Re: [PATCH RFC v2 net-next 10/16] bpf: add eBPF verifier

On Wed, Jul 23, 2014 at 5:48 PM, Alexei Starovoitov [off-list ref] wrote:

check_reg_arg() is indeed incorrect here. Will fix. That was a good catch.
Thank you for review!

I fixed this missing check_reg_arg() check, addressed feedback for
other patches, rebased it and pushed it:
https://git.kernel.org/cgit/linux/kernel/git/ast/bpf.git/log/

Also replaced dump into syslog from verifier with print into user supplied
buffer, so that user space can see why verifier rejected the program.
The more I play with it the more I like fd-based user interface. Cannot
thank enough Andy for suggesting it :)
Currently map_id is still there as kernel internal id that programs are
using to access maps. I'll try to replace it with direct pointer.
That will save one idr_find() lookup in critical path. That's a good
reason to add new 16-byte eBPF instruction. It will be
"load 64-bit immediate" that we've discussed before.