[23/31] HWPOISON: add memory cgroup filter
in [Kernel]
|
Prev: drivers/net/au1000_eth.c: Remove misuse of DECLARE_MAC_BUF
Next: [PATCH] hwmon: Add driver for VIA CPU core temperature
From: KAMEZAWA Hiroyuki on 9 Dec 2009 19:00 On Wed, 9 Dec 2009 12:47:27 -0800 Paul Menage <menage(a)google.com> wrote: > On Tue, Dec 8, 2009 at 1:16 PM, Andi Kleen <andi(a)firstfloor.org> wrote: > > > > The hwpoison test suite need to inject hwpoison to a collection of > > selected task pages, and must not touch pages not owned by them and > > thus kill important system processes such as init. (But it's OK to > > mis-hwpoison free/unowned pages as well as shared clean pages. > > Mis-hwpoison of shared dirty pages will kill all tasks, so the test > > suite will target all or non of such tasks in the first place.) > > While the functionality sounds useful, the interface (passing an inode > number) feels a bit ugly to me. Also, if that group is deleted and a > new cgroup created, you could end up reusing the inode number. > I agree. > How about an approach where you write either the cgroup path (relative > to the memcg mount) or an fd open on the desired cgroup? Then you > could store a (counted) css reference rather than an inode number, > which would make the filter function cleaner too, since it would just > need to compare css objects. > Sounds reasonable. Thanks, -Kame -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Andi Kleen on 9 Dec 2009 20:50 > While the functionality sounds useful, the interface (passing an inode > number) feels a bit ugly to me. Also, if that group is deleted and a > new cgroup created, you could end up reusing the inode number. Please note this is just a testing interface, doesn't need to be 100% fool-proof. It'll never be used in production. > > How about an approach where you write either the cgroup path (relative > to the memcg mount) or an fd open on the desired cgroup? Then you > could store a (counted) css reference rather than an inode number, > which would make the filter function cleaner too, since it would just > need to compare css objects. Sounds complicated, I assume it would be much more code? I would prefer to keep the testing interfaces as simple as possible. -Andi -- ak(a)linux.intel.com -- Speaking for myself only. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Balbir Singh on 9 Dec 2009 21:30 * Andi Kleen <andi(a)firstfloor.org> [2009-12-10 02:42:12]: > > While the functionality sounds useful, the interface (passing an inode > > number) feels a bit ugly to me. Also, if that group is deleted and a > > new cgroup created, you could end up reusing the inode number. > > Please note this is just a testing interface, doesn't need to be > 100% fool-proof. It'll never be used in production. > > > > > How about an approach where you write either the cgroup path (relative > > to the memcg mount) or an fd open on the desired cgroup? Then you > > could store a (counted) css reference rather than an inode number, > > which would make the filter function cleaner too, since it would just > > need to compare css objects. > > Sounds complicated, I assume it would be much more code? > I would prefer to keep the testing interfaces as simple as possible. > We do this for cgroupstats and the code is not very complicated. In case you want to look, the user space code is at Documentation/accounting/getdelays.c and the kernel code is in kernel/taskstats.c -- Balbir -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Wu Fengguang on 10 Dec 2009 21:20 On Thu, Dec 10, 2009 at 10:21:13AM +0800, Balbir Singh wrote: > * Andi Kleen <andi(a)firstfloor.org> [2009-12-10 02:42:12]: > > > > While the functionality sounds useful, the interface (passing an inode > > > number) feels a bit ugly to me. Also, if that group is deleted and a > > > new cgroup created, you could end up reusing the inode number. > > > > Please note this is just a testing interface, doesn't need to be > > 100% fool-proof. It'll never be used in production. > > > > > > > > How about an approach where you write either the cgroup path (relative > > > to the memcg mount) or an fd open on the desired cgroup? Then you > > > could store a (counted) css reference rather than an inode number, > > > which would make the filter function cleaner too, since it would just > > > need to compare css objects. > > > > Sounds complicated, I assume it would be much more code? > > I would prefer to keep the testing interfaces as simple as possible. > > > > We do this for cgroupstats and the code is not very complicated. In > case you want to look, the user space code is at > Documentation/accounting/getdelays.c and the kernel code is in > kernel/taskstats.c Balbir, thanks for the tip. We could keep an fd open on the desired cgroup, in user space: #!/bin/bash mkdir /cgroup/hwpoison && \ exec 9<>/cgroup/hwpoison/tasks || exit 1 A bit simpler than an in-kernel fget_light() or CSS refcount :) Thanks, Fengguang -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
From: Andi Kleen on 14 Dec 2009 08:00
Wu Fengguang <fengguang.wu(a)intel.com> writes: > > We could keep an fd open on the desired cgroup, in user space: > > #!/bin/bash > > mkdir /cgroup/hwpoison && \ > exec 9<>/cgroup/hwpoison/tasks || exit 1 > > A bit simpler than an in-kernel fget_light() or CSS refcount :) FYI, I decided to not do any of this in .33, but just keep the ugly-but-working inode hack. We can look at fixing that for .34. These interfaces are debugfs, so can be changed. -Andi -- ak(a)linux.intel.com -- Speaking for myself only. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo(a)vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ |