On Fri, Sep 25, 2015 at 01:57:58PM +0200, Lucas Stach wrote: > This rewrites the submit interface to copy the command stream from user > memory. This mitigates a potential attack vector of the old interface > where userspace could submit a command buffer that would be validated by > the kernel, but is still mapped into userspace. This could be exploited > by changing the command stream after validation but before the actual > GPU execution. > > A nice side effect is that validation and reloc patching can now operate > on cached memory and only the final result is copied to a writecombined > command buffer which should make those operations a bit more efficient. > > A simplification to the interface is the removal of the ability to push > multiple command buffers per submit. As we don't use it for context > restore buffers and the fact that userspace doesn't need to work with > a fixed command buffer size anymore, with the potential risk to overflow > its size in the middle of an atomic stream section, there is no need > for this complication anymore. > > Signed-off-by: Lucas Stach <l.stach@xxxxxxxxxxxxxx> Applied, thanks. -- FTTC broadband for 0.8mile line: currently at 9.6Mbps down 400kbps up according to speedtest.net. _______________________________________________ dri-devel mailing list dri-devel@xxxxxxxxxxxxxxxxxxxxx http://lists.freedesktop.org/mailman/listinfo/dri-devel
