On Wed, Mar 15, 2023 at 05:41:53PM +0100, David Hildenbrand wrote: > On 15.03.23 17:10, Luis Chamberlain wrote: > > On Wed, Mar 15, 2023 at 01:24:41PM +0100, David Hildenbrand wrote: > > > On 11.03.23 06:17, Luis Chamberlain wrote: > > > > A long time ago we had some issues with userspace doing stupid stuff. > > > > Well, it turns out even the kernel and do stupid stuff too, as we're > > > > learning with the ACPI modules aliaes and that hammering tons of loads. > > > > > > > > So add a bit of code which gets us a bit more in the defensive about > > > > these situations. > > > > > > > > To experiment, this also adds in-kernel alias support to see if this helps > > > > with some larger systems. > > > > > > > > This is all based on some old code which tried to add defensive > > > > mechanisms the last of which was here and I had dropped the ball: > > > > > > > > https://lore.kernel.org/all/20171208001540.23696-1-mcgrof@xxxxxxxxxx/ > > > > > > > > I've only compile tested this for now. Will need to stress to test > > > > with kmod tests 0008 and 0009 to see if there's any differences. > > > > I'll have to re-test and re-gnuplot stuff there. But early feedback > > > > is appreciated, hence the RFC. > > > > > > > > David Hildenbrand had reported a while ago issues with userspace > > > > doing insane things with allocations bringing a system down to > > > > its knees. This is part of the motivation for this series. > > > > > > > > > I'll try to grab a system where I can reproduce the issue and give your > > > patches a churn. > > > > Great, then please wait for v2 RFC as the first patch was missing an > > obvious mutex grab / release, I already have some memory pressure data > > that shows gains. Hope to post soon. > > I expect to have a machine (with a crazy number of CPUs/devices) available > in a couple of days (1-2), so no need to rush. > > The original machine I was able to reproduce with is blocked for a little > bit longer; so I hope the alternative I looked up will similarly trigger the > issue easily. OK give this a spin: https://git.kernel.org/pub/scm/linux/kernel/git/mcgrof/linux.git/log/?h=20230316-module-alloc-opts I'm seeing about ~86 MiB saving on the upper bound on memory usage while hammering on kmod test 0008, and this is on a small system. Probably won't help *that* much but am curious... if it helps somewhat. Luis
