Add libbpf to maintain Ubuntu 16.04 compatibility #2403
Conversation
| #include "ebpf_common.h" | ||
|
|
||
|
|
||
| /* If we operate in user space we only need to include bpf.h and |
There was a problem hiding this comment.
Will this compile with any kernel version?
There was a problem hiding this comment.
Well this is what libbpf says: "No ties to any specific kernel, transparent handling of older kernels. Libbpf is designed to be kernel-agnostic and work across multitude of kernel versions. It has built-in mechanisms to gracefully handle older kernels, that are missing some of the features, by working around or gracefully degrading functionality. Thus libbpf is not tied to a specific kernel version and can/should be packaged and versioned independently."
I try to use their custom headers everywhere I can.
| @@ -141,6 +141,15 @@ else() | |||
| set (SUPPORTS_KERNEL False) | |||
| endif() | |||
|
|
|||
| if (EXISTS "${CMAKE_CURRENT_SOURCE_DIR}/runtime/usr/include/bpf") | |||
| message(STATUS "Found libbpf") | |||
| else() | |||
There was a problem hiding this comment.
did you test this on MacOS?
There was a problem hiding this comment.
Let's see what Travis will say.
backends/ebpf/runtime/ebpf_kernel.h
Outdated
| ({ \ | ||
| char ____fmt[] = fmt; \ | ||
| bpf_trace_printk(____fmt, sizeof(____fmt), \ | ||
| ##__VA_ARGS__); \ | ||
| }) | ||
|
|
||
| /* helper functions called from eBPF programs written in C */ | ||
| static void *(*bpf_map_lookup_elem)(void *map, void *key) = | ||
| static void *(*bpf_map_lookup_elem)(void *map, const void *key) = |
There was a problem hiding this comment.
I was hoping we would not need this junk anymore.
Isn't there a header with this stuff in libbpf?
There was a problem hiding this comment.
I will see if I can strip more later. Last time I checked those were still necessary.
|
There are some problems with your Python scripts. |
|
Trying to support kernel eBPF tests in Ubuntu 16.04 is just pure masochism. At this point we also need to build the iproute2 package from source... These are the same problems we had with the xdp back ends so at least we have the fixes already. |
Cmake improvements. Fix some more things. Some work on runtime.mk
|
Trying to fix the build unearthed a bunch of other issues. Right now, I have no idea what to do about the failing tests. One of them requires a change to how tables are registered, the other is the LPM_TRIE map getting rejected. They are failing for reasons we have missed before. |
|
I thought we didn't support lpm in the ebpf backends. |
|
There is a test that fails for the kernel |
|
The other tests that fail are the extern object ones, they do not play nicely for various reasons. |
Add a comment.
More junk-a-way Make lpm ebpf fail properly.
|
I think this is ready for review now. I also created some issues based on problems I have encountered. |
|
@fruffy : is this finished? |
|
From my point of view it is. The remaining issues should probably be handled separately. Merging this will break the xdp back end, but I have a pull request ready for it. |
This pull request adds libbpf as dependency to make the eBPF back end more stable. It makes the generated code less dependent on the kernel headers of a particular distribution. This may help with the problem that different distributions of Ubuntu cannot compile the code generated by p4c-ebpf. Also fixes #1961.
Let's see if this works out.
Changes:
lpm_ebpf.p4, also marked this test as xfail.ebpf_kernel.hfile, using only the imports fromlibbpf.BPF_USER_MAP_UPDATE_ELEMto the user-space test framework.