On Mon, Jan 9, 2017 at 7:38 PM, Emil Velikov <emil.l.veli...@gmail.com> wrote:
> On 28 July 2016 at 13:45, Marek Olšák <mar...@gmail.com> wrote:
>> On Fri, Jul 22, 2016 at 6:22 PM, Rob Herring <r...@kernel.org> wrote:
>>> In order to prevent multiple pipe_screens being created in the same
>>> process, lookup of the DRM FD and reference counting of the pipe_screen
>>> are needed. Several implementations of this exist in various gallium
>>> drivers/winsys already. This creates a common version which is opt-in
>>> for winsys implementations.
>>>
>>> Signed-off-by: Rob Herring <r...@kernel.org>
>>> ---
>>> src/gallium/auxiliary/Makefile.sources | 2 +
>>> src/gallium/auxiliary/util/u_screen.c | 114
>>> +++++++++++++++++++++++++++++++++
>>> src/gallium/auxiliary/util/u_screen.h | 32 +++++++++
>>> src/gallium/include/pipe/p_screen.h | 3 +
>>> 4 files changed, 151 insertions(+)
>>> create mode 100644 src/gallium/auxiliary/util/u_screen.c
>>> create mode 100644 src/gallium/auxiliary/util/u_screen.h
>>>
>>> diff --git a/src/gallium/auxiliary/Makefile.sources
>>> b/src/gallium/auxiliary/Makefile.sources
>>> index e0311bf..197ed36 100644
>>> --- a/src/gallium/auxiliary/Makefile.sources
>>> +++ b/src/gallium/auxiliary/Makefile.sources
>>> @@ -284,6 +284,8 @@ C_SOURCES := \
>>> util/u_ringbuffer.h \
>>> util/u_sampler.c \
>>> util/u_sampler.h \
>>> + util/u_screen.c \
>>> + util/u_screen.h \
>>> util/u_simple_shaders.c \
>>> util/u_simple_shaders.h \
>>> util/u_slab.c \
>>> diff --git a/src/gallium/auxiliary/util/u_screen.c
>>> b/src/gallium/auxiliary/util/u_screen.c
>>> new file mode 100644
>>> index 0000000..47bad11
>>> --- /dev/null
>>> +++ b/src/gallium/auxiliary/util/u_screen.c
>>> @@ -0,0 +1,114 @@
>>> +/*
>>> + * Copyright 2016 Linaro, Ltd., Rob Herring <r...@kernel.org>
>>> + *
>>> + * Permission is hereby granted, free of charge, to any person obtaining a
>>> + * copy of this software and associated documentation files (the
>>> + * "Software"), to deal in the Software without restriction, including
>>> + * without limitation the rights to use, copy, modify, merge, publish,
>>> + * distribute, sub license, and/or sell copies of the Software, and to
>>> + * permit persons to whom the Software is furnished to do so, subject to
>>> + * the following conditions:
>>> + *
>>> + * The above copyright notice and this permission notice (including the
>>> + * next paragraph) shall be included in all copies or substantial portions
>>> + * of the Software.
>>> + *
>>> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
>>> + * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
>>> + * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
>>> + * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
>>> + * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
>>> + * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
>>> + * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
>>> + */
>>> +
>>> +/**
>>> + * Functions for managing pipe_screen's
>>> + */
>>> +
>>> +#include <sys/stat.h>
>>> +
>>> +#include "os/os_thread.h"
>>> +
>>> +#include "pipe/p_screen.h"
>>> +#include "util/u_hash_table.h"
>>> +#include "util/u_inlines.h"
>>> +#include "util/u_pointer.h"
>>> +#include "util/u_screen.h"
>>> +
>>> +static struct util_hash_table *fd_tab = NULL;
>>> +pipe_static_mutex(fd_tab_mutex);
>>> +
>>> +static unsigned hash_fd(void *key)
>>> +{
>>> + int fd = pointer_to_intptr(key);
>>> + struct stat stat;
>>> + fstat(fd, &stat);
>>> +
>>> + return stat.st_dev ^ stat.st_ino ^ stat.st_rdev;
>>> +}
>>> +
>>> +static int compare_fd(void *key1, void *key2)
>>> +{
>>> + int fd1 = pointer_to_intptr(key1);
>>> + int fd2 = pointer_to_intptr(key2);
>>> + struct stat stat1, stat2;
>>> + fstat(fd1, &stat1);
>>> + fstat(fd2, &stat2);
>>> +
>>> + return stat1.st_dev != stat2.st_dev ||
>>> + stat1.st_ino != stat2.st_ino ||
>>> + stat1.st_rdev != stat2.st_rdev;
>>> +}
>>
>> I think fstat won't work for amdgpu. We moved away from it some time
>> ago, though it's not documented why. libdrm_amdgpu uses
>> drmGetPrimaryDeviceNameFromFd instead, so I think Mesa should too.
>>
> libdrm_amdgpu uses a different scheme than others where it allows
> sharing between card and renderD.
> It stores a reference to both nodes (if applicable) does not nessesary
> acrobatics, if needed, in order to present a single/unique
> amdgpu_device_handle per actual HW device.
>
> Whether that's possible/good idea for others is to be confirmed, so
> let's keep that as follow-up work ?
I wouldn't like to change the behavior for amdgpu. A possible
workaround is to let drivers supply the hash table key. Then, amdgpu
can use the device handle and other drivers can use fd.
Marek
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
