bug-hurd
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: user-space device drivers - hurd


From: Sergio Lopez
Subject: Re: user-space device drivers - hurd
Date: Fri, 16 Sep 2005 01:03:12 +0200

Hello,

El jue, 15-09-2005 a las 08:13 -0700, Shakthi Kannan escribió:

> 3. I want to try to strip down the gnumach kernel and
> drivers to a minimal. Or is the default one the
> minimum? I'd appreciate developers' inputs and their
> experience they had on what makes GNUMach slow and
> what needs to be fixed or what cannot be fixed. I am
> particularly interested only in hardware and device
> drivers.
> 

AFAIK, some of the things that slow down Mach/Hurd are:

1. The way in which memory objects are accessed by the translators
(vm_map+memcpy)

2. The fact that Mach/Hurd can't transfer more than one page (4K) at
time to/from external pagers.

3. IPC scheduling

jemarch is working on the last (by implementing "migrating threads" on
GNU Mach) and I'm working on the first two (but help is really
appreciated :-)

Also, there's something that would be _very_ usefull and is related to
hardware/drivers: upgrading the glue code of GNU Mach to support drivers
from linux-2.6 (2.4 will be also fine ;-) and the architecture dependant
code (i386/*).

> 4. Do we write drivers in user-space or should I try
> to use the existing device driver interface for
> writing drivers? 
> 

IMHO, user-space drivers could be a nice feature in a future (for
non-critical low-traffic devices), but right now they can't provide
nothing better than more slowness and less stability.

Happy Hacking!

-- 
Sergio Lopez
koro@sinrega.org





reply via email to

[Prev in Thread] Current Thread [Next in Thread]