Jerome Glisse wrote: > Thomas Hellström wrote: >> Jerome Glisse wrote: >> >>> Hi, >>> >>> If my driver emit fence function fail should there be a way >>> to clean unfenced buffer ? Or should i do this by myself in >>> the function which validate buffer ? If so then i915 execbuffer >>> is likely buggy in this regards (don't try to unload module >>> then as it will trigger BUG_ON(!list_empty(&bm->unfenced)) >>> >>> Cheers, >>> Jerome Glisse >>> >>> >>> >> Hi. >> If fencing fails, the function that validates the buffer must clean >> the unfenced list. >> There is a function to clean the unfenced list in drm_bo.c. >> >> This is typically called if the function receives a signal while >> waiting to validate a buffer, which is a very common case if the X >> server is the caller. >> >> If command submission has already occured when fencing fails, >> the calling function must wait for GPU idle before cleaning the list. >> >> /Thomas >> >>> ------------------------------------------------------------------------- >>> >>> This SF.net email is sponsored by: Microsoft >>> Defy all challenges. Microsoft(R) Visual Studio 2005. >>> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ >>> -- >>> _______________________________________________ >>> Dri-devel mailing list >>> [email protected] >>> https://lists.sourceforge.net/lists/listinfo/dri-devel >>> >>> > > Is it drm_putback_buffer_objects ? If so then i don't see it call > in any error path in i915_dma.c is this right (for i915 i mean). Yes. That's the function. I haven't had a thorough look into the i915 superioctl error handling yet, so it's possible that fencing errors are handled differently there.
It's currently used by the Intel Poulsbo driver, but I'm not sure Intel has released the latest Poulsbo drm source yet. /Thomas > > Cheers, > Jerome Glisse ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2005. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ -- _______________________________________________ Dri-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/dri-devel
