[PATCH v5 00/21] Host1x sync point UAPI should not be used for tracking DRM jobs

Thierry Reding thierry.reding at gmail.com
Tue Mar 23 18:21:58 UTC 2021


On Sat, Feb 27, 2021 at 02:19:39PM +0300, Dmitry Osipenko wrote:
> 03.02.2021 14:18, Mikko Perttunen пишет:
> ...
> >> I'll need more time to think about it.
> >>
> > 
> > How about something like this:
> > 
> > Turn the syncpt_incr field back into an array of structs like
> > 
> > #define DRM_TEGRA_SUBMIT_SYNCPT_INCR_REPLACE_SYNCOBJ        (1<<0)
> > #define DRM_TEGRA_SUBMIT_SYNCPT_INCR_PATCH_DYNAMIC_SYNCPT    (1<<1)
> > 
> > struct drm_tegra_submit_syncpt_incr {
> >     /* can be left as zero if using dynamic syncpt */
> >     __u32 syncpt_id;
> >     __u32 flags;
> > 
> >     struct {
> >         __u32 syncobj;
> >         __u32 value;
> >     } fence;
> > 
> >     /* patch word as such:
> >          * *word = *word | (syncpt_id << shift)
> >          */
> >     struct {
> >         __u32 gather_offset_words;
> >         __u32 shift;
> >     } patch;
> > };
> > 
> > So this will work similarly to the buffer reloc system; the kernel
> > driver will allocate a job syncpoint and patch in the syncpoint ID if
> > requested, and allows outputting syncobjs for each increment.
> 
> I haven't got any great ideas so far, but it feels that will be easier
> and cleaner if we could have separate job paths (and job IOCTLS) based
> on hardware generation since the workloads a too different. The needs of
> a newer h/w are too obscure for me and absence of userspace code,
> firmware sources and full h/w documentation do not help.
> 
> There still should be quite a lot to share, but things like
> mapping-to-channel and VM sync points are too far away from older h/w,
> IMO. This means that code path before drm-sched and path for job-timeout
> handling should be separate.
> 
> Maybe later on it will become cleaner that we actually could unify it
> all nicely, but for now it doesn't look like a good idea to me.

Sorry for jumping in rather randomly here and elsewhere, but it's been a
long time since the discussion and I just want to share my thoughts on a
couple of topics in order to hopefully help move this forward somehow.

For UAPI, "unifying it later" doesn't really work. So I think the only
realistic option is to make a best attempt at getting the UABI right so
that it works for all existing use-cases and ideally perhaps even as of
yet unknown use-cases in the future. As with all APIs this means that
there's going to be a need to abstract away some of the hardware details
so that we don't have to deal with too many low-level details in
userspace, because otherwise the UAPI is just going to be a useless
mess.

I think a proposal such as the above to allow both implicit and explicit
syncpoints makes sense. For what it's worth, it's fairly similar to what
we had come up with last time we tried destaging the ABI, although back
at the time I'm not sure we had even considered explicit syncpoint usage
yet. I think where reasonably possible this kind of optional behaviour
is acceptable, but I don't think having two completely separate paths is
going to help in any way. If anything it's just going to make it more
difficult to maintain the code and get it to a usable state in the first
place.

Like I said elsewhere, the programming model for host1x hasn't changed
since Tegra20. It's rather evolved and gained a couple more features,
but that doesn't change anything about how userspace uses it.

Thierry
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://lists.freedesktop.org/archives/dri-devel/attachments/20210323/032c67c8/attachment.sig>


More information about the dri-devel mailing list