linux-kernel - Re: [PATCH] android: binder: Disable preemption while holding the global binder lock

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: <20160914074152.GC5020@twins.programming.kicks-ass.net>
Date:   Wed, 14 Sep 2016 09:41:52 +0200
From:   Peter Zijlstra <peterz@...radead.org>
To:     Arve Hjønnevåg <arve@...roid.com>
Cc:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Thomas Gleixner <tglx@...utronix.de>,
        "devel@...verdev.osuosl.org" <devel@...verdev.osuosl.org>,
        Riley Andrews <riandrews@...roid.com>,
        LKML <linux-kernel@...r.kernel.org>,
        Christoph Hellwig <hch@...radead.org>,
        Todd Kjos <tkjos@...gle.com>
Subject: Re: [PATCH] android: binder: Disable preemption while holding the
 global binder lock

On Wed, Sep 14, 2016 at 09:10:01AM +0200, Peter Zijlstra wrote:
> We could define a meaningful something for CFS and implement that, but
> it isn't currently done.

So the generalization of the Priority Inheritance Protocol is Proxy
Execution Protocol, which basically lets the boosted task run _as_ the
task on the block chain as picked by the schedule function (treating all
of them as runnable). Where 'as' means it really consumes scheduling
resources of the (blocked) donor task.

Since the scheduling function for FIFO is: pick the highest prio one and
go for it, this trivially reduces to PI for FIFO.

Now, Proxy Execution Protocol is 'trivial' to implement on UP, but has a
few wobbles on SMP.

But we can use it to define a sensible definition for a WFQ class
scheduler (like CFS). For these the scheduling function basically runs
the boosted task as every donor task on the block chain gets their slice.
Alternatively, since it treats all 'blocked' tasks as runnable, the
total weight of the boosted task is its own weight plus the sum of
weight on the block chain.

Which is something that shouldn't be too hard to implement, but very
much isn't what happens today.