Re: [RFC / PATCH] [SPI] get_module while using it.

[David Brownell <david-b-yBeKhBN/0LDR7s880joybQ@public.gmane.org>]

On Tuesday 15 April 2008, Sebastian Siewior wrote:
> * David Brownell | 2008-04-14 16:23:36 [-0700]:
> 
> >On Monday 14 April 2008, Sebastian Siewior wrote:
> >> Without this it is possible to unload the SPI-driver while an
> >> spidev user open()ed the spidev device.
> >
> >Does that misbehave in any way?  That sort of unloading of
>
> Yes. Unload, read() -> Ooops.

What's the oops, and with which controller driver?

If the adapter is unloaded it's *supposed* to first complete
all pending transfers (and prevent submission of new ones).

And removing the subsidiary devices is supposed to unbind
them ... though looking at it quickly, I suspect that the
"spidev" driver doesn't check for unbound devices in its
various I/O paths.

So this sounds like a bug in the controller driver, or else
in how spidev handles unbind.

- Dave


 
> >adapter code is possible with USB and I2C and hasn't been
> >perceived as much of a problem.  I'm not sure I see a strong
> >reason for SPI to act very different.
>
> True. usb returns with -ENODEV after read/whatever. We could add
> something like this to spi/spidev. This might also fit better into
> hot-plugging.

-------------------------------------------------------------------------
This SF.net email is sponsored by the 2008 JavaOne(SM) Conference 
Don't miss this year's exciting event. There's still time to save $100. 
Use priority code J8TL2D2. 
http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone