Messages in this thread | | | From | (david parsons) | Subject | Re: [OT] RE: UDI and Free Software | Date | 7 Oct 1998 13:23:41 -0700 |
| |
In article <linux.kernel.199810071736.NAA05114@dcl.mit.edu>, Theodore Y. Ts'o <tytso@MIT.EDU> wrote: > > > to native drivers... For some hardware it may not make a difference > > (for example, a scanner). But for other hardware (such as a gigabit > > ethernet board), my guess is that UDI will be a disaster as far as > > performance is concerned. > > UDI has no concept of other interface libraries or access API's - so > it lacks stuff like check_region that is central to the Linux philosophy. > UDI on Linux would also be useless for parallel scanners - UDI can't > describe talking to the parport management layer > >Exactly. So you'd have to do an extra memory copy in the Linux<->UDI >interface layer.
I guess it depends on the driver. I looked at one of the scsi drivers to see just what check_region() was used for (2.0.28:advansys.c, which is a device driver from a vendor, and thus more likely to be udi'zed if a udi driver interface went into the kernel and to my application programmer eyes it looks like all it uses check_region() for is while probing for devices. Now, if we need to use check_region() to make sure we don't stomp on other device while we're doing probing, I'm assming that all the other Unices out there need to do something much the same <provided, of course, that they aren't stuuupid(tm) and will let drivers trash the system while they're loading> while they're detecting devices, but alas there is no large SAMPLE.IMPLEMENTATION directory on the website for UDI.)
>It'll be slow, but you an make it work.
I've looked through a few of the scsi drivers, seeing what check_region() is used for. It looks like it's used for initialization and nothing else? If this is the case, do we care about Blinding Speed(tm) -- There may be people out there agitating for the fastest possible driver initialization, but, at least for disk devices you spend a lot more time actually waiting for the disks to identify themselves than you do poking at the controller.
____ david parsons \bi/ If UDI doesn't support the abstractions needed to \/ initialize Linux device drivers, that sounds like a good place to propose changing the standard to fit.
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |