Re: [RFC v4 net-next 2/4] net: dsa: mt7530: add interrupt support

From: DENG Qingfang <dqfext@gmail.com>
To: Andrew Lunn <andrew@lunn.ch>
Cc: "Marc Zyngier" <maz@kernel.org>,
	"David S. Miller" <davem@davemloft.net>,
	"Florian Fainelli" <f.fainelli@gmail.com>,
	"Heiner Kallweit" <hkallweit1@gmail.com>,
	"Jakub Kicinski" <kuba@kernel.org>,
	"Landen Chao" <Landen.Chao@mediatek.com>,
	"Matthias Brugger" <matthias.bgg@gmail.com>,
	"Russell King" <linux@armlinux.org.uk>,
	"Sean Wang" <sean.wang@mediatek.com>,
	"Vivien Didelot" <vivien.didelot@gmail.com>,
	"Vladimir Oltean" <olteanv@gmail.com>,
	"Rob Herring" <robh+dt@kernel.org>,
	"Linus Walleij" <linus.walleij@linaro.org>,
	"Greg Kroah-Hartman" <gregkh@linuxfoundation.org>,
	"Sergio Paracuellos" <sergio.paracuellos@gmail.com>,
	linux-kernel@vger.kernel.org, linux-mediatek@lists.infradead.org,
	linux-staging@lists.linux.dev, devicetree@vger.kernel.org,
	netdev@vger.kernel.org, "Weijie Gao" <weijie.gao@mediatek.com>,
	"Chuanhong Guo" <gch981213@gmail.com>,
	"René van Dorst" <opensource@vdorst.com>,
	"Frank Wunderlich" <frank-w@public-files.de>,
	"Thomas Gleixner" <tglx@linutronix.de>,
	"Greg Ungerer" <gerg@kernel.org>
Subject: Re: [RFC v4 net-next 2/4] net: dsa: mt7530: add interrupt support
Date: Tue, 13 Apr 2021 23:29:20 +0800	[thread overview]
Message-ID: <20210413152920.2190769-1-dqfext@gmail.com> (raw)
In-Reply-To: <YHWUK+tG3v9ZU/DT@lunn.ch>

On Tue, Apr 13, 2021 at 02:52:59PM +0200, Andrew Lunn wrote:
> > I guess this is depends whether the most usual case is to have all
> > these interrupts being actively in use or not. Most interrupts only
> > use a limited portion of their interrupt space at any given time.
> > Allocating all interrupts and creating mappings upfront is a waste of
> > memory.
> > 
> > If the use case here is that all these interrupts will be wired and
> > used in most cases, then upfront allocation is probably not a problem.
> 
> Hi Marc
> 
> The interrupts are generally used. Since this is an Ethernet switch,
> generally the port is administratively up, even if there is no cable
> plugged in. Once/if a cable is plugged in and there is a link peer,
> the PHY will interrupt to indicate this.
> 
> The only real case i can think of when the interrupts are not used is
> when the switch has more ports than connected to the front panel. This
> can happen in industrial settings, but not SOHO. Those ports which
> don't go anywhere are never configured up and so the interrupt is
> never used.

Hi Andrew

This is what the extra check (BIT(p) & ds->phys_mii_mask) avoids.

Currently the mv88e6xxx driver does not have this check, and creates
15 PHY IRQ mappings on my 88E6176 unconditionally, leaving a gap in
/proc/interrupts:

...
 57:          0          0  mv88e6xxx-g1   3 Edge      mv88e6xxx-f1072004.mdio-mii:00-g1-atu-prob
 59:          0          0  mv88e6xxx-g1   5 Edge      mv88e6xxx-f1072004.mdio-mii:00-g1-vtu-prob
 61:          8          5  mv88e6xxx-g1   7 Edge      mv88e6xxx-f1072004.mdio-mii:00-g2
 63:          8          4  mv88e6xxx-g2   0 Edge      mv88e6xxx-1:00
 64:          0          0  mv88e6xxx-g2   1 Edge      mv88e6xxx-1:01
 65:          0          0  mv88e6xxx-g2   2 Edge      mv88e6xxx-1:02
 66:          0          0  mv88e6xxx-g2   3 Edge      mv88e6xxx-1:03
 67:          0          2  mv88e6xxx-g2   4 Edge      mv88e6xxx-1:04
// IRQ 68~77 are created but not used
 78:          0          0  mv88e6xxx-g2  15 Edge      mv88e6xxx-f1072004.mdio-mii:00-watchdog
...

You may as well add irq_set_nested_thread(irq, true) to irq_domain_map
so all IRQs share a single thread.

> 
>       Andrew