On Thu, Dec 08, 2016 at 10:47:51AM -0800, Luis R. Rodriguez wrote: > Upon running into an old kmod v19 issue with mount (get_fs_type()) a few of us > hunted for the cause of the issue. Although the issue ended up being a > userspace issue, a stress test driver was written to help reproduce the issue, > and along the way a few other fixes and sanity checks were implemented. > > I've taken the time to generalize the stress test driver as a kselftest driver > with a 9 test cases. The last two test cases reveal an existing issue which > is not yet addressed upstream, even if you have kmod v19 present. A fix is > proposed in the last patch. Orignally we had discarded this patch as too > complex due to the alias handling, but upon further analysis of test cases > and memory pressure issues, it seems worth considering. Other than the > last patch I don't think much of the other patches are controversial, but > sending as RFC first just in case. > > If its not clear, an end goal here is to make module loading a bit more > deterministic with stronger sanity checks and stress tests. Please note, > the stress test diver requires 4 GiB of RAM to run all tests without running > out of memory. A lot of this has to do with the memory requirements needed > for a dynamic test for multiple threads, but note that the final memory > pressure and OOMs actually don't come from this allocation, but instead > from many finit_module() calls, this consumes quite a bit of memory, specially > if you have a lot of dependencies which also need to be loaded prior to > your needed module -- as is the case for filesystem drivers. > > These patches are available on my linux-next git-tree on my branch > 20161208-kmod-test-driver-try2 [0], which is based on linux-next tag > next-20161208. Patches are also available based on v4.9-rc8 [1] for > those looking for a bit more stable tree given x86_64 on linux-next is > hosed at the moment. > > Since kmod.c doesn't seem to get much love, and since I've been digging > quite a bit into it for other users (firmware) I suppose I could volunteer > myself to maintain this code as well, unless there are oppositions to this. > > [0] https://git.kernel.org/cgit/linux/kernel/git/mcgrof/linux-next.git/log/?h=20161208-kmod-test-driver-try2 > [1] https://git.kernel.org/cgit/linux/kernel/git/mcgrof/linux.git/log/?h=20161208-kmod-test-driver > > Luis R. Rodriguez (10): > kmod: add test driver to stress test the module loader > module: fix memory leak on early load_module() failures > kmod: add dynamic max concurrent thread count > kmod: provide wrappers for kmod_concurrent inc/dec > kmod: return -EBUSY if modprobe limit is reached > kmod: provide sanity check on kmod_concurrent access > kmod: use simplified rate limit printk > sysctl: add support for unsigned int properly > kmod: add helpers for getting kmod count and limit > kmod: add a sanity check on module loading > A lot of good discussions have come up form this, and so also a few more patches. I'm going to split up the work into changes which make sense now and leave debug work for a follow up later. Luis -- To unsubscribe from this list: send the line "unsubscribe linux-kselftest" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
