correctly.
--
You are receiving this mail because:
You are the assignee for the bug.
-- next part --
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/dri-devel/attachments/20150402/665d0259/attachment.html>
)
libvorbis.so.0 => /usr/lib/libvorbis.so.0 (0x7f5b85094000)
libnsl.so.1 => /usr/lib/libnsl.so.1 (0x7f5b84e7c000)
libattr.so.1 => /usr/lib/libattr.so.1 (0x7f5b84c77000)
--
You are receiving this mail because:
You are the assignee for the bug.
-- next part --
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/dri-devel/attachments/20150402/17c986d4/attachment-0001.html>
On 2 April 2015 at 19:07, wrote:
> Thanks Emil. Please check the comments embedded and for the rest I will
> update the code.
>
>> Hi Jilai,
>>
>> Just a few questions, not really a review as I'm not that familiar
>> with the code.
>>
>>
>>> +++ b/drivers/gpu/drm/msm/Kconfig
>>> @@ -27,6 +27,16 @
---- next part --
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/dri-devel/attachments/20150402/9b6d9867/attachment.html>
Hi Dave,
Fix display on issue to Exynos5250 based Snow(1366x768) board.
Please kindly let me know if there is any problem.
Thanks,
Inki Dae
The following changes since commit 91fd89660ba2e8ee59a587294fa9b17761691b05:
Merge branch 'drm-fixes-4.0' of git://people.freedesktop.org/~agd5f/l
LANE;
> >> + q->io_modes = VB2_DMABUF;
> >> + q->drv_priv = dev;
> >> + q->buf_struct_size = sizeof(struct msm_wb_v4l2_buffer);
> >> + q->ops = &msm_wb_vb2_ops;
> >> + q->mem_ops = &msm_wb_vb2_mem_ops;
> >> + q-
Am Donnerstag, den 02.04.2015, 12:14 -0500 schrieb Robert Nelson:
> On Thu, Apr 2, 2015 at 10:29 AM, Lucas Stach
> wrote:
> > IOMMUv2 support isn't implemented yet, so don't pretend it is there.
> >
> > Signed-off-by: Lucas Stach
> > ---
> > drivers/staging/etnaviv/etnaviv_gpu.c | 10 +
Hi Dave,
The following changes since commit c517d838eb7d07bbe9507871fab3931deccff539:
Linux 4.0-rc1 (2015-02-22 18:21:14 -0800)
are available in the git repository at:
git://anongit.freedesktop.org/tegra/linux tags/drm/panel/for-4.1-rc1
for you to fetch changes up to 1c550fa193d08920255240
From: Fabio Estevam
Staticize dw_hdmi_bridge_funcs to fix the following sparse warning:
drivers/gpu/drm/bridge/dw_hdmi.c:1458:25: warning: symbol
'dw_hdmi_bridge_funcs' was not declared. Should it be static?
Signed-off-by: Fabio Estevam
---
drivers/gpu/drm/bridge/dw_hdmi.c | 2 +-
1 file cha
Hi Dave,
The following changes since commit 13a7a6ac0a11197edcd0f756a035f472b42cdf8b:
Linux 4.0-rc2 (2015-03-03 09:04:59 -0800)
are available in the git repository at:
git://anongit.freedesktop.org/tegra/linux tags/drm/tegra/for-4.1-rc1
for you to fetch changes up to 535a65db484ff4e4654a03
intf() into just "msm_wb".
>>> >
>>> >> + ret = v4l2_device_register(NULL, &dev->v4l2_dev);
>>> >> + if (ret)
>>> >> + goto free_dev;
>>> >> +
>>> >> + /* default ARGB
intf() into just "msm_wb".
>>> >
>>> >> + ret = v4l2_device_register(NULL, &dev->v4l2_dev);
>>> >> + if (ret)
>>> >> + goto free_dev;
>>> >> +
>>> >> + /* default ARGB
Because the helper function which calls this callback checks whether
it is registered or not. It is not necessary if it does nothing.
So it would be better to remove the function for clarity.
