gtt()
error: we previously assumed 'sg' could be
On 14/11/2016 08:56, Chris Wilson wrote:
The scheduler needs to know the dependencies of each request for the
lifetime of the request, as it may choose to reschedule the requests at
any time and must ensure the dependency tree is not broken. This is in
additional to using the fence to only
On 14/11/2016 08:56, Chris Wilson wrote:
Defer the transfer from the client's timeline onto the execution
timeline from the point of readiness to the point of actual submission.
For example, in execlists, a request is finally submitted to hardware
when the hardware is ready, and only put onto
On 14/11/2016 08:56, Chris Wilson wrote:
Localise the static struct lock_class_key to the caller of
i915_sw_fence_init() so that we create a lock_class instance for each
unique sw_fence rather than all sw_fences sharing the same
lock_class. This eliminate some lockdep false positive when using
On Thu, 06 Oct 2016, Tomeu Vizoso wrote:
> diff --git a/drivers/gpu/drm/i915/intel_display.c
> b/drivers/gpu/drm/i915/intel_display.c
> index 23a6c7213eca..7412a05fa5d9 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
>
From: Tom O'Rourke
This patch makes SLPC enabled by default on
platforms with hardware/firmware support.
v1: Removing warning "enable_slpc < 0" as it is
set to -1 with this patch now. This was caught by CI BAT.
v2-v4: Rebase.
v5: Sanitizing SLPC option based on
This tests whether SLPC in GuC is configured properly through shared data.
It checks whether GTPERF is running in default state, post reset and post
system suspend/resume.
This test will be extended further based on enablement of other SLPC tasks.
Signed-off-by: Sagar Arun Kamble
v2: Removing checks for vma obj and kmap_atomic validity. (Chris)
v3: Rebase.
v4: Updated to make sure SLPC enable keeps min/max freq softlimits
unchanged after initializing once. (Chris)
v5: s/first_enable/i915_load_enable. Updating freq softlimits after
checking that SLPC has
From: Tom O'Rourke
SLPC behavior can be changed through set of parameters.
These parameters can be updated and queried from i915 though
Host to GuC SLPC events. This patch add parameter values and
events for setting/unsetting parameters.
v1: Use host2guc_slpc
update
From: Tom O'Rourke
If slpc enabled, then add enable SLPC flag to guc
control parameter during guc load.
v1: Use intel_slpc_enabled() (Paulo)
v2-v4: Rebase.
v5: Changed intel_slpc_enabled() to i915.enable_slpc. (Sagar)
Signed-off-by: Tom O'Rourke
From: Tom O'Rourke
When frequency requests are made by SLPC, host driver
should not attempt to make frequency requests due to
potential conflicts.
Host-based turbo operations are already avoided when
SLPC is used. This change covers other frequency
requests such as from
v1: Updated tasks and frequency post reset.
Added DFPS param update for MAX_FPS and FPS Stall.
v2-v3: Rebase.
v4: Updated with GuC firmware v9.
v5: Rebase. Replaced H2G interrupts for parameter override with memory
setup with required parameters.
Signed-off-by: Sagar Arun Kamble
From: Tom O'Rourke
This patch adds has_slpc to skylake info.
The SLPC interface has changed and could continue to
change. Only GuC versions known to be compatible are
supported here.
On Skylake, GuC firmware v6 is supported. Other
platforms and versions can be added
From: Tom O'Rourke
Expose host2guc_action for use by SLPC in intel_slpc.c.
Expose functions to allocate and release objects used
by GuC to be used for SLPC shared memory object.
v1: Updated function names as they need to be made extern. (ChrisW)
v2-v4: Rebase
v5:
From: Tom O'Rourke
Send SLPC shutdown event during disable, suspend, and reset
operations. Sending shutdown event while already shutdown
is OK.
v1: Return void instead of ignored error code (Paulo)
Removed WARN_ON for checking msb of gtt address of
shared gem
From: Tom O'Rourke
Adds has_slpc to broxton info and adds broxton firmware version check
to sanitize_slpc_option.
v1: Adjusted slpc version check for major version 8.
