Re: DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC
On 9/11/20 1:15 AM, Dmitry Osipenko wrote: 09.09.2020 11:10, Mikko Perttunen пишет: On 9/9/20 2:45 AM, Dmitry Osipenko wrote: 05.09.2020 13:34, Mikko Perttunen пишет: ... +/* Submission */ + +/** Patch address of the specified mapping in the submitted gather. */ +#define DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC (1<<0) Shouldn't the kernel driver be aware about what relocations need to be patched? Could you please explain the purpose of this flag? Sure, the kernel knows if it returned the IOVA to the user or not, so we could remove this flag and determine it implicitly. I don't think there is much harm in it though; if we have the flag an application can decide to ignore the iova field and just pass WRITE_RELOC always, and it's not really any extra code on kernel side. Sounds like there is no real practical use for this flag other than for testing purposes, correct? Patching depending just on if the MAP IOCTL returned an IOVA or not seems a bit "spooky action at a distance"-ish to me, but maybe it's not so bad.. I'll consider removing it. Mikko ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC
09.09.2020 11:10, Mikko Perttunen пишет: > On 9/9/20 2:45 AM, Dmitry Osipenko wrote: >> 05.09.2020 13:34, Mikko Perttunen пишет: >> ... >>> +/* Submission */ >>> + >>> +/** Patch address of the specified mapping in the submitted gather. */ >>> +#define DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC (1<<0) >> >> Shouldn't the kernel driver be aware about what relocations need to be >> patched? Could you please explain the purpose of this flag? >> > > Sure, the kernel knows if it returned the IOVA to the user or not, so we > could remove this flag and determine it implicitly. I don't think there > is much harm in it though; if we have the flag an application can decide > to ignore the iova field and just pass WRITE_RELOC always, and it's not > really any extra code on kernel side. Sounds like there is no real practical use for this flag other than for testing purposes, correct? ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC
On 9/9/20 2:45 AM, Dmitry Osipenko wrote: 05.09.2020 13:34, Mikko Perttunen пишет: ... +/* Submission */ + +/** Patch address of the specified mapping in the submitted gather. */ +#define DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC (1<<0) Shouldn't the kernel driver be aware about what relocations need to be patched? Could you please explain the purpose of this flag? Sure, the kernel knows if it returned the IOVA to the user or not, so we could remove this flag and determine it implicitly. I don't think there is much harm in it though; if we have the flag an application can decide to ignore the iova field and just pass WRITE_RELOC always, and it's not really any extra code on kernel side. Mikko ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel
Re: DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC
05.09.2020 13:34, Mikko Perttunen пишет: ... > +/* Submission */ > + > +/** Patch address of the specified mapping in the submitted gather. */ > +#define DRM_TEGRA_SUBMIT_BUF_WRITE_RELOC (1<<0) Shouldn't the kernel driver be aware about what relocations need to be patched? Could you please explain the purpose of this flag? ___ dri-devel mailing list dri-devel@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/dri-devel