Date: Tue, 11 Dec 2001 06:14:45 -0800 From: Terry Lambert <tlambert2@mindspring.com> To: Danny Braniss <danny@cs.huji.ac.il> Cc: Bernd Walter <ticso@cicely8.cicely.de>, Mike Smith <msmith@FreeBSD.ORG>, freebsd-hackers@FreeBSD.ORG Subject: Re: irq Message-ID: <3C1614D5.B5C3B4C5@mindspring.com> References: <E16Dk3D-000AIb-00@pampa.cs.huji.ac.il>
next in thread | previous in thread | raw e-mail | index | archive | help
Danny Braniss wrote: > > > It looks like IRQ sharing is the only issue here. > > Q: are all interrupt handlers for the shared irq called, or only the > 'correct' one? All interrupt handlers are polled -- asked if they have work to do, on the basis of having caused the interrupt. Some cards do not have a hardware "I caused an interrupt" register, and use a differential (e.g. ring head vs. tail inequal after interrupt) to tell if there is work to do. If these cards were to share interrupts, then they most likely do work every interrupt, and less work per interrupt, then would otherwise be the case (i.e. it would defeat hardware interrupt coelescing thresholds, among other things). > [...] > > > But I would still blame software (in particular, the firmware on > > your Meteor). > > > > -- Terry > > isn't firmware another word for hardware :-)? No, because you can fix firmware in software, but you can't fix hardware in software, and get anywhere near the performance (e.g. CMD640B, etc.). For example, you could use the PCI bus management code to ensure broken cards got seperate interrupts from all other cards. -- Terry To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?3C1614D5.B5C3B4C5>