On Di, 2015-06-16 at 14:55 +1000, Dave Airlie wrote: > Hey Gerd, > > This probably doesn't matter yet, but they virtgpu_vq buffer > allocation code you added isn't threadsafe. Right, I'll add a lock (or do you have a patch already?) > I'm not sure what was wrong with the original code I wrote, you never > contacted me with reasons for rewriting it, but the Linux kmalloc > isn't that slow, but if we want some form of cached allocator you > probably should just use kmem_cache_create instead of what you have > there. That came up during patch review. Problem isn't kmalloc performance, kmalloc (ENOMEM) but error handling. When we allocate stuff in advance a bunch of nasty error paths simply can't happen. cheers, Gerd _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
