Re: [uml-devel] __exitcall

[Jon S. <jon...@ya...>]

--- Jeff Dike <jd...@ka...> wrote:
> Don't you have to handle this case anyway?  And if
> you do, doesn't that mean
> that UML exitcalls don't need to be working?
> 
> If UML abandons a device in an unknown state, can
> you just reset it to get
> it back to a known state?

1) If UML is shutdown unexpectedly I leave DMA buffers
allocated in the host.  You will have to take a manual
action to free them when you are sure the hardware is
reset.
2) If UML is shutdown and runs exitcalls. The hardware
drivers receive their shutdown messages. This causes
them to shutdown their DMA, free resources, etc. They
will call dma_free_coherent() which I map out to the
host driver to free the buffers.
3) My PCI support is generic. It works for all
possible PCI devices.  There is no way for me to know
how to reset all possible PCI devices. This is what
each drivers' exitcall routine does and why I need to
call them.

I could leave exitscall out, but then I would never be
able to free any DMA buffers in the host.  The only
time I should free buffers in the host is when I get a
dma_free_coherent() call in UML since the drivers
won't call this unless the hardware is done with the
buffer. The only way to get the drivers to shutdown
their hardware is to call exitcalls.


=====
Jon Smirl
jon...@ya...

__________________________________________________
Do you Yahoo!?
Yahoo! Mail Plus - Powerful. Affordable. Sign up now.
http://mailplus.yahoo.com