Signed-off-by: Hyungwon Hwang
---
drivers/gpu/drm/exynos/exynos_drm_dsi.c | 7 ---
1 file changed, 7
Am Donnerstag, den 02.04.2015, 17:45 +0100 schrieb Russell King - ARM
Linux:
> On Thu, Apr 02, 2015 at 06:29:24PM +0200, Lucas Stach wrote:
> > The start of the commands must always be 64bit aligned, it's the same
> > for all pipes. The size can be dword aligned if the last command in the
> > strea
Am Donnerstag, den 02.04.2015, 17:22 +0100 schrieb Russell King - ARM
Linux:
> On Thu, Apr 02, 2015 at 05:30:32PM +0200, Lucas Stach wrote:
> > The intention clearly was to do the same thing for WC and UC buffers,
> > not for cached ones.
>
> Err, from one of my previous commits:
>
> staging:
Am Donnerstag, den 02.04.2015, 17:20 +0100 schrieb Russell King - ARM
Linux:
> On Thu, Apr 02, 2015 at 05:30:29PM +0200, Lucas Stach wrote:
> > It is legal for the userspace to pass in a command stream of a size
> > aligned to 32 bit, if that is where the last user command ends. The
> > kernel then
Thanks Emil. Please check the comments embedded and for the rest I will
update the code.
> Hi Jilai,
>
> Just a few questions, not really a review as I'm not that familiar
> with the code.
>
>
>> +++ b/drivers/gpu/drm/msm/Kconfig
>> @@ -27,6 +27,16 @@ config DRM_MSM_FBDEV
>> support. Not
Hi,
On 03/31/2015 04:11 AM, Gustavo Padovan wrote:
> From: Gustavo Padovan
>
> Run dpms operations through the atomic intefaces. This basically removes
> the .dpms() callback from econders and crtcs and use .disable() and
> .enable() to turn the crtc on and off.
>
> Signed-off-by: Gustavo Padov
Am Donnerstag, den 02.04.2015, 16:43 +0100 schrieb Russell King - ARM
Linux:
> On Thu, Apr 02, 2015 at 05:29:02PM +0200, Lucas Stach wrote:
> > Hey all,
> >
> > this is the Etnaviv DRM driver for Vivante embedded GPUs. It is heavily
> > influenced by the MSM driver, as can be clearly seen with the
Thanks Paul. Some comments embedded and for the rest I will update the
code accordingly.
> A few nits follow.
>
> On Wed, 2015-04-01 at 17:12 -0400, Jilai Wang wrote:
>> --- a/drivers/gpu/drm/msm/Kconfig
>> +++ b/drivers/gpu/drm/msm/Kconfig
>
>> +config DRM_MSM_WB
>> +bool "Enable writeback su
should be changed:
+ BUG_ON(size >= MAX_INLINE_CMD_SIZE);
--
Marc-André Lureau
-- next part --
An HTML attachment was scrubbed...
URL:
<http://lists.freedesktop.org/archives/dri-devel/attachments/20150402/8b8205a9/attachment-0001.html>
Add HDMI HDCP support including HDCP PartI/II/III authentication.
V1: Initial Change
V2: Address Bjorn&Rob's comments
Refactor the authentication process to use single work instead
of multiple work for different authentication stages.
V3: Update to align with qcom SCM api.
Signed-off-by: J
On Thu, Apr 02, 2015 at 06:29:24PM +0200, Lucas Stach wrote:
> The start of the commands must always be 64bit aligned, it's the same
> for all pipes. The size can be dword aligned if the last command in the
> stream is something like SET_STATE with a length of 2. In that case one
> needs to insert
Hi,
On 03/28/2015 12:58 AM, Gustavo Padovan wrote:
> From: Gustavo Padovan
>
> Now that no one is using the functions exported by exynos_drm_plane due
> to the atomic conversion we can make remove some of the them or make them
> static.
>
> Signed-off-by: Gustavo Padovan
> ---
> drivers/gpu/d
On Thu, Apr 02, 2015 at 05:30:44PM +0200, Lucas Stach wrote:
> While this isn't the case on i.MX6 a single GPU pipe can have
> multiple rendering backend states, which can be selected by the
> pipe switch command, so there is no strict mapping between the
> user "pipes" and the PIPE_2D/PIPE_3D exec
This adds the device nodes for 2D, 3D and VG GPU cores.
Signed-off-by: Lucas Stach
---
arch/arm/boot/dts/imx6dl.dtsi | 5 +
arch/arm/boot/dts/imx6q.dtsi | 14 ++
arch/arm/boot/dts/imx6qdl.dtsi | 19 +++
3 files changed, 38 insertions(+)
diff --git a/arch/arm/
- correct license to the proper GPLv2
- add correct author names
- remove double MODULE_DEVICE_TABLE
- update driver date
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_drv.c | 13 +++--
1 file changed, 7 insertions(+), 6 deletions(-)
diff --git a/drivers/staging/etnaviv
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_buffer.c | 2 +-
drivers/staging/etnaviv/etnaviv_cmd_parser.c | 16
drivers/staging/etnaviv/etnaviv_drv.c| 3 +--
drivers/staging/etnaviv/etnaviv_drv.h| 3 +--
drivers/staging/etnaviv/etnaviv_ge
From: Christian Gmeiner
There is no need to spam the kernel logs with the GPU specs and features
at startup. If someone wants to know about this stuff debugfs should be
the right place to look at.
Also use better format specifiers to make it easier for humans to read.
Signed-off-by: Christian G
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_drv.c| 6 +++---
drivers/staging/etnaviv/etnaviv_drv.h| 7 +++
drivers/staging/etnaviv/etnaviv_gem.c| 8
drivers/staging/etnaviv/etnaviv_gem_submit.c | 2 +-
drivers/staging/etnaviv/etnaviv_gpu.c
This is a simple rename without functional changes.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_drv.c | 10 +-
drivers/staging/etnaviv/etnaviv_drv.h | 12 ++--
drivers/staging/etnaviv/etnaviv_gem.c | 2 +-
drivers/staging/etnaviv/etnaviv_gem_
Dumb buffers must be only used as backing storage for scanout
only surfaces. Any acceleration operation on them is not allowed.
So there is no point in having dumb buffer support in a driver
that isn't able to drive any scanout hardware.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/et
Drop the last remaining MSM bits and things we don't need for
Vivante GPUs. Those include shifting and or-ing of reloc addresses
and IB buffers.
Signed-off-by: Russell King
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gem_submit.c | 9 +
include/uapi/drm/etnaviv_drm.h
In case the component bind fails the mutex should not be left locked.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_drv.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/staging/etnaviv/etnaviv_drv.c
b/drivers/staging/etnaviv/etnaviv_drv.c
index 8
While this isn't the case on i.MX6 a single GPU pipe can have
multiple rendering backend states, which can be selected by the
pipe switch command, so there is no strict mapping between the
user "pipes" and the PIPE_2D/PIPE_3D execution states.
We need to respect this in the public userspace API. T
The platform should specify the appropriate IRQ flags and it's
a really bad idea to override them in individual drivers.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gpu.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gp
The GPU cores are possibly scattered in the SoC address space, so the
current abstraction of having a parent node for the master device and
the cores as child nodes doesn't fit too well.
Instead take the same approach as with imx-drm to have a logical master
node that refers to the other component
At least the GC2000 I'm testing with seems to have a bug that all vertex
streams have to be mapped either through the MMU or without it. Mixing
between both mapping types in a single draw command results in corrupted
vertex data.
As we can not quarantee that a buffer may be mappable without the MM
The MMU needs to be flushed when changing the render context to get
rid of stale TLB entries left behind by the last context.
While we do not support context switching between different processes yet
this commit fixes memory corruptions seen when executing different 3D
applications one after anoth
The MMU is per GPU (pipe), rather than per DRM device, so it makes
a lot more sense to use the GPU device instead of the DRM device for
the MMU to hang off from.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gpu.c | 2 +-
drivers/staging/etnaviv/etnaviv_mmu.c | 2 +-
drivers/sta
Each pipe has it's own MMU, so there is no point in pretending
to have a single one at the DRM driver level. All MMU management has
to happen on a per-pipe level.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_drv.c | 7 ---
drivers/staging/etnaviv/etnaviv_drv.h | 5 -
dr
As single buffer object may be mapped into different address spaces at the
same time. For now we only have two different address spaces for the 3D and
2D pipe, but this may change as soon as we implement per-process page tables.
