Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 22 Feb 2018 13:45:09 -0500 (EST)
From:   Alan Stern <stern@rowland.harvard.edu>
To:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>
cc:     Mauro Carvalho Chehab <mchehab@s-opensource.com>,
        Frederic Weisbecker <frederic@kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        <linux-usb@vger.kernel.org>
Subject: Re: [RFC PATCH] usb: hcd: complete URBs in threaded-IRQ context
 instead of tasklet
In-Reply-To: <20180222161949.3prjzbvkss4hjj2c@linutronix.de>
Message-ID: <Pine.LNX.4.44L0.1802221336520.1295-100000@iolanthe.rowland.org>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 22 Feb 2018, Sebastian Andrzej Siewior wrote:

> On 2018-02-16 16:46:41 [-0500], Alan Stern wrote:
> > > The theaded interrupt runs SCHED_FIFO priority 50 by default. The only
> > > thing that can interrupt it are interrupts, a softirq (not ksoftirqd)
> > > and other tasks with a higher priority than 50.
> > > There should be no downside performance wise.
> > 
> > Maybe.  It would be nice to see some real measurements.
> 
> I had an usb3 flash stick behind the EHCI controller which was passed
> through from the host to a kvm guest. The performance numbers in the
> guest were equal (some noise was there) with and without the patch.
> The numbers with the patch were worse if lockdep was enabled which isn't
> much of a surprise.
> If you have anything specific requirements for a measurement then please
> let me know and I see what I can do.

No, I didn't have anything more specific in mind.

In principle then, using threaded-interrupt bottom halves instead of
tasklets should be fine.  I don't object to making such a change.

However, using a work queue for root-hub URBs is pretty ugly.  It would
be better to reinstate the code that dropped hcd_root_hub_lock around
root-hub givebacks, which was removed by commit 94dfd7edfd5c ("USB:
HCD: support giveback of URB in tasklet context"); then it would be 
safe to give back those URBs in the bottom half.

Alan Stern