Date: Sun, 21 Mar 1999 13:55:20 -0800 From: Alan DuBoff <aland@blueneptune.com> To: Scott Mitchell <scott@dcs.qmw.ac.uk> Cc: freebsd-mobile@FreeBSD.ORG Subject: Re: PCCARD support -- welcome to the revolution! Message-ID: <36F56AC7.96D0374C@blueneptune.com> References: <19990318174601.C7591@dcs.qmw.ac.uk>
next in thread | previous in thread | raw e-mail | index | archive | help
First, sorry I've taken a few days to respond, I've been real busy with the current project I'm working on. At 05:46 PM 3/18/99 +0000, Scott Mitchell wrote: >So what are we aiming to achieve? In approximate order of difficulty, it >seems to me that what we need to work on is: > > - New drivers, for PCCARD controllers and cards. These may be home-grown, > brought in from PAO, ported from Linux, whatever. There seems to be a > lot of this going on already, where people have cards that don't work > and are hacking up drivers for them. This will definitely be an important area, and I think most of us would agree that the majority of drivers will not be written from the ground up at this point, rather that most would be ported from other OSs, such as Linux. Linux is getting more drivers than most other PC unix platforms, so it seems fair that FreeBSD should look at making gains from those potential resources. UnixWare has some ability to use Linux drivers, not sure of the exact status of that since the promotional copy of UnixWare 7 I got at LinuxWorld (odd place to get it) doesn't appear to come with PCMCIA support on the CD, rather you have to get it from UnixWare 2.1 somehow and add it to UnixWare 7. That in itself has prevented me from giving UnixWare a complete evaluation. > - Reworking the support for a whole bunch of purely PCCARD issues. The > ones that spring to mind (in that they've bugged me lately) are: > - Multifunction card support > - CardBus support > - Improved userland utilities (pccardd/pccardc) > - Kernel support for e.g. reading the CIS of a card > - Graceful recovery from unexpected card removal Also, very very important. I'm not totally qualified to speak on the specific issues as I'm running 2.2.8 and it's working well. > etc, etc. I'm sure there's a lot more. I hope that a lot of this can > be worked on with needing to rewite the _whole_ PCCARD support > framework, although my list already seems to be picking on a goodly > chunk of it... I don't like the idea of having to rewrite drivers for changes to the kernel, and most likely because I don't understand the driver model well enough. This is a terrible limitation, depending on what changes need to be done, and problems will arise in the future, most likely worse than they are now, where pccard support will break. This is bad, bottom line. > - Fitting PCMCIA/CardBus into the 'new bus' and 'dynamic device' support > that is apparently somewhere in the pipeline. I expect this _will_ > involve a rewrite of most of the support code and drivers :( For now I > guess we should be tracking what's going on in that area, and making > appropriate suggestions to ease the pain of transition when the time > comes. There are issues with CardBus, and they change for each vendor. The support of CardBus devices has been limited and since most vendors now write drivers for Win9x/NT, that support comes first (and sometimes last:-() I saw a gent from Japan seems to have written a driver for the 32 bit CardBus 3c-575-TX card (big brother for the 16 bit 3c-574-TX I have). I would hate to duplicate effort in porting something if someone was already working on it. Laptop support is important to me. I want to be able to run my OSs on my laptop and be able to take it to a client site and hook it up to their net so I can work remotely as well as onsite. Currently I have NT, FreeBSD 2.2.8, Solaris 7, and Linux 5.2/GNOME installed on different partitions/disks for my laptop. No questions, NT has the best support from the vendors of devices, I've yet to run into a card that doesn't have NT support. Sun has stated they will not support laptops anymore. This is a real bad move for them, and something that everyone here should think about. Solaris x86 runs well on a lot of laptops, but there are no user configurable parameters for their pcic, so you cannot make any changes to it. Sun seems to be aimed at the high end server market for Intel, and I think that is a limitation in the long run for them, since more and more laptops are being deployed daily. FreeBSD is so far the best PC unix I've run for a single processor. That is changing now, but I still give edge to Solaris x86 on SMP platforms. I would like to see some comparisons of FreeBSD and Solaris running with, say, 4 processors in a server. My guess is that Sun still is able to manipulate their wisdom they've learned from years of SMP support on Solaris Sparc, to come out the winner on the high end. Please do not take this statement as a discount to FreeBSD, I need to see some actual systems running and actual numbers for FreeBSD in 2-way, 4-way, and 8-way servers. Both FreeBSD and Linux are new to this arena. Not to get off on a tangent, laptops as we know them today, only have single processors in them for the most part, and FreeBSD is the best PC unix for that environment, so I see laptop support as being very critical for FreeBSD users. In this same light, I have to wonder if there is a disadvantage of running 3.x and -current code on a laptop, due to the SMP support added to 3.x. The support must be there for 3.x and -current so laptops don't get lost in the dust, that is exactly what has happened to Solaris 7 running on laptops. Linux has decent laptop support, and a lot of market hype to echo it. This should not prevent FreeBSD from finding it's way and/or working with Linux machines as a whole. Linux may be in the same boat, that pre-2.2 kernel may be a better choice for a laptop. I've not run the 2.2 kernel on a laptop, so I can't speak. The ability to run Linux binaries is an important feature for FreeBSD, since applications like Oracle have trial offers for Linux but not FreeBSD. In Oracle's case I hear you can get them to sell you a FreeBSD version as they do have it running in house, although it's not readily available or something similar to that. I don't know if the ability to reuse pccard drivers from Linux would be a big advantage or not, my initial guess is that something will be lost in the emulation, most notably performance (maybe not that important for many situations). If this technology proves useful for SCO, they could be big winners in the long run since they are latched onto IBM for the Merced chip, and IBM is without a doubt one of the largest suppliers of laptops in existence. If SCO could convince IBM to preload their 32 bit version on laptops today, that could give them a definite laptop edge. This remains to be seen, and IBM has already stated they WILL preload Linux. With first class laptop support, FreeBSD could very easily work it's way into a situation where it could get vendors to preload it on different laptops. To me the pccard support works well on 2.2.8, I'm not completely up to date on all the changes that need to be made for what you are proposing for -current. I need to understand that better before committing my time, but I am willing to help in understanding this to see how I can help the effort. -- Alan DuBoff - Conductor Software Orchestration, Inc. aland@SoftOrchestra.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-mobile" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?36F56AC7.96D0374C>