On Mon, Jun 15, 2020 at 01:30:42PM -0300, Vitor Massaru Iha wrote:
On Fri, 2020-06-12 at 15:36 -0700, Kees Cook wrote:
Why drop the __initconst?
I removed __initconst because of these warnings below, as it is used for the kernel during the module initialization, and I do not use the module initialization in this tests. Does this have any side effects in these tests?
WARNING: modpost: vmlinux.o(.text.unlikely+0x131b7): Section mismatch in reference from the function test_s8_overflow() to the variable .init.rodata:s8_tests The function test_s8_overflow() references the variable __initconst s8_tests. This is often because test_s8_overflow lacks a __initconst annotation or the annotation of s8_tests is wrong.
Ah, right. I assume there are build modes where the tests don't only live in the __init section any more due to how Kunit does its runtime? (Before all the tests ran from __init via module_init(), IIRC.)
So, yes! I like it. :) Most of my comments here have nothing to do with specifically this patch (sorry)! But I'd love to see a v2.
Thanks for doing this! I'm glad to see more TAP output. :)
Thanks Kees, I'm learning a lot from you, and as I said privately with Brendan, I've never seen so much macro in a code. I learned a lot from it.
Heh, yes, for that I must beg forgiveness. ;) The macros are rather wild, but it seemed the best way to avoid a ton of cut/paste code duplication.