This repository provides test environment to test various mechanism for communications between userland and kernelland. Mechanisms were choosen regardless of how 'dirty' they are considered to be. (ex : /pipe).
The following test environments are provided:
- Netlink with pre-allocated buffer of fixed size for reception and no resynchronisation (/netlink_simple)
- Generic Netlink for pseudo-synchronous I/O (transactions), and asynchronous I/O (/genetlink)
- A character device module with read/write callback (/rwdev)
- A character device module with mmap callback (/mmapdev)
- Relay API with debugfs files (/relay)
- POSIX pipe IPC (/pipe)
Some mechanism provide two implementations:
- An implementation for user to kernel communication;
- An implementation for kernel to user communication.
Each implementation has:
- a kern directory containing the module to be added to the kernel;
- a user directory containing a binary to communicate with the module.
In user land, output can be read in stdout. In kernel land, output can be read with 'dmesg'.
KernToUser : First launch user/nlreader then load nlkern module.
KernToUser : First launch user/nlreader then load nlkern module.
First, load genlbench module, then genlbench_test. Finally, user/benchclient will run tests.
Tests were run on an Intel(R) Core(TM) i5-4210U CPU @ 1.70GHz with 2 core and 2 threads per core. Results are provided in clock cycles/messages. Messages are all 1000 to 1024 bytes long.
| User Sending | Kernel receiving | |
|---|---|---|
| netlink_simple | 15 000 | NO DATA |
| netlink_synchro | NOT IMPLEMENTED | NOT IMPLEMENTED |
| rwdev | 52 405 | 182 |
| mmapdev | ||
| relay | NOT POSSIBLE | NOT POSSIBLE |
| pipe | NOT IMPLEMENTED | NOT IMPLEMENTED |
| Kernel Sending | User receiving | |
|---|---|---|
| netlink_simple | 3 498 | 3 724 |
| netlink_synchro | 1 502 | 5 601 |
| rwdev | NOT IMPLEMENTED | NOT IMPLEMENTED |
| mmapdev | NOT IMPLEMENTED | NOT IMPLEMENTED |
| relay | 158 | 6078 |
| pipe | 4 620 | 4 622 |