content-class: urn:content-classes:message
MIME-Version: 1.0
Content-Type: text/plain;
	charset="us-ascii"
Content-Transfer-Encoding: 8BIT
Subject: RE: Updated MSI Patches
Date: Tue, 12 Aug 2003 18:34:28 -0700
Message-ID: <7F740D512C7C1046AB53446D3720017304AE9B@scsmsx402.sc.intel.com>
Thread-Topic: Updated MSI Patches
Thread-Index: AcNhNlcjRMlFc1eSQX6cRLw+PlGFygAAXqXA
From: "Nakajima, Jun" <jun.nakajima@intel.com>
To: "Zwane Mwaikambo" <zwane@linuxpower.ca>
Cc: "Jeff Garzik" <jgarzik@pobox.com>,
       "Nguyen, Tom L" <tom.l.nguyen@intel.com>,
       "Linux Kernel" <linux-kernel@vger.kernel.org>,
       "long" <tlnguyen@snoqualmie.dp.intel.com>
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1781
Lines: 55

Salability means many things. I'm not sure which aspect you are talking
about, but a good thing with MSI is that it does not depend on I/O
APICs. A typical I/O APIC has 24 RTEs, and we need about 10 I/O APICs to
consume the vectors. 

You can make this scale on SMP systems. The vector-based approach would
be easier to extend it per-CPU because vectors are inherently per-CPU,
compared with IRQ-based. Today we have IRQ balancing as you know, and
the key part is to bind IRQ to a particular CPU (BTW, with our patch it
happens automatically because the balancer does not care if they are IRQ
or vector). 

Thanks,
Jun

> -----Original Message-----
> From: Zwane Mwaikambo [mailto:zwane@linuxpower.ca]
> Sent: Tuesday, August 12, 2003 5:49 PM
> To: Nakajima, Jun
> Cc: Jeff Garzik; Nguyen, Tom L; Linux Kernel; long
> Subject: RE: Updated MSI Patches
> 
> On Tue, 12 Aug 2003, Nakajima, Jun wrote:
> 
> > static unsigned int startup_edge_ioapic_vector(unsigned int vector)
> > {
> > 	int was_pending = 0;
> > 	unsigned long flags;
> > 	int irq = vector_to_irq (vector);
> >
> > 	spin_lock_irqsave(&ioapic_lock, flags);
> > 	if (irq < 16) {
> > 		disable_8259A_irq(irq);
> > 		if (i8259A_irq_pending(irq))
> > 			was_pending = 1;
> > 	}
> > 	__unmask_IO_APIC_irq(irq);
> > 	spin_unlock_irqrestore(&ioapic_lock, flags);
> >
> > 	return was_pending;
> > }
> 
> Hmm that doesn't look too bad, i like, ok another question, do you
think
> this could be made to scale on systems with lots of devices requiring
> vectors?
> 
> Thanks,
> 	Zwane

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