Assorted fixes for kernel 4.14 #658
Conversation
Off by one; 6e71b04a8224 was first tagged in 4.15-rc1. Signed-off-by: Timo Beckers <timo@isovalent.com>
Since the test is gated by haveProgAttach(), it should not specify AttachType, which is a feature available as of 4.17. Also, CGroupSKB doesn't require specifying AttachType. Signed-off-by: Timo Beckers <timo@isovalent.com>
Kernels up until 4.16 return kernel pointers in the constant field of map load instructions and don't set the source register to pseudo values correctly. This results in an incorrect Tag value calculation on the userspace side. Remove the use of a map fd in TestProgramInstructions to fix the test between kernel versions 4.13 and 4.16. Signed-off-by: Timo Beckers <timo@isovalent.com>
| @@ -687,17 +687,12 @@ func TestProgramBindMap(t *testing.T) { | |||
| } | |||
|
|
|||
| func TestProgramInstructions(t *testing.T) { | |||
There was a problem hiding this comment.
Does comparing the Tag in our tests make sense in the first place if the output is so unpredictable?
There was a problem hiding this comment.
Might be worth an issue as well.
There was a problem hiding this comment.
We compare Tag instead of comparing Instructions. I don't think it does any harm, and when a check like this starts to fail, it usually leads to an interesting discovery, like in this case.
Might be worth an issue as well.
What do you mean?
There was a problem hiding this comment.
Can we sanitise the Instructions ourselves? Should we store map ID in metadata?
There was a problem hiding this comment.
For the test case as it was before this patch; not really. Given the following prog:
0: LdImmDW dst: r0 imm: -1
2: LoadMapPtr dst: r1 fd: 6
4: Mov32Imm dst: r0 imm: 0
5: Exit
This is what the xlated insns look like:
0: LdImmDW dst: r0 imm: -1
2: LdImmDW dst: r1 imm: -103942818276608
4: Mov32Imm dst: r0 imm: 0
5: Exit
Src of instruction 2 is unset. Sanitizing the pointer is no problem, but no way to figure out this is a map load if I understand correctly.
TestMapLoadPinnedWithOptionsas of 4.15mustCgroupFixturesTestProgramInstructionsTogether with #657, this makes tests pass on our 4.14 ci-kernels image. To be enabled later when we move to self-hosted CI runners.