[PATCH libdrm] add libsync.h helper

Gustavo Padovan gustavo at padovan.org
Mon Oct 31 15:13:00 UTC 2016


2016-10-31 Rob Clark <robdclark at gmail.com>:

> On Mon, Oct 31, 2016 at 9:55 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> > On Mon, Oct 31, 2016 at 09:44:07AM -0400, Rob Clark wrote:
> >> From: Rob Clark <robclark at freedesktop.org>
> >>
> >> Rather than cut/pasting these couple ioctl wrappers everywhere, just
> >> stuff them as static-inline into a header.
> >>
> >> Signed-off-by: Rob Clark <robclark at freedesktop.org>
> >> ---
> >> This is probably mostly used from mesa, but some drivers, test apps, etc
> >> may also want to use it from libdrm.
> >>
> >>  Makefile.sources |  1 +
> >>  libsync.h        | 74 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >>  2 files changed, 75 insertions(+)
> >>  create mode 100644 libsync.h
> >>
> >> diff --git a/Makefile.sources b/Makefile.sources
> >> index a57036a..10aa1d0 100644
> >> --- a/Makefile.sources
> >> +++ b/Makefile.sources
> >> @@ -13,6 +13,7 @@ LIBDRM_FILES := \
> >>       util_math.h
> >>
> >>  LIBDRM_H_FILES := \
> >> +     libsync.h \
> >>       xf86drm.h \
> >>       xf86drmMode.h
> >>
> >> diff --git a/libsync.h b/libsync.h
> >> new file mode 100644
> >> index 0000000..fc23b7f
> >> --- /dev/null
> >> +++ b/libsync.h
> >> @@ -0,0 +1,74 @@
> >> +/*
> >> + *  sync abstraction
> >> + *  Copyright 2015-2016 Collabora Ltd.
> >> + *
> >> + *  Based on the implementation from the Android Open Source Project,
> >> + *
> >> + *  Copyright 2012 Google, Inc
> >> + *
> >> + *  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, sublicense,
> >> + *  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 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 NONINFRINGEMENT. IN NO EVENT SHALL
> >> + *  THE AUTHORS OR COPYRIGHT HOLDERS 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.
> >> + */
> >> +
> >> +#ifndef _LIBSYNC_H
> >> +#define _LIBSYNC_H
> >> +
> >> +#include <stdint.h>
> >> +#include <string.h>
> >> +#include <sys/ioctl.h>
> >> +#include <sys/poll.h>
> >> +
> >> +// todo prob should just #include <linux/sync_file.h> ?
> >> +struct sync_merge_data {
> >> +     char    name[32];
> >> +     int32_t fd2;
> >> +     int32_t fence;
> >> +     uint32_t        flags;
> >> +     uint32_t        pad;
> >> +};
> >> +#define SYNC_IOC_MAGIC               '>'
> >> +#define SYNC_IOC_MERGE               _IOWR(SYNC_IOC_MAGIC, 3, struct sync_merge_data)
> >> +
> >> +
> >> +
> >> +static inline int sync_wait(int fd, int timeout)
> >> +{
> >> +     struct pollfd fds;
> >> +
> >> +     fds.fd = fd;
> >> +     fds.events = POLLIN | POLLERR;
> >
> > POLLERR is implied and ignored in fds.events.
> >
> >> +     return poll(&fds, 1, timeout);
> >
> > Returns: -1 on error, 0 on timeout, 1 if signaled.
> >
> > This function is horrible wrt to -EINTR for example :) Hmm, mixing
> > poll() and looping over signals until timeout doesn't look as
> > straightforward.
> 
> hmm, this was just basically cut/paste from android libsync (and iirc
> removing the legacy ioctls)..  although I did just realize there was a
> newer version which turned this into:
> 
>    ret = poll(..);
>    if (ret > 0)
>       return 0;
>    return ret;
> 
> perhaps not super-awesome for -EINTR handling.. and tbh I'm 100% of
> the behavior when this used to be simply "return ioctl(fd,
> SYNC_IOC_WAIT, &to);"..

The latest version take a few more cases in account and should behave
exatly like SYNC_IOC_WAIT.

int sync_wait(int fd, int timeout)                                              
{                                                                               
    struct pollfd fds;                                                          
    int ret;                                                                    
                                                                                
    fds.fd = fd;                                                                
    fds.events = POLLIN;                                                        
    ret = poll(&fds, 1, timeout);                                               
    if (ret > 0) {                                                              
        if (fds.revents & (POLLERR | POLLNVAL))                                 
            errno = EINVAL;                                                     
            return -1;                                                          
        return 0;                                                               
    } else if (ret == 0) {                                                      
        errno = ETIME;                                                          
        return -1;                                                              
    }                                                                           
    return ret;                                                                 
}


> 
> >> +}
> >> +
> >> +static inline int sync_merge(const char *name, int fd1, int fd2)
> >> +{
> >> +     struct sync_merge_data data = {};
> >> +     int err;
> >
> > What I liked was doing
> >
> > if (fd2 < 0)
> >         return dup(fd1);
> >
> > if (fd1 < 0)
> >         return dup(fd2);
> >
> > That makes accumulating the fences in the caller much easier (i.e. they
> > start with
> >         batch.fence_in = -1;
> > then
> >         batch.fence_in = sync_merge(batch.fence_in, fence);
> > finished by
> >         execbuf(&batch);
> >         close(batch.fence_in);
> >         batch.fence_in = -1;
> 
> I guess that would end up ignoring the fence name.. although not sure
> that I care.  But probably we should either make the android version
> behave the same, or pick new names for these fxns.  I think having
> same names but slightly different behaviour would be confusing.
> 
> Oh, and now that I'm actually looking at that code, the extra null
> terminator bit after the strncpy() is kinda unneeded..  should just
> do:
> 
>        strncpy(data.name, name, sizeof(data.name));
> 
> (without the -1) instead
> 
> BR,
> -R
> 
> > -Chris
> >
> > --
> > Chris Wilson, Intel Open Source Technology Centre
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/dri-devel


More information about the dri-devel mailing list