Allow this by having each buffer object manage a list of all it's
map
Allows userspace to properly synchronize with the GPU when accessing
buffers.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gem.c | 18 +-
1 file changed, 9 insertions(+), 9 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gem.c
b/drivers/staging/etnav
Other users may only have the gpu pointer available and it's easy to convert
from pipe to gpu in the ioctl path.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_drv.c | 24 +---
drivers/staging/etnaviv/etnaviv_drv.h | 5 +++--
2 files changed, 16 insertions(+)
This provides a bit more type safety.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gem.h | 7 ++-
1 file changed, 6 insertions(+), 1 deletion(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gem.h
b/drivers/staging/etnaviv/etnaviv_gem.h
index cfade337d4db..fadd5198b3e8 100
Avoids memory corruptions seen due to stale TLB entries.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_buffer.c | 5 -
1 file changed, 4 insertions(+), 1 deletion(-)
diff --git a/drivers/staging/etnaviv/etnaviv_buffer.c
b/drivers/staging/etnaviv/etnaviv_buffer.c
index 391b
The intention clearly was to do the same thing for WC and UC buffers,
not for cached ones.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gem.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gem.c
b/drivers/staging/etnaviv/etn
This is taken from the Vivante kernel driver and seems to improve
stability and performance.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gpu.c | 10 ++
1 file changed, 10 insertions(+)
diff --git a/drivers/staging/etnaviv/etnaviv_gpu.c
b/drivers/staging/etnaviv/etnav
These two GPUs seems to be in the transition state between two
generations and report a wrong instuction count. Fix it up in the
kernel like the Vivante driver does.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_gpu.c | 7 ++-
1 file changed, 6 insertions(+), 1 deletion(-)
It is legal for the userspace to pass in a command stream of a size
aligned to 32 bit, if that is where the last user command ends. The
kernel then needs to insert a LINK command at the end of the stream,
which needs to be aligned to 64 bit, so the kernel may insert an
additional 32bits of padding
The hardware interprets a value of 0 as the maximium number of rectangles
which is 256. Allow this in the command parser.
Signed-off-by: Lucas Stach
---
drivers/staging/etnaviv/etnaviv_cmd_parser.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/staging/etnaviv/etnaviv_cmd_parser.c
From: Russell King
On iMX6, memory starts at 256MB physical, and if we have 2GB of memory,
this causes it to extend beyond 0x8000. All memory is available for
DMA coherent allocations, which can result in the command buffers being
allocated from addresses above 0x8000.
However, the Viva
From: Russell King
Add initial runtime PM support. We manage the runtime PM state based
on the events: when we allocate an event upon command submission, we
"get" the runtime PM state, causing it to resume if not already resumed.
When we receive an interrupt, and free an event, we "put" the run
From: Russell King
Add support to append an END command to the GPU command stream so we
cleanly shutdown the GPU on driver removal and suspend. This properly
quiesces the GPU, allowing it to come back up on SoCs such as iMX6.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_buf
From: Russell King
We need to reprogram various registers when coming out of runtime PM,
many of which are those which are setup by the main initialisation.
Abstract this code out and arrange for the runtime PM resume method
to call it.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/e
From: Russell King
Power management of each GPU core should be a matter for the driver of
each GPU core itself. Move the basic power management calls out of
etnaviv_drv.c and into etnaviv_gpu.c so that we can add runtime PM.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.
From: Russell King
Hold the mutex while calling component_{un,}bind_all() so that
components can perform initialisation in their bind and unbind
callbacks from the component helper.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.c | 11 ++-
drivers/staging/etnaviv/
From: Russell King
The etnaviv_gpu structure can have a longer lifetime than the GPU
command buffer, MMU and drm_device structures. When these other
structures are freed (via the unbind method) we may be tempted to
access these via other functions after they've been freed. Leaving
pointers in t
From: Russell King
Remove the etnaviv specific power rail support, which is mostly disabled
anyway. This really wants to be using the power domain support instead,
which allows the SoC specifics to be abstracted from the driver, rather
than a home-cooked version of it.
For example, on Dove, it
From: Russell King
If we fail to allocate an event, we leave the submitted fence number
incremented. This can cause an already running hangcheck timer to
believe that we should be waiting for further events when no event has
actually been queued.
Resolve this by moving the fence allocation (whi
From: Russell King
GC600 does not set busy bits in its idle register for modules which are
not present. Add handling to ensure that we don't misinterpret these
zero bits as indicating that these modules are busy.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gpu.c | 15 +
From: Russell King
There is no need to restrict access to the get_param ioctl; this is
used to obtain information about the device and has no side effects.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/
From: Russell King
Parse the submitted command buffer for allowable GPU commands, and
validate that all commands fit wholely within the submitted buffer.
We allow the following commands:
- load state
- any of the draw commands
- stall
- nop
which denies attempts to link, call, return, etc from
From: Russell King
Add a poisoned bad page to map unused MMU entries to. This gives us a
certain amount of protection against bad command streams causing us to
hit regions of memory we really shouldn't be touching.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_iommu.c | 29 +
From: Russell King
The DMA API usage by etnaviv is really not up to scratch. It does not
respect the buffer ownership rules, which are vitally necessary when
using DMA_BIDIRECTIONAL, as mapping /only/ cleans the cache lines,
causing dirty data to be written back to RAM and an unmap /only/
invali
From: Russell King
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_mmu.c | 11 ---
1 file changed, 4 insertions(+), 7 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_mmu.c
b/drivers/staging/etnaviv/etnaviv_mmu.c
index 5647768c2be4..0a2f9dd5bb7a 100644
--- a/d
From: Russell King
We never pass the GPU addresses of BOs to userspace, so userspace can
never specify the correct address. Hence, this code serves no useful
purpose, and can be removed.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.h| 1 -
drivers/staging/etnav
From: Russell King
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 89 ---
1 file changed, 52 insertions(+), 37 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gem.c
b/drivers/staging/etnaviv/etnaviv_gem.c
index 2f2bf5619ffd..8
From: Russell King
Gem objects compared fences using simple <= tests. This is
problematical when the fences wrap past (uint32_t)~0. Resolve this
by using our helpers.
However, this is complicated by the use of '0' to indicate that the
fence has not been set (eg, because we are not writing to a
From: Russell King
Use the etnaviv definitions for feature constants, rather than BIT()s.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gpu.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gpu.c
b/drivers/staging/etna
From: Russell King
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.c | 19 +++
1 file changed, 19 insertions(+)
diff --git a/drivers/staging/etnaviv/etnaviv_drv.c
b/drivers/staging/etnaviv/etnaviv_drv.c
index 568615154845..77e05b80f18d 100644
--- a/drivers/
From: Russell King
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.c | 50 ---
drivers/staging/etnaviv/etnaviv_gpu.c | 18 ++---
2 files changed, 42 insertions(+), 26 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_drv.c
b/
From: Russell King
Increase the iommu page table size to 512KiB so we can map a maximum of
512MiB of memory. GC600 (at least) seems happy with this change.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_iommu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git
From: Russell King
Add the workarounds found in the GALCORE code (and found to be required)
for the GC320 2D core found on iMX6 to etnaviv.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gpu.c | 24
1 file changed, 24 insertions(+)
diff --git a/driver
From: Russell King
Rather than waiting indefinitely for the GPU to reset, bound this and
if it doesn't appear to be successful, bail out and report why we
failed.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gpu.c | 32 +++-
1 file changed, 27 ins
From: Russell King
If we queue up multiple buffers, each with their own event, where the
first buffer takes a while to execute, but subsequent buffers do not,
we can end up receiving multiple events simultaneously. (eg, 0, 1, 2).
In this case, we only look at event 2, which updates the last fen
From: Russell King
If we queue up a large command buffer (32K) containing about 164 1080p
blit operations, it can take the GPU several seconds to complete before
raising the next event. Our existing hangcheck code decides after a
second that the GPU is stuck, and provokes a retirement of the eve
From: Russell King
Provide a function to safely take down the hangcheck timer and
workqueue.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gpu.c | 10 +++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/drivers/staging/etnaviv/etnaviv_gpu.c
b/drivers/sta
From: Russell King
We need to synchronously take down the hangcheck timer, and then cancel
the recovery work when we're unbinding the GPU to avoid these timers and
workers running after we clean up.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gpu.c | 6 --
1 file change
From: Russell King
Report messages against the component device rather than the subsystem
device, so that the component responsible for the message is identified
rather than having to be separately formatted in the string. This
ensures that many of the debug messages are properly attributed to t
From: Russell King
Call the DRM device 'drm' in the etnaviv_gpu structure, so that we can
add a struct device pointer.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_buffer.c | 2 +-
drivers/staging/etnaviv/etnaviv_gpu.c| 74
drivers/stagi
From: Russell King
Add support for mapping userspace memory to the GPU. This is useful for
cases where we have some malloc()'d memory which we wish to pass to the
GPU or shmem memory via Xv, and wish to avoid the overhead of an
additional memcpy(), especially as is is expensive to memcpy()ing a
From: Russell King
Move the scatterlist creation from etnaviv_gem_shmem_get_pages() into
etnaviv_gem_get_pages() as we always want a scatterlist internally for
the IOMMU code. It makes little sense to have each get_pages() method
re-implement this code.
