Re: [PATCH] shm: fix a race between shm_exit() and shm_init()

[Richard Weinberger <richard@nod.at>]

On Mittwoch 03 August 2011 15:33:27 Kay Sievers wrote:
> On Wed, Aug 3, 2011 at 15:13, Thadeu Lima de Souza Cascardo
> 
> <cascardo@minaslivre.org> wrote:
> > As Marc Zyngier has pointed out, the culprit is driver_init, which
> > calls devices_init. I'd say other calls from driver_init also call into
> > kset and kobject, which will dispatch uevent helper.
> > 
> > I tried to reproduce the problem using UML and only was successful after
> > setting CONFIG_UEVENT_HELPER_PATH. I guess we could obsolete this
> > feature and plan its removal soon. Is anybody still using this?
> 
> Forking binaries from the kernel for frequent events is something
> fundamentally wrong to do. It is not even rate-limited or has any
> upper bounds. It is known to create out-of-memory situations on
> machines with many devices and tiny RAM. Configs that virtualizations
> with many guests use pretty often. They can not even bootup with
> /sbin/hotplug enabled then.
> 
> It would be nice to remove that broken thing, no common system or
> distro uses it since quite some years now. But I wouldn't be surprised
> if some people still use it for whatever hack they need locally.
> 

True.

I'll remove CONFIG_UEVENT_HELPER_PATH from UML's defconfig.

Thanks,
//richard