[Mesa-dev] [PATCH v4 03/11] gallium: add common pipe_screen reference counting functions

[Ilia Mirkin]

On Fri, Jul 22, 2016 at 2:01 PM, Rob Herring <robh at kernel.org> wrote:
> On Fri, Jul 22, 2016 at 11:46 AM, Ilia Mirkin <imirkin at alum.mit.edu> wrote:
>> On Fri, Jul 22, 2016 at 12:22 PM, Rob Herring <robh at 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 <robh at 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 <robh at 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;
>>> +}
>>> +
>>> +struct pipe_screen *
>>> +pipe_screen_reference(int fd)
>>> +{
>>> +   struct pipe_screen *pscreen;
>>> +
>>> +   if (!fd_tab) {
>>> +      fd_tab = util_hash_table_create(hash_fd, compare_fd);
>>
>> Do you need to grab the fd_tab_mutex around this? What if two
>> pipe_screen_reference() calls race to be the first ones?
>
> No, but only because the loader_mutex serializes things. That's not
> obvious though so putting fd_tab_mutex around it would make this
> function more robust.
>
>>> +      return NULL;
>>> +   }
>>> +
>>> +   pipe_mutex_lock(fd_tab_mutex);
>>> +   pscreen = util_hash_table_get(fd_tab, intptr_to_pointer(fd));
>>> +   if (pscreen)
>>> +      pipe_reference(NULL, &pscreen->reference);
>>> +   pipe_mutex_unlock(fd_tab_mutex);
>>> +
>>> +   return pscreen;
>>> +}
>>> +
>>> +boolean
>>> +pipe_screen_unreference(struct pipe_screen *pscreen)
>>> +{
>>> +   boolean destroy;
>>> +
>>> +   if (!pscreen)
>>> +      return FALSE;
>>> +
>>> +   /* Work-around until all pipe_screens have ref counting */
>>> +   if (!pipe_is_referenced(&pscreen->reference)) {
>>> +      pscreen->destroy(pscreen);
>>> +      return TRUE;
>>> +   }
>>> +
>>> +   pipe_mutex_lock(fd_tab_mutex);
>>> +   destroy = pipe_reference(&pscreen->reference, NULL);
>>> +   if (destroy) {
>>> +      pscreen->destroy(pscreen);
>>> +      util_hash_table_remove(fd_tab, intptr_to_pointer(pscreen->fd));
>>> +      close(pscreen->fd);
>>
>> It seems a little odd that you're closing a fd that you didn't
>> open/dup in this library. It's a bit of asymmetry in the calling
>> convention.
>
> Yes, the asymmetry in general compared to the RFC bugs me a bit too.
> The flip side is duplicating the close in all the destroy() functions
> seems pointless.
>
> Also, I just noticed I have a use after free problem here. The
> destroy() call needs to be last.

Yeah, so it actually occurs to me that at least nouveau was most
likely leaking its dup'd fd in the current code. One issue is that the
fd has to be alive until it is removed from the fd table, otherwise
we'll have fstat issues.

Anyways, this is fine as-is. And yes, you need to flip the screen
destroy order. Although closing the fd before calling destroy could
also cause annoyance - Perhaps save off the fd before calling destroy,
and then close it?

  -ilia