Added message if version mismatch happens for easier debug. (Sagar)
v2-v3: Rebase.
v4: Commit
From: Tom O'Rourke
i915.enable_slpc is used to override the default for slpc usage.
The expected values are -1=auto, 0=disabled [default], 1=enabled.
slpc_enable_sanitize() converts i915.enable_slpc to either 0 or 1.
Interpretation of default value is based on
From: Tom O'Rourke
Update sysfs and debugfs functions to set SLPC
parameters when setting max/min frequency.
v1: Update for SLPC 2015.2.4 (params for both slice and unslice)
Replace HAS_SLPC with intel_slpc_active() (Paulo)
v2-v4: Rebase.
v5: Removed typecasting
From: Tom O'Rourke
On platforms with SLPC support: call intel_slpc_*() functions from
intel_*_gt_powersave() functions and GuC setup functions and do not
use rps functions. intel_slpc_enable is tied to GuC setup.
With SLPC, intel_enable_gt_powersave will only handle RC6
From: Tom O'Rourke
i915_slpc_info shows the contents of SLPC shared data
parsed into text format.
v1: Reformat slpc info (Radek)
squashed query task state info
in slpc info, kunmap before seq_print (Paulo)
return void instead of ignored return value (Paulo)
From: Tom O'Rourke
Adds debugfs hooks for enabling/disabling each slpc task.
The enable/disable debugfs files are
i915_slpc_gtperf, i915_slpc_balancer, and i915_slpc_dcc.
Each of these can take the values:
"default", "enabled", or "disabled"
v1: update for SLPC
From: Tom O'Rourke
Communication with SLPC is via Host to GuC interrupt.
This patch defines the data structure to be passed as input
and received as output from SLPC. This patch also defines the
events to be sent as input and status values output by GuC
on processing SLPC
From: Tom O'Rourke
Add has_slpc capablity flag to indicate GuC firmware
supports single loop power control (SLPC). SLPC is
a replacement for some host-based power management
features.
v1: fix whitespace (Sagar)
Reviewed-by: David Weinehall
With GuC based SLPC, frequency control will be moved to GuC and Host will
continue to control RC6 and Ring frequency setup. SLPC can be enabled in
the GuC setup path and can happen in parallel in GuC with other i915 setup.
Hence we can do away with deferred RPS enabling. This needs separate
SLPC (Single Loop Power Controller) is a replacement for some host-based
power management features. The SLPC implementation runs in firmware on GuC.
This series has been tested with SKL and APL GuC firmware v9 (not released
yet). Current GuC firmware's SLPC interfaces are updated in v9.
The
From: Tom O'Rourke
Send host2guc SLPC reset event to GuC post GuC load.
Post this, i915 can ascertain if SLPC has started running successfully
through shared data. This check is done during intel_init_gt_powersave.
This allows to get initial configuration setup by SLPC
From: Tom O'Rourke
When SLPC is controlling requested frequency, the rps.cur_freq
value is not used to make the frequency request.
Requested frequency from register RPNSWREQ has the value
most recently requested by SLPC firmware. Adding new sysfs
interface
From: Tom O'Rourke
The SLPC interface is dependent on GuC version.
Only GuC versions known to be compatible are supported here.
On SKL, BXT, KBL, GuC firmware v9 is supported. Other
platforms and versions can be added here later.
v1: Updated with modified
From: Tom O'Rourke
SLPC shared data is used to pass information
to/from SLPC in GuC firmware.
For Skylake, platform sku type and slice count
are identified from device id and fuse values.
