Re: [Qemu-devel] [PATCH] Flush pending AIO on reboot and shutdown.

[Samuel Thibault <samuel.thibault@ens-lyon.org>]

Avi Kivity wrote:
> Why not? IIUC triple fault is exported as a cpu pin, which the chipset/
> motherboard wires into the board-level reset line.

Ah ok, I didn't know it was propagated further than the cpu.  Probably a
good thing actually :)

> I think the bios can expect the IDE hardware (or any other device) to be
> quiescent on entry. It doesn't make sense to initiate an explicit reset
> immediately after boot; that would slow things down.

Well, the Bios reset entry could be entered by "accident" too :)
Actually windows drivers used to break my video board so badly that I
had to manually reset my machine before booting Linux.

Well, anyway, I agree with Anthony: the IDE emulation should behave
like hardware: try to cancel pending requests and wait for everything
to finish, instead of possibly writing what the bios puts in the being
written pages :) And it should reset the DMA engines too.

The block layer itself could have a last sync check.

Samuel