Re: [Madwifi-devel] [Call For Ideas] "madwifi.org labs": setting up a testbed
Status: Beta
Brought to you by:
otaku
From: bruno r. <br...@th...> - 2008-03-10 03:31:46
|
On Sunday 09 March 2008 23:52:16 John Denker wrote: > > that would probably mean some central server with a lot of serial > > connections to different devices. (maybe netconsole could be used to > > avoid that need?). > > I doubt netconsole would be satisfactory. For sure I don't know how > to make it satisfactory for oops messages. It's too dependent on having > the full interrupt system working. true. so having a serial console is a must. > You don't need a central server. It suffices to have the serial > console of each machine logged by *some* other machine. If each > machine has only one serial port, this requires that if machine > A is logging machine B, then B is logging A over the same port, > which is tricky but doable. If each machine has two serial ports > then there are lots of easy solutions, including round-robin > daisy-chain. The logs can be collected via ordinary ethernet > connections. i don't think that will work. what happens when a machine in the chain crashes? - and we do have to expect crashes when we are debugging and testing kernel drivers. > Coordinating devices is the easy part. The hard part is when > people are testing two different devices that use the same > driver. Who gets to do an insmod? for the same reasons i think we need exclusive access to hardware as a whole. that's why i think of a central server with many serial connections to several embedded devices. the server can be used to compile drivers for different archs, and could hold root filesystems for the devices if necessary. - from there you log into an embedded device and crash it ;) > > one > > more thing we also need is a way to trigger a reset of the devices from > > the server in case things go wrong (they will!). does anyone know good > > solutions for that without needing to build lots of custom hardware? > > This is the "watchdog" function. There are lots of off-the-shelf > watchdog cards, so you don't need to mess with custom hardware. > > Forsooth, many Intel platforms have the i8xx TCO Timer/Watchdog > feature already built into the mobo. See > http://www.av8n.com/computer/htm/kernel-lockup.htm#main-hard > for additional details. ok, that might be a good solution for platforms which have a watchdog. although i'm not too sure if we can trust the watchdog to always work (???)... we have to avoid to have to call people all the time to reset devices... so that means we need embedded devices with the following minimal requirements: * serial console * hardware watchdog * miniPCI or PCIexpress slot bruno |