Support for other platforms needs to be added.
v1: Update for SLPC interface
Cc: Chris Wilson
Cc: Daniel Vetter
Signed-off-by: Abdiel Janulgue
---
tests/kms_busy.c | 75 +++-
1 file changed, 4 insertions(+), 71 deletions(-)
diff --git
On 11.11.2016 18:16, Daniel Vetter wrote:
> On Fri, Nov 11, 2016 at 07:41:10PM +0200, Abdiel Janulgue wrote:
>> A lot of igt testcases need some GPU workload to make sure a race
>> window is big enough. Unfortunately having a fixed amount of
>> workload leads to spurious test failures or overtly
Cc: Chris Wilson
Cc: Daniel Vetter
Signed-off-by: Abdiel Janulgue
---
tests/kms_flip.c | 185 ++-
1 file changed, 4 insertions(+), 181 deletions(-)
diff --git
A lot of igt testcases need some GPU workload to make sure a race
window is big enough. Unfortunately having a fixed amount of
workload leads to spurious test failures or overtly long runtimes
on some fast/slow platforms. This library contains functionality
to submit GPU workloads that should
Cc: Chris Wilson
Cc: Daniel Vetter
Signed-off-by: Abdiel Janulgue
---
tests/gem_wait.c | 125 ---
1 file changed, 7 insertions(+), 118 deletions(-)
diff --git
== Series Details ==
Series: series starting with [1/2] drm/i915: Skip clflushes for all non-page
backed objects
URL : https://patchwork.freedesktop.org/series/15255/
State : warning
== Summary ==
Series 15255v1 Series without cover letter
== Series Details ==
Series: dma-buf: Replace reservation shared fence array with a compressed radix
tree
URL : https://patchwork.freedesktop.org/series/15252/
State : success
== Summary ==
Series 15252v1 dma-buf: Replace reservation shared fence array with a
compressed radix tree
== Series Details ==
Series: drm/i915: Prune the reservation shared fence array
URL : https://patchwork.freedesktop.org/series/15254/
State : warning
== Summary ==
Series 15254v1 drm/i915: Prune the reservation shared fence array
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
lib/igt_kms.c | 22 ++
tests/kms_atomic_transition.c | 30 --
2 files changed, 50 insertions(+), 2
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic_transition.c | 22 +-
1 file changed, 17 insertions(+), 5 deletions(-)
diff --git a/tests/kms_atomic_transition.c
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic.c | 124 +
1 file changed, 115 insertions(+), 9 deletions(-)
diff --git a/tests/kms_atomic.c
Hi Chris,
Am 14.11.2016 um 09:31 schrieb Chris Wilson:
The primary operation on the shared fence arrays is insertion and
retrieval. Retrieval is reasonably fast, as we just copy the array, but
insertion into the shared fence array is slow as we must iterate over all
current fences to discard
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic_transition.c | 92 ++-
1 file changed, 82 insertions(+), 10 deletions(-)
diff --git
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic_transition.c | 10 +-
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/tests/kms_atomic_transition.c b/tests/kms_atomic_transition.c
From: Robert Foss
Add support dor the IN_FENCE_FD property to enable setting in fences for atomic
commits.
Signed-off-by: Robert Foss
---
lib/igt_kms.c | 20
lib/igt_kms.h | 5 +
2 files changed, 25 insertions(+)
From: Gustavo Padovan
Add support for the OUT_FENCE_PTR property to enable setting out fences for
atomic commits.
Signed-off-by: Gustavo Padovan
---
lib/igt_kms.c | 20 +++-
lib/igt_kms.h | 3 +++
2 files
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic.c | 123 -
1 file changed, 37 insertions(+), 86 deletions(-)
diff --git a/tests/kms_atomic.c
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
lib/igt_kms.c | 6 +++---
lib/igt_kms.h | 5 +
2 files changed, 8 insertions(+), 3 deletions(-)
diff --git a/lib/igt_kms.c b/lib/igt_kms.c
index aa9fd16..8aaff5b
From: Gustavo Padovan
Hi,
That is the first version of the igt tests for DRM fences[1]. The
first four patches are just fix/improvements on the kms_atomic
infrastructure.
These patches depends on Robert Foss tests for sw_sync and a branch
with those tests
From: Gustavo Padovan
If the event never arrives we can timeout with select and end the test.