However, we still allow a get_pages() met
From: Russell King
We can easily exhaust the MMU space since we leave mappings in place
until the underlying buffers are freed.
Solve this by reaping inactive MMU entries using the drm_mm scanning
facility to select candidate(s), which will then have their MMU
mappings released.
Signed-off-by:
From: Russell King
If etnaviv_iommu_map() fails, we returned an error, but we didn't clean
up the allocated drm_mm node. Simplify the return path and add the
necessary failure clean up.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_mmu.c | 28 +---
1
From: Russell King
In order to avoid flushing the GPU MMU every time we unmap and remap,
allocate MMU addresses in a round-robin fashion. When we have to wrap
back to the beginning, indicate that the MMU needs to be flushed.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_mmu.
From: Russell King
Bypass the iommu when we are dealing with single-entry scatterlists.
The etnaviv iommu code needs to be more inteligent: as it currently
stands, it is unusable as it always allocates from the bottom upwards.
This causes entries to be re-used without the MMU TLB being flushed,
From: Russell King
Move the code which sets up and tears down the MMU mappings (iow,
allocates a node in the drm_mm, and then calls the iommu to setup
the actual mapping, and the reverse) into etnaviv_mmu.c
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 33
From: Russell King
Ensure that we unmap all MMU entries when unmapping a region. We
fail to do this because we assume that the return value from the
unmap method should be zero. It should be the size of entry which
has been unmapped.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/et
From: Russell King
We model the GPU MMU using the iommu layer, which supports exporting
the iommu domain geometry. Use this feature to publish the size of
the MMU window, and initialise the MMU drm_mm object according to
the available MMU window size.
As we only allocate a MMU page table which
From: Russell King
The GPU memory management (managed by a drm_mm object) is used to
track which areas of the MMU address space are in-use. Therefore,
this should be tied to the MMU object, rather than the GPU object.
This means we could (as comments suggest) have multiple MMU objects,
one for
From: Russell King
Add a flag to indicate that the GPU MMU needs to be flushed before
executing the next set of command buffers. This is necessary to
ensure that the GPU sees updated page table entries which may have
been modified by GEM.
It is expected that userspace will have flushed the cach
From: Russell King
We test for write-combine and non-cacheable mappings before calling the
DMA API. This is werid, because non-cacheable mappings are DMA coherent
by definition, whereas cacheable mappings need cache maintanence
provided by the DMA API.
This seems to be a typo: ETNA_BO_CACHED sh
From: Russell King
Provide a get_pages() method for gem objects, which allows our objects
to provide their own method to obtain the struct page array and
scatterlist.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 67 -
drivers/staging
From: Russell King
Move the locking into etnaviv_gem_prime.c and pass an etnaviv_gem_object
rather than drm_gem_object. As this becomes an internal gem function,
move the prototype into etnaviv_gem.h.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.h | 2 --
drivers
From: Russell King
Convert the internal get_pages()/put_pages() functions to take an
etnaviv_obj rather tha converting between drm_gem_object and our
private one.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 33 -
1 file changed, 16 in
From: Russell King
Clean up the etnaviv prime import handling by combining msm_gem_import()
and msm_gem_prime_import_sg_table(), and then giving it an etnaviv_
prefix.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_drv.c | 2 +-
drivers/staging/etnaviv/etnaviv_drv.h
From: Russell King
Move the prime import code out into etnaviv_gem_prime.c, which keeps
all this functionality together.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 55 -
drivers/staging/etnaviv/etnaviv_gem_prime.c | 55
From: Russell King
etnaviv_gem_new_private() creates a private non-shmem gem object which
can be used to create these kinds of objects. Fix up msm_gem_import()
to use it.
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 32 ++--
drivers/stagi
From: Russell King
As the tail of etnaviv_free_obj() is identical to etnaviv_free_cmd(),
we can eliminate etnaviv_free_obj() entirely by moving it into
etnaviv_gem_free_object().
Signed-off-by: Russell King
---
drivers/staging/etnaviv/etnaviv_gem.c | 19 +++
1 file changed, 3 i
1 - 100 of 202 matches
Mail list logo
