linux-kernel - Re: [PATCH] flush_cpu_workqueue: don't flush an empty ->worklist

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20070115125401.GA134@tv-sign.ru>
Date:	Mon, 15 Jan 2007 15:54:01 +0300
From:	Oleg Nesterov <oleg@...sign.ru>
To:	Srivatsa Vaddagiri <vatsa@...ibm.com>
Cc:	Andrew Morton <akpm@...l.org>, David Howells <dhowells@...hat.com>,
	Christoph Hellwig <hch@...radead.org>,
	Ingo Molnar <mingo@...e.hu>,
	Linus Torvalds <torvalds@...l.org>,
	linux-kernel@...r.kernel.org, Gautham shenoy <ego@...ibm.com>,
	"Pallipadi, Venkatesh" <venkatesh.pallipadi@...el.com>
Subject: Re: [PATCH] flush_cpu_workqueue: don't flush an empty ->worklist

On 01/15, Srivatsa Vaddagiri wrote:
>
> On Mon, Jan 15, 2007 at 02:54:10AM +0300, Oleg Nesterov wrote:
> > How about the pseudo-code below?
> 
> Some quick comments:
> 
> - singlethread_cpu needs to be hotplug safe (broken currently)

Why? Could you explain?

> - Any reason why cpu_populated_map is not modified on CPU_DEAD?

Because CPU_DEAD/CPU_UP_CANCELED doesn't wait for cwq->thread to exit.
cpu_populated_map never shrinks, it only grows on CPU_UP_PREPARE.

We can change this, but it needs some more code, and I am not sure
we need it. Note that a "false" bit in cpu_populated_map only means
that flush_work/flush_workqueue/destroy_workqueu will do lock/unlock
of cwq->lock, nothing more.

> - I feel more comfortable if workqueue_cpu_callback were to take
>   workqueue_mutex in LOCK_ACQ and release it in LOCK_RELEASE
>   notifications.

The whole purpose of this change to avoid this!

	Gautham R Shenoy wrote:
	>
	> Oh! I was refering to the other set of workqueue deadlock woes. The
	> ones caused when subsystems (like cpufreq) try to create/destroy
	> workqueue from the cpuhotplug callback path.
	>
	> Creation/destruction of workqueue requires us to take workqueue_mutex,
	> which would have already been taken during CPU_LOCK_ACQUIRE.

With this change workqueue_mutex is only taken to protect workqueues
list, why should we hold it for (say) CPU_UP_PREPARE->CPU_ONLINE path?

>                   This will provide stable access to cpu_populated_map
>   to functions like __create_workqueue.

I think this is not needed.

> Finally, I wonder if these changes will be unnecessary if we move to
> process freezer based hotplug locking ...

This change ir not strictly necessary but imho make the code better and
shrinks .text by 379 bytes.

But I believe that freezer will change nothing for workqueue. We still
need take_over_work(), and hacks like migrate_sequence. And no, CPU_DEAD
can't just thaw cwq->thread which was bound to the dead CPU to complete
kthread_stop(), we should thaw all processes.

Oleg.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/