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic_transition.c | 8
1 file changed, 8 insertions(+)
diff --git
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
lib/igt_kms.c | 34 +-
1 file changed, 17 insertions(+), 17 deletions(-)
diff --git a/lib/igt_kms.c b/lib/igt_kms.c
index 989704e..aa9fd16
From: Gustavo Padovan
Now other gpus are supported too.
Signed-off-by: Gustavo Padovan
---
lib/drmtest.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/drmtest.c b/lib/drmtest.c
index 884fe7c..9f3ac7f
From: Gustavo Padovan
Support the virtio GPU on drmtest.
Signed-off-by: Gustavo Padovan
---
lib/drmtest.c | 9 +
lib/drmtest.h | 1 +
2 files changed, 10 insertions(+)
diff --git a/lib/drmtest.c b/lib/drmtest.c
index
From: Gustavo Padovan
Signed-off-by: Gustavo Padovan
---
tests/kms_atomic_transition.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/tests/kms_atomic_transition.c b/tests/kms_atomic_transition.c
index
On Fri, Nov 11, 2016 at 08:50:17AM +, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin
>
> Scatterlist entries have an unsigned int for the offset so
> correct the sg_alloc_table_from_pages function accordingly.
>
> Since these are offsets withing a page, unsigned int
On Fri, Nov 11, 2016 at 12:36:13PM +, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin
>
> With the addition of __sg_alloc_table_from_pages we can control
> the maximum coallescing size and eliminate a separate path for
> allocating backing store here.
>
> Similar to
On Fri, Nov 11, 2016 at 02:17:44PM +, Tvrtko Ursulin wrote:
> From: Tvrtko Ursulin
>
> Drivers like i915 benefit from being able to control the maxium
> size of the sg coallesced segment while building the scatter-
> gather list.
>
> Introduce and export the
[corrected my email in cc]
On 12/11/2016 02:21, Jeff McGee wrote:
On Wed, Nov 09, 2016 at 11:11:07AM -0800, Anusha Srivatsa wrote:
Replace i915.enable_guc_loading and i915.enable_guc_submission
with a single parameter - i915.enable_guc. Where:
-1 : Platform default (Only load GuC)
0 : Do not
Hi Chris,
[auto build test WARNING on drm/drm-next]
[also build test WARNING on next-20161114]
[cannot apply to v4.9-rc5]
[if your patch is applied to the wrong git tree, please drop us a note to help
improve the system]
url:
https://github.com/0day-ci/linux/commits/Chris-Wilson/dma-buf
On Mon, Nov 14, 2016 at 09:01:00AM +, Tvrtko Ursulin wrote:
>
> On 14/11/2016 08:56, Chris Wilson wrote:
> >Localise the static struct lock_class_key to the caller of
> >i915_sw_fence_init() so that we create a lock_class instance for each
> >unique sw_fence rather than all sw_fences sharing
On 14/11/2016 08:56, Chris Wilson wrote:
Localise the static struct lock_class_key to the caller of
i915_sw_fence_init() so that we create a lock_class instance for each
unique sw_fence rather than all sw_fences sharing the same
lock_class. This eliminate some lockdep false positive when using
Userspace is faced with a dilemma. The kernel requires implicit fencing
to manage resource usage (we always must wait for the GPU to finish
before releasing its PTE) and for third parties. However, userspace may
wish to avoid this serialisation if it is either using explicit fencing
between
The scheduler needs to know the dependencies of each request for the
lifetime of the request, as it may choose to reschedule the requests at
any time and must ensure the dependency tree is not broken. This is in
additional to using the fence to only allow execution after all
dependencies have been
In order to simplify the lockdep annotation, as they become more complex
in the future with deferred execution and multiple paths through the
same functions, create a separate lockclass for the user timeline and
the hardware execution timeline.
We should only ever be locking the user timeline and
Localise the static struct lock_class_key to the caller of
i915_sw_fence_init() so that we create a lock_class instance for each
unique sw_fence rather than all sw_fences sharing the same
lock_class. This eliminate some lockdep false positive when using fences
from within fence callbacks.
101 - 200 of 216 matches
Mail list logo