| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
| |
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
|
|
|
|
| |
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
|
|
|
|
| |
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
|
|
|
|
|
|
|
| |
Also keep the output in the rnndb directory to make it easier
to find when moving things to other repos.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
|
|
|
|
| |
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
|
|
|
|
| |
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
|
|\
| |
| | |
Add apalis headers
|
|/ |
|
|
|
|
| |
Russell King discovered these.
|
|\
| |
| | |
add CHIP_SPECS and CHIP_SPECS_2 bitfields
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
|/
|
|
|
|
|
| |
The strucutre of the those two registers was found here:
https://github.com/gcwnow/linux/blob/jz-3.12/drivers/gpu/vivante/v4/gc_hal_kernel_hardware.c#L46
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
|
|
|
|
|
| |
We need to make sure to unmap all memory before we close the fd
to galcore. This allows the kernel driver to call drv_release
and clean up after the process exits.
|
|
|
|
|
| |
Make sure that we tell the kernel to clean up the HAL context
before we close the connection.
|
|
|
|
|
|
|
|
|
| |
Right now the kernel will complain when etna_bo_del
is called without a queue and the unlock is delayed (async=1 is
returned), as the free happens immediately.
So make sure that FREE_VIDEO_MEMORY is sent as event even when
no queue is passed.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The standard galcore user memory mapping function fails when asked to
map read-only memory, such as X shmem buffers. The reason these fail
is because get_user_pages() is always attempted with "write" mode
enabled, which is not permissible with read-only mappings.
To allow this to work (so that Xv can be implemented efficiently) we
must be able to map read-only buffers. This adds an extension to
etnaviv, found in my Dove galcore driver and Jon Nettleton's iMX6
kernels which permit this without changing the existing galcore APIs.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
| |
dmabufs are DMA buffers which can be passed between kernel drivers
under the control of userspace by means of a file descriptor. Add
support to import dmabufs into etnaviv, creating an etna_bo for
them which can be released using the normal methods.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
|
| |
The 4.6.9 iMX GCABI has a .physical member added to the MapUserMemory
structure, which must be initialised to ~0 to map by virtual address.
Introduce a new conditional "GCABI_MAPUSERMEMORY_HAS_PHYSICAL" to
indicate the presence of this member and arrange to initialise it
appropriately in viv_map_user_memory().
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
| |
Add header files and etnaviv configuration for the 4.6.9 iMX galcore
version, found in (at least) Jon Nettleton's 3.14.14 kernel.
Signed-off-by: Jon Nettleton <jon.nettleton@gmail.com>
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Jon Nettleton's kernel galcore driver can return EINTR while waiting
for user signals. With this galcore version, if we are waiting on a
command buffer galcore signal (to indicate that the buffer is now
available for re-use), and we are sent a user signal (eg, SIGALRM),
then ioctl() may return EINTR.
This causes etnaviv to believe that the galcore signal was received,
and it is safe to proceed with the next command buffer. In reality,
this can cause the GPU to scribble over memory which it should not,
and can cause the entire platform to lock up.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
| |
Add a viv_ioctl() interface to provide an etnaviv version of
gcoOS_DeviceControl(). This allows us to temporarily extend galcore
with features that it otherwise would not have (such as my DMA buf
import, and support for user memory mapping of regions marked
read-only.)
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
| |
etna_create() allocates six signals, one for the context and one per
allocated command buffer. etna_free() cleans up the command buffers,
but fails to clean up the signals which etna_create() allocated. Fix
this oversight.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
| |
I don't know if this will have any effect but the check was clearly
the wrong way around.
Thanks to imirkin on IRC for noticing this.
|
|
|
|
|
|
|
|
|
| |
We must ensure that the subdirectories are built in the correct
order, otherwise the link fails. Disable parallel make in the
top level makefile, so that only the lower level makefiles
benefit from parallel make.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
| |
It is not very friendly for make -n to fail. Fix this.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
| |
Rather than printing command buffers as one word per line, print 8
words per line. This allows a more condensed view of the command
stream.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Avoid writing to stdout; depending on the program using etnaviv, stdout
may not be the console, and, in the case of programs such as Xorg, may
not be something which should receive any output.
As the C library uses stderr internally for assert(), it is reasonable
to expect that this will be suitable for debug output. Therefore,
direct etnaviv debugging output to stderr.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
|
| |
Testing on Dove reveals that the context buffers are leaked each time
Etnaviv is closed down. This is caused by the ifdef which surrounds
the call to enta_bo_del() always being false, and so the code to free
this buffer is never compiled.
Fix the ifdef, and also fix the code so that it builds.
Signed-off-by: Russell King <rmk@arm.linux.org.uk>
|
|
|
|
| |
Much easier to read than the others.
|
| |
|
|\
| |
| | |
Fix fbdemos segmentation fault when using single-buffered fbdev devices
|
|/ |
|
|\
| |
| | |
add bitfield descriptions for CHIP_IDENTITY
|
|/
|
|
| |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| | |
When detiling, the RS requires dimensions for the window
size that are a multiple of 4. With two pixel pipes this
amounts to 8.
Hence pad the height of the output buffers to the next multiple
of 8.
|
| |
| |
| |
| | |
Suggested by @kneo and @stefan-langenmaier at about the same time.
|
|/
|
|
| |
Revert behavior in case of one pixel pipe.
|
|\
| |
| | |
mesatest_gles fixes
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
|/
|
|
| |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
|\
| |
| | |
GC2000 works
|
|/
|
|
| |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
|\
| |
| | |
some gc2000 related stuff
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
| |
| |
| |
| |
| |
| |
| |
| | |
With this commit the PE pipe addresses and the
RS pipe addresses are the same --> on a GC2000
the image is combined correctly.
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|
| |
| |
| |
| | |
Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
|