Re: FW: + qla3xxx-is-bust.patch added to -mm tree

[Jeff Garzik <jeff@garzik.org>]

Ron Mercer wrote:
> Jeff,
> 
> Regarding your June 22nd post:
> 
>> 14) Surely there is a better way to down the adapter than masking the 
>> interrupts and resetting the adapter?  If this is ever used in non-MSI
> 
>> situations (common in Linux today), there is the possibility of 
>> screaming interrupts, in shared-interrupt situations.
> 
> 
> The actual downing of the adapter frees interrupts before issuing a
> reset.  Perhaps you are referring to the ql_reset_work() path?  This
> handles chip reset requests generated by the iSCSI driver or iSCSI
> firmware.  What you see is that interrupts are masked and reset is
> allowed to continue.  I have made a change where the interrupt is freed
> before the reset continues.  I will post a new patch that covers all
> issues tomorrow.

Freeing a kernel interrupt reservation does not imply that your hardware 
has stopped delivering them...  in a shared interrupt situation, the 
code sequence

	free_irq()
	perform some action

could cause the kernel to call _another_ driver's interrupt handling 
routine over and over again, until _your_ hardware stops raising the 
interrupt condition.

